diff options
Diffstat (limited to 'src/gallium')
527 files changed, 64067 insertions, 11775 deletions
diff --git a/src/gallium/Makefile.template b/src/gallium/Makefile.template index 98487d43bd6..63983c52201 100644 --- a/src/gallium/Makefile.template +++ b/src/gallium/Makefile.template @@ -31,8 +31,8 @@ INCLUDES = \ default: depend lib$(LIBNAME).a -lib$(LIBNAME).a: $(OBJECTS) Makefile $(TOP)/src/gallium/Makefile.template - $(MKLIB) -o $(LIBNAME) -static $(OBJECTS) +lib$(LIBNAME).a: $(OBJECTS) $(EXTRA_OBJECTS) Makefile $(TOP)/src/gallium/Makefile.template + $(MKLIB) -o $(LIBNAME) -static $(OBJECTS) $(EXTRA_OBJECTS) depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(SYMLINKS) rm -f depend @@ -41,7 +41,7 @@ depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(SYMLINKS) # Emacs tags tags: - etags `find . -name \*.[ch]` `find ../include` + etags `find . -name \*.[ch]` `find $(TOP)/src/gallium/include -name \*.h` # Remove .o and backup files clean: diff --git a/src/gallium/SConscript b/src/gallium/SConscript index b6ceaf3edf9..89c69d7205e 100644 --- a/src/gallium/SConscript +++ b/src/gallium/SConscript @@ -22,6 +22,7 @@ SConscript([ 'auxiliary/draw/SConscript', 'auxiliary/pipebuffer/SConscript', 'auxiliary/indices/SConscript', + 'auxiliary/rbug/SConscript', ]) for driver in env['drivers']: @@ -30,6 +31,7 @@ for driver in env['drivers']: SConscript('state_trackers/python/SConscript') SConscript('state_trackers/glx/xlib/SConscript') SConscript('state_trackers/dri/SConscript') +SConscript('state_trackers/xorg/SConscript') if platform == 'windows': SConscript('state_trackers/wgl/SConscript') diff --git a/src/gallium/auxiliary/cso_cache/cso_cache.c b/src/gallium/auxiliary/cso_cache/cso_cache.c index 0bc77a57287..e6dce3f0e5b 100644 --- a/src/gallium/auxiliary/cso_cache/cso_cache.c +++ b/src/gallium/auxiliary/cso_cache/cso_cache.c @@ -361,6 +361,10 @@ void cso_for_each_state(struct cso_cache *sc, enum cso_cache_type type, void cso_cache_delete(struct cso_cache *sc) { assert(sc); + + if (!sc) + return; + /* delete driver data */ cso_for_each_state(sc, CSO_BLEND, delete_blend_state, 0); cso_for_each_state(sc, CSO_DEPTH_STENCIL_ALPHA, delete_depth_stencil_state, 0); diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c index f388bf5d951..36c882acb75 100644 --- a/src/gallium/auxiliary/cso_cache/cso_context.c +++ b/src/gallium/auxiliary/cso_cache/cso_context.c @@ -55,14 +55,14 @@ struct cso_context { void *samplers[PIPE_MAX_SAMPLERS]; unsigned nr_samplers; - void *samplers_saved[PIPE_MAX_SAMPLERS]; unsigned nr_samplers_saved; + void *samplers_saved[PIPE_MAX_SAMPLERS]; struct pipe_texture *textures[PIPE_MAX_SAMPLERS]; uint nr_textures; - struct pipe_texture *textures_saved[PIPE_MAX_SAMPLERS]; uint nr_textures_saved; + struct pipe_texture *textures_saved[PIPE_MAX_SAMPLERS]; /** Current and saved state. * The saved state is used as a 1-deep stack. diff --git a/src/gallium/auxiliary/draw/draw_pipe.c b/src/gallium/auxiliary/draw/draw_pipe.c index 3cfae2de831..1c6d657297c 100644 --- a/src/gallium/auxiliary/draw/draw_pipe.c +++ b/src/gallium/auxiliary/draw/draw_pipe.c @@ -248,22 +248,22 @@ void draw_pipeline_run( struct draw_context *draw, ( DRAW_PIPE_RESET_STIPPLE | \ DRAW_PIPE_EDGE_FLAG_0 | \ DRAW_PIPE_EDGE_FLAG_2 ), \ - verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \ - verts + stride * (i1), \ + verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (i1), \ verts + stride * (i3)); \ - do_triangle( draw, \ - ( DRAW_PIPE_EDGE_FLAG_0 | \ - DRAW_PIPE_EDGE_FLAG_1 ), \ - verts + stride * ((i1) & ~DRAW_PIPE_FLAG_MASK), \ - verts + stride * (i2), \ - verts + stride * (i3)) + do_triangle( draw, \ + ( DRAW_PIPE_EDGE_FLAG_0 | \ + DRAW_PIPE_EDGE_FLAG_1 ), \ + verts + stride * ((i1) & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (i2), \ + verts + stride * (i3)) #define TRIANGLE(flags,i0,i1,i2) \ do_triangle( draw, \ flags, /* flags */ \ - verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \ - verts + stride * (i1), \ - verts + stride * (i2)) + verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (i1), \ + verts + stride * (i2)) #define LINE(flags,i0,i1) \ do_line( draw, \ diff --git a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c index 30a6d2919d9..283502cdf3e 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c +++ b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c @@ -256,7 +256,10 @@ pstip_transform_inst(struct tgsi_transform_context *ctx, uint size = 4; immed = tgsi_default_full_immediate(); immed.Immediate.NrTokens = 1 + size; /* one for the token itself */ - immed.u.Pointer = (void *) value; + immed.u[0].Float = value[0]; + immed.u[1].Float = value[1]; + immed.u[2].Float = value[2]; + immed.u[3].Float = value[3]; ctx->emit_immediate(ctx, &immed); } diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c index a5d840b96ea..1a5269c0de9 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c +++ b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c @@ -159,8 +159,19 @@ vbuf_tri( struct draw_stage *stage, check_space( vbuf, 3 ); - for (i = 0; i < 3; i++) { - vbuf->indices[vbuf->nr_indices++] = emit_vertex( vbuf, prim->v[i] ); + if (vbuf->stage.draw->rasterizer->flatshade_first) { + /* Put provoking vertex in position expected by the driver. + * Emit last provoking vertex in first pos. + * Swap verts 0 & 1 to preserve polygon winding. + */ + vbuf->indices[vbuf->nr_indices++] = emit_vertex( vbuf, prim->v[2] ); + vbuf->indices[vbuf->nr_indices++] = emit_vertex( vbuf, prim->v[0] ); + vbuf->indices[vbuf->nr_indices++] = emit_vertex( vbuf, prim->v[1] ); + } + else { + for (i = 0; i < 3; i++) { + vbuf->indices[vbuf->nr_indices++] = emit_vertex( vbuf, prim->v[i] ); + } } } diff --git a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c index d84bab9eaae..7d76a7dbf39 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c +++ b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c @@ -28,6 +28,30 @@ /* Authors: Keith Whitwell <keith@tungstengraphics.com> */ +/** + * Notes on wide points and sprite mode: + * + * In wide point/sprite mode we effectively need to convert each incoming + * vertex into four outgoing vertices specifying the corners of a quad. + * Since we don't (yet) have geometry shaders, we have to handle this here + * in the draw module. + * + * For sprites, it also means that this is where we have to handle texcoords + * for the vertices of the quad. OpenGL's GL_COORD_REPLACE state specifies + * if/how enabled texcoords are automatically generated for sprites. We pass + * that info through gallium in the pipe_rasterizer_state::sprite_coord_mode + * array. + * + * Additionally, GLSL's gl_PointCoord fragment attribute has to be handled + * here as well. This is basically an additional texture/generic attribute + * that varies .x from 0 to 1 horizontally across the point and varies .y + * vertically from 0 to 1 down the sprite. + * + * With geometry shaders, the state tracker could create a GS to do + * most/all of this. + */ + + #include "util/u_math.h" #include "util/u_memory.h" #include "pipe/p_defines.h" @@ -52,7 +76,7 @@ struct widepoint_stage { int psize_slot; - int point_coord_fs_input; /**< input for pointcoord (and fog) */ + int point_coord_fs_input; /**< input for pointcoord */ }; @@ -64,8 +88,6 @@ widepoint_stage( struct draw_stage *stage ) } - - /** * Set the vertex texcoords for sprite mode. * Coords may be left untouched or set to a right-side-up or upside-down @@ -89,10 +111,12 @@ static void set_texcoords(const struct widepoint_stage *wide, } if (wide->point_coord_fs_input >= 0) { - /* put gl_PointCoord into extra vertex output's zw components */ - uint k = wide->stage.draw->extra_vp_outputs.slot; - v->data[k][2] = tc[0]; - v->data[k][3] = tc[1]; + /* put gl_PointCoord into the extra vertex slot */ + uint slot = wide->stage.draw->extra_vp_outputs.slot; + v->data[slot][0] = tc[0]; + v->data[slot][1] = tc[1]; + v->data[slot][2] = 0.0F; + v->data[slot][3] = 1.0F; } } @@ -181,6 +205,16 @@ static void widepoint_point( struct draw_stage *stage, } +static int +find_pntc_input_attrib(struct draw_context *draw) +{ + /* Scan the fragment program's input decls to find the pointcoord + * attribute. The xy components will store the point coord. + */ + return 0; /* XXX fix this */ +} + + static void widepoint_first_point( struct draw_stage *stage, struct prim_header *header ) { @@ -219,8 +253,8 @@ static void widepoint_first_point( struct draw_stage *stage, } wide->num_texcoords = j; - /* find fragment shader PointCoord/Fog input */ - wide->point_coord_fs_input = 0; /* XXX fix this! */ + /* find fragment shader PointCoord input */ + wide->point_coord_fs_input = find_pntc_input_attrib(draw); /* setup extra vp output (point coord implemented as a texcoord) */ draw->extra_vp_outputs.semantic_name = TGSI_SEMANTIC_GENERIC; diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h index 81e4eae401c..41fcb16a0a5 100644 --- a/src/gallium/auxiliary/draw/draw_private.h +++ b/src/gallium/auxiliary/draw/draw_private.h @@ -44,7 +44,6 @@ #include "pipe/p_state.h" #include "pipe/p_defines.h" -#include "tgsi/tgsi_exec.h" #include "tgsi/tgsi_scan.h" @@ -55,6 +54,8 @@ struct draw_vertex_shader; struct draw_context; struct draw_stage; struct vbuf_render; +struct tgsi_exec_machine; +struct tgsi_sampler; /** @@ -185,7 +186,7 @@ struct draw_context uint position_output; /** TGSI program interpreter runtime state */ - struct tgsi_exec_machine machine; + struct tgsi_exec_machine *machine; uint num_samplers; struct tgsi_sampler **samplers; diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c index 9ea0cbe5990..dbb5ac71821 100644 --- a/src/gallium/auxiliary/draw/draw_pt.c +++ b/src/gallium/auxiliary/draw/draw_pt.c @@ -36,6 +36,7 @@ #include "draw/draw_vs.h" #include "tgsi/tgsi_dump.h" #include "util/u_math.h" +#include "util/u_prim.h" static unsigned trim( unsigned count, unsigned first, unsigned incr ) { @@ -278,7 +279,7 @@ void draw_arrays(struct draw_context *draw, unsigned prim, unsigned start, unsigned count) { - unsigned reduced_prim = draw_pt_reduced_prim(prim); + unsigned reduced_prim = u_reduced_prim(prim); if (reduced_prim != draw->reduced_prim) { draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); draw->reduced_prim = reduced_prim; diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h index 8ef0ea8011f..7a17a9fb6b2 100644 --- a/src/gallium/auxiliary/draw/draw_pt.h +++ b/src/gallium/auxiliary/draw/draw_pt.h @@ -228,7 +228,6 @@ void draw_pt_post_vs_destroy( struct pt_post_vs *pvs ); * Utils: */ void draw_pt_split_prim(unsigned prim, unsigned *first, unsigned *incr); -unsigned draw_pt_reduced_prim(unsigned prim); #endif diff --git a/src/gallium/auxiliary/draw/draw_pt_decompose.h b/src/gallium/auxiliary/draw/draw_pt_decompose.h index 3fb06956878..4ca5b520204 100644 --- a/src/gallium/auxiliary/draw/draw_pt_decompose.h +++ b/src/gallium/auxiliary/draw/draw_pt_decompose.h @@ -47,10 +47,19 @@ static void FUNC( ARGS, case PIPE_PRIM_TRIANGLES: for (i = 0; i+2 < count; i += 3) { - TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, - (i + 0), - (i + 1), - (i + 2 )); + if (flatfirst) { + /* put provoking vertex in last pos for clipper */ + TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, + (i + 1), + (i + 2), + (i + 0 )); + } + else { + TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, + (i + 0), + (i + 1), + (i + 2 )); + } } break; @@ -58,9 +67,9 @@ static void FUNC( ARGS, if (flatfirst) { for (i = 0; i+2 < count; i++) { TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, - (i + 0), (i + 1 + (i&1)), - (i + 2 - (i&1))); + (i + 2 - (i&1)), + (i + 0) ); } } else { @@ -78,9 +87,9 @@ static void FUNC( ARGS, if (flatfirst) { for (i = 0; i+2 < count; i++) { TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, - (i + 1), (i + 2), - (0 )); + 0, + (i + 1) ); } } else { @@ -96,20 +105,40 @@ static void FUNC( ARGS, case PIPE_PRIM_QUADS: - for (i = 0; i+3 < count; i += 4) { - QUAD( (i + 0), - (i + 1), - (i + 2), - (i + 3)); + if (flatfirst) { + for (i = 0; i+3 < count; i += 4) { + QUAD( (i + 1), + (i + 2), + (i + 3), + (i + 0) ); + } + } + else { + for (i = 0; i+3 < count; i += 4) { + QUAD( (i + 0), + (i + 1), + (i + 2), + (i + 3)); + } } break; case PIPE_PRIM_QUAD_STRIP: - for (i = 0; i+3 < count; i += 2) { - QUAD( (i + 2), - (i + 0), - (i + 1), - (i + 3)); + if (flatfirst) { + for (i = 0; i+3 < count; i += 2) { + QUAD( (i + 1), + (i + 3), + (i + 2), + (i + 0) ); + } + } + else { + for (i = 0; i+3 < count; i += 2) { + QUAD( (i + 2), + (i + 0), + (i + 1), + (i + 3)); + } } break; diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c index 6b7d02a19bc..e7fe6b3b768 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c @@ -130,6 +130,10 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle, unsigned output_format; switch (vinfo->attrib[i].emit) { + case EMIT_4UB: + output_format = PIPE_FORMAT_R8G8B8A8_UNORM; + emit_sz = 4 * sizeof(unsigned char); + break; case EMIT_4F: output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; emit_sz = 4 * sizeof(float); @@ -153,6 +157,8 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle, output_format = PIPE_FORMAT_R32_FLOAT; emit_sz = 1 * sizeof(float); break; + case EMIT_OMIT: + continue; default: assert(0); output_format = PIPE_FORMAT_NONE; diff --git a/src/gallium/auxiliary/draw/draw_pt_util.c b/src/gallium/auxiliary/draw/draw_pt_util.c index 3bc7939c556..b61fa291436 100644 --- a/src/gallium/auxiliary/draw/draw_pt_util.c +++ b/src/gallium/auxiliary/draw/draw_pt_util.c @@ -75,28 +75,3 @@ void draw_pt_split_prim(unsigned prim, unsigned *first, unsigned *incr) break; } } - - -unsigned draw_pt_reduced_prim(unsigned prim) -{ - switch (prim) { - case PIPE_PRIM_POINTS: - return PIPE_PRIM_POINTS; - case PIPE_PRIM_LINES: - case PIPE_PRIM_LINE_STRIP: - case PIPE_PRIM_LINE_LOOP: - return PIPE_PRIM_LINES; - case PIPE_PRIM_TRIANGLES: - case PIPE_PRIM_TRIANGLE_STRIP: - case PIPE_PRIM_TRIANGLE_FAN: - case PIPE_PRIM_POLYGON: - case PIPE_PRIM_QUADS: - case PIPE_PRIM_QUAD_STRIP: - return PIPE_PRIM_TRIANGLES; - default: - assert(0); - return PIPE_PRIM_POINTS; - } -} - - diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache.c b/src/gallium/auxiliary/draw/draw_pt_vcache.c index 5d268a22264..1a0527be63a 100644 --- a/src/gallium/auxiliary/draw/draw_pt_vcache.c +++ b/src/gallium/auxiliary/draw/draw_pt_vcache.c @@ -31,6 +31,7 @@ */ #include "util/u_memory.h" +#include "util/u_prim.h" #include "draw/draw_context.h" #include "draw/draw_private.h" #include "draw/draw_pt.h" @@ -193,16 +194,30 @@ vcache_ef_quad( struct vcache_frontend *vcache, unsigned i2, unsigned i3 ) { - vcache_triangle_flags( vcache, - ( DRAW_PIPE_RESET_STIPPLE | - DRAW_PIPE_EDGE_FLAG_0 | - DRAW_PIPE_EDGE_FLAG_2 ), - i0, i1, i3 ); - - vcache_triangle_flags( vcache, - ( DRAW_PIPE_EDGE_FLAG_0 | - DRAW_PIPE_EDGE_FLAG_1 ), - i1, i2, i3 ); + if (vcache->draw->rasterizer->flatshade_first) { + vcache_triangle_flags( vcache, + ( DRAW_PIPE_RESET_STIPPLE | + DRAW_PIPE_EDGE_FLAG_0 | + DRAW_PIPE_EDGE_FLAG_1 ), + i0, i1, i2 ); + + vcache_triangle_flags( vcache, + ( DRAW_PIPE_EDGE_FLAG_2 | + DRAW_PIPE_EDGE_FLAG_1 ), + i0, i2, i3 ); + } + else { + vcache_triangle_flags( vcache, + ( DRAW_PIPE_RESET_STIPPLE | + DRAW_PIPE_EDGE_FLAG_0 | + DRAW_PIPE_EDGE_FLAG_2 ), + i0, i1, i3 ); + + vcache_triangle_flags( vcache, + ( DRAW_PIPE_EDGE_FLAG_0 | + DRAW_PIPE_EDGE_FLAG_1 ), + i1, i2, i3 ); + } } /* At least for now, we're back to using a template include file for @@ -453,7 +468,7 @@ vcache_prepare( struct draw_pt_front_end *frontend, } vcache->input_prim = prim; - vcache->output_prim = draw_pt_reduced_prim(prim); + vcache->output_prim = u_reduced_prim(prim); vcache->middle = middle; vcache->opt = opt; diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h b/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h index ec05bbeab40..62822a3d562 100644 --- a/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h +++ b/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h @@ -118,21 +118,39 @@ static void FUNC( struct draw_pt_front_end *frontend, case PIPE_PRIM_QUADS: for (i = 0; i+3 < count; i += 4) { - QUAD( vcache, - get_elt(elts, i + 0), - get_elt(elts, i + 1), - get_elt(elts, i + 2), - get_elt(elts, i + 3)); + if (flatfirst) { + QUAD( vcache, + get_elt(elts, i + 0), + get_elt(elts, i + 1), + get_elt(elts, i + 2), + get_elt(elts, i + 3) ); + } + else { + QUAD( vcache, + get_elt(elts, i + 0), + get_elt(elts, i + 1), + get_elt(elts, i + 2), + get_elt(elts, i + 3) ); + } } break; case PIPE_PRIM_QUAD_STRIP: for (i = 0; i+3 < count; i += 2) { - QUAD( vcache, - get_elt(elts, i + 2), - get_elt(elts, i + 0), - get_elt(elts, i + 1), - get_elt(elts, i + 3)); + if (flatfirst) { + QUAD( vcache, + get_elt(elts, i + 0), + get_elt(elts, i + 1), + get_elt(elts, i + 3), + get_elt(elts, i + 2) ); + } + else { + QUAD( vcache, + get_elt(elts, i + 2), + get_elt(elts, i + 0), + get_elt(elts, i + 1), + get_elt(elts, i + 3) ); + } } break; @@ -144,19 +162,38 @@ static void FUNC( struct draw_pt_front_end *frontend, const ushort edge_first = DRAW_PIPE_EDGE_FLAG_2; const ushort edge_middle = DRAW_PIPE_EDGE_FLAG_0; const ushort edge_last = DRAW_PIPE_EDGE_FLAG_1; + ushort edge_next, edge_finish; - flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle; + if (flatfirst) { + flags = DRAW_PIPE_RESET_STIPPLE | edge_middle | edge_last; + edge_next = edge_last; + edge_finish = edge_first; + } + else { + flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle; + edge_next = edge_middle; + edge_finish = edge_last; + } - for (i = 0; i+2 < count; i++, flags = edge_middle) { + for (i = 0; i+2 < count; i++, flags = edge_next) { if (i + 3 == count) - flags |= edge_last; + flags |= edge_finish; - TRIANGLE( vcache, - flags, - get_elt(elts, i + 1), - get_elt(elts, i + 2), - get_elt(elts, 0)); + if (flatfirst) { + TRIANGLE( vcache, + flags, + get_elt(elts, 0), + get_elt(elts, i + 1), + get_elt(elts, i + 2) ); + } + else { + TRIANGLE( vcache, + flags, + get_elt(elts, i + 1), + get_elt(elts, i + 2), + get_elt(elts, 0)); + } } } break; diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c index c057cd67fde..790e89ed820 100644 --- a/src/gallium/auxiliary/draw/draw_vs.c +++ b/src/gallium/auxiliary/draw/draw_vs.c @@ -43,6 +43,8 @@ #include "translate/translate.h" #include "translate/translate_cache.h" +#include "tgsi/tgsi_exec.h" + @@ -146,16 +148,8 @@ draw_delete_vertex_shader(struct draw_context *draw, boolean draw_vs_init( struct draw_context *draw ) { - tgsi_exec_machine_init(&draw->vs.machine); - - /* FIXME: give this machine thing a proper constructor: - */ - draw->vs.machine.Inputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16); - if (!draw->vs.machine.Inputs) - return FALSE; - - draw->vs.machine.Outputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16); - if (!draw->vs.machine.Outputs) + draw->vs.machine = tgsi_exec_machine_create(); + if (!draw->vs.machine) return FALSE; draw->vs.emit_cache = translate_cache_create(); @@ -178,12 +172,6 @@ draw_vs_init( struct draw_context *draw ) void draw_vs_destroy( struct draw_context *draw ) { - if (draw->vs.machine.Inputs) - align_free(draw->vs.machine.Inputs); - - if (draw->vs.machine.Outputs) - align_free(draw->vs.machine.Outputs); - if (draw->vs.fetch_cache) translate_cache_destroy(draw->vs.fetch_cache); @@ -196,8 +184,7 @@ draw_vs_destroy( struct draw_context *draw ) if (draw->vs.aligned_constant_storage) align_free((void*)draw->vs.aligned_constant_storage); - tgsi_exec_machine_free_data(&draw->vs.machine); - + tgsi_exec_machine_destroy(draw->vs.machine); } diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 9e37a26c1e2..62e04a65f30 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -1758,24 +1758,24 @@ emit_instruction( struct aos_compilation *cp, case TGSI_OPCODE_SUB: return emit_SUB(cp, inst); - case TGSI_OPCODE_LERP: + case TGSI_OPCODE_LRP: // return emit_LERP(cp, inst); return FALSE; - case TGSI_OPCODE_FRAC: + case TGSI_OPCODE_FRC: return emit_FRC(cp, inst); case TGSI_OPCODE_CLAMP: // return emit_CLAMP(cp, inst); return FALSE; - case TGSI_OPCODE_FLOOR: + case TGSI_OPCODE_FLR: return emit_FLR(cp, inst); case TGSI_OPCODE_ROUND: return emit_RND(cp, inst); - case TGSI_OPCODE_EXPBASE2: + case TGSI_OPCODE_EX2: #if FAST_MATH return emit_EXPBASE2(cp, inst); #elif 0 @@ -1787,13 +1787,13 @@ emit_instruction( struct aos_compilation *cp, return FALSE; #endif - case TGSI_OPCODE_LOGBASE2: + case TGSI_OPCODE_LG2: return emit_LG2(cp, inst); - case TGSI_OPCODE_POWER: + case TGSI_OPCODE_POW: return emit_POW(cp, inst); - case TGSI_OPCODE_CROSSPRODUCT: + case TGSI_OPCODE_XPD: return emit_XPD(cp, inst); case TGSI_OPCODE_ABS: @@ -1891,8 +1891,9 @@ static boolean note_immediate( struct aos_compilation *cp, unsigned pos = cp->num_immediates++; unsigned j; + assert( imm->Immediate.NrTokens <= 4 + 1 ); for (j = 0; j < imm->Immediate.NrTokens - 1; j++) { - cp->vaos->machine->immediate[pos][j] = imm->u.ImmediateFloat32[j].Float; + cp->vaos->machine->immediate[pos][j] = imm->u[j].Float; } return TRUE; diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c index dbbc33fffa7..41cc8026131 100644 --- a/src/gallium/auxiliary/draw/draw_vs_exec.c +++ b/src/gallium/auxiliary/draw/draw_vs_exec.c @@ -41,6 +41,7 @@ #include "tgsi/tgsi_parse.h" #include "tgsi/tgsi_scan.h" +#include "tgsi/tgsi_exec.h" struct exec_vertex_shader { @@ -114,6 +115,12 @@ vs_exec_run_linear( struct draw_vertex_shader *shader, #endif for (slot = 0; slot < shader->info.num_inputs; slot++) { +#if 0 + assert(!util_is_inf_or_nan(input[slot][0])); + assert(!util_is_inf_or_nan(input[slot][1])); + assert(!util_is_inf_or_nan(input[slot][2])); + assert(!util_is_inf_or_nan(input[slot][3])); +#endif machine->Inputs[slot].xyzw[0].f[j] = input[slot][0]; machine->Inputs[slot].xyzw[1].f[j] = input[slot][1]; machine->Inputs[slot].xyzw[2].f[j] = input[slot][2]; @@ -195,7 +202,7 @@ draw_create_vs_exec(struct draw_context *draw, vs->base.run_linear = vs_exec_run_linear; vs->base.delete = vs_exec_delete; vs->base.create_varient = draw_vs_varient_generic; - vs->machine = &draw->vs.machine; + vs->machine = draw->vs.machine; return &vs->base; } diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c index 727977bc3af..b3535c0e48e 100644 --- a/src/gallium/auxiliary/draw/draw_vs_llvm.c +++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c @@ -119,7 +119,7 @@ draw_create_vs_llvm(struct draw_context *draw, vs->base.create_varient = draw_vs_varient_generic; vs->base.run_linear = vs_llvm_run_linear; vs->base.delete = vs_llvm_delete; - vs->machine = &draw->vs.machine; + vs->machine = draw->vs.machine; { struct gallivm_ir *ir = gallivm_ir_new(GALLIVM_VS); diff --git a/src/gallium/auxiliary/draw/draw_vs_ppc.c b/src/gallium/auxiliary/draw/draw_vs_ppc.c index d35db57d571..ad184bd696d 100644 --- a/src/gallium/auxiliary/draw/draw_vs_ppc.c +++ b/src/gallium/auxiliary/draw/draw_vs_ppc.c @@ -48,6 +48,7 @@ #include "rtasm/rtasm_ppc.h" #include "tgsi/tgsi_ppc.h" #include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_exec.h" diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c index 77ba5152f9f..702051387ac 100644 --- a/src/gallium/auxiliary/draw/draw_vs_sse.c +++ b/src/gallium/auxiliary/draw/draw_vs_sse.c @@ -48,27 +48,16 @@ #include "rtasm/rtasm_x86sse.h" #include "tgsi/tgsi_sse2.h" #include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_exec.h" #define SSE_MAX_VERTICES 4 -typedef void (PIPE_CDECL *codegen_function) ( - const struct tgsi_exec_vector *input, /* 1 */ - struct tgsi_exec_vector *output, /* 2 */ - float (*constant)[4], /* 3 */ - struct tgsi_exec_vector *temporary, /* 4 */ - float (*immediates)[4], /* 5 */ - const float (*aos_input)[4], /* 6 */ - uint num_inputs, /* 7 */ - uint input_stride, /* 8 */ - float (*aos_output)[4], /* 9 */ - uint num_outputs, /* 10 */ - uint output_stride ); /* 11 */ struct draw_sse_vertex_shader { struct draw_vertex_shader base; struct x86_function sse2_program; - codegen_function func; + tgsi_sse2_vs_func func; struct tgsi_exec_machine *machine; }; @@ -78,6 +67,10 @@ static void vs_sse_prepare( struct draw_vertex_shader *base, struct draw_context *draw ) { + struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base; + struct tgsi_exec_machine *machine = shader->machine; + + machine->Samplers = draw->vs.samplers; } @@ -118,11 +111,9 @@ vs_sse_run_linear( struct draw_vertex_shader *base, /* run compiled shader */ - shader->func(machine->Inputs, - machine->Outputs, - (float (*)[4])constants, - machine->Temps, - (float (*)[4])shader->base.immediates, + shader->func(machine, + constants, + shader->base.immediates, input, base->info.num_inputs, input_stride, @@ -184,7 +175,7 @@ draw_create_vs_sse(struct draw_context *draw, vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 * sizeof(float), 16); - vs->machine = &draw->vs.machine; + vs->machine = draw->vs.machine; x86_init_func( &vs->sse2_program ); @@ -194,7 +185,7 @@ draw_create_vs_sse(struct draw_context *draw, TRUE )) goto fail; - vs->func = (codegen_function) x86_get_func( &vs->sse2_program ); + vs->func = (tgsi_sse2_vs_func) x86_get_func( &vs->sse2_program ); if (!vs->func) { goto fail; } diff --git a/src/gallium/auxiliary/gallivm/instructionssoa.cpp b/src/gallium/auxiliary/gallivm/instructionssoa.cpp index 925e948763e..721b7d2d833 100644 --- a/src/gallium/auxiliary/gallivm/instructionssoa.cpp +++ b/src/gallium/auxiliary/gallivm/instructionssoa.cpp @@ -24,6 +24,8 @@ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. * **************************************************************************/ + +#include <cstdio> #include "instructionssoa.h" #include "storagesoa.h" @@ -126,7 +128,7 @@ void InstructionsSoa::createFunctionMap() m_functionsMap[TGSI_OPCODE_DP4] = "dp4"; m_functionsMap[TGSI_OPCODE_MIN] = "min"; m_functionsMap[TGSI_OPCODE_MAX] = "max"; - m_functionsMap[TGSI_OPCODE_POWER] = "pow"; + m_functionsMap[TGSI_OPCODE_POW] = "pow"; m_functionsMap[TGSI_OPCODE_LIT] = "lit"; m_functionsMap[TGSI_OPCODE_RSQ] = "rsq"; m_functionsMap[TGSI_OPCODE_SLT] = "slt"; @@ -309,7 +311,7 @@ std::vector<llvm::Value*> InstructionsSoa::mul(const std::vector<llvm::Value*> i std::vector<llvm::Value*> InstructionsSoa::pow(const std::vector<llvm::Value*> in1, const std::vector<llvm::Value*> in2) { - llvm::Function *func = function(TGSI_OPCODE_POWER); + llvm::Function *func = function(TGSI_OPCODE_POW); return callBuiltin(func, in1, in2); } diff --git a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp index 5b08200d142..bf84401e112 100644 --- a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp +++ b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp @@ -160,10 +160,11 @@ translate_immediate(Storage *storage, { float vec[4]; int i; + assert( imm->Immediate.NrTokens <= 4 + 1 ); for (i = 0; i < imm->Immediate.NrTokens - 1; ++i) { switch (imm->Immediate.DataType) { case TGSI_IMM_FLOAT32: - vec[i] = imm->u.ImmediateFloat32[i].Float; + vec[i] = imm->u[i].Float; break; default: assert(0); @@ -179,10 +180,11 @@ translate_immediateir(StorageSoa *storage, { float vec[4]; int i; + assert( imm->Immediate.NrTokens <= 4 + 1 ); for (i = 0; i < imm->Immediate.NrTokens - 1; ++i) { switch (imm->Immediate.DataType) { case TGSI_IMM_FLOAT32: - vec[i] = imm->u.ImmediateFloat32[i].Float; + vec[i] = imm->u[i].Float; break; default: assert(0); @@ -336,7 +338,7 @@ translate_instruction(llvm::Module *module, out = instr->sub(inputs[0], inputs[1]); } break; - case TGSI_OPCODE_LERP: { + case TGSI_OPCODE_LRP: { out = instr->lerp(inputs[0], inputs[1], inputs[2]); } break; @@ -348,17 +350,11 @@ translate_instruction(llvm::Module *module, out = instr->cnd0(inputs[0], inputs[1], inputs[2]); } break; - case TGSI_OPCODE_DOT2ADD: { + case TGSI_OPCODE_DP2A: { out = instr->dot2add(inputs[0], inputs[1], inputs[2]); } break; - case TGSI_OPCODE_INDEX: - break; - case TGSI_OPCODE_NEGATE: { - out = instr->neg(inputs[0]); - } - break; - case TGSI_OPCODE_FRAC: { + case TGSI_OPCODE_FRC: { out = instr->frc(inputs[0]); } break; @@ -366,30 +362,28 @@ translate_instruction(llvm::Module *module, out = instr->clamp(inputs[0]); } break; - case TGSI_OPCODE_FLOOR: { + case TGSI_OPCODE_FLR: { out = instr->floor(inputs[0]); } break; case TGSI_OPCODE_ROUND: break; - case TGSI_OPCODE_EXPBASE2: { + case TGSI_OPCODE_EX2: { out = instr->ex2(inputs[0]); } break; - case TGSI_OPCODE_LOGBASE2: { + case TGSI_OPCODE_LG2: { out = instr->lg2(inputs[0]); } break; - case TGSI_OPCODE_POWER: { + case TGSI_OPCODE_POW: { out = instr->pow(inputs[0], inputs[1]); } break; - case TGSI_OPCODE_CROSSPRODUCT: { + case TGSI_OPCODE_XPD: { out = instr->cross(inputs[0], inputs[1]); } break; - case TGSI_OPCODE_MULTIPLYMATRIX: - break; case TGSI_OPCODE_ABS: { out = instr->abs(inputs[0]); } @@ -522,7 +516,7 @@ translate_instruction(llvm::Module *module, return; //just update the state } break; - case TGSI_OPCODE_LOOP: + case TGSI_OPCODE_BGNFOR: break; case TGSI_OPCODE_REP: break; @@ -538,7 +532,7 @@ translate_instruction(llvm::Module *module, return; //just update the state } break; - case TGSI_OPCODE_ENDLOOP: + case TGSI_OPCODE_ENDFOR: break; case TGSI_OPCODE_ENDREP: break; @@ -580,7 +574,7 @@ translate_instruction(llvm::Module *module, break; case TGSI_OPCODE_ENDPRIM: break; - case TGSI_OPCODE_BGNLOOP2: { + case TGSI_OPCODE_BGNLOOP: { instr->beginLoop(); storage->setCurrentBlock(instr->currentBlock()); return; @@ -593,7 +587,7 @@ translate_instruction(llvm::Module *module, return; } break; - case TGSI_OPCODE_ENDLOOP2: { + case TGSI_OPCODE_ENDLOOP: { instr->endLoop(); storage->setCurrentBlock(instr->currentBlock()); return; @@ -617,14 +611,6 @@ translate_instruction(llvm::Module *module, break; case TGSI_OPCODE_NOP: break; - case TGSI_OPCODE_M4X3: - break; - case TGSI_OPCODE_M3X4: - break; - case TGSI_OPCODE_M3X3: - break; - case TGSI_OPCODE_M3X2: - break; case TGSI_OPCODE_CALLNZ: break; case TGSI_OPCODE_IFC: @@ -778,44 +764,38 @@ translate_instructionir(llvm::Module *module, out = instr->sub(inputs[0], inputs[1]); } break; - case TGSI_OPCODE_LERP: { + case TGSI_OPCODE_LRP: { } break; case TGSI_OPCODE_CND: break; case TGSI_OPCODE_CND0: break; - case TGSI_OPCODE_DOT2ADD: + case TGSI_OPCODE_DP2A: break; - case TGSI_OPCODE_INDEX: - break; - case TGSI_OPCODE_NEGATE: - break; - case TGSI_OPCODE_FRAC: { + case TGSI_OPCODE_FRC: { } break; case TGSI_OPCODE_CLAMP: break; - case TGSI_OPCODE_FLOOR: { + case TGSI_OPCODE_FLR: { } break; case TGSI_OPCODE_ROUND: break; - case TGSI_OPCODE_EXPBASE2: { + case TGSI_OPCODE_EX2: { } break; - case TGSI_OPCODE_LOGBASE2: { + case TGSI_OPCODE_LG2: { } break; - case TGSI_OPCODE_POWER: { + case TGSI_OPCODE_POW: { out = instr->pow(inputs[0], inputs[1]); } break; - case TGSI_OPCODE_CROSSPRODUCT: { + case TGSI_OPCODE_XPD: { } break; - case TGSI_OPCODE_MULTIPLYMATRIX: - break; case TGSI_OPCODE_ABS: { out = instr->abs(inputs[0]); } @@ -910,7 +890,7 @@ translate_instructionir(llvm::Module *module, case TGSI_OPCODE_IF: { } break; - case TGSI_OPCODE_LOOP: + case TGSI_OPCODE_BGNFOR: break; case TGSI_OPCODE_REP: break; @@ -920,7 +900,7 @@ translate_instructionir(llvm::Module *module, case TGSI_OPCODE_ENDIF: { } break; - case TGSI_OPCODE_ENDLOOP: + case TGSI_OPCODE_ENDFOR: break; case TGSI_OPCODE_ENDREP: break; @@ -961,13 +941,13 @@ translate_instructionir(llvm::Module *module, break; case TGSI_OPCODE_ENDPRIM: break; - case TGSI_OPCODE_BGNLOOP2: { + case TGSI_OPCODE_BGNLOOP: { } break; case TGSI_OPCODE_BGNSUB: { } break; - case TGSI_OPCODE_ENDLOOP2: { + case TGSI_OPCODE_ENDLOOP: { } break; case TGSI_OPCODE_ENDSUB: { @@ -983,14 +963,6 @@ translate_instructionir(llvm::Module *module, break; case TGSI_OPCODE_NOP: break; - case TGSI_OPCODE_M4X3: - break; - case TGSI_OPCODE_M3X4: - break; - case TGSI_OPCODE_M3X3: - break; - case TGSI_OPCODE_M3X2: - break; case TGSI_OPCODE_NRM4: break; case TGSI_OPCODE_CALLNZ: diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c index 0760d607161..6bdce5fcb06 100644 --- a/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c +++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c @@ -53,6 +53,8 @@ static INLINE struct malloc_buffer * malloc_buffer(struct pb_buffer *buf) { assert(buf); + if (!buf) + return NULL; assert(buf->vtbl == &malloc_buffer_vtbl); return (struct malloc_buffer *)buf; } diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_ondemand.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_ondemand.c index bc3093f620f..cb32d251367 100644 --- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_ondemand.c +++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_ondemand.c @@ -74,6 +74,8 @@ static INLINE struct pb_ondemand_buffer * pb_ondemand_buffer(struct pb_buffer *buf) { assert(buf); + if (!buf) + return NULL; assert(buf->vtbl == &pb_ondemand_buffer_vtbl); return (struct pb_ondemand_buffer *)buf; } diff --git a/src/gallium/auxiliary/rbug/Makefile b/src/gallium/auxiliary/rbug/Makefile new file mode 100644 index 00000000000..cd12e8468fc --- /dev/null +++ b/src/gallium/auxiliary/rbug/Makefile @@ -0,0 +1,14 @@ +TOP = ../../../.. +include $(TOP)/configs/current + +LIBNAME = rbug + +C_SOURCES = \ + rbug_connection.c \ + rbug_core.c \ + rbug_texture.c \ + rbug_context.c \ + rbug_shader.c \ + rbug_demarshal.c + +include ../../Makefile.template diff --git a/src/gallium/auxiliary/rbug/README b/src/gallium/auxiliary/rbug/README new file mode 100644 index 00000000000..33d76371de4 --- /dev/null +++ b/src/gallium/auxiliary/rbug/README @@ -0,0 +1,21 @@ + GALLIUM REMOTE DEBUGGING COMMON CODE + += About = + +This directory contains the common code for the Gallium 3D remote debugging +driver and clients. The code is two parts the connection managment code and +the (de)marsheller. + +The code currently uses tcp and ip4v for connections. + +Information about driver integration can be found in: + +src/gallium/drivers/trace/README + +for information about applications look in: + +progs/rbug/README + + +-- +Jakob Bornecrantz <jakob@vmware.com> diff --git a/src/gallium/auxiliary/rbug/SConscript b/src/gallium/auxiliary/rbug/SConscript new file mode 100644 index 00000000000..4a9afb45d3c --- /dev/null +++ b/src/gallium/auxiliary/rbug/SConscript @@ -0,0 +1,14 @@ +Import('*') + +rbug = env.ConvenienceLibrary( + target = 'rbug', + source = [ + 'rbug_core.c', + 'rbug_shader.c', + 'rbug_context.c', + 'rbug_texture.c', + 'rbug_demarshal.c', + 'rbug_connection.c', + ]) + +auxiliaries.insert(0, rbug) diff --git a/src/gallium/auxiliary/rbug/rbug.h b/src/gallium/auxiliary/rbug/rbug.h new file mode 100644 index 00000000000..259bfc6c79c --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug.h @@ -0,0 +1,33 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/* + * Include all for users the remote debugger protocol code. + */ + +#include "rbug/rbug_core.h" +#include "rbug/rbug_shader.h" +#include "rbug/rbug_context.h" +#include "rbug/rbug_texture.h" +#include "rbug/rbug_connection.h" diff --git a/src/gallium/auxiliary/rbug/rbug_connection.c b/src/gallium/auxiliary/rbug/rbug_connection.c new file mode 100644 index 00000000000..52acb700af9 --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_connection.c @@ -0,0 +1,167 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +#include "rbug/rbug.h" +#include "rbug/rbug_internal.h" + +#include "util/u_network.h" + +struct rbug_connection +{ + int socket; + uint32_t send_serial; + uint32_t recv_serial; + enum rbug_opcode opcode; +}; + +/** + * Create a rbug connection from a socket created with u_socket. + * + * Result: + * A new allocated connection using socket as communication path + */ +struct rbug_connection * +rbug_from_socket(int socket) +{ + struct rbug_connection *c = CALLOC_STRUCT(rbug_connection); + c->socket = socket; + return c; +} + +/** + * Free a connection, also closes socket. + */ +void +rbug_disconnect(struct rbug_connection *c) +{ + u_socket_close(c->socket); + FREE(c); +} + +/** + * Waits for a message to be fully received. + * Also returns the serial for the message, serial is not touched for replys. + * + * Result: + * demarshaled message on success, NULL on connection error + */ +struct rbug_header * +rbug_get_message(struct rbug_connection *c, uint32_t *serial) +{ + struct rbug_proto_header header; + struct rbug_header *out; + struct rbug_proto_header *data; + size_t length = 0; + size_t read = 0; + int ret; + + + ret = u_socket_peek(c->socket, &header, sizeof(header)); + if (ret <= 0) { + return NULL; + } + + length = (size_t)header.length * 4; + data = MALLOC(length); + if (!data) { + return NULL; + } + + do { + uint8_t *ptr = ((uint8_t*)data) + read; + ret = u_socket_recv(c->socket, ptr, length - read); + + if (ret <= 0) { + FREE(data); + return NULL; + } + + read += ret; + } while(read < length); + + out = rbug_demarshal(data); + if (!out) + FREE(data); + else if (serial) + *serial = c->recv_serial++; + else + c->recv_serial++; + + return out; +} + +/** + * Frees a message and associated data. + */ +void +rbug_free_header(struct rbug_header *header) +{ + if (!header) + return; + + FREE(header->__message); + FREE(header); +} + +/** + * Internal function used by rbug_send_* functions. + * + * Start sending a message. + */ +int +rbug_connection_send_start(struct rbug_connection *c, enum rbug_opcode opcode, uint32_t length) +{ + c->opcode = opcode; + return 0; +} + +/** + * Internal function used by rbug_send_* functions. + * + * Write data to the socket. + */ +int +rbug_connection_write(struct rbug_connection *c, void *to, uint32_t size) +{ + int ret = u_socket_send(c->socket, to, size); + return ret; +} + +/** + * Internal function used by rbug_send_* functions. + * + * Finish writeing data to the socket. + * Ups the send_serial and sets the serial argument if supplied. + */ +int rbug_connection_send_finish(struct rbug_connection *c, uint32_t *serial) +{ + if (c->opcode < 0) + return 0; + else if (serial) + *serial = c->send_serial++; + else + c->send_serial++; + + return 0; +} diff --git a/src/gallium/auxiliary/rbug/rbug_connection.h b/src/gallium/auxiliary/rbug/rbug_connection.h new file mode 100644 index 00000000000..1f2c9ff347c --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_connection.h @@ -0,0 +1,45 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/* + * This file contains the function defentions for connection see c file for + * more comments covering function use. + */ + +#ifndef _RBUG_CONNECTION_H_ +#define _RBUG_CONNECTION_H_ + +#include "rbug/rbug_proto.h" + +struct rbug_connection * rbug_from_socket(int socket); + +void rbug_disconnect(struct rbug_connection *c); + +struct rbug_header * rbug_get_message(struct rbug_connection *c, uint32_t *serial); + +void rbug_free_header(struct rbug_header *header); + +struct rbug_header * rbug_demarshal(struct rbug_proto_header *header); + +#endif diff --git a/src/gallium/auxiliary/rbug/rbug_context.c b/src/gallium/auxiliary/rbug/rbug_context.c new file mode 100644 index 00000000000..1832425658f --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_context.c @@ -0,0 +1,759 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/* + * This file holds the function implementation for one of the rbug extensions. + * Prototypes and declerations of functions and structs is in the same folder + * in the header file matching this file's name. + * + * The functions starting rbug_send_* encodes a call to the write format and + * sends that to the supplied connection, while functions starting with + * rbug_demarshal_* demarshal data in the wire protocol. + * + * Functions ending with _reply are replies to requests. + */ + +#include "rbug_internal.h" +#include "rbug/rbug_context.h" + +int rbug_send_context_list(struct rbug_connection *__con, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_CONTEXT_LIST)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_CONTEXT_LIST, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_context_info(struct rbug_connection *__con, + rbug_context_t context, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* context */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_CONTEXT_INFO)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_context_t, context); /* context */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_CONTEXT_INFO, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_context_draw_block(struct rbug_connection *__con, + rbug_context_t context, + rbug_block_t block, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* context */ + LEN(4); /* block */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_CONTEXT_DRAW_BLOCK)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_context_t, context); /* context */ + WRITE(4, rbug_block_t, block); /* block */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_CONTEXT_DRAW_BLOCK, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_context_draw_step(struct rbug_connection *__con, + rbug_context_t context, + rbug_block_t step, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* context */ + LEN(4); /* step */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_CONTEXT_DRAW_STEP)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_context_t, context); /* context */ + WRITE(4, rbug_block_t, step); /* step */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_CONTEXT_DRAW_STEP, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_context_draw_unblock(struct rbug_connection *__con, + rbug_context_t context, + rbug_block_t unblock, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* context */ + LEN(4); /* unblock */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_CONTEXT_DRAW_UNBLOCK)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_context_t, context); /* context */ + WRITE(4, rbug_block_t, unblock); /* unblock */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_CONTEXT_DRAW_UNBLOCK, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_context_draw_rule(struct rbug_connection *__con, + rbug_context_t context, + rbug_shader_t vertex, + rbug_shader_t fragment, + rbug_texture_t texture, + rbug_texture_t surface, + rbug_block_t block, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* context */ + LEN(8); /* vertex */ + LEN(8); /* fragment */ + LEN(8); /* texture */ + LEN(8); /* surface */ + LEN(4); /* block */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_CONTEXT_DRAW_RULE)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_context_t, context); /* context */ + WRITE(8, rbug_shader_t, vertex); /* vertex */ + WRITE(8, rbug_shader_t, fragment); /* fragment */ + WRITE(8, rbug_texture_t, texture); /* texture */ + WRITE(8, rbug_texture_t, surface); /* surface */ + WRITE(4, rbug_block_t, block); /* block */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_CONTEXT_DRAW_RULE, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_context_flush(struct rbug_connection *__con, + rbug_context_t context, + int32_t flags, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* context */ + LEN(4); /* flags */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_CONTEXT_FLUSH)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_context_t, context); /* context */ + WRITE(4, int32_t, flags); /* flags */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_CONTEXT_FLUSH, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_context_list_reply(struct rbug_connection *__con, + uint32_t serial, + rbug_context_t *contexts, + uint32_t contexts_len, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(4); /* serial */ + LEN_ARRAY(8, contexts); /* contexts */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_CONTEXT_LIST_REPLY)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(4, uint32_t, serial); /* serial */ + WRITE_ARRAY(8, rbug_context_t, contexts); /* contexts */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_CONTEXT_LIST_REPLY, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_context_info_reply(struct rbug_connection *__con, + uint32_t serial, + rbug_shader_t vertex, + rbug_shader_t fragment, + rbug_texture_t *texs, + uint32_t texs_len, + rbug_texture_t *cbufs, + uint32_t cbufs_len, + rbug_texture_t zsbuf, + rbug_block_t blocker, + rbug_block_t blocked, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(4); /* serial */ + LEN(8); /* vertex */ + LEN(8); /* fragment */ + LEN_ARRAY(8, texs); /* texs */ + LEN_ARRAY(8, cbufs); /* cbufs */ + LEN(8); /* zsbuf */ + LEN(4); /* blocker */ + LEN(4); /* blocked */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_CONTEXT_INFO_REPLY)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(4, uint32_t, serial); /* serial */ + WRITE(8, rbug_shader_t, vertex); /* vertex */ + WRITE(8, rbug_shader_t, fragment); /* fragment */ + WRITE_ARRAY(8, rbug_texture_t, texs); /* texs */ + WRITE_ARRAY(8, rbug_texture_t, cbufs); /* cbufs */ + WRITE(8, rbug_texture_t, zsbuf); /* zsbuf */ + WRITE(4, rbug_block_t, blocker); /* blocker */ + WRITE(4, rbug_block_t, blocked); /* blocked */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_CONTEXT_INFO_REPLY, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_context_draw_blocked(struct rbug_connection *__con, + rbug_context_t context, + rbug_block_t block, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* context */ + LEN(4); /* block */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_CONTEXT_DRAW_BLOCKED)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_context_t, context); /* context */ + WRITE(4, rbug_block_t, block); /* block */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_CONTEXT_DRAW_BLOCKED, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +struct rbug_proto_context_list * rbug_demarshal_context_list(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_context_list *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_CONTEXT_LIST) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + + return ret; +} + +struct rbug_proto_context_info * rbug_demarshal_context_info(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_context_info *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_CONTEXT_INFO) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_context_t, context); /* context */ + + return ret; +} + +struct rbug_proto_context_draw_block * rbug_demarshal_context_draw_block(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_context_draw_block *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_CONTEXT_DRAW_BLOCK) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_context_t, context); /* context */ + READ(4, rbug_block_t, block); /* block */ + + return ret; +} + +struct rbug_proto_context_draw_step * rbug_demarshal_context_draw_step(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_context_draw_step *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_CONTEXT_DRAW_STEP) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_context_t, context); /* context */ + READ(4, rbug_block_t, step); /* step */ + + return ret; +} + +struct rbug_proto_context_draw_unblock * rbug_demarshal_context_draw_unblock(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_context_draw_unblock *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_CONTEXT_DRAW_UNBLOCK) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_context_t, context); /* context */ + READ(4, rbug_block_t, unblock); /* unblock */ + + return ret; +} + +struct rbug_proto_context_draw_rule * rbug_demarshal_context_draw_rule(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_context_draw_rule *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_CONTEXT_DRAW_RULE) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_context_t, context); /* context */ + READ(8, rbug_shader_t, vertex); /* vertex */ + READ(8, rbug_shader_t, fragment); /* fragment */ + READ(8, rbug_texture_t, texture); /* texture */ + READ(8, rbug_texture_t, surface); /* surface */ + READ(4, rbug_block_t, block); /* block */ + + return ret; +} + +struct rbug_proto_context_flush * rbug_demarshal_context_flush(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_context_flush *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_CONTEXT_FLUSH) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_context_t, context); /* context */ + READ(4, int32_t, flags); /* flags */ + + return ret; +} + +struct rbug_proto_context_list_reply * rbug_demarshal_context_list_reply(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_context_list_reply *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_CONTEXT_LIST_REPLY) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(4, uint32_t, serial); /* serial */ + READ_ARRAY(8, rbug_context_t, contexts); /* contexts */ + + return ret; +} + +struct rbug_proto_context_info_reply * rbug_demarshal_context_info_reply(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_context_info_reply *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_CONTEXT_INFO_REPLY) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(4, uint32_t, serial); /* serial */ + READ(8, rbug_shader_t, vertex); /* vertex */ + READ(8, rbug_shader_t, fragment); /* fragment */ + READ_ARRAY(8, rbug_texture_t, texs); /* texs */ + READ_ARRAY(8, rbug_texture_t, cbufs); /* cbufs */ + READ(8, rbug_texture_t, zsbuf); /* zsbuf */ + READ(4, rbug_block_t, blocker); /* blocker */ + READ(4, rbug_block_t, blocked); /* blocked */ + + return ret; +} + +struct rbug_proto_context_draw_blocked * rbug_demarshal_context_draw_blocked(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_context_draw_blocked *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_CONTEXT_DRAW_BLOCKED) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_context_t, context); /* context */ + READ(4, rbug_block_t, block); /* block */ + + return ret; +} diff --git a/src/gallium/auxiliary/rbug/rbug_context.h b/src/gallium/auxiliary/rbug/rbug_context.h new file mode 100644 index 00000000000..da61c2365b0 --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_context.h @@ -0,0 +1,212 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/* + * This file holds structs decelerations and function prototypes for one of + * the rbug extensions. Implementation of the functions is in the same folder + * in the c file matching this file's name. + * + * The structs what is returned from the demarshal functions. The functions + * starting rbug_send_* encodes a call to the write format and sends that to + * the supplied connection, while functions starting with rbug_demarshal_* + * demarshal data from the wire protocol. + * + * Structs and functions ending with _reply are replies to requests. + */ + +#ifndef _RBUG_PROTO_CONTEXT_H_ +#define _RBUG_PROTO_CONTEXT_H_ + +#include "rbug/rbug_proto.h" +#include "rbug/rbug_core.h" + +typedef enum +{ + RBUG_BLOCK_BEFORE = 1, + RBUG_BLOCK_AFTER = 2, + RBUG_BLOCK_RULE = 4, + RBUG_BLOCK_MASK = 7, +} rbug_block_t; + +struct rbug_proto_context_list +{ + struct rbug_header header; +}; + +struct rbug_proto_context_info +{ + struct rbug_header header; + rbug_context_t context; +}; + +struct rbug_proto_context_draw_block +{ + struct rbug_header header; + rbug_context_t context; + rbug_block_t block; +}; + +struct rbug_proto_context_draw_step +{ + struct rbug_header header; + rbug_context_t context; + rbug_block_t step; +}; + +struct rbug_proto_context_draw_unblock +{ + struct rbug_header header; + rbug_context_t context; + rbug_block_t unblock; +}; + +struct rbug_proto_context_draw_rule +{ + struct rbug_header header; + rbug_context_t context; + rbug_shader_t vertex; + rbug_shader_t fragment; + rbug_texture_t texture; + rbug_texture_t surface; + rbug_block_t block; +}; + +struct rbug_proto_context_flush +{ + struct rbug_header header; + rbug_context_t context; + int32_t flags; +}; + +struct rbug_proto_context_list_reply +{ + struct rbug_header header; + uint32_t serial; + rbug_context_t *contexts; + uint32_t contexts_len; +}; + +struct rbug_proto_context_info_reply +{ + struct rbug_header header; + uint32_t serial; + rbug_shader_t vertex; + rbug_shader_t fragment; + rbug_texture_t *texs; + uint32_t texs_len; + rbug_texture_t *cbufs; + uint32_t cbufs_len; + rbug_texture_t zsbuf; + rbug_block_t blocker; + rbug_block_t blocked; +}; + +struct rbug_proto_context_draw_blocked +{ + struct rbug_header header; + rbug_context_t context; + rbug_block_t block; +}; + +int rbug_send_context_list(struct rbug_connection *__con, + uint32_t *__serial); + +int rbug_send_context_info(struct rbug_connection *__con, + rbug_context_t context, + uint32_t *__serial); + +int rbug_send_context_draw_block(struct rbug_connection *__con, + rbug_context_t context, + rbug_block_t block, + uint32_t *__serial); + +int rbug_send_context_draw_step(struct rbug_connection *__con, + rbug_context_t context, + rbug_block_t step, + uint32_t *__serial); + +int rbug_send_context_draw_unblock(struct rbug_connection *__con, + rbug_context_t context, + rbug_block_t unblock, + uint32_t *__serial); + +int rbug_send_context_draw_rule(struct rbug_connection *__con, + rbug_context_t context, + rbug_shader_t vertex, + rbug_shader_t fragment, + rbug_texture_t texture, + rbug_texture_t surface, + rbug_block_t block, + uint32_t *__serial); + +int rbug_send_context_flush(struct rbug_connection *__con, + rbug_context_t context, + int32_t flags, + uint32_t *__serial); + +int rbug_send_context_list_reply(struct rbug_connection *__con, + uint32_t serial, + rbug_context_t *contexts, + uint32_t contexts_len, + uint32_t *__serial); + +int rbug_send_context_info_reply(struct rbug_connection *__con, + uint32_t serial, + rbug_shader_t vertex, + rbug_shader_t fragment, + rbug_texture_t *texs, + uint32_t texs_len, + rbug_texture_t *cbufs, + uint32_t cbufs_len, + rbug_texture_t zsbuf, + rbug_block_t blocker, + rbug_block_t blocked, + uint32_t *__serial); + +int rbug_send_context_draw_blocked(struct rbug_connection *__con, + rbug_context_t context, + rbug_block_t block, + uint32_t *__serial); + +struct rbug_proto_context_list * rbug_demarshal_context_list(struct rbug_proto_header *header); + +struct rbug_proto_context_info * rbug_demarshal_context_info(struct rbug_proto_header *header); + +struct rbug_proto_context_draw_block * rbug_demarshal_context_draw_block(struct rbug_proto_header *header); + +struct rbug_proto_context_draw_step * rbug_demarshal_context_draw_step(struct rbug_proto_header *header); + +struct rbug_proto_context_draw_unblock * rbug_demarshal_context_draw_unblock(struct rbug_proto_header *header); + +struct rbug_proto_context_draw_rule * rbug_demarshal_context_draw_rule(struct rbug_proto_header *header); + +struct rbug_proto_context_flush * rbug_demarshal_context_flush(struct rbug_proto_header *header); + +struct rbug_proto_context_list_reply * rbug_demarshal_context_list_reply(struct rbug_proto_header *header); + +struct rbug_proto_context_info_reply * rbug_demarshal_context_info_reply(struct rbug_proto_header *header); + +struct rbug_proto_context_draw_blocked * rbug_demarshal_context_draw_blocked(struct rbug_proto_header *header); + +#endif diff --git a/src/gallium/auxiliary/rbug/rbug_core.c b/src/gallium/auxiliary/rbug/rbug_core.c new file mode 100644 index 00000000000..876ae5a0ce6 --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_core.c @@ -0,0 +1,359 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/* + * This file holds the function implementation for one of the rbug extensions. + * Prototypes and declerations of functions and structs is in the same folder + * in the header file matching this file's name. + * + * The functions starting rbug_send_* encodes a call to the write format and + * sends that to the supplied connection, while functions starting with + * rbug_demarshal_* demarshal data in the wire protocol. + * + * Functions ending with _reply are replies to requests. + */ + +#include "rbug_internal.h" +#include "rbug/rbug_core.h" + +int rbug_send_noop(struct rbug_connection *__con, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_NOOP)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_NOOP, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_ping(struct rbug_connection *__con, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_PING)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_PING, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_error(struct rbug_connection *__con, + uint32_t error, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(4); /* error */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_ERROR)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(4, uint32_t, error); /* error */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_ERROR, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_ping_reply(struct rbug_connection *__con, + uint32_t serial, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(4); /* serial */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_PING_REPLY)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(4, uint32_t, serial); /* serial */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_PING_REPLY, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_error_reply(struct rbug_connection *__con, + uint32_t serial, + uint32_t error, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(4); /* serial */ + LEN(4); /* error */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_ERROR_REPLY)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(4, uint32_t, serial); /* serial */ + WRITE(4, uint32_t, error); /* error */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_ERROR_REPLY, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +struct rbug_proto_noop * rbug_demarshal_noop(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_noop *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_NOOP) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + + return ret; +} + +struct rbug_proto_ping * rbug_demarshal_ping(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_ping *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_PING) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + + return ret; +} + +struct rbug_proto_error * rbug_demarshal_error(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_error *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_ERROR) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(4, uint32_t, error); /* error */ + + return ret; +} + +struct rbug_proto_ping_reply * rbug_demarshal_ping_reply(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_ping_reply *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_PING_REPLY) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(4, uint32_t, serial); /* serial */ + + return ret; +} + +struct rbug_proto_error_reply * rbug_demarshal_error_reply(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_error_reply *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_ERROR_REPLY) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(4, uint32_t, serial); /* serial */ + READ(4, uint32_t, error); /* error */ + + return ret; +} diff --git a/src/gallium/auxiliary/rbug/rbug_core.h b/src/gallium/auxiliary/rbug/rbug_core.h new file mode 100644 index 00000000000..99a36a0163e --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_core.h @@ -0,0 +1,105 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/* + * This file holds structs decelerations and function prototypes for one of + * the rbug extensions. Implementation of the functions is in the same folder + * in the c file matching this file's name. + * + * The structs what is returned from the demarshal functions. The functions + * starting rbug_send_* encodes a call to the write format and sends that to + * the supplied connection, while functions starting with rbug_demarshal_* + * demarshal data from the wire protocol. + * + * Structs and functions ending with _reply are replies to requests. + */ + +#ifndef _RBUG_PROTO_CORE_H_ +#define _RBUG_PROTO_CORE_H_ + +#include "rbug/rbug_proto.h" + +typedef uint64_t rbug_shader_t; +typedef uint64_t rbug_context_t; +typedef uint64_t rbug_texture_t; + +struct rbug_proto_noop +{ + struct rbug_header header; +}; + +struct rbug_proto_ping +{ + struct rbug_header header; +}; + +struct rbug_proto_error +{ + struct rbug_header header; + uint32_t error; +}; + +struct rbug_proto_ping_reply +{ + struct rbug_header header; + uint32_t serial; +}; + +struct rbug_proto_error_reply +{ + struct rbug_header header; + uint32_t serial; + uint32_t error; +}; + +int rbug_send_noop(struct rbug_connection *__con, + uint32_t *__serial); + +int rbug_send_ping(struct rbug_connection *__con, + uint32_t *__serial); + +int rbug_send_error(struct rbug_connection *__con, + uint32_t error, + uint32_t *__serial); + +int rbug_send_ping_reply(struct rbug_connection *__con, + uint32_t serial, + uint32_t *__serial); + +int rbug_send_error_reply(struct rbug_connection *__con, + uint32_t serial, + uint32_t error, + uint32_t *__serial); + +struct rbug_proto_noop * rbug_demarshal_noop(struct rbug_proto_header *header); + +struct rbug_proto_ping * rbug_demarshal_ping(struct rbug_proto_header *header); + +struct rbug_proto_error * rbug_demarshal_error(struct rbug_proto_header *header); + +struct rbug_proto_ping_reply * rbug_demarshal_ping_reply(struct rbug_proto_header *header); + +struct rbug_proto_error_reply * rbug_demarshal_error_reply(struct rbug_proto_header *header); + +#endif diff --git a/src/gallium/auxiliary/rbug/rbug_demarshal.c b/src/gallium/auxiliary/rbug/rbug_demarshal.c new file mode 100644 index 00000000000..47390fbcee7 --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_demarshal.c @@ -0,0 +1,93 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +#include "rbug.h" + +/** + * Small function that looks at the proto_header and selects the correct + * demarshal functions and return the result. + */ +struct rbug_header * rbug_demarshal(struct rbug_proto_header *header) +{ + switch(header->opcode) { + case RBUG_OP_NOOP: + return (struct rbug_header *)rbug_demarshal_noop(header); + case RBUG_OP_PING: + return (struct rbug_header *)rbug_demarshal_ping(header); + case RBUG_OP_ERROR: + return (struct rbug_header *)rbug_demarshal_error(header); + case RBUG_OP_PING_REPLY: + return (struct rbug_header *)rbug_demarshal_ping_reply(header); + case RBUG_OP_ERROR_REPLY: + return (struct rbug_header *)rbug_demarshal_error_reply(header); + case RBUG_OP_TEXTURE_LIST: + return (struct rbug_header *)rbug_demarshal_texture_list(header); + case RBUG_OP_TEXTURE_INFO: + return (struct rbug_header *)rbug_demarshal_texture_info(header); + case RBUG_OP_TEXTURE_WRITE: + return (struct rbug_header *)rbug_demarshal_texture_write(header); + case RBUG_OP_TEXTURE_READ: + return (struct rbug_header *)rbug_demarshal_texture_read(header); + case RBUG_OP_TEXTURE_LIST_REPLY: + return (struct rbug_header *)rbug_demarshal_texture_list_reply(header); + case RBUG_OP_TEXTURE_INFO_REPLY: + return (struct rbug_header *)rbug_demarshal_texture_info_reply(header); + case RBUG_OP_TEXTURE_READ_REPLY: + return (struct rbug_header *)rbug_demarshal_texture_read_reply(header); + case RBUG_OP_CONTEXT_LIST: + return (struct rbug_header *)rbug_demarshal_context_list(header); + case RBUG_OP_CONTEXT_INFO: + return (struct rbug_header *)rbug_demarshal_context_info(header); + case RBUG_OP_CONTEXT_DRAW_BLOCK: + return (struct rbug_header *)rbug_demarshal_context_draw_block(header); + case RBUG_OP_CONTEXT_DRAW_STEP: + return (struct rbug_header *)rbug_demarshal_context_draw_step(header); + case RBUG_OP_CONTEXT_DRAW_UNBLOCK: + return (struct rbug_header *)rbug_demarshal_context_draw_unblock(header); + case RBUG_OP_CONTEXT_DRAW_RULE: + return (struct rbug_header *)rbug_demarshal_context_draw_rule(header); + case RBUG_OP_CONTEXT_FLUSH: + return (struct rbug_header *)rbug_demarshal_context_flush(header); + case RBUG_OP_CONTEXT_LIST_REPLY: + return (struct rbug_header *)rbug_demarshal_context_list_reply(header); + case RBUG_OP_CONTEXT_INFO_REPLY: + return (struct rbug_header *)rbug_demarshal_context_info_reply(header); + case RBUG_OP_CONTEXT_DRAW_BLOCKED: + return (struct rbug_header *)rbug_demarshal_context_draw_blocked(header); + case RBUG_OP_SHADER_LIST: + return (struct rbug_header *)rbug_demarshal_shader_list(header); + case RBUG_OP_SHADER_INFO: + return (struct rbug_header *)rbug_demarshal_shader_info(header); + case RBUG_OP_SHADER_DISABLE: + return (struct rbug_header *)rbug_demarshal_shader_disable(header); + case RBUG_OP_SHADER_REPLACE: + return (struct rbug_header *)rbug_demarshal_shader_replace(header); + case RBUG_OP_SHADER_LIST_REPLY: + return (struct rbug_header *)rbug_demarshal_shader_list_reply(header); + case RBUG_OP_SHADER_INFO_REPLY: + return (struct rbug_header *)rbug_demarshal_shader_info_reply(header); + default: + return NULL; + } +} diff --git a/src/gallium/auxiliary/rbug/rbug_internal.h b/src/gallium/auxiliary/rbug/rbug_internal.h new file mode 100644 index 00000000000..4aba1a810f6 --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_internal.h @@ -0,0 +1,100 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/* + * This file is internal to the rbug protocol code, and contains asorted + * features needed by the code. + */ + +#ifndef _RBUG_INTERNAL_H_ +#define _RBUG_INTERNAL_H_ + +#include "rbug/rbug_proto.h" + +#include "util/u_memory.h" +#include "util/u_debug.h" +#include <errno.h> + +int rbug_connection_send_start(struct rbug_connection *con, enum rbug_opcode opcode, uint32_t length); +int rbug_connection_write(struct rbug_connection *con, void *data, uint32_t size); +int rbug_connection_send_finish(struct rbug_connection *con, uint32_t *c); + +/** + * Only works with multiples of 2 + */ +#define PAD(from, to) \ +do { \ + from = (from + to - 1) & ~(to - 1); \ +} while(0) + +#define LEN(size) \ +do { \ + PAD(__len, size); \ + __len += size; \ +} while(0) + +#define LEN_ARRAY(size, name) \ +do { \ + LEN(4); \ + PAD(__len, size); \ + __len += size * name##_len; \ +} while(0) + +#define WRITE(size, type, name) \ +do { \ + PAD(__pos, size); \ + *((type *)(&__data[__pos])) = name; \ + __pos += size; \ +} while(0) + +#define WRITE_ARRAY(size, type, name) \ +do { \ + WRITE(4, uint32_t, name##_len); \ + PAD(__pos, size); \ + memcpy(&__data[__pos], name, size * name##_len); \ + __pos += size * name##_len; \ +} while(0) + +#define READ(size, type, name) \ +do { \ + PAD(pos, size); \ + pos += size; \ + if (pos > len) \ + break; \ + ret->name = *((type *)(&data[pos - size])); \ +} while(0) + +#define READ_ARRAY(size, type, name) \ +do { \ + READ(4, uint32_t, name##_len); \ + if (pos > len) \ + break; \ + PAD(pos, size); \ + pos += size * ret->name##_len; \ + if (pos > len) \ + break; \ + ret->name = (type *)&data[pos - size * ret->name##_len]; \ +} while(0) + +#endif diff --git a/src/gallium/auxiliary/rbug/rbug_proto.h b/src/gallium/auxiliary/rbug/rbug_proto.h new file mode 100644 index 00000000000..d273be0166d --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_proto.h @@ -0,0 +1,94 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/* + * This file holds common definitions of the gallium remote debugging protocol. + */ + +#ifndef _RBUG_PROTO_H_ +#define _RBUG_PROTO_H_ + +#include "pipe/p_compiler.h" + +/** + * Uniqe indentifier for each command. + * + * Replys are designated by negative. + */ +enum rbug_opcode +{ + RBUG_OP_NOOP = 0, + RBUG_OP_PING = 1, + RBUG_OP_ERROR = 2, + RBUG_OP_PING_REPLY = -1, + RBUG_OP_ERROR_REPLY = -2, + RBUG_OP_TEXTURE_LIST = 256, + RBUG_OP_TEXTURE_INFO = 257, + RBUG_OP_TEXTURE_WRITE = 258, + RBUG_OP_TEXTURE_READ = 259, + RBUG_OP_TEXTURE_LIST_REPLY = -256, + RBUG_OP_TEXTURE_INFO_REPLY = -257, + RBUG_OP_TEXTURE_READ_REPLY = -259, + RBUG_OP_CONTEXT_LIST = 512, + RBUG_OP_CONTEXT_INFO = 513, + RBUG_OP_CONTEXT_DRAW_BLOCK = 514, + RBUG_OP_CONTEXT_DRAW_STEP = 515, + RBUG_OP_CONTEXT_DRAW_UNBLOCK = 516, + RBUG_OP_CONTEXT_DRAW_RULE = 518, + RBUG_OP_CONTEXT_FLUSH = 519, + RBUG_OP_CONTEXT_LIST_REPLY = -512, + RBUG_OP_CONTEXT_INFO_REPLY = -513, + RBUG_OP_CONTEXT_DRAW_BLOCKED = 517, + RBUG_OP_SHADER_LIST = 768, + RBUG_OP_SHADER_INFO = 769, + RBUG_OP_SHADER_DISABLE = 770, + RBUG_OP_SHADER_REPLACE = 771, + RBUG_OP_SHADER_LIST_REPLY = -768, + RBUG_OP_SHADER_INFO_REPLY = -769, +}; + +/** + * Header for demarshaled message. + */ +struct rbug_header +{ + enum rbug_opcode opcode; + void *__message; +}; + +/** + * Header for a message in wire format. + */ +struct rbug_proto_header +{ + int32_t opcode; + uint32_t length; +}; + +/** + * Forward declare connection here, as this file is included by all users. + */ +struct rbug_connection; + +#endif diff --git a/src/gallium/auxiliary/rbug/rbug_shader.c b/src/gallium/auxiliary/rbug/rbug_shader.c new file mode 100644 index 00000000000..fccd2f55efd --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_shader.c @@ -0,0 +1,468 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/* + * This file holds the function implementation for one of the rbug extensions. + * Prototypes and declerations of functions and structs is in the same folder + * in the header file matching this file's name. + * + * The functions starting rbug_send_* encodes a call to the write format and + * sends that to the supplied connection, while functions starting with + * rbug_demarshal_* demarshal data in the wire protocol. + * + * Functions ending with _reply are replies to requests. + */ + +#include "rbug_internal.h" +#include "rbug/rbug_shader.h" + +int rbug_send_shader_list(struct rbug_connection *__con, + rbug_context_t context, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* context */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_SHADER_LIST)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_context_t, context); /* context */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_SHADER_LIST, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_shader_info(struct rbug_connection *__con, + rbug_context_t context, + rbug_shader_t shader, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* context */ + LEN(8); /* shader */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_SHADER_INFO)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_context_t, context); /* context */ + WRITE(8, rbug_shader_t, shader); /* shader */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_SHADER_INFO, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_shader_disable(struct rbug_connection *__con, + rbug_context_t context, + rbug_shader_t shader, + uint8_t disable, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* context */ + LEN(8); /* shader */ + LEN(1); /* disable */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_SHADER_DISABLE)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_context_t, context); /* context */ + WRITE(8, rbug_shader_t, shader); /* shader */ + WRITE(1, uint8_t, disable); /* disable */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_SHADER_DISABLE, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_shader_replace(struct rbug_connection *__con, + rbug_context_t context, + rbug_shader_t shader, + uint32_t *tokens, + uint32_t tokens_len, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* context */ + LEN(8); /* shader */ + LEN_ARRAY(4, tokens); /* tokens */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_SHADER_REPLACE)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_context_t, context); /* context */ + WRITE(8, rbug_shader_t, shader); /* shader */ + WRITE_ARRAY(4, uint32_t, tokens); /* tokens */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_SHADER_REPLACE, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_shader_list_reply(struct rbug_connection *__con, + uint32_t serial, + rbug_shader_t *shaders, + uint32_t shaders_len, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(4); /* serial */ + LEN_ARRAY(8, shaders); /* shaders */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_SHADER_LIST_REPLY)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(4, uint32_t, serial); /* serial */ + WRITE_ARRAY(8, rbug_shader_t, shaders); /* shaders */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_SHADER_LIST_REPLY, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_shader_info_reply(struct rbug_connection *__con, + uint32_t serial, + uint32_t *original, + uint32_t original_len, + uint32_t *replaced, + uint32_t replaced_len, + uint8_t disabled, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(4); /* serial */ + LEN_ARRAY(4, original); /* original */ + LEN_ARRAY(4, replaced); /* replaced */ + LEN(1); /* disabled */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_SHADER_INFO_REPLY)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(4, uint32_t, serial); /* serial */ + WRITE_ARRAY(4, uint32_t, original); /* original */ + WRITE_ARRAY(4, uint32_t, replaced); /* replaced */ + WRITE(1, uint8_t, disabled); /* disabled */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_SHADER_INFO_REPLY, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +struct rbug_proto_shader_list * rbug_demarshal_shader_list(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_shader_list *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_SHADER_LIST) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_context_t, context); /* context */ + + return ret; +} + +struct rbug_proto_shader_info * rbug_demarshal_shader_info(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_shader_info *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_SHADER_INFO) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_context_t, context); /* context */ + READ(8, rbug_shader_t, shader); /* shader */ + + return ret; +} + +struct rbug_proto_shader_disable * rbug_demarshal_shader_disable(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_shader_disable *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_SHADER_DISABLE) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_context_t, context); /* context */ + READ(8, rbug_shader_t, shader); /* shader */ + READ(1, uint8_t, disable); /* disable */ + + return ret; +} + +struct rbug_proto_shader_replace * rbug_demarshal_shader_replace(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_shader_replace *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_SHADER_REPLACE) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_context_t, context); /* context */ + READ(8, rbug_shader_t, shader); /* shader */ + READ_ARRAY(4, uint32_t, tokens); /* tokens */ + + return ret; +} + +struct rbug_proto_shader_list_reply * rbug_demarshal_shader_list_reply(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_shader_list_reply *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_SHADER_LIST_REPLY) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(4, uint32_t, serial); /* serial */ + READ_ARRAY(8, rbug_shader_t, shaders); /* shaders */ + + return ret; +} + +struct rbug_proto_shader_info_reply * rbug_demarshal_shader_info_reply(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_shader_info_reply *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_SHADER_INFO_REPLY) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(4, uint32_t, serial); /* serial */ + READ_ARRAY(4, uint32_t, original); /* original */ + READ_ARRAY(4, uint32_t, replaced); /* replaced */ + READ(1, uint8_t, disabled); /* disabled */ + + return ret; +} diff --git a/src/gallium/auxiliary/rbug/rbug_shader.h b/src/gallium/auxiliary/rbug/rbug_shader.h new file mode 100644 index 00000000000..b5d886781d8 --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_shader.h @@ -0,0 +1,142 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/* + * This file holds structs decelerations and function prototypes for one of + * the rbug extensions. Implementation of the functions is in the same folder + * in the c file matching this file's name. + * + * The structs what is returned from the demarshal functions. The functions + * starting rbug_send_* encodes a call to the write format and sends that to + * the supplied connection, while functions starting with rbug_demarshal_* + * demarshal data from the wire protocol. + * + * Structs and functions ending with _reply are replies to requests. + */ + +#ifndef _RBUG_PROTO_SHADER_H_ +#define _RBUG_PROTO_SHADER_H_ + +#include "rbug/rbug_proto.h" +#include "rbug/rbug_core.h" + +struct rbug_proto_shader_list +{ + struct rbug_header header; + rbug_context_t context; +}; + +struct rbug_proto_shader_info +{ + struct rbug_header header; + rbug_context_t context; + rbug_shader_t shader; +}; + +struct rbug_proto_shader_disable +{ + struct rbug_header header; + rbug_context_t context; + rbug_shader_t shader; + uint8_t disable; +}; + +struct rbug_proto_shader_replace +{ + struct rbug_header header; + rbug_context_t context; + rbug_shader_t shader; + uint32_t *tokens; + uint32_t tokens_len; +}; + +struct rbug_proto_shader_list_reply +{ + struct rbug_header header; + uint32_t serial; + rbug_shader_t *shaders; + uint32_t shaders_len; +}; + +struct rbug_proto_shader_info_reply +{ + struct rbug_header header; + uint32_t serial; + uint32_t *original; + uint32_t original_len; + uint32_t *replaced; + uint32_t replaced_len; + uint8_t disabled; +}; + +int rbug_send_shader_list(struct rbug_connection *__con, + rbug_context_t context, + uint32_t *__serial); + +int rbug_send_shader_info(struct rbug_connection *__con, + rbug_context_t context, + rbug_shader_t shader, + uint32_t *__serial); + +int rbug_send_shader_disable(struct rbug_connection *__con, + rbug_context_t context, + rbug_shader_t shader, + uint8_t disable, + uint32_t *__serial); + +int rbug_send_shader_replace(struct rbug_connection *__con, + rbug_context_t context, + rbug_shader_t shader, + uint32_t *tokens, + uint32_t tokens_len, + uint32_t *__serial); + +int rbug_send_shader_list_reply(struct rbug_connection *__con, + uint32_t serial, + rbug_shader_t *shaders, + uint32_t shaders_len, + uint32_t *__serial); + +int rbug_send_shader_info_reply(struct rbug_connection *__con, + uint32_t serial, + uint32_t *original, + uint32_t original_len, + uint32_t *replaced, + uint32_t replaced_len, + uint8_t disabled, + uint32_t *__serial); + +struct rbug_proto_shader_list * rbug_demarshal_shader_list(struct rbug_proto_header *header); + +struct rbug_proto_shader_info * rbug_demarshal_shader_info(struct rbug_proto_header *header); + +struct rbug_proto_shader_disable * rbug_demarshal_shader_disable(struct rbug_proto_header *header); + +struct rbug_proto_shader_replace * rbug_demarshal_shader_replace(struct rbug_proto_header *header); + +struct rbug_proto_shader_list_reply * rbug_demarshal_shader_list_reply(struct rbug_proto_header *header); + +struct rbug_proto_shader_info_reply * rbug_demarshal_shader_info_reply(struct rbug_proto_header *header); + +#endif diff --git a/src/gallium/auxiliary/rbug/rbug_texture.c b/src/gallium/auxiliary/rbug/rbug_texture.c new file mode 100644 index 00000000000..5a918fe6bc0 --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_texture.c @@ -0,0 +1,631 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/* + * This file holds the function implementation for one of the rbug extensions. + * Prototypes and declerations of functions and structs is in the same folder + * in the header file matching this file's name. + * + * The functions starting rbug_send_* encodes a call to the write format and + * sends that to the supplied connection, while functions starting with + * rbug_demarshal_* demarshal data in the wire protocol. + * + * Functions ending with _reply are replies to requests. + */ + +#include "rbug_internal.h" +#include "rbug/rbug_texture.h" + +int rbug_send_texture_list(struct rbug_connection *__con, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_TEXTURE_LIST)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_TEXTURE_LIST, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_texture_info(struct rbug_connection *__con, + rbug_texture_t texture, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* texture */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_TEXTURE_INFO)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_texture_t, texture); /* texture */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_TEXTURE_INFO, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_texture_write(struct rbug_connection *__con, + rbug_texture_t texture, + uint32_t face, + uint32_t level, + uint32_t zslice, + uint32_t x, + uint32_t y, + uint32_t w, + uint32_t h, + uint8_t *data, + uint32_t data_len, + uint32_t stride, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* texture */ + LEN(4); /* face */ + LEN(4); /* level */ + LEN(4); /* zslice */ + LEN(4); /* x */ + LEN(4); /* y */ + LEN(4); /* w */ + LEN(4); /* h */ + LEN_ARRAY(1, data); /* data */ + LEN(4); /* stride */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_TEXTURE_WRITE)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_texture_t, texture); /* texture */ + WRITE(4, uint32_t, face); /* face */ + WRITE(4, uint32_t, level); /* level */ + WRITE(4, uint32_t, zslice); /* zslice */ + WRITE(4, uint32_t, x); /* x */ + WRITE(4, uint32_t, y); /* y */ + WRITE(4, uint32_t, w); /* w */ + WRITE(4, uint32_t, h); /* h */ + WRITE_ARRAY(1, uint8_t, data); /* data */ + WRITE(4, uint32_t, stride); /* stride */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_TEXTURE_WRITE, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_texture_read(struct rbug_connection *__con, + rbug_texture_t texture, + uint32_t face, + uint32_t level, + uint32_t zslice, + uint32_t x, + uint32_t y, + uint32_t w, + uint32_t h, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(8); /* texture */ + LEN(4); /* face */ + LEN(4); /* level */ + LEN(4); /* zslice */ + LEN(4); /* x */ + LEN(4); /* y */ + LEN(4); /* w */ + LEN(4); /* h */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_TEXTURE_READ)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(8, rbug_texture_t, texture); /* texture */ + WRITE(4, uint32_t, face); /* face */ + WRITE(4, uint32_t, level); /* level */ + WRITE(4, uint32_t, zslice); /* zslice */ + WRITE(4, uint32_t, x); /* x */ + WRITE(4, uint32_t, y); /* y */ + WRITE(4, uint32_t, w); /* w */ + WRITE(4, uint32_t, h); /* h */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_TEXTURE_READ, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_texture_list_reply(struct rbug_connection *__con, + uint32_t serial, + rbug_texture_t *textures, + uint32_t textures_len, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(4); /* serial */ + LEN_ARRAY(8, textures); /* textures */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_TEXTURE_LIST_REPLY)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(4, uint32_t, serial); /* serial */ + WRITE_ARRAY(8, rbug_texture_t, textures); /* textures */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_TEXTURE_LIST_REPLY, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_texture_info_reply(struct rbug_connection *__con, + uint32_t serial, + uint32_t target, + uint32_t format, + uint32_t *width, + uint32_t width_len, + uint32_t *height, + uint32_t height_len, + uint32_t *depth, + uint32_t depth_len, + uint32_t blockw, + uint32_t blockh, + uint32_t blocksize, + uint32_t last_level, + uint32_t nr_samples, + uint32_t tex_usage, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(4); /* serial */ + LEN(4); /* target */ + LEN(4); /* format */ + LEN_ARRAY(4, width); /* width */ + LEN_ARRAY(4, height); /* height */ + LEN_ARRAY(4, depth); /* depth */ + LEN(4); /* blockw */ + LEN(4); /* blockh */ + LEN(4); /* blocksize */ + LEN(4); /* last_level */ + LEN(4); /* nr_samples */ + LEN(4); /* tex_usage */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_TEXTURE_INFO_REPLY)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(4, uint32_t, serial); /* serial */ + WRITE(4, uint32_t, target); /* target */ + WRITE(4, uint32_t, format); /* format */ + WRITE_ARRAY(4, uint32_t, width); /* width */ + WRITE_ARRAY(4, uint32_t, height); /* height */ + WRITE_ARRAY(4, uint32_t, depth); /* depth */ + WRITE(4, uint32_t, blockw); /* blockw */ + WRITE(4, uint32_t, blockh); /* blockh */ + WRITE(4, uint32_t, blocksize); /* blocksize */ + WRITE(4, uint32_t, last_level); /* last_level */ + WRITE(4, uint32_t, nr_samples); /* nr_samples */ + WRITE(4, uint32_t, tex_usage); /* tex_usage */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_TEXTURE_INFO_REPLY, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +int rbug_send_texture_read_reply(struct rbug_connection *__con, + uint32_t serial, + uint32_t format, + uint32_t blockw, + uint32_t blockh, + uint32_t blocksize, + uint8_t *data, + uint32_t data_len, + uint32_t stride, + uint32_t *__serial) +{ + uint32_t __len = 0; + uint32_t __pos = 0; + uint8_t *__data = NULL; + int __ret = 0; + + LEN(8); /* header */ + LEN(4); /* serial */ + LEN(4); /* format */ + LEN(4); /* blockw */ + LEN(4); /* blockh */ + LEN(4); /* blocksize */ + LEN_ARRAY(1, data); /* data */ + LEN(4); /* stride */ + + /* align */ + PAD(__len, 8); + + __data = (uint8_t*)MALLOC(__len); + if (!__data) + return -ENOMEM; + + WRITE(4, int32_t, ((int32_t)RBUG_OP_TEXTURE_READ_REPLY)); + WRITE(4, uint32_t, ((uint32_t)(__len / 4))); + WRITE(4, uint32_t, serial); /* serial */ + WRITE(4, uint32_t, format); /* format */ + WRITE(4, uint32_t, blockw); /* blockw */ + WRITE(4, uint32_t, blockh); /* blockh */ + WRITE(4, uint32_t, blocksize); /* blocksize */ + WRITE_ARRAY(1, uint8_t, data); /* data */ + WRITE(4, uint32_t, stride); /* stride */ + + /* final pad */ + PAD(__pos, 8); + + if (__pos != __len) { + __ret = -EINVAL; + } else { + rbug_connection_send_start(__con, RBUG_OP_TEXTURE_READ_REPLY, __len); + rbug_connection_write(__con, __data, __len); + __ret = rbug_connection_send_finish(__con, __serial); + } + + FREE(__data); + return __ret; +} + +struct rbug_proto_texture_list * rbug_demarshal_texture_list(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_texture_list *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_TEXTURE_LIST) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + + return ret; +} + +struct rbug_proto_texture_info * rbug_demarshal_texture_info(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_texture_info *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_TEXTURE_INFO) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_texture_t, texture); /* texture */ + + return ret; +} + +struct rbug_proto_texture_write * rbug_demarshal_texture_write(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_texture_write *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_TEXTURE_WRITE) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_texture_t, texture); /* texture */ + READ(4, uint32_t, face); /* face */ + READ(4, uint32_t, level); /* level */ + READ(4, uint32_t, zslice); /* zslice */ + READ(4, uint32_t, x); /* x */ + READ(4, uint32_t, y); /* y */ + READ(4, uint32_t, w); /* w */ + READ(4, uint32_t, h); /* h */ + READ_ARRAY(1, uint8_t, data); /* data */ + READ(4, uint32_t, stride); /* stride */ + + return ret; +} + +struct rbug_proto_texture_read * rbug_demarshal_texture_read(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_texture_read *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_TEXTURE_READ) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(8, rbug_texture_t, texture); /* texture */ + READ(4, uint32_t, face); /* face */ + READ(4, uint32_t, level); /* level */ + READ(4, uint32_t, zslice); /* zslice */ + READ(4, uint32_t, x); /* x */ + READ(4, uint32_t, y); /* y */ + READ(4, uint32_t, w); /* w */ + READ(4, uint32_t, h); /* h */ + + return ret; +} + +struct rbug_proto_texture_list_reply * rbug_demarshal_texture_list_reply(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_texture_list_reply *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_TEXTURE_LIST_REPLY) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(4, uint32_t, serial); /* serial */ + READ_ARRAY(8, rbug_texture_t, textures); /* textures */ + + return ret; +} + +struct rbug_proto_texture_info_reply * rbug_demarshal_texture_info_reply(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_texture_info_reply *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_TEXTURE_INFO_REPLY) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(4, uint32_t, serial); /* serial */ + READ(4, uint32_t, target); /* target */ + READ(4, uint32_t, format); /* format */ + READ_ARRAY(4, uint32_t, width); /* width */ + READ_ARRAY(4, uint32_t, height); /* height */ + READ_ARRAY(4, uint32_t, depth); /* depth */ + READ(4, uint32_t, blockw); /* blockw */ + READ(4, uint32_t, blockh); /* blockh */ + READ(4, uint32_t, blocksize); /* blocksize */ + READ(4, uint32_t, last_level); /* last_level */ + READ(4, uint32_t, nr_samples); /* nr_samples */ + READ(4, uint32_t, tex_usage); /* tex_usage */ + + return ret; +} + +struct rbug_proto_texture_read_reply * rbug_demarshal_texture_read_reply(struct rbug_proto_header *header) +{ + uint32_t len = 0; + uint32_t pos = 0; + uint8_t *data = NULL; + struct rbug_proto_texture_read_reply *ret; + + if (!header) + return NULL; + if (header->opcode != (int16_t)RBUG_OP_TEXTURE_READ_REPLY) + return NULL; + + pos = 0; + len = header->length * 4; + data = (uint8_t*)&header[1]; + ret = MALLOC(sizeof(*ret)); + if (!ret) + return NULL; + + ret->header.__message = header; + ret->header.opcode = header->opcode; + + READ(4, uint32_t, serial); /* serial */ + READ(4, uint32_t, format); /* format */ + READ(4, uint32_t, blockw); /* blockw */ + READ(4, uint32_t, blockh); /* blockh */ + READ(4, uint32_t, blocksize); /* blocksize */ + READ_ARRAY(1, uint8_t, data); /* data */ + READ(4, uint32_t, stride); /* stride */ + + return ret; +} diff --git a/src/gallium/auxiliary/rbug/rbug_texture.h b/src/gallium/auxiliary/rbug/rbug_texture.h new file mode 100644 index 00000000000..fbb247e1d4f --- /dev/null +++ b/src/gallium/auxiliary/rbug/rbug_texture.h @@ -0,0 +1,206 @@ +/* + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/* + * This file holds structs decelerations and function prototypes for one of + * the rbug extensions. Implementation of the functions is in the same folder + * in the c file matching this file's name. + * + * The structs what is returned from the demarshal functions. The functions + * starting rbug_send_* encodes a call to the write format and sends that to + * the supplied connection, while functions starting with rbug_demarshal_* + * demarshal data from the wire protocol. + * + * Structs and functions ending with _reply are replies to requests. + */ + +#ifndef _RBUG_PROTO_TEXTURE_H_ +#define _RBUG_PROTO_TEXTURE_H_ + +#include "rbug/rbug_proto.h" +#include "rbug/rbug_core.h" + +struct rbug_proto_texture_list +{ + struct rbug_header header; +}; + +struct rbug_proto_texture_info +{ + struct rbug_header header; + rbug_texture_t texture; +}; + +struct rbug_proto_texture_write +{ + struct rbug_header header; + rbug_texture_t texture; + uint32_t face; + uint32_t level; + uint32_t zslice; + uint32_t x; + uint32_t y; + uint32_t w; + uint32_t h; + uint8_t *data; + uint32_t data_len; + uint32_t stride; +}; + +struct rbug_proto_texture_read +{ + struct rbug_header header; + rbug_texture_t texture; + uint32_t face; + uint32_t level; + uint32_t zslice; + uint32_t x; + uint32_t y; + uint32_t w; + uint32_t h; +}; + +struct rbug_proto_texture_list_reply +{ + struct rbug_header header; + uint32_t serial; + rbug_texture_t *textures; + uint32_t textures_len; +}; + +struct rbug_proto_texture_info_reply +{ + struct rbug_header header; + uint32_t serial; + uint32_t target; + uint32_t format; + uint32_t *width; + uint32_t width_len; + uint32_t *height; + uint32_t height_len; + uint32_t *depth; + uint32_t depth_len; + uint32_t blockw; + uint32_t blockh; + uint32_t blocksize; + uint32_t last_level; + uint32_t nr_samples; + uint32_t tex_usage; +}; + +struct rbug_proto_texture_read_reply +{ + struct rbug_header header; + uint32_t serial; + uint32_t format; + uint32_t blockw; + uint32_t blockh; + uint32_t blocksize; + uint8_t *data; + uint32_t data_len; + uint32_t stride; +}; + +int rbug_send_texture_list(struct rbug_connection *__con, + uint32_t *__serial); + +int rbug_send_texture_info(struct rbug_connection *__con, + rbug_texture_t texture, + uint32_t *__serial); + +int rbug_send_texture_write(struct rbug_connection *__con, + rbug_texture_t texture, + uint32_t face, + uint32_t level, + uint32_t zslice, + uint32_t x, + uint32_t y, + uint32_t w, + uint32_t h, + uint8_t *data, + uint32_t data_len, + uint32_t stride, + uint32_t *__serial); + +int rbug_send_texture_read(struct rbug_connection *__con, + rbug_texture_t texture, + uint32_t face, + uint32_t level, + uint32_t zslice, + uint32_t x, + uint32_t y, + uint32_t w, + uint32_t h, + uint32_t *__serial); + +int rbug_send_texture_list_reply(struct rbug_connection *__con, + uint32_t serial, + rbug_texture_t *textures, + uint32_t textures_len, + uint32_t *__serial); + +int rbug_send_texture_info_reply(struct rbug_connection *__con, + uint32_t serial, + uint32_t target, + uint32_t format, + uint32_t *width, + uint32_t width_len, + uint32_t *height, + uint32_t height_len, + uint32_t *depth, + uint32_t depth_len, + uint32_t blockw, + uint32_t blockh, + uint32_t blocksize, + uint32_t last_level, + uint32_t nr_samples, + uint32_t tex_usage, + uint32_t *__serial); + +int rbug_send_texture_read_reply(struct rbug_connection *__con, + uint32_t serial, + uint32_t format, + uint32_t blockw, + uint32_t blockh, + uint32_t blocksize, + uint8_t *data, + uint32_t data_len, + uint32_t stride, + uint32_t *__serial); + +struct rbug_proto_texture_list * rbug_demarshal_texture_list(struct rbug_proto_header *header); + +struct rbug_proto_texture_info * rbug_demarshal_texture_info(struct rbug_proto_header *header); + +struct rbug_proto_texture_write * rbug_demarshal_texture_write(struct rbug_proto_header *header); + +struct rbug_proto_texture_read * rbug_demarshal_texture_read(struct rbug_proto_header *header); + +struct rbug_proto_texture_list_reply * rbug_demarshal_texture_list_reply(struct rbug_proto_header *header); + +struct rbug_proto_texture_info_reply * rbug_demarshal_texture_info_reply(struct rbug_proto_header *header); + +struct rbug_proto_texture_read_reply * rbug_demarshal_texture_read_reply(struct rbug_proto_header *header); + +#endif diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c index 57fcf6de2ab..1acf3c373eb 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c @@ -993,6 +993,15 @@ void sse_pmovmskb( struct x86_function *p, emit_modrm(p, dst, src); } +void sse_movmskps( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src) +{ + DUMP_RR( dst, src ); + emit_2ub(p, X86_TWOB, 0x50); + emit_modrm(p, dst, src); +} + /*********************************************************************** * SSE2 instructions */ diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h index 1b5eaaca850..731a6517968 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h @@ -223,6 +223,7 @@ void sse_unpckhps( struct x86_function *p, struct x86_reg dst, struct x86_reg sr void sse_unpcklps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_pmovmskb( struct x86_function *p, struct x86_reg dest, struct x86_reg src ); void sse2_punpcklbw( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_movmskps( struct x86_function *p, struct x86_reg dst, struct x86_reg src); void x86_add( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void x86_and( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); diff --git a/src/gallium/auxiliary/tgsi/Makefile b/src/gallium/auxiliary/tgsi/Makefile index b4900e8dbaa..5f0a580b096 100644 --- a/src/gallium/auxiliary/tgsi/Makefile +++ b/src/gallium/auxiliary/tgsi/Makefile @@ -16,6 +16,7 @@ C_SOURCES = \ tgsi_sse2.c \ tgsi_text.c \ tgsi_transform.c \ + tgsi_ureg.c \ tgsi_util.c include ../../Makefile.template diff --git a/src/gallium/auxiliary/tgsi/SConscript b/src/gallium/auxiliary/tgsi/SConscript index 8200cce42f5..b6bc2924f06 100644 --- a/src/gallium/auxiliary/tgsi/SConscript +++ b/src/gallium/auxiliary/tgsi/SConscript @@ -16,6 +16,7 @@ tgsi = env.ConvenienceLibrary( 'tgsi_sse2.c', 'tgsi_text.c', 'tgsi_transform.c', + 'tgsi_ureg.c', 'tgsi_util.c', ]) diff --git a/src/gallium/auxiliary/tgsi/tgsi-instruction-set.txt b/src/gallium/auxiliary/tgsi/tgsi-instruction-set.txt index a3f4947c734..eb492076b7d 100644 --- a/src/gallium/auxiliary/tgsi/tgsi-instruction-set.txt +++ b/src/gallium/auxiliary/tgsi/tgsi-instruction-set.txt @@ -187,11 +187,7 @@ TGSI Instruction Specification 1.2.6 CND0 - Condition Zero - dst.x = (src2.x >= 0.0) ? src0.x : src1.x - dst.y = (src2.y >= 0.0) ? src0.y : src1.y - dst.z = (src2.z >= 0.0) ? src0.z : src1.z - dst.w = (src2.w >= 0.0) ? src0.w : src1.w - + Removed. Use (CMP src2, src1, src0) instead. 1.2.7 DOT2ADD - 2-component Dot Product And Add @@ -665,9 +661,18 @@ TGSI Instruction Specification TBD -1.9.8 LOOP - Loop +1.9.8 BGNFOR - Begin a For-Loop - TBD + dst.x = floor(src.x) + dst.y = floor(src.y) + dst.z = floor(src.z) + + if (dst.y <= 0) + pc = [matching ENDFOR] + 1 + endif + + Note: The destination must be a loop register. + The source must be a constant register. 1.9.9 REP - Repeat @@ -685,9 +690,16 @@ TGSI Instruction Specification TBD -1.9.12 ENDLOOP - End Loop +1.9.12 ENDFOR - End a For-Loop - TBD + dst.x = dst.x + dst.z + dst.y = dst.y - 1.0 + + if (dst.y > 0) + pc = [matching BGNFOR instruction] + 1 + endif + + Note: The destination must be a loop register. 1.9.13 ENDREP - End Repeat @@ -840,7 +852,7 @@ TGSI Instruction Specification ---------- -1.13.1 BGNLOOP2 - Begin Loop +1.13.1 BGNLOOP - Begin a Loop TBD @@ -850,7 +862,7 @@ TGSI Instruction Specification TBD -1.13.3 ENDLOOP2 - End Loop +1.13.3 ENDLOOP - End a Loop TBD @@ -1015,12 +1027,12 @@ TGSI Instruction Specification 1.18.1 EXPP - Approximate Exponential Base 2 - Alias for EXP. + Use EXP. See also 1.19.3. 1.18.2 LOGP - Logarithm Base 2 - Alias for LG2. + Use LOG. See also 1.19.4. 1.19 vs_2_0 @@ -1037,6 +1049,16 @@ TGSI Instruction Specification Alias for ARR. +1.19.3 EXPP - Approximate Exponential Base 2 + + Use EX2. + + +1.19.4 LOGP - Logarithm Base 2 + + Use LG2. + + 2 Explanation of symbols used ============================== diff --git a/src/gallium/auxiliary/tgsi/tgsi_build.c b/src/gallium/auxiliary/tgsi/tgsi_build.c index 94d9c638e1c..e0cfc54420e 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_build.c +++ b/src/gallium/auxiliary/tgsi/tgsi_build.c @@ -335,7 +335,10 @@ tgsi_default_full_immediate( void ) struct tgsi_full_immediate fullimm; fullimm.Immediate = tgsi_default_immediate(); - fullimm.u.Pointer = (void *) 0; + fullimm.u[0].Float = 0.0f; + fullimm.u[1].Float = 0.0f; + fullimm.u[2].Float = 0.0f; + fullimm.u[3].Float = 0.0f; return fullimm; } @@ -352,19 +355,19 @@ immediate_grow( header_bodysize_grow( header ); } -struct tgsi_immediate_float32 +union tgsi_immediate_data tgsi_build_immediate_float32( float value, struct tgsi_immediate *immediate, struct tgsi_header *header ) { - struct tgsi_immediate_float32 immediate_float32; + union tgsi_immediate_data immediate_data; - immediate_float32.Float = value; + immediate_data.Float = value; immediate_grow( immediate, header ); - return immediate_float32; + return immediate_data; } unsigned @@ -384,16 +387,18 @@ tgsi_build_full_immediate( *immediate = tgsi_build_immediate( header ); + assert( full_imm->Immediate.NrTokens <= 4 + 1 ); + for( i = 0; i < full_imm->Immediate.NrTokens - 1; i++ ) { - struct tgsi_immediate_float32 *if32; + union tgsi_immediate_data *data; if( maxsize <= size ) return 0; - if32 = (struct tgsi_immediate_float32 *) &tokens[size]; + data = (union tgsi_immediate_data *) &tokens[size]; size++; - *if32 = tgsi_build_immediate_float32( - full_imm->u.ImmediateFloat32[i].Float, + *data = tgsi_build_immediate_float32( + full_imm->u[i].Float, immediate, header ); } diff --git a/src/gallium/auxiliary/tgsi/tgsi_build.h b/src/gallium/auxiliary/tgsi/tgsi_build.h index 9a3a077cf2b..17d977b0597 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_build.h +++ b/src/gallium/auxiliary/tgsi/tgsi_build.h @@ -119,7 +119,7 @@ tgsi_build_immediate( struct tgsi_full_immediate tgsi_default_full_immediate( void ); -struct tgsi_immediate_float32 +union tgsi_immediate_data tgsi_build_immediate_float32( float value, struct tgsi_immediate *immediate, diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c b/src/gallium/auxiliary/tgsi/tgsi_dump.c index a6994ecd48b..111d95b6665 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_dump.c +++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c @@ -33,12 +33,20 @@ #include "tgsi_info.h" #include "tgsi_iterate.h" + +/** Number of spaces to indent for IF/LOOP/etc */ +static const int indent_spaces = 3; + + struct dump_ctx { struct tgsi_iterate_context iter; uint instno; + int indent; + uint indentation; + void (*printf)(struct dump_ctx *ctx, const char *format, ...); }; @@ -82,7 +90,7 @@ static const char *processor_type_names[] = "GEOM" }; -static const char *file_names[] = +static const char *file_names[TGSI_FILE_COUNT] = { "NULL", "CONST", @@ -91,7 +99,8 @@ static const char *file_names[] = "TEMP", "SAMP", "ADDR", - "IMM" + "IMM", + "LOOP" }; static const char *interpolate_names[] = @@ -295,10 +304,12 @@ iter_immediate( ENM( imm->Immediate.DataType, immediate_type_names ); TXT( " { " ); + + assert( imm->Immediate.NrTokens <= 4 + 1 ); for (i = 0; i < imm->Immediate.NrTokens - 1; i++) { switch (imm->Immediate.DataType) { case TGSI_IMM_FLOAT32: - FLT( imm->u.ImmediateFloat32[i].Float ); + FLT( imm->u[i].Float ); break; default: assert( 0 ); @@ -332,13 +343,19 @@ iter_instruction( { struct dump_ctx *ctx = (struct dump_ctx *) iter; uint instno = ctx->instno++; - + const struct tgsi_opcode_info *info = tgsi_get_opcode_info( inst->Instruction.Opcode ); uint i; boolean first_reg = TRUE; INSTID( instno ); TXT( ": " ); - TXT( tgsi_get_opcode_info( inst->Instruction.Opcode )->mnemonic ); + + ctx->indent -= info->pre_dedent; + for(i = 0; (int)i < ctx->indent; ++i) + TXT( " " ); + ctx->indent += info->post_indent; + + TXT( info->mnemonic ); switch (inst->Instruction.Saturate) { case TGSI_SAT_NONE: @@ -470,14 +487,22 @@ iter_instruction( switch (inst->Instruction.Opcode) { case TGSI_OPCODE_IF: case TGSI_OPCODE_ELSE: - case TGSI_OPCODE_BGNLOOP2: - case TGSI_OPCODE_ENDLOOP2: + case TGSI_OPCODE_BGNLOOP: + case TGSI_OPCODE_ENDLOOP: case TGSI_OPCODE_CAL: TXT( " :" ); UID( inst->InstructionExtLabel.Label ); break; } + /* update indentation */ + if (inst->Instruction.Opcode == TGSI_OPCODE_IF || + inst->Instruction.Opcode == TGSI_OPCODE_ELSE || + inst->Instruction.Opcode == TGSI_OPCODE_BGNFOR || + inst->Instruction.Opcode == TGSI_OPCODE_BGNLOOP) { + ctx->indentation += indent_spaces; + } + EOL(); return TRUE; @@ -491,7 +516,9 @@ tgsi_dump_instruction( struct dump_ctx ctx; ctx.instno = instno; + ctx.indent = 0; ctx.printf = dump_ctx_printf; + ctx.indentation = 0; iter_instruction( &ctx.iter, (struct tgsi_full_instruction *)inst ); } @@ -523,7 +550,9 @@ tgsi_dump( ctx.iter.epilog = NULL; ctx.instno = 0; + ctx.indent = 0; ctx.printf = dump_ctx_printf; + ctx.indentation = 0; tgsi_iterate_shader( tokens, &ctx.iter ); } @@ -575,7 +604,9 @@ tgsi_dump_str( ctx.base.iter.epilog = NULL; ctx.base.instno = 0; + ctx.base.indent = 0; ctx.base.printf = &str_dump_ctx_printf; + ctx.base.indentation = 0; ctx.str = str; ctx.str[0] = 0; diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump_c.c b/src/gallium/auxiliary/tgsi/tgsi_dump_c.c index 3dc61c48ca3..4a9c02b1413 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_dump_c.c +++ b/src/gallium/auxiliary/tgsi/tgsi_dump_c.c @@ -69,7 +69,7 @@ static const char *TGSI_TOKEN_TYPES[] = "TOKEN_TYPE_INSTRUCTION" }; -static const char *TGSI_FILES[] = +static const char *TGSI_FILES[TGSI_FILE_COUNT] = { "FILE_NULL", "FILE_CONSTANT", @@ -78,7 +78,8 @@ static const char *TGSI_FILES[] = "FILE_TEMPORARY", "FILE_SAMPLER", "FILE_ADDRESS", - "FILE_IMMEDIATE" + "FILE_IMMEDIATE", + "FILE_LOOP" }; static const char *TGSI_INTERPOLATES[] = @@ -283,12 +284,13 @@ dump_immediate_verbose( UIX( imm->Immediate.Padding ); } + assert( imm->Immediate.NrTokens <= 4 + 1 ); for( i = 0; i < imm->Immediate.NrTokens - 1; i++ ) { EOL(); switch( imm->Immediate.DataType ) { case TGSI_IMM_FLOAT32: TXT( "\nFloat: " ); - FLT( imm->u.ImmediateFloat32[i].Float ); + FLT( imm->u[i].Float ); break; default: diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c index 259894e72e0..c79c56debd6 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_exec.c +++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c @@ -304,14 +304,14 @@ tgsi_exec_machine_bind_shader( case TGSI_TOKEN_TYPE_IMMEDIATE: { uint size = parse.FullToken.FullImmediate.Immediate.NrTokens - 1; - assert( size % 4 == 0 ); - assert( mach->ImmLimit + size / 4 <= TGSI_EXEC_NUM_IMMEDIATES ); + assert( size <= 4 ); + assert( mach->ImmLimit + 1 <= TGSI_EXEC_NUM_IMMEDIATES ); for( i = 0; i < size; i++ ) { - mach->Imms[mach->ImmLimit + i / 4][i % 4] = - parse.FullToken.FullImmediate.u.ImmediateFloat32[i].Float; + mach->Imms[mach->ImmLimit][i] = + parse.FullToken.FullImmediate.u[i].Float; } - mach->ImmLimit += size / 4; + mach->ImmLimit += 1; } break; @@ -373,13 +373,18 @@ tgsi_exec_machine_bind_shader( } -void -tgsi_exec_machine_init( - struct tgsi_exec_machine *mach ) +struct tgsi_exec_machine * +tgsi_exec_machine_create( void ) { + struct tgsi_exec_machine *mach; uint i; - mach->Temps = (struct tgsi_exec_vector *) tgsi_align_128bit( mach->_Temps); + mach = align_malloc( sizeof *mach, 16 ); + if (!mach) + goto fail; + + memset(mach, 0, sizeof(*mach)); + mach->Addrs = &mach->Temps[TGSI_EXEC_TEMP_ADDR]; /* Setup constants. */ @@ -401,22 +406,24 @@ tgsi_exec_machine_init( (void) print_chan; (void) print_temp; #endif + + return mach; + +fail: + align_free(mach); + return NULL; } void -tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach) +tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach) { - if (mach->Instructions) { + if (mach) { FREE(mach->Instructions); - mach->Instructions = NULL; - mach->NumInstructions = 0; - } - if (mach->Declarations) { FREE(mach->Declarations); - mach->Declarations = NULL; - mach->NumDeclarations = 0; } + + align_free(mach); } @@ -2015,8 +2022,7 @@ exec_instruction( switch (inst->Instruction.Opcode) { case TGSI_OPCODE_ARL: - case TGSI_OPCODE_FLOOR: - /* TGSI_OPCODE_FLR */ + case TGSI_OPCODE_FLR: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); micro_flr( &r[0], &r[0] ); @@ -2299,8 +2305,7 @@ exec_instruction( } break; - case TGSI_OPCODE_LERP: - /* TGSI_OPCODE_LRP */ + case TGSI_OPCODE_LRP: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH(&r[0], 0, chan_index); FETCH(&r[1], 1, chan_index); @@ -2324,18 +2329,7 @@ exec_instruction( } break; - case TGSI_OPCODE_CND0: - FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { - FETCH(&r[0], 0, chan_index); - FETCH(&r[1], 1, chan_index); - FETCH(&r[2], 2, chan_index); - micro_le(&r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[2], &r[0], &r[1]); - STORE(&r[0], 0, chan_index); - } - break; - - case TGSI_OPCODE_DOT2ADD: - /* TGSI_OPCODE_DP2A */ + case TGSI_OPCODE_DP2A: FETCH( &r[0], 0, CHAN_X ); FETCH( &r[1], 1, CHAN_X ); micro_mul( &r[0], &r[0], &r[1] ); @@ -2353,18 +2347,7 @@ exec_instruction( } break; - case TGSI_OPCODE_INDEX: - /* XXX: considered for removal */ - assert (0); - break; - - case TGSI_OPCODE_NEGATE: - /* XXX: considered for removal */ - assert (0); - break; - - case TGSI_OPCODE_FRAC: - /* TGSI_OPCODE_FRC */ + case TGSI_OPCODE_FRC: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); micro_frc( &r[0], &r[0] ); @@ -2392,8 +2375,7 @@ exec_instruction( } break; - case TGSI_OPCODE_EXPBASE2: - /* TGSI_OPCODE_EX2 */ + case TGSI_OPCODE_EX2: FETCH(&r[0], 0, CHAN_X); #if FAST_MATH @@ -2407,8 +2389,7 @@ exec_instruction( } break; - case TGSI_OPCODE_LOGBASE2: - /* TGSI_OPCODE_LG2 */ + case TGSI_OPCODE_LG2: FETCH( &r[0], 0, CHAN_X ); micro_lg2( &r[0], &r[0] ); FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { @@ -2416,8 +2397,7 @@ exec_instruction( } break; - case TGSI_OPCODE_POWER: - /* TGSI_OPCODE_POW */ + case TGSI_OPCODE_POW: FETCH(&r[0], 0, CHAN_X); FETCH(&r[1], 1, CHAN_X); @@ -2428,8 +2408,7 @@ exec_instruction( } break; - case TGSI_OPCODE_CROSSPRODUCT: - /* TGSI_OPCODE_XPD */ + case TGSI_OPCODE_XPD: FETCH(&r[0], 0, CHAN_Y); FETCH(&r[1], 1, CHAN_Z); @@ -2471,11 +2450,6 @@ exec_instruction( } break; - case TGSI_OPCODE_MULTIPLYMATRIX: - /* XXX: considered for removal */ - assert (0); - break; - case TGSI_OPCODE_ABS: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH(&r[0], 0, chan_index); @@ -2782,19 +2756,32 @@ exec_instruction( if (mach->ExecMask) { /* do the call */ - /* push the Cond, Loop, Cont stacks */ + /* First, record the depths of the execution stacks. + * This is important for deeply nested/looped return statements. + * We have to unwind the stacks by the correct amount. For a + * real code generator, we could determine the number of entries + * to pop off each stack with simple static analysis and avoid + * implementing this data structure at run time. + */ + mach->CallStack[mach->CallStackTop].CondStackTop = mach->CondStackTop; + mach->CallStack[mach->CallStackTop].LoopStackTop = mach->LoopStackTop; + mach->CallStack[mach->CallStackTop].ContStackTop = mach->ContStackTop; + /* note that PC was already incremented above */ + mach->CallStack[mach->CallStackTop].ReturnAddr = *pc; + + mach->CallStackTop++; + + /* Second, push the Cond, Loop, Cont, Func stacks */ assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING); mach->CondStack[mach->CondStackTop++] = mach->CondMask; assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING); mach->ContStack[mach->ContStackTop++] = mach->ContMask; - assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING); mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask; - /* note that PC was already incremented above */ - mach->CallStack[mach->CallStackTop++] = *pc; + /* Finally, jump to the subroutine */ *pc = inst->InstructionExtLabel.Label; } break; @@ -2811,18 +2798,24 @@ exec_instruction( *pc = -1; return; } - *pc = mach->CallStack[--mach->CallStackTop]; - /* pop the Cond, Loop, Cont stacks */ - assert(mach->CondStackTop > 0); - mach->CondMask = mach->CondStack[--mach->CondStackTop]; - assert(mach->LoopStackTop > 0); - mach->LoopMask = mach->LoopStack[--mach->LoopStackTop]; - assert(mach->ContStackTop > 0); - mach->ContMask = mach->ContStack[--mach->ContStackTop]; + assert(mach->CallStackTop > 0); + mach->CallStackTop--; + + mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop; + mach->CondMask = mach->CondStack[mach->CondStackTop]; + + mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop; + mach->LoopMask = mach->LoopStack[mach->LoopStackTop]; + + mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop; + mach->ContMask = mach->ContStack[mach->ContStackTop]; + assert(mach->FuncStackTop > 0); mach->FuncMask = mach->FuncStack[--mach->FuncStackTop]; + *pc = mach->CallStack[mach->CallStackTop].ReturnAddr; + UPDATE_EXEC_MASK(mach); } break; @@ -3119,26 +3112,81 @@ exec_instruction( mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]] = 0; break; - case TGSI_OPCODE_LOOP: + case TGSI_OPCODE_BGNFOR: + assert(mach->LoopCounterStackTop < TGSI_EXEC_MAX_LOOP_NESTING); + for (chan_index = 0; chan_index < 3; chan_index++) { + FETCH( &mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[chan_index], 0, chan_index ); + } + STORE( &mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[CHAN_Y], 0, CHAN_X ); + ++mach->LoopCounterStackTop; /* fall-through (for now) */ - case TGSI_OPCODE_BGNLOOP2: + case TGSI_OPCODE_BGNLOOP: /* push LoopMask and ContMasks */ assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING); mach->ContStack[mach->ContStackTop++] = mach->ContMask; + assert(mach->LoopLabelStackTop < TGSI_EXEC_MAX_LOOP_NESTING); + mach->LoopLabelStack[mach->LoopLabelStackTop++] = *pc - 1; + break; + + case TGSI_OPCODE_ENDFOR: + assert(mach->LoopCounterStackTop > 0); + micro_sub( &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], + &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], + &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C] ); + /* update LoopMask */ + if( mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X].f[0] <= 0) { + mach->LoopMask &= ~0x1; + } + if( mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X].f[1] <= 0 ) { + mach->LoopMask &= ~0x2; + } + if( mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X].f[2] <= 0 ) { + mach->LoopMask &= ~0x4; + } + if( mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X].f[3] <= 0 ) { + mach->LoopMask &= ~0x8; + } + micro_add( &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y], + &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y], + &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Z]); + assert(mach->LoopLabelStackTop > 0); + inst = mach->Instructions + mach->LoopLabelStack[mach->LoopLabelStackTop - 1]; + STORE( &mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[CHAN_Y], 0, CHAN_X ); + /* Restore ContMask, but don't pop */ + assert(mach->ContStackTop > 0); + mach->ContMask = mach->ContStack[mach->ContStackTop - 1]; + UPDATE_EXEC_MASK(mach); + if (mach->ExecMask) { + /* repeat loop: jump to instruction just past BGNLOOP */ + assert(mach->LoopLabelStackTop > 0); + *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1; + } + else { + /* exit loop: pop LoopMask */ + assert(mach->LoopStackTop > 0); + mach->LoopMask = mach->LoopStack[--mach->LoopStackTop]; + /* pop ContMask */ + assert(mach->ContStackTop > 0); + mach->ContMask = mach->ContStack[--mach->ContStackTop]; + assert(mach->LoopLabelStackTop > 0); + --mach->LoopLabelStackTop; + assert(mach->LoopCounterStackTop > 0); + --mach->LoopCounterStackTop; + } + UPDATE_EXEC_MASK(mach); break; - + case TGSI_OPCODE_ENDLOOP: - /* fall-through (for now at least) */ - case TGSI_OPCODE_ENDLOOP2: /* Restore ContMask, but don't pop */ assert(mach->ContStackTop > 0); mach->ContMask = mach->ContStack[mach->ContStackTop - 1]; UPDATE_EXEC_MASK(mach); if (mach->ExecMask) { /* repeat loop: jump to instruction just past BGNLOOP */ - *pc = inst->InstructionExtLabel.Label + 1; + assert(mach->LoopLabelStackTop > 0); + *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1; } else { /* exit loop: pop LoopMask */ @@ -3147,6 +3195,8 @@ exec_instruction( /* pop ContMask */ assert(mach->ContStackTop > 0); mach->ContMask = mach->ContStack[--mach->ContStackTop]; + assert(mach->LoopLabelStackTop > 0); + --mach->LoopLabelStackTop; } UPDATE_EXEC_MASK(mach); break; @@ -3214,7 +3264,6 @@ tgsi_exec_machine_run( struct tgsi_exec_machine *mach ) mach->FuncMask = 0xf; mach->ExecMask = 0xf; - mach->CondStackTop = 0; /* temporarily subvert this assertion */ assert(mach->CondStackTop == 0); assert(mach->LoopStackTop == 0); assert(mach->ContStackTop == 0); diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h index 182e5e228b5..c72f76809d4 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_exec.h +++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h @@ -29,6 +29,7 @@ #define TGSI_EXEC_H #include "pipe/p_compiler.h" +#include "pipe/p_state.h" #if defined __cplusplus extern "C" { @@ -94,7 +95,6 @@ struct tgsi_exec_labels #define TGSI_EXEC_NUM_TEMPS 128 -#define TGSI_EXEC_NUM_TEMP_EXTRAS 6 #define TGSI_EXEC_NUM_IMMEDIATES 256 /* @@ -162,9 +162,14 @@ struct tgsi_exec_labels #define TGSI_EXEC_MASK_I (TGSI_EXEC_NUM_TEMPS + 3) #define TGSI_EXEC_MASK_C 2 +/* 4 register buffer for various purposes */ #define TGSI_EXEC_TEMP_R0 (TGSI_EXEC_NUM_TEMPS + 4) +#define TGSI_EXEC_NUM_TEMP_R 4 + +#define TGSI_EXEC_TEMP_ADDR (TGSI_EXEC_NUM_TEMPS + 8) +#define TGSI_EXEC_NUM_ADDRS 1 +#define TGSI_EXEC_NUM_TEMP_EXTRAS 9 -#define TGSI_EXEC_TEMP_ADDR (TGSI_EXEC_NUM_TEMPS + 5) #define TGSI_EXEC_MAX_COND_NESTING 20 @@ -181,30 +186,38 @@ struct tgsi_exec_labels */ #define TGSI_EXEC_MAX_CONST_BUFFER 4096 + +/** function call/activation record */ +struct tgsi_call_record +{ + uint CondStackTop; + uint LoopStackTop; + uint ContStackTop; + uint ReturnAddr; +}; + + /** * Run-time virtual machine state for executing TGSI shader. */ struct tgsi_exec_machine { /* Total = program temporaries + internal temporaries - * + 1 padding to align to 16 bytes */ - struct tgsi_exec_vector _Temps[TGSI_EXEC_NUM_TEMPS + - TGSI_EXEC_NUM_TEMP_EXTRAS + 1]; + struct tgsi_exec_vector Temps[TGSI_EXEC_NUM_TEMPS + + TGSI_EXEC_NUM_TEMP_EXTRAS]; + + float Imms[TGSI_EXEC_NUM_IMMEDIATES][4]; + + struct tgsi_exec_vector Inputs[PIPE_MAX_ATTRIBS]; + struct tgsi_exec_vector Outputs[PIPE_MAX_ATTRIBS]; - /* - * This will point to _Temps after aligning to 16B boundary. - */ - struct tgsi_exec_vector *Temps; struct tgsi_exec_vector *Addrs; struct tgsi_sampler **Samplers; - float Imms[TGSI_EXEC_NUM_IMMEDIATES][4]; unsigned ImmLimit; const float (*Consts)[4]; - struct tgsi_exec_vector *Inputs; - struct tgsi_exec_vector *Outputs; const struct tgsi_token *Tokens; /**< Declarations, instructions */ unsigned Processor; /**< TGSI_PROCESSOR_x */ @@ -230,6 +243,14 @@ struct tgsi_exec_machine uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING]; int LoopStackTop; + /** Loop label stack */ + uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING]; + int LoopLabelStackTop; + + /** Loop counter stack (x = count, y = current, z = step) */ + struct tgsi_exec_vector LoopCounterStack[TGSI_EXEC_MAX_LOOP_NESTING]; + int LoopCounterStackTop; + /** Loop continue mask stack (see comments in tgsi_exec.c) */ uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING]; int ContStackTop; @@ -239,7 +260,7 @@ struct tgsi_exec_machine int FuncStackTop; /** Function call stack for saving/restoring the program counter */ - uint CallStack[TGSI_EXEC_MAX_CALL_NESTING]; + struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING]; int CallStackTop; struct tgsi_full_instruction *Instructions; @@ -251,9 +272,11 @@ struct tgsi_exec_machine struct tgsi_exec_labels Labels; }; +struct tgsi_exec_machine * +tgsi_exec_machine_create( void ); + void -tgsi_exec_machine_init( - struct tgsi_exec_machine *mach ); +tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach); void diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.c b/src/gallium/auxiliary/tgsi/tgsi_info.c index 37f2b66d1f6..17af4cb7ad2 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_info.c +++ b/src/gallium/auxiliary/tgsi/tgsi_info.c @@ -26,136 +26,156 @@ **************************************************************************/ #include "util/u_debug.h" +#include "util/u_memory.h" #include "tgsi_info.h" static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] = { - { 1, 1, 0, 0, "ARL", NULL, NULL }, - { 1, 1, 0, 0, "MOV", NULL, NULL }, - { 1, 1, 0, 0, "LIT", NULL, NULL }, - { 1, 1, 0, 0, "RCP", "RECIP", NULL }, - { 1, 1, 0, 0, "RSQ", "RECIPSQRT", NULL }, - { 1, 1, 0, 0, "EXP", "EXPP", NULL }, - { 1, 1, 0, 0, "LOG", NULL, NULL }, - { 1, 2, 0, 0, "MUL", NULL, NULL }, - { 1, 2, 0, 0, "ADD", NULL, NULL }, - { 1, 2, 0, 0, "DP3", "DOT3", NULL }, - { 1, 2, 0, 0, "DP4", "DOT4", NULL }, - { 1, 2, 0, 0, "DST", NULL, NULL }, - { 1, 2, 0, 0, "MIN", NULL, NULL }, - { 1, 2, 0, 0, "MAX", NULL, NULL }, - { 1, 2, 0, 0, "SLT", "SETLT", NULL }, - { 1, 2, 0, 0, "SGE", "SETGE", NULL }, - { 1, 3, 0, 0, "MAD", "MADD", NULL }, - { 1, 2, 0, 0, "SUB", NULL, NULL }, - { 1, 3, 0, 0, "LRP", "LERP", NULL }, - { 1, 3, 0, 0, "CND", NULL, NULL }, - { 1, 3, 0, 0, "CND0", NULL, NULL }, - { 1, 3, 0, 0, "DP2A", "DP2ADD", "DOT2ADD" }, - { 1, 2, 0, 0, "INDEX", NULL, NULL }, - { 1, 1, 0, 0, "NEGATE", NULL, NULL }, - { 1, 1, 0, 0, "FRC", "FRAC", NULL }, - { 1, 3, 0, 0, "CLAMP", NULL, NULL }, - { 1, 1, 0, 0, "FLR", "FLOOR", NULL }, - { 1, 1, 0, 0, "ROUND", NULL, NULL }, - { 1, 1, 0, 0, "EX2", "EXPBASE2", NULL }, - { 1, 1, 0, 0, "LG2", "LOGBASE2", "LOGP" }, - { 1, 2, 0, 0, "POW", "POWER", NULL }, - { 1, 2, 0, 0, "XPD", "CRS", "CROSSPRODUCT" }, - { 1, 2, 0, 0, "M4X4", "MULTIPLYMATRIX", NULL }, - { 1, 1, 0, 0, "ABS", NULL, NULL }, - { 1, 1, 0, 0, "RCC", NULL, NULL }, - { 1, 2, 0, 0, "DPH", NULL, NULL }, - { 1, 1, 0, 0, "COS", NULL, NULL }, - { 1, 1, 0, 0, "DDX", "DSX", NULL }, - { 1, 1, 0, 0, "DDY", "DSY", NULL }, - { 0, 0, 0, 0, "KILP", NULL, NULL }, - { 1, 1, 0, 0, "PK2H", NULL, NULL }, - { 1, 1, 0, 0, "PK2US", NULL, NULL }, - { 1, 1, 0, 0, "PK4B", NULL, NULL }, - { 1, 1, 0, 0, "PK4UB", NULL, NULL }, - { 1, 2, 0, 0, "RFL", NULL, NULL }, - { 1, 2, 0, 0, "SEQ", NULL, NULL }, - { 1, 2, 0, 0, "SFL", NULL, NULL }, - { 1, 2, 0, 0, "SGT", NULL, NULL }, - { 1, 1, 0, 0, "SIN", NULL, NULL }, - { 1, 2, 0, 0, "SLE", NULL, NULL }, - { 1, 2, 0, 0, "SNE", NULL, NULL }, - { 1, 2, 0, 0, "STR", NULL, NULL }, - { 1, 2, 1, 0, "TEX", "TEXLD", NULL }, - { 1, 4, 1, 0, "TXD", "TEXLDD", NULL }, - { 1, 2, 1, 0, "TXP", NULL, NULL }, - { 1, 1, 0, 0, "UP2H", NULL, NULL }, - { 1, 1, 0, 0, "UP2US", NULL, NULL }, - { 1, 1, 0, 0, "UP4B", NULL, NULL }, - { 1, 1, 0, 0, "UP4UB", NULL, NULL }, - { 1, 3, 0, 0, "X2D", NULL, NULL }, - { 1, 1, 0, 0, "ARA", NULL, NULL }, - { 1, 1, 0, 0, "ARR", "MOVA", NULL }, - { 0, 1, 0, 0, "BRA", NULL, NULL }, - { 0, 0, 0, 1, "CAL", "CALL", NULL }, - { 0, 0, 0, 0, "RET", NULL, NULL }, - { 1, 1, 0, 0, "SGN", "SSG", NULL }, - { 1, 3, 0, 0, "CMP", NULL, NULL }, - { 1, 1, 0, 0, "SCS", "SINCOS", NULL }, - { 1, 2, 1, 0, "TXB", "TEXLDB", NULL }, - { 1, 1, 0, 0, "NRM", NULL, NULL }, - { 1, 2, 0, 0, "DIV", NULL, NULL }, - { 1, 2, 0, 0, "DP2", NULL, NULL }, - { 1, 2, 1, 0, "TXL", NULL, NULL }, - { 0, 0, 0, 0, "BRK", "BREAK", NULL }, - { 0, 1, 0, 1, "IF", NULL, NULL }, - { 0, 0, 0, 0, "LOOP", NULL, NULL }, - { 0, 1, 0, 0, "REP", NULL, NULL }, - { 0, 0, 0, 1, "ELSE", NULL, NULL }, - { 0, 0, 0, 0, "ENDIF", NULL, NULL }, - { 0, 0, 0, 0, "ENDLOOP", NULL, NULL }, - { 0, 0, 0, 0, "ENDREP", NULL, NULL }, - { 0, 1, 0, 0, "PUSHA", NULL, NULL }, - { 1, 0, 0, 0, "POPA", NULL, NULL }, - { 1, 1, 0, 0, "CEIL", NULL, NULL }, - { 1, 1, 0, 0, "I2F", NULL, NULL }, - { 1, 1, 0, 0, "NOT", NULL, NULL }, - { 1, 1, 0, 0, "INT", "TRUNC", NULL }, - { 1, 2, 0, 0, "SHL", NULL, NULL }, - { 1, 2, 0, 0, "SHR", NULL, NULL }, - { 1, 2, 0, 0, "AND", NULL, NULL }, - { 1, 2, 0, 0, "OR", NULL, NULL }, - { 1, 2, 0, 0, "MOD", NULL, NULL }, - { 1, 2, 0, 0, "XOR", NULL, NULL }, - { 1, 3, 0, 0, "SAD", NULL, NULL }, - { 1, 2, 1, 0, "TXF", NULL, NULL }, - { 1, 2, 1, 0, "TXQ", NULL, NULL }, - { 0, 0, 0, 0, "CONT", NULL, NULL }, - { 0, 0, 0, 0, "EMIT", NULL, NULL }, - { 0, 0, 0, 0, "ENDPRIM", NULL, NULL }, - { 0, 0, 0, 1, "BGNLOOP2", NULL, NULL }, - { 0, 0, 0, 0, "BGNSUB", NULL, NULL }, - { 0, 0, 0, 1, "ENDLOOP2", NULL, NULL }, - { 0, 0, 0, 0, "ENDSUB", NULL, NULL }, - { 1, 1, 0, 0, "NOISE1", NULL, NULL }, - { 1, 1, 0, 0, "NOISE2", NULL, NULL }, - { 1, 1, 0, 0, "NOISE3", NULL, NULL }, - { 1, 1, 0, 0, "NOISE4", NULL, NULL }, - { 0, 0, 0, 0, "NOP", NULL, NULL }, - { 1, 2, 0, 0, "M4X3", NULL, NULL }, - { 1, 2, 0, 0, "M3X4", NULL, NULL }, - { 1, 2, 0, 0, "M3X3", NULL, NULL }, - { 1, 2, 0, 0, "M3X2", NULL, NULL }, - { 1, 1, 0, 0, "NRM4", NULL, NULL }, - { 0, 1, 0, 0, "CALLNZ", NULL, NULL }, - { 0, 1, 0, 0, "IFC", NULL, NULL }, - { 0, 1, 0, 0, "BREAKC", NULL, NULL }, - { 0, 1, 0, 0, "KIL", "TEXKILL", NULL }, - { 0, 0, 0, 0, "END", NULL, NULL }, - { 1, 1, 0, 0, "SWZ", NULL, NULL } + { 1, 1, 0, 0, 0, 0, "ARL", TGSI_OPCODE_ARL }, + { 1, 1, 0, 0, 0, 0, "MOV", TGSI_OPCODE_MOV }, + { 1, 1, 0, 0, 0, 0, "LIT", TGSI_OPCODE_LIT }, + { 1, 1, 0, 0, 0, 0, "RCP", TGSI_OPCODE_RCP }, + { 1, 1, 0, 0, 0, 0, "RSQ", TGSI_OPCODE_RSQ }, + { 1, 1, 0, 0, 0, 0, "EXP", TGSI_OPCODE_EXP }, + { 1, 1, 0, 0, 0, 0, "LOG", TGSI_OPCODE_LOG }, + { 1, 2, 0, 0, 0, 0, "MUL", TGSI_OPCODE_MUL }, + { 1, 2, 0, 0, 0, 0, "ADD", TGSI_OPCODE_ADD }, + { 1, 2, 0, 0, 0, 0, "DP3", TGSI_OPCODE_DP3 }, + { 1, 2, 0, 0, 0, 0, "DP4", TGSI_OPCODE_DP4 }, + { 1, 2, 0, 0, 0, 0, "DST", TGSI_OPCODE_DST }, + { 1, 2, 0, 0, 0, 0, "MIN", TGSI_OPCODE_MIN }, + { 1, 2, 0, 0, 0, 0, "MAX", TGSI_OPCODE_MAX }, + { 1, 2, 0, 0, 0, 0, "SLT", TGSI_OPCODE_SLT }, + { 1, 2, 0, 0, 0, 0, "SGE", TGSI_OPCODE_SGE }, + { 1, 3, 0, 0, 0, 0, "MAD", TGSI_OPCODE_MAD }, + { 1, 2, 0, 0, 0, 0, "SUB", TGSI_OPCODE_SUB }, + { 1, 3, 0, 0, 0, 0, "LRP", TGSI_OPCODE_LRP }, + { 1, 3, 0, 0, 0, 0, "CND", TGSI_OPCODE_CND }, + { 0, 0, 0, 0, 0, 0, "", 20 }, /* removed */ + { 1, 3, 0, 0, 0, 0, "DP2A", TGSI_OPCODE_DP2A }, + { 0, 0, 0, 0, 0, 0, "", 22 }, /* removed */ + { 0, 0, 0, 0, 0, 0, "", 23 }, /* removed */ + { 1, 1, 0, 0, 0, 0, "FRC", TGSI_OPCODE_FRC }, + { 1, 3, 0, 0, 0, 0, "CLAMP", TGSI_OPCODE_CLAMP }, + { 1, 1, 0, 0, 0, 0, "FLR", TGSI_OPCODE_FLR }, + { 1, 1, 0, 0, 0, 0, "ROUND", TGSI_OPCODE_ROUND }, + { 1, 1, 0, 0, 0, 0, "EX2", TGSI_OPCODE_EX2 }, + { 1, 1, 0, 0, 0, 0, "LG2", TGSI_OPCODE_LG2 }, + { 1, 2, 0, 0, 0, 0, "POW", TGSI_OPCODE_POW }, + { 1, 2, 0, 0, 0, 0, "XPD", TGSI_OPCODE_XPD }, + { 0, 0, 0, 0, 0, 0, "", 32 }, /* removed */ + { 1, 1, 0, 0, 0, 0, "ABS", TGSI_OPCODE_ABS }, + { 1, 1, 0, 0, 0, 0, "RCC", TGSI_OPCODE_RCC }, + { 1, 2, 0, 0, 0, 0, "DPH", TGSI_OPCODE_DPH }, + { 1, 1, 0, 0, 0, 0, "COS", TGSI_OPCODE_COS }, + { 1, 1, 0, 0, 0, 0, "DDX", TGSI_OPCODE_DDX }, + { 1, 1, 0, 0, 0, 0, "DDY", TGSI_OPCODE_DDY }, + { 0, 0, 0, 0, 0, 0, "KILP", TGSI_OPCODE_KILP }, + { 1, 1, 0, 0, 0, 0, "PK2H", TGSI_OPCODE_PK2H }, + { 1, 1, 0, 0, 0, 0, "PK2US", TGSI_OPCODE_PK2US }, + { 1, 1, 0, 0, 0, 0, "PK4B", TGSI_OPCODE_PK4B }, + { 1, 1, 0, 0, 0, 0, "PK4UB", TGSI_OPCODE_PK4UB }, + { 1, 2, 0, 0, 0, 0, "RFL", TGSI_OPCODE_RFL }, + { 1, 2, 0, 0, 0, 0, "SEQ", TGSI_OPCODE_SEQ }, + { 1, 2, 0, 0, 0, 0, "SFL", TGSI_OPCODE_SFL }, + { 1, 2, 0, 0, 0, 0, "SGT", TGSI_OPCODE_SGT }, + { 1, 1, 0, 0, 0, 0, "SIN", TGSI_OPCODE_SIN }, + { 1, 2, 0, 0, 0, 0, "SLE", TGSI_OPCODE_SLE }, + { 1, 2, 0, 0, 0, 0, "SNE", TGSI_OPCODE_SNE }, + { 1, 2, 0, 0, 0, 0, "STR", TGSI_OPCODE_STR }, + { 1, 2, 1, 0, 0, 0, "TEX", TGSI_OPCODE_TEX }, + { 1, 4, 1, 0, 0, 0, "TXD", TGSI_OPCODE_TXD }, + { 1, 2, 1, 0, 0, 0, "TXP", TGSI_OPCODE_TXP }, + { 1, 1, 0, 0, 0, 0, "UP2H", TGSI_OPCODE_UP2H }, + { 1, 1, 0, 0, 0, 0, "UP2US", TGSI_OPCODE_UP2US }, + { 1, 1, 0, 0, 0, 0, "UP4B", TGSI_OPCODE_UP4B }, + { 1, 1, 0, 0, 0, 0, "UP4UB", TGSI_OPCODE_UP4UB }, + { 1, 3, 0, 0, 0, 0, "X2D", TGSI_OPCODE_X2D }, + { 1, 1, 0, 0, 0, 0, "ARA", TGSI_OPCODE_ARA }, + { 1, 1, 0, 0, 0, 0, "ARR", TGSI_OPCODE_ARR }, + { 0, 1, 0, 0, 0, 0, "BRA", TGSI_OPCODE_BRA }, + { 0, 0, 0, 1, 0, 0, "CAL", TGSI_OPCODE_CAL }, + { 0, 0, 0, 0, 0, 0, "RET", TGSI_OPCODE_RET }, + { 1, 1, 0, 0, 0, 0, "SSG", TGSI_OPCODE_SSG }, + { 1, 3, 0, 0, 0, 0, "CMP", TGSI_OPCODE_CMP }, + { 1, 1, 0, 0, 0, 0, "SCS", TGSI_OPCODE_SCS }, + { 1, 2, 1, 0, 0, 0, "TXB", TGSI_OPCODE_TXB }, + { 1, 1, 0, 0, 0, 0, "NRM", TGSI_OPCODE_NRM }, + { 1, 2, 0, 0, 0, 0, "DIV", TGSI_OPCODE_DIV }, + { 1, 2, 0, 0, 0, 0, "DP2", TGSI_OPCODE_DP2 }, + { 1, 2, 1, 0, 0, 0, "TXL", TGSI_OPCODE_TXL }, + { 0, 0, 0, 0, 0, 0, "BRK", TGSI_OPCODE_BRK }, + { 0, 1, 0, 1, 0, 1, "IF", TGSI_OPCODE_IF }, + { 1, 1, 0, 0, 0, 1, "BGNFOR", TGSI_OPCODE_BGNFOR }, + { 0, 1, 0, 0, 0, 1, "REP", TGSI_OPCODE_REP }, + { 0, 0, 0, 1, 1, 1, "ELSE", TGSI_OPCODE_ELSE }, + { 0, 0, 0, 0, 1, 0, "ENDIF", TGSI_OPCODE_ENDIF }, + { 1, 0, 0, 0, 1, 0, "ENDFOR", TGSI_OPCODE_ENDFOR }, + { 0, 0, 0, 0, 1, 0, "ENDREP", TGSI_OPCODE_ENDREP }, + { 0, 1, 0, 0, 0, 0, "PUSHA", TGSI_OPCODE_PUSHA }, + { 1, 0, 0, 0, 0, 0, "POPA", TGSI_OPCODE_POPA }, + { 1, 1, 0, 0, 0, 0, "CEIL", TGSI_OPCODE_CEIL }, + { 1, 1, 0, 0, 0, 0, "I2F", TGSI_OPCODE_I2F }, + { 1, 1, 0, 0, 0, 0, "NOT", TGSI_OPCODE_NOT }, + { 1, 1, 0, 0, 0, 0, "TRUNC", TGSI_OPCODE_TRUNC }, + { 1, 2, 0, 0, 0, 0, "SHL", TGSI_OPCODE_SHL }, + { 1, 2, 0, 0, 0, 0, "SHR", TGSI_OPCODE_SHR }, + { 1, 2, 0, 0, 0, 0, "AND", TGSI_OPCODE_AND }, + { 1, 2, 0, 0, 0, 0, "OR", TGSI_OPCODE_OR }, + { 1, 2, 0, 0, 0, 0, "MOD", TGSI_OPCODE_MOD }, + { 1, 2, 0, 0, 0, 0, "XOR", TGSI_OPCODE_XOR }, + { 1, 3, 0, 0, 0, 0, "SAD", TGSI_OPCODE_SAD }, + { 1, 2, 1, 0, 0, 0, "TXF", TGSI_OPCODE_TXF }, + { 1, 2, 1, 0, 0, 0, "TXQ", TGSI_OPCODE_TXQ }, + { 0, 0, 0, 0, 0, 0, "CONT", TGSI_OPCODE_CONT }, + { 0, 0, 0, 0, 0, 0, "EMIT", TGSI_OPCODE_EMIT }, + { 0, 0, 0, 0, 0, 0, "ENDPRIM", TGSI_OPCODE_ENDPRIM }, + { 0, 0, 0, 1, 0, 1, "BGNLOOP", TGSI_OPCODE_BGNLOOP }, + { 0, 0, 0, 0, 0, 1, "BGNSUB", TGSI_OPCODE_BGNSUB }, + { 0, 0, 0, 1, 1, 0, "ENDLOOP", TGSI_OPCODE_ENDLOOP }, + { 0, 0, 0, 0, 1, 0, "ENDSUB", TGSI_OPCODE_ENDSUB }, + { 1, 1, 0, 0, 0, 0, "NOISE1", TGSI_OPCODE_NOISE1 }, + { 1, 1, 0, 0, 0, 0, "NOISE2", TGSI_OPCODE_NOISE2 }, + { 1, 1, 0, 0, 0, 0, "NOISE3", TGSI_OPCODE_NOISE3 }, + { 1, 1, 0, 0, 0, 0, "NOISE4", TGSI_OPCODE_NOISE4 }, + { 0, 0, 0, 0, 0, 0, "NOP", TGSI_OPCODE_NOP }, + { 0, 0, 0, 0, 0, 0, "", 108 }, /* removed */ + { 0, 0, 0, 0, 0, 0, "", 109 }, /* removed */ + { 0, 0, 0, 0, 0, 0, "", 110 }, /* removed */ + { 0, 0, 0, 0, 0, 0, "", 111 }, /* removed */ + { 1, 1, 0, 0, 0, 0, "NRM4", TGSI_OPCODE_NRM4 }, + { 0, 1, 0, 0, 0, 0, "CALLNZ", TGSI_OPCODE_CALLNZ }, + { 0, 1, 0, 0, 0, 0, "IFC", TGSI_OPCODE_IFC }, + { 0, 1, 0, 0, 0, 0, "BREAKC", TGSI_OPCODE_BREAKC }, + { 0, 1, 0, 0, 0, 0, "KIL", TGSI_OPCODE_KIL }, + { 0, 0, 0, 0, 0, 0, "END", TGSI_OPCODE_END }, + { 1, 1, 0, 0, 0, 0, "SWZ", TGSI_OPCODE_SWZ } }; const struct tgsi_opcode_info * tgsi_get_opcode_info( uint opcode ) { + static boolean firsttime = 1; + + if (firsttime) { + unsigned i; + firsttime = 0; + for (i = 0; i < Elements(opcode_info); i++) + assert(opcode_info[i].opcode == i); + } + if (opcode < TGSI_OPCODE_LAST) return &opcode_info[opcode]; + assert( 0 ); return NULL; } + + +const char * +tgsi_get_opcode_name( uint opcode ) +{ + const struct tgsi_opcode_info *info = tgsi_get_opcode_info(opcode); + return info->mnemonic; +} + diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.h b/src/gallium/auxiliary/tgsi/tgsi_info.h index 077e25acd7f..74713c3b98a 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_info.h +++ b/src/gallium/auxiliary/tgsi/tgsi_info.h @@ -36,18 +36,23 @@ extern "C" { struct tgsi_opcode_info { - uint num_dst; - uint num_src; - boolean is_tex; - boolean is_branch; + unsigned num_dst:3; + unsigned num_src:3; + unsigned is_tex:1; + unsigned is_branch:1; + int pre_dedent:2; + int post_indent:2; const char *mnemonic; - const char *alt_mnemonic1; - const char *alt_mnemonic2; + uint opcode; }; const struct tgsi_opcode_info * tgsi_get_opcode_info( uint opcode ); +const char * +tgsi_get_opcode_name( uint opcode ); + + #if defined __cplusplus } #endif diff --git a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h new file mode 100644 index 00000000000..e7bcf4bf754 --- /dev/null +++ b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h @@ -0,0 +1,172 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ +#ifndef OP12_TEX +#define OP12_TEX(a) OP12(a) +#endif + +#ifndef OP14_TEX +#define OP14_TEX(a) OP14(a) +#endif + +#ifndef OP00_LBL +#define OP00_LBL(a) OP00(a) +#endif + +#ifndef OP01_LBL +#define OP01_LBL(a) OP01(a) +#endif + +OP11(ARL) +OP11(MOV) +OP11(LIT) +OP11(RCP) +OP11(RSQ) +OP11(EXP) +OP11(LOG) +OP12(MUL) +OP12(ADD) +OP12(DP3) +OP12(DP4) +OP12(DST) +OP12(MIN) +OP12(MAX) +OP12(SLT) +OP12(SGE) +OP13(MAD) +OP12(SUB) +OP13(LRP) +OP13(CND) +OP13(DP2A) +OP11(FRC) +OP13(CLAMP) +OP11(FLR) +OP11(ROUND) +OP11(EX2) +OP11(LG2) +OP12(POW) +OP12(XPD) +OP11(ABS) +OP11(RCC) +OP12(DPH) +OP11(COS) +OP11(DDX) +OP11(DDY) +OP00(KILP) +OP11(PK2H) +OP11(PK2US) +OP11(PK4B) +OP11(PK4UB) +OP12(RFL) +OP12(SEQ) +OP12(SFL) +OP12(SGT) +OP11(SIN) +OP12(SLE) +OP12(SNE) +OP12(STR) +OP12_TEX(TEX) +OP14_TEX(TXD) +OP12_TEX(TXP) +OP11(UP2H) +OP11(UP2US) +OP11(UP4B) +OP11(UP4UB) +OP13(X2D) +OP11(ARA) +OP11(ARR) +OP01(BRA) +OP00_LBL(CAL) +OP00(RET) +OP11(SSG) +OP13(CMP) +OP11(SCS) +OP12_TEX(TXB) +OP11(NRM) +OP12(DIV) +OP12(DP2) +OP12_TEX(TXL) +OP00(BRK) +OP01_LBL(IF) +OP11(BGNFOR) +OP01(REP) +OP00_LBL(ELSE) +OP00(ENDIF) +OP10(ENDFOR) +OP00(ENDREP) +OP01(PUSHA) +OP10(POPA) +OP11(CEIL) +OP11(I2F) +OP11(NOT) +OP11(TRUNC) +OP12(SHL) +OP12(SHR) +OP12(AND) +OP12(OR) +OP12(MOD) +OP12(XOR) +OP13(SAD) +OP12_TEX(TXF) +OP12_TEX(TXQ) +OP00(CONT) +OP00(EMIT) +OP00(ENDPRIM) +OP00_LBL(BGNLOOP) +OP00(BGNSUB) +OP00_LBL(ENDLOOP) +OP00(ENDSUB) +OP11(NOISE1) +OP11(NOISE2) +OP11(NOISE3) +OP11(NOISE4) +OP00(NOP) +OP11(NRM4) +OP01(CALLNZ) +OP01(IFC) +OP01(BREAKC) +OP01(KIL) +OP00(END) +OP11(SWZ) + + +#undef OP00 +#undef OP01 +#undef OP10 +#undef OP11 +#undef OP12 +#undef OP13 + +#ifdef OP14 +#undef OP14 +#endif + +#undef OP00_LBL +#undef OP01_LBL + +#undef OP12_TEX +#undef OP14_TEX + diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.c b/src/gallium/auxiliary/tgsi/tgsi_parse.c index 7f2cfb7988f..4870f82b6bd 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_parse.c +++ b/src/gallium/auxiliary/tgsi/tgsi_parse.c @@ -42,9 +42,6 @@ void tgsi_full_token_free( union tgsi_full_token *full_token ) { - if( full_token->Token.Type == TGSI_TOKEN_TYPE_IMMEDIATE ) { - FREE( (void *) full_token->FullImmediate.u.Pointer ); - } } unsigned @@ -156,14 +153,8 @@ tgsi_parse_token( case TGSI_IMM_FLOAT32: { uint imm_count = imm->Immediate.NrTokens - 1; - struct tgsi_immediate_float32 *data; - - data = (struct tgsi_immediate_float32 *) MALLOC(sizeof(struct tgsi_immediate_float32) * imm_count); - if (data) { - for (i = 0; i < imm_count; i++) { - next_token(ctx, &data[i]); - } - imm->u.ImmediateFloat32 = data; + for (i = 0; i < imm_count; i++) { + next_token(ctx, &imm->u[i]); } } break; diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.h b/src/gallium/auxiliary/tgsi/tgsi_parse.h index 5268c029006..a26ee5ba862 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_parse.h +++ b/src/gallium/auxiliary/tgsi/tgsi_parse.h @@ -73,11 +73,7 @@ struct tgsi_full_declaration struct tgsi_full_immediate { struct tgsi_immediate Immediate; - union - { - const void *Pointer; - const struct tgsi_immediate_float32 *ImmediateFloat32; - } u; + union tgsi_immediate_data u[4]; }; #define TGSI_FULL_MAX_DST_REGISTERS 2 diff --git a/src/gallium/auxiliary/tgsi/tgsi_ppc.c b/src/gallium/auxiliary/tgsi/tgsi_ppc.c index 8466d9bc229..4b1c7d4e01b 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_ppc.c +++ b/src/gallium/auxiliary/tgsi/tgsi_ppc.c @@ -38,6 +38,7 @@ #include "util/u_math.h" #include "util/u_memory.h" #include "util/u_sse.h" +#include "tgsi/tgsi_info.h" #include "tgsi/tgsi_parse.h" #include "tgsi/tgsi_util.h" #include "tgsi_dump.h" @@ -619,17 +620,17 @@ emit_unaryop(struct gen_context *gen, struct tgsi_full_instruction *inst) ppc_vandc(gen->f, v1, v0, bit31_vec); /* v1 = v0 & ~bit31 */ } break; - case TGSI_OPCODE_FLOOR: + case TGSI_OPCODE_FLR: ppc_vrfim(gen->f, v1, v0); /* v1 = floor(v0) */ break; - case TGSI_OPCODE_FRAC: + case TGSI_OPCODE_FRC: ppc_vrfim(gen->f, v1, v0); /* tmp = floor(v0) */ ppc_vsubfp(gen->f, v1, v0, v1); /* v1 = v0 - v1 */ break; - case TGSI_OPCODE_EXPBASE2: + case TGSI_OPCODE_EX2: ppc_vexptefp(gen->f, v1, v0); /* v1 = 2^v0 */ break; - case TGSI_OPCODE_LOGBASE2: + case TGSI_OPCODE_LG2: /* XXX this may be broken! */ ppc_vlogefp(gen->f, v1, v0); /* v1 = log2(v0) */ break; @@ -1120,10 +1121,10 @@ emit_instruction(struct gen_context *gen, case TGSI_OPCODE_MOV: case TGSI_OPCODE_SWZ: case TGSI_OPCODE_ABS: - case TGSI_OPCODE_FLOOR: - case TGSI_OPCODE_FRAC: - case TGSI_OPCODE_EXPBASE2: - case TGSI_OPCODE_LOGBASE2: + case TGSI_OPCODE_FLR: + case TGSI_OPCODE_FRC: + case TGSI_OPCODE_EX2: + case TGSI_OPCODE_LG2: emit_unaryop(gen, inst); break; case TGSI_OPCODE_RSQ: @@ -1326,8 +1327,10 @@ tgsi_emit_ppc(const struct tgsi_token *tokens, ok = emit_instruction(&gen, &parse.FullToken.FullInstruction); if (!ok) { - debug_printf("failed to translate tgsi opcode %d to PPC (%s)\n", - parse.FullToken.FullInstruction.Instruction.Opcode, + uint opcode = parse.FullToken.FullInstruction.Instruction.Opcode; + debug_printf("failed to translate tgsi opcode %d (%s) to PPC (%s)\n", + opcode, + tgsi_get_opcode_name(opcode), parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX ? "vertex shader" : "fragment shader"); } @@ -1342,7 +1345,7 @@ tgsi_emit_ppc(const struct tgsi_token *tokens, assert(num_immediates < TGSI_EXEC_NUM_IMMEDIATES); for (i = 0; i < size; i++) { immediates[num_immediates][i] = - parse.FullToken.FullImmediate.u.ImmediateFloat32[i].Float; + parse.FullToken.FullImmediate.u[i].Float; } num_immediates++; } diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.c b/src/gallium/auxiliary/tgsi/tgsi_sanity.c index 6f1f5c2b4b0..8a13885da9b 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_sanity.c +++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.c @@ -131,7 +131,7 @@ is_register_used( return (ctx->regs_used[file][index / BITS_IN_REG_FLAG] & (1 << (index % BITS_IN_REG_FLAG))) ? TRUE : FALSE; } -static const char *file_names[] = +static const char *file_names[TGSI_FILE_COUNT] = { "NULL", "CONST", @@ -140,7 +140,8 @@ static const char *file_names[] = "TEMP", "SAMP", "ADDR", - "IMM" + "IMM", + "LOOP" }; static boolean @@ -198,10 +199,10 @@ iter_instruction( } if (info->num_dst != inst->Instruction.NumDstRegs) { - report_error( ctx, "Invalid number of destination operands, should be %u", info->num_dst ); + report_error( ctx, "%s: Invalid number of destination operands, should be %u", info->mnemonic, info->num_dst ); } if (info->num_src != inst->Instruction.NumSrcRegs) { - report_error( ctx, "Invalid number of source operands, should be %u", info->num_src ); + report_error( ctx, "%s: Invalid number of source operands, should be %u", info->mnemonic, info->num_src ); } /* Check destination and source registers' validity. @@ -234,9 +235,29 @@ iter_instruction( index, "indirect", FALSE ); - if (file != TGSI_FILE_ADDRESS || index != 0) - report_warning( ctx, "Indirect register not ADDR[0]" ); + if (!(file == TGSI_FILE_ADDRESS || file == TGSI_FILE_LOOP) || index != 0) { + report_warning(ctx, "Indirect register neither ADDR[0] nor LOOP[0]"); + } + } + } + + switch (inst->Instruction.Opcode) { + case TGSI_OPCODE_BGNFOR: + case TGSI_OPCODE_ENDFOR: + if (inst->FullDstRegisters[0].DstRegister.File != TGSI_FILE_LOOP || + inst->FullDstRegisters[0].DstRegister.Index != 0) { + report_error(ctx, "Destination register must be LOOP[0]"); + } + break; + } + + switch (inst->Instruction.Opcode) { + case TGSI_OPCODE_BGNFOR: + if (inst->FullSrcRegisters[0].SrcRegister.File != TGSI_FILE_CONSTANT && + inst->FullSrcRegisters[0].SrcRegister.File != TGSI_FILE_IMMEDIATE) { + report_error(ctx, "Source register file must be either CONST or IMM"); } + break; } ctx->num_instructions++; diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c index a13368c80c8..3cdf8b9f359 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c +++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c @@ -32,9 +32,11 @@ #include "util/u_debug.h" #include "pipe/p_shader_tokens.h" #include "util/u_math.h" +#include "util/u_memory.h" #if defined(PIPE_ARCH_SSE) #include "util/u_sse.h" #endif +#include "tgsi/tgsi_info.h" #include "tgsi/tgsi_parse.h" #include "tgsi/tgsi_util.h" #include "tgsi_exec.h" @@ -100,37 +102,55 @@ get_const_base( void ) { return x86_make_reg( file_REG32, - reg_CX ); + reg_AX ); } static struct x86_reg -get_input_base( void ) +get_machine_base( void ) { return x86_make_reg( file_REG32, - reg_AX ); + reg_CX ); +} + +static struct x86_reg +get_input_base( void ) +{ + return x86_make_disp( + get_machine_base(), + Offset(struct tgsi_exec_machine, Inputs) ); } static struct x86_reg get_output_base( void ) { - return x86_make_reg( - file_REG32, - reg_DX ); + return x86_make_disp( + get_machine_base(), + Offset(struct tgsi_exec_machine, Outputs) ); } static struct x86_reg get_temp_base( void ) { + return x86_make_disp( + get_machine_base(), + Offset(struct tgsi_exec_machine, Temps) ); +} + +static struct x86_reg +get_coef_base( void ) +{ return x86_make_reg( file_REG32, reg_BX ); } static struct x86_reg -get_coef_base( void ) +get_sampler_base( void ) { - return get_output_base(); + return x86_make_reg( + file_REG32, + reg_DI ); } static struct x86_reg @@ -138,7 +158,7 @@ get_immediate_base( void ) { return x86_make_reg( file_REG32, - reg_DI ); + reg_DX ); } @@ -168,6 +188,15 @@ get_const( } static struct x86_reg +get_sampler_ptr( + unsigned unit ) +{ + return x86_make_disp( + get_sampler_base(), + unit * sizeof( struct tgsi_sampler * ) ); +} + +static struct x86_reg get_input( unsigned vec, unsigned chan ) @@ -241,12 +270,14 @@ emit_const( /* 'vec' is the offset from the address register's value. * We're loading CONST[ADDR+vec] into an xmm register. */ - struct x86_reg r0 = get_input_base(); - struct x86_reg r1 = get_output_base(); + struct x86_reg r0 = get_immediate_base(); + struct x86_reg r1 = get_coef_base(); uint i; assert( indirectFile == TGSI_FILE_ADDRESS ); assert( indirectIndex == 0 ); + assert( r0.mod == mod_REG ); + assert( r1.mod == mod_REG ); x86_push( func, r0 ); x86_push( func, r1 ); @@ -520,24 +551,15 @@ emit_coef_dady( * that the stack pointer is 16 byte aligned, as expected. */ static void -emit_func_call_dst( +emit_func_call( struct x86_function *func, - unsigned xmm_save, - unsigned xmm_dst, + unsigned xmm_save_mask, + const struct x86_reg *arg, + unsigned nr_args, void (PIPE_CDECL *code)() ) { struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX ); unsigned i, n; - unsigned xmm_mask; - - /* Bitmask of the xmm registers to save */ - xmm_mask = (1 << xmm_save) - 1; - xmm_mask &= ~(1 << xmm_dst); - - sse_movaps( - func, - get_temp( TEMP_R0, 0 ), - make_xmm( xmm_dst ) ); x86_push( func, @@ -549,8 +571,10 @@ emit_func_call_dst( func, x86_make_reg( file_REG32, reg_DX) ); + /* Store XMM regs to the stack + */ for(i = 0, n = 0; i < 8; ++i) - if(xmm_mask & (1 << i)) + if(xmm_save_mask & (1 << i)) ++n; x86_sub_imm( @@ -559,26 +583,42 @@ emit_func_call_dst( n*16); for(i = 0, n = 0; i < 8; ++i) - if(xmm_mask & (1 << i)) { + if(xmm_save_mask & (1 << i)) { sse_movups( func, x86_make_disp( x86_make_reg( file_REG32, reg_SP ), n*16 ), make_xmm( i ) ); ++n; } + + for (i = 0; i < nr_args; i++) { + /* Load the address of the buffer we use for passing arguments and + * receiving results: + */ + x86_lea( + func, + ecx, + arg[i] ); - x86_lea( - func, - ecx, - get_temp( TEMP_R0, 0 ) ); - - x86_push( func, ecx ); + /* Push actual function arguments (currently just the pointer to + * the buffer above), and call the function: + */ + x86_push( func, ecx ); + } + x86_mov_reg_imm( func, ecx, (unsigned long) code ); x86_call( func, ecx ); - x86_pop(func, ecx ); - + + /* Pop the arguments (or just add an immediate to esp) + */ + for (i = 0; i < nr_args; i++) { + x86_pop(func, ecx ); + } + + /* Pop the saved XMM regs: + */ for(i = 0, n = 0; i < 8; ++i) - if(xmm_mask & (1 << i)) { + if(xmm_save_mask & (1 << i)) { sse_movups( func, make_xmm( i ), @@ -602,34 +642,86 @@ emit_func_call_dst( x86_pop( func, x86_make_reg( file_REG32, reg_AX) ); +} + +static void +emit_func_call_dst_src1( + struct x86_function *func, + unsigned xmm_save, + unsigned xmm_dst, + unsigned xmm_src0, + void (PIPE_CDECL *code)() ) +{ + struct x86_reg store = get_temp( TEMP_R0, 0 ); + unsigned xmm_mask = ((1 << xmm_save) - 1) & ~(1 << xmm_dst); + + /* Store our input parameters (in xmm regs) to the buffer we use + * for passing arguments. We will pass a pointer to this buffer as + * the actual function argument. + */ + sse_movaps( + func, + store, + make_xmm( xmm_src0 ) ); + + emit_func_call( func, + xmm_mask, + &store, + 1, + code ); sse_movaps( func, make_xmm( xmm_dst ), - get_temp( TEMP_R0, 0 ) ); + store ); } + static void -emit_func_call_dst_src( +emit_func_call_dst_src2( struct x86_function *func, unsigned xmm_save, unsigned xmm_dst, - unsigned xmm_src, + unsigned xmm_src0, + unsigned xmm_src1, void (PIPE_CDECL *code)() ) { + struct x86_reg store = get_temp( TEMP_R0, 0 ); + unsigned xmm_mask = ((1 << xmm_save) - 1) & ~(1 << xmm_dst); + + /* Store two inputs to parameter buffer. + */ sse_movaps( func, - get_temp( TEMP_R0, 1 ), - make_xmm( xmm_src ) ); + store, + make_xmm( xmm_src0 ) ); - emit_func_call_dst( + sse_movaps( func, - xmm_save, - xmm_dst, - code ); + x86_make_disp( store, 4 * sizeof(float) ), + make_xmm( xmm_src1 ) ); + + + /* Emit the call + */ + emit_func_call( func, + xmm_mask, + &store, + 1, + code ); + + /* Retrieve the results: + */ + sse_movaps( + func, + make_xmm( xmm_dst ), + store ); } + + + #if defined(PIPE_ARCH_SSE) /* @@ -782,10 +874,11 @@ emit_cos( unsigned xmm_save, unsigned xmm_dst ) { - emit_func_call_dst( + emit_func_call_dst_src1( func, xmm_save, xmm_dst, + xmm_dst, cos4f ); } @@ -812,10 +905,11 @@ emit_ex2( unsigned xmm_save, unsigned xmm_dst ) { - emit_func_call_dst( + emit_func_call_dst_src1( func, xmm_save, xmm_dst, + xmm_dst, ex24f ); } @@ -857,10 +951,11 @@ emit_flr( unsigned xmm_save, unsigned xmm_dst ) { - emit_func_call_dst( + emit_func_call_dst_src1( func, xmm_save, xmm_dst, + xmm_dst, flr4f ); } @@ -880,10 +975,11 @@ emit_frc( unsigned xmm_save, unsigned xmm_dst ) { - emit_func_call_dst( + emit_func_call_dst_src1( func, xmm_save, xmm_dst, + xmm_dst, frc4f ); } @@ -910,10 +1006,11 @@ emit_lg2( unsigned xmm_save, unsigned xmm_dst ) { - emit_func_call_dst( + emit_func_call_dst_src1( func, xmm_save, xmm_dst, + xmm_dst, lg24f ); } @@ -975,13 +1072,15 @@ emit_pow( struct x86_function *func, unsigned xmm_save, unsigned xmm_dst, - unsigned xmm_src ) + unsigned xmm_src0, + unsigned xmm_src1 ) { - emit_func_call_dst_src( + emit_func_call_dst_src2( func, xmm_save, xmm_dst, - xmm_src, + xmm_src0, + xmm_src1, pow4f ); } @@ -1017,10 +1116,11 @@ emit_rnd( unsigned xmm_save, unsigned xmm_dst ) { - emit_func_call_dst( + emit_func_call_dst_src1( func, xmm_save, xmm_dst, + xmm_dst, rnd4f ); } @@ -1099,10 +1199,11 @@ emit_sgn( unsigned xmm_save, unsigned xmm_dst ) { - emit_func_call_dst( + emit_func_call_dst_src1( func, xmm_save, xmm_dst, + xmm_dst, sgn4f ); } @@ -1121,10 +1222,11 @@ emit_sin (struct x86_function *func, unsigned xmm_save, unsigned xmm_dst) { - emit_func_call_dst( + emit_func_call_dst_src1( func, xmm_save, xmm_dst, + xmm_dst, sin4f ); } @@ -1140,6 +1242,12 @@ emit_sub( make_xmm( xmm_src ) ); } + + + + + + /** * Register fetch. */ @@ -1298,20 +1406,164 @@ emit_store( #define STORE( FUNC, INST, XMM, INDEX, CHAN )\ emit_store( FUNC, XMM, &(INST).FullDstRegisters[INDEX], &(INST), CHAN ) + +static void PIPE_CDECL +fetch_texel( struct tgsi_sampler **sampler, + float *store ) +{ +#if 0 + uint j; + + debug_printf("%s sampler: %p (%p) store: %p\n", + __FUNCTION__, + sampler, *sampler, + store ); + + debug_printf("lodbias %f\n", store[12]); + + for (j = 0; j < 4; j++) + debug_printf("sample %d texcoord %f %f\n", + j, + store[0+j], + store[4+j]); +#endif + + { + float rgba[NUM_CHANNELS][QUAD_SIZE]; + (*sampler)->get_samples(*sampler, + &store[0], + &store[4], + &store[8], + 0.0f, /*store[12], lodbias */ + rgba); + + memcpy( store, rgba, 16 * sizeof(float)); + } + +#if 0 + for (j = 0; j < 4; j++) + debug_printf("sample %d result %f %f %f %f\n", + j, + store[0+j], + store[4+j], + store[8+j], + store[12+j]); +#endif +} + /** * High-level instruction translators. */ static void +emit_tex( struct x86_function *func, + const struct tgsi_full_instruction *inst, + boolean lodbias, + boolean projected) +{ + const uint unit = inst->FullSrcRegisters[1].SrcRegister.Index; + struct x86_reg args[2]; + unsigned count; + unsigned i; + + switch (inst->InstructionExtTexture.Texture) { + case TGSI_TEXTURE_1D: + count = 1; + break; + case TGSI_TEXTURE_2D: + case TGSI_TEXTURE_RECT: + count = 2; + break; + case TGSI_TEXTURE_SHADOW1D: + case TGSI_TEXTURE_SHADOW2D: + case TGSI_TEXTURE_SHADOWRECT: + case TGSI_TEXTURE_3D: + case TGSI_TEXTURE_CUBE: + count = 3; + break; + default: + assert(0); + return; + } + + if (lodbias) { + FETCH( func, *inst, 3, 0, 3 ); + } + else { + emit_tempf( + func, + 3, + TGSI_EXEC_TEMP_00000000_I, + TGSI_EXEC_TEMP_00000000_C ); + + } + + /* store lodbias whether enabled or not -- fetch_texel currently + * respects it always. + */ + sse_movaps( func, + get_temp( TEMP_R0, 3 ), + make_xmm( 3 ) ); + + + if (projected) { + FETCH( func, *inst, 3, 0, 3 ); + + emit_rcp( func, 3, 3 ); + } + + for (i = 0; i < count; i++) { + FETCH( func, *inst, i, 0, i ); + + if (projected) { + sse_mulps( + func, + make_xmm( i ), + make_xmm( 3 ) ); + } + + /* Store in the argument buffer: + */ + sse_movaps( + func, + get_temp( TEMP_R0, i ), + make_xmm( i ) ); + } + + args[0] = get_temp( TEMP_R0, 0 ); + args[1] = get_sampler_ptr( unit ); + + + emit_func_call( func, + 0, + args, + Elements(args), + fetch_texel ); + + /* If all four channels are enabled, could use a pointer to + * dst[0].x instead of TEMP_R0 for store? + */ + FOR_EACH_DST0_ENABLED_CHANNEL( *inst, i ) { + + sse_movaps( + func, + make_xmm( 0 ), + get_temp( TEMP_R0, i ) ); + + STORE( func, *inst, 0, 0, i ); + } +} + + +static void emit_kil( struct x86_function *func, const struct tgsi_full_src_register *reg ) { unsigned uniquemask; - unsigned registers[4]; - unsigned nextregister = 0; - unsigned firstchan = ~0; + unsigned unique_count = 0; unsigned chan_index; + unsigned i; /* This mask stores component bits that were already tested. Note that * we test if the value is less than zero, so 1.0 and 0.0 need not to be @@ -1331,18 +1583,11 @@ emit_kil( uniquemask |= 1 << swizzle; /* allocate register */ - registers[chan_index] = nextregister; emit_fetch( func, - nextregister, + unique_count++, reg, chan_index ); - nextregister++; - - /* mark the first channel used */ - if( firstchan == ~0 ) { - firstchan = chan_index; - } } } @@ -1353,32 +1598,32 @@ emit_kil( func, x86_make_reg( file_REG32, reg_DX ) ); - FOR_EACH_CHANNEL( chan_index ) { - if( uniquemask & (1 << chan_index) ) { - sse_cmpps( + for (i = 0 ; i < unique_count; i++ ) { + struct x86_reg dataXMM = make_xmm(i); + + sse_cmpps( + func, + dataXMM, + get_temp( + TGSI_EXEC_TEMP_00000000_I, + TGSI_EXEC_TEMP_00000000_C ), + cc_LessThan ); + + if( i == 0 ) { + sse_movmskps( func, - make_xmm( registers[chan_index] ), - get_temp( - TGSI_EXEC_TEMP_00000000_I, - TGSI_EXEC_TEMP_00000000_C ), - cc_LessThan ); - - if( chan_index == firstchan ) { - sse_pmovmskb( - func, - x86_make_reg( file_REG32, reg_AX ), - make_xmm( registers[chan_index] ) ); - } - else { - sse_pmovmskb( - func, - x86_make_reg( file_REG32, reg_DX ), - make_xmm( registers[chan_index] ) ); - x86_or( - func, - x86_make_reg( file_REG32, reg_AX ), - x86_make_reg( file_REG32, reg_DX ) ); - } + x86_make_reg( file_REG32, reg_AX ), + dataXMM ); + } + else { + sse_movmskps( + func, + x86_make_reg( file_REG32, reg_DX ), + dataXMM ); + x86_or( + func, + x86_make_reg( file_REG32, reg_AX ), + x86_make_reg( file_REG32, reg_DX ) ); } } @@ -1581,7 +1826,7 @@ emit_instruction( get_temp( TGSI_EXEC_TEMP_MINUS_128_I, TGSI_EXEC_TEMP_MINUS_128_C ) ); - emit_pow( func, 3, 1, 2 ); + emit_pow( func, 3, 1, 1, 2 ); FETCH( func, *inst, 0, 0, CHAN_X ); sse_xorps( func, @@ -1828,8 +2073,7 @@ emit_instruction( } break; - case TGSI_OPCODE_LERP: - /* TGSI_OPCODE_LRP */ + case TGSI_OPCODE_LRP: FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( func, *inst, 0, 0, chan_index ); FETCH( func, *inst, 1, 1, chan_index ); @@ -1845,12 +2089,7 @@ emit_instruction( return 0; break; - case TGSI_OPCODE_CND0: - return 0; - break; - - case TGSI_OPCODE_DOT2ADD: - /* TGSI_OPCODE_DP2A */ + case TGSI_OPCODE_DP2A: FETCH( func, *inst, 0, 0, CHAN_X ); /* xmm0 = src[0].x */ FETCH( func, *inst, 1, 1, CHAN_X ); /* xmm1 = src[1].x */ emit_mul( func, 0, 1 ); /* xmm0 = xmm0 * xmm1 */ @@ -1865,16 +2104,7 @@ emit_instruction( } break; - case TGSI_OPCODE_INDEX: - return 0; - break; - - case TGSI_OPCODE_NEGATE: - return 0; - break; - - case TGSI_OPCODE_FRAC: - /* TGSI_OPCODE_FRC */ + case TGSI_OPCODE_FRC: FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( func, *inst, 0, 0, chan_index ); emit_frc( func, 0, 0 ); @@ -1886,8 +2116,7 @@ emit_instruction( return 0; break; - case TGSI_OPCODE_FLOOR: - /* TGSI_OPCODE_FLR */ + case TGSI_OPCODE_FLR: FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( func, *inst, 0, 0, chan_index ); emit_flr( func, 0, 0 ); @@ -1903,8 +2132,7 @@ emit_instruction( } break; - case TGSI_OPCODE_EXPBASE2: - /* TGSI_OPCODE_EX2 */ + case TGSI_OPCODE_EX2: FETCH( func, *inst, 0, 0, CHAN_X ); emit_ex2( func, 0, 0 ); FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { @@ -1912,8 +2140,7 @@ emit_instruction( } break; - case TGSI_OPCODE_LOGBASE2: - /* TGSI_OPCODE_LG2 */ + case TGSI_OPCODE_LG2: FETCH( func, *inst, 0, 0, CHAN_X ); emit_lg2( func, 0, 0 ); FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { @@ -1921,18 +2148,16 @@ emit_instruction( } break; - case TGSI_OPCODE_POWER: - /* TGSI_OPCODE_POW */ + case TGSI_OPCODE_POW: FETCH( func, *inst, 0, 0, CHAN_X ); FETCH( func, *inst, 1, 1, CHAN_X ); - emit_pow( func, 0, 0, 1 ); + emit_pow( func, 0, 0, 0, 1 ); FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { STORE( func, *inst, 0, 0, chan_index ); } break; - case TGSI_OPCODE_CROSSPRODUCT: - /* TGSI_OPCODE_XPD */ + case TGSI_OPCODE_XPD: if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) || IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ) { FETCH( func, *inst, 1, 1, CHAN_Z ); @@ -1978,10 +2203,6 @@ emit_instruction( } break; - case TGSI_OPCODE_MULTIPLYMATRIX: - return 0; - break; - case TGSI_OPCODE_ABS: FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( func, *inst, 0, 0, chan_index ); @@ -2094,21 +2315,7 @@ emit_instruction( break; case TGSI_OPCODE_TEX: - if (0) { - /* Disable dummy texture code: - */ - emit_tempf( - func, - 0, - TEMP_ONE_I, - TEMP_ONE_C ); - FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { - STORE( func, *inst, 0, 0, chan_index ); - } - } - else { - return 0; - } + emit_tex( func, inst, FALSE, FALSE ); break; case TGSI_OPCODE_TXD: @@ -2206,7 +2413,7 @@ emit_instruction( break; case TGSI_OPCODE_TXB: - return 0; + emit_tex( func, inst, TRUE, FALSE ); break; case TGSI_OPCODE_NRM: @@ -2314,9 +2521,13 @@ emit_instruction( break; case TGSI_OPCODE_TXL: - return 0; + emit_tex( func, inst, TRUE, FALSE ); break; + case TGSI_OPCODE_TXP: + emit_tex( func, inst, FALSE, TRUE ); + break; + case TGSI_OPCODE_BRK: return 0; break; @@ -2325,7 +2536,7 @@ emit_instruction( return 0; break; - case TGSI_OPCODE_LOOP: + case TGSI_OPCODE_BGNFOR: return 0; break; @@ -2341,7 +2552,7 @@ emit_instruction( return 0; break; - case TGSI_OPCODE_ENDLOOP: + case TGSI_OPCODE_ENDFOR: return 0; break; @@ -2496,7 +2707,7 @@ emit_declaration( static void aos_to_soa( struct x86_function *func, uint arg_aos, - uint arg_soa, + uint arg_machine, uint arg_num, uint arg_stride ) { @@ -2511,7 +2722,10 @@ static void aos_to_soa( struct x86_function *func, x86_push( func, x86_make_reg( file_REG32, reg_BX ) ); x86_mov( func, aos_input, x86_fn_arg( func, arg_aos ) ); - x86_mov( func, soa_input, x86_fn_arg( func, arg_soa ) ); + x86_mov( func, soa_input, x86_fn_arg( func, arg_machine ) ); + x86_lea( func, soa_input, + x86_make_disp( soa_input, + Offset(struct tgsi_exec_machine, Inputs) ) ); x86_mov( func, num_inputs, x86_fn_arg( func, arg_num ) ); x86_mov( func, stride, x86_fn_arg( func, arg_stride ) ); @@ -2553,28 +2767,30 @@ static void aos_to_soa( struct x86_function *func, x86_jcc( func, cc_NE, inner_loop ); /* Restore EBX */ - x86_pop( func, aos_input ); + x86_pop( func, x86_make_reg( file_REG32, reg_BX ) ); } -static void soa_to_aos( struct x86_function *func, uint aos, uint soa, uint num, uint stride ) +static void soa_to_aos( struct x86_function *func, + uint arg_aos, + uint arg_machine, + uint arg_num, + uint arg_stride ) { - struct x86_reg soa_output; - struct x86_reg aos_output; - struct x86_reg num_outputs; - struct x86_reg temp; + struct x86_reg soa_output = x86_make_reg( file_REG32, reg_AX ); + struct x86_reg aos_output = x86_make_reg( file_REG32, reg_BX ); + struct x86_reg num_outputs = x86_make_reg( file_REG32, reg_CX ); + struct x86_reg temp = x86_make_reg( file_REG32, reg_DX ); int inner_loop; - soa_output = x86_make_reg( file_REG32, reg_AX ); - aos_output = x86_make_reg( file_REG32, reg_BX ); - num_outputs = x86_make_reg( file_REG32, reg_CX ); - temp = x86_make_reg( file_REG32, reg_DX ); - /* Save EBX */ - x86_push( func, aos_output ); + x86_push( func, x86_make_reg( file_REG32, reg_BX ) ); - x86_mov( func, soa_output, x86_fn_arg( func, soa ) ); - x86_mov( func, aos_output, x86_fn_arg( func, aos ) ); - x86_mov( func, num_outputs, x86_fn_arg( func, num ) ); + x86_mov( func, aos_output, x86_fn_arg( func, arg_aos ) ); + x86_mov( func, soa_output, x86_fn_arg( func, arg_machine ) ); + x86_lea( func, soa_output, + x86_make_disp( soa_output, + Offset(struct tgsi_exec_machine, Outputs) ) ); + x86_mov( func, num_outputs, x86_fn_arg( func, arg_num ) ); /* do */ inner_loop = x86_get_label( func ); @@ -2591,7 +2807,7 @@ static void soa_to_aos( struct x86_function *func, uint aos, uint soa, uint num, sse_unpcklps( func, make_xmm( 3 ), make_xmm( 4 ) ); sse_unpckhps( func, make_xmm( 5 ), make_xmm( 4 ) ); - x86_mov( func, temp, x86_fn_arg( func, stride ) ); + x86_mov( func, temp, x86_fn_arg( func, arg_stride ) ); x86_push( func, aos_output ); sse_movlps( func, x86_make_disp( aos_output, 0 ), make_xmm( 0 ) ); sse_movlps( func, x86_make_disp( aos_output, 8 ), make_xmm( 3 ) ); @@ -2615,20 +2831,13 @@ static void soa_to_aos( struct x86_function *func, uint aos, uint soa, uint num, x86_jcc( func, cc_NE, inner_loop ); /* Restore EBX */ - x86_pop( func, aos_output ); + x86_pop( func, x86_make_reg( file_REG32, reg_BX ) ); } /** * Translate a TGSI vertex/fragment shader to SSE2 code. * Slightly different things are done for vertex vs. fragment shaders. * - * Note that fragment shaders are responsible for interpolating shader - * inputs. Because on x86 we have only 4 GP registers, and here we - * have 5 shader arguments (input, output, const, temp and coef), the - * code is split into two phases -- DECLARATION and INSTRUCTION phase. - * GP register holding the output argument is aliased with the coeff - * argument, as outputs are not needed in the DECLARATION phase. - * * \param tokens the TGSI input shader * \param func the output SSE code/function * \param immediates buffer to place immediates, later passed to SSE func @@ -2642,7 +2851,6 @@ tgsi_emit_sse2( boolean do_swizzles ) { struct tgsi_parse_context parse; - boolean instruction_phase = FALSE; unsigned ok = 1; uint num_immediates = 0; @@ -2654,74 +2862,48 @@ tgsi_emit_sse2( /* Can't just use EDI, EBX without save/restoring them: */ - x86_push( - func, - get_immediate_base() ); - - x86_push( - func, - get_temp_base() ); - + x86_push( func, x86_make_reg( file_REG32, reg_BX ) ); + x86_push( func, x86_make_reg( file_REG32, reg_DI ) ); /* * Different function args for vertex/fragment shaders: */ - if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_FRAGMENT) { - /* DECLARATION phase, do not load output argument. */ - x86_mov( - func, - get_input_base(), - x86_fn_arg( func, 1 ) ); - /* skipping outputs argument here */ - x86_mov( - func, - get_const_base(), - x86_fn_arg( func, 3 ) ); - x86_mov( - func, - get_temp_base(), - x86_fn_arg( func, 4 ) ); - x86_mov( - func, - get_coef_base(), - x86_fn_arg( func, 5 ) ); - x86_mov( - func, - get_immediate_base(), - x86_fn_arg( func, 6 ) ); - } - else { - assert(parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX); - + if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX) { if (do_swizzles) aos_to_soa( func, - 6, /* aos_input */ - 1, /* machine->input */ - 7, /* num_inputs */ - 8 ); /* input_stride */ + 4, /* aos_input */ + 1, /* machine */ + 5, /* num_inputs */ + 6 ); /* input_stride */ + } + x86_mov( + func, + get_machine_base(), + x86_fn_arg( func, 1 ) ); + x86_mov( + func, + get_const_base(), + x86_fn_arg( func, 2 ) ); + x86_mov( + func, + get_immediate_base(), + x86_fn_arg( func, 3 ) ); + + if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_FRAGMENT) { x86_mov( - func, - get_input_base(), - x86_fn_arg( func, 1 ) ); - x86_mov( - func, - get_output_base(), - x86_fn_arg( func, 2 ) ); - x86_mov( - func, - get_const_base(), - x86_fn_arg( func, 3 ) ); - x86_mov( - func, - get_temp_base(), - x86_fn_arg( func, 4 ) ); - x86_mov( - func, - get_immediate_base(), - x86_fn_arg( func, 5 ) ); + func, + get_coef_base(), + x86_fn_arg( func, 4 ) ); } + x86_mov( + func, + get_sampler_base(), + x86_make_disp( get_machine_base(), + Offset( struct tgsi_exec_machine, Samplers ) ) ); + + while( !tgsi_parse_end_of_tokens( &parse ) && ok ) { tgsi_parse_token( &parse ); @@ -2735,24 +2917,15 @@ tgsi_emit_sse2( break; case TGSI_TOKEN_TYPE_INSTRUCTION: - if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_FRAGMENT) { - if( !instruction_phase ) { - /* INSTRUCTION phase, overwrite coeff with output. */ - instruction_phase = TRUE; - x86_mov( - func, - get_output_base(), - x86_fn_arg( func, 2 ) ); - } - } - ok = emit_instruction( func, &parse.FullToken.FullInstruction ); if (!ok) { - debug_printf("failed to translate tgsi opcode %d to SSE (%s)\n", - parse.FullToken.FullInstruction.Instruction.Opcode, + uint opcode = parse.FullToken.FullInstruction.Instruction.Opcode; + debug_printf("failed to translate tgsi opcode %d (%s) to SSE (%s)\n", + opcode, + tgsi_get_opcode_name(opcode), parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX ? "vertex shader" : "fragment shader"); } @@ -2767,7 +2940,7 @@ tgsi_emit_sse2( assert(num_immediates < TGSI_EXEC_NUM_IMMEDIATES); for( i = 0; i < size; i++ ) { immediates[num_immediates][i] = - parse.FullToken.FullImmediate.u.ImmediateFloat32[i].Float; + parse.FullToken.FullImmediate.u[i].Float; } #if 0 debug_printf("SSE FS immediate[%d] = %f %f %f %f\n", @@ -2789,18 +2962,17 @@ tgsi_emit_sse2( if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX) { if (do_swizzles) - soa_to_aos( func, 9, 2, 10, 11 ); + soa_to_aos( func, + 7, /* aos_output */ + 1, /* machine */ + 8, /* num_outputs */ + 9 ); /* output_stride */ } /* Can't just use EBX, EDI without save/restoring them: */ - x86_pop( - func, - get_temp_base() ); - - x86_pop( - func, - get_immediate_base() ); + x86_pop( func, x86_make_reg( file_REG32, reg_DI ) ); + x86_pop( func, x86_make_reg( file_REG32, reg_BX ) ); emit_ret( func ); diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.h b/src/gallium/auxiliary/tgsi/tgsi_sse2.h index af838b2a25b..d81ee3d00ec 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_sse2.h +++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.h @@ -34,6 +34,7 @@ extern "C" { struct tgsi_token; struct x86_function; +struct tgsi_interp_coef; unsigned tgsi_emit_sse2( @@ -42,6 +43,33 @@ tgsi_emit_sse2( float (*immediates)[4], boolean do_swizzles ); + +/* This is the function prototype generated when do_swizzles is false + * -- effectively for fragment shaders. + */ +typedef void (PIPE_CDECL *tgsi_sse2_fs_function) ( + struct tgsi_exec_machine *machine, /* 1 */ + const float (*constant)[4], /* 2 */ + const float (*immediate)[4], /* 3 */ + const struct tgsi_interp_coef *coef /* 4 */ + ); + + +/* This is the function prototype generated when do_swizzles is true + * -- effectively for vertex shaders. + */ +typedef void (PIPE_CDECL *tgsi_sse2_vs_func) ( + struct tgsi_exec_machine *machine, /* 1 */ + const float (*constant)[4], /* 2 */ + const float (*immediate)[4], /* 3 */ + const float (*aos_input)[4], /* 4 */ + uint num_inputs, /* 5 */ + uint input_stride, /* 6 */ + float (*aos_output)[4], /* 7 */ + uint num_outputs, /* 8 */ + uint output_stride ); /* 9 */ + + #if defined __cplusplus } #endif diff --git a/src/gallium/auxiliary/tgsi/tgsi_text.c b/src/gallium/auxiliary/tgsi/tgsi_text.c index a76bbc91400..d438450b1e4 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_text.c +++ b/src/gallium/auxiliary/tgsi/tgsi_text.c @@ -231,7 +231,8 @@ static const char *file_names[TGSI_FILE_COUNT] = "TEMP", "SAMP", "ADDR", - "IMM" + "IMM", + "LOOP" }; static boolean @@ -789,16 +790,6 @@ match_inst_mnemonic(const char **pcur, if (str_match_no_case(pcur, info->mnemonic)) { return TRUE; } - if (info->alt_mnemonic1) { - if (str_match_no_case(pcur, info->alt_mnemonic1)) { - return TRUE; - } - if (info->alt_mnemonic2) { - if (str_match_no_case(pcur, info->alt_mnemonic2)) { - return TRUE; - } - } - } return FALSE; } @@ -1091,7 +1082,10 @@ static boolean parse_immediate( struct translate_ctx *ctx ) imm = tgsi_default_full_immediate(); imm.Immediate.NrTokens += 4; imm.Immediate.DataType = TGSI_IMM_FLOAT32; - imm.u.Pointer = values; + imm.u[0].Float = values[0]; + imm.u[1].Float = values[1]; + imm.u[2].Float = values[2]; + imm.u[3].Float = values[3]; advance = tgsi_build_full_immediate( &imm, diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.c b/src/gallium/auxiliary/tgsi/tgsi_ureg.c new file mode 100644 index 00000000000..f7096bd8e2c --- /dev/null +++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.c @@ -0,0 +1,938 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE, INC AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "pipe/p_context.h" +#include "pipe/p_state.h" +#include "tgsi/tgsi_ureg.h" +#include "tgsi/tgsi_info.h" +#include "tgsi/tgsi_dump.h" +#include "util/u_memory.h" +#include "util/u_math.h" + +union tgsi_any_token { + struct tgsi_version version; + struct tgsi_header header; + struct tgsi_processor processor; + struct tgsi_token token; + struct tgsi_declaration decl; + struct tgsi_declaration_range decl_range; + struct tgsi_declaration_semantic decl_semantic; + struct tgsi_immediate imm; + union tgsi_immediate_data imm_data; + struct tgsi_instruction insn; + struct tgsi_instruction_ext_nv insn_ext_nv; + struct tgsi_instruction_ext_label insn_ext_label; + struct tgsi_instruction_ext_texture insn_ext_texture; + struct tgsi_instruction_ext_predicate insn_ext_predicate; + struct tgsi_src_register src; + struct tgsi_src_register_ext_swz src_ext_swz; + struct tgsi_src_register_ext_mod src_ext_mod; + struct tgsi_dimension dim; + struct tgsi_dst_register dst; + struct tgsi_dst_register_ext_concode dst_ext_code; + struct tgsi_dst_register_ext_modulate dst_ext_mod; + struct tgsi_dst_register_ext_predicate dst_ext_pred; + unsigned value; +}; + + +struct ureg_tokens { + union tgsi_any_token *tokens; + unsigned size; + unsigned order; + unsigned count; +}; + +#define UREG_MAX_INPUT PIPE_MAX_ATTRIBS +#define UREG_MAX_OUTPUT PIPE_MAX_ATTRIBS +#define UREG_MAX_IMMEDIATE 32 +#define UREG_MAX_TEMP 256 +#define UREG_MAX_ADDR 2 + +#define DOMAIN_DECL 0 +#define DOMAIN_INSN 1 + +struct ureg_program +{ + unsigned processor; + struct pipe_context *pipe; + + struct { + unsigned semantic_name; + unsigned semantic_index; + unsigned interp; + } input[UREG_MAX_INPUT]; + unsigned nr_inputs; + + struct { + unsigned semantic_name; + unsigned semantic_index; + } output[UREG_MAX_OUTPUT]; + unsigned nr_outputs; + + struct { + float v[4]; + unsigned nr; + } immediate[UREG_MAX_IMMEDIATE]; + unsigned nr_immediates; + + struct ureg_src sampler[PIPE_MAX_SAMPLERS]; + unsigned nr_samplers; + + unsigned temps_active[UREG_MAX_TEMP / 32]; + unsigned nr_temps; + + unsigned nr_addrs; + + unsigned nr_constants; + unsigned nr_instructions; + + struct ureg_tokens domain[2]; +}; + +static union tgsi_any_token error_tokens[32]; + +static void tokens_error( struct ureg_tokens *tokens ) +{ + tokens->tokens = error_tokens; + tokens->size = Elements(error_tokens); + tokens->count = 0; +} + + +static void tokens_expand( struct ureg_tokens *tokens, + unsigned count ) +{ + unsigned old_size = tokens->size * sizeof(unsigned); + + if (tokens->tokens == error_tokens) + goto fail; + + while (tokens->count + count > tokens->size) { + tokens->size = (1 << ++tokens->order); + } + + tokens->tokens = REALLOC(tokens->tokens, + old_size, + tokens->size * sizeof(unsigned)); + if (tokens->tokens == NULL) + goto fail; + + return; + +fail: + tokens_error(tokens); +} + +static void set_bad( struct ureg_program *ureg ) +{ + tokens_error(&ureg->domain[0]); +} + + + +static union tgsi_any_token *get_tokens( struct ureg_program *ureg, + unsigned domain, + unsigned count ) +{ + struct ureg_tokens *tokens = &ureg->domain[domain]; + union tgsi_any_token *result; + + if (tokens->count + count > tokens->size) + tokens_expand(tokens, count); + + result = &tokens->tokens[tokens->count]; + tokens->count += count; + return result; +} + + +static union tgsi_any_token *retrieve_token( struct ureg_program *ureg, + unsigned domain, + unsigned nr ) +{ + if (ureg->domain[domain].tokens == error_tokens) + return &error_tokens[0]; + + return &ureg->domain[domain].tokens[nr]; +} + + + +static INLINE struct ureg_dst +ureg_dst_register( unsigned file, + unsigned index ) +{ + struct ureg_dst dst; + + dst.File = file; + dst.WriteMask = TGSI_WRITEMASK_XYZW; + dst.Indirect = 0; + dst.IndirectIndex = 0; + dst.IndirectSwizzle = 0; + dst.Saturate = 0; + dst.Index = index; + dst.Pad1 = 0; + dst.Pad2 = 0; + + return dst; +} + +static INLINE struct ureg_src +ureg_src_register( unsigned file, + unsigned index ) +{ + struct ureg_src src; + + src.File = file; + src.SwizzleX = TGSI_SWIZZLE_X; + src.SwizzleY = TGSI_SWIZZLE_Y; + src.SwizzleZ = TGSI_SWIZZLE_Z; + src.SwizzleW = TGSI_SWIZZLE_W; + src.Pad = 0; + src.Indirect = 0; + src.IndirectIndex = 0; + src.IndirectSwizzle = 0; + src.Absolute = 0; + src.Index = index; + src.Negate = 0; + + return src; +} + + + + +static struct ureg_src +ureg_DECL_input( struct ureg_program *ureg, + unsigned name, + unsigned index, + unsigned interp_mode ) +{ + unsigned i; + + for (i = 0; i < ureg->nr_inputs; i++) { + if (ureg->input[i].semantic_name == name && + ureg->input[i].semantic_index == index) + goto out; + } + + if (ureg->nr_inputs < UREG_MAX_INPUT) { + ureg->input[i].semantic_name = name; + ureg->input[i].semantic_index = index; + ureg->input[i].interp = interp_mode; + ureg->nr_inputs++; + } + else { + set_bad( ureg ); + } + +out: + return ureg_src_register( TGSI_FILE_INPUT, i ); +} + + + +struct ureg_src +ureg_DECL_fs_input( struct ureg_program *ureg, + unsigned name, + unsigned index, + unsigned interp ) +{ + assert(ureg->processor == TGSI_PROCESSOR_FRAGMENT); + return ureg_DECL_input( ureg, name, index, interp ); +} + + +struct ureg_src +ureg_DECL_vs_input( struct ureg_program *ureg, + unsigned name, + unsigned index ) +{ + assert(ureg->processor == TGSI_PROCESSOR_VERTEX); + return ureg_DECL_input( ureg, name, index, TGSI_INTERPOLATE_CONSTANT ); +} + + +struct ureg_dst +ureg_DECL_output( struct ureg_program *ureg, + unsigned name, + unsigned index ) +{ + unsigned i; + + for (i = 0; i < ureg->nr_outputs; i++) { + if (ureg->output[i].semantic_name == name && + ureg->output[i].semantic_index == index) + goto out; + } + + if (ureg->nr_outputs < UREG_MAX_OUTPUT) { + ureg->output[i].semantic_name = name; + ureg->output[i].semantic_index = index; + ureg->nr_outputs++; + } + else { + set_bad( ureg ); + } + +out: + return ureg_dst_register( TGSI_FILE_OUTPUT, i ); +} + + +/* Returns a new constant register. Keep track of which have been + * referred to so that we can emit decls later. + * + * There is nothing in this code to bind this constant to any tracked + * value or manage any constant_buffer contents -- that's the + * resposibility of the calling code. + */ +struct ureg_src ureg_DECL_constant(struct ureg_program *ureg ) +{ + return ureg_src_register( TGSI_FILE_CONSTANT, ureg->nr_constants++ ); +} + + +/* Allocate a new temporary. Temporaries greater than UREG_MAX_TEMP + * are legal, but will not be released. + */ +struct ureg_dst ureg_DECL_temporary( struct ureg_program *ureg ) +{ + unsigned i; + + for (i = 0; i < UREG_MAX_TEMP; i += 32) { + int bit = ffs(~ureg->temps_active[i/32]); + if (bit != 0) { + i += bit - 1; + goto out; + } + } + + /* No reusable temps, so allocate a new one: + */ + i = ureg->nr_temps++; + +out: + if (i < UREG_MAX_TEMP) + ureg->temps_active[i/32] |= 1 << (i % 32); + + if (i >= ureg->nr_temps) + ureg->nr_temps = i + 1; + + return ureg_dst_register( TGSI_FILE_TEMPORARY, i ); +} + + +void ureg_release_temporary( struct ureg_program *ureg, + struct ureg_dst tmp ) +{ + if(tmp.File == TGSI_FILE_TEMPORARY) + if (tmp.Index < UREG_MAX_TEMP) + ureg->temps_active[tmp.Index/32] &= ~(1 << (tmp.Index % 32)); +} + + +/* Allocate a new address register. + */ +struct ureg_dst ureg_DECL_address( struct ureg_program *ureg ) +{ + if (ureg->nr_addrs < UREG_MAX_ADDR) + return ureg_dst_register( TGSI_FILE_ADDRESS, ureg->nr_addrs++ ); + + assert( 0 ); + return ureg_dst_register( TGSI_FILE_ADDRESS, 0 ); +} + +/* Allocate a new sampler. + */ +struct ureg_src ureg_DECL_sampler( struct ureg_program *ureg, + unsigned nr ) +{ + unsigned i; + + for (i = 0; i < ureg->nr_samplers; i++) + if (ureg->sampler[i].Index == nr) + return ureg->sampler[i]; + + if (i < PIPE_MAX_SAMPLERS) { + ureg->sampler[i] = ureg_src_register( TGSI_FILE_SAMPLER, nr ); + ureg->nr_samplers++; + return ureg->sampler[i]; + } + + assert( 0 ); + return ureg->sampler[0]; +} + + + + +static int match_or_expand_immediate( const float *v, + unsigned nr, + float *v2, + unsigned *nr2, + unsigned *swizzle ) +{ + unsigned i, j; + + *swizzle = 0; + + for (i = 0; i < nr; i++) { + boolean found = FALSE; + + for (j = 0; j < *nr2 && !found; j++) { + if (v[i] == v2[j]) { + *swizzle |= j << (i * 2); + found = TRUE; + } + } + + if (!found) { + if (*nr2 >= 4) + return FALSE; + + v2[*nr2] = v[i]; + *swizzle |= *nr2 << (i * 2); + (*nr2)++; + } + } + + return TRUE; +} + + + + +struct ureg_src ureg_DECL_immediate( struct ureg_program *ureg, + const float *v, + unsigned nr ) +{ + unsigned i, j; + unsigned swizzle; + + /* Could do a first pass where we examine all existing immediates + * without expanding. + */ + + for (i = 0; i < ureg->nr_immediates; i++) { + if (match_or_expand_immediate( v, + nr, + ureg->immediate[i].v, + &ureg->immediate[i].nr, + &swizzle )) + goto out; + } + + if (ureg->nr_immediates < UREG_MAX_IMMEDIATE) { + i = ureg->nr_immediates++; + if (match_or_expand_immediate( v, + nr, + ureg->immediate[i].v, + &ureg->immediate[i].nr, + &swizzle )) + goto out; + } + + set_bad( ureg ); + +out: + /* Make sure that all referenced elements are from this immediate. + * Has the effect of making size-one immediates into scalars. + */ + for (j = nr; j < 4; j++) + swizzle |= (swizzle & 0x3) << (j * 2); + + return ureg_swizzle( ureg_src_register( TGSI_FILE_IMMEDIATE, i ), + (swizzle >> 0) & 0x3, + (swizzle >> 2) & 0x3, + (swizzle >> 4) & 0x3, + (swizzle >> 6) & 0x3); +} + + +void +ureg_emit_src( struct ureg_program *ureg, + struct ureg_src src ) +{ + unsigned size = (1 + + (src.Absolute ? 1 : 0) + + (src.Indirect ? 1 : 0)); + + union tgsi_any_token *out = get_tokens( ureg, DOMAIN_INSN, size ); + unsigned n = 0; + + assert(src.File != TGSI_FILE_NULL); + assert(src.File != TGSI_FILE_OUTPUT); + assert(src.File < TGSI_FILE_COUNT); + + out[n].value = 0; + out[n].src.File = src.File; + out[n].src.SwizzleX = src.SwizzleX; + out[n].src.SwizzleY = src.SwizzleY; + out[n].src.SwizzleZ = src.SwizzleZ; + out[n].src.SwizzleW = src.SwizzleW; + out[n].src.Index = src.Index; + out[n].src.Negate = src.Negate; + n++; + + if (src.Absolute) { + out[0].src.Extended = 1; + out[0].src.Negate = 0; + out[n].value = 0; + out[n].src_ext_mod.Type = TGSI_SRC_REGISTER_EXT_TYPE_MOD; + out[n].src_ext_mod.Absolute = 1; + out[n].src_ext_mod.Negate = src.Negate; + n++; + } + + if (src.Indirect) { + out[0].src.Indirect = 1; + out[n].value = 0; + out[n].src.File = TGSI_FILE_ADDRESS; + out[n].src.SwizzleX = src.IndirectSwizzle; + out[n].src.SwizzleY = src.IndirectSwizzle; + out[n].src.SwizzleZ = src.IndirectSwizzle; + out[n].src.SwizzleW = src.IndirectSwizzle; + out[n].src.Index = src.IndirectIndex; + n++; + } + + assert(n == size); +} + + +void +ureg_emit_dst( struct ureg_program *ureg, + struct ureg_dst dst ) +{ + unsigned size = (1 + + (dst.Indirect ? 1 : 0)); + + union tgsi_any_token *out = get_tokens( ureg, DOMAIN_INSN, size ); + unsigned n = 0; + + assert(dst.File != TGSI_FILE_NULL); + assert(dst.File != TGSI_FILE_CONSTANT); + assert(dst.File != TGSI_FILE_INPUT); + assert(dst.File != TGSI_FILE_SAMPLER); + assert(dst.File != TGSI_FILE_IMMEDIATE); + assert(dst.File < TGSI_FILE_COUNT); + + out[n].value = 0; + out[n].dst.File = dst.File; + out[n].dst.WriteMask = dst.WriteMask; + out[n].dst.Indirect = dst.Indirect; + out[n].dst.Index = dst.Index; + n++; + + if (dst.Indirect) { + out[n].value = 0; + out[n].src.File = TGSI_FILE_ADDRESS; + out[n].src.SwizzleX = dst.IndirectSwizzle; + out[n].src.SwizzleY = dst.IndirectSwizzle; + out[n].src.SwizzleZ = dst.IndirectSwizzle; + out[n].src.SwizzleW = dst.IndirectSwizzle; + out[n].src.Index = dst.IndirectIndex; + n++; + } + + assert(n == size); +} + + + +unsigned +ureg_emit_insn(struct ureg_program *ureg, + unsigned opcode, + boolean saturate, + unsigned num_dst, + unsigned num_src ) +{ + union tgsi_any_token *out; + + out = get_tokens( ureg, DOMAIN_INSN, 1 ); + out[0].value = 0; + out[0].insn.Type = TGSI_TOKEN_TYPE_INSTRUCTION; + out[0].insn.NrTokens = 0; + out[0].insn.Opcode = opcode; + out[0].insn.Saturate = saturate; + out[0].insn.NumDstRegs = num_dst; + out[0].insn.NumSrcRegs = num_src; + out[0].insn.Padding = 0; + out[0].insn.Extended = 0; + + ureg->nr_instructions++; + + return ureg->domain[DOMAIN_INSN].count - 1; +} + + +void +ureg_emit_label(struct ureg_program *ureg, + unsigned insn_token, + unsigned *label_token ) +{ + union tgsi_any_token *out, *insn; + + if(!label_token) + return; + + out = get_tokens( ureg, DOMAIN_INSN, 1 ); + insn = retrieve_token( ureg, DOMAIN_INSN, insn_token ); + + insn->insn.Extended = 1; + + out[0].value = 0; + out[0].insn_ext_label.Type = TGSI_INSTRUCTION_EXT_TYPE_LABEL; + + *label_token = ureg->domain[DOMAIN_INSN].count - 1; +} + +/* Will return a number which can be used in a label to point to the + * next instruction to be emitted. + */ +unsigned +ureg_get_instruction_number( struct ureg_program *ureg ) +{ + return ureg->nr_instructions; +} + +/* Patch a given label (expressed as a token number) to point to a + * given instruction (expressed as an instruction number). + */ +void +ureg_fixup_label(struct ureg_program *ureg, + unsigned label_token, + unsigned instruction_number ) +{ + union tgsi_any_token *out = retrieve_token( ureg, DOMAIN_INSN, label_token ); + + assert(out->insn_ext_label.Type == TGSI_INSTRUCTION_EXT_TYPE_LABEL); + out->insn_ext_label.Label = instruction_number; +} + + +void +ureg_emit_texture(struct ureg_program *ureg, + unsigned insn_token, + unsigned target ) +{ + union tgsi_any_token *out, *insn; + + out = get_tokens( ureg, DOMAIN_INSN, 1 ); + insn = retrieve_token( ureg, DOMAIN_INSN, insn_token ); + + insn->insn.Extended = 1; + + out[0].value = 0; + out[0].insn_ext_texture.Type = TGSI_INSTRUCTION_EXT_TYPE_TEXTURE; + out[0].insn_ext_texture.Texture = target; +} + + +void +ureg_fixup_insn_size(struct ureg_program *ureg, + unsigned insn ) +{ + union tgsi_any_token *out = retrieve_token( ureg, DOMAIN_INSN, insn ); + + assert(out->insn.Type == TGSI_TOKEN_TYPE_INSTRUCTION); + out->insn.NrTokens = ureg->domain[DOMAIN_INSN].count - insn - 1; +} + + +void +ureg_insn(struct ureg_program *ureg, + unsigned opcode, + const struct ureg_dst *dst, + unsigned nr_dst, + const struct ureg_src *src, + unsigned nr_src ) +{ + unsigned insn, i; + boolean saturate; + +#ifdef DEBUG + { + const struct tgsi_opcode_info *info = tgsi_get_opcode_info( opcode ); + assert(info); + if(info) { + assert(nr_dst == info->num_dst); + assert(nr_src == info->num_src); + } + } +#endif + + saturate = nr_dst ? dst[0].Saturate : FALSE; + + insn = ureg_emit_insn( ureg, opcode, saturate, nr_dst, nr_src ); + + for (i = 0; i < nr_dst; i++) + ureg_emit_dst( ureg, dst[i] ); + + for (i = 0; i < nr_src; i++) + ureg_emit_src( ureg, src[i] ); + + ureg_fixup_insn_size( ureg, insn ); +} + + + +static void emit_decl( struct ureg_program *ureg, + unsigned file, + unsigned index, + unsigned semantic_name, + unsigned semantic_index, + unsigned interp ) +{ + union tgsi_any_token *out = get_tokens( ureg, DOMAIN_DECL, 3 ); + + out[0].value = 0; + out[0].decl.Type = TGSI_TOKEN_TYPE_DECLARATION; + out[0].decl.NrTokens = 3; + out[0].decl.File = file; + out[0].decl.UsageMask = TGSI_WRITEMASK_XYZW; /* FIXME! */ + out[0].decl.Interpolate = interp; + out[0].decl.Semantic = 1; + + out[1].value = 0; + out[1].decl_range.First = + out[1].decl_range.Last = index; + + out[2].value = 0; + out[2].decl_semantic.SemanticName = semantic_name; + out[2].decl_semantic.SemanticIndex = semantic_index; + +} + + +static void emit_decl_range( struct ureg_program *ureg, + unsigned file, + unsigned first, + unsigned count ) +{ + union tgsi_any_token *out = get_tokens( ureg, DOMAIN_DECL, 2 ); + + out[0].value = 0; + out[0].decl.Type = TGSI_TOKEN_TYPE_DECLARATION; + out[0].decl.NrTokens = 2; + out[0].decl.File = file; + out[0].decl.UsageMask = 0xf; + out[0].decl.Interpolate = TGSI_INTERPOLATE_CONSTANT; + out[0].decl.Semantic = 0; + + out[1].value = 0; + out[1].decl_range.First = first; + out[1].decl_range.Last = first + count - 1; +} + +static void emit_immediate( struct ureg_program *ureg, + const float *v ) +{ + union tgsi_any_token *out = get_tokens( ureg, DOMAIN_DECL, 5 ); + + out[0].value = 0; + out[0].imm.Type = TGSI_TOKEN_TYPE_IMMEDIATE; + out[0].imm.NrTokens = 5; + out[0].imm.DataType = TGSI_IMM_FLOAT32; + out[0].imm.Padding = 0; + out[0].imm.Extended = 0; + + out[1].imm_data.Float = v[0]; + out[2].imm_data.Float = v[1]; + out[3].imm_data.Float = v[2]; + out[4].imm_data.Float = v[3]; +} + + + + +static void emit_decls( struct ureg_program *ureg ) +{ + unsigned i; + + for (i = 0; i < ureg->nr_inputs; i++) { + emit_decl( ureg, + TGSI_FILE_INPUT, + i, + ureg->input[i].semantic_name, + ureg->input[i].semantic_index, + ureg->input[i].interp ); + } + + for (i = 0; i < ureg->nr_outputs; i++) { + emit_decl( ureg, + TGSI_FILE_OUTPUT, + i, + ureg->output[i].semantic_name, + ureg->output[i].semantic_index, + TGSI_INTERPOLATE_CONSTANT ); + } + + for (i = 0; i < ureg->nr_samplers; i++) { + emit_decl_range( ureg, + TGSI_FILE_SAMPLER, + ureg->sampler[i].Index, 1 ); + } + + if (ureg->nr_constants) { + emit_decl_range( ureg, + TGSI_FILE_CONSTANT, + 0, ureg->nr_constants ); + } + + if (ureg->nr_temps) { + emit_decl_range( ureg, + TGSI_FILE_TEMPORARY, + 0, ureg->nr_temps ); + } + + if (ureg->nr_addrs) { + emit_decl_range( ureg, + TGSI_FILE_ADDRESS, + 0, ureg->nr_addrs ); + } + + for (i = 0; i < ureg->nr_immediates; i++) { + emit_immediate( ureg, + ureg->immediate[i].v ); + } +} + +/* Append the instruction tokens onto the declarations to build a + * contiguous stream suitable to send to the driver. + */ +static void copy_instructions( struct ureg_program *ureg ) +{ + unsigned nr_tokens = ureg->domain[DOMAIN_INSN].count; + union tgsi_any_token *out = get_tokens( ureg, + DOMAIN_DECL, + nr_tokens ); + + memcpy(out, + ureg->domain[DOMAIN_INSN].tokens, + nr_tokens * sizeof out[0] ); +} + + +static void +fixup_header_size(struct ureg_program *ureg) +{ + union tgsi_any_token *out = retrieve_token( ureg, DOMAIN_DECL, 1 ); + + out->header.BodySize = ureg->domain[DOMAIN_DECL].count - 3; +} + + +static void +emit_header( struct ureg_program *ureg ) +{ + union tgsi_any_token *out = get_tokens( ureg, DOMAIN_DECL, 3 ); + + out[0].version.MajorVersion = 1; + out[0].version.MinorVersion = 1; + out[0].version.Padding = 0; + + out[1].header.HeaderSize = 2; + out[1].header.BodySize = 0; + + out[2].processor.Processor = ureg->processor; + out[2].processor.Padding = 0; +} + + +const struct tgsi_token *ureg_finalize( struct ureg_program *ureg ) +{ + const struct tgsi_token *tokens; + + emit_header( ureg ); + emit_decls( ureg ); + copy_instructions( ureg ); + fixup_header_size( ureg ); + + if (ureg->domain[0].tokens == error_tokens || + ureg->domain[1].tokens == error_tokens) { + debug_printf("%s: error in generated shader\n", __FUNCTION__); + assert(0); + return NULL; + } + + tokens = &ureg->domain[DOMAIN_DECL].tokens[0].token; + + if (0) { + debug_printf("%s: emitted shader %d tokens:\n", __FUNCTION__, + ureg->domain[DOMAIN_DECL].count); + tgsi_dump( tokens, 0 ); + } + + return tokens; +} + + +void *ureg_create_shader( struct ureg_program *ureg, + struct pipe_context *pipe ) +{ + struct pipe_shader_state state; + + state.tokens = ureg_finalize(ureg); + if(!state.tokens) + return NULL; + + if (ureg->processor == TGSI_PROCESSOR_VERTEX) + return pipe->create_vs_state( pipe, &state ); + else + return pipe->create_fs_state( pipe, &state ); +} + + + + +struct ureg_program *ureg_create( unsigned processor ) +{ + struct ureg_program *ureg = CALLOC_STRUCT( ureg_program ); + if (ureg == NULL) + return NULL; + + ureg->processor = processor; + return ureg; +} + + +void ureg_destroy( struct ureg_program *ureg ) +{ + unsigned i; + + for (i = 0; i < Elements(ureg->domain); i++) { + if (ureg->domain[i].tokens && + ureg->domain[i].tokens != error_tokens) + FREE(ureg->domain[i].tokens); + } + + FREE(ureg); +} diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.h b/src/gallium/auxiliary/tgsi/tgsi_ureg.h new file mode 100644 index 00000000000..acbca59040c --- /dev/null +++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.h @@ -0,0 +1,590 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE, INC AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef TGSI_UREG_H +#define TGSI_UREG_H + +#include "pipe/p_compiler.h" +#include "pipe/p_shader_tokens.h" + +#ifdef __cplusplus +extern "C" { +#endif + +struct ureg_program; + +/* Almost a tgsi_src_register, but we need to pull in the Absolute + * flag from the _ext token. Indirect flag always implies ADDR[0]. + */ +struct ureg_src +{ + unsigned File : 4; /* TGSI_FILE_ */ + unsigned SwizzleX : 2; /* TGSI_SWIZZLE_ */ + unsigned SwizzleY : 2; /* TGSI_SWIZZLE_ */ + unsigned SwizzleZ : 2; /* TGSI_SWIZZLE_ */ + unsigned SwizzleW : 2; /* TGSI_SWIZZLE_ */ + unsigned Pad : 1; /* BOOL */ + unsigned Indirect : 1; /* BOOL */ + unsigned Absolute : 1; /* BOOL */ + int Index : 16; /* SINT */ + unsigned Negate : 1; /* BOOL */ + int IndirectIndex : 16; /* SINT */ + int IndirectSwizzle : 2; /* TGSI_SWIZZLE_ */ +}; + +/* Very similar to a tgsi_dst_register, removing unsupported fields + * and adding a Saturate flag. It's easier to push saturate into the + * destination register than to try and create a _SAT varient of each + * instruction function. + */ +struct ureg_dst +{ + unsigned File : 4; /* TGSI_FILE_ */ + unsigned WriteMask : 4; /* TGSI_WRITEMASK_ */ + unsigned Indirect : 1; /* BOOL */ + unsigned Saturate : 1; /* BOOL */ + int Index : 16; /* SINT */ + unsigned Pad1 : 5; + unsigned Pad2 : 1; /* BOOL */ + int IndirectIndex : 16; /* SINT */ + int IndirectSwizzle : 2; /* TGSI_SWIZZLE_ */ +}; + +struct pipe_context; + +struct ureg_program * +ureg_create( unsigned processor ); + +const struct tgsi_token * +ureg_finalize( struct ureg_program * ); + +void * +ureg_create_shader( struct ureg_program *, + struct pipe_context *pipe ); + +void +ureg_destroy( struct ureg_program * ); + + +/*********************************************************************** + * Convenience routine: + */ +static INLINE void * +ureg_create_shader_and_destroy( struct ureg_program *p, + struct pipe_context *pipe ) +{ + void *result = ureg_create_shader( p, pipe ); + ureg_destroy( p ); + return result; +} + + + +/*********************************************************************** + * Build shader declarations: + */ + +struct ureg_src +ureg_DECL_fs_input( struct ureg_program *, + unsigned semantic_name, + unsigned semantic_index, + unsigned interp_mode ); + +struct ureg_src +ureg_DECL_vs_input( struct ureg_program *, + unsigned semantic_name, + unsigned semantic_index ); + +struct ureg_dst +ureg_DECL_output( struct ureg_program *, + unsigned semantic_name, + unsigned semantic_index ); + +struct ureg_src +ureg_DECL_immediate( struct ureg_program *, + const float *v, + unsigned nr ); + +struct ureg_src +ureg_DECL_constant( struct ureg_program * ); + +struct ureg_dst +ureg_DECL_temporary( struct ureg_program * ); + +void +ureg_release_temporary( struct ureg_program *ureg, + struct ureg_dst tmp ); + +struct ureg_dst +ureg_DECL_address( struct ureg_program * ); + +/* Supply an index to the sampler declaration as this is the hook to + * the external pipe_sampler state. Users of this function probably + * don't want just any sampler, but a specific one which they've set + * up state for in the context. + */ +struct ureg_src +ureg_DECL_sampler( struct ureg_program *, + unsigned index ); + + +static INLINE struct ureg_src +ureg_imm4f( struct ureg_program *ureg, + float a, float b, + float c, float d) +{ + float v[4]; + v[0] = a; + v[1] = b; + v[2] = c; + v[3] = d; + return ureg_DECL_immediate( ureg, v, 4 ); +} + +static INLINE struct ureg_src +ureg_imm3f( struct ureg_program *ureg, + float a, float b, + float c) +{ + float v[3]; + v[0] = a; + v[1] = b; + v[2] = c; + return ureg_DECL_immediate( ureg, v, 3 ); +} + +static INLINE struct ureg_src +ureg_imm2f( struct ureg_program *ureg, + float a, float b) +{ + float v[2]; + v[0] = a; + v[1] = b; + return ureg_DECL_immediate( ureg, v, 2 ); +} + +static INLINE struct ureg_src +ureg_imm1f( struct ureg_program *ureg, + float a) +{ + float v[1]; + v[0] = a; + return ureg_DECL_immediate( ureg, v, 1 ); +} + +/*********************************************************************** + * Functions for patching up labels + */ + + +/* Will return a number which can be used in a label to point to the + * next instruction to be emitted. + */ +unsigned +ureg_get_instruction_number( struct ureg_program *ureg ); + + +/* Patch a given label (expressed as a token number) to point to a + * given instruction (expressed as an instruction number). + * + * Labels are obtained from instruction emitters, eg ureg_CAL(). + * Instruction numbers are obtained from ureg_get_instruction_number(), + * above. + */ +void +ureg_fixup_label(struct ureg_program *ureg, + unsigned label_token, + unsigned instruction_number ); + + +/* Generic instruction emitter. Use if you need to pass the opcode as + * a parameter, rather than using the emit_OP() varients below. + */ +void +ureg_insn(struct ureg_program *ureg, + unsigned opcode, + const struct ureg_dst *dst, + unsigned nr_dst, + const struct ureg_src *src, + unsigned nr_src ); + + +/*********************************************************************** + * Internal instruction helpers, don't call these directly: + */ + +unsigned +ureg_emit_insn(struct ureg_program *ureg, + unsigned opcode, + boolean saturate, + unsigned num_dst, + unsigned num_src ); + +void +ureg_emit_label(struct ureg_program *ureg, + unsigned insn_token, + unsigned *label_token ); + +void +ureg_emit_texture(struct ureg_program *ureg, + unsigned insn_token, + unsigned target ); + +void +ureg_emit_dst( struct ureg_program *ureg, + struct ureg_dst dst ); + +void +ureg_emit_src( struct ureg_program *ureg, + struct ureg_src src ); + +void +ureg_fixup_insn_size(struct ureg_program *ureg, + unsigned insn ); + + +#define OP00( op ) \ +static INLINE void ureg_##op( struct ureg_program *ureg ) \ +{ \ + unsigned opcode = TGSI_OPCODE_##op; \ + unsigned insn = ureg_emit_insn( ureg, opcode, FALSE, 0, 0 ); \ + ureg_fixup_insn_size( ureg, insn ); \ +} + +#define OP01( op ) \ +static INLINE void ureg_##op( struct ureg_program *ureg, \ + struct ureg_src src ) \ +{ \ + unsigned opcode = TGSI_OPCODE_##op; \ + unsigned insn = ureg_emit_insn( ureg, opcode, FALSE, 0, 1 ); \ + ureg_emit_src( ureg, src ); \ + ureg_fixup_insn_size( ureg, insn ); \ +} + +#define OP00_LBL( op ) \ +static INLINE void ureg_##op( struct ureg_program *ureg, \ + unsigned *label_token ) \ +{ \ + unsigned opcode = TGSI_OPCODE_##op; \ + unsigned insn = ureg_emit_insn( ureg, opcode, FALSE, 0, 0 ); \ + ureg_emit_label( ureg, insn, label_token ); \ + ureg_fixup_insn_size( ureg, insn ); \ +} + +#define OP01_LBL( op ) \ +static INLINE void ureg_##op( struct ureg_program *ureg, \ + struct ureg_src src, \ + unsigned *label_token ) \ +{ \ + unsigned opcode = TGSI_OPCODE_##op; \ + unsigned insn = ureg_emit_insn( ureg, opcode, FALSE, 0, 1 ); \ + ureg_emit_label( ureg, insn, label_token ); \ + ureg_emit_src( ureg, src ); \ + ureg_fixup_insn_size( ureg, insn ); \ +} + +#define OP10( op ) \ +static INLINE void ureg_##op( struct ureg_program *ureg, \ + struct ureg_dst dst ) \ +{ \ + unsigned opcode = TGSI_OPCODE_##op; \ + unsigned insn = ureg_emit_insn( ureg, opcode, dst.Saturate, 1, 0 ); \ + ureg_emit_dst( ureg, dst ); \ + ureg_fixup_insn_size( ureg, insn ); \ +} + + +#define OP11( op ) \ +static INLINE void ureg_##op( struct ureg_program *ureg, \ + struct ureg_dst dst, \ + struct ureg_src src ) \ +{ \ + unsigned opcode = TGSI_OPCODE_##op; \ + unsigned insn = ureg_emit_insn( ureg, opcode, dst.Saturate, 1, 1 ); \ + ureg_emit_dst( ureg, dst ); \ + ureg_emit_src( ureg, src ); \ + ureg_fixup_insn_size( ureg, insn ); \ +} + +#define OP12( op ) \ +static INLINE void ureg_##op( struct ureg_program *ureg, \ + struct ureg_dst dst, \ + struct ureg_src src0, \ + struct ureg_src src1 ) \ +{ \ + unsigned opcode = TGSI_OPCODE_##op; \ + unsigned insn = ureg_emit_insn( ureg, opcode, dst.Saturate, 1, 2 ); \ + ureg_emit_dst( ureg, dst ); \ + ureg_emit_src( ureg, src0 ); \ + ureg_emit_src( ureg, src1 ); \ + ureg_fixup_insn_size( ureg, insn ); \ +} + +#define OP12_TEX( op ) \ +static INLINE void ureg_##op( struct ureg_program *ureg, \ + struct ureg_dst dst, \ + unsigned target, \ + struct ureg_src src0, \ + struct ureg_src src1 ) \ +{ \ + unsigned opcode = TGSI_OPCODE_##op; \ + unsigned insn = ureg_emit_insn( ureg, opcode, dst.Saturate, 1, 2 ); \ + ureg_emit_texture( ureg, insn, target ); \ + ureg_emit_dst( ureg, dst ); \ + ureg_emit_src( ureg, src0 ); \ + ureg_emit_src( ureg, src1 ); \ + ureg_fixup_insn_size( ureg, insn ); \ +} + +#define OP13( op ) \ +static INLINE void ureg_##op( struct ureg_program *ureg, \ + struct ureg_dst dst, \ + struct ureg_src src0, \ + struct ureg_src src1, \ + struct ureg_src src2 ) \ +{ \ + unsigned opcode = TGSI_OPCODE_##op; \ + unsigned insn = ureg_emit_insn( ureg, opcode, dst.Saturate, 1, 3 ); \ + ureg_emit_dst( ureg, dst ); \ + ureg_emit_src( ureg, src0 ); \ + ureg_emit_src( ureg, src1 ); \ + ureg_emit_src( ureg, src2 ); \ + ureg_fixup_insn_size( ureg, insn ); \ +} + +#define OP14_TEX( op ) \ +static INLINE void ureg_##op( struct ureg_program *ureg, \ + struct ureg_dst dst, \ + unsigned target, \ + struct ureg_src src0, \ + struct ureg_src src1, \ + struct ureg_src src2, \ + struct ureg_src src3 ) \ +{ \ + unsigned opcode = TGSI_OPCODE_##op; \ + unsigned insn = ureg_emit_insn( ureg, opcode, dst.Saturate, 1, 4 ); \ + ureg_emit_texture( ureg, insn, target ); \ + ureg_emit_dst( ureg, dst ); \ + ureg_emit_src( ureg, src0 ); \ + ureg_emit_src( ureg, src1 ); \ + ureg_emit_src( ureg, src2 ); \ + ureg_emit_src( ureg, src3 ); \ + ureg_fixup_insn_size( ureg, insn ); \ +} + + +/* Use a template include to generate a correctly-typed ureg_OP() + * function for each TGSI opcode: + */ +#include "tgsi_opcode_tmp.h" + + +/*********************************************************************** + * Inline helpers for manipulating register structs: + */ +static INLINE struct ureg_src +ureg_negate( struct ureg_src reg ) +{ + assert(reg.File != TGSI_FILE_NULL); + reg.Negate ^= 1; + return reg; +} + +static INLINE struct ureg_src +ureg_abs( struct ureg_src reg ) +{ + assert(reg.File != TGSI_FILE_NULL); + reg.Absolute = 1; + reg.Negate = 0; + return reg; +} + +static INLINE struct ureg_src +ureg_swizzle( struct ureg_src reg, + int x, int y, int z, int w ) +{ + unsigned swz = ( (reg.SwizzleX << 0) | + (reg.SwizzleY << 2) | + (reg.SwizzleZ << 4) | + (reg.SwizzleW << 6)); + + assert(reg.File != TGSI_FILE_NULL); + assert(x < 4); + assert(y < 4); + assert(z < 4); + assert(w < 4); + + reg.SwizzleX = (swz >> (x*2)) & 0x3; + reg.SwizzleY = (swz >> (y*2)) & 0x3; + reg.SwizzleZ = (swz >> (z*2)) & 0x3; + reg.SwizzleW = (swz >> (w*2)) & 0x3; + return reg; +} + +static INLINE struct ureg_src +ureg_scalar( struct ureg_src reg, int x ) +{ + return ureg_swizzle(reg, x, x, x, x); +} + +static INLINE struct ureg_dst +ureg_writemask( struct ureg_dst reg, + unsigned writemask ) +{ + assert(reg.File != TGSI_FILE_NULL); + reg.WriteMask &= writemask; + return reg; +} + +static INLINE struct ureg_dst +ureg_saturate( struct ureg_dst reg ) +{ + assert(reg.File != TGSI_FILE_NULL); + reg.Saturate = 1; + return reg; +} + +static INLINE struct ureg_dst +ureg_dst_indirect( struct ureg_dst reg, struct ureg_src addr ) +{ + assert(reg.File != TGSI_FILE_NULL); + assert(addr.File == TGSI_FILE_ADDRESS); + reg.Indirect = 1; + reg.IndirectIndex = addr.Index; + reg.IndirectSwizzle = addr.SwizzleX; + return reg; +} + +static INLINE struct ureg_src +ureg_src_indirect( struct ureg_src reg, struct ureg_src addr ) +{ + assert(reg.File != TGSI_FILE_NULL); + assert(addr.File == TGSI_FILE_ADDRESS); + reg.Indirect = 1; + reg.IndirectIndex = addr.Index; + reg.IndirectSwizzle = addr.SwizzleX; + return reg; +} + +static INLINE struct ureg_dst +ureg_dst( struct ureg_src src ) +{ + struct ureg_dst dst; + + dst.File = src.File; + dst.WriteMask = TGSI_WRITEMASK_XYZW; + dst.Indirect = src.Indirect; + dst.IndirectIndex = src.IndirectIndex; + dst.IndirectSwizzle = src.IndirectSwizzle; + dst.Saturate = 0; + dst.Index = src.Index; + dst.Pad1 = 0; + dst.Pad2 = 0; + + return dst; +} + +static INLINE struct ureg_src +ureg_src( struct ureg_dst dst ) +{ + struct ureg_src src; + + src.File = dst.File; + src.SwizzleX = TGSI_SWIZZLE_X; + src.SwizzleY = TGSI_SWIZZLE_Y; + src.SwizzleZ = TGSI_SWIZZLE_Z; + src.SwizzleW = TGSI_SWIZZLE_W; + src.Pad = 0; + src.Indirect = dst.Indirect; + src.IndirectIndex = dst.IndirectIndex; + src.IndirectSwizzle = dst.IndirectSwizzle; + src.Absolute = 0; + src.Index = dst.Index; + src.Negate = 0; + + return src; +} + + + +static INLINE struct ureg_dst +ureg_dst_undef( void ) +{ + struct ureg_dst dst; + + dst.File = TGSI_FILE_NULL; + dst.WriteMask = 0; + dst.Indirect = 0; + dst.IndirectIndex = 0; + dst.IndirectSwizzle = 0; + dst.Saturate = 0; + dst.Index = 0; + dst.Pad1 = 0; + dst.Pad2 = 0; + + return dst; +} + +static INLINE struct ureg_src +ureg_src_undef( void ) +{ + struct ureg_src src; + + src.File = TGSI_FILE_NULL; + src.SwizzleX = 0; + src.SwizzleY = 0; + src.SwizzleZ = 0; + src.SwizzleW = 0; + src.Pad = 0; + src.Indirect = 0; + src.IndirectIndex = 0; + src.IndirectSwizzle = 0; + src.Absolute = 0; + src.Index = 0; + src.Negate = 0; + + return src; +} + +static INLINE boolean +ureg_src_is_undef( struct ureg_src src ) +{ + return src.File == TGSI_FILE_NULL; +} + +static INLINE boolean +ureg_dst_is_undef( struct ureg_dst dst ) +{ + return dst.File == TGSI_FILE_NULL; +} + + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/src/gallium/auxiliary/tgsi/tgsi_util.h b/src/gallium/auxiliary/tgsi/tgsi_util.h index 7877f345587..21eb656327e 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_util.h +++ b/src/gallium/auxiliary/tgsi/tgsi_util.h @@ -32,6 +32,10 @@ extern "C" { #endif +struct tgsi_src_register; +struct tgsi_src_register_ext_swz; +struct tgsi_full_src_register; + void * tgsi_align_128bit( void *unaligned ); diff --git a/src/gallium/auxiliary/util/.gitignore b/src/gallium/auxiliary/util/.gitignore new file mode 100644 index 00000000000..29c586c9b51 --- /dev/null +++ b/src/gallium/auxiliary/util/.gitignore @@ -0,0 +1,2 @@ +u_format_access.c +u_format_table.c diff --git a/src/gallium/auxiliary/util/Makefile b/src/gallium/auxiliary/util/Makefile index 2995aba1b91..ae8d330a787 100644 --- a/src/gallium/auxiliary/util/Makefile +++ b/src/gallium/auxiliary/util/Makefile @@ -5,17 +5,22 @@ LIBNAME = util C_SOURCES = \ u_debug.c \ + u_debug_dump.c \ u_debug_symbol.c \ u_debug_stack.c \ u_blit.c \ u_cache.c \ u_draw_quad.c \ + u_format.c \ + u_format_access.c \ + u_format_table.c \ u_gen_mipmap.c \ u_handle_table.c \ u_hash_table.c \ u_hash.c \ u_keymap.c \ u_linear.c \ + u_network.c \ u_math.c \ u_mm.c \ u_rect.c \ @@ -31,3 +36,9 @@ C_SOURCES = \ u_simple_screen.c include ../../Makefile.template + +u_format_table.c: u_format_table.py u_format_parse.py u_format.csv + python u_format_table.py u_format.csv > $@ + +u_format_access.c: u_format_access.py u_format_parse.py u_format.csv + python u_format_access.py u_format.csv > $@ diff --git a/src/gallium/auxiliary/util/SConscript b/src/gallium/auxiliary/util/SConscript index d3ac7f747fd..28a5ab42569 100644 --- a/src/gallium/auxiliary/util/SConscript +++ b/src/gallium/auxiliary/util/SConscript @@ -1,5 +1,23 @@ Import('*') +env.Clone() + +env.Append(CPPPATH = ['.']) + +env.CodeGenerate( + target = 'u_format_table.c', + script = 'u_format_table.py', + source = ['u_format.csv'], + command = 'python $SCRIPT $SOURCE > $TARGET' +) + +env.CodeGenerate( + target = 'u_format_access.c', + script = 'u_format_access.py', + source = ['u_format.csv'], + command = 'python $SCRIPT $SOURCE > $TARGET' +) + util = env.ConvenienceLibrary( target = 'util', source = [ @@ -7,16 +25,21 @@ util = env.ConvenienceLibrary( 'u_blit.c', 'u_cache.c', 'u_debug.c', + 'u_debug_dump.c', 'u_debug_memory.c', 'u_debug_profile.c', 'u_debug_stack.c', 'u_debug_symbol.c', 'u_draw_quad.c', + 'u_format.c', + 'u_format_access.c', + 'u_format_table.c', 'u_gen_mipmap.c', 'u_handle_table.c', 'u_hash.c', 'u_hash_table.c', 'u_keymap.c', + 'u_network.c', 'u_math.c', 'u_mm.c', 'u_rect.c', diff --git a/src/gallium/auxiliary/util/u_cache.c b/src/gallium/auxiliary/util/u_cache.c index 41cd38171fa..47c16b1c927 100644 --- a/src/gallium/auxiliary/util/u_cache.c +++ b/src/gallium/auxiliary/util/u_cache.c @@ -137,6 +137,8 @@ util_cache_set(struct util_cache *cache, struct util_cache_entry *entry; assert(cache); + if (!cache) + return; entry = util_cache_entry_get(cache, key); util_cache_entry_destroy(cache, entry); @@ -158,6 +160,8 @@ util_cache_get(struct util_cache *cache, struct util_cache_entry *entry; assert(cache); + if (!cache) + return NULL; entry = util_cache_entry_get(cache, key); if(!entry->key && !entry->value) @@ -176,7 +180,9 @@ util_cache_clear(struct util_cache *cache) uint32_t i; assert(cache); - + if (!cache) + return; + for(i = 0; i < cache->size; ++i) util_cache_entry_destroy(cache, &cache->entries[i]); } @@ -186,6 +192,8 @@ void util_cache_destroy(struct util_cache *cache) { assert(cache); + if (!cache) + return; #ifdef DEBUG if(cache->count >= 20*cache->size) { diff --git a/src/gallium/auxiliary/util/u_debug.h b/src/gallium/auxiliary/util/u_debug.h index d42b65ce281..1380d98d7ee 100644 --- a/src/gallium/auxiliary/util/u_debug.h +++ b/src/gallium/auxiliary/util/u_debug.h @@ -88,6 +88,7 @@ _debug_printf(const char *format, ...) * - avoid outputing large strings (512 bytes is the current maximum length * that is guaranteed to be printed in all platforms) */ +#if !defined(PIPE_OS_HAIKU) static INLINE void debug_printf(const char *format, ...) { @@ -101,6 +102,7 @@ debug_printf(const char *format, ...) #endif } +#endif /* !PIPE_OS_HAIKU */ /* * ... isn't portable so we need to pass arguments in parentheses. diff --git a/src/gallium/auxiliary/util/u_debug_dump.c b/src/gallium/auxiliary/util/u_debug_dump.c new file mode 100644 index 00000000000..6bdecde048e --- /dev/null +++ b/src/gallium/auxiliary/util/u_debug_dump.c @@ -0,0 +1,189 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "util/u_memory.h" +#include "util/u_debug.h" +#include "util/u_debug_dump.h" + + +#define DEBUG_DUMP_INVALID_NAME "<invalid>" + + +#if 0 +static const char * +debug_dump_strip_prefix(const char *name, + const char *prefix) +{ + const char *stripped; + assert(name); + assert(prefix); + stripped = name; + while(*prefix) { + if(*stripped != *prefix) + return name; + + ++stripped; + ++prefix; + } + return stripped; +} +#endif + +static const char * +debug_dump_enum_continuous(unsigned value, + unsigned num_names, + const char **names) +{ + if (value >= num_names) + return DEBUG_DUMP_INVALID_NAME; + return names[value]; +} + + +#define DEFINE_DEBUG_DUMP_CONTINUOUS(_name) \ + const char * \ + debug_dump_##_name(unsigned value, boolean shortened) \ + { \ + if(shortened) \ + return debug_dump_enum_continuous(value, Elements(debug_dump_##_name##_short_names), debug_dump_##_name##_short_names); \ + else \ + return debug_dump_enum_continuous(value, Elements(debug_dump_##_name##_names), debug_dump_##_name##_names); \ + } + + +static const char * +debug_dump_blend_factor_names[] = { + DEBUG_DUMP_INVALID_NAME, /* 0x0 */ + "PIPE_BLENDFACTOR_ONE", + "PIPE_BLENDFACTOR_SRC_COLOR", + "PIPE_BLENDFACTOR_SRC_ALPHA", + "PIPE_BLENDFACTOR_DST_ALPHA", + "PIPE_BLENDFACTOR_DST_COLOR", + "PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE", + "PIPE_BLENDFACTOR_CONST_COLOR", + "PIPE_BLENDFACTOR_CONST_ALPHA", + "PIPE_BLENDFACTOR_SRC1_COLOR", + "PIPE_BLENDFACTOR_SRC1_ALPHA", + DEBUG_DUMP_INVALID_NAME, /* 0x0b */ + DEBUG_DUMP_INVALID_NAME, /* 0x0c */ + DEBUG_DUMP_INVALID_NAME, /* 0x0d */ + DEBUG_DUMP_INVALID_NAME, /* 0x0e */ + DEBUG_DUMP_INVALID_NAME, /* 0x0f */ + DEBUG_DUMP_INVALID_NAME, /* 0x10 */ + "PIPE_BLENDFACTOR_ZERO", + "PIPE_BLENDFACTOR_INV_SRC_COLOR", + "PIPE_BLENDFACTOR_INV_SRC_ALPHA", + "PIPE_BLENDFACTOR_INV_DST_ALPHA", + "PIPE_BLENDFACTOR_INV_DST_COLOR", + DEBUG_DUMP_INVALID_NAME, /* 0x16 */ + "PIPE_BLENDFACTOR_INV_CONST_COLOR", + "PIPE_BLENDFACTOR_INV_CONST_ALPHA", + "PIPE_BLENDFACTOR_INV_SRC1_COLOR", + "PIPE_BLENDFACTOR_INV_SRC1_ALPHA" +}; + +static const char * +debug_dump_blend_factor_short_names[] = { + DEBUG_DUMP_INVALID_NAME, /* 0x0 */ + "one", + "src_color", + "src_alpha", + "dst_alpha", + "dst_color", + "src_alpha_saturate", + "const_color", + "const_alpha", + "src1_color", + "src1_alpha", + DEBUG_DUMP_INVALID_NAME, /* 0x0b */ + DEBUG_DUMP_INVALID_NAME, /* 0x0c */ + DEBUG_DUMP_INVALID_NAME, /* 0x0d */ + DEBUG_DUMP_INVALID_NAME, /* 0x0e */ + DEBUG_DUMP_INVALID_NAME, /* 0x0f */ + DEBUG_DUMP_INVALID_NAME, /* 0x10 */ + "zero", + "inv_src_color", + "inv_src_alpha", + "inv_dst_alpha", + "inv_dst_color", + DEBUG_DUMP_INVALID_NAME, /* 0x16 */ + "inv_const_color", + "inv_const_alpha", + "inv_src1_color", + "inv_src1_alpha" +}; + +DEFINE_DEBUG_DUMP_CONTINUOUS(blend_factor) + + +static const char * +debug_dump_blend_func_names[] = { + "PIPE_BLEND_ADD", + "PIPE_BLEND_SUBTRACT", + "PIPE_BLEND_REVERSE_SUBTRACT", + "PIPE_BLEND_MIN", + "PIPE_BLEND_MAX" +}; + +static const char * +debug_dump_blend_func_short_names[] = { + "add", + "sub", + "rev_sub", + "min", + "max" +}; + +DEFINE_DEBUG_DUMP_CONTINUOUS(blend_func) + + +static const char * +debug_dump_func_names[] = { + "PIPE_FUNC_NEVER", + "PIPE_FUNC_LESS", + "PIPE_FUNC_EQUAL", + "PIPE_FUNC_LEQUAL", + "PIPE_FUNC_GREATER", + "PIPE_FUNC_NOTEQUAL", + "PIPE_FUNC_GEQUAL", + "PIPE_FUNC_ALWAYS" +}; + +static const char * +debug_dump_func_short_names[] = { + "never", + "less", + "equal", + "less_equal", + "greater", + "not_equal", + "greater_equal", + "always" +}; + +DEFINE_DEBUG_DUMP_CONTINUOUS(func) diff --git a/src/gallium/auxiliary/util/u_debug_dump.h b/src/gallium/auxiliary/util/u_debug_dump.h new file mode 100644 index 00000000000..102935559c1 --- /dev/null +++ b/src/gallium/auxiliary/util/u_debug_dump.h @@ -0,0 +1,65 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Dump data in human/machine readable format. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#ifndef U_DEBUG_DUMP_H_ +#define U_DEBUG_DUMP_H_ + + +#include "pipe/p_compiler.h" +#include "pipe/p_state.h" + + +#ifdef __cplusplus +extern "C" { +#endif + + +const char * +debug_dump_blend_factor(unsigned value, boolean shortened); + +const char * +debug_dump_blend_func(unsigned value, boolean shortened); + +const char * +debug_dump_func(unsigned value, boolean shortened); + + +/* FIXME: Move the other debug_dump_xxx functions out of u_debug.h into here. */ + + +#ifdef __cplusplus +} +#endif + +#endif /* U_DEBUG_H_ */ diff --git a/src/gallium/auxiliary/util/u_format.c b/src/gallium/auxiliary/util/u_format.c new file mode 100644 index 00000000000..98ea13b60b5 --- /dev/null +++ b/src/gallium/auxiliary/util/u_format.c @@ -0,0 +1,46 @@ +/************************************************************************** + * + * Copyright 2009 Vmware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "u_format.h" + + +const struct util_format_description * +util_format_description(enum pipe_format format) +{ + const struct util_format_description *desc = util_format_description_table; + + while(TRUE) { + if(desc->format == format) + return desc; + + if(desc->format == PIPE_FORMAT_NONE) + return NULL; + + ++desc; + }; +} diff --git a/src/gallium/auxiliary/util/u_format.csv b/src/gallium/auxiliary/util/u_format.csv new file mode 100644 index 00000000000..00a46d0cc48 --- /dev/null +++ b/src/gallium/auxiliary/util/u_format.csv @@ -0,0 +1,99 @@ +PIPE_FORMAT_A8R8G8B8_UNORM , arith , 1, 1, un8 , un8 , un8 , un8 , zyxw, rgb +PIPE_FORMAT_X8R8G8B8_UNORM , arith , 1, 1, un8 , un8 , un8 , un8 , zyx1, rgb +PIPE_FORMAT_B8G8R8A8_UNORM , arith , 1, 1, un8 , un8 , un8 , un8 , yzwx, rgb +PIPE_FORMAT_B8G8R8X8_UNORM , arith , 1, 1, un8 , un8 , un8 , un8 , yzw1, rgb +PIPE_FORMAT_A1R5G5B5_UNORM , arith , 1, 1, un1 , un5 , un5 , un5 , zyxw, rgb +PIPE_FORMAT_A4R4G4B4_UNORM , arith , 1, 1, un4 , un4 , un4 , un4 , zyxw, rgb +PIPE_FORMAT_R5G6B5_UNORM , arith , 1, 1, un5 , un6 , un5 , , zyx1, rgb +PIPE_FORMAT_A2B10G10R10_UNORM , arith , 1, 1, un10, un10, un10, un2 , xyzw, rgb +PIPE_FORMAT_L8_UNORM , arith , 1, 1, un8 , , , , xxx1, rgb +PIPE_FORMAT_A8_UNORM , arith , 1, 1, un8 , , , , 000x, rgb +PIPE_FORMAT_I8_UNORM , arith , 1, 1, un8 , , , , xxxx, rgb +PIPE_FORMAT_A8L8_UNORM , arith , 1, 1, un8 , un8 , , , xxxy, rgb +PIPE_FORMAT_L16_UNORM , arith , 1, 1, un16, , , , xxx1, rgb +PIPE_FORMAT_Z16_UNORM , array , 1, 1, un16, , , , x___, zs +PIPE_FORMAT_Z32_UNORM , array , 1, 1, un32, , , , x___, zs +PIPE_FORMAT_Z32_FLOAT , array , 1, 1, f32 , , , , x___, zs +PIPE_FORMAT_S8Z24_UNORM , arith , 1, 1, un8 , un24, , , yx__, zs +PIPE_FORMAT_Z24S8_UNORM , arith , 1, 1, un24, un8 , , , xy__, zs +PIPE_FORMAT_X8Z24_UNORM , arith , 1, 1, un8 , un24, , , y___, zs +PIPE_FORMAT_Z24X8_UNORM , arith , 1, 1, un24, un8 , , , x___, zs +PIPE_FORMAT_S8_UNORM , array , 1, 1, un8 , , , , _x__, zs +PIPE_FORMAT_R64_FLOAT , array , 1, 1, f64 , , , , x001, rgb +PIPE_FORMAT_R64G64_FLOAT , array , 1, 1, f64 , f64 , , , xy01, rgb +PIPE_FORMAT_R64G64B64_FLOAT , array , 1, 1, f64 , f64 , f64 , , xyz1, rgb +PIPE_FORMAT_R64G64B64A64_FLOAT , array , 1, 1, f64 , f64 , f64 , f64 , xyzw, rgb +PIPE_FORMAT_R32_FLOAT , array , 1, 1, f32 , , , , x001, rgb +PIPE_FORMAT_R32G32_FLOAT , array , 1, 1, f32 , f32 , , , xy01, rgb +PIPE_FORMAT_R32G32B32_FLOAT , array , 1, 1, f32 , f32 , f32 , , xyz1, rgb +PIPE_FORMAT_R32G32B32A32_FLOAT , array , 1, 1, f32 , f32 , f32 , f32 , xyzw, rgb +PIPE_FORMAT_R32_UNORM , array , 1, 1, un32, , , , x001, rgb +PIPE_FORMAT_R32G32_UNORM , array , 1, 1, un32, un32, , , xy01, rgb +PIPE_FORMAT_R32G32B32_UNORM , array , 1, 1, un32, un32, un32, , xyz1, rgb +PIPE_FORMAT_R32G32B32A32_UNORM , array , 1, 1, un32, un32, un32, un32, xyzw, rgb +PIPE_FORMAT_R32_USCALED , array , 1, 1, u32 , , , , x001, rgb +PIPE_FORMAT_R32G32_USCALED , array , 1, 1, u32 , u32 , , , xy01, rgb +PIPE_FORMAT_R32G32B32_USCALED , array , 1, 1, u32 , u32 , u32 , , xyz1, rgb +PIPE_FORMAT_R32G32B32A32_USCALED , array , 1, 1, u32 , u32 , u32 , u32 , xyzw, rgb +PIPE_FORMAT_R32_SNORM , array , 1, 1, sn32, , , , x001, rgb +PIPE_FORMAT_R32G32_SNORM , array , 1, 1, sn32, sn32, , , xy01, rgb +PIPE_FORMAT_R32G32B32_SNORM , array , 1, 1, sn32, sn32, sn32, , xyz1, rgb +PIPE_FORMAT_R32G32B32A32_SNORM , array , 1, 1, sn32, sn32, sn32, sn32, xyzw, rgb +PIPE_FORMAT_R32_SSCALED , array , 1, 1, s32 , , , , x001, rgb +PIPE_FORMAT_R32G32_SSCALED , array , 1, 1, s32 , s32 , , , xy01, rgb +PIPE_FORMAT_R32G32B32_SSCALED , array , 1, 1, s32 , s32 , s32 , , xyz1, rgb +PIPE_FORMAT_R32G32B32A32_SSCALED , array , 1, 1, s32 , s32 , s32 , s32 , xyzw, rgb +PIPE_FORMAT_R16_UNORM , array , 1, 1, un16, , , , x001, rgb +PIPE_FORMAT_R16G16_UNORM , array , 1, 1, un16, un16, , , xy01, rgb +PIPE_FORMAT_R16G16B16_UNORM , array , 1, 1, un16, un16, un16, , xyz1, rgb +PIPE_FORMAT_R16G16B16A16_UNORM , array , 1, 1, un16, un16, un16, un16, xyzw, rgb +PIPE_FORMAT_R16_USCALED , array , 1, 1, u16 , , , , x001, rgb +PIPE_FORMAT_R16G16_USCALED , array , 1, 1, u16 , u16 , , , xy01, rgb +PIPE_FORMAT_R16G16B16_USCALED , array , 1, 1, u16 , u16 , u16 , , xyz1, rgb +PIPE_FORMAT_R16G16B16A16_USCALED , array , 1, 1, u16 , u16 , u16 , u16 , xyzw, rgb +PIPE_FORMAT_R16_SNORM , array , 1, 1, sn16, , , , x001, rgb +PIPE_FORMAT_R16G16_SNORM , array , 1, 1, sn16, sn16, , , xy01, rgb +PIPE_FORMAT_R16G16B16_SNORM , array , 1, 1, sn16, sn16, sn16, , xyz1, rgb +PIPE_FORMAT_R16G16B16A16_SNORM , array , 1, 1, sn16, sn16, sn16, sn16, xyzw, rgb +PIPE_FORMAT_R16_SSCALED , array , 1, 1, s16 , , , , x001, rgb +PIPE_FORMAT_R16G16_SSCALED , array , 1, 1, s16 , s16 , , , xy01, rgb +PIPE_FORMAT_R16G16B16_SSCALED , array , 1, 1, s16 , s16 , s16 , , xyz1, rgb +PIPE_FORMAT_R16G16B16A16_SSCALED , array , 1, 1, s16 , s16 , s16 , s16 , xyzw, rgb +PIPE_FORMAT_R8_UNORM , array , 1, 1, un8 , , , , x001, rgb +PIPE_FORMAT_R8G8_UNORM , array , 1, 1, un8 , un8 , , , xy01, rgb +PIPE_FORMAT_R8G8B8_UNORM , array , 1, 1, un8 , un8 , un8 , , xyz1, rgb +PIPE_FORMAT_R8G8B8A8_UNORM , array , 1, 1, un8 , un8 , un8 , un8 , xyzw, rgb +PIPE_FORMAT_R8G8B8X8_UNORM , array , 1, 1, un8 , un8 , un8 , un8 , xyz1, rgb +PIPE_FORMAT_R8_USCALED , array , 1, 1, u8 , , , , x001, rgb +PIPE_FORMAT_R8G8_USCALED , array , 1, 1, u8 , u8 , , , xy01, rgb +PIPE_FORMAT_R8G8B8_USCALED , array , 1, 1, u8 , u8 , u8 , , xyz1, rgb +PIPE_FORMAT_R8G8B8A8_USCALED , array , 1, 1, u8 , u8 , u8 , u8 , xyzw, rgb +PIPE_FORMAT_R8G8B8X8_USCALED , array , 1, 1, u8 , u8 , u8 , u8 , xyz1, rgb +PIPE_FORMAT_R8_SNORM , array , 1, 1, sn8 , , , , x001, rgb +PIPE_FORMAT_R8G8_SNORM , array , 1, 1, sn8 , sn8 , , , xy01, rgb +PIPE_FORMAT_R8G8B8_SNORM , array , 1, 1, sn8 , sn8 , sn8 , , xyz1, rgb +PIPE_FORMAT_R8G8B8A8_SNORM , array , 1, 1, sn8 , sn8 , sn8 , sn8 , xyzw, rgb +PIPE_FORMAT_R8G8B8X8_SNORM , array , 1, 1, sn8 , sn8 , sn8 , sn8 , xyz1, rgb +PIPE_FORMAT_B6G5R5_SNORM , arith , 1, 1, sn5 , sn5 , sn6 , , zyx1, rgb +PIPE_FORMAT_A8B8G8R8_SNORM , arith , 1, 1, sn8 , sn8 , sn8 , sn8 , zyxw, rgb +PIPE_FORMAT_X8B8G8R8_SNORM , arith , 1, 1, sn8 , sn8 , sn8 , sn8 , zyx1, rgb +PIPE_FORMAT_R8_SSCALED , array , 1, 1, s8 , , , , x001, rgb +PIPE_FORMAT_R8G8_SSCALED , array , 1, 1, s8 , s8 , , , xy01, rgb +PIPE_FORMAT_R8G8B8_SSCALED , array , 1, 1, s8 , s8 , s8 , , xyz1, rgb +PIPE_FORMAT_R8G8B8A8_SSCALED , array , 1, 1, s8 , s8 , s8 , s8 , xyzw, rgb +PIPE_FORMAT_R8G8B8X8_SSCALED , array , 1, 1, s8 , s8 , s8 , s8 , xyz1, rgb +PIPE_FORMAT_R32_FIXED , array , 1, 1, h32 , , , , x001, rgb +PIPE_FORMAT_R32G32_FIXED , array , 1, 1, h32 , h32 , , , xy01, rgb +PIPE_FORMAT_R32G32B32_FIXED , array , 1, 1, h32 , h32 , h32 , , xyz1, rgb +PIPE_FORMAT_R32G32B32A32_FIXED , array , 1, 1, h32 , h32 , h32 , h32 , xyzw, rgb +PIPE_FORMAT_L8_SRGB , arith , 1, 1, u8 , , , , xxx1, srgb +PIPE_FORMAT_A8L8_SRGB , arith , 1, 1, u8 , u8 , , , xxxy, srgb +PIPE_FORMAT_R8G8B8_SRGB , arith , 1, 1, u8 , u8 , u8 , , xyz1, srgb +PIPE_FORMAT_R8G8B8A8_SRGB , arith , 1, 1, u8 , u8 , u8 , u8 , xyzw, srgb +PIPE_FORMAT_R8G8B8X8_SRGB , arith , 1, 1, u8 , u8 , u8 , u8 , xyz1, srgb +PIPE_FORMAT_A8R8G8B8_SRGB , arith , 1, 1, u8 , u8 , u8 , u8 , wxyz, srgb +PIPE_FORMAT_X8R8G8B8_SRGB , arith , 1, 1, u8 , u8 , u8 , u8 , 1xyz, srgb +PIPE_FORMAT_B8G8R8A8_SRGB , arith , 1, 1, u8 , u8 , u8 , u8 , zyxw, srgb +PIPE_FORMAT_B8G8R8X8_SRGB , arith , 1, 1, u8 , u8 , u8 , u8 , zyx1, srgb +PIPE_FORMAT_X8UB8UG8SR8S_NORM , arith , 1, 1, sn8 , sn8 , un8 , x8 , 1zyx, rgb +PIPE_FORMAT_B6UG5SR5S_NORM , arith , 1, 1, sn5 , sn5 , un6 , , xyz1, rgb diff --git a/src/gallium/auxiliary/util/u_format.h b/src/gallium/auxiliary/util/u_format.h new file mode 100644 index 00000000000..7b5b7fcda5b --- /dev/null +++ b/src/gallium/auxiliary/util/u_format.h @@ -0,0 +1,138 @@ +/************************************************************************** + * + * Copyright 2009 Vmware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef U_FORMAT_H +#define U_FORMAT_H + + +#include "pipe/p_format.h" + + +enum util_format_layout { + UTIL_FORMAT_LAYOUT_SCALAR = 0, + UTIL_FORMAT_LAYOUT_ARITH = 1, + UTIL_FORMAT_LAYOUT_ARRAY = 2, + UTIL_FORMAT_LAYOUT_YUV = 3, + UTIL_FORMAT_LAYOUT_DXT = 4 +}; + + +struct util_format_block +{ + /** Block width in pixels */ + unsigned width; + + /** Block height in pixels */ + unsigned height; + + /** Block size in bytes */ + unsigned bits; +}; + + +enum util_format_type { + UTIL_FORMAT_TYPE_VOID = 0, + UTIL_FORMAT_TYPE_UNSIGNED = 1, + UTIL_FORMAT_TYPE_SIGNED = 2, + UTIL_FORMAT_TYPE_FIXED = 3, + UTIL_FORMAT_TYPE_FLOAT = 4 +}; + + +enum util_format_swizzle { + UTIL_FORMAT_SWIZZLE_X = 0, + UTIL_FORMAT_SWIZZLE_Y = 1, + UTIL_FORMAT_SWIZZLE_Z = 2, + UTIL_FORMAT_SWIZZLE_W = 3, + UTIL_FORMAT_SWIZZLE_0 = 4, + UTIL_FORMAT_SWIZZLE_1 = 5, + UTIL_FORMAT_SWIZZLE_NONE = 6 +}; + + +enum util_format_colorspace { + UTIL_FORMAT_COLORSPACE_RGB = 0, + UTIL_FORMAT_COLORSPACE_SRGB = 1, + UTIL_FORMAT_COLORSPACE_YUV = 2, + UTIL_FORMAT_COLORSPACE_ZS = 3, +}; + + +struct util_format_channel_description +{ + unsigned type:6; + unsigned normalized:1; + unsigned size:9; +}; + + +struct util_format_description +{ + enum pipe_format format; + const char *name; + struct util_format_block block; + enum util_format_layout layout; + struct util_format_channel_description channel[4]; + unsigned char swizzle[4]; + enum util_format_colorspace colorspace; +}; + + +extern const struct util_format_description +util_format_description_table[]; + + +const struct util_format_description * +util_format_description(enum pipe_format format); + + +void +util_format_read_4f(enum pipe_format format, + float *dst, unsigned dst_stride, + const void *src, unsigned src_stride, + unsigned x, unsigned y, unsigned w, unsigned h); + +void +util_format_write_4f(enum pipe_format format, + const float *src, unsigned src_stride, + void *dst, unsigned dst_stride, + unsigned x, unsigned y, unsigned w, unsigned h); + +void +util_format_read_4ub(enum pipe_format format, + uint8_t *dst, unsigned dst_stride, + const void *src, unsigned src_stride, + unsigned x, unsigned y, unsigned w, unsigned h); + +void +util_format_write_4ub(enum pipe_format format, + const uint8_t *src, unsigned src_stride, + void *dst, unsigned dst_stride, + unsigned x, unsigned y, unsigned w, unsigned h); + +#endif /* ! U_FORMAT_H */ diff --git a/src/gallium/auxiliary/util/u_format_access.py b/src/gallium/auxiliary/util/u_format_access.py new file mode 100644 index 00000000000..eeb1a9657fd --- /dev/null +++ b/src/gallium/auxiliary/util/u_format_access.py @@ -0,0 +1,498 @@ +#!/usr/bin/env python + +''' +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Pixel format accessor functions. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ +''' + + +import sys + +from u_format_parse import * + + +def short_name(format): + '''Make up a short norm for a format, suitable to be used as suffix in + function names.''' + + name = format.name + if name.startswith('PIPE_FORMAT_'): + name = name[len('PIPE_FORMAT_'):] + name = name.lower() + return name + + +def is_format_supported(format): + '''Determines whether we actually have the plumbing necessary to generate the + to read/write to/from this format.''' + + # FIXME: Ideally we would support any format combination here. + + # XXX: It should be straightforward to support srgb + if format.colorspace not in ('rgb', 'zs'): + return False + + if format.layout not in (ARITH, ARRAY): + return False + + for i in range(4): + type = format.in_types[i] + if type.kind not in (VOID, UNSIGNED, FLOAT): + return False + + # We can only read a color from a depth/stencil format if the depth channel is present + if format.colorspace == 'zs' and format.out_swizzle[0] == SWIZZLE_NONE: + return False + + return True + + +def native_type(format): + '''Get the native appropriate for a format.''' + + if format.layout == ARITH: + # For arithmetic pixel formats return the integer type that matches the whole pixel + return 'uint%u_t' % format.block_size() + elif format.layout == ARRAY: + # For array pixel formats return the integer type that matches the color channel + type = format.in_types[0] + if type.kind == UNSIGNED: + return 'uint%u_t' % type.size + elif type.kind == SIGNED: + return 'int%u_t' % type.size + elif type.kind == FLOAT: + if type.size == 32: + return 'float' + elif type.size == 64: + return 'double' + else: + assert False + else: + assert False + else: + assert False + + +def intermediate_native_type(bits, sign): + '''Find a native type adequate to hold intermediate results of the request bit size.''' + + bytes = 4 # don't use anything smaller than 32bits + while bytes * 8 < bits: + bytes *= 2 + bits = bytes*8 + + if sign: + return 'int%u_t' % bits + else: + return 'uint%u_t' % bits + + +def get_one_shift(type): + '''Get the number of the bit that matches unity for this type.''' + if type.kind == 'FLOAT': + assert False + if not type.norm: + return 0 + if type.kind == UNSIGNED: + return type.size + if type.kind == SIGNED: + return type.size - 1 + if type.kind == FIXED: + return type.size / 2 + assert False + + +def get_one(type): + '''Get the value of unity for this type.''' + if type.kind == 'FLOAT' or not type.norm: + return 1 + else: + return (1 << get_one_shift(type)) - 1 + + +def generate_clamp(): + '''Code generate the clamping functions for each type. + + We don't use a macro so that arguments with side effects, + like *src_pixel++ are correctly handled. + ''' + + for suffix, native_type in [ + ('', 'double'), + ('f', 'float'), + ('ui', 'unsigned int'), + ('si', 'int'), + ]: + print 'static INLINE %s' % native_type + print 'clamp%s(%s value, %s lbound, %s ubound)' % (suffix, native_type, native_type, native_type) + print '{' + print ' if(value < lbound)' + print ' return lbound;' + print ' if(value > ubound)' + print ' return ubound;' + print ' return value;' + print '}' + print + + +def clamp_expr(src_type, dst_type, dst_native_type, value): + '''Generate the expression to clamp the value in the source type to the + destination type range.''' + + if src_type == dst_type: + return value + + # Pick the approriate clamp function + if src_type.kind == FLOAT: + if src_type.size == 32: + func = 'clampf' + elif src_type.size == 64: + func = 'clamp' + else: + assert False + elif src_type.kind == UNSIGNED: + func = 'clampui' + elif src_type.kind == SIGNED: + func = 'clampsi' + else: + assert False + + # Clamp floats to [-1, 1] or [0, 1] range + if src_type.kind == FLOAT and dst_type.norm: + max = 1 + if src_type.sign and dst_type.sign: + min = -1 + else: + min = 0 + return '%s(%s, %s, %s)' % (func, value, min, max) + + # FIXME: Also clamp scaled values + + return value + + +def conversion_expr(src_type, dst_type, dst_native_type, value): + '''Generate the expression to convert a value between two types.''' + + if src_type == dst_type: + return value + + if src_type.kind == FLOAT and dst_type.kind == FLOAT: + return '(%s)%s' % (dst_native_type, value) + + if not src_type.norm and not dst_type.norm: + return '(%s)%s' % (dst_native_type, value) + + value = clamp_expr(src_type, dst_type, dst_native_type, value) + + if dst_type.kind == FLOAT: + if src_type.norm: + one = get_one(src_type) + if src_type.size <= 23: + scale = '(1.0f/0x%x)' % one + else: + # bigger than single precision mantissa, use double + scale = '(1.0/0x%x)' % one + value = '(%s * %s)' % (value, scale) + return '(%s)%s' % (dst_native_type, value) + + if src_type.kind == FLOAT: + if dst_type.norm: + dst_one = get_one(dst_type) + if dst_type.size <= 23: + scale = '0x%x' % dst_one + else: + # bigger than single precision mantissa, use double + scale = '(double)0x%x' % dst_one + value = '(%s * %s)' % (value, scale) + return '(%s)%s' % (dst_native_type, value) + + if src_type.kind == dst_type.kind: + src_one = get_one(src_type) + dst_one = get_one(dst_type) + + if src_one > dst_one and src_type.norm and dst_type.norm: + # We can just bitshift + src_shift = get_one_shift(src_type) + dst_shift = get_one_shift(dst_type) + value = '(%s >> %s)' % (value, src_shift - dst_shift) + else: + # We need to rescale using an intermediate type big enough to hold the multiplication of both + tmp_native_type = intermediate_native_type(src_type.size + dst_type.size, src_type.sign and dst_type.sign) + value = '(%s)%s' % (tmp_native_type, value) + value = '%s * 0x%x / 0x%x' % (value, dst_one, src_one) + value = '(%s)%s' % (dst_native_type, value) + return value + + assert False + + +def generate_format_read(format, dst_type, dst_native_type, dst_suffix): + '''Generate the function to read pixels from a particular format''' + + name = short_name(format) + + src_native_type = native_type(format) + + print 'static void' + print 'util_format_%s_read_%s(%s *dst, unsigned dst_stride, const uint8_t *src, unsigned src_stride, unsigned x0, unsigned y0, unsigned w, unsigned h)' % (name, dst_suffix, dst_native_type) + print '{' + print ' unsigned x, y;' + print ' const uint8_t *src_row = src + y0*src_stride;' + print ' %s *dst_row = dst;' % dst_native_type + print ' for (y = 0; y < h; ++y) {' + print ' const %s *src_pixel = (const %s *)(src_row + x0*%u);' % (src_native_type, src_native_type, format.stride()) + print ' %s *dst_pixel = dst_row;' %dst_native_type + print ' for (x = 0; x < w; ++x) {' + + names = ['']*4 + if format.colorspace == 'rgb': + for i in range(4): + swizzle = format.out_swizzle[i] + if swizzle < 4: + names[swizzle] += 'rgba'[i] + elif format.colorspace == 'zs': + swizzle = format.out_swizzle[0] + if swizzle < 4: + names[swizzle] = 'z' + else: + assert False + else: + assert False + + if format.layout == ARITH: + print ' %s pixel = *src_pixel++;' % src_native_type + shift = 0; + for i in range(4): + src_type = format.in_types[i] + width = src_type.size + if names[i]: + value = 'pixel' + mask = (1 << width) - 1 + if shift: + value = '(%s >> %u)' % (value, shift) + if shift + width < format.block_size(): + value = '(%s & 0x%x)' % (value, mask) + value = conversion_expr(src_type, dst_type, dst_native_type, value) + print ' %s %s = %s;' % (dst_native_type, names[i], value) + shift += width + elif format.layout == ARRAY: + for i in range(4): + src_type = format.in_types[i] + if names[i]: + value = '(*src_pixel++)' + value = conversion_expr(src_type, dst_type, dst_native_type, value) + print ' %s %s = %s;' % (dst_native_type, names[i], value) + else: + assert False + + for i in range(4): + if format.colorspace == 'rgb': + swizzle = format.out_swizzle[i] + if swizzle < 4: + value = names[swizzle] + elif swizzle == SWIZZLE_0: + value = '0' + elif swizzle == SWIZZLE_1: + value = '1' + else: + assert False + elif format.colorspace == 'zs': + if i < 3: + value = 'z' + else: + value = '1' + else: + assert False + print ' *dst_pixel++ = %s; /* %s */' % (value, 'rgba'[i]) + + print ' }' + print ' src_row += src_stride;' + print ' dst_row += dst_stride/sizeof(%s);' % dst_native_type + print ' }' + print '}' + print + + +def generate_format_write(format, src_type, src_native_type, src_suffix): + '''Generate the function to write pixels to a particular format''' + + name = short_name(format) + + dst_native_type = native_type(format) + + print 'static void' + print 'util_format_%s_write_%s(const %s *src, unsigned src_stride, uint8_t *dst, unsigned dst_stride, unsigned x0, unsigned y0, unsigned w, unsigned h)' % (name, src_suffix, src_native_type) + print '{' + print ' unsigned x, y;' + print ' uint8_t *dst_row = dst + y0*dst_stride;' + print ' const %s *src_row = src;' % src_native_type + print ' for (y = 0; y < h; ++y) {' + print ' %s *dst_pixel = (%s *)(dst_row + x0*%u);' % (dst_native_type, dst_native_type, format.stride()) + print ' const %s *src_pixel = src_row;' %src_native_type + print ' for (x = 0; x < w; ++x) {' + + inv_swizzle = [None]*4 + if format.colorspace == 'rgb': + for i in range(4): + swizzle = format.out_swizzle[i] + if swizzle < 4: + inv_swizzle[swizzle] = i + elif format.colorspace == 'zs': + swizzle = format.out_swizzle[0] + if swizzle < 4: + inv_swizzle[swizzle] = 0 + else: + assert False + + if format.layout == ARITH: + print ' %s pixel = 0;' % dst_native_type + shift = 0; + for i in range(4): + dst_type = format.in_types[i] + width = dst_type.size + if inv_swizzle[i] is not None: + value = 'src_pixel[%u]' % inv_swizzle[i] + value = conversion_expr(src_type, dst_type, dst_native_type, value) + if shift: + value = '(%s << %u)' % (value, shift) + print ' pixel |= %s;' % value + shift += width + print ' *dst_pixel++ = pixel;' + elif format.layout == ARRAY: + for i in range(4): + dst_type = format.in_types[i] + if inv_swizzle[i] is not None: + value = 'src_pixel[%u]' % inv_swizzle[i] + value = conversion_expr(src_type, dst_type, dst_native_type, value) + print ' *dst_pixel++ = %s;' % value + else: + assert False + print ' src_pixel += 4;' + + print ' }' + print ' dst_row += dst_stride;' + print ' src_row += src_stride/sizeof(%s);' % src_native_type + print ' }' + print '}' + print + + +def generate_read(formats, dst_type, dst_native_type, dst_suffix): + '''Generate the dispatch function to read pixels from any format''' + + for format in formats: + if is_format_supported(format): + generate_format_read(format, dst_type, dst_native_type, dst_suffix) + + print 'void' + print 'util_format_read_%s(enum pipe_format format, %s *dst, unsigned dst_stride, const void *src, unsigned src_stride, unsigned x, unsigned y, unsigned w, unsigned h)' % (dst_suffix, dst_native_type) + print '{' + print ' void (*func)(%s *dst, unsigned dst_stride, const uint8_t *src, unsigned src_stride, unsigned x0, unsigned y0, unsigned w, unsigned h);' % dst_native_type + print ' switch(format) {' + for format in formats: + if is_format_supported(format): + print ' case %s:' % format.name + print ' func = &util_format_%s_read_%s;' % (short_name(format), dst_suffix) + print ' break;' + print ' default:' + print ' debug_printf("unsupported format\\n");' + print ' return;' + print ' }' + print ' func(dst, dst_stride, (const uint8_t *)src, src_stride, x, y, w, h);' + print '}' + print + + +def generate_write(formats, src_type, src_native_type, src_suffix): + '''Generate the dispatch function to write pixels to any format''' + + for format in formats: + if is_format_supported(format): + generate_format_write(format, src_type, src_native_type, src_suffix) + + print 'void' + print 'util_format_write_%s(enum pipe_format format, const %s *src, unsigned src_stride, void *dst, unsigned dst_stride, unsigned x, unsigned y, unsigned w, unsigned h)' % (src_suffix, src_native_type) + + print '{' + print ' void (*func)(const %s *src, unsigned src_stride, uint8_t *dst, unsigned dst_stride, unsigned x0, unsigned y0, unsigned w, unsigned h);' % src_native_type + print ' switch(format) {' + for format in formats: + if is_format_supported(format): + print ' case %s:' % format.name + print ' func = &util_format_%s_write_%s;' % (short_name(format), src_suffix) + print ' break;' + print ' default:' + print ' debug_printf("unsupported format\\n");' + print ' return;' + print ' }' + print ' func(src, src_stride, (uint8_t *)dst, dst_stride, x, y, w, h);' + print '}' + print + + +def main(): + formats = [] + for arg in sys.argv[1:]: + formats.extend(parse(arg)) + + print '/* This file is autogenerated by u_format_access.py from u_format.csv. Do not edit directly. */' + print + # This will print the copyright message on the top of this file + print __doc__.strip() + print + print '#include "pipe/p_compiler.h"' + print '#include "u_format.h"' + print '#include "u_math.h"' + print + + generate_clamp() + + type = Type(FLOAT, False, 32) + native_type = 'float' + suffix = '4f' + + generate_read(formats, type, native_type, suffix) + generate_write(formats, type, native_type, suffix) + + type = Type(UNSIGNED, True, 8) + native_type = 'uint8_t' + suffix = '4ub' + + generate_read(formats, type, native_type, suffix) + generate_write(formats, type, native_type, suffix) + + +if __name__ == '__main__': + main() diff --git a/src/gallium/auxiliary/util/u_format_parse.py b/src/gallium/auxiliary/util/u_format_parse.py new file mode 100755 index 00000000000..493aff71127 --- /dev/null +++ b/src/gallium/auxiliary/util/u_format_parse.py @@ -0,0 +1,141 @@ +#!/usr/bin/env python + +''' +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ +''' + + +import sys + + +VOID, UNSIGNED, SIGNED, FIXED, FLOAT = range(5) + +SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W, SWIZZLE_0, SWIZZLE_1, SWIZZLE_NONE, = range(7) + +ARITH = 'arith' +ARRAY = 'array' + + +class Type: + '''Describe the type of a color channel.''' + + def __init__(self, kind, norm, size): + self.kind = kind + self.norm = norm + self.size = size + self.sign = kind in (SIGNED, FIXED, FLOAT) + + def __str__(self): + s = str(self.kind) + if self.norm: + s += 'n' + s += str(self.size) + return s + + def __eq__(self, other): + return self.kind == other.kind and self.norm == other.norm and self.size == other.size + + +class Format: + '''Describe a pixel format.''' + + def __init__(self, name, layout, block_width, block_height, in_types, out_swizzle, colorspace): + self.name = name + self.layout = layout + self.block_width = block_width + self.block_height = block_height + self.in_types = in_types + self.out_swizzle = out_swizzle + self.name = name + self.colorspace = colorspace + + def __str__(self): + return self.name + + def block_size(self): + size = 0 + for type in self.in_types: + size += type.size + return size + + def stride(self): + return self.block_size()/8 + + +_kind_parse_map = { + '': VOID, + 'x': VOID, + 'u': UNSIGNED, + 's': SIGNED, + 'h': FIXED, + 'f': FLOAT, +} + +_swizzle_parse_map = { + 'x': SWIZZLE_X, + 'y': SWIZZLE_Y, + 'z': SWIZZLE_Z, + 'w': SWIZZLE_W, + '0': SWIZZLE_0, + '1': SWIZZLE_1, + '_': SWIZZLE_NONE, +} + +def parse(filename): + '''Parse the format descrition in CSV format in terms of the + Type and Format classes above.''' + + stream = open(filename) + formats = [] + for line in stream: + line = line.rstrip() + fields = [field.strip() for field in line.split(',')] + name = fields[0] + layout = fields[1] + block_width, block_height = map(int, fields[2:4]) + in_types = [] + for field in fields[4:8]: + if field: + kind = _kind_parse_map[field[0]] + if field[1] == 'n': + norm = True + size = int(field[2:]) + else: + norm = False + size = int(field[1:]) + else: + kind = VOID + norm = False + size = 0 + in_type = Type(kind, norm, size) + in_types.append(in_type) + out_swizzle = [_swizzle_parse_map[swizzle] for swizzle in fields[8]] + colorspace = fields[9] + formats.append(Format(name, layout, block_width, block_height, in_types, out_swizzle, colorspace)) + return formats + diff --git a/src/gallium/auxiliary/util/u_format_table.py b/src/gallium/auxiliary/util/u_format_table.py new file mode 100755 index 00000000000..8834568e8ee --- /dev/null +++ b/src/gallium/auxiliary/util/u_format_table.py @@ -0,0 +1,144 @@ +#!/usr/bin/env python + +''' +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ +''' + + +import sys + +from u_format_parse import * + + +def layout_map(layout): + return 'UTIL_FORMAT_LAYOUT_' + str(layout).upper() + + +def colorspace_map(colorspace): + return 'UTIL_FORMAT_COLORSPACE_' + str(colorspace).upper() + + +colorspace_channels_map = { + 'rgb': 'rgba', + 'rgba': 'rgba', + 'zs': 'zs', + 'yuv': ['y1', 'y2', 'u', 'v'], + 'dxt': [] +} + + +kind_map = { + VOID: "UTIL_FORMAT_TYPE_VOID", + UNSIGNED: "UTIL_FORMAT_TYPE_UNSIGNED", + SIGNED: "UTIL_FORMAT_TYPE_SIGNED", + FIXED: "UTIL_FORMAT_TYPE_FIXED", + FLOAT: "UTIL_FORMAT_TYPE_FLOAT", +} + + +def bool_map(value): + if value: + return "TRUE" + else: + return "FALSE" + + +swizzle_map = { + SWIZZLE_X: "UTIL_FORMAT_SWIZZLE_X", + SWIZZLE_Y: "UTIL_FORMAT_SWIZZLE_Y", + SWIZZLE_Z: "UTIL_FORMAT_SWIZZLE_Z", + SWIZZLE_W: "UTIL_FORMAT_SWIZZLE_W", + SWIZZLE_0: "UTIL_FORMAT_SWIZZLE_0", + SWIZZLE_1: "UTIL_FORMAT_SWIZZLE_1", + SWIZZLE_NONE: "UTIL_FORMAT_SWIZZLE_NONE", +} + + +def write_format_table(formats): + print '/* This file is autogenerated by u_format_table.py from u_format.csv. Do not edit directly. */' + print + # This will print the copyright message on the top of this file + print __doc__.strip() + print + print '#include "u_format.h"' + print + print 'const struct util_format_description' + print 'util_format_description_table[] = ' + print "{" + for format in formats: + print " {" + print " %s," % (format.name,) + print " \"%s\"," % (format.name,) + print " {%u, %u, %u}, /* block */" % (format.block_width, format.block_height, format.block_size()) + print " %s," % (layout_map(format.layout),) + print " {" + for i in range(4): + type = format.in_types[i] + if i < 3: + sep = "," + else: + sep = "" + print " {%s, %s, %u}%s /* %s */" % (kind_map[type.kind], bool_map(type.norm), type.size, sep, "xyzw"[i]) + print " }," + print " {" + for i in range(4): + swizzle = format.out_swizzle[i] + if i < 3: + sep = "," + else: + sep = "" + try: + comment = layout_channels_map[format.layout][i] + except: + comment = 'ignored' + print " %s%s /* %s */" % (swizzle_map[swizzle], sep, comment) + print " }," + print " %s," % (colorspace_map(format.colorspace),) + print " }," + print " {" + print " PIPE_FORMAT_NONE," + print " \"PIPE_FORMAT_NONE\"," + print " {0, 0, 0}," + print " 0," + print " {{0, 0, 0}, {0, 0, 0}, {0, 0, 0}, {0, 0, 0}}," + print " {0, 0, 0, 0}," + print " 0" + print " }," + print "};" + + +def main(): + + formats = [] + for arg in sys.argv[1:]: + formats.extend(parse(arg)) + write_format_table(formats) + + +if __name__ == '__main__': + main() diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.c b/src/gallium/auxiliary/util/u_gen_mipmap.c index 833c0b83385..f06c0e463d0 100644 --- a/src/gallium/auxiliary/util/u_gen_mipmap.c +++ b/src/gallium/auxiliary/util/u_gen_mipmap.c @@ -46,10 +46,6 @@ #include "util/u_gen_mipmap.h" #include "util/u_simple_shaders.h" -#include "tgsi/tgsi_build.h" -#include "tgsi/tgsi_dump.h" -#include "tgsi/tgsi_parse.h" - #include "cso_cache/cso_context.h" diff --git a/src/gallium/auxiliary/util/u_handle_table.c b/src/gallium/auxiliary/util/u_handle_table.c index 6da7353e259..3703718a621 100644 --- a/src/gallium/auxiliary/util/u_handle_table.c +++ b/src/gallium/auxiliary/util/u_handle_table.c @@ -87,6 +87,8 @@ handle_table_set_destroy(struct handle_table *ht, void (*destroy)(void *object)) { assert(ht); + if (!ht) + return; ht->destroy = destroy; } @@ -155,7 +157,7 @@ handle_table_add(struct handle_table *ht, assert(ht); assert(object); - if(!object) + if(!object || !ht) return 0; /* linear search for an empty handle */ @@ -193,7 +195,7 @@ handle_table_set(struct handle_table *ht, assert(ht); assert(handle); - if(!handle) + if(!handle || !ht) return 0; assert(object); @@ -222,7 +224,7 @@ handle_table_get(struct handle_table *ht, assert(ht); assert(handle); - if(!handle || handle > ht->size) + if(!handle || !ht || handle > ht->size) return NULL; object = ht->objects[handle - 1]; @@ -240,7 +242,7 @@ handle_table_remove(struct handle_table *ht, assert(ht); assert(handle); - if(!handle || handle > ht->size) + if(!handle || !ht || handle > ht->size) return; index = handle - 1; @@ -283,6 +285,9 @@ handle_table_destroy(struct handle_table *ht) unsigned index; assert(ht); + if (!ht) + return; + if(ht->destroy) for(index = 0; index < ht->size; ++index) handle_table_clear(ht, index); diff --git a/src/gallium/auxiliary/util/u_hash_table.c b/src/gallium/auxiliary/util/u_hash_table.c index 2f83e318e44..8c2a8f454cc 100644 --- a/src/gallium/auxiliary/util/u_hash_table.c +++ b/src/gallium/auxiliary/util/u_hash_table.c @@ -148,6 +148,8 @@ hash_table_set(struct hash_table *ht, struct cso_hash_iter iter; assert(ht); + if (!ht) + return PIPE_ERROR_BAD_INPUT; key_hash = ht->hash(key); @@ -183,6 +185,8 @@ hash_table_get(struct hash_table *ht, struct hash_table_item *item; assert(ht); + if (!ht) + return NULL; key_hash = ht->hash(key); @@ -203,6 +207,8 @@ hash_table_remove(struct hash_table *ht, struct hash_table_item *item; assert(ht); + if (!ht) + return; key_hash = ht->hash(key); @@ -225,7 +231,9 @@ hash_table_clear(struct hash_table *ht) struct hash_table_item *item; assert(ht); - + if (!ht) + return; + iter = cso_hash_first_node(ht->cso); while (!cso_hash_iter_is_null(iter)) { item = (struct hash_table_item *)cso_hash_take(ht->cso, cso_hash_iter_key(iter)); @@ -243,9 +251,11 @@ hash_table_foreach(struct hash_table *ht, struct cso_hash_iter iter; struct hash_table_item *item; enum pipe_error result; - + assert(ht); - + if (!ht) + return PIPE_ERROR_BAD_INPUT; + iter = cso_hash_first_node(ht->cso); while (!cso_hash_iter_is_null(iter)) { item = (struct hash_table_item *)cso_hash_iter_data(iter); @@ -264,9 +274,11 @@ hash_table_destroy(struct hash_table *ht) { struct cso_hash_iter iter; struct hash_table_item *item; - + assert(ht); - + if (!ht) + return; + iter = cso_hash_first_node(ht->cso); while (!cso_hash_iter_is_null(iter)) { item = (struct hash_table_item *)cso_hash_iter_data(iter); diff --git a/src/gallium/auxiliary/util/u_keymap.c b/src/gallium/auxiliary/util/u_keymap.c index 3f70809efdc..508a2ee0634 100644 --- a/src/gallium/auxiliary/util/u_keymap.c +++ b/src/gallium/auxiliary/util/u_keymap.c @@ -194,6 +194,8 @@ util_keymap_insert(struct keymap *map, const void *key, struct cso_hash_iter iter; assert(map); + if (!map) + return FALSE; key_hash = hash(key, map->key_size); @@ -234,6 +236,8 @@ util_keymap_lookup(const struct keymap *map, const void *key) struct keymap_item *item; assert(map); + if (!map) + return NULL; key_hash = hash(key, map->key_size); @@ -258,6 +262,8 @@ util_keymap_remove(struct keymap *map, const void *key, void *user) struct keymap_item *item; assert(map); + if (!map) + return; key_hash = hash(key, map->key_size); @@ -267,6 +273,8 @@ util_keymap_remove(struct keymap *map, const void *key, void *user) item = hash_table_item(iter); assert(item); + if (!item) + return; map->delete_func(map, item->key, item->value, user); FREE(item->key); FREE(item); @@ -288,7 +296,9 @@ util_keymap_remove_all(struct keymap *map, void *user) struct keymap_item *item; assert(map); - + if (!map) + return; + iter = cso_hash_first_node(map->cso); while (!cso_hash_iter_is_null(iter)) { item = (struct keymap_item *) diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h index e5003af01d8..4c6c2bc00e1 100644 --- a/src/gallium/auxiliary/util/u_math.h +++ b/src/gallium/auxiliary/util/u_math.h @@ -53,11 +53,11 @@ __inline double ceil(double val) { double ceil_val; - if((val - (long) val) == 0) { + if ((val - (long) val) == 0) { ceil_val = val; } else { - if(val > 0) { + if (val > 0) { ceil_val = (long) val + 1; } else { @@ -73,11 +73,11 @@ __inline double floor(double val) { double floor_val; - if((val - (long) val) == 0) { + if ((val - (long) val) == 0) { floor_val = val; } else { - if(val > 0) { + if (val > 0) { floor_val = (long) val; } else { @@ -189,7 +189,10 @@ static INLINE double log2( double x ) extern float pow2_table[POW2_TABLE_SIZE]; - +/** + * Initialize math module. This should be called before using any + * other functions in this module. + */ extern void util_init_math(void); @@ -216,23 +219,24 @@ util_fast_exp2(float x) int32_t ipart; float fpart, mpart; union fi epart; - + if(x > 129.00000f) return 3.402823466e+38f; - - if(x < -126.99999f) + + if (x < -126.99999f) return 0.0f; ipart = (int32_t) x; fpart = x - (float) ipart; - + /* same as * epart.f = (float) (1 << ipart) - * but faster and without integer overflow for ipart > 31 */ + * but faster and without integer overflow for ipart > 31 + */ epart.i = (ipart + 127 ) << 23; - + mpart = pow2_table[POW2_TABLE_OFFSET + (int)(fpart * POW2_TABLE_SCALE)]; - + return epart.f * mpart; } @@ -254,6 +258,9 @@ util_fast_exp(float x) extern float log2_table[LOG2_TABLE_SIZE]; +/** + * Fast approximation to log2(x). + */ static INLINE float util_fast_log2(float x) { @@ -267,6 +274,9 @@ util_fast_log2(float x) } +/** + * Fast approximation to x^y. + */ static INLINE float util_fast_pow(float x, float y) { @@ -274,7 +284,6 @@ util_fast_pow(float x, float y) } - /** * Floor(x), returned as int. */ @@ -284,8 +293,8 @@ util_ifloor(float f) int ai, bi; double af, bf; union fi u; - af = (3 << 22) + 0.5 + (double)f; - bf = (3 << 22) + 0.5 - (double)f; + af = (3 << 22) + 0.5 + (double) f; + bf = (3 << 22) + 0.5 - (double) f; u.f = (float) af; ai = u.i; u.f = (float) bf; bi = u.i; return (ai - bi) >> 1; @@ -305,9 +314,9 @@ util_iround(float f) #elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86) int r; _asm { - fld f - fistp r - } + fld f + fistp r + } return r; #else if (f >= 0.0f) @@ -340,7 +349,7 @@ static INLINE unsigned long ffs( unsigned long u ) { unsigned long i; - if(_BitScanForward(&i, u)) + if (_BitScanForward(&i, u)) return i + 1; else return 0; @@ -351,7 +360,7 @@ unsigned ffs( unsigned u ) { unsigned i; - if( u == 0 ) { + if (u == 0) { return 0; } @@ -365,6 +374,22 @@ unsigned ffs( unsigned u ) #define ffs __builtin_ffs #endif +#ifdef __MINGW32__ +#define ffs __builtin_ffs +#endif + + +/* Could also binary search for the highest bit. + */ +static INLINE unsigned +util_unsigned_logbase2(unsigned n) +{ + unsigned log2 = 0; + while (n >>= 1) + ++log2; + return log2; +} + /** * Return float bits. @@ -378,7 +403,10 @@ fui( float f ) } - +/** + * Convert ubyte to float in [0, 1]. + * XXX a 256-entry lookup table would be slightly faster. + */ static INLINE float ubyte_to_float(ubyte ub) { @@ -409,7 +437,23 @@ float_to_ubyte(float f) } +/** + * Calc log base 2 + */ +static INLINE unsigned +util_logbase2(unsigned n) +{ + unsigned log2 = 0; + while (n >>= 1) + ++log2; + return log2; +} + +/** + * Clamp X to [MIN, MAX]. + * This is a macro to allow float, int, uint, etc. types. + */ #define CLAMP( X, MIN, MAX ) ( (X)<(MIN) ? (MIN) : ((X)>(MAX) ? (MAX) : (X)) ) #define MIN2( A, B ) ( (A)<(B) ? (A) : (B) ) @@ -422,6 +466,11 @@ align(int value, int alignment) return (value + alignment - 1) & ~(alignment - 1); } +static INLINE unsigned +minify(unsigned value) +{ + return MAX2(1, value >> 1); +} #ifndef COPY_4V #define COPY_4V( DST, SRC ) \ diff --git a/src/gallium/auxiliary/util/u_memory.h b/src/gallium/auxiliary/util/u_memory.h index 0b18d043adb..c3f8c918338 100644 --- a/src/gallium/auxiliary/util/u_memory.h +++ b/src/gallium/auxiliary/util/u_memory.h @@ -100,8 +100,14 @@ ExFreePool(void *P); #define MALLOC( SIZE ) malloc( SIZE ) #define CALLOC( COUNT, SIZE ) calloc( COUNT, SIZE ) #define FREE( PTR ) free( PTR ) -#define REALLOC( OLDPTR, OLDSIZE, NEWSIZE ) realloc( OLDPTR, NEWSIZE ) +static INLINE void * +_REALLOC( void *old_ptr, unsigned old_size, unsigned new_size ) +{ + (void) old_size; + return realloc(old_ptr, new_size); +} +#define REALLOC( a, b, c ) _REALLOC( a, b, c ) #endif diff --git a/src/gallium/auxiliary/util/u_mm.c b/src/gallium/auxiliary/util/u_mm.c index 151a480d34d..4b75d4ba1d0 100644 --- a/src/gallium/auxiliary/util/u_mm.c +++ b/src/gallium/auxiliary/util/u_mm.c @@ -33,30 +33,32 @@ void u_mmDumpMemInfo(const struct mem_block *heap) { - debug_printf("Memory heap %p:\n", (void *)heap); + debug_printf("Memory heap %p:\n", (void *) heap); if (heap == 0) { debug_printf(" heap == 0\n"); - } else { + } + else { const struct mem_block *p; - for(p = heap->next; p != heap; p = p->next) { - debug_printf(" Offset:%08x, Size:%08x, %c%c\n",p->ofs,p->size, - p->free ? 'F':'.', - p->reserved ? 'R':'.'); + for (p = heap->next; p != heap; p = p->next) { + debug_printf(" Offset:%08x, Size:%08x, %c%c\n", p->ofs, p->size, + p->free ? 'F':'.', + p->reserved ? 'R':'.'); } debug_printf("\nFree list:\n"); - for(p = heap->next_free; p != heap; p = p->next_free) { - debug_printf(" FREE Offset:%08x, Size:%08x, %c%c\n",p->ofs,p->size, - p->free ? 'F':'.', - p->reserved ? 'R':'.'); + for (p = heap->next_free; p != heap; p = p->next_free) { + debug_printf(" FREE Offset:%08x, Size:%08x, %c%c\n", p->ofs, p->size, + p->free ? 'F':'.', + p->reserved ? 'R':'.'); } } debug_printf("End of memory blocks\n"); } + struct mem_block * u_mmInit(int ofs, int size) { diff --git a/src/gallium/auxiliary/util/u_mm.h b/src/gallium/auxiliary/util/u_mm.h index ce20e487635..6b158aae6e4 100644 --- a/src/gallium/auxiliary/util/u_mm.h +++ b/src/gallium/auxiliary/util/u_mm.h @@ -84,7 +84,7 @@ extern struct mem_block *u_mmFindBlock(struct mem_block *heap, int start); extern void u_mmDestroy(struct mem_block *mmInit); /** - * For debuging purpose. + * For debugging purposes. */ extern void u_mmDumpMemInfo(const struct mem_block *mmInit); diff --git a/src/gallium/auxiliary/util/u_network.c b/src/gallium/auxiliary/util/u_network.c new file mode 100644 index 00000000000..bc4b7584067 --- /dev/null +++ b/src/gallium/auxiliary/util/u_network.c @@ -0,0 +1,188 @@ + +#include "pipe/p_compiler.h" +#include "util/u_network.h" +#include "util/u_debug.h" + +#if defined(PIPE_SUBSYSTEM_WINDOWS_USER) +# include <winsock2.h> +# include <windows.h> +#elif defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU) +# include <sys/socket.h> +# include <netinet/in.h> +# include <unistd.h> +# include <fcntl.h> +# include <netdb.h> +#else +# warning "No socket implementation" +#endif + +boolean +u_socket_init() +{ +#if defined(PIPE_SUBSYSTEM_WINDOWS_USER) + WORD wVersionRequested; + WSADATA wsaData; + int err; + + /* Use the MAKEWORD(lowbyte, highbyte) macro declared in Windef.h */ + wVersionRequested = MAKEWORD(1, 1); + + err = WSAStartup(wVersionRequested, &wsaData); + if (err != 0) { + debug_printf("WSAStartup failed with error: %d\n", err); + return FALSE; + } + return TRUE; +#elif defined(PIPE_HAVE_SOCKETS) + return TRUE; +#else + return FALSE; +#endif +} + +void +u_socket_stop() +{ +#if defined(PIPE_SUBSYSTEM_WINDOWS_USER) + WSACleanup(); +#endif +} + +void +u_socket_close(int s) +{ + if (s < 0) + return; + +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU) + shutdown(s, SHUT_RDWR); + close(s); +#elif defined(PIPE_SUBSYSTEM_WINDOWS_USER) + shutdown(s, SD_BOTH); + closesocket(s); +#else + assert(0); +#endif +} + +int u_socket_accept(int s) +{ +#if defined(PIPE_HAVE_SOCKETS) + return accept(s, NULL, NULL); +#else + return -1; +#endif +} + +int +u_socket_send(int s, void *data, size_t size) +{ +#if defined(PIPE_HAVE_SOCKETS) + return send(s, data, size, 0); +#else + return -1; +#endif +} + +int +u_socket_peek(int s, void *data, size_t size) +{ +#if defined(PIPE_HAVE_SOCKETS) + return recv(s, data, size, MSG_PEEK); +#else + return -1; +#endif +} + +int +u_socket_recv(int s, void *data, size_t size) +{ +#if defined(PIPE_HAVE_SOCKETS) + return recv(s, data, size, 0); +#else + return -1; +#endif +} + +int +u_socket_connect(const char *hostname, uint16_t port) +{ +#if defined(PIPE_HAVE_SOCKETS) + int s; + struct sockaddr_in sa; + struct hostent *host = NULL; + + memset(&sa, 0, sizeof(struct sockaddr_in)); + host = gethostbyname(hostname); + if (!host) + return -1; + + memcpy((char *)&sa.sin_addr,host->h_addr,host->h_length); + sa.sin_family= host->h_addrtype; + sa.sin_port = htons(port); + + s = socket(host->h_addrtype, SOCK_STREAM, IPPROTO_TCP); + if (s < 0) + return -1; + + if (connect(s, (struct sockaddr *)&sa, sizeof(sa))) { + u_socket_close(s); + return -1; + } + + return s; +#else + assert(0); + return -1; +#endif +} + +int +u_socket_listen_on_port(uint16_t portnum) +{ +#if defined(PIPE_HAVE_SOCKETS) + int s; + struct sockaddr_in sa; + memset(&sa, 0, sizeof(struct sockaddr_in)); + + sa.sin_family = AF_INET; + sa.sin_port = htons(portnum); + + s = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP); + if (s < 0) + return -1; + + if (bind(s, (struct sockaddr *)&sa, sizeof(struct sockaddr_in)) == -1) { + u_socket_close(s); + return -1; + } + + listen(s, 0); + + return s; +#else + assert(0); + return -1; +#endif +} + +void +u_socket_block(int s, boolean block) +{ +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU) + int old = fcntl(s, F_GETFL, 0); + if (old == -1) + return; + + /* TODO obey block */ + if (block) + fcntl(s, F_SETFL, old & ~O_NONBLOCK); + else + fcntl(s, F_SETFL, old | O_NONBLOCK); +#elif defined(PIPE_SUBSYSTEM_WINDOWS_USER) + u_long iMode = block ? 0 : 1; + ioctlsocket(s, FIONBIO, &iMode); +#else + assert(0); +#endif +} diff --git a/src/gallium/auxiliary/util/u_network.h b/src/gallium/auxiliary/util/u_network.h new file mode 100644 index 00000000000..8c778f492ca --- /dev/null +++ b/src/gallium/auxiliary/util/u_network.h @@ -0,0 +1,24 @@ + +#ifndef _U_NETWORK_H_ +#define _U_NETWORK_H_ + +#include "pipe/p_compiler.h" + +#if defined(PIPE_SUBSYSTEM_WINDOWS_USER) +# define PIPE_HAVE_SOCKETS +#elif defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU) +# define PIPE_HAVE_SOCKETS +#endif + +boolean u_socket_init(void); +void u_socket_stop(void); +void u_socket_close(int s); +int u_socket_listen_on_port(uint16_t portnum); +int u_socket_accept(int s); +int u_socket_connect(const char *host, uint16_t port); +int u_socket_send(int s, void *data, size_t size); +int u_socket_peek(int s, void *data, size_t size); +int u_socket_recv(int s, void *data, size_t size); +void u_socket_block(int s, boolean block); + +#endif diff --git a/src/gallium/auxiliary/util/u_prim.h b/src/gallium/auxiliary/util/u_prim.h index d7c3995dbf0..a9b533eea70 100644 --- a/src/gallium/auxiliary/util/u_prim.h +++ b/src/gallium/auxiliary/util/u_prim.h @@ -119,7 +119,7 @@ static INLINE boolean u_trim_pipe_prim( unsigned pipe_prim, unsigned *nr ) } -static INLINE boolean u_reduced_prim( unsigned pipe_prim ) +static INLINE unsigned u_reduced_prim( unsigned pipe_prim ) { switch (pipe_prim) { case PIPE_PRIM_POINTS: diff --git a/src/gallium/auxiliary/util/u_rect.c b/src/gallium/auxiliary/util/u_rect.c index 74259d453b1..9866b6fc8a0 100644 --- a/src/gallium/auxiliary/util/u_rect.c +++ b/src/gallium/auxiliary/util/u_rect.c @@ -43,7 +43,7 @@ * src_pitch may be negative to do vertical flip of pixels from source. */ void -pipe_copy_rect(ubyte * dst, +util_copy_rect(ubyte * dst, const struct pipe_format_block *block, unsigned dst_stride, unsigned dst_x, @@ -91,7 +91,7 @@ pipe_copy_rect(ubyte * dst, } void -pipe_fill_rect(ubyte * dst, +util_fill_rect(ubyte * dst, const struct pipe_format_block *block, unsigned dst_stride, unsigned dst_x, @@ -204,7 +204,7 @@ util_surface_copy(struct pipe_context *pipe, if (src_map && dst_map) { /* If do_flip, invert src_y position and pass negative src stride */ - pipe_copy_rect(dst_map, + util_copy_rect(dst_map, &dst_trans->block, dst_trans->stride, 0, 0, @@ -263,7 +263,7 @@ util_surface_fill(struct pipe_context *pipe, case 1: case 2: case 4: - pipe_fill_rect(dst_map, &dst_trans->block, dst_trans->stride, + util_fill_rect(dst_map, &dst_trans->block, dst_trans->stride, 0, 0, width, height, value); break; case 8: diff --git a/src/gallium/auxiliary/util/u_rect.h b/src/gallium/auxiliary/util/u_rect.h index 59e842e16d1..daa50834d36 100644 --- a/src/gallium/auxiliary/util/u_rect.h +++ b/src/gallium/auxiliary/util/u_rect.h @@ -42,13 +42,13 @@ struct pipe_surface; extern void -pipe_copy_rect(ubyte * dst, const struct pipe_format_block *block, +util_copy_rect(ubyte * dst, const struct pipe_format_block *block, unsigned dst_stride, unsigned dst_x, unsigned dst_y, unsigned width, unsigned height, const ubyte * src, int src_stride, unsigned src_x, int src_y); extern void -pipe_fill_rect(ubyte * dst, const struct pipe_format_block *block, +util_fill_rect(ubyte * dst, const struct pipe_format_block *block, unsigned dst_stride, unsigned dst_x, unsigned dst_y, unsigned width, unsigned height, uint32_t value); diff --git a/src/gallium/auxiliary/util/u_simple_screen.c b/src/gallium/auxiliary/util/u_simple_screen.c index 8114b53cd0d..f01296b40fc 100644 --- a/src/gallium/auxiliary/util/u_simple_screen.c +++ b/src/gallium/auxiliary/util/u_simple_screen.c @@ -65,12 +65,13 @@ pass_surface_buffer_create(struct pipe_screen *screen, unsigned width, unsigned height, enum pipe_format format, unsigned usage, + unsigned tex_usage, unsigned *stride) { struct pipe_buffer *buffer = screen->winsys->surface_buffer_create(screen->winsys, width, height, - format, usage, stride); + format, usage, tex_usage, stride); buffer->screen = screen; diff --git a/src/gallium/auxiliary/util/u_simple_shaders.c b/src/gallium/auxiliary/util/u_simple_shaders.c index acc5b83c622..ab754296fa8 100644 --- a/src/gallium/auxiliary/util/u_simple_shaders.c +++ b/src/gallium/auxiliary/util/u_simple_shaders.c @@ -42,9 +42,7 @@ #include "util/u_memory.h" #include "util/u_simple_shaders.h" -#include "tgsi/tgsi_build.h" -#include "tgsi/tgsi_dump.h" -#include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_ureg.h" @@ -58,93 +56,31 @@ util_make_vertex_passthrough_shader(struct pipe_context *pipe, const uint *semantic_indexes) { - struct pipe_shader_state shader; - struct tgsi_token tokens[100]; - struct tgsi_header *header; - struct tgsi_processor *processor; - struct tgsi_full_declaration decl; - struct tgsi_full_instruction inst; - const uint procType = TGSI_PROCESSOR_VERTEX; - uint ti, i; + struct ureg_program *ureg; + uint i; - /* shader header - */ - *(struct tgsi_version *) &tokens[0] = tgsi_build_version(); + ureg = ureg_create( TGSI_PROCESSOR_VERTEX ); + if (ureg == NULL) + return NULL; - header = (struct tgsi_header *) &tokens[1]; - *header = tgsi_build_header(); - - processor = (struct tgsi_processor *) &tokens[2]; - *processor = tgsi_build_processor( procType, header ); - - ti = 3; - - /* declare inputs */ - for (i = 0; i < num_attribs; i++) { - decl = tgsi_default_full_declaration(); - decl.Declaration.File = TGSI_FILE_INPUT; - - decl.Declaration.Semantic = 1; - decl.Semantic.SemanticName = semantic_names[i]; - decl.Semantic.SemanticIndex = semantic_indexes[i]; - - decl.DeclarationRange.First = - decl.DeclarationRange.Last = i; - ti += tgsi_build_full_declaration(&decl, - &tokens[ti], - header, - Elements(tokens) - ti); - } - - /* declare outputs */ for (i = 0; i < num_attribs; i++) { - decl = tgsi_default_full_declaration(); - decl.Declaration.File = TGSI_FILE_OUTPUT; - decl.Declaration.Semantic = 1; - decl.Semantic.SemanticName = semantic_names[i]; - decl.Semantic.SemanticIndex = semantic_indexes[i]; - decl.DeclarationRange.First = - decl.DeclarationRange.Last = i; - ti += tgsi_build_full_declaration(&decl, - &tokens[ti], - header, - Elements(tokens) - ti); + struct ureg_src src; + struct ureg_dst dst; + + src = ureg_DECL_vs_input( ureg, + semantic_names[i], + semantic_indexes[i]); + + dst = ureg_DECL_output( ureg, + semantic_names[i], + semantic_indexes[i]); + + ureg_MOV( ureg, dst, src ); } - /* emit MOV instructions */ - for (i = 0; i < num_attribs; i++) { - /* MOVE out[i], in[i]; */ - inst = tgsi_default_full_instruction(); - inst.Instruction.Opcode = TGSI_OPCODE_MOV; - inst.Instruction.NumDstRegs = 1; - inst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT; - inst.FullDstRegisters[0].DstRegister.Index = i; - inst.Instruction.NumSrcRegs = 1; - inst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT; - inst.FullSrcRegisters[0].SrcRegister.Index = i; - ti += tgsi_build_full_instruction(&inst, - &tokens[ti], - header, - Elements(tokens) - ti ); - } - - /* END instruction */ - inst = tgsi_default_full_instruction(); - inst.Instruction.Opcode = TGSI_OPCODE_END; - inst.Instruction.NumDstRegs = 0; - inst.Instruction.NumSrcRegs = 0; - ti += tgsi_build_full_instruction(&inst, - &tokens[ti], - header, - Elements(tokens) - ti ); + ureg_END( ureg ); -#if 0 /*debug*/ - tgsi_dump(tokens, 0); -#endif - - shader.tokens = tokens; - - return pipe->create_vs_state(pipe, &shader); + return ureg_create_shader_and_destroy( ureg, pipe ); } @@ -161,130 +97,29 @@ void * util_make_fragment_tex_shader_writemask(struct pipe_context *pipe, unsigned writemask ) { - struct pipe_shader_state shader; - struct tgsi_token tokens[100]; - struct tgsi_header *header; - struct tgsi_processor *processor; - struct tgsi_full_declaration decl; - struct tgsi_full_instruction inst; - const uint procType = TGSI_PROCESSOR_FRAGMENT; - uint ti; - - /* shader header - */ - *(struct tgsi_version *) &tokens[0] = tgsi_build_version(); - - header = (struct tgsi_header *) &tokens[1]; - *header = tgsi_build_header(); - - processor = (struct tgsi_processor *) &tokens[2]; - *processor = tgsi_build_processor( procType, header ); - - ti = 3; - - /* declare TEX[0] input */ - decl = tgsi_default_full_declaration(); - decl.Declaration.File = TGSI_FILE_INPUT; - /* XXX this could be linear... */ - decl.Declaration.Interpolate = TGSI_INTERPOLATE_PERSPECTIVE; - decl.Declaration.Semantic = 1; - decl.Semantic.SemanticName = TGSI_SEMANTIC_GENERIC; - decl.Semantic.SemanticIndex = 0; - decl.DeclarationRange.First = - decl.DeclarationRange.Last = 0; - ti += tgsi_build_full_declaration(&decl, - &tokens[ti], - header, - Elements(tokens) - ti); - - /* declare color[0] output */ - decl = tgsi_default_full_declaration(); - decl.Declaration.File = TGSI_FILE_OUTPUT; - decl.Declaration.Semantic = 1; - decl.Semantic.SemanticName = TGSI_SEMANTIC_COLOR; - decl.Semantic.SemanticIndex = 0; - decl.DeclarationRange.First = - decl.DeclarationRange.Last = 0; - ti += tgsi_build_full_declaration(&decl, - &tokens[ti], - header, - Elements(tokens) - ti); - - /* declare sampler */ - decl = tgsi_default_full_declaration(); - decl.Declaration.File = TGSI_FILE_SAMPLER; - decl.DeclarationRange.First = - decl.DeclarationRange.Last = 0; - ti += tgsi_build_full_declaration(&decl, - &tokens[ti], - header, - Elements(tokens) - ti); - - - if (writemask != TGSI_WRITEMASK_XYZW) { - struct tgsi_full_immediate imm; - static const float value[4] = { 0, 0, 0, 1 }; - - imm = tgsi_default_full_immediate(); - imm.Immediate.NrTokens += 4; - imm.Immediate.DataType = TGSI_IMM_FLOAT32; - imm.u.Pointer = value; - - ti += tgsi_build_full_immediate(&imm, - &tokens[ti], - header, - Elements(tokens) - ti ); - - /* MOV instruction */ - inst = tgsi_default_full_instruction(); - inst.Instruction.Opcode = TGSI_OPCODE_MOV; - inst.Instruction.NumDstRegs = 1; - inst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT; - inst.FullDstRegisters[0].DstRegister.Index = 0; - inst.Instruction.NumSrcRegs = 1; - inst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_IMMEDIATE; - inst.FullSrcRegisters[0].SrcRegister.Index = 0; - ti += tgsi_build_full_instruction(&inst, - &tokens[ti], - header, - Elements(tokens) - ti ); - } + struct ureg_program *ureg; + struct ureg_src sampler; + struct ureg_src tex; + struct ureg_dst out; - /* TEX instruction */ - inst = tgsi_default_full_instruction(); - inst.Instruction.Opcode = TGSI_OPCODE_TEX; - inst.Instruction.NumDstRegs = 1; - inst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT; - inst.FullDstRegisters[0].DstRegister.Index = 0; - inst.FullDstRegisters[0].DstRegister.WriteMask = writemask; - inst.Instruction.NumSrcRegs = 2; - inst.InstructionExtTexture.Texture = TGSI_TEXTURE_2D; - inst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT; - inst.FullSrcRegisters[0].SrcRegister.Index = 0; - inst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_SAMPLER; - inst.FullSrcRegisters[1].SrcRegister.Index = 0; - ti += tgsi_build_full_instruction(&inst, - &tokens[ti], - header, - Elements(tokens) - ti ); - - /* END instruction */ - inst = tgsi_default_full_instruction(); - inst.Instruction.Opcode = TGSI_OPCODE_END; - inst.Instruction.NumDstRegs = 0; - inst.Instruction.NumSrcRegs = 0; - ti += tgsi_build_full_instruction(&inst, - &tokens[ti], - header, - Elements(tokens) - ti ); - -#if 0 /*debug*/ - tgsi_dump(tokens, 0); -#endif - - shader.tokens = tokens; - - return pipe->create_fs_state(pipe, &shader); + ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT ); + if (ureg == NULL) + return NULL; + + sampler = ureg_DECL_sampler( ureg, 0 ); + + tex = ureg_DECL_fs_input( ureg, + TGSI_SEMANTIC_GENERIC, 0, + TGSI_INTERPOLATE_PERSPECTIVE ); + + out = ureg_DECL_output( ureg, + TGSI_SEMANTIC_COLOR, + 0 ); + + ureg_TEX( ureg, out, TGSI_TEXTURE_2D, tex, sampler ); + ureg_END( ureg ); + + return ureg_create_shader_and_destroy( ureg, pipe ); } void * @@ -296,95 +131,29 @@ util_make_fragment_tex_shader(struct pipe_context *pipe ) - - /** * Make simple fragment color pass-through shader. */ void * util_make_fragment_passthrough_shader(struct pipe_context *pipe) { - struct pipe_shader_state shader; - struct tgsi_token tokens[40]; - struct tgsi_header *header; - struct tgsi_processor *processor; - struct tgsi_full_declaration decl; - struct tgsi_full_instruction inst; - const uint procType = TGSI_PROCESSOR_FRAGMENT; - uint ti; - - /* shader header - */ - *(struct tgsi_version *) &tokens[0] = tgsi_build_version(); - - header = (struct tgsi_header *) &tokens[1]; - *header = tgsi_build_header(); - - processor = (struct tgsi_processor *) &tokens[2]; - *processor = tgsi_build_processor( procType, header ); - - ti = 3; - - /* declare input */ - decl = tgsi_default_full_declaration(); - decl.Declaration.File = TGSI_FILE_INPUT; - decl.Declaration.Semantic = 1; - decl.Semantic.SemanticName = TGSI_SEMANTIC_COLOR; - decl.Semantic.SemanticIndex = 0; - decl.DeclarationRange.First = - decl.DeclarationRange.Last = 0; - ti += tgsi_build_full_declaration(&decl, - &tokens[ti], - header, - Elements(tokens) - ti); - - /* declare output */ - decl = tgsi_default_full_declaration(); - decl.Declaration.File = TGSI_FILE_OUTPUT; - decl.Declaration.Semantic = 1; - decl.Semantic.SemanticName = TGSI_SEMANTIC_COLOR; - decl.Semantic.SemanticIndex = 0; - decl.DeclarationRange.First = - decl.DeclarationRange.Last = 0; - ti += tgsi_build_full_declaration(&decl, - &tokens[ti], - header, - Elements(tokens) - ti); - - - /* MOVE out[0], in[0]; */ - inst = tgsi_default_full_instruction(); - inst.Instruction.Opcode = TGSI_OPCODE_MOV; - inst.Instruction.NumDstRegs = 1; - inst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT; - inst.FullDstRegisters[0].DstRegister.Index = 0; - inst.Instruction.NumSrcRegs = 1; - inst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT; - inst.FullSrcRegisters[0].SrcRegister.Index = 0; - ti += tgsi_build_full_instruction(&inst, - &tokens[ti], - header, - Elements(tokens) - ti ); - - /* END instruction */ - inst = tgsi_default_full_instruction(); - inst.Instruction.Opcode = TGSI_OPCODE_END; - inst.Instruction.NumDstRegs = 0; - inst.Instruction.NumSrcRegs = 0; - ti += tgsi_build_full_instruction(&inst, - &tokens[ti], - header, - Elements(tokens) - ti ); - - assert(ti < Elements(tokens)); - -#if 0 /*debug*/ - tgsi_dump(tokens, 0); -#endif - - shader.tokens = tokens; - - return pipe->create_fs_state(pipe, &shader); + struct ureg_program *ureg; + struct ureg_src src; + struct ureg_dst dst; + + ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT ); + if (ureg == NULL) + return NULL; + + src = ureg_DECL_fs_input( ureg, TGSI_SEMANTIC_COLOR, 0, + TGSI_INTERPOLATE_PERSPECTIVE ); + + dst = ureg_DECL_output( ureg, TGSI_SEMANTIC_COLOR, 0 ); + + ureg_MOV( ureg, dst, src ); + ureg_END( ureg ); + + return ureg_create_shader_and_destroy( ureg, pipe ); } diff --git a/src/gallium/auxiliary/util/u_stream_stdc.c b/src/gallium/auxiliary/util/u_stream_stdc.c index d8f648e5dd1..5cd05b29047 100644 --- a/src/gallium/auxiliary/util/u_stream_stdc.c +++ b/src/gallium/auxiliary/util/u_stream_stdc.c @@ -32,7 +32,7 @@ #include "pipe/p_config.h" -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_HAIKU) #include <stdio.h> diff --git a/src/gallium/auxiliary/util/u_tile.c b/src/gallium/auxiliary/util/u_tile.c index aa16fe771fe..1235a67d264 100644 --- a/src/gallium/auxiliary/util/u_tile.c +++ b/src/gallium/auxiliary/util/u_tile.c @@ -62,7 +62,7 @@ pipe_get_tile_raw(struct pipe_transfer *pt, if(!src) return; - pipe_copy_rect(dst, &pt->block, dst_stride, 0, 0, w, h, src, pt->stride, x, y); + util_copy_rect(dst, &pt->block, dst_stride, 0, 0, w, h, src, pt->stride, x, y); screen->transfer_unmap(screen, pt); } @@ -90,7 +90,7 @@ pipe_put_tile_raw(struct pipe_transfer *pt, if(!dst) return; - pipe_copy_rect(dst, &pt->block, pt->stride, x, y, w, h, src, src_stride, 0, 0); + util_copy_rect(dst, &pt->block, pt->stride, x, y, w, h, src, src_stride, 0, 0); screen->transfer_unmap(screen, pt); } diff --git a/src/gallium/auxiliary/util/u_time.c b/src/gallium/auxiliary/util/u_time.c index 5268cbf79ce..b958a986353 100644 --- a/src/gallium/auxiliary/util/u_time.c +++ b/src/gallium/auxiliary/util/u_time.c @@ -35,7 +35,7 @@ #include "pipe/p_config.h" -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) #include <sys/time.h> #elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) #include <windows.h> @@ -77,7 +77,7 @@ util_time_get_frequency(void) void util_time_get(struct util_time *t) { -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) gettimeofday(&t->tv, NULL); #elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) LONGLONG temp; @@ -102,7 +102,7 @@ util_time_add(const struct util_time *t1, int64_t usecs, struct util_time *t2) { -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) t2->tv.tv_sec = t1->tv.tv_sec + usecs / 1000000; t2->tv.tv_usec = t1->tv.tv_usec + usecs % 1000000; #elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE) @@ -124,7 +124,7 @@ int64_t util_time_diff(const struct util_time *t1, const struct util_time *t2) { -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) return (t2->tv.tv_usec - t1->tv.tv_usec) + (t2->tv.tv_sec - t1->tv.tv_sec)*1000000; #elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE) @@ -144,7 +144,7 @@ util_time_micros( void ) util_time_get(&t1); -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) return t1.tv.tv_usec + t1.tv.tv_sec*1000000LL; #elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE) util_time_get_frequency(); @@ -166,7 +166,7 @@ static INLINE int util_time_compare(const struct util_time *t1, const struct util_time *t2) { -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) if (t1->tv.tv_sec < t2->tv.tv_sec) return -1; else if(t1->tv.tv_sec > t2->tv.tv_sec) diff --git a/src/gallium/auxiliary/util/u_time.h b/src/gallium/auxiliary/util/u_time.h index 6bca6077a2a..a6189a247bb 100644 --- a/src/gallium/auxiliary/util/u_time.h +++ b/src/gallium/auxiliary/util/u_time.h @@ -38,11 +38,16 @@ #include "pipe/p_config.h" -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) #include <time.h> /* timeval */ #include <unistd.h> /* usleep */ #endif +#if defined(PIPE_OS_HAIKU) +#include <sys/time.h> /* timeval */ +#include <unistd.h> +#endif + #include "pipe/p_compiler.h" @@ -58,7 +63,7 @@ extern "C" { */ struct util_time { -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) struct timeval tv; #else int64_t counter; @@ -89,7 +94,7 @@ util_time_timeout(const struct util_time *start, const struct util_time *end, const struct util_time *curr); -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) #define util_time_sleep usleep #else void diff --git a/src/gallium/auxiliary/util/u_timed_winsys.c b/src/gallium/auxiliary/util/u_timed_winsys.c index 77b2a3a1c87..178acdca4df 100644 --- a/src/gallium/auxiliary/util/u_timed_winsys.c +++ b/src/gallium/auxiliary/util/u_timed_winsys.c @@ -212,13 +212,14 @@ timed_surface_buffer_create(struct pipe_winsys *winsys, unsigned width, unsigned height, enum pipe_format format, unsigned usage, + unsigned tex_usage, unsigned *stride) { struct pipe_winsys *backend = timed_winsys(winsys)->backend; uint64_t start = time_start(); struct pipe_buffer *ret = backend->surface_buffer_create( backend, width, height, - format, usage, stride ); + format, usage, tex_usage, stride ); time_finish(winsys, start, 7, __FUNCTION__); diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h index 1f6860da119..d5f5c7bbba8 100644 --- a/src/gallium/drivers/cell/common.h +++ b/src/gallium/drivers/cell/common.h @@ -327,7 +327,7 @@ struct cell_command_sampler opcode_t opcode; /**< CELL_CMD_STATE_SAMPLER */ uint unit; struct pipe_sampler_state state; - uint32_t pad_[1]; + uint32_t pad_[2]; }; diff --git a/src/gallium/drivers/cell/ppu/cell_gen_fp.c b/src/gallium/drivers/cell/ppu/cell_gen_fp.c index 5a889a6119d..312621fd534 100644 --- a/src/gallium/drivers/cell/ppu/cell_gen_fp.c +++ b/src/gallium/drivers/cell/ppu/cell_gen_fp.c @@ -1767,7 +1767,7 @@ emit_instruction(struct codegen *gen, case TGSI_OPCODE_MAD: return emit_MAD(gen, inst); case TGSI_OPCODE_LERP: - return emit_LERP(gen, inst); + return emit_LRP(gen, inst); case TGSI_OPCODE_DP3: return emit_DP3(gen, inst); case TGSI_OPCODE_DP4: @@ -1810,9 +1810,9 @@ emit_instruction(struct codegen *gen, return emit_function_call(gen, inst, "spu_sin", 1, TRUE); case TGSI_OPCODE_POW: return emit_function_call(gen, inst, "spu_pow", 2, TRUE); - case TGSI_OPCODE_EXPBASE2: + case TGSI_OPCODE_EX2: return emit_function_call(gen, inst, "spu_exp2", 1, TRUE); - case TGSI_OPCODE_LOGBASE2: + case TGSI_OPCODE_LG2: return emit_function_call(gen, inst, "spu_log2", 1, TRUE); case TGSI_OPCODE_TEX: /* fall-through for now */ @@ -1834,9 +1834,9 @@ emit_instruction(struct codegen *gen, case TGSI_OPCODE_ENDIF: return emit_ENDIF(gen, inst); - case TGSI_OPCODE_BGNLOOP2: + case TGSI_OPCODE_BGNLOOP: return emit_BGNLOOP(gen, inst); - case TGSI_OPCODE_ENDLOOP2: + case TGSI_OPCODE_ENDLOOP: return emit_ENDLOOP(gen, inst); case TGSI_OPCODE_BRK: return emit_BRK(gen, inst); @@ -1875,9 +1875,9 @@ emit_immediate(struct codegen *gen, const struct tgsi_full_immediate *immed) assert(gen->num_imm < MAX_TEMPS); for (ch = 0; ch < 4; ch++) { - float val = immed->u.ImmediateFloat32[ch].Float; + float val = immed->u[ch].Float; - if (ch > 0 && val == immed->u.ImmediateFloat32[ch - 1].Float) { + if (ch > 0 && val == immed->u[ch - 1].Float) { /* re-use previous register */ gen->imm_regs[gen->num_imm][ch] = gen->imm_regs[gen->num_imm][ch - 1]; } diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c index e26594448f0..6a63a0e6ced 100644 --- a/src/gallium/drivers/cell/ppu/cell_texture.c +++ b/src/gallium/drivers/cell/ppu/cell_texture.c @@ -44,13 +44,6 @@ -static unsigned -minify(unsigned d) -{ - return MAX2(1, d>>1); -} - - static void cell_texture_layout(struct cell_texture *ct) { @@ -424,7 +417,8 @@ cell_transfer_map(struct pipe_screen *screen, struct pipe_transfer *transfer) if (!ctrans->map) return NULL; /* out of memory */ - if (transfer->usage & PIPE_TRANSFER_READ) { + if (transfer->usage == PIPE_TRANSFER_READ || + transfer->usage == PIPE_TRANSFER_READ_WRITE) { /* need to untwiddle the texture to make a linear version */ const uint bpp = pf_get_size(ct->base.format); if (bpp == 4) { @@ -465,7 +459,8 @@ cell_transfer_unmap(struct pipe_screen *screen, PIPE_BUFFER_USAGE_CPU_READ); } - if (transfer->usage & PIPE_TRANSFER_WRITE) { + if (transfer->usage == PIPE_TRANSFER_WRITE || + transfer->usage == PIPE_TRANSFER_READ_WRITE) { /* The user wrote new texture data into the mapped buffer. * We need to convert the new linear data into the twiddled/tiled format. */ diff --git a/src/gallium/drivers/cell/spu/spu_exec.c b/src/gallium/drivers/cell/spu/spu_exec.c index e27df2dfb38..0eaae2e451b 100644 --- a/src/gallium/drivers/cell/spu/spu_exec.c +++ b/src/gallium/drivers/cell/spu/spu_exec.c @@ -952,7 +952,6 @@ exec_instruction( break; case TGSI_OPCODE_RCP: - /* TGSI_OPCODE_RECIP */ FETCH( &r[0], 0, CHAN_X ); r[0].q = micro_div(mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].q, r[0].q); FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { @@ -961,7 +960,6 @@ exec_instruction( break; case TGSI_OPCODE_RSQ: - /* TGSI_OPCODE_RECIPSQRT */ FETCH( &r[0], 0, CHAN_X ); r[0].q = micro_sqrt(r[0].q); r[0].q = micro_div(mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].q, r[0].q); @@ -1115,7 +1113,6 @@ exec_instruction( break; case TGSI_OPCODE_MAD: - /* TGSI_OPCODE_MADD */ FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); @@ -1136,8 +1133,7 @@ exec_instruction( } break; - case TGSI_OPCODE_LERP: - /* TGSI_OPCODE_LRP */ + case TGSI_OPCODE_LRP: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH(&r[0], 0, chan_index); FETCH(&r[1], 1, chan_index); @@ -1154,25 +1150,11 @@ exec_instruction( ASSERT (0); break; - case TGSI_OPCODE_CND0: + case TGSI_OPCODE_DP2A: ASSERT (0); break; - case TGSI_OPCODE_DOT2ADD: - /* TGSI_OPCODE_DP2A */ - ASSERT (0); - break; - - case TGSI_OPCODE_INDEX: - ASSERT (0); - break; - - case TGSI_OPCODE_NEGATE: - ASSERT (0); - break; - - case TGSI_OPCODE_FRAC: - /* TGSI_OPCODE_FRC */ + case TGSI_OPCODE_FRC: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); r[0].q = micro_frc(r[0].q); @@ -1184,8 +1166,7 @@ exec_instruction( ASSERT (0); break; - case TGSI_OPCODE_FLOOR: - /* TGSI_OPCODE_FLR */ + case TGSI_OPCODE_FLR: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); r[0].q = micro_flr(r[0].q); @@ -1201,8 +1182,7 @@ exec_instruction( } break; - case TGSI_OPCODE_EXPBASE2: - /* TGSI_OPCODE_EX2 */ + case TGSI_OPCODE_EX2: FETCH(&r[0], 0, CHAN_X); r[0].q = micro_pow(mach->Temps[TEMP_2_I].xyzw[TEMP_2_C].q, r[0].q); @@ -1212,8 +1192,7 @@ exec_instruction( } break; - case TGSI_OPCODE_LOGBASE2: - /* TGSI_OPCODE_LG2 */ + case TGSI_OPCODE_LG2: FETCH( &r[0], 0, CHAN_X ); r[0].q = micro_lg2(r[0].q); FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { @@ -1221,8 +1200,7 @@ exec_instruction( } break; - case TGSI_OPCODE_POWER: - /* TGSI_OPCODE_POW */ + case TGSI_OPCODE_POW: FETCH(&r[0], 0, CHAN_X); FETCH(&r[1], 1, CHAN_X); @@ -1233,7 +1211,7 @@ exec_instruction( } break; - case TGSI_OPCODE_CROSSPRODUCT: + case TGSI_OPCODE_XPD: /* TGSI_OPCODE_XPD */ FETCH(&r[0], 0, CHAN_Y); FETCH(&r[1], 1, CHAN_Z); @@ -1275,10 +1253,6 @@ exec_instruction( } break; - case TGSI_OPCODE_MULTIPLYMATRIX: - ASSERT (0); - break; - case TGSI_OPCODE_ABS: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH(&r[0], 0, chan_index); @@ -1780,9 +1754,9 @@ exec_instruction( mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]] = 0; break; - case TGSI_OPCODE_LOOP: + case TGSI_OPCODE_BGNFOR: /* fall-through (for now) */ - case TGSI_OPCODE_BGNLOOP2: + case TGSI_OPCODE_BGNLOOP: /* push LoopMask and ContMasks */ ASSERT(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; @@ -1790,9 +1764,9 @@ exec_instruction( mach->ContStack[mach->ContStackTop++] = mach->ContMask; break; - case TGSI_OPCODE_ENDLOOP: + case TGSI_OPCODE_ENDFOR: /* fall-through (for now at least) */ - case TGSI_OPCODE_ENDLOOP2: + case TGSI_OPCODE_ENDLOOP: /* Restore ContMask, but don't pop */ ASSERT(mach->ContStackTop > 0); mach->ContMask = mach->ContStack[mach->ContStackTop - 1]; diff --git a/src/gallium/drivers/cell/spu/spu_render.c b/src/gallium/drivers/cell/spu/spu_render.c index 7c225e2f27c..5ffb7073abf 100644 --- a/src/gallium/drivers/cell/spu/spu_render.c +++ b/src/gallium/drivers/cell/spu/spu_render.c @@ -32,6 +32,7 @@ #include "spu_main.h" #include "spu_render.h" +#include "spu_shuffle.h" #include "spu_tri.h" #include "spu_tile.h" #include "cell/common.h" @@ -267,15 +268,75 @@ cmd_render(const struct cell_command_render *render, uint *pos_incr) uint drawn = 0; - /* loop over tris */ - for (j = 0; j < render->num_indexes; j += 3) { - const float *v0, *v1, *v2; - - v0 = (const float *) (vertices + indexes[j+0] * vertex_size); - v1 = (const float *) (vertices + indexes[j+1] * vertex_size); - v2 = (const float *) (vertices + indexes[j+2] * vertex_size); - - drawn += tri_draw(v0, v1, v2, tx, ty); + const qword vertex_sizes = (qword)spu_splats(vertex_size); + const qword verticess = (qword)spu_splats((uint)vertices); + + ASSERT_ALIGN16(&indexes[0]); + + const uint num_indexes = render->num_indexes; + + /* loop over tris + * &indexes[0] will be 16 byte aligned. This loop is heavily unrolled + * avoiding variable rotates when extracting vertex indices. + */ + for (j = 0; j < num_indexes; j += 24) { + /* Load three vectors, containing 24 ushort indices */ + const qword* lower_qword = (qword*)&indexes[j]; + const qword indices0 = lower_qword[0]; + const qword indices1 = lower_qword[1]; + const qword indices2 = lower_qword[2]; + + /* stores three indices for each tri n in slots 0, 1 and 2 of vsn */ + /* Straightforward rotates for these */ + qword vs0 = indices0; + qword vs1 = si_shlqbyi(indices0, 6); + qword vs3 = si_shlqbyi(indices1, 2); + qword vs4 = si_shlqbyi(indices1, 8); + qword vs6 = si_shlqbyi(indices2, 4); + qword vs7 = si_shlqbyi(indices2, 10); + + /* For tri 2 and 5, the three indices are split across two machine + * words - rotate and combine */ + const qword tmp2a = si_shlqbyi(indices0, 12); + const qword tmp2b = si_rotqmbyi(indices1, 12|16); + qword vs2 = si_selb(tmp2a, tmp2b, si_fsmh(si_from_uint(0x20))); + + const qword tmp5a = si_shlqbyi(indices1, 14); + const qword tmp5b = si_rotqmbyi(indices2, 14|16); + qword vs5 = si_selb(tmp5a, tmp5b, si_fsmh(si_from_uint(0x60))); + + /* unpack indices from halfword slots to word slots */ + vs0 = si_shufb(vs0, vs0, SHUFB8(0,A,0,B,0,C,0,0)); + vs1 = si_shufb(vs1, vs1, SHUFB8(0,A,0,B,0,C,0,0)); + vs2 = si_shufb(vs2, vs2, SHUFB8(0,A,0,B,0,C,0,0)); + vs3 = si_shufb(vs3, vs3, SHUFB8(0,A,0,B,0,C,0,0)); + vs4 = si_shufb(vs4, vs4, SHUFB8(0,A,0,B,0,C,0,0)); + vs5 = si_shufb(vs5, vs5, SHUFB8(0,A,0,B,0,C,0,0)); + vs6 = si_shufb(vs6, vs6, SHUFB8(0,A,0,B,0,C,0,0)); + vs7 = si_shufb(vs7, vs7, SHUFB8(0,A,0,B,0,C,0,0)); + + /* Calculate address of vertex in vertices[] */ + vs0 = si_mpya(vs0, vertex_sizes, verticess); + vs1 = si_mpya(vs1, vertex_sizes, verticess); + vs2 = si_mpya(vs2, vertex_sizes, verticess); + vs3 = si_mpya(vs3, vertex_sizes, verticess); + vs4 = si_mpya(vs4, vertex_sizes, verticess); + vs5 = si_mpya(vs5, vertex_sizes, verticess); + vs6 = si_mpya(vs6, vertex_sizes, verticess); + vs7 = si_mpya(vs7, vertex_sizes, verticess); + + /* Select the appropriate call based on the number of vertices + * remaining */ + switch(num_indexes - j) { + default: drawn += tri_draw(vs7, tx, ty); + case 21: drawn += tri_draw(vs6, tx, ty); + case 18: drawn += tri_draw(vs5, tx, ty); + case 15: drawn += tri_draw(vs4, tx, ty); + case 12: drawn += tri_draw(vs3, tx, ty); + case 9: drawn += tri_draw(vs2, tx, ty); + case 6: drawn += tri_draw(vs1, tx, ty); + case 3: drawn += tri_draw(vs0, tx, ty); + } } //printf("SPU %u: drew %u of %u\n", spu.init.id, drawn, render->num_indexes/3); diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c index d727268475e..58be001be4c 100644 --- a/src/gallium/drivers/cell/spu/spu_tri.c +++ b/src/gallium/drivers/cell/spu/spu_tri.c @@ -133,7 +133,15 @@ struct setup_stage { uint tx, ty; /**< position of current tile (x, y) */ - int cliprect_minx, cliprect_maxx, cliprect_miny, cliprect_maxy; + union { + struct { + int cliprect_minx; + int cliprect_miny; + int cliprect_maxx; + int cliprect_maxy; + }; + qword cliprect; + }; struct interp_coef coef[PIPE_MAX_SHADER_INPUTS]; @@ -432,6 +440,41 @@ print_vertex(const struct vertex_header *v) } #endif +/* Returns the minimum of each slot of two vec_float4s as qwords. + * i.e. return[n] = min(q0[n],q1[n]); + */ +static qword +minfq(qword q0, qword q1) +{ + const qword q0q1m = si_fcgt(q0, q1); + return si_selb(q0, q1, q0q1m); +} + +/* Returns the minimum of each slot of three vec_float4s as qwords. + * i.e. return[n] = min(q0[n],q1[n],q2[n]); + */ +static qword +min3fq(qword q0, qword q1, qword q2) +{ + return minfq(minfq(q0, q1), q2); +} + +/* Returns the maximum of each slot of two vec_float4s as qwords. + * i.e. return[n] = min(q0[n],q1[n],q2[n]); + */ +static qword +maxfq(qword q0, qword q1) { + const qword q0q1m = si_fcgt(q0, q1); + return si_selb(q1, q0, q0q1m); +} + +/* Returns the maximum of each slot of three vec_float4s as qwords. + * i.e. return[n] = min(q0[n],q1[n],q2[n]); + */ +static qword +max3fq(qword q0, qword q1, qword q2) { + return maxfq(maxfq(q0, q1), q2); +} /** * Sort vertices from top to bottom. @@ -440,9 +483,7 @@ print_vertex(const struct vertex_header *v) * \return FALSE if tri is totally outside tile, TRUE otherwise */ static boolean -setup_sort_vertices(const struct vertex_header *v0, - const struct vertex_header *v1, - const struct vertex_header *v2) +setup_sort_vertices(const qword vs) { float area, sign; @@ -455,57 +496,57 @@ setup_sort_vertices(const struct vertex_header *v0, } #endif - /* determine bottom to top order of vertices */ { + /* Load the float values for various processing... */ + const qword f0 = (qword)(((const struct vertex_header*)si_to_ptr(vs))->data[0]); + const qword f1 = (qword)(((const struct vertex_header*)si_to_ptr(si_rotqbyi(vs, 4)))->data[0]); + const qword f2 = (qword)(((const struct vertex_header*)si_to_ptr(si_rotqbyi(vs, 8)))->data[0]); + + /* Check if triangle is completely outside the tile bounds + * Find the min and max x and y positions of the three poits */ + const qword minf = min3fq(f0, f1, f2); + const qword maxf = max3fq(f0, f1, f2); + + /* Compare min and max against cliprect vals */ + const qword maxsmins = si_shufb(maxf, minf, SHUFB4(A,B,a,b)); + const qword outside = si_fcgt(maxsmins, si_csflt(setup.cliprect, 0)); + + /* Use a little magic to work out of the tri is visible or not */ + if(si_to_uint(si_xori(si_gb(outside), 0xc))) return FALSE; + + /* determine bottom to top order of vertices */ /* A table of shuffle patterns for putting vertex_header pointers into correct order. Quite magical. */ - const vec_uchar16 sort_order_patterns[] = { - SHUFFLE4(A,B,C,C), - SHUFFLE4(C,A,B,C), - SHUFFLE4(A,C,B,C), - SHUFFLE4(B,C,A,C), - SHUFFLE4(B,A,C,C), - SHUFFLE4(C,B,A,C) }; - - /* The vertex_header pointers, packed for easy shuffling later */ - const vec_uint4 vs = {(unsigned)v0, (unsigned)v1, (unsigned)v2}; + const qword sort_order_patterns[] = { + SHUFB4(A,B,C,C), + SHUFB4(C,A,B,C), + SHUFB4(A,C,B,C), + SHUFB4(B,C,A,C), + SHUFB4(B,A,C,C), + SHUFB4(C,B,A,C) }; /* Collate y values into two vectors for comparison. Using only one shuffle constant! ;) */ - const vec_float4 y_02_ = spu_shuffle(v0->data[0], v2->data[0], SHUFFLE4(0,B,b,C)); - const vec_float4 y_10_ = spu_shuffle(v1->data[0], v0->data[0], SHUFFLE4(0,B,b,C)); - const vec_float4 y_012 = spu_shuffle(y_02_, v1->data[0], SHUFFLE4(0,B,b,C)); - const vec_float4 y_120 = spu_shuffle(y_10_, v2->data[0], SHUFFLE4(0,B,b,C)); + const qword y_02_ = si_shufb(f0, f2, SHUFB4(0,B,b,C)); + const qword y_10_ = si_shufb(f1, f0, SHUFB4(0,B,b,C)); + const qword y_012 = si_shufb(y_02_, f1, SHUFB4(0,B,b,C)); + const qword y_120 = si_shufb(y_10_, f2, SHUFB4(0,B,b,C)); /* Perform comparison: {y0,y1,y2} > {y1,y2,y0} */ - const vec_uint4 compare = spu_cmpgt(y_012, y_120); + const qword compare = si_fcgt(y_012, y_120); /* Compress the result of the comparison into 4 bits */ - const vec_uint4 gather = spu_gather(compare); + const qword gather = si_gb(compare); /* Subtract one to attain the index into the LUT. Magical. */ - const unsigned int index = spu_extract(gather, 0) - 1; + const unsigned int index = si_to_uint(gather) - 1; /* Load the appropriate pattern and construct the desired vector. */ - setup.vertex_headers = (qword)spu_shuffle(vs, vs, sort_order_patterns[index]); + setup.vertex_headers = si_shufb(vs, vs, sort_order_patterns[index]); /* Using the result of the comparison, set sign. Very magical. */ - sign = ((si_to_uint(si_cntb((qword)gather)) == 2) ? 1.0f : -1.0f); + sign = ((si_to_uint(si_cntb(gather)) == 2) ? 1.0f : -1.0f); } - /* Check if triangle is completely outside the tile bounds */ - if (spu_extract(setup.vmin->data[0], 1) > setup.cliprect_maxy) - return FALSE; - if (spu_extract(setup.vmax->data[0], 1) < setup.cliprect_miny) - return FALSE; - if (spu_extract(setup.vmin->data[0], 0) < setup.cliprect_minx && - spu_extract(setup.vmid->data[0], 0) < setup.cliprect_minx && - spu_extract(setup.vmax->data[0], 0) < setup.cliprect_minx) - return FALSE; - if (spu_extract(setup.vmin->data[0], 0) > setup.cliprect_maxx && - spu_extract(setup.vmid->data[0], 0) > setup.cliprect_maxx && - spu_extract(setup.vmax->data[0], 0) > setup.cliprect_maxx) - return FALSE; - setup.ebot.ds = spu_sub(setup.vmid->data[0], setup.vmin->data[0]); setup.emaj.ds = spu_sub(setup.vmax->data[0], setup.vmin->data[0]); setup.etop.ds = spu_sub(setup.vmax->data[0], setup.vmid->data[0]); @@ -761,21 +802,19 @@ subtriangle(struct edge *eleft, struct edge *eright, unsigned lines) * The tile data should have already been fetched. */ boolean -tri_draw(const float *v0, const float *v1, const float *v2, +tri_draw(const qword vs, uint tx, uint ty) { setup.tx = tx; setup.ty = ty; /* set clipping bounds to tile bounds */ - setup.cliprect_minx = tx * TILE_SIZE; - setup.cliprect_miny = ty * TILE_SIZE; - setup.cliprect_maxx = (tx + 1) * TILE_SIZE; - setup.cliprect_maxy = (ty + 1) * TILE_SIZE; - - if (!setup_sort_vertices((struct vertex_header *) v0, - (struct vertex_header *) v1, - (struct vertex_header *) v2)) { + const qword clipbase = (qword)((vec_uint4){tx, ty}); + const qword clipmin = si_mpyui(clipbase, TILE_SIZE); + const qword clipmax = si_ai(clipmin, TILE_SIZE); + setup.cliprect = si_shufb(clipmin, clipmax, SHUFB4(A,B,a,b)); + + if(!setup_sort_vertices(vs)) { return FALSE; /* totally clipped */ } diff --git a/src/gallium/drivers/cell/spu/spu_tri.h b/src/gallium/drivers/cell/spu/spu_tri.h index aa694dd7c93..82e3b19ad7e 100644 --- a/src/gallium/drivers/cell/spu/spu_tri.h +++ b/src/gallium/drivers/cell/spu/spu_tri.h @@ -31,7 +31,7 @@ extern boolean -tri_draw(const float *v0, const float *v1, const float *v2, uint tx, uint ty); +tri_draw(const qword vs, uint tx, uint ty); #endif /* SPU_TRI_H */ diff --git a/src/gallium/drivers/i915simple/Makefile b/src/gallium/drivers/i915simple/Makefile index 8870b398661..fb533c17961 100644 --- a/src/gallium/drivers/i915simple/Makefile +++ b/src/gallium/drivers/i915simple/Makefile @@ -5,6 +5,7 @@ LIBNAME = i915simple C_SOURCES = \ i915_blit.c \ + i915_buffer.c \ i915_clear.c \ i915_flush.c \ i915_context.c \ diff --git a/src/gallium/drivers/i915simple/SConscript b/src/gallium/drivers/i915simple/SConscript index 2366e1247f2..778c4ed0fde 100644 --- a/src/gallium/drivers/i915simple/SConscript +++ b/src/gallium/drivers/i915simple/SConscript @@ -6,6 +6,7 @@ i915simple = env.ConvenienceLibrary( target = 'i915simple', source = [ 'i915_blit.c', + 'i915_buffer.c', 'i915_clear.c', 'i915_context.c', 'i915_debug.c', diff --git a/src/gallium/drivers/i915simple/i915_batch.h b/src/gallium/drivers/i915simple/i915_batch.h index a433cf054de..b813784723f 100644 --- a/src/gallium/drivers/i915simple/i915_batch.h +++ b/src/gallium/drivers/i915simple/i915_batch.h @@ -28,89 +28,20 @@ #ifndef I915_BATCH_H #define I915_BATCH_H -#include "i915_winsys.h" +#include "intel_batchbuffer.h" -struct i915_batchbuffer -{ - struct pipe_buffer *buffer; - struct i915_winsys *winsys; +#define BEGIN_BATCH(dwords, relocs) \ + (intel_batchbuffer_check(i915->batch, dwords, relocs)) - unsigned char *map; - unsigned char *ptr; +#define OUT_BATCH(dword) \ + intel_batchbuffer_dword(i915->batch, dword) - size_t size; - size_t actual_size; +#define OUT_RELOC(buf, usage, offset) \ + intel_batchbuffer_reloc(i915->batch, buf, usage, offset) - size_t relocs; - size_t max_relocs; -}; - -static INLINE boolean -i915_batchbuffer_check( struct i915_batchbuffer *batch, - size_t dwords, - size_t relocs ) -{ - /** TODO JB: Check relocs */ - return dwords * 4 <= batch->size - (batch->ptr - batch->map); -} - -static INLINE size_t -i915_batchbuffer_space( struct i915_batchbuffer *batch ) -{ - return batch->size - (batch->ptr - batch->map); -} - -static INLINE void -i915_batchbuffer_dword( struct i915_batchbuffer *batch, - unsigned dword ) -{ - if (i915_batchbuffer_space(batch) < 4) - return; - - *(unsigned *)batch->ptr = dword; - batch->ptr += 4; -} - -static INLINE void -i915_batchbuffer_write( struct i915_batchbuffer *batch, - void *data, - size_t size ) -{ - if (i915_batchbuffer_space(batch) < size) - return; - - memcpy(data, batch->ptr, size); - batch->ptr += size; -} - -static INLINE void -i915_batchbuffer_reloc( struct i915_batchbuffer *batch, - struct pipe_buffer *buffer, - size_t flags, - size_t offset ) -{ - batch->winsys->batch_reloc( batch->winsys, buffer, flags, offset ); -} - -static INLINE void -i915_batchbuffer_flush( struct i915_batchbuffer *batch, - struct pipe_fence_handle **fence ) -{ - batch->winsys->batch_flush( batch->winsys, fence ); -} - -#define BEGIN_BATCH( dwords, relocs ) \ - (i915_batchbuffer_check( i915->batch, dwords, relocs )) - -#define OUT_BATCH( dword ) \ - i915_batchbuffer_dword( i915->batch, dword ) - -#define OUT_RELOC( buf, flags, delta ) \ - i915_batchbuffer_reloc( i915->batch, buf, flags, delta ) - -#define FLUSH_BATCH(fence) do { \ - i915->winsys->batch_flush( i915->winsys, fence ); \ - i915->hardware_dirty = ~0; \ +#define FLUSH_BATCH(fence) do { \ + intel_batchbuffer_flush(i915->batch, fence); \ + i915->hardware_dirty = ~0; \ } while (0) #endif diff --git a/src/gallium/drivers/i915simple/i915_blit.c b/src/gallium/drivers/i915simple/i915_blit.c index 448a4708ce8..83dfc335288 100644 --- a/src/gallium/drivers/i915simple/i915_blit.c +++ b/src/gallium/drivers/i915simple/i915_blit.c @@ -26,8 +26,6 @@ **************************************************************************/ -#include "i915_context.h" -#include "i915_winsys.h" #include "i915_blit.h" #include "i915_reg.h" #include "i915_batch.h" @@ -37,33 +35,33 @@ void i915_fill_blit(struct i915_context *i915, - unsigned cpp, - unsigned short dst_pitch, - struct pipe_buffer *dst_buffer, - unsigned dst_offset, - short x, short y, - short w, short h, - unsigned color) + unsigned cpp, + unsigned short dst_pitch, + struct intel_buffer *dst_buffer, + unsigned dst_offset, + short x, short y, + short w, short h, + unsigned color) { unsigned BR13, CMD; I915_DBG(i915, - "%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n", - __FUNCTION__, - dst_buffer, dst_pitch, dst_offset, x, y, w, h); + "%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n", + __FUNCTION__, + dst_buffer, dst_pitch, dst_offset, x, y, w, h); switch (cpp) { case 1: case 2: case 3: BR13 = (((int) dst_pitch) & 0xffff) | - (0xF0 << 16) | (1 << 24); + (0xF0 << 16) | (1 << 24); CMD = XY_COLOR_BLT_CMD; break; case 4: BR13 = (((int) dst_pitch) & 0xffff) | - (0xF0 << 16) | (1 << 24) | (1 << 25); + (0xF0 << 16) | (1 << 24) | (1 << 25); CMD = (XY_COLOR_BLT_CMD | XY_COLOR_BLT_WRITE_ALPHA | XY_COLOR_BLT_WRITE_RGB); break; @@ -79,25 +77,24 @@ i915_fill_blit(struct i915_context *i915, OUT_BATCH(BR13); OUT_BATCH((y << 16) | x); OUT_BATCH(((y + h) << 16) | (x + w)); - OUT_RELOC( dst_buffer, I915_BUFFER_ACCESS_WRITE, dst_offset); + OUT_RELOC(dst_buffer, INTEL_USAGE_2D_TARGET, dst_offset); OUT_BATCH(color); FLUSH_BATCH(NULL); } - void -i915_copy_blit( struct i915_context *i915, - unsigned do_flip, - unsigned cpp, - unsigned short src_pitch, - struct pipe_buffer *src_buffer, - unsigned src_offset, - unsigned short dst_pitch, - struct pipe_buffer *dst_buffer, - unsigned dst_offset, - short src_x, short src_y, - short dst_x, short dst_y, - short w, short h ) +i915_copy_blit(struct i915_context *i915, + unsigned do_flip, + unsigned cpp, + unsigned short src_pitch, + struct intel_buffer *src_buffer, + unsigned src_offset, + unsigned short dst_pitch, + struct intel_buffer *dst_buffer, + unsigned dst_offset, + short src_x, short src_y, + short dst_x, short dst_y, + short w, short h) { unsigned CMD, BR13; int dst_y2 = dst_y + h; @@ -105,32 +102,30 @@ i915_copy_blit( struct i915_context *i915, I915_DBG(i915, - "%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n", - __FUNCTION__, - src_buffer, src_pitch, src_offset, src_x, src_y, - dst_buffer, dst_pitch, dst_offset, dst_x, dst_y, w, h); + "%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n", + __FUNCTION__, + src_buffer, src_pitch, src_offset, src_x, src_y, + dst_buffer, dst_pitch, dst_offset, dst_x, dst_y, w, h); switch (cpp) { case 1: case 2: case 3: BR13 = (((int) dst_pitch) & 0xffff) | - (0xCC << 16) | (1 << 24); + (0xCC << 16) | (1 << 24); CMD = XY_SRC_COPY_BLT_CMD; break; case 4: BR13 = (((int) dst_pitch) & 0xffff) | - (0xCC << 16) | (1 << 24) | (1 << 25); - CMD = - (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA | - XY_SRC_COPY_BLT_WRITE_RGB); + (0xCC << 16) | (1 << 24) | (1 << 25); + CMD = (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA | + XY_SRC_COPY_BLT_WRITE_RGB); break; default: return; } - if (dst_y2 < dst_y || - dst_x2 < dst_x) { + if (dst_y2 < dst_y || dst_x2 < dst_x) { return; } @@ -140,7 +135,6 @@ i915_copy_blit( struct i915_context *i915, */ assert (dst_pitch > 0 && src_pitch > 0); - if (!BEGIN_BATCH(8, 2)) { FLUSH_BATCH(NULL); assert(BEGIN_BATCH(8, 2)); @@ -149,11 +143,9 @@ i915_copy_blit( struct i915_context *i915, OUT_BATCH(BR13); OUT_BATCH((dst_y << 16) | dst_x); OUT_BATCH((dst_y2 << 16) | dst_x2); - OUT_RELOC(dst_buffer, I915_BUFFER_ACCESS_WRITE, dst_offset); + OUT_RELOC(dst_buffer, INTEL_USAGE_2D_TARGET, dst_offset); OUT_BATCH((src_y << 16) | src_x); OUT_BATCH(((int) src_pitch & 0xffff)); - OUT_RELOC(src_buffer, I915_BUFFER_ACCESS_READ, src_offset); + OUT_RELOC(src_buffer, INTEL_USAGE_2D_SOURCE, src_offset); FLUSH_BATCH(NULL); } - - diff --git a/src/gallium/drivers/i915simple/i915_blit.h b/src/gallium/drivers/i915simple/i915_blit.h index 0bb34538611..8ce3220cfd9 100644 --- a/src/gallium/drivers/i915simple/i915_blit.h +++ b/src/gallium/drivers/i915simple/i915_blit.h @@ -32,24 +32,24 @@ extern void i915_copy_blit(struct i915_context *i915, unsigned do_flip, - unsigned cpp, - unsigned short src_pitch, - struct pipe_buffer *src_buffer, - unsigned src_offset, - unsigned short dst_pitch, - struct pipe_buffer *dst_buffer, - unsigned dst_offset, - short srcx, short srcy, - short dstx, short dsty, - short w, short h ); + unsigned cpp, + unsigned short src_pitch, + struct intel_buffer *src_buffer, + unsigned src_offset, + unsigned short dst_pitch, + struct intel_buffer *dst_buffer, + unsigned dst_offset, + short srcx, short srcy, + short dstx, short dsty, + short w, short h); extern void i915_fill_blit(struct i915_context *i915, - unsigned cpp, - unsigned short dst_pitch, - struct pipe_buffer *dst_buffer, - unsigned dst_offset, - short x, short y, - short w, short h, unsigned color); + unsigned cpp, + unsigned short dst_pitch, + struct intel_buffer *dst_buffer, + unsigned dst_offset, + short x, short y, + short w, short h, unsigned color); #endif diff --git a/src/gallium/drivers/i915simple/i915_buffer.c b/src/gallium/drivers/i915simple/i915_buffer.c new file mode 100644 index 00000000000..effeba12972 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_buffer.c @@ -0,0 +1,136 @@ +/************************************************************************** + * + * Copyright © 2009 Jakob Bornecrantz + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "util/u_memory.h" +#include "i915_screen.h" +#include "i915_buffer.h" + +struct intel_buffer; + +struct i915_buffer +{ + struct pipe_buffer base; + + struct intel_buffer *ibuf; /** hw buffer */ + + void *data; /**< user and malloc data */ + boolean own; /**< we own the data incase of malloc */ +}; + +static INLINE struct i915_buffer * +i915_buffer(struct pipe_buffer *buffer) +{ + return (struct i915_buffer *)buffer; +} + +static struct pipe_buffer * +i915_buffer_create(struct pipe_screen *screen, + unsigned alignment, + unsigned usage, + unsigned size) +{ + struct i915_buffer *buf = CALLOC_STRUCT(i915_buffer); + + if (!buf) + return NULL; + + pipe_reference_init(&buf->base.reference, 1); + buf->base.alignment = alignment; + buf->base.screen = screen; + buf->base.usage = usage; + buf->base.size = size; + buf->data = MALLOC(size); + buf->own = TRUE; + + if (!buf->data) + goto err; + + return &buf->base; + +err: + FREE(buf); + return NULL; +} + +static struct pipe_buffer * +i915_user_buffer_create(struct pipe_screen *screen, + void *ptr, + unsigned bytes) +{ + struct i915_buffer *buf = CALLOC_STRUCT(i915_buffer); + + if (!buf) + return NULL; + + pipe_reference_init(&buf->base.reference, 1); + buf->base.alignment = 0; + buf->base.screen = screen; + buf->base.usage = 0; + buf->base.size = bytes; + buf->data = ptr; + buf->own = FALSE; + + return &buf->base; +} + +static void * +i915_buffer_map(struct pipe_screen *screen, + struct pipe_buffer *buffer, + unsigned usage) +{ + struct i915_buffer *buf = i915_buffer(buffer); + assert(!buf->ibuf); + return buf->data; +} + +static void +i915_buffer_unmap(struct pipe_screen *screen, + struct pipe_buffer *buffer) +{ + struct i915_buffer *buf = i915_buffer(buffer); + assert(!buf->ibuf); +} + +static void +i915_buffer_destroy(struct pipe_buffer *buffer) +{ + struct i915_buffer *buf = i915_buffer(buffer); + assert(!buf->ibuf); + + if (buf->own) + FREE(buf->data); + FREE(buf); +} + +void i915_init_screen_buffer_functions(struct i915_screen *screen) +{ + screen->base.buffer_create = i915_buffer_create; + screen->base.user_buffer_create = i915_user_buffer_create; + screen->base.buffer_map = i915_buffer_map; + screen->base.buffer_map_range = NULL; + screen->base.buffer_flush_mapped_range = NULL; + screen->base.buffer_unmap = i915_buffer_unmap; + screen->base.buffer_destroy = i915_buffer_destroy; +} diff --git a/src/gallium/drivers/i915simple/i915_buffer.h b/src/gallium/drivers/i915simple/i915_buffer.h new file mode 100644 index 00000000000..80fda7c62fd --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_buffer.h @@ -0,0 +1,31 @@ +/************************************************************************** + * + * Copyright © 2009 Jakob Bornecrantz + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef I915_BUFFER_H +#define I915_BUFFER_H + +void i915_init_screen_buffer_functions(struct i915_screen *screen); + +#endif diff --git a/src/gallium/drivers/i915simple/i915_context.c b/src/gallium/drivers/i915simple/i915_context.c index ccf9bb31fb0..b43f7352456 100644 --- a/src/gallium/drivers/i915simple/i915_context.c +++ b/src/gallium/drivers/i915simple/i915_context.c @@ -26,8 +26,8 @@ **************************************************************************/ #include "i915_context.h" -#include "i915_winsys.h" #include "i915_state.h" +#include "i915_screen.h" #include "i915_batch.h" #include "i915_texture.h" #include "i915_reg.h" @@ -40,66 +40,58 @@ #include "pipe/p_screen.h" -static void i915_destroy( struct pipe_context *pipe ) -{ - struct i915_context *i915 = i915_context( pipe ); - - draw_destroy( i915->draw ); - - if(i915->winsys->destroy) - i915->winsys->destroy(i915->winsys); - - FREE( i915 ); -} +/* + * Draw functions + */ static boolean i915_draw_range_elements(struct pipe_context *pipe, - struct pipe_buffer *indexBuffer, - unsigned indexSize, - unsigned min_index, - unsigned max_index, - unsigned prim, unsigned start, unsigned count) + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned min_index, + unsigned max_index, + unsigned prim, unsigned start, unsigned count) { - struct i915_context *i915 = i915_context( pipe ); + struct i915_context *i915 = i915_context(pipe); struct draw_context *draw = i915->draw; unsigned i; if (i915->dirty) - i915_update_derived( i915 ); + i915_update_derived(i915); /* * Map vertex buffers */ for (i = 0; i < i915->num_vertex_buffers; i++) { - void *buf - = pipe_buffer_map(pipe->screen, - i915->vertex_buffer[i].buffer, - PIPE_BUFFER_USAGE_CPU_READ); + void *buf = pipe_buffer_map(pipe->screen, i915->vertex_buffer[i].buffer, + PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_vertex_buffer(draw, i, buf); } - /* Map index buffer, if present */ + + /* + * Map index buffer, if present + */ if (indexBuffer) { - void *mapped_indexes - = pipe_buffer_map(pipe->screen, indexBuffer, - PIPE_BUFFER_USAGE_CPU_READ); + void *mapped_indexes = pipe_buffer_map(pipe->screen, indexBuffer, + PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_element_buffer_range(draw, indexSize, - min_index, - max_index, - mapped_indexes); - } - else { - /* no index/element buffer */ + min_index, + max_index, + mapped_indexes); + } else { draw_set_mapped_element_buffer(draw, 0, NULL); } draw_set_mapped_constant_buffer(draw, i915->current.constants[PIPE_SHADER_VERTEX], - ( i915->current.num_user_constants[PIPE_SHADER_VERTEX] * - 4 * sizeof(float) )); + (i915->current.num_user_constants[PIPE_SHADER_VERTEX] * + 4 * sizeof(float))); - /* draw! */ + /* + * Do the drawing + */ draw_arrays(i915->draw, prim, start, count); /* @@ -109,6 +101,7 @@ i915_draw_range_elements(struct pipe_context *pipe, pipe_buffer_unmap(pipe->screen, i915->vertex_buffer[i].buffer); draw_set_mapped_vertex_buffer(draw, i, NULL); } + if (indexBuffer) { pipe_buffer_unmap(pipe->screen, indexBuffer); draw_set_mapped_element_buffer_range(draw, 0, start, start + count - 1, NULL); @@ -118,51 +111,81 @@ i915_draw_range_elements(struct pipe_context *pipe, } static boolean -i915_draw_elements( struct pipe_context *pipe, - struct pipe_buffer *indexBuffer, - unsigned indexSize, - unsigned prim, unsigned start, unsigned count) +i915_draw_elements(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned prim, unsigned start, unsigned count) { - return i915_draw_range_elements( pipe, indexBuffer, - indexSize, - 0, 0xffffffff, - prim, start, count ); + return i915_draw_range_elements(pipe, indexBuffer, + indexSize, + 0, 0xffffffff, + prim, start, count); } -static boolean i915_draw_arrays( struct pipe_context *pipe, - unsigned prim, unsigned start, unsigned count) +static boolean +i915_draw_arrays(struct pipe_context *pipe, + unsigned prim, unsigned start, unsigned count) { return i915_draw_elements(pipe, NULL, 0, prim, start, count); } +/* + * Is referenced functions + */ + + static unsigned int -i915_is_texture_referenced( struct pipe_context *pipe, - struct pipe_texture *texture, - unsigned face, unsigned level) +i915_is_texture_referenced(struct pipe_context *pipe, + struct pipe_texture *texture, + unsigned face, unsigned level) { /** - * FIXME: Optimize. + * FIXME: Return the corrent result. We can't alays return referenced + * since it causes a double flush within the vbo module. */ - +#if 0 return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE; +#else + return 0; +#endif } static unsigned int -i915_is_buffer_referenced( struct pipe_context *pipe, - struct pipe_buffer *buf) +i915_is_buffer_referenced(struct pipe_context *pipe, + struct pipe_buffer *buf) { /** - * FIXME: Optimize. + * FIXME: Return the corrent result. We can't alays return referenced + * since it causes a double flush within the vbo module. */ - +#if 0 return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE; +#else + return 0; +#endif } -struct pipe_context *i915_create_context( struct pipe_screen *screen, - struct pipe_winsys *pipe_winsys, - struct i915_winsys *i915_winsys ) +/* + * Generic context functions + */ + + +static void i915_destroy(struct pipe_context *pipe) +{ + struct i915_context *i915 = i915_context(pipe); + + draw_destroy(i915->draw); + + if(i915->batch) + i915->iws->batchbuffer_destroy(i915->batch); + + FREE(i915); +} + +struct pipe_context * +i915_create_context(struct pipe_screen *screen) { struct i915_context *i915; @@ -170,21 +193,20 @@ struct pipe_context *i915_create_context( struct pipe_screen *screen, if (i915 == NULL) return NULL; - i915->winsys = i915_winsys; - i915->pipe.winsys = pipe_winsys; - i915->pipe.screen = screen; + i915->iws = i915_screen(screen)->iws; + i915->base.winsys = NULL; + i915->base.screen = screen; - i915->pipe.destroy = i915_destroy; + i915->base.destroy = i915_destroy; - i915->pipe.clear = i915_clear; + i915->base.clear = i915_clear; + i915->base.draw_arrays = i915_draw_arrays; + i915->base.draw_elements = i915_draw_elements; + i915->base.draw_range_elements = i915_draw_range_elements; - i915->pipe.draw_arrays = i915_draw_arrays; - i915->pipe.draw_elements = i915_draw_elements; - i915->pipe.draw_range_elements = i915_draw_range_elements; - - i915->pipe.is_texture_referenced = i915_is_texture_referenced; - i915->pipe.is_buffer_referenced = i915_is_buffer_referenced; + i915->base.is_texture_referenced = i915_is_texture_referenced; + i915->base.is_buffer_referenced = i915_is_buffer_referenced; /* * Create drawing context and plug our rendering stage into it. @@ -193,27 +215,23 @@ struct pipe_context *i915_create_context( struct pipe_screen *screen, assert(i915->draw); if (!debug_get_bool_option("I915_NO_VBUF", FALSE)) { draw_set_rasterize_stage(i915->draw, i915_draw_vbuf_stage(i915)); - } - else { + } else { draw_set_rasterize_stage(i915->draw, i915_draw_render_stage(i915)); } i915_init_surface_functions(i915); i915_init_state_functions(i915); i915_init_flush_functions(i915); - i915_init_texture_functions(i915); - draw_install_aaline_stage(i915->draw, &i915->pipe); - draw_install_aapoint_stage(i915->draw, &i915->pipe); + draw_install_aaline_stage(i915->draw, &i915->base); + draw_install_aapoint_stage(i915->draw, &i915->base); i915->dirty = ~0; i915->hardware_dirty = ~0; /* Batch stream debugging is a bit hacked up at the moment: */ - i915->batch = i915_winsys->batch_get(i915_winsys); - i915->batch->winsys = i915_winsys; + i915->batch = i915->iws->batchbuffer_create(i915->iws); - return &i915->pipe; + return &i915->base; } - diff --git a/src/gallium/drivers/i915simple/i915_context.h b/src/gallium/drivers/i915simple/i915_context.h index b6983ba86e7..234b441ce6e 100644 --- a/src/gallium/drivers/i915simple/i915_context.h +++ b/src/gallium/drivers/i915simple/i915_context.h @@ -38,6 +38,11 @@ #include "tgsi/tgsi_scan.h" +struct intel_winsys; +struct intel_buffer; +struct intel_batchbuffer; + + #define I915_TEX_UNITS 8 #define I915_DYNAMIC_MODES4 0 @@ -182,7 +187,6 @@ struct i915_sampler_state { unsigned maxlod; }; - struct i915_texture { struct pipe_texture base; @@ -192,7 +196,8 @@ struct i915_texture { unsigned depth_stride; /* per-image on i945? */ unsigned total_nblocksy; - unsigned tiled; + unsigned sw_tiled; /**< tiled with software flags */ + unsigned hw_tiled; /**< tiled with hardware fences */ unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS]; @@ -206,15 +211,15 @@ struct i915_texture { /* The data is held here: */ - struct pipe_buffer *buffer; + struct intel_buffer *buffer; }; -struct i915_batchbuffer; - struct i915_context { - struct pipe_context pipe; - struct i915_winsys *winsys; + struct pipe_context base; + + struct intel_winsys *iws; + struct draw_context *draw; /* The most recent drawing state as set by the driver: @@ -243,10 +248,10 @@ struct i915_context unsigned num_vertex_elements; unsigned num_vertex_buffers; - struct i915_batchbuffer *batch; + struct intel_batchbuffer *batch; /** Vertex buffer */ - struct pipe_buffer *vbo; + struct intel_buffer *vbo; size_t vbo_offset; unsigned vbo_flushed; diff --git a/src/gallium/drivers/i915simple/i915_debug.c b/src/gallium/drivers/i915simple/i915_debug.c index e08582efaba..ce92d1af9a7 100644 --- a/src/gallium/drivers/i915simple/i915_debug.c +++ b/src/gallium/drivers/i915simple/i915_debug.c @@ -27,7 +27,6 @@ #include "i915_reg.h" #include "i915_context.h" -#include "i915_winsys.h" #include "i915_debug.h" #include "i915_batch.h" #include "pipe/internal/p_winsys_screen.h" @@ -864,7 +863,7 @@ static boolean i915_debug_packet( struct debug_stream *stream ) void -i915_dump_batchbuffer( struct i915_batchbuffer *batch ) +i915_dump_batchbuffer( struct intel_batchbuffer *batch ) { struct debug_stream stream; unsigned *start = (unsigned*)batch->map; diff --git a/src/gallium/drivers/i915simple/i915_debug.h b/src/gallium/drivers/i915simple/i915_debug.h index 16ca7277c7a..dd9b86e17b5 100644 --- a/src/gallium/drivers/i915simple/i915_debug.h +++ b/src/gallium/drivers/i915simple/i915_debug.h @@ -104,9 +104,9 @@ I915_DBG( #endif -struct i915_batchbuffer; +struct intel_batchbuffer; -void i915_dump_batchbuffer( struct i915_batchbuffer *i915 ); +void i915_dump_batchbuffer( struct intel_batchbuffer *i915 ); void i915_debug_init( struct i915_context *i915 ); diff --git a/src/gallium/drivers/i915simple/i915_flush.c b/src/gallium/drivers/i915simple/i915_flush.c index 472e0ab7740..1582168eba5 100644 --- a/src/gallium/drivers/i915simple/i915_flush.c +++ b/src/gallium/drivers/i915simple/i915_flush.c @@ -45,6 +45,7 @@ static void i915_flush( struct pipe_context *pipe, draw_flush(i915->draw); +#if 0 /* Do we need to emit an MI_FLUSH command to flush the hardware * caches? */ @@ -63,6 +64,13 @@ static void i915_flush( struct pipe_context *pipe, } OUT_BATCH( flush ); } +#endif + +#if 0 + if (i915->batch->map == i915->batch->ptr) { + return; + } +#endif /* If there are no flags, just flush pending commands to hardware: */ @@ -74,5 +82,5 @@ static void i915_flush( struct pipe_context *pipe, void i915_init_flush_functions( struct i915_context *i915 ) { - i915->pipe.flush = i915_flush; + i915->base.flush = i915_flush; } diff --git a/src/gallium/drivers/i915simple/i915_fpc_translate.c b/src/gallium/drivers/i915simple/i915_fpc_translate.c index 961c1bf2134..89504ced276 100644 --- a/src/gallium/drivers/i915simple/i915_fpc_translate.c +++ b/src/gallium/drivers/i915simple/i915_fpc_translate.c @@ -975,8 +975,9 @@ i915_translate_instructions(struct i915_fp_compile *p, = &parse.FullToken.FullImmediate; const uint pos = p->num_immediates++; uint j; + assert( imm->Immediate.NrTokens <= 4 + 1 ); for (j = 0; j < imm->Immediate.NrTokens - 1; j++) { - p->immediates[pos][j] = imm->u.ImmediateFloat32[j].Float; + p->immediates[pos][j] = imm->u[j].Float; } } break; diff --git a/src/gallium/drivers/i915simple/i915_prim_emit.c b/src/gallium/drivers/i915simple/i915_prim_emit.c index 8f1f58b2dd1..d9a5c40ab97 100644 --- a/src/gallium/drivers/i915simple/i915_prim_emit.c +++ b/src/gallium/drivers/i915simple/i915_prim_emit.c @@ -32,7 +32,6 @@ #include "util/u_pack_color.h" #include "i915_context.h" -#include "i915_winsys.h" #include "i915_reg.h" #include "i915_state.h" #include "i915_batch.h" diff --git a/src/gallium/drivers/i915simple/i915_prim_vbuf.c b/src/gallium/drivers/i915simple/i915_prim_vbuf.c index 9bdd91f2881..508f4560e48 100644 --- a/src/gallium/drivers/i915simple/i915_prim_vbuf.c +++ b/src/gallium/drivers/i915simple/i915_prim_vbuf.c @@ -42,13 +42,11 @@ #include "draw/draw_vbuf.h" #include "util/u_debug.h" #include "pipe/p_inlines.h" -#include "pipe/internal/p_winsys_screen.h" #include "util/u_math.h" #include "util/u_memory.h" #include "i915_context.h" #include "i915_reg.h" -#include "i915_winsys.h" #include "i915_batch.h" #include "i915_state.h" @@ -59,7 +57,7 @@ struct i915_vbuf_render { struct vbuf_render base; - struct i915_context *i915; + struct i915_context *i915; /** Vertex size in bytes */ size_t vertex_size; @@ -74,7 +72,7 @@ struct i915_vbuf_render { unsigned fallback; /* Stuff for the vbo */ - struct pipe_buffer *vbo; + struct intel_buffer *vbo; size_t vbo_size; size_t vbo_offset; void *vbo_ptr; @@ -87,36 +85,34 @@ struct i915_vbuf_render { * Basically a cast wrapper. */ static INLINE struct i915_vbuf_render * -i915_vbuf_render( struct vbuf_render *render ) +i915_vbuf_render(struct vbuf_render *render) { assert(render); return (struct i915_vbuf_render *)render; } - static const struct vertex_info * -i915_vbuf_render_get_vertex_info( struct vbuf_render *render ) +i915_vbuf_render_get_vertex_info(struct vbuf_render *render) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; if (i915->dirty) { /* make sure we have up to date vertex layout */ - i915_update_derived( i915 ); + i915_update_derived(i915); } return &i915->current.vertex_info; } - static boolean -i915_vbuf_render_allocate_vertices( struct vbuf_render *render, - ushort vertex_size, - ushort nr_vertices ) +i915_vbuf_render_allocate_vertices(struct vbuf_render *render, + ushort vertex_size, + ushort nr_vertices) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; - struct pipe_screen *screen = i915->pipe.screen; + struct intel_winsys *iws = i915->iws; size_t size = (size_t)vertex_size * (size_t)nr_vertices; /* FIXME: handle failure */ @@ -125,17 +121,17 @@ i915_vbuf_render_allocate_vertices( struct vbuf_render *render, if (i915_render->vbo_size > size + i915_render->vbo_offset && !i915->vbo_flushed) { } else { i915->vbo_flushed = 0; - if (i915_render->vbo) - pipe_buffer_reference(&i915_render->vbo, NULL); + if (i915_render->vbo) { + iws->buffer_destroy(iws, i915_render->vbo); + i915_render->vbo = NULL; + } } if (!i915_render->vbo) { i915_render->vbo_size = MAX2(size, i915_render->vbo_alloc_size); i915_render->vbo_offset = 0; - i915_render->vbo = pipe_buffer_create(screen, - 64, - I915_BUFFER_USAGE_LIT_VERTEX, - i915_render->vbo_size); + i915_render->vbo = iws->buffer_create(iws, i915_render->vbo_size, 64, + INTEL_NEW_VERTEX); } @@ -149,40 +145,37 @@ i915_vbuf_render_allocate_vertices( struct vbuf_render *render, return TRUE; } - static void * -i915_vbuf_render_map_vertices( struct vbuf_render *render ) +i915_vbuf_render_map_vertices(struct vbuf_render *render) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; - struct pipe_screen *screen = i915->pipe.screen; + struct intel_winsys *iws = i915->iws; if (i915->vbo_flushed) debug_printf("%s bad vbo flush occured stalling on hw\n"); - i915_render->vbo_ptr = pipe_buffer_map(screen, - i915_render->vbo, - PIPE_BUFFER_USAGE_CPU_WRITE); + i915_render->vbo_ptr = iws->buffer_map(iws, i915_render->vbo, TRUE); return (unsigned char *)i915_render->vbo_ptr + i915->vbo_offset; } static void -i915_vbuf_render_unmap_vertices( struct vbuf_render *render, - ushort min_index, - ushort max_index ) +i915_vbuf_render_unmap_vertices(struct vbuf_render *render, + ushort min_index, + ushort max_index) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; - struct pipe_screen *screen = i915->pipe.screen; + struct intel_winsys *iws = i915->iws; i915_render->vbo_max_used = MAX2(i915_render->vbo_max_used, i915_render->vertex_size * (max_index + 1)); - pipe_buffer_unmap(screen, i915_render->vbo); + iws->buffer_unmap(iws, i915_render->vbo); } static boolean -i915_vbuf_render_set_primitive( struct vbuf_render *render, - unsigned prim ) +i915_vbuf_render_set_primitive(struct vbuf_render *render, + unsigned prim) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); i915_render->prim = prim; @@ -234,15 +227,13 @@ i915_vbuf_render_set_primitive( struct vbuf_render *render, } } - - /** * Used for fallbacks in draw_arrays */ static void -draw_arrays_generate_indices( struct vbuf_render *render, - unsigned start, uint nr, - unsigned type ) +draw_arrays_generate_indices(struct vbuf_render *render, + unsigned start, uint nr, + unsigned type) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; @@ -251,29 +242,29 @@ draw_arrays_generate_indices( struct vbuf_render *render, switch(type) { case 0: for (i = start; i+1 < end; i += 2) - OUT_BATCH( (i+0) | (i+1) << 16 ); + OUT_BATCH((i+0) | (i+1) << 16); if (i < end) - OUT_BATCH( i ); + OUT_BATCH(i); break; case PIPE_PRIM_LINE_LOOP: if (nr >= 2) { - for (i = start + 1; i < end; i++) - OUT_BATCH( (i-0) | (i+0) << 16 ); - OUT_BATCH( (i-0) | ( start) << 16 ); + for (i = start + 1; i < end; i++) + OUT_BATCH((i-0) | (i+0) << 16); + OUT_BATCH((i-0) | ( start) << 16); } break; case PIPE_PRIM_QUADS: for (i = start; i + 3 < end; i += 4) { - OUT_BATCH( (i+0) | (i+1) << 16 ); - OUT_BATCH( (i+3) | (i+1) << 16 ); - OUT_BATCH( (i+2) | (i+3) << 16 ); + OUT_BATCH((i+0) | (i+1) << 16); + OUT_BATCH((i+3) | (i+1) << 16); + OUT_BATCH((i+2) | (i+3) << 16); } break; case PIPE_PRIM_QUAD_STRIP: for (i = start; i + 3 < end; i += 2) { - OUT_BATCH( (i+0) | (i+1) << 16 ); - OUT_BATCH( (i+3) | (i+2) << 16 ); - OUT_BATCH( (i+0) | (i+3) << 16 ); + OUT_BATCH((i+0) | (i+1) << 16); + OUT_BATCH((i+3) | (i+2) << 16); + OUT_BATCH((i+0) | (i+3) << 16); } break; default: @@ -282,16 +273,16 @@ draw_arrays_generate_indices( struct vbuf_render *render, } static unsigned -draw_arrays_calc_nr_indices( uint nr, unsigned type ) +draw_arrays_calc_nr_indices(uint nr, unsigned type) { switch (type) { case 0: return nr; case PIPE_PRIM_LINE_LOOP: if (nr >= 2) - return nr * 2; + return nr * 2; else - return 0; + return 0; case PIPE_PRIM_QUADS: return (nr / 4) * 6; case PIPE_PRIM_QUAD_STRIP: @@ -303,64 +294,64 @@ draw_arrays_calc_nr_indices( uint nr, unsigned type ) } static void -draw_arrays_fallback( struct vbuf_render *render, - unsigned start, - uint nr ) +draw_arrays_fallback(struct vbuf_render *render, + unsigned start, + uint nr) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; unsigned nr_indices; if (i915->dirty) - i915_update_derived( i915 ); + i915_update_derived(i915); if (i915->hardware_dirty) - i915_emit_hardware_state( i915 ); + i915_emit_hardware_state(i915); - nr_indices = draw_arrays_calc_nr_indices( nr, i915_render->fallback ); + nr_indices = draw_arrays_calc_nr_indices(nr, i915_render->fallback); if (!nr_indices) return; - if (!BEGIN_BATCH( 1 + (nr_indices + 1)/2, 1 )) { + if (!BEGIN_BATCH(1 + (nr_indices + 1)/2, 1)) { FLUSH_BATCH(NULL); /* Make sure state is re-emitted after a flush: */ - i915_update_derived( i915 ); - i915_emit_hardware_state( i915 ); + i915_update_derived(i915); + i915_emit_hardware_state(i915); i915->vbo_flushed = 1; - if (!BEGIN_BATCH( 1 + (nr_indices + 1)/2, 1 )) { - assert(0); - goto out; + if (!BEGIN_BATCH(1 + (nr_indices + 1)/2, 1)) { + assert(0); + goto out; } } - OUT_BATCH( _3DPRIMITIVE | - PRIM_INDIRECT | - i915_render->hwprim | - PRIM_INDIRECT_ELTS | - nr_indices ); + OUT_BATCH(_3DPRIMITIVE | + PRIM_INDIRECT | + i915_render->hwprim | + PRIM_INDIRECT_ELTS | + nr_indices); - draw_arrays_generate_indices( render, start, nr, i915_render->fallback ); + draw_arrays_generate_indices(render, start, nr, i915_render->fallback); out: return; } static void -i915_vbuf_render_draw_arrays( struct vbuf_render *render, - unsigned start, - uint nr ) +i915_vbuf_render_draw_arrays(struct vbuf_render *render, + unsigned start, + uint nr) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); if (i915_render->fallback) { - draw_arrays_fallback( render, start, nr ); + draw_arrays_fallback(render, start, nr); return; } /* JB: TODO submit direct cmds */ - draw_arrays_fallback( render, start, nr ); + draw_arrays_fallback(render, start, nr); } /** @@ -368,10 +359,10 @@ i915_vbuf_render_draw_arrays( struct vbuf_render *render, * If type is zero normal operation assumed. */ static void -draw_generate_indices( struct vbuf_render *render, - const ushort *indices, - uint nr_indices, - unsigned type ) +draw_generate_indices(struct vbuf_render *render, + const ushort *indices, + uint nr_indices, + unsigned type) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; @@ -380,31 +371,31 @@ draw_generate_indices( struct vbuf_render *render, switch(type) { case 0: for (i = 0; i + 1 < nr_indices; i += 2) { - OUT_BATCH( indices[i] | indices[i+1] << 16 ); + OUT_BATCH(indices[i] | indices[i+1] << 16); } if (i < nr_indices) { - OUT_BATCH( indices[i] ); + OUT_BATCH(indices[i]); } break; case PIPE_PRIM_LINE_LOOP: if (nr_indices >= 2) { - for (i = 1; i < nr_indices; i++) - OUT_BATCH( indices[i-1] | indices[i] << 16 ); - OUT_BATCH( indices[i-1] | indices[0] << 16 ); + for (i = 1; i < nr_indices; i++) + OUT_BATCH(indices[i-1] | indices[i] << 16); + OUT_BATCH(indices[i-1] | indices[0] << 16); } break; case PIPE_PRIM_QUADS: for (i = 0; i + 3 < nr_indices; i += 4) { - OUT_BATCH( indices[i+0] | indices[i+1] << 16 ); - OUT_BATCH( indices[i+3] | indices[i+1] << 16 ); - OUT_BATCH( indices[i+2] | indices[i+3] << 16 ); + OUT_BATCH(indices[i+0] | indices[i+1] << 16); + OUT_BATCH(indices[i+3] | indices[i+1] << 16); + OUT_BATCH(indices[i+2] | indices[i+3] << 16); } break; case PIPE_PRIM_QUAD_STRIP: for (i = 0; i + 3 < nr_indices; i += 2) { - OUT_BATCH( indices[i+0] | indices[i+1] << 16 ); - OUT_BATCH( indices[i+3] | indices[i+2] << 16 ); - OUT_BATCH( indices[i+0] | indices[i+3] << 16 ); + OUT_BATCH(indices[i+0] | indices[i+1] << 16); + OUT_BATCH(indices[i+3] | indices[i+2] << 16); + OUT_BATCH(indices[i+0] | indices[i+3] << 16); } break; default: @@ -414,16 +405,16 @@ draw_generate_indices( struct vbuf_render *render, } static unsigned -draw_calc_nr_indices( uint nr_indices, unsigned type ) +draw_calc_nr_indices(uint nr_indices, unsigned type) { switch (type) { case 0: return nr_indices; case PIPE_PRIM_LINE_LOOP: if (nr_indices >= 2) - return nr_indices * 2; + return nr_indices * 2; else - return 0; + return 0; case PIPE_PRIM_QUADS: return (nr_indices / 4) * 6; case PIPE_PRIM_QUAD_STRIP: @@ -435,9 +426,9 @@ draw_calc_nr_indices( uint nr_indices, unsigned type ) } static void -i915_vbuf_render_draw( struct vbuf_render *render, - const ushort *indices, - uint nr_indices) +i915_vbuf_render_draw(struct vbuf_render *render, + const ushort *indices, + uint nr_indices) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; @@ -445,48 +436,47 @@ i915_vbuf_render_draw( struct vbuf_render *render, save_nr_indices = nr_indices; - nr_indices = draw_calc_nr_indices( nr_indices, i915_render->fallback ); + nr_indices = draw_calc_nr_indices(nr_indices, i915_render->fallback); if (!nr_indices) return; if (i915->dirty) - i915_update_derived( i915 ); + i915_update_derived(i915); if (i915->hardware_dirty) - i915_emit_hardware_state( i915 ); + i915_emit_hardware_state(i915); - if (!BEGIN_BATCH( 1 + (nr_indices + 1)/2, 1 )) { + if (!BEGIN_BATCH(1 + (nr_indices + 1)/2, 1)) { FLUSH_BATCH(NULL); /* Make sure state is re-emitted after a flush: */ - i915_update_derived( i915 ); - i915_emit_hardware_state( i915 ); + i915_update_derived(i915); + i915_emit_hardware_state(i915); i915->vbo_flushed = 1; - if (!BEGIN_BATCH( 1 + (nr_indices + 1)/2, 1 )) { - assert(0); - goto out; + if (!BEGIN_BATCH(1 + (nr_indices + 1)/2, 1)) { + assert(0); + goto out; } } - OUT_BATCH( _3DPRIMITIVE | - PRIM_INDIRECT | - i915_render->hwprim | - PRIM_INDIRECT_ELTS | - nr_indices ); - draw_generate_indices( render, - indices, - save_nr_indices, - i915_render->fallback ); + OUT_BATCH(_3DPRIMITIVE | + PRIM_INDIRECT | + i915_render->hwprim | + PRIM_INDIRECT_ELTS | + nr_indices); + draw_generate_indices(render, + indices, + save_nr_indices, + i915_render->fallback); out: return; } - static void -i915_vbuf_render_release_vertices( struct vbuf_render *render ) +i915_vbuf_render_release_vertices(struct vbuf_render *render) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; @@ -499,23 +489,21 @@ i915_vbuf_render_release_vertices( struct vbuf_render *render ) i915->dirty |= I915_NEW_VBO; } - static void -i915_vbuf_render_destroy( struct vbuf_render *render ) +i915_vbuf_render_destroy(struct vbuf_render *render) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); FREE(i915_render); } - /** * Create a new primitive render. */ static struct vbuf_render * -i915_vbuf_render_create( struct i915_context *i915 ) +i915_vbuf_render_create(struct i915_context *i915) { struct i915_vbuf_render *i915_render = CALLOC_STRUCT(i915_vbuf_render); - struct pipe_screen *screen = i915->pipe.screen; + struct intel_winsys *iws = i915->iws; i915_render->i915 = i915; @@ -539,23 +527,19 @@ i915_vbuf_render_create( struct i915_context *i915 ) i915_render->vbo_alloc_size = 128 * 4096; i915_render->vbo_size = i915_render->vbo_alloc_size; i915_render->vbo_offset = 0; - i915_render->vbo = pipe_buffer_create(screen, - 64, - I915_BUFFER_USAGE_LIT_VERTEX, - i915_render->vbo_size); - i915_render->vbo_ptr = pipe_buffer_map(screen, - i915_render->vbo, - PIPE_BUFFER_USAGE_CPU_WRITE); - pipe_buffer_unmap(screen, i915_render->vbo); + i915_render->vbo = iws->buffer_create(iws, i915_render->vbo_size, 64, + INTEL_NEW_VERTEX); + /* TODO JB: is this realy needed? */ + i915_render->vbo_ptr = iws->buffer_map(iws, i915_render->vbo, TRUE); + iws->buffer_unmap(iws, i915_render->vbo); return &i915_render->base; } - /** * Create a new primitive vbuf/render stage. */ -struct draw_stage *i915_draw_vbuf_stage( struct i915_context *i915 ) +struct draw_stage *i915_draw_vbuf_stage(struct i915_context *i915) { struct vbuf_render *render; struct draw_stage *stage; @@ -564,7 +548,7 @@ struct draw_stage *i915_draw_vbuf_stage( struct i915_context *i915 ) if(!render) return NULL; - stage = draw_vbuf_stage( i915->draw, render ); + stage = draw_vbuf_stage(i915->draw, render); if(!stage) { render->destroy(render); return NULL; diff --git a/src/gallium/drivers/i915simple/i915_screen.c b/src/gallium/drivers/i915simple/i915_screen.c index f4aa8e60d81..9f017a14cca 100644 --- a/src/gallium/drivers/i915simple/i915_screen.c +++ b/src/gallium/drivers/i915simple/i915_screen.c @@ -26,33 +26,36 @@ **************************************************************************/ -#include "util/u_memory.h" -#include "util/u_simple_screen.h" -#include "pipe/internal/p_winsys_screen.h" #include "pipe/p_inlines.h" +#include "util/u_memory.h" #include "util/u_string.h" #include "i915_reg.h" #include "i915_context.h" #include "i915_screen.h" +#include "i915_buffer.h" #include "i915_texture.h" -#include "i915_winsys.h" +#include "intel_winsys.h" + + +/* + * Probe functions + */ static const char * -i915_get_vendor( struct pipe_screen *pscreen ) +i915_get_vendor(struct pipe_screen *screen) { return "Tungsten Graphics, Inc."; } - static const char * -i915_get_name( struct pipe_screen *pscreen ) +i915_get_name(struct pipe_screen *screen) { static char buffer[128]; const char *chipset; - switch (i915_screen(pscreen)->pci_id) { + switch (i915_screen(screen)->pci_id) { case PCI_CHIP_I915_G: chipset = "915G"; break; @@ -86,7 +89,6 @@ i915_get_name( struct pipe_screen *pscreen ) return buffer; } - static int i915_get_param(struct pipe_screen *screen, int param) { @@ -122,7 +124,6 @@ i915_get_param(struct pipe_screen *screen, int param) } } - static float i915_get_paramf(struct pipe_screen *screen, int param) { @@ -148,13 +149,12 @@ i915_get_paramf(struct pipe_screen *screen, int param) } } - static boolean -i915_is_format_supported( struct pipe_screen *screen, - enum pipe_format format, - enum pipe_texture_target target, - unsigned tex_usage, - unsigned geom_flags ) +i915_is_format_supported(struct pipe_screen *screen, + enum pipe_format format, + enum pipe_texture_target target, + unsigned tex_usage, + unsigned geom_flags) { static const enum pipe_format tex_supported[] = { PIPE_FORMAT_R8G8B8A8_UNORM, @@ -173,7 +173,6 @@ i915_is_format_supported( struct pipe_screen *screen, PIPE_FORMAT_A8R8G8B8_UNORM, PIPE_FORMAT_R5G6B5_UNORM, PIPE_FORMAT_S8Z24_UNORM, - /*PIPE_FORMAT_R16G16B16A16_SNORM,*/ PIPE_FORMAT_NONE /* list terminator */ }; const enum pipe_format *list; @@ -193,120 +192,73 @@ i915_is_format_supported( struct pipe_screen *screen, } -static void -i915_destroy_screen( struct pipe_screen *screen ) -{ - struct pipe_winsys *winsys = screen->winsys; - - if(winsys->destroy) - winsys->destroy(winsys); - - FREE(screen); -} +/* + * Fence functions + */ -static struct pipe_transfer* -i915_get_tex_transfer(struct pipe_screen *screen, - struct pipe_texture *texture, - unsigned face, unsigned level, unsigned zslice, - enum pipe_transfer_usage usage, unsigned x, unsigned y, - unsigned w, unsigned h) +static void +i915_fence_reference(struct pipe_screen *screen, + struct pipe_fence_handle **ptr, + struct pipe_fence_handle *fence) { - struct i915_texture *tex = (struct i915_texture *)texture; - struct i915_transfer *trans; - unsigned offset; /* in bytes */ + struct i915_screen *is = i915_screen(screen); - if (texture->target == PIPE_TEXTURE_CUBE) { - offset = tex->image_offset[level][face]; - } - else if (texture->target == PIPE_TEXTURE_3D) { - offset = tex->image_offset[level][zslice]; - } - else { - offset = tex->image_offset[level][0]; - assert(face == 0); - assert(zslice == 0); - } - - trans = CALLOC_STRUCT(i915_transfer); - if (trans) { - pipe_texture_reference(&trans->base.texture, texture); - trans->base.format = trans->base.format; - trans->base.width = w; - trans->base.height = h; - trans->base.block = texture->block; - trans->base.nblocksx = texture->nblocksx[level]; - trans->base.nblocksy = texture->nblocksy[level]; - trans->base.stride = tex->stride; - trans->offset = offset; - trans->base.usage = usage; - } - return &trans->base; + is->iws->fence_reference(is->iws, ptr, fence); } -static void -i915_tex_transfer_destroy(struct pipe_transfer *trans) +static int +i915_fence_signalled(struct pipe_screen *screen, + struct pipe_fence_handle *fence, + unsigned flags) { - pipe_texture_reference(&trans->texture, NULL); - FREE(trans); + struct i915_screen *is = i915_screen(screen); + + return is->iws->fence_signalled(is->iws, fence); } -static void * -i915_transfer_map( struct pipe_screen *screen, - struct pipe_transfer *transfer ) +static int +i915_fence_finish(struct pipe_screen *screen, + struct pipe_fence_handle *fence, + unsigned flags) { - struct i915_texture *tex = (struct i915_texture *)transfer->texture; - char *map; - unsigned flags = 0; + struct i915_screen *is = i915_screen(screen); - if (transfer->usage != PIPE_TRANSFER_WRITE) - flags |= PIPE_BUFFER_USAGE_CPU_READ; + return is->iws->fence_finish(is->iws, fence); +} - if (transfer->usage != PIPE_TRANSFER_READ) - flags |= PIPE_BUFFER_USAGE_CPU_WRITE; - map = pipe_buffer_map( screen, tex->buffer, flags ); - if (map == NULL) - return NULL; +/* + * Generic functions + */ - if (transfer->texture && - (flags & PIPE_BUFFER_USAGE_CPU_WRITE)) - { - /* Do something to notify contexts of a texture change. - */ - /* i915_screen(screen)->timestamp++; */ - } - - return map + i915_transfer(transfer)->offset + - transfer->y / transfer->block.height * transfer->stride + - transfer->x / transfer->block.width * transfer->block.size; -} static void -i915_transfer_unmap(struct pipe_screen *screen, - struct pipe_transfer *transfer) +i915_destroy_screen(struct pipe_screen *screen) { - struct i915_texture *tex = (struct i915_texture *)transfer->texture; - pipe_buffer_unmap( screen, tex->buffer ); -} + struct i915_screen *is = i915_screen(screen); + if (is->iws) + is->iws->destroy(is->iws); + FREE(is); +} /** * Create a new i915_screen object */ struct pipe_screen * -i915_create_screen(struct pipe_winsys *winsys, uint pci_id) +i915_create_screen(struct intel_winsys *iws, uint pci_id) { - struct i915_screen *i915screen = CALLOC_STRUCT(i915_screen); + struct i915_screen *is = CALLOC_STRUCT(i915_screen); - if (!i915screen) + if (!is) return NULL; switch (pci_id) { case PCI_CHIP_I915_G: case PCI_CHIP_I915_GM: - i915screen->is_i945 = FALSE; + is->is_i945 = FALSE; break; case PCI_CHIP_I945_G: @@ -315,7 +267,7 @@ i915_create_screen(struct pipe_winsys *winsys, uint pci_id) case PCI_CHIP_G33_G: case PCI_CHIP_Q33_G: case PCI_CHIP_Q35_G: - i915screen->is_i945 = TRUE; + is->is_i945 = TRUE; break; default: @@ -324,24 +276,25 @@ i915_create_screen(struct pipe_winsys *winsys, uint pci_id) return NULL; } - i915screen->pci_id = pci_id; + is->pci_id = pci_id; + is->iws = iws; + + is->base.winsys = NULL; - i915screen->screen.winsys = winsys; + is->base.destroy = i915_destroy_screen; - i915screen->screen.destroy = i915_destroy_screen; + is->base.get_name = i915_get_name; + is->base.get_vendor = i915_get_vendor; + is->base.get_param = i915_get_param; + is->base.get_paramf = i915_get_paramf; + is->base.is_format_supported = i915_is_format_supported; - i915screen->screen.get_name = i915_get_name; - i915screen->screen.get_vendor = i915_get_vendor; - i915screen->screen.get_param = i915_get_param; - i915screen->screen.get_paramf = i915_get_paramf; - i915screen->screen.is_format_supported = i915_is_format_supported; - i915screen->screen.get_tex_transfer = i915_get_tex_transfer; - i915screen->screen.tex_transfer_destroy = i915_tex_transfer_destroy; - i915screen->screen.transfer_map = i915_transfer_map; - i915screen->screen.transfer_unmap = i915_transfer_unmap; + is->base.fence_reference = i915_fence_reference; + is->base.fence_signalled = i915_fence_signalled; + is->base.fence_finish = i915_fence_finish; - i915_init_screen_texture_functions(&i915screen->screen); - u_simple_screen_init(&i915screen->screen); + i915_init_screen_texture_functions(is); + i915_init_screen_buffer_functions(is); - return &i915screen->screen; + return &is->base; } diff --git a/src/gallium/drivers/i915simple/i915_screen.h b/src/gallium/drivers/i915simple/i915_screen.h index 5284c325951..5126485caa7 100644 --- a/src/gallium/drivers/i915simple/i915_screen.h +++ b/src/gallium/drivers/i915simple/i915_screen.h @@ -25,17 +25,14 @@ * **************************************************************************/ - #ifndef I915_SCREEN_H #define I915_SCREEN_H - +#include "pipe/p_state.h" #include "pipe/p_screen.h" -#ifdef __cplusplus -extern "C" { -#endif +struct intel_winsys; /** @@ -43,13 +40,14 @@ extern "C" { */ struct i915_screen { - struct pipe_screen screen; + struct pipe_screen base; + + struct intel_winsys *iws; boolean is_i945; uint pci_id; }; - /** * Subclass of pipe_transfer */ @@ -61,7 +59,11 @@ struct i915_transfer }; -/** cast wrappers */ +/* + * Cast wrappers + */ + + static INLINE struct i915_screen * i915_screen(struct pipe_screen *pscreen) { @@ -69,14 +71,10 @@ i915_screen(struct pipe_screen *pscreen) } static INLINE struct i915_transfer * -i915_transfer( struct pipe_transfer *transfer ) +i915_transfer(struct pipe_transfer *transfer) { return (struct i915_transfer *)transfer; } -#ifdef __cplusplus -} -#endif - #endif /* I915_SCREEN_H */ diff --git a/src/gallium/drivers/i915simple/i915_state.c b/src/gallium/drivers/i915simple/i915_state.c index 273e74002aa..0087dfa410f 100644 --- a/src/gallium/drivers/i915simple/i915_state.c +++ b/src/gallium/drivers/i915simple/i915_state.c @@ -518,7 +518,7 @@ static void i915_set_constant_buffer(struct pipe_context *pipe, const struct pipe_constant_buffer *buf) { struct i915_context *i915 = i915_context(pipe); - struct pipe_winsys *ws = pipe->winsys; + struct pipe_screen *screen = pipe->screen; draw_flush(i915->draw); assert(shader < PIPE_SHADER_TYPES); @@ -536,10 +536,10 @@ static void i915_set_constant_buffer(struct pipe_context *pipe, if (buf) { void *mapped; if (buf->buffer && buf->buffer->size && - (mapped = ws->buffer_map(ws, buf->buffer, + (mapped = pipe_buffer_map(screen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ))) { memcpy(i915->current.constants[shader], mapped, buf->buffer->size); - ws->buffer_unmap(ws, buf->buffer); + pipe_buffer_unmap(screen, buf->buffer); i915->current.num_user_constants[shader] = buf->buffer->size / (4 * sizeof(float)); } @@ -751,38 +751,38 @@ static void i915_set_edgeflags(struct pipe_context *pipe, void i915_init_state_functions( struct i915_context *i915 ) { - i915->pipe.set_edgeflags = i915_set_edgeflags; - i915->pipe.create_blend_state = i915_create_blend_state; - i915->pipe.bind_blend_state = i915_bind_blend_state; - i915->pipe.delete_blend_state = i915_delete_blend_state; - - i915->pipe.create_sampler_state = i915_create_sampler_state; - i915->pipe.bind_sampler_states = i915_bind_sampler_states; - i915->pipe.delete_sampler_state = i915_delete_sampler_state; - - i915->pipe.create_depth_stencil_alpha_state = i915_create_depth_stencil_state; - i915->pipe.bind_depth_stencil_alpha_state = i915_bind_depth_stencil_state; - i915->pipe.delete_depth_stencil_alpha_state = i915_delete_depth_stencil_state; - - i915->pipe.create_rasterizer_state = i915_create_rasterizer_state; - i915->pipe.bind_rasterizer_state = i915_bind_rasterizer_state; - i915->pipe.delete_rasterizer_state = i915_delete_rasterizer_state; - i915->pipe.create_fs_state = i915_create_fs_state; - i915->pipe.bind_fs_state = i915_bind_fs_state; - i915->pipe.delete_fs_state = i915_delete_fs_state; - i915->pipe.create_vs_state = i915_create_vs_state; - i915->pipe.bind_vs_state = i915_bind_vs_state; - i915->pipe.delete_vs_state = i915_delete_vs_state; - - i915->pipe.set_blend_color = i915_set_blend_color; - i915->pipe.set_clip_state = i915_set_clip_state; - i915->pipe.set_constant_buffer = i915_set_constant_buffer; - i915->pipe.set_framebuffer_state = i915_set_framebuffer_state; - - i915->pipe.set_polygon_stipple = i915_set_polygon_stipple; - i915->pipe.set_scissor_state = i915_set_scissor_state; - i915->pipe.set_sampler_textures = i915_set_sampler_textures; - i915->pipe.set_viewport_state = i915_set_viewport_state; - i915->pipe.set_vertex_buffers = i915_set_vertex_buffers; - i915->pipe.set_vertex_elements = i915_set_vertex_elements; + i915->base.set_edgeflags = i915_set_edgeflags; + i915->base.create_blend_state = i915_create_blend_state; + i915->base.bind_blend_state = i915_bind_blend_state; + i915->base.delete_blend_state = i915_delete_blend_state; + + i915->base.create_sampler_state = i915_create_sampler_state; + i915->base.bind_sampler_states = i915_bind_sampler_states; + i915->base.delete_sampler_state = i915_delete_sampler_state; + + i915->base.create_depth_stencil_alpha_state = i915_create_depth_stencil_state; + i915->base.bind_depth_stencil_alpha_state = i915_bind_depth_stencil_state; + i915->base.delete_depth_stencil_alpha_state = i915_delete_depth_stencil_state; + + i915->base.create_rasterizer_state = i915_create_rasterizer_state; + i915->base.bind_rasterizer_state = i915_bind_rasterizer_state; + i915->base.delete_rasterizer_state = i915_delete_rasterizer_state; + i915->base.create_fs_state = i915_create_fs_state; + i915->base.bind_fs_state = i915_bind_fs_state; + i915->base.delete_fs_state = i915_delete_fs_state; + i915->base.create_vs_state = i915_create_vs_state; + i915->base.bind_vs_state = i915_bind_vs_state; + i915->base.delete_vs_state = i915_delete_vs_state; + + i915->base.set_blend_color = i915_set_blend_color; + i915->base.set_clip_state = i915_set_clip_state; + i915->base.set_constant_buffer = i915_set_constant_buffer; + i915->base.set_framebuffer_state = i915_set_framebuffer_state; + + i915->base.set_polygon_stipple = i915_set_polygon_stipple; + i915->base.set_scissor_state = i915_set_scissor_state; + i915->base.set_sampler_textures = i915_set_sampler_textures; + i915->base.set_viewport_state = i915_set_viewport_state; + i915->base.set_vertex_buffers = i915_set_vertex_buffers; + i915->base.set_vertex_elements = i915_set_vertex_elements; } diff --git a/src/gallium/drivers/i915simple/i915_state_emit.c b/src/gallium/drivers/i915simple/i915_state_emit.c index 1e1fb968b47..a3d4e3b04e5 100644 --- a/src/gallium/drivers/i915simple/i915_state_emit.c +++ b/src/gallium/drivers/i915simple/i915_state_emit.c @@ -28,7 +28,6 @@ #include "i915_reg.h" #include "i915_context.h" -#include "i915_winsys.h" #include "i915_batch.h" #include "i915_reg.h" @@ -107,7 +106,7 @@ i915_emit_hardware_state(struct i915_context *i915 ) 6 ) * 3/2; /* plus 50% margin */ const unsigned relocs = ( I915_TEX_UNITS + - 3 + 3 ) * 3/2; /* plus 50% margin */ #if 0 @@ -123,9 +122,9 @@ i915_emit_hardware_state(struct i915_context *i915 ) if (i915->hardware_dirty & I915_HW_INVARIENT) { OUT_BATCH(_3DSTATE_AA_CMD | - AA_LINE_ECAAR_WIDTH_ENABLE | - AA_LINE_ECAAR_WIDTH_1_0 | - AA_LINE_REGION_WIDTH_ENABLE | AA_LINE_REGION_WIDTH_1_0); + AA_LINE_ECAAR_WIDTH_ENABLE | + AA_LINE_ECAAR_WIDTH_1_0 | + AA_LINE_REGION_WIDTH_ENABLE | AA_LINE_REGION_WIDTH_1_0); OUT_BATCH(_3DSTATE_DFLT_DIFFUSE_CMD); OUT_BATCH(0); @@ -137,24 +136,24 @@ i915_emit_hardware_state(struct i915_context *i915 ) OUT_BATCH(0); OUT_BATCH(_3DSTATE_COORD_SET_BINDINGS | - CSB_TCB(0, 0) | - CSB_TCB(1, 1) | - CSB_TCB(2, 2) | - CSB_TCB(3, 3) | - CSB_TCB(4, 4) | - CSB_TCB(5, 5) | - CSB_TCB(6, 6) | - CSB_TCB(7, 7)); + CSB_TCB(0, 0) | + CSB_TCB(1, 1) | + CSB_TCB(2, 2) | + CSB_TCB(3, 3) | + CSB_TCB(4, 4) | + CSB_TCB(5, 5) | + CSB_TCB(6, 6) | + CSB_TCB(7, 7)); OUT_BATCH(_3DSTATE_RASTER_RULES_CMD | - ENABLE_POINT_RASTER_RULE | - OGL_POINT_RASTER_RULE | - ENABLE_LINE_STRIP_PROVOKE_VRTX | - ENABLE_TRI_FAN_PROVOKE_VRTX | - LINE_STRIP_PROVOKE_VRTX(1) | - TRI_FAN_PROVOKE_VRTX(2) | - ENABLE_TEXKILL_3D_4D | - TEXKILL_4D); + ENABLE_POINT_RASTER_RULE | + OGL_POINT_RASTER_RULE | + ENABLE_LINE_STRIP_PROVOKE_VRTX | + ENABLE_TRI_FAN_PROVOKE_VRTX | + LINE_STRIP_PROVOKE_VRTX(1) | + TRI_FAN_PROVOKE_VRTX(2) | + ENABLE_TEXKILL_3D_4D | + TEXKILL_4D); /* Need to initialize this to zero. */ @@ -173,21 +172,21 @@ i915_emit_hardware_state(struct i915_context *i915 ) if (i915->hardware_dirty & I915_HW_IMMEDIATE) { OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 | - I1_LOAD_S(0) | - I1_LOAD_S(1) | - I1_LOAD_S(2) | - I1_LOAD_S(4) | - I1_LOAD_S(5) | - I1_LOAD_S(6) | - (5)); + I1_LOAD_S(0) | + I1_LOAD_S(1) | + I1_LOAD_S(2) | + I1_LOAD_S(4) | + I1_LOAD_S(5) | + I1_LOAD_S(6) | + (5)); if(i915->vbo) OUT_RELOC(i915->vbo, - I915_BUFFER_ACCESS_READ, + INTEL_USAGE_VERTEX, i915->current.immediate[I915_IMMEDIATE_S0]); else - /* FIXME: we should not do this */ - OUT_BATCH(0); + /* FIXME: we should not do this */ + OUT_BATCH(0); OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S1]); OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S2]); OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S4]); @@ -200,7 +199,7 @@ i915_emit_hardware_state(struct i915_context *i915 ) { int i; for (i = 0; i < I915_MAX_DYNAMIC; i++) { - OUT_BATCH(i915->current.dynamic[i]); + OUT_BATCH(i915->current.dynamic[i]); } } @@ -211,68 +210,68 @@ i915_emit_hardware_state(struct i915_context *i915 ) struct pipe_surface *depth_surface = i915->framebuffer.zsbuf; if (cbuf_surface) { - unsigned ctile = BUF_3D_USE_FENCE; + unsigned ctile = BUF_3D_USE_FENCE; struct i915_texture *tex = (struct i915_texture *) cbuf_surface->texture; assert(tex); - if (tex && tex->tiled) { - ctile = BUF_3D_TILED_SURFACE; - } + if (tex && tex->sw_tiled) { + ctile = BUF_3D_TILED_SURFACE; + } - OUT_BATCH(_3DSTATE_BUF_INFO_CMD); + OUT_BATCH(_3DSTATE_BUF_INFO_CMD); - OUT_BATCH(BUF_3D_ID_COLOR_BACK | - BUF_3D_PITCH(tex->stride) | /* pitch in bytes */ - ctile); + OUT_BATCH(BUF_3D_ID_COLOR_BACK | + BUF_3D_PITCH(tex->stride) | /* pitch in bytes */ + ctile); - OUT_RELOC(tex->buffer, - I915_BUFFER_ACCESS_WRITE, - cbuf_surface->offset); + OUT_RELOC(tex->buffer, + INTEL_USAGE_RENDER, + cbuf_surface->offset); } /* What happens if no zbuf?? */ if (depth_surface) { - unsigned ztile = BUF_3D_USE_FENCE; + unsigned ztile = BUF_3D_USE_FENCE; struct i915_texture *tex = (struct i915_texture *) depth_surface->texture; assert(tex); - if (tex && tex->tiled) { - ztile = BUF_3D_TILED_SURFACE; - } + if (tex && tex->sw_tiled) { + ztile = BUF_3D_TILED_SURFACE; + } - OUT_BATCH(_3DSTATE_BUF_INFO_CMD); + OUT_BATCH(_3DSTATE_BUF_INFO_CMD); - OUT_BATCH(BUF_3D_ID_DEPTH | - BUF_3D_PITCH(tex->stride) | /* pitch in bytes */ - ztile); + OUT_BATCH(BUF_3D_ID_DEPTH | + BUF_3D_PITCH(tex->stride) | /* pitch in bytes */ + ztile); - OUT_RELOC(tex->buffer, - I915_BUFFER_ACCESS_WRITE, - depth_surface->offset); + OUT_RELOC(tex->buffer, + INTEL_USAGE_RENDER, + depth_surface->offset); } { - unsigned cformat, zformat = 0; + unsigned cformat, zformat = 0; - if (cbuf_surface) + if (cbuf_surface) cformat = cbuf_surface->format; else cformat = PIPE_FORMAT_A8R8G8B8_UNORM; /* arbitrary */ cformat = translate_format(cformat); - if (depth_surface) - zformat = translate_depth_format( i915->framebuffer.zsbuf->format ); + if (depth_surface) + zformat = translate_depth_format( i915->framebuffer.zsbuf->format ); - OUT_BATCH(_3DSTATE_DST_BUF_VARS_CMD); - OUT_BATCH(DSTORG_HORT_BIAS(0x8) | /* .5 */ - DSTORG_VERT_BIAS(0x8) | /* .5 */ - LOD_PRECLAMP_OGL | - TEX_DEFAULT_COLOR_OGL | - cformat | - zformat ); + OUT_BATCH(_3DSTATE_DST_BUF_VARS_CMD); + OUT_BATCH(DSTORG_HORT_BIAS(0x8) | /* .5 */ + DSTORG_VERT_BIAS(0x8) | /* .5 */ + LOD_PRECLAMP_OGL | + TEX_DEFAULT_COLOR_OGL | + cformat | + zformat ); } } @@ -290,16 +289,13 @@ i915_emit_hardware_state(struct i915_context *i915 ) OUT_BATCH(enabled); for (unit = 0; unit < I915_TEX_UNITS; unit++) { if (enabled & (1 << unit)) { - struct pipe_buffer *buf = - i915->texture[unit]->buffer; + struct intel_buffer *buf = i915->texture[unit]->buffer; uint offset = 0; assert(buf); count++; - OUT_RELOC(buf, - I915_BUFFER_ACCESS_READ, - offset); + OUT_RELOC(buf, INTEL_USAGE_SAMPLER, offset); OUT_BATCH(i915->current.texbuffer[unit][0]); /* MS3 */ OUT_BATCH(i915->current.texbuffer[unit][1]); /* MS4 */ } @@ -315,20 +311,20 @@ i915_emit_hardware_state(struct i915_context *i915 ) if (i915->hardware_dirty & I915_HW_SAMPLER) { if (i915->current.sampler_enable_nr) { - int i; - - OUT_BATCH( _3DSTATE_SAMPLER_STATE | - (3 * i915->current.sampler_enable_nr) ); - - OUT_BATCH( i915->current.sampler_enable_flags ); - - for (i = 0; i < I915_TEX_UNITS; i++) { - if (i915->current.sampler_enable_flags & (1<<i)) { - OUT_BATCH( i915->current.sampler[i][0] ); - OUT_BATCH( i915->current.sampler[i][1] ); - OUT_BATCH( i915->current.sampler[i][2] ); - } - } + int i; + + OUT_BATCH( _3DSTATE_SAMPLER_STATE | + (3 * i915->current.sampler_enable_nr) ); + + OUT_BATCH( i915->current.sampler_enable_flags ); + + for (i = 0; i < I915_TEX_UNITS; i++) { + if (i915->current.sampler_enable_flags & (1<<i)) { + OUT_BATCH( i915->current.sampler[i][0] ); + OUT_BATCH( i915->current.sampler[i][1] ); + OUT_BATCH( i915->current.sampler[i][2] ); + } + } } } #endif diff --git a/src/gallium/drivers/i915simple/i915_state_sampler.c b/src/gallium/drivers/i915simple/i915_state_sampler.c index 3667ed1afa7..c5e9084d12e 100644 --- a/src/gallium/drivers/i915simple/i915_state_sampler.c +++ b/src/gallium/drivers/i915simple/i915_state_sampler.c @@ -247,7 +247,7 @@ i915_update_texture(struct i915_context *i915, assert(format); assert(pitch); - if (tex->tiled) { + if (tex->sw_tiled) { assert(!((pitch - 1) & pitch)); tiled = MS3_TILED_SURFACE; } diff --git a/src/gallium/drivers/i915simple/i915_surface.c b/src/gallium/drivers/i915simple/i915_surface.c index 09b2c499b8f..ab8331f3e64 100644 --- a/src/gallium/drivers/i915simple/i915_surface.c +++ b/src/gallium/drivers/i915simple/i915_surface.c @@ -89,6 +89,6 @@ i915_surface_fill(struct pipe_context *pipe, void i915_init_surface_functions(struct i915_context *i915) { - i915->pipe.surface_copy = i915_surface_copy; - i915->pipe.surface_fill = i915_surface_fill; + i915->base.surface_copy = i915_surface_copy; + i915->base.surface_fill = i915_surface_fill; } diff --git a/src/gallium/drivers/i915simple/i915_texture.c b/src/gallium/drivers/i915simple/i915_texture.c index ca8e87af8d1..6a6c6542717 100644 --- a/src/gallium/drivers/i915simple/i915_texture.c +++ b/src/gallium/drivers/i915simple/i915_texture.c @@ -42,12 +42,14 @@ #include "i915_texture.h" #include "i915_debug.h" #include "i915_screen.h" -#include "i915_winsys.h" +#include "intel_winsys.h" + /* * Helper function and arrays */ + /** * Initial offset for Cube map. */ @@ -72,11 +74,6 @@ static const int step_offsets[6][2] = { {-1, 1} }; -static unsigned minify( unsigned d ) -{ - return MAX2(1, d>>1); -} - static unsigned power_of_two(unsigned x) { @@ -138,7 +135,7 @@ i915_miptree_set_level_info(struct i915_texture *tex, static void i915_miptree_set_image_offset(struct i915_texture *tex, - unsigned level, unsigned img, unsigned x, unsigned y) + unsigned level, unsigned img, unsigned x, unsigned y) { if (img == 0 && level == 0) assert(x == 0 && y == 0); @@ -155,49 +152,189 @@ i915_miptree_set_image_offset(struct i915_texture *tex, /* - * Layout functions + * i915 layout functions, some used by i945 */ /** - * Special case to deal with display targets. + * Special case to deal with scanout textures. */ static boolean -i915_displaytarget_layout(struct i915_texture *tex) +i915_scanout_layout(struct i915_texture *tex) { struct pipe_texture *pt = &tex->base; if (pt->last_level > 0 || pt->block.size != 4) return 0; - i915_miptree_set_level_info( tex, 0, 1, - tex->base.width[0], - tex->base.height[0], - 1 ); - i915_miptree_set_image_offset( tex, 0, 0, 0, 0 ); + i915_miptree_set_level_info(tex, 0, 1, + tex->base.width[0], + tex->base.height[0], + 1); + i915_miptree_set_image_offset(tex, 0, 0, 0, 0); - if (tex->base.width[0] >= 128) { + if (tex->base.width[0] >= 240) { + tex->stride = power_of_two(tex->base.nblocksx[0] * pt->block.size); + tex->total_nblocksy = round_up(tex->base.nblocksy[0], 8); + tex->hw_tiled = INTEL_TILE_X; + } else if (tex->base.width[0] == 64 && tex->base.height[0] == 64) { tex->stride = power_of_two(tex->base.nblocksx[0] * pt->block.size); tex->total_nblocksy = round_up(tex->base.nblocksy[0], 8); -#if 0 /* used for tiled display targets */ - tex->tiled = 1; -#endif } else { - tex->stride = round_up(tex->base.nblocksx[0] * pt->block.size, 64); - tex->total_nblocksy = tex->base.nblocksy[0]; + return FALSE; } - /* - printf("%s size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__, + debug_printf("%s size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__, tex->base.width[0], tex->base.height[0], pt->block.size, tex->stride, tex->total_nblocksy, tex->stride * tex->total_nblocksy); - */ - return 1; + return TRUE; +} + +static void +i915_miptree_layout_2d(struct i915_texture *tex) +{ + struct pipe_texture *pt = &tex->base; + unsigned level; + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + unsigned nblocksx = pt->nblocksx[0]; + unsigned nblocksy = pt->nblocksy[0]; + + tex->stride = round_up(pt->nblocksx[0] * pt->block.size, 4); + tex->total_nblocksy = 0; + + for (level = 0; level <= pt->last_level; level++) { + i915_miptree_set_level_info(tex, level, 1, width, height, 1); + i915_miptree_set_image_offset(tex, level, 0, 0, tex->total_nblocksy); + + nblocksy = round_up(MAX2(2, nblocksy), 2); + + tex->total_nblocksy += nblocksy; + + width = minify(width); + height = minify(height); + nblocksx = pf_get_nblocksx(&pt->block, width); + nblocksy = pf_get_nblocksy(&pt->block, height); + } +} + +static void +i915_miptree_layout_3d(struct i915_texture *tex) +{ + struct pipe_texture *pt = &tex->base; + unsigned level; + + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + unsigned depth = pt->depth[0]; + unsigned nblocksx = pt->nblocksx[0]; + unsigned nblocksy = pt->nblocksy[0]; + unsigned stack_nblocksy = 0; + + /* Calculate the size of a single slice. + */ + tex->stride = round_up(pt->nblocksx[0] * pt->block.size, 4); + + /* XXX: hardware expects/requires 9 levels at minimum. + */ + for (level = 0; level <= MAX2(8, pt->last_level); level++) { + i915_miptree_set_level_info(tex, level, depth, width, height, depth); + + stack_nblocksy += MAX2(2, nblocksy); + + width = minify(width); + height = minify(height); + depth = minify(depth); + nblocksx = pf_get_nblocksx(&pt->block, width); + nblocksy = pf_get_nblocksy(&pt->block, height); + } + + /* Fixup depth image_offsets: + */ + depth = pt->depth[0]; + for (level = 0; level <= pt->last_level; level++) { + unsigned i; + for (i = 0; i < depth; i++) + i915_miptree_set_image_offset(tex, level, i, 0, i * stack_nblocksy); + + depth = minify(depth); + } + + /* Multiply slice size by texture depth for total size. It's + * remarkable how wasteful of memory the i915 texture layouts + * are. They are largely fixed in the i945. + */ + tex->total_nblocksy = stack_nblocksy * pt->depth[0]; +} + +static void +i915_miptree_layout_cube(struct i915_texture *tex) +{ + struct pipe_texture *pt = &tex->base; + unsigned width = pt->width[0], height = pt->height[0]; + const unsigned nblocks = pt->nblocksx[0]; + unsigned level; + unsigned face; + + assert(width == height); /* cubemap images are square */ + + /* double pitch for cube layouts */ + tex->stride = round_up(nblocks * pt->block.size * 2, 4); + tex->total_nblocksy = nblocks * 4; + + for (level = 0; level <= pt->last_level; level++) { + i915_miptree_set_level_info(tex, level, 6, width, height, 1); + width /= 2; + height /= 2; + } + + for (face = 0; face < 6; face++) { + unsigned x = initial_offsets[face][0] * nblocks; + unsigned y = initial_offsets[face][1] * nblocks; + unsigned d = nblocks; + + for (level = 0; level <= pt->last_level; level++) { + i915_miptree_set_image_offset(tex, level, face, x, y); + d >>= 1; + x += step_offsets[face][0] * d; + y += step_offsets[face][1] * d; + } + } +} + +static boolean +i915_miptree_layout(struct i915_texture * tex) +{ + struct pipe_texture *pt = &tex->base; + + switch (pt->target) { + case PIPE_TEXTURE_1D: + case PIPE_TEXTURE_2D: + i915_miptree_layout_2d(tex); + break; + case PIPE_TEXTURE_3D: + i915_miptree_layout_3d(tex); + break; + case PIPE_TEXTURE_CUBE: + i915_miptree_layout_cube(tex); + break; + default: + assert(0); + return FALSE; + } + + return TRUE; } + +/* + * i945 layout functions + */ + + static void -i945_miptree_layout_2d( struct i915_texture *tex ) +i945_miptree_layout_2d(struct i915_texture *tex) { struct pipe_texture *pt = &tex->base; const int align_x = 2, align_y = 4; @@ -209,10 +346,10 @@ i945_miptree_layout_2d( struct i915_texture *tex ) unsigned nblocksx = pt->nblocksx[0]; unsigned nblocksy = pt->nblocksy[0]; - /* used for tiled display targets */ - if (0) - if (i915_displaytarget_layout(tex)) - return; + /* used for scanouts that need special layouts */ + if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_PRIMARY) + if (i915_scanout_layout(tex)) + return; tex->stride = round_up(pt->nblocksx[0] * pt->block.size, 4); @@ -223,11 +360,11 @@ i945_miptree_layout_2d( struct i915_texture *tex ) */ if (pt->last_level > 0) { unsigned mip1_nblocksx - = align(pf_get_nblocksx(&pt->block, minify(width)), align_x) + = align(pf_get_nblocksx(&pt->block, minify(width)), align_x) + pf_get_nblocksx(&pt->block, minify(minify(width))); if (mip1_nblocksx > nblocksx) - tex->stride = mip1_nblocksx * pt->block.size; + tex->stride = mip1_nblocksx * pt->block.size; } /* Pitch must be a whole number of dwords @@ -249,10 +386,10 @@ i945_miptree_layout_2d( struct i915_texture *tex ) /* Layout_below: step right after second mipmap level. */ if (level == 1) { - x += align(nblocksx, align_x); + x += align(nblocksx, align_x); } else { - y += nblocksy; + y += nblocksy; } width = minify(width); @@ -263,6 +400,63 @@ i945_miptree_layout_2d( struct i915_texture *tex ) } static void +i945_miptree_layout_3d(struct i915_texture *tex) +{ + struct pipe_texture *pt = &tex->base; + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + unsigned depth = pt->depth[0]; + unsigned nblocksx = pt->nblocksx[0]; + unsigned nblocksy = pt->nblocksy[0]; + unsigned pack_x_pitch, pack_x_nr; + unsigned pack_y_pitch; + unsigned level; + + tex->stride = round_up(pt->nblocksx[0] * pt->block.size, 4); + tex->total_nblocksy = 0; + + pack_y_pitch = MAX2(pt->nblocksy[0], 2); + pack_x_pitch = tex->stride / pt->block.size; + pack_x_nr = 1; + + for (level = 0; level <= pt->last_level; level++) { + int x = 0; + int y = 0; + unsigned q, j; + + i915_miptree_set_level_info(tex, level, depth, width, height, depth); + + for (q = 0; q < depth;) { + for (j = 0; j < pack_x_nr && q < depth; j++, q++) { + i915_miptree_set_image_offset(tex, level, q, x, y + tex->total_nblocksy); + x += pack_x_pitch; + } + + x = 0; + y += pack_y_pitch; + } + + tex->total_nblocksy += y; + + if (pack_x_pitch > 4) { + pack_x_pitch >>= 1; + pack_x_nr <<= 1; + assert(pack_x_pitch * pack_x_nr * pt->block.size <= tex->stride); + } + + if (pack_y_pitch > 2) { + pack_y_pitch >>= 1; + } + + width = minify(width); + height = minify(height); + depth = minify(depth); + nblocksx = pf_get_nblocksx(&pt->block, width); + nblocksy = pf_get_nblocksy(&pt->block, height); + } +} + +static void i945_miptree_layout_cube(struct i915_texture *tex) { struct pipe_texture *pt = &tex->base; @@ -364,226 +558,44 @@ i945_miptree_layout_cube(struct i915_texture *tex) } static boolean -i915_miptree_layout(struct i915_texture * tex) -{ - struct pipe_texture *pt = &tex->base; - unsigned level; - - switch (pt->target) { - case PIPE_TEXTURE_CUBE: { - const unsigned nblocks = pt->nblocksx[0]; - unsigned face; - unsigned width = pt->width[0], height = pt->height[0]; - - assert(width == height); /* cubemap images are square */ - - /* double pitch for cube layouts */ - tex->stride = round_up(nblocks * pt->block.size * 2, 4); - tex->total_nblocksy = nblocks * 4; - - for (level = 0; level <= pt->last_level; level++) { - i915_miptree_set_level_info(tex, level, 6, - width, height, - 1); - width /= 2; - height /= 2; - } - - for (face = 0; face < 6; face++) { - unsigned x = initial_offsets[face][0] * nblocks; - unsigned y = initial_offsets[face][1] * nblocks; - unsigned d = nblocks; - - for (level = 0; level <= pt->last_level; level++) { - i915_miptree_set_image_offset(tex, level, face, x, y); - d >>= 1; - x += step_offsets[face][0] * d; - y += step_offsets[face][1] * d; - } - } - break; - } - case PIPE_TEXTURE_3D:{ - unsigned width = pt->width[0]; - unsigned height = pt->height[0]; - unsigned depth = pt->depth[0]; - unsigned nblocksx = pt->nblocksx[0]; - unsigned nblocksy = pt->nblocksy[0]; - unsigned stack_nblocksy = 0; - - /* Calculate the size of a single slice. - */ - tex->stride = round_up(pt->nblocksx[0] * pt->block.size, 4); - - /* XXX: hardware expects/requires 9 levels at minimum. - */ - for (level = 0; level <= MAX2(8, pt->last_level); - level++) { - i915_miptree_set_level_info(tex, level, depth, - width, height, depth); - - - stack_nblocksy += MAX2(2, nblocksy); - - width = minify(width); - height = minify(height); - depth = minify(depth); - nblocksx = pf_get_nblocksx(&pt->block, width); - nblocksy = pf_get_nblocksy(&pt->block, height); - } - - /* Fixup depth image_offsets: - */ - depth = pt->depth[0]; - for (level = 0; level <= pt->last_level; level++) { - unsigned i; - for (i = 0; i < depth; i++) - i915_miptree_set_image_offset(tex, level, i, - 0, i * stack_nblocksy); - - depth = minify(depth); - } - - - /* Multiply slice size by texture depth for total size. It's - * remarkable how wasteful of memory the i915 texture layouts - * are. They are largely fixed in the i945. - */ - tex->total_nblocksy = stack_nblocksy * pt->depth[0]; - break; - } - - default:{ - unsigned width = pt->width[0]; - unsigned height = pt->height[0]; - unsigned nblocksx = pt->nblocksx[0]; - unsigned nblocksy = pt->nblocksy[0]; - - tex->stride = round_up(pt->nblocksx[0] * pt->block.size, 4); - tex->total_nblocksy = 0; - - for (level = 0; level <= pt->last_level; level++) { - i915_miptree_set_level_info(tex, level, 1, - width, height, 1); - i915_miptree_set_image_offset(tex, level, 0, - 0, tex->total_nblocksy); - - nblocksy = round_up(MAX2(2, nblocksy), 2); - - tex->total_nblocksy += nblocksy; - - width = minify(width); - height = minify(height); - nblocksx = pf_get_nblocksx(&pt->block, width); - nblocksy = pf_get_nblocksy(&pt->block, height); - } - break; - } - } - /* - DBG("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__, - tex->pitch, - tex->total_nblocksy, pt->block.size, tex->stride * tex->total_nblocksy); - */ - - return TRUE; -} - - -static boolean i945_miptree_layout(struct i915_texture * tex) { struct pipe_texture *pt = &tex->base; - unsigned level; switch (pt->target) { + case PIPE_TEXTURE_1D: + case PIPE_TEXTURE_2D: + i945_miptree_layout_2d(tex); + break; + case PIPE_TEXTURE_3D: + i945_miptree_layout_3d(tex); + break; case PIPE_TEXTURE_CUBE: i945_miptree_layout_cube(tex); break; - case PIPE_TEXTURE_3D:{ - unsigned width = pt->width[0]; - unsigned height = pt->height[0]; - unsigned depth = pt->depth[0]; - unsigned nblocksx = pt->nblocksx[0]; - unsigned nblocksy = pt->nblocksy[0]; - unsigned pack_x_pitch, pack_x_nr; - unsigned pack_y_pitch; - - tex->stride = round_up(pt->nblocksx[0] * pt->block.size, 4); - tex->total_nblocksy = 0; - - pack_y_pitch = MAX2(pt->nblocksy[0], 2); - pack_x_pitch = tex->stride / pt->block.size; - pack_x_nr = 1; - - for (level = 0; level <= pt->last_level; level++) { - unsigned nr_images = pt->target == PIPE_TEXTURE_3D ? depth : 6; - int x = 0; - int y = 0; - unsigned q, j; - - i915_miptree_set_level_info(tex, level, nr_images, - width, height, depth); - - for (q = 0; q < nr_images;) { - for (j = 0; j < pack_x_nr && q < nr_images; j++, q++) { - i915_miptree_set_image_offset(tex, level, q, x, y + tex->total_nblocksy); - x += pack_x_pitch; - } - - x = 0; - y += pack_y_pitch; - } - - - tex->total_nblocksy += y; - - if (pack_x_pitch > 4) { - pack_x_pitch >>= 1; - pack_x_nr <<= 1; - assert(pack_x_pitch * pack_x_nr * pt->block.size <= tex->stride); - } - - if (pack_y_pitch > 2) { - pack_y_pitch >>= 1; - } - - width = minify(width); - height = minify(height); - depth = minify(depth); - nblocksx = pf_get_nblocksx(&pt->block, width); - nblocksy = pf_get_nblocksy(&pt->block, height); - } - break; - } - - case PIPE_TEXTURE_1D: - case PIPE_TEXTURE_2D: -// case PIPE_TEXTURE_RECTANGLE: - i945_miptree_layout_2d(tex); - break; default: assert(0); return FALSE; } - /* - DBG("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__, - tex->pitch, - tex->total_nblocksy, pt->block.size, tex->stride * tex->total_nblocksy); - */ - return TRUE; } +/* + * Screen texture functions + */ + + static struct pipe_texture * i915_texture_create(struct pipe_screen *screen, const struct pipe_texture *templat) { - struct i915_screen *i915screen = i915_screen(screen); + struct i915_screen *is = i915_screen(screen); + struct intel_winsys *iws = is->iws; struct i915_texture *tex = CALLOC_STRUCT(i915_texture); size_t tex_size; + unsigned buf_usage = 0; if (!tex) return NULL; @@ -595,23 +607,35 @@ i915_texture_create(struct pipe_screen *screen, tex->base.nblocksx[0] = pf_get_nblocksx(&tex->base.block, tex->base.width[0]); tex->base.nblocksy[0] = pf_get_nblocksy(&tex->base.block, tex->base.height[0]); - if (i915screen->is_i945) { + if (is->is_i945) { if (!i945_miptree_layout(tex)) - goto fail; + goto fail; } else { if (!i915_miptree_layout(tex)) - goto fail; + goto fail; } tex_size = tex->stride * tex->total_nblocksy; - tex->buffer = screen->buffer_create(screen, 64, - PIPE_BUFFER_USAGE_PIXEL, - tex_size); + + /* for scanouts and cursors, cursors arn't scanouts */ + if (templat->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY && templat->width[0] != 64) + buf_usage = INTEL_NEW_SCANOUT; + else + buf_usage = INTEL_NEW_TEXTURE; + + tex->buffer = iws->buffer_create(iws, tex_size, 64, buf_usage); if (!tex->buffer) goto fail; + /* setup any hw fences */ + if (tex->hw_tiled) { + assert(tex->sw_tiled == INTEL_TILE_NONE); + iws->buffer_set_fence_reg(iws, tex->buffer, tex->stride, tex->hw_tiled); + } + + #if 0 void *ptr = ws->buffer_map(ws, tex->buffer, PIPE_BUFFER_USAGE_CPU_WRITE); @@ -626,18 +650,56 @@ fail: return NULL; } +static struct pipe_texture * +i915_texture_blanket(struct pipe_screen * screen, + const struct pipe_texture *base, + const unsigned *stride, + struct pipe_buffer *buffer) +{ +#if 0 + struct i915_texture *tex; + assert(screen); + + /* Only supports one type */ + if (base->target != PIPE_TEXTURE_2D || + base->last_level != 0 || + base->depth[0] != 1) { + return NULL; + } + + tex = CALLOC_STRUCT(i915_texture); + if (!tex) + return NULL; + + tex->base = *base; + pipe_reference_init(&tex->base.reference, 1); + tex->base.screen = screen; + + tex->stride = stride[0]; + + i915_miptree_set_level_info(tex, 0, 1, base->width[0], base->height[0], 1); + i915_miptree_set_image_offset(tex, 0, 0, 0, 0); + + pipe_buffer_reference(&tex->buffer, buffer); + + return &tex->base; +#else + return NULL; +#endif +} static void i915_texture_destroy(struct pipe_texture *pt) { struct i915_texture *tex = (struct i915_texture *)pt; + struct intel_winsys *iws = i915_screen(pt->screen)->iws; uint i; /* DBG("%s deleting %p\n", __FUNCTION__, (void *) tex); */ - pipe_buffer_reference(&tex->buffer, NULL); + iws->buffer_destroy(iws, tex->buffer); for (i = 0; i < PIPE_MAX_TEXTURE_LEVELS; i++) if (tex->image_offset[i]) @@ -646,6 +708,12 @@ i915_texture_destroy(struct pipe_texture *pt) FREE(tex); } + +/* + * Screen surface functions + */ + + static struct pipe_surface * i915_get_tex_surface(struct pipe_screen *screen, struct pipe_texture *pt, @@ -681,11 +749,122 @@ i915_get_tex_surface(struct pipe_screen *screen, return ps; } -static struct pipe_texture * -i915_texture_blanket(struct pipe_screen * screen, - const struct pipe_texture *base, - const unsigned *stride, - struct pipe_buffer *buffer) +static void +i915_tex_surface_destroy(struct pipe_surface *surf) +{ + pipe_texture_reference(&surf->texture, NULL); + FREE(surf); +} + + +/* + * Screen transfer functions + */ + + +static struct pipe_transfer* +i915_get_tex_transfer(struct pipe_screen *screen, + struct pipe_texture *texture, + unsigned face, unsigned level, unsigned zslice, + enum pipe_transfer_usage usage, unsigned x, unsigned y, + unsigned w, unsigned h) +{ + struct i915_texture *tex = (struct i915_texture *)texture; + struct i915_transfer *trans; + unsigned offset; /* in bytes */ + + if (texture->target == PIPE_TEXTURE_CUBE) { + offset = tex->image_offset[level][face]; + } + else if (texture->target == PIPE_TEXTURE_3D) { + offset = tex->image_offset[level][zslice]; + } + else { + offset = tex->image_offset[level][0]; + assert(face == 0); + assert(zslice == 0); + } + + trans = CALLOC_STRUCT(i915_transfer); + if (trans) { + pipe_texture_reference(&trans->base.texture, texture); + trans->base.format = trans->base.format; + trans->base.x = x; + trans->base.y = y; + trans->base.width = w; + trans->base.height = h; + trans->base.block = texture->block; + trans->base.nblocksx = texture->nblocksx[level]; + trans->base.nblocksy = texture->nblocksy[level]; + trans->base.stride = tex->stride; + trans->offset = offset; + trans->base.usage = usage; + } + return &trans->base; +} + +static void * +i915_transfer_map(struct pipe_screen *screen, + struct pipe_transfer *transfer) +{ + struct i915_texture *tex = (struct i915_texture *)transfer->texture; + struct intel_winsys *iws = i915_screen(tex->base.screen)->iws; + char *map; + boolean write = FALSE; + + if (transfer->usage != PIPE_TRANSFER_READ) + write = TRUE; + + map = iws->buffer_map(iws, tex->buffer, write); + if (map == NULL) + return NULL; + + return map + i915_transfer(transfer)->offset + + transfer->y / transfer->block.height * transfer->stride + + transfer->x / transfer->block.width * transfer->block.size; +} + +static void +i915_transfer_unmap(struct pipe_screen *screen, + struct pipe_transfer *transfer) +{ + struct i915_texture *tex = (struct i915_texture *)transfer->texture; + struct intel_winsys *iws = i915_screen(tex->base.screen)->iws; + iws->buffer_unmap(iws, tex->buffer); +} + +static void +i915_tex_transfer_destroy(struct pipe_transfer *trans) +{ + pipe_texture_reference(&trans->texture, NULL); + FREE(trans); +} + + +/* + * Other texture functions + */ + + +void +i915_init_screen_texture_functions(struct i915_screen *is) +{ + is->base.texture_create = i915_texture_create; + is->base.texture_blanket = i915_texture_blanket; + is->base.texture_destroy = i915_texture_destroy; + is->base.get_tex_surface = i915_get_tex_surface; + is->base.tex_surface_destroy = i915_tex_surface_destroy; + is->base.get_tex_transfer = i915_get_tex_transfer; + is->base.transfer_map = i915_transfer_map; + is->base.transfer_unmap = i915_transfer_unmap; + is->base.tex_transfer_destroy = i915_tex_transfer_destroy; +} + +struct pipe_texture * +i915_texture_blanket_intel(struct pipe_screen *screen, + struct pipe_texture *base, + unsigned stride, + struct intel_buffer *buffer) { struct i915_texture *tex; assert(screen); @@ -705,52 +884,28 @@ i915_texture_blanket(struct pipe_screen * screen, pipe_reference_init(&tex->base.reference, 1); tex->base.screen = screen; - tex->stride = stride[0]; + tex->stride = stride; i915_miptree_set_level_info(tex, 0, 1, base->width[0], base->height[0], 1); i915_miptree_set_image_offset(tex, 0, 0, 0, 0); - pipe_buffer_reference(&tex->buffer, buffer); + tex->buffer = buffer; return &tex->base; } -void -i915_init_texture_functions(struct i915_context *i915) -{ -// i915->pipe.texture_update = i915_texture_update; -} - -static void -i915_tex_surface_destroy(struct pipe_surface *surf) -{ - pipe_texture_reference(&surf->texture, NULL); - FREE(surf); -} - -void -i915_init_screen_texture_functions(struct pipe_screen *screen) -{ - screen->texture_create = i915_texture_create; - screen->texture_destroy = i915_texture_destroy; - screen->get_tex_surface = i915_get_tex_surface; - screen->texture_blanket = i915_texture_blanket; - screen->tex_surface_destroy = i915_tex_surface_destroy; -} - -boolean i915_get_texture_buffer( struct pipe_texture *texture, - struct pipe_buffer **buf, - unsigned *stride ) +boolean +i915_get_texture_buffer_intel(struct pipe_texture *texture, + struct intel_buffer **buffer, + unsigned *stride) { struct i915_texture *tex = (struct i915_texture *)texture; - if (!tex) + if (!texture) return FALSE; - pipe_buffer_reference(buf, tex->buffer); - - if (stride) - *stride = tex->stride; + *stride = tex->stride; + *buffer = tex->buffer; return TRUE; } diff --git a/src/gallium/drivers/i915simple/i915_texture.h b/src/gallium/drivers/i915simple/i915_texture.h index 7225016a9f4..51a1dd984c8 100644 --- a/src/gallium/drivers/i915simple/i915_texture.h +++ b/src/gallium/drivers/i915simple/i915_texture.h @@ -28,16 +28,9 @@ #ifndef I915_TEXTURE_H #define I915_TEXTURE_H -struct i915_context; -struct pipe_screen; - +struct i915_screen; extern void -i915_init_texture_functions(struct i915_context *i915); - - -extern void -i915_init_screen_texture_functions(struct pipe_screen *screen); - +i915_init_screen_texture_functions(struct i915_screen *is); #endif /* I915_TEXTURE_H */ diff --git a/src/gallium/drivers/i915simple/i915_winsys.h b/src/gallium/drivers/i915simple/i915_winsys.h deleted file mode 100644 index ff5b34f193a..00000000000 --- a/src/gallium/drivers/i915simple/i915_winsys.h +++ /dev/null @@ -1,143 +0,0 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ - -/** - * \file - * This is the interface that i915simple requires any window system - * hosting it to implement. This is the only include file in i915simple - * which is public. - * - * This isn't currently true as the winsys needs i915_batchbuffer.h - */ - -#ifndef I915_WINSYS_H -#define I915_WINSYS_H - - -#include "pipe/p_defines.h" - - -#ifdef __cplusplus -extern "C" { -#endif - - -/* Pipe drivers are independent of both GL and the window system. - * The window system provides a buffer manager and a set of additional - * hooks for things like command buffer submission, etc. - * - * There clearly has to be some agreement between the window system - * driver and the hardware driver about the format of command buffers, - * etc. - */ - -struct i915_batchbuffer; -struct pipe_texture; -struct pipe_buffer; -struct pipe_fence_handle; -struct pipe_winsys; -struct pipe_screen; - - -/** - * Additional winsys interface for i915simple. - * - * It is an over-simple batchbuffer mechanism. Will want to improve the - * performance of this, perhaps based on the cmdstream stuff. It - * would be pretty impossible to implement swz on top of this - * interface. - * - * Will also need additions/changes to implement static/dynamic - * indirect state. - */ -struct i915_winsys { - - void (*destroy)( struct i915_winsys *sws ); - - /** - * Get the current batch buffer from the winsys. - */ - struct i915_batchbuffer *(*batch_get)( struct i915_winsys *sws ); - - /** - * Emit a relocation to a buffer. - * - * Used not only when the buffer addresses are not pinned, but also to - * ensure refered buffers will not be destroyed until the current batch - * buffer execution is finished. - * - * The access flags is a combination of I915_BUFFER_ACCESS_WRITE and - * I915_BUFFER_ACCESS_READ macros. - */ - void (*batch_reloc)( struct i915_winsys *sws, - struct pipe_buffer *buf, - unsigned access_flags, - unsigned delta ); - - /** - * Flush the batch. - */ - void (*batch_flush)( struct i915_winsys *sws, - struct pipe_fence_handle **fence ); -}; - -#define I915_BUFFER_ACCESS_WRITE 0x1 -#define I915_BUFFER_ACCESS_READ 0x2 - -#define I915_BUFFER_USAGE_LIT_VERTEX (PIPE_BUFFER_USAGE_CUSTOM << 0) - - -/** - * Create i915 pipe_screen. - */ -struct pipe_screen *i915_create_screen( struct pipe_winsys *winsys, - uint pci_id ); - -/** - * Create a i915 pipe_context. - */ -struct pipe_context *i915_create_context( struct pipe_screen *screen, - struct pipe_winsys *winsys, - struct i915_winsys *i915 ); - -/** - * Used for the winsys to get the buffer used for a texture - * and also the stride used for the texture. - * - * Buffer is referenced for you so you need to unref after use. - * - * This is needed for example kms. - */ -boolean i915_get_texture_buffer( struct pipe_texture *texture, - struct pipe_buffer **buf, - unsigned *stride ); - -#ifdef __cplusplus -} -#endif - -#endif diff --git a/src/gallium/drivers/i915simple/intel_batchbuffer.h b/src/gallium/drivers/i915simple/intel_batchbuffer.h new file mode 100644 index 00000000000..db12dfd2ac2 --- /dev/null +++ b/src/gallium/drivers/i915simple/intel_batchbuffer.h @@ -0,0 +1,87 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTEL_BATCH_H +#define INTEL_BATCH_H + +#include "intel_winsys.h" + +static INLINE boolean +intel_batchbuffer_check(struct intel_batchbuffer *batch, + size_t dwords, + size_t relocs) +{ + return dwords * 4 <= batch->size - (batch->ptr - batch->map) && + relocs <= (batch->max_relocs - batch->relocs); +} + +static INLINE size_t +intel_batchbuffer_space(struct intel_batchbuffer *batch) +{ + return batch->size - (batch->ptr - batch->map); +} + +static INLINE void +intel_batchbuffer_dword(struct intel_batchbuffer *batch, + unsigned dword) +{ + if (intel_batchbuffer_space(batch) < 4) + return; + + *(unsigned *)batch->ptr = dword; + batch->ptr += 4; +} + +static INLINE void +intel_batchbuffer_write(struct intel_batchbuffer *batch, + void *data, + size_t size) +{ + if (intel_batchbuffer_space(batch) < size) + return; + + memcpy(data, batch->ptr, size); + batch->ptr += size; +} + +static INLINE int +intel_batchbuffer_reloc(struct intel_batchbuffer *batch, + struct intel_buffer *buffer, + enum intel_buffer_usage usage, + size_t offset) +{ + return batch->iws->batchbuffer_reloc(batch, buffer, usage, offset); +} + +static INLINE void +intel_batchbuffer_flush(struct intel_batchbuffer *batch, + struct pipe_fence_handle **fence) +{ + batch->iws->batchbuffer_flush(batch, fence); +} + +#endif diff --git a/src/gallium/drivers/i915simple/intel_winsys.h b/src/gallium/drivers/i915simple/intel_winsys.h new file mode 100644 index 00000000000..f949f52a9ce --- /dev/null +++ b/src/gallium/drivers/i915simple/intel_winsys.h @@ -0,0 +1,219 @@ +/************************************************************************** + * + * Copyright © 2009 Jakob Bornecrantz + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTEL_WINSYS_H +#define INTEL_WINSYS_H + +#include "pipe/p_compiler.h" + +struct intel_winsys; +struct intel_buffer; +struct intel_batchbuffer; +struct pipe_texture; +struct pipe_fence_handle; + +enum intel_buffer_usage +{ + /* use on textures */ + INTEL_USAGE_RENDER = 0x01, + INTEL_USAGE_SAMPLER = 0x02, + INTEL_USAGE_2D_TARGET = 0x04, + INTEL_USAGE_2D_SOURCE = 0x08, + /* use on vertex */ + INTEL_USAGE_VERTEX = 0x10, +}; + +enum intel_buffer_type +{ + INTEL_NEW_TEXTURE, + INTEL_NEW_SCANOUT, /**< a texture used for scanning out from */ + INTEL_NEW_VERTEX, +}; + +enum intel_buffer_tile +{ + INTEL_TILE_NONE, + INTEL_TILE_X, + INTEL_TILE_Y, +}; + +struct intel_batchbuffer { + + struct intel_winsys *iws; + + /** + * Values exported to speed up the writing the batchbuffer, + * instead of having to go trough a accesor function for + * each dword written. + */ + /*{@*/ + uint8_t *map; + uint8_t *ptr; + size_t size; + + size_t relocs; + size_t max_relocs; + /*@}*/ +}; + +struct intel_winsys { + + /** + * Batchbuffer functions. + */ + /*@{*/ + /** + * Create a new batchbuffer. + */ + struct intel_batchbuffer *(*batchbuffer_create)(struct intel_winsys *iws); + + /** + * Emit a relocation to a buffer. + * Target position in batchbuffer is the same as ptr. + * + * @batch + * @reloc buffer address to be inserted into target. + * @usage how is the hardware going to use the buffer. + * @offset add this to the reloc buffers address + * @target buffer where to write the address, null for batchbuffer. + */ + int (*batchbuffer_reloc)(struct intel_batchbuffer *batch, + struct intel_buffer *reloc, + enum intel_buffer_usage usage, + unsigned offset); + + /** + * Flush a bufferbatch. + */ + void (*batchbuffer_flush)(struct intel_batchbuffer *batch, + struct pipe_fence_handle **fence); + + /** + * Destroy a batchbuffer. + */ + void (*batchbuffer_destroy)(struct intel_batchbuffer *batch); + /*@}*/ + + + /** + * Buffer functions. + */ + /*@{*/ + /** + * Create a buffer. + */ + struct intel_buffer *(*buffer_create)(struct intel_winsys *iws, + unsigned size, unsigned alignment, + enum intel_buffer_type type); + + /** + * Fence a buffer with a fence reg. + * Not to be confused with pipe_fence_handle. + */ + int (*buffer_set_fence_reg)(struct intel_winsys *iws, + struct intel_buffer *buffer, + unsigned stride, + enum intel_buffer_tile tile); + + /** + * Map a buffer. + */ + void *(*buffer_map)(struct intel_winsys *iws, + struct intel_buffer *buffer, + boolean write); + + /** + * Unmap a buffer. + */ + void (*buffer_unmap)(struct intel_winsys *iws, + struct intel_buffer *buffer); + + void (*buffer_destroy)(struct intel_winsys *iws, + struct intel_buffer *buffer); + /*@}*/ + + + /** + * Fence functions. + */ + /*@{*/ + /** + * Reference fence and set ptr to fence. + */ + void (*fence_reference)(struct intel_winsys *iws, + struct pipe_fence_handle **ptr, + struct pipe_fence_handle *fence); + + /** + * Check if a fence has finished. + */ + int (*fence_signalled)(struct intel_winsys *iws, + struct pipe_fence_handle *fence); + + /** + * Wait on a fence to finish. + */ + int (*fence_finish)(struct intel_winsys *iws, + struct pipe_fence_handle *fence); + /*@}*/ + + + /** + * Destroy the winsys. + */ + void (*destroy)(struct intel_winsys *iws); +}; + + +/** + * Create i915 pipe_screen. + */ +struct pipe_screen *i915_create_screen(struct intel_winsys *iws, unsigned pci_id); + +/** + * Create a i915 pipe_context. + */ +struct pipe_context *i915_create_context(struct pipe_screen *screen); + +/** + * Get the intel_winsys buffer backing the texture. + * + * TODO UGLY + */ +boolean i915_get_texture_buffer_intel(struct pipe_texture *texture, + struct intel_buffer **buffer, + unsigned *stride); + +/** + * Wrap a intel_winsys buffer with a texture blanket. + * + * TODO UGLY + */ +struct pipe_texture * i915_texture_blanket_intel(struct pipe_screen *screen, + struct pipe_texture *tmplt, + unsigned pitch, + struct intel_buffer *buffer); + +#endif diff --git a/src/gallium/drivers/i965simple/brw_draw.c b/src/gallium/drivers/i965simple/brw_draw.c index 648aaa0da52..49d80cb41c5 100644 --- a/src/gallium/drivers/i965simple/brw_draw.c +++ b/src/gallium/drivers/i965simple/brw_draw.c @@ -35,6 +35,7 @@ #include "pipe/p_context.h" #include "pipe/internal/p_winsys_screen.h" +#include "util/u_prim.h" static unsigned hw_prim[PIPE_PRIM_POLYGON+1] = { _3DPRIM_POINTLIST, @@ -50,20 +51,6 @@ static unsigned hw_prim[PIPE_PRIM_POLYGON+1] = { }; -static const int reduced_prim[PIPE_PRIM_POLYGON+1] = { - PIPE_PRIM_POINTS, - PIPE_PRIM_LINES, - PIPE_PRIM_LINES, - PIPE_PRIM_LINES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES -}; - - /* When the primitive changes, set a state bit and re-validate. Not * the nicest and would rather deal with this by having all the * programs be immune to the active primitive (ie. cope with all @@ -85,8 +72,8 @@ static void brw_set_prim(struct brw_context *brw, int prim) brw->primitive = prim; brw->state.dirty.brw |= BRW_NEW_PRIMITIVE; - if (reduced_prim[prim] != brw->reduced_primitive) { - brw->reduced_primitive = reduced_prim[prim]; + if (u_reduced_prim(prim) != brw->reduced_primitive) { + brw->reduced_primitive = u_reduced_prim(prim); brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE; } diff --git a/src/gallium/drivers/i965simple/brw_surface.c b/src/gallium/drivers/i965simple/brw_surface.c index 511779dbfa0..724a69b2eee 100644 --- a/src/gallium/drivers/i965simple/brw_surface.c +++ b/src/gallium/drivers/i965simple/brw_surface.c @@ -60,7 +60,7 @@ brw_surface_copy(struct pipe_context *pipe, src, PIPE_BUFFER_USAGE_CPU_READ ); - pipe_copy_rect(dst_map, + util_copy_rect(dst_map, &dst->block, dst->stride, dstx, dsty, @@ -99,7 +99,7 @@ brw_surface_fill(struct pipe_context *pipe, dst, PIPE_BUFFER_USAGE_CPU_WRITE ); - pipe_fill_rect(dst_map, &dst->block, dst->stride, dstx, dsty, width, height, value); + util_fill_rect(dst_map, &dst->block, dst->stride, dstx, dsty, width, height, value); pipe->screen->surface_unmap(pipe->screen, dst); } diff --git a/src/gallium/drivers/i965simple/brw_tex_layout.c b/src/gallium/drivers/i965simple/brw_tex_layout.c index 8aea8c05581..998ffaeac4a 100644 --- a/src/gallium/drivers/i965simple/brw_tex_layout.c +++ b/src/gallium/drivers/i965simple/brw_tex_layout.c @@ -65,11 +65,6 @@ unsigned intel_compressed_alignment(unsigned internalFormat) } #endif -static unsigned minify( unsigned d ) -{ - return MAX2(1, d>>1); -} - static void intel_miptree_set_image_offset(struct brw_texture *tex, unsigned level, diff --git a/src/gallium/drivers/i965simple/brw_vs_emit.c b/src/gallium/drivers/i965simple/brw_vs_emit.c index e03d6534821..3ee82d95b3a 100644 --- a/src/gallium/drivers/i965simple/brw_vs_emit.c +++ b/src/gallium/drivers/i965simple/brw_vs_emit.c @@ -1294,10 +1294,10 @@ void brw_vs_emit(struct brw_vs_compile *c) case TGSI_TOKEN_TYPE_IMMEDIATE: { struct tgsi_full_immediate *imm = &parse.FullToken.FullImmediate; assert(imm->Immediate.NrTokens == 4 + 1); - c->prog_data.imm_buf[c->prog_data.num_imm][0] = imm->u.ImmediateFloat32[0].Float; - c->prog_data.imm_buf[c->prog_data.num_imm][1] = imm->u.ImmediateFloat32[1].Float; - c->prog_data.imm_buf[c->prog_data.num_imm][2] = imm->u.ImmediateFloat32[2].Float; - c->prog_data.imm_buf[c->prog_data.num_imm][3] = imm->u.ImmediateFloat32[3].Float; + c->prog_data.imm_buf[c->prog_data.num_imm][0] = imm->u[0].Float; + c->prog_data.imm_buf[c->prog_data.num_imm][1] = imm->u[1].Float; + c->prog_data.imm_buf[c->prog_data.num_imm][2] = imm->u[2].Float; + c->prog_data.imm_buf[c->prog_data.num_imm][3] = imm->u[3].Float; c->prog_data.num_imm++; } break; diff --git a/src/gallium/drivers/i965simple/brw_wm_glsl.c b/src/gallium/drivers/i965simple/brw_wm_glsl.c index ab6410aa607..db759639328 100644 --- a/src/gallium/drivers/i965simple/brw_wm_glsl.c +++ b/src/gallium/drivers/i965simple/brw_wm_glsl.c @@ -947,7 +947,7 @@ static void brw_wm_emit_instruction( struct brw_wm_compile *c, #endif break; - case TGSI_OPCODE_LOOP: + case TGSI_OPCODE_BGNFOR: c->loop_inst[c->loop_insn++] = brw_DO(p, BRW_EXECUTE_8); break; case TGSI_OPCODE_BRK: @@ -958,11 +958,11 @@ static void brw_wm_emit_instruction( struct brw_wm_compile *c, brw_CONT(p); brw_set_predicate_control(p, BRW_PREDICATE_NONE); break; - case TGSI_OPCODE_ENDLOOP: + case TGSI_OPCODE_ENDFOR: c->loop_insn--; c->inst0 = c->inst1 = brw_WHILE(p, c->loop_inst[c->loop_insn]); /* patch all the BREAK instructions from - last BEGINLOOP */ + last BGNFOR */ while (c->inst0 > c->loop_inst[c->loop_insn]) { c->inst0--; if (c->inst0->header.opcode == BRW_OPCODE_BREAK) { diff --git a/src/gallium/drivers/identity/Makefile b/src/gallium/drivers/identity/Makefile new file mode 100644 index 00000000000..e32b9102e59 --- /dev/null +++ b/src/gallium/drivers/identity/Makefile @@ -0,0 +1,12 @@ +TOP = ../../../.. +include $(TOP)/configs/current + +LIBNAME = identity + +C_SOURCES = \ + id_objects.c \ + id_context.c \ + id_screen.c \ + id_drm.c + +include ../../Makefile.template diff --git a/src/gallium/drivers/identity/SConscript b/src/gallium/drivers/identity/SConscript new file mode 100644 index 00000000000..7f079dd0a8b --- /dev/null +++ b/src/gallium/drivers/identity/SConscript @@ -0,0 +1,13 @@ +Import('*') + +env = env.Clone() + +identity = env.ConvenienceLibrary( + target = 'identity', + source = [ + 'id_screen.c', + 'id_context.c', + 'id_objects.c', + ]) + +Export('identity') diff --git a/src/gallium/drivers/identity/id_context.c b/src/gallium/drivers/identity/id_context.c new file mode 100644 index 00000000000..4e700089e33 --- /dev/null +++ b/src/gallium/drivers/identity/id_context.c @@ -0,0 +1,719 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "pipe/p_context.h" +#include "util/u_memory.h" + +#include "id_public.h" +#include "id_context.h" +#include "id_objects.h" + + +static void +identity_destroy(struct pipe_context *_pipe) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->destroy(pipe); + + free(id_pipe); +} + +static void +identity_set_edgeflags(struct pipe_context *_pipe, + const unsigned *bitfield) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->set_edgeflags(pipe, + bitfield); +} + +static boolean +identity_draw_arrays(struct pipe_context *_pipe, + unsigned prim, + unsigned start, + unsigned count) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + return pipe->draw_arrays(pipe, + prim, + start, + count); +} + +static boolean +identity_draw_elements(struct pipe_context *_pipe, + struct pipe_buffer *_indexBuffer, + unsigned indexSize, + unsigned prim, + unsigned start, + unsigned count) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct identity_buffer *id_buffer = identity_buffer(_indexBuffer); + struct pipe_context *pipe = id_pipe->pipe; + struct pipe_buffer *indexBuffer = id_buffer->buffer; + + return pipe->draw_elements(pipe, + indexBuffer, + indexSize, + prim, + start, + count); +} + +static boolean +identity_draw_range_elements(struct pipe_context *_pipe, + struct pipe_buffer *_indexBuffer, + unsigned indexSize, + unsigned minIndex, + unsigned maxIndex, + unsigned mode, + unsigned start, + unsigned count) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct identity_buffer *id_buffer = identity_buffer(_indexBuffer); + struct pipe_context *pipe = id_pipe->pipe; + struct pipe_buffer *indexBuffer = id_buffer->buffer; + + return pipe->draw_range_elements(pipe, + indexBuffer, + indexSize, + minIndex, + maxIndex, + mode, + start, + count); +} + +static struct pipe_query * +identity_create_query(struct pipe_context *_pipe, + unsigned query_type) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + return pipe->create_query(pipe, + query_type); +} + +static void +identity_destroy_query(struct pipe_context *_pipe, + struct pipe_query *query) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->destroy_query(pipe, + query); +} + +static void +identity_begin_query(struct pipe_context *_pipe, + struct pipe_query *query) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->begin_query(pipe, + query); +} + +static void +identity_end_query(struct pipe_context *_pipe, + struct pipe_query *query) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->end_query(pipe, + query); +} + +static boolean +identity_get_query_result(struct pipe_context *_pipe, + struct pipe_query *query, + boolean wait, + uint64_t *result) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + return pipe->get_query_result(pipe, + query, + wait, + result); +} + +static void * +identity_create_blend_state(struct pipe_context *_pipe, + const struct pipe_blend_state *blend) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + return pipe->create_blend_state(pipe, + blend); +} + +static void +identity_bind_blend_state(struct pipe_context *_pipe, + void *blend) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->bind_blend_state(pipe, + blend); +} + +static void +identity_delete_blend_state(struct pipe_context *_pipe, + void *blend) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->delete_blend_state(pipe, + blend); +} + +static void * +identity_create_sampler_state(struct pipe_context *_pipe, + const struct pipe_sampler_state *sampler) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + return pipe->create_sampler_state(pipe, + sampler); +} + +static void +identity_bind_sampler_states(struct pipe_context *_pipe, + unsigned num, + void **samplers) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->bind_sampler_states(pipe, + num, + samplers); +} + +static void +identity_delete_sampler_state(struct pipe_context *_pipe, + void *sampler) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->delete_sampler_state(pipe, + sampler); +} + +static void * +identity_create_rasterizer_state(struct pipe_context *_pipe, + const struct pipe_rasterizer_state *rasterizer) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + return pipe->create_rasterizer_state(pipe, + rasterizer); +} + +static void +identity_bind_rasterizer_state(struct pipe_context *_pipe, + void *rasterizer) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->bind_rasterizer_state(pipe, + rasterizer); +} + +static void +identity_delete_rasterizer_state(struct pipe_context *_pipe, + void *rasterizer) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->delete_rasterizer_state(pipe, + rasterizer); +} + +static void * +identity_create_depth_stencil_alpha_state(struct pipe_context *_pipe, + const struct pipe_depth_stencil_alpha_state *depth_stencil_alpha) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + return pipe->create_depth_stencil_alpha_state(pipe, + depth_stencil_alpha); +} + +static void +identity_bind_depth_stencil_alpha_state(struct pipe_context *_pipe, + void *depth_stencil_alpha) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->bind_depth_stencil_alpha_state(pipe, + depth_stencil_alpha); +} + +static void +identity_delete_depth_stencil_alpha_state(struct pipe_context *_pipe, + void *depth_stencil_alpha) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->delete_depth_stencil_alpha_state(pipe, + depth_stencil_alpha); +} + +static void * +identity_create_fs_state(struct pipe_context *_pipe, + const struct pipe_shader_state *fs) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + return pipe->create_fs_state(pipe, + fs); +} + +static void +identity_bind_fs_state(struct pipe_context *_pipe, + void *fs) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->bind_fs_state(pipe, + fs); +} + +static void +identity_delete_fs_state(struct pipe_context *_pipe, + void *fs) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->delete_fs_state(pipe, + fs); +} + +static void * +identity_create_vs_state(struct pipe_context *_pipe, + const struct pipe_shader_state *vs) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + return pipe->create_vs_state(pipe, + vs); +} + +static void +identity_bind_vs_state(struct pipe_context *_pipe, + void *vs) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->bind_vs_state(pipe, + vs); +} + +static void +identity_delete_vs_state(struct pipe_context *_pipe, + void *vs) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->delete_vs_state(pipe, + vs); +} + +static void +identity_set_blend_color(struct pipe_context *_pipe, + const struct pipe_blend_color *blend_color) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->set_blend_color(pipe, + blend_color); +} + +static void +identity_set_clip_state(struct pipe_context *_pipe, + const struct pipe_clip_state *clip) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->set_clip_state(pipe, + clip); +} + +static void +identity_set_constant_buffer(struct pipe_context *_pipe, + uint shader, + uint index, + const struct pipe_constant_buffer *_buffer) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + struct pipe_constant_buffer unwrapped_buffer; + struct pipe_constant_buffer *buffer = NULL; + + /* unwrap the input state */ + if (_buffer) { + unwrapped_buffer.buffer = identity_buffer_unwrap(_buffer->buffer); + buffer = &unwrapped_buffer; + } + + pipe->set_constant_buffer(pipe, + shader, + index, + buffer); +} + +static void +identity_set_framebuffer_state(struct pipe_context *_pipe, + const struct pipe_framebuffer_state *_state) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + struct pipe_framebuffer_state unwrapped_state; + struct pipe_framebuffer_state *state = NULL; + unsigned i; + + /* unwrap the input state */ + if (_state) { + memcpy(&unwrapped_state, _state, sizeof(unwrapped_state)); + for(i = 0; i < _state->nr_cbufs; i++) + unwrapped_state.cbufs[i] = identity_surface_unwrap(_state->cbufs[i]); + for (; i < PIPE_MAX_COLOR_BUFS; i++) + unwrapped_state.cbufs[i] = NULL; + unwrapped_state.zsbuf = identity_surface_unwrap(_state->zsbuf); + state = &unwrapped_state; + } + + pipe->set_framebuffer_state(pipe, + state); +} + +static void +identity_set_polygon_stipple(struct pipe_context *_pipe, + const struct pipe_poly_stipple *poly_stipple) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->set_polygon_stipple(pipe, + poly_stipple); +} + +static void +identity_set_scissor_state(struct pipe_context *_pipe, + const struct pipe_scissor_state *scissor) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->set_scissor_state(pipe, + scissor); +} + +static void +identity_set_viewport_state(struct pipe_context *_pipe, + const struct pipe_viewport_state *viewport) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->set_viewport_state(pipe, + viewport); +} + +static void +identity_set_sampler_textures(struct pipe_context *_pipe, + unsigned num_textures, + struct pipe_texture **_textures) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + struct pipe_texture *unwrapped_textures[PIPE_MAX_SAMPLERS]; + struct pipe_texture **textures = NULL; + unsigned i; + + if (_textures) { + for (i = 0; i < num_textures; i++) + unwrapped_textures[i] = identity_texture_unwrap(_textures[i]); + for (; i < PIPE_MAX_SAMPLERS; i++) + unwrapped_textures[i] = NULL; + + textures = unwrapped_textures; + } + + pipe->set_sampler_textures(pipe, + num_textures, + textures); +} + +static void +identity_set_vertex_buffers(struct pipe_context *_pipe, + unsigned num_buffers, + const struct pipe_vertex_buffer *_buffers) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + struct pipe_vertex_buffer unwrapped_buffers[PIPE_MAX_SHADER_INPUTS]; + struct pipe_vertex_buffer *buffers = NULL; + unsigned i; + + if (num_buffers) { + memcpy(unwrapped_buffers, _buffers, num_buffers * sizeof(*_buffers)); + for (i = 0; i < num_buffers; i++) + unwrapped_buffers[i].buffer = identity_buffer_unwrap(_buffers[i].buffer); + buffers = unwrapped_buffers; + } + + pipe->set_vertex_buffers(pipe, + num_buffers, + buffers); +} + +static void +identity_set_vertex_elements(struct pipe_context *_pipe, + unsigned num_elements, + const struct pipe_vertex_element *vertex_elements) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->set_vertex_elements(pipe, + num_elements, + vertex_elements); +} + +static void +identity_surface_copy(struct pipe_context *_pipe, + struct pipe_surface *_dst, + unsigned dstx, + unsigned dsty, + struct pipe_surface *_src, + unsigned srcx, + unsigned srcy, + unsigned width, + unsigned height) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct identity_surface *id_surface_dst = identity_surface(_dst); + struct identity_surface *id_surface_src = identity_surface(_src); + struct pipe_context *pipe = id_pipe->pipe; + struct pipe_surface *dst = id_surface_dst->surface; + struct pipe_surface *src = id_surface_src->surface; + + pipe->surface_copy(pipe, + dst, + dstx, + dsty, + src, + srcx, + srcy, + width, + height); +} + +static void +identity_surface_fill(struct pipe_context *_pipe, + struct pipe_surface *_dst, + unsigned dstx, + unsigned dsty, + unsigned width, + unsigned height, + unsigned value) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct identity_surface *id_surface_dst = identity_surface(_dst); + struct pipe_context *pipe = id_pipe->pipe; + struct pipe_surface *dst = id_surface_dst->surface; + + pipe->surface_fill(pipe, + dst, + dstx, + dsty, + width, + height, + value); +} + +static void +identity_clear(struct pipe_context *_pipe, + unsigned buffers, + const float *rgba, + double depth, + unsigned stencil) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->clear(pipe, + buffers, + rgba, + depth, + stencil); +} + +static void +identity_flush(struct pipe_context *_pipe, + unsigned flags, + struct pipe_fence_handle **fence) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct pipe_context *pipe = id_pipe->pipe; + + pipe->flush(pipe, + flags, + fence); +} + +static unsigned int +identity_is_texture_referenced(struct pipe_context *_pipe, + struct pipe_texture *_texture, + unsigned face, + unsigned level) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct identity_texture *id_texture = identity_texture(_texture); + struct pipe_context *pipe = id_pipe->pipe; + struct pipe_texture *texture = id_texture->texture; + + return pipe->is_texture_referenced(pipe, + texture, + face, + level); +} + +static unsigned int +identity_is_buffer_referenced(struct pipe_context *_pipe, + struct pipe_buffer *_buffer) +{ + struct identity_context *id_pipe = identity_context(_pipe); + struct identity_buffer *id_buffer = identity_buffer(_buffer); + struct pipe_context *pipe = id_pipe->pipe; + struct pipe_buffer *buffer = id_buffer->buffer; + + return pipe->is_buffer_referenced(pipe, + buffer); +} + +struct pipe_context * +identity_context_create(struct pipe_screen *_screen, struct pipe_context *pipe) +{ + struct identity_context *id_pipe; + (void)identity_screen(_screen); + + id_pipe = CALLOC_STRUCT(identity_context); + if (!id_pipe) { + return NULL; + } + + id_pipe->base.winsys = NULL; + id_pipe->base.screen = _screen; + id_pipe->base.priv = pipe->priv; + id_pipe->base.draw = NULL; + + id_pipe->base.destroy = identity_destroy; + id_pipe->base.set_edgeflags = identity_set_edgeflags; + id_pipe->base.draw_arrays = identity_draw_arrays; + id_pipe->base.draw_elements = identity_draw_elements; + id_pipe->base.draw_range_elements = identity_draw_range_elements; + id_pipe->base.create_query = identity_create_query; + id_pipe->base.destroy_query = identity_destroy_query; + id_pipe->base.begin_query = identity_begin_query; + id_pipe->base.end_query = identity_end_query; + id_pipe->base.get_query_result = identity_get_query_result; + id_pipe->base.create_blend_state = identity_create_blend_state; + id_pipe->base.bind_blend_state = identity_bind_blend_state; + id_pipe->base.delete_blend_state = identity_delete_blend_state; + id_pipe->base.create_sampler_state = identity_create_sampler_state; + id_pipe->base.bind_sampler_states = identity_bind_sampler_states; + id_pipe->base.delete_sampler_state = identity_delete_sampler_state; + id_pipe->base.create_rasterizer_state = identity_create_rasterizer_state; + id_pipe->base.bind_rasterizer_state = identity_bind_rasterizer_state; + id_pipe->base.delete_rasterizer_state = identity_delete_rasterizer_state; + id_pipe->base.create_depth_stencil_alpha_state = identity_create_depth_stencil_alpha_state; + id_pipe->base.bind_depth_stencil_alpha_state = identity_bind_depth_stencil_alpha_state; + id_pipe->base.delete_depth_stencil_alpha_state = identity_delete_depth_stencil_alpha_state; + id_pipe->base.create_fs_state = identity_create_fs_state; + id_pipe->base.bind_fs_state = identity_bind_fs_state; + id_pipe->base.delete_fs_state = identity_delete_fs_state; + id_pipe->base.create_vs_state = identity_create_vs_state; + id_pipe->base.bind_vs_state = identity_bind_vs_state; + id_pipe->base.delete_vs_state = identity_delete_vs_state; + id_pipe->base.set_blend_color = identity_set_blend_color; + id_pipe->base.set_clip_state = identity_set_clip_state; + id_pipe->base.set_constant_buffer = identity_set_constant_buffer; + id_pipe->base.set_framebuffer_state = identity_set_framebuffer_state; + id_pipe->base.set_polygon_stipple = identity_set_polygon_stipple; + id_pipe->base.set_scissor_state = identity_set_scissor_state; + id_pipe->base.set_viewport_state = identity_set_viewport_state; + id_pipe->base.set_sampler_textures = identity_set_sampler_textures; + id_pipe->base.set_vertex_buffers = identity_set_vertex_buffers; + id_pipe->base.set_vertex_elements = identity_set_vertex_elements; + id_pipe->base.surface_copy = identity_surface_copy; + id_pipe->base.surface_fill = identity_surface_fill; + id_pipe->base.clear = identity_clear; + id_pipe->base.flush = identity_flush; + id_pipe->base.is_texture_referenced = identity_is_texture_referenced; + id_pipe->base.is_buffer_referenced = identity_is_buffer_referenced; + + id_pipe->pipe = pipe; + + return &id_pipe->base; +} diff --git a/src/gallium/drivers/identity/id_context.h b/src/gallium/drivers/identity/id_context.h new file mode 100644 index 00000000000..75b73fc7df6 --- /dev/null +++ b/src/gallium/drivers/identity/id_context.h @@ -0,0 +1,48 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef ID_CONTEXT_H +#define ID_CONTEXT_H + +#include "pipe/p_state.h" +#include "pipe/p_context.h" + + +struct identity_context { + struct pipe_context base; /**< base class */ + + struct pipe_context *pipe; +}; + + +static INLINE struct identity_context * +identity_context(struct pipe_context *pipe) +{ + return (struct identity_context *)pipe; +} + +#endif /* ID_CONTEXT_H */ diff --git a/src/gallium/drivers/identity/id_drm.c b/src/gallium/drivers/identity/id_drm.c new file mode 100644 index 00000000000..14f68ac0d00 --- /dev/null +++ b/src/gallium/drivers/identity/id_drm.c @@ -0,0 +1,173 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "state_tracker/drm_api.h" + +#include "util/u_memory.h" +#include "identity/id_drm.h" +#include "identity/id_screen.h" +#include "identity/id_public.h" +#include "identity/id_screen.h" +#include "identity/id_objects.h" + +struct identity_drm_api +{ + struct drm_api base; + + struct drm_api *api; +}; + +static INLINE struct identity_drm_api * +identity_drm_api(struct drm_api *_api) +{ + return (struct identity_drm_api *)_api; +} + +static struct pipe_screen * +identity_drm_create_screen(struct drm_api *_api, int fd, + struct drm_create_screen_arg *arg) +{ + struct identity_drm_api *id_api = identity_drm_api(_api); + struct drm_api *api = id_api->api; + struct pipe_screen *screen; + + if (arg && arg->mode != DRM_CREATE_NORMAL) + return NULL; + + screen = api->create_screen(api, fd, arg); + + return identity_screen_create(screen); +} + +static struct pipe_context * +identity_drm_create_context(struct drm_api *_api, + struct pipe_screen *_screen) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_drm_api *id_api = identity_drm_api(_api); + struct pipe_screen *screen = id_screen->screen; + struct drm_api *api = id_api->api; + struct pipe_context *pipe; + + pipe = api->create_context(api, screen); + + pipe = identity_context_create(_screen, pipe); + + return pipe; +} + +static struct pipe_texture * +identity_drm_texture_from_shared_handle(struct drm_api *_api, + struct pipe_screen *_screen, + struct pipe_texture *templ, + const char *name, + unsigned stride, + unsigned handle) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_drm_api *id_api = identity_drm_api(_api); + struct pipe_screen *screen = id_screen->screen; + struct drm_api *api = id_api->api; + struct pipe_texture *result; + + result = api->texture_from_shared_handle(api, screen, templ, name, stride, handle); + + result = identity_texture_create(identity_screen(_screen), result); + + return result; +} + +static boolean +identity_drm_shared_handle_from_texture(struct drm_api *_api, + struct pipe_screen *_screen, + struct pipe_texture *_texture, + unsigned *stride, + unsigned *handle) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_texture *id_texture = identity_texture(_texture); + struct identity_drm_api *id_api = identity_drm_api(_api); + struct pipe_screen *screen = id_screen->screen; + struct pipe_texture *texture = id_texture->texture; + struct drm_api *api = id_api->api; + + return api->shared_handle_from_texture(api, screen, texture, stride, handle); +} + +static boolean +identity_drm_local_handle_from_texture(struct drm_api *_api, + struct pipe_screen *_screen, + struct pipe_texture *_texture, + unsigned *stride, + unsigned *handle) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_texture *id_texture = identity_texture(_texture); + struct identity_drm_api *id_api = identity_drm_api(_api); + struct pipe_screen *screen = id_screen->screen; + struct pipe_texture *texture = id_texture->texture; + struct drm_api *api = id_api->api; + + return api->local_handle_from_texture(api, screen, texture, stride, handle); +} + +static void +identity_drm_destroy(struct drm_api *_api) +{ + struct identity_drm_api *id_api = identity_drm_api(_api); + struct drm_api *api = id_api->api; + api->destroy(api); + + free(id_api); +} + +struct drm_api * +identity_drm_create(struct drm_api *api) +{ + struct identity_drm_api *id_api; + + if (!api) + goto error; + + id_api = CALLOC_STRUCT(identity_drm_api); + + if (!id_api) + goto error; + + id_api->base.create_screen = identity_drm_create_screen; + id_api->base.create_context = identity_drm_create_context; + id_api->base.texture_from_shared_handle = identity_drm_texture_from_shared_handle; + id_api->base.shared_handle_from_texture = identity_drm_shared_handle_from_texture; + id_api->base.local_handle_from_texture = identity_drm_local_handle_from_texture; + id_api->base.destroy = identity_drm_destroy; + id_api->api = api; + + return &id_api->base; + +error: + return api; +} diff --git a/src/gallium/drivers/identity/id_drm.h b/src/gallium/drivers/identity/id_drm.h new file mode 100644 index 00000000000..cf2ad2ce07f --- /dev/null +++ b/src/gallium/drivers/identity/id_drm.h @@ -0,0 +1,35 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef ID_DRM_H +#define ID_DRM_H + +struct drm_api; + +struct drm_api* identity_drm_create(struct drm_api *api); + +#endif /* ID_DRM_H */ diff --git a/src/gallium/drivers/identity/id_objects.c b/src/gallium/drivers/identity/id_objects.c new file mode 100644 index 00000000000..e893e599408 --- /dev/null +++ b/src/gallium/drivers/identity/id_objects.c @@ -0,0 +1,182 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "util/u_memory.h" + +#include "id_public.h" +#include "id_screen.h" +#include "id_objects.h" + +struct pipe_buffer * +identity_buffer_create(struct identity_screen *id_screen, + struct pipe_buffer *buffer) +{ + struct identity_buffer *id_buffer; + + if(!buffer) + goto error; + + assert(buffer->screen == id_screen->screen); + + id_buffer = CALLOC_STRUCT(identity_buffer); + if(!id_buffer) + goto error; + + memcpy(&id_buffer->base, buffer, sizeof(struct pipe_buffer)); + + pipe_reference_init(&id_buffer->base.reference, 1); + id_buffer->base.screen = &id_screen->base; + id_buffer->buffer = buffer; + + return &id_buffer->base; + +error: + pipe_buffer_reference(&buffer, NULL); + return NULL; +} + +void +identity_buffer_destroy(struct identity_buffer *id_buffer) +{ + pipe_buffer_reference(&id_buffer->buffer, NULL); + FREE(id_buffer); +} + + +struct pipe_texture * +identity_texture_create(struct identity_screen *id_screen, + struct pipe_texture *texture) +{ + struct identity_texture *id_texture; + + if(!texture) + goto error; + + assert(texture->screen == id_screen->screen); + + id_texture = CALLOC_STRUCT(identity_texture); + if(!id_texture) + goto error; + + memcpy(&id_texture->base, texture, sizeof(struct pipe_texture)); + + pipe_reference_init(&id_texture->base.reference, 1); + id_texture->base.screen = &id_screen->base; + id_texture->texture = texture; + + return &id_texture->base; + +error: + pipe_texture_reference(&texture, NULL); + return NULL; +} + +void +identity_texture_destroy(struct identity_texture *id_texture) +{ + pipe_texture_reference(&id_texture->texture, NULL); + FREE(id_texture); +} + + +struct pipe_surface * +identity_surface_create(struct identity_texture *id_texture, + struct pipe_surface *surface) +{ + struct identity_surface *id_surface; + + if(!surface) + goto error; + + assert(surface->texture == id_texture->texture); + + id_surface = CALLOC_STRUCT(identity_surface); + if(!id_surface) + goto error; + + memcpy(&id_surface->base, surface, sizeof(struct pipe_surface)); + + pipe_reference_init(&id_surface->base.reference, 1); + id_surface->base.texture = NULL; + pipe_texture_reference(&id_surface->base.texture, &id_texture->base); + id_surface->surface = surface; + + return &id_surface->base; + +error: + pipe_surface_reference(&surface, NULL); + return NULL; +} + +void +identity_surface_destroy(struct identity_surface *id_surface) +{ + pipe_texture_reference(&id_surface->base.texture, NULL); + pipe_surface_reference(&id_surface->surface, NULL); + FREE(id_surface); +} + + +struct pipe_transfer * +identity_transfer_create(struct identity_texture *id_texture, + struct pipe_transfer *transfer) +{ + struct identity_transfer *id_transfer; + + if(!transfer) + goto error; + + assert(transfer->texture == id_texture->texture); + + id_transfer = CALLOC_STRUCT(identity_transfer); + if(!id_transfer) + goto error; + + memcpy(&id_transfer->base, transfer, sizeof(struct pipe_transfer)); + + id_transfer->base.texture = NULL; + pipe_texture_reference(&id_transfer->base.texture, &id_texture->base); + id_transfer->transfer = transfer; + assert(id_transfer->base.texture == &id_texture->base); + + return &id_transfer->base; + +error: + transfer->texture->screen->tex_transfer_destroy(transfer); + return NULL; +} + +void +identity_transfer_destroy(struct identity_transfer *id_transfer) +{ + struct identity_screen *id_screen = identity_screen(id_transfer->base.texture->screen); + struct pipe_screen *screen = id_screen->screen; + + pipe_texture_reference(&id_transfer->base.texture, NULL); + screen->tex_transfer_destroy(id_transfer->transfer); + FREE(id_transfer); +} diff --git a/src/gallium/drivers/identity/id_objects.h b/src/gallium/drivers/identity/id_objects.h new file mode 100644 index 00000000000..ce58faa3c7c --- /dev/null +++ b/src/gallium/drivers/identity/id_objects.h @@ -0,0 +1,169 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef ID_OBJECTS_H +#define ID_OBJECTS_H + + +#include "pipe/p_compiler.h" +#include "pipe/p_state.h" + +#include "id_screen.h" + + +struct identity_buffer +{ + struct pipe_buffer base; + + struct pipe_buffer *buffer; +}; + + +struct identity_texture +{ + struct pipe_texture base; + + struct pipe_texture *texture; +}; + + +struct identity_surface +{ + struct pipe_surface base; + + struct pipe_surface *surface; +}; + + +struct identity_transfer +{ + struct pipe_transfer base; + + struct pipe_transfer *transfer; +}; + + +static INLINE struct identity_buffer * +identity_buffer(struct pipe_buffer *_buffer) +{ + if(!_buffer) + return NULL; + (void)identity_screen(_buffer->screen); + return (struct identity_buffer *)_buffer; +} + +static INLINE struct identity_texture * +identity_texture(struct pipe_texture *_texture) +{ + if(!_texture) + return NULL; + (void)identity_screen(_texture->screen); + return (struct identity_texture *)_texture; +} + +static INLINE struct identity_surface * +identity_surface(struct pipe_surface *_surface) +{ + if(!_surface) + return NULL; + (void)identity_texture(_surface->texture); + return (struct identity_surface *)_surface; +} + +static INLINE struct identity_transfer * +identity_transfer(struct pipe_transfer *_transfer) +{ + if(!_transfer) + return NULL; + (void)identity_texture(_transfer->texture); + return (struct identity_transfer *)_transfer; +} + + +static INLINE struct pipe_buffer * +identity_buffer_unwrap(struct pipe_buffer *_buffer) +{ + if(!_buffer) + return NULL; + return identity_buffer(_buffer)->buffer; +} + +static INLINE struct pipe_texture * +identity_texture_unwrap(struct pipe_texture *_texture) +{ + if(!_texture) + return NULL; + return identity_texture(_texture)->texture; +} + +static INLINE struct pipe_surface * +identity_surface_unwrap(struct pipe_surface *_surface) +{ + if(!_surface) + return NULL; + return identity_surface(_surface)->surface; +} + +static INLINE struct pipe_transfer * +identity_transfer_unwrap(struct pipe_transfer *_transfer) +{ + if(!_transfer) + return NULL; + return identity_transfer(_transfer)->transfer; +} + + +struct pipe_buffer * +identity_buffer_create(struct identity_screen *id_screen, + struct pipe_buffer *buffer); + +void +identity_buffer_destroy(struct identity_buffer *id_buffer); + +struct pipe_texture * +identity_texture_create(struct identity_screen *id_screen, + struct pipe_texture *texture); + +void +identity_texture_destroy(struct identity_texture *id_texture); + +struct pipe_surface * +identity_surface_create(struct identity_texture *id_texture, + struct pipe_surface *surface); + +void +identity_surface_destroy(struct identity_surface *id_surface); + +struct pipe_transfer * +identity_transfer_create(struct identity_texture *id_texture, + struct pipe_transfer *transfer); + +void +identity_transfer_destroy(struct identity_transfer *id_transfer); + + +#endif /* ID_OBJECTS_H */ diff --git a/src/gallium/drivers/identity/id_public.h b/src/gallium/drivers/identity/id_public.h new file mode 100644 index 00000000000..cac14cfd604 --- /dev/null +++ b/src/gallium/drivers/identity/id_public.h @@ -0,0 +1,40 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef ID_PUBLIC_H +#define ID_PUBLIC_H + +struct pipe_screen; +struct pipe_context; + +struct pipe_screen * +identity_screen_create(struct pipe_screen *screen); + +struct pipe_context * +identity_context_create(struct pipe_screen *screen, struct pipe_context *pipe); + +#endif /* PT_PUBLIC_H */ diff --git a/src/gallium/drivers/identity/id_screen.c b/src/gallium/drivers/identity/id_screen.c new file mode 100644 index 00000000000..26439637d08 --- /dev/null +++ b/src/gallium/drivers/identity/id_screen.c @@ -0,0 +1,483 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "pipe/p_screen.h" +#include "pipe/p_state.h" +#include "util/u_memory.h" + +#include "id_public.h" +#include "id_screen.h" +#include "id_objects.h" + + +static void +identity_screen_destroy(struct pipe_screen *_screen) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + + screen->destroy(screen); + + FREE(id_screen); +} + +static const char * +identity_screen_get_name(struct pipe_screen *_screen) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + + return screen->get_name(screen); +} + +static const char * +identity_screen_get_vendor(struct pipe_screen *_screen) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + + return screen->get_vendor(screen); +} + +static int +identity_screen_get_param(struct pipe_screen *_screen, + int param) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + + return screen->get_param(screen, + param); +} + +static float +identity_screen_get_paramf(struct pipe_screen *_screen, + int param) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + + return screen->get_paramf(screen, + param); +} + +static boolean +identity_screen_is_format_supported(struct pipe_screen *_screen, + enum pipe_format format, + enum pipe_texture_target target, + unsigned tex_usage, + unsigned geom_flags) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + + return screen->is_format_supported(screen, + format, + target, + tex_usage, + geom_flags); +} + +static struct pipe_texture * +identity_screen_texture_create(struct pipe_screen *_screen, + const struct pipe_texture *templat) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + struct pipe_texture *result; + + result = screen->texture_create(screen, + templat); + + if (result) + return identity_texture_create(id_screen, result); + return NULL; +} + +static struct pipe_texture * +identity_screen_texture_blanket(struct pipe_screen *_screen, + const struct pipe_texture *templat, + const unsigned *stride, + struct pipe_buffer *_buffer) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_buffer *id_buffer = identity_buffer(_buffer); + struct pipe_screen *screen = id_screen->screen; + struct pipe_buffer *buffer = id_buffer->buffer; + struct pipe_texture *result; + + result = screen->texture_blanket(screen, + templat, + stride, + buffer); + + if (result) + return identity_texture_create(id_screen, result); + return NULL; +} + +static void +identity_screen_texture_destroy(struct pipe_texture *_texture) +{ + identity_texture_destroy(identity_texture(_texture)); +} + +static struct pipe_surface * +identity_screen_get_tex_surface(struct pipe_screen *_screen, + struct pipe_texture *_texture, + unsigned face, + unsigned level, + unsigned zslice, + unsigned usage) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_texture *id_texture = identity_texture(_texture); + struct pipe_screen *screen = id_screen->screen; + struct pipe_texture *texture = id_texture->texture; + struct pipe_surface *result; + + result = screen->get_tex_surface(screen, + texture, + face, + level, + zslice, + usage); + + if (result) + return identity_surface_create(id_texture, result); + return NULL; +} + +static void +identity_screen_tex_surface_destroy(struct pipe_surface *_surface) +{ + identity_surface_destroy(identity_surface(_surface)); +} + +static struct pipe_transfer * +identity_screen_get_tex_transfer(struct pipe_screen *_screen, + struct pipe_texture *_texture, + unsigned face, + unsigned level, + unsigned zslice, + enum pipe_transfer_usage usage, + unsigned x, + unsigned y, + unsigned w, + unsigned h) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_texture *id_texture = identity_texture(_texture); + struct pipe_screen *screen = id_screen->screen; + struct pipe_texture *texture = id_texture->texture; + struct pipe_transfer *result; + + result = screen->get_tex_transfer(screen, + texture, + face, + level, + zslice, + usage, + x, + y, + w, + h); + + if (result) + return identity_transfer_create(id_texture, result); + return NULL; +} + +static void +identity_screen_tex_transfer_destroy(struct pipe_transfer *_transfer) +{ + identity_transfer_destroy(identity_transfer(_transfer)); +} + +static void * +identity_screen_transfer_map(struct pipe_screen *_screen, + struct pipe_transfer *_transfer) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_transfer *id_transfer = identity_transfer(_transfer); + struct pipe_screen *screen = id_screen->screen; + struct pipe_transfer *transfer = id_transfer->transfer; + + return screen->transfer_map(screen, + transfer); +} + +static void +identity_screen_transfer_unmap(struct pipe_screen *_screen, + struct pipe_transfer *_transfer) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_transfer *id_transfer = identity_transfer(_transfer); + struct pipe_screen *screen = id_screen->screen; + struct pipe_transfer *transfer = id_transfer->transfer; + + screen->transfer_unmap(screen, + transfer); +} + +static struct pipe_buffer * +identity_screen_buffer_create(struct pipe_screen *_screen, + unsigned alignment, + unsigned usage, + unsigned size) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + struct pipe_buffer *result; + + result = screen->buffer_create(screen, + alignment, + usage, + size); + + if (result) + return identity_buffer_create(id_screen, result); + return NULL; +} + +static struct pipe_buffer * +identity_screen_user_buffer_create(struct pipe_screen *_screen, + void *ptr, + unsigned bytes) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + struct pipe_buffer *result; + + result = screen->user_buffer_create(screen, + ptr, + bytes); + + if (result) + return identity_buffer_create(id_screen, result); + return NULL; +} + +static struct pipe_buffer * +identity_screen_surface_buffer_create(struct pipe_screen *_screen, + unsigned width, + unsigned height, + enum pipe_format format, + unsigned usage, + unsigned tex_usage, + unsigned *stride) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + struct pipe_buffer *result; + + result = screen->surface_buffer_create(screen, + width, + height, + format, + usage, + tex_usage, + stride); + + if (result) + return identity_buffer_create(id_screen, result); + return NULL; +} + +static void * +identity_screen_buffer_map(struct pipe_screen *_screen, + struct pipe_buffer *_buffer, + unsigned usage) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_buffer *id_buffer = identity_buffer(_buffer); + struct pipe_screen *screen = id_screen->screen; + struct pipe_buffer *buffer = id_buffer->buffer; + + return screen->buffer_map(screen, + buffer, + usage); +} + +static void * +identity_screen_buffer_map_range(struct pipe_screen *_screen, + struct pipe_buffer *_buffer, + unsigned offset, + unsigned length, + unsigned usage) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_buffer *id_buffer = identity_buffer(_buffer); + struct pipe_screen *screen = id_screen->screen; + struct pipe_buffer *buffer = id_buffer->buffer; + + return screen->buffer_map_range(screen, + buffer, + offset, + length, + usage); +} + +static void +identity_screen_buffer_flush_mapped_range(struct pipe_screen *_screen, + struct pipe_buffer *_buffer, + unsigned offset, + unsigned length) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_buffer *id_buffer = identity_buffer(_buffer); + struct pipe_screen *screen = id_screen->screen; + struct pipe_buffer *buffer = id_buffer->buffer; + + screen->buffer_flush_mapped_range(screen, + buffer, + offset, + length); +} + +static void +identity_screen_buffer_unmap(struct pipe_screen *_screen, + struct pipe_buffer *_buffer) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_buffer *id_buffer = identity_buffer(_buffer); + struct pipe_screen *screen = id_screen->screen; + struct pipe_buffer *buffer = id_buffer->buffer; + + screen->buffer_unmap(screen, + buffer); +} + +static void +identity_screen_buffer_destroy(struct pipe_buffer *_buffer) +{ + identity_buffer_destroy(identity_buffer(_buffer)); +} + +static void +identity_screen_flush_frontbuffer(struct pipe_screen *_screen, + struct pipe_surface *_surface, + void *context_private) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct identity_surface *id_surface = identity_surface(_surface); + struct pipe_screen *screen = id_screen->screen; + struct pipe_surface *surface = id_surface->surface; + + screen->flush_frontbuffer(screen, + surface, + context_private); +} + +static void +identity_screen_fence_reference(struct pipe_screen *_screen, + struct pipe_fence_handle **ptr, + struct pipe_fence_handle *fence) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + + screen->fence_reference(screen, + ptr, + fence); +} + +static int +identity_screen_fence_signalled(struct pipe_screen *_screen, + struct pipe_fence_handle *fence, + unsigned flags) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + + return screen->fence_signalled(screen, + fence, + flags); +} + +static int +identity_screen_fence_finish(struct pipe_screen *_screen, + struct pipe_fence_handle *fence, + unsigned flags) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + + return screen->fence_finish(screen, + fence, + flags); +} + +struct pipe_screen * +identity_screen_create(struct pipe_screen *screen) +{ + struct identity_screen *id_screen; + + id_screen = CALLOC_STRUCT(identity_screen); + if (!id_screen) { + return NULL; + } + + id_screen->base.winsys = NULL; + + id_screen->base.destroy = identity_screen_destroy; + id_screen->base.get_name = identity_screen_get_name; + id_screen->base.get_vendor = identity_screen_get_vendor; + id_screen->base.get_param = identity_screen_get_param; + id_screen->base.get_paramf = identity_screen_get_paramf; + id_screen->base.is_format_supported = identity_screen_is_format_supported; + id_screen->base.texture_create = identity_screen_texture_create; + id_screen->base.texture_blanket = identity_screen_texture_blanket; + id_screen->base.texture_destroy = identity_screen_texture_destroy; + id_screen->base.get_tex_surface = identity_screen_get_tex_surface; + id_screen->base.tex_surface_destroy = identity_screen_tex_surface_destroy; + id_screen->base.get_tex_transfer = identity_screen_get_tex_transfer; + id_screen->base.tex_transfer_destroy = identity_screen_tex_transfer_destroy; + id_screen->base.transfer_map = identity_screen_transfer_map; + id_screen->base.transfer_unmap = identity_screen_transfer_unmap; + id_screen->base.buffer_create = identity_screen_buffer_create; + id_screen->base.user_buffer_create = identity_screen_user_buffer_create; + id_screen->base.surface_buffer_create = identity_screen_surface_buffer_create; + if (screen->buffer_map) + id_screen->base.buffer_map = identity_screen_buffer_map; + if (screen->buffer_map_range) + id_screen->base.buffer_map_range = identity_screen_buffer_map_range; + if (screen->buffer_flush_mapped_range) + id_screen->base.buffer_flush_mapped_range = identity_screen_buffer_flush_mapped_range; + if (screen->buffer_unmap) + id_screen->base.buffer_unmap = identity_screen_buffer_unmap; + id_screen->base.buffer_destroy = identity_screen_buffer_destroy; + id_screen->base.flush_frontbuffer = identity_screen_flush_frontbuffer; + id_screen->base.fence_reference = identity_screen_fence_reference; + id_screen->base.fence_signalled = identity_screen_fence_signalled; + id_screen->base.fence_finish = identity_screen_fence_finish; + + id_screen->screen = screen; + + return &id_screen->base; +} diff --git a/src/gallium/drivers/identity/id_screen.h b/src/gallium/drivers/identity/id_screen.h new file mode 100644 index 00000000000..2c4f1290894 --- /dev/null +++ b/src/gallium/drivers/identity/id_screen.h @@ -0,0 +1,48 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef ID_SCREEN_H +#define ID_SCREEN_H + +#include "pipe/p_screen.h" +#include "pipe/p_defines.h" + + +struct identity_screen { + struct pipe_screen base; + + struct pipe_screen *screen; +}; + + +static INLINE struct identity_screen * +identity_screen(struct pipe_screen *screen) +{ + return (struct identity_screen *)screen; +} + +#endif /* ID_SCREEN_H */ diff --git a/src/gallium/drivers/llvmpipe/Makefile b/src/gallium/drivers/llvmpipe/Makefile new file mode 100644 index 00000000000..5ac09de79ea --- /dev/null +++ b/src/gallium/drivers/llvmpipe/Makefile @@ -0,0 +1,54 @@ +TOP = ../../../.. +include $(TOP)/configs/current + +LIBNAME = llvmpipe + +CFLAGS += -D__STDC_CONSTANT_MACROS -D__STDC_LIMIT_MACROS + +C_SOURCES = \ + lp_bld_alpha.c \ + lp_bld_arit.c \ + lp_bld_blend_aos.c \ + lp_bld_blend_logicop.c \ + lp_bld_blend_soa.c \ + lp_bld_const.c \ + lp_bld_conv.c \ + lp_bld_debug.c \ + lp_bld_depth.c \ + lp_bld_flow.c \ + lp_bld_format_aos.c \ + lp_bld_interp.c \ + lp_bld_intr.c \ + lp_bld_logic.c \ + lp_bld_swizzle.c \ + lp_bld_struct.c \ + lp_bld_tgsi_soa.c \ + lp_bld_type.c \ + lp_buffer.c \ + lp_clear.c \ + lp_context.c \ + lp_draw_arrays.c \ + lp_flush.c \ + lp_jit.c \ + lp_prim_setup.c \ + lp_prim_vbuf.c \ + lp_setup.c \ + lp_query.c \ + lp_screen.c \ + lp_state_blend.c \ + lp_state_clip.c \ + lp_state_derived.c \ + lp_state_fs.c \ + lp_state_rasterizer.c \ + lp_state_sampler.c \ + lp_state_surface.c \ + lp_state_vertex.c \ + lp_state_vs.c \ + lp_surface.c \ + lp_tex_cache.c \ + lp_tex_sample.c \ + lp_texture.c \ + lp_tile_cache.c \ + lp_tile_soa.c + +include ../../Makefile.template diff --git a/src/gallium/drivers/llvmpipe/README b/src/gallium/drivers/llvmpipe/README new file mode 100644 index 00000000000..498d21dea6c --- /dev/null +++ b/src/gallium/drivers/llvmpipe/README @@ -0,0 +1,142 @@ +LLVMPIPE -- a fork of softpipe that employs LLVM for code generation. + + +Status +====== + +Done so far is: + + - the whole fragment pipeline is code generated in a single function + + - depth testing + + - fragment shader TGSI translation + - same level of support as the TGSI SSE2 exec machine, with the exception + we don't fallback to TGSI interpretation when an unsupported opcode is + found, but just ignore it + - texture sampling via an intrinsic call + - done in SoA layout + - input interpolation also code generated + + - alpha testing + + - blend (including logic ops) + - both in SoA and AoS layouts, but only the former used for now + + - code is generic + - intermediates can be vectors of floats, ubytes, fixed point, etc, and of + any width and length + - not all operations are implemented for these types yet though + +Most mesa/progs/demos/* work. Speed is on par with Keith's softpipe-opt branch, +which includes hand written fast implementations for common cases. + +To do (probably by this order): + + - code generate stipple and stencil testing + + - code generate texture sampling + + - translate TGSI control flow instructions, and all other remaining opcodes + + - code generate the triangle setup and rasterization + + +Requirements +============ + + - Linux + + - udis86, http://udis86.sourceforge.net/ . Use my repository, which decodes + opcodes not yet supported by upstream. + + git clone git://people.freedesktop.org/~jrfonseca/udis86 + cd udis86 + ./configure --with-pic + make + sudo make install + + - LLVM 2.5. On Debian based distributions do: + + aptitude install llvm-dev + + There is a typo in one of the llvm-dev 2.5 headers, that causes compilation + errors in the debug build: + + --- /usr/include/llvm-c/Core.h.orig 2009-08-10 15:38:54.000000000 +0100 + +++ /usr/include/llvm-c/Core.h 2009-08-10 15:38:25.000000000 +0100 + @@ -831,7 +831,7 @@ + template<typename T> + inline T **unwrap(LLVMValueRef *Vals, unsigned Length) { + #if DEBUG + - for (LLVMValueRef *I = Vals, E = Vals + Length; I != E; ++I) + + for (LLVMValueRef *I = Vals, *E = Vals + Length; I != E; ++I) + cast<T>(*I); + #endif + return reinterpret_cast<T**>(Vals); + + - A x86 or amd64 processor with support for sse2, sse3, and sse4.1 SIMD + instructions. This is necessary because we emit several SSE intrinsics for + convenience. See /proc/cpuinfo to know what your CPU supports. + + - scons + + +Building +======== + +To build everything invoke scons as: + + scons debug=yes statetrackers=mesa drivers=llvmpipe winsys=xlib dri=false -k + +Alternatively, you can build it with GNU make, if you prefer, by invoking it as + + make linux-llvm + +but the rest of these instructions assume scons is used. + + +Using +===== + +Building will create a drop-in alternative for libGL.so. To use it set the +environment variables: + + export LD_LIBRARY_PATH=$PWD/build/linux-x86_64-debug/lib:$LD_LIBRARY_PATH + +or + + export LD_LIBRARY_PATH=$PWD/build/linux-x86-debug/lib:$LD_LIBRARY_PATH + + +Unit testing +============ + +Building will also create several unit tests in +build/linux-???-debug/gallium/drivers/llvmpipe: + + - lp_test_blend: blending + - lp_test_conv: SIMD vector conversion + - lp_test_format: pixel unpacking/packing + +Some of this tests can output results and benchmarks to a tab-seperated-file +for posterior analysis, e.g.: + + build/linux-x86_64-debug/gallium/drivers/llvmpipe/lp_test_blend -o blend.tsv + + +Development Notes +================= + +- When looking to this code by the first time start in lp_state_fs.c, and + then skim through the lp_bld_* functions called in there, and the comments + at the top of the lp_bld_*.c functions. + +- All lp_bld_*.[ch] are isolated from the rest of the driver, and could/may be + put in a standalone Gallium state -> LLVM IR translation module. + +- We use LLVM-C bindings for now. They are not documented, but follow the C++ + interfaces very closely, and appear to be complete enough for code + generation. See + http://npcontemplation.blogspot.com/2008/06/secret-of-llvm-c-bindings.html + for a standalone example. diff --git a/src/gallium/drivers/llvmpipe/SConscript b/src/gallium/drivers/llvmpipe/SConscript new file mode 100644 index 00000000000..5c29bdac56e --- /dev/null +++ b/src/gallium/drivers/llvmpipe/SConscript @@ -0,0 +1,81 @@ +Import('*') + +env = env.Clone() + +env.Tool('llvm') +if 'LLVM_VERSION' not in env: + print 'warning: LLVM not found: not building llvmpipe' + Return() + +env.Tool('udis86') + +llvmpipe = env.ConvenienceLibrary( + target = 'llvmpipe', + source = [ + 'lp_bld_alpha.c', + 'lp_bld_arit.c', + 'lp_bld_blend_aos.c', + 'lp_bld_blend_logicop.c', + 'lp_bld_blend_soa.c', + 'lp_bld_const.c', + 'lp_bld_conv.c', + 'lp_bld_debug.c', + 'lp_bld_depth.c', + 'lp_bld_flow.c', + 'lp_bld_format_aos.c', + 'lp_bld_interp.c', + 'lp_bld_intr.c', + 'lp_bld_struct.c', + 'lp_bld_logic.c', + 'lp_bld_swizzle.c', + 'lp_bld_tgsi_soa.c', + 'lp_bld_type.c', + 'lp_buffer.c', + 'lp_clear.c', + 'lp_context.c', + 'lp_draw_arrays.c', + 'lp_flush.c', + 'lp_jit.c', + 'lp_prim_setup.c', + 'lp_prim_vbuf.c', + 'lp_setup.c', + 'lp_query.c', + 'lp_screen.c', + 'lp_state_blend.c', + 'lp_state_clip.c', + 'lp_state_derived.c', + 'lp_state_fs.c', + 'lp_state_rasterizer.c', + 'lp_state_sampler.c', + 'lp_state_surface.c', + 'lp_state_vertex.c', + 'lp_state_vs.c', + 'lp_surface.c', + 'lp_tex_cache.c', + 'lp_tex_sample.c', + 'lp_texture.c', + 'lp_tile_cache.c', + 'lp_tile_soa.c', + ]) + + +env = env.Clone() + +env.Prepend(LIBS = [llvmpipe] + auxiliaries) + +env.Program( + target = 'lp_test_format', + source = ['lp_test_format.c'], +) + +env.Program( + target = 'lp_test_blend', + source = ['lp_test_blend.c', 'lp_test_main.c'], +) + +env.Program( + target = 'lp_test_conv', + source = ['lp_test_conv.c', 'lp_test_main.c'], +) + +Export('llvmpipe') diff --git a/src/gallium/drivers/llvmpipe/lp_bld_alpha.c b/src/gallium/drivers/llvmpipe/lp_bld_alpha.c new file mode 100644 index 00000000000..49c2f911af7 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_alpha.c @@ -0,0 +1,64 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Alpha testing to LLVM IR translation. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#include "pipe/p_state.h" + +#include "lp_bld_type.h" +#include "lp_bld_const.h" +#include "lp_bld_arit.h" +#include "lp_bld_logic.h" +#include "lp_bld_flow.h" +#include "lp_bld_debug.h" +#include "lp_bld_alpha.h" + + +void +lp_build_alpha_test(LLVMBuilderRef builder, + const struct pipe_alpha_state *state, + union lp_type type, + struct lp_build_mask_context *mask, + LLVMValueRef alpha, + LLVMValueRef ref) +{ + struct lp_build_context bld; + + lp_build_context_init(&bld, builder, type); + + if(state->enabled) { + LLVMValueRef test = lp_build_cmp(&bld, state->func, alpha, ref); + + lp_build_name(test, "alpha_mask"); + + lp_build_mask_update(mask, test); + } +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_alpha.h b/src/gallium/drivers/llvmpipe/lp_bld_alpha.h new file mode 100644 index 00000000000..9dbcdb4daab --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_alpha.h @@ -0,0 +1,54 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Alpha testing to LLVM IR translation. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#ifndef LP_BLD_ALPHA_H +#define LP_BLD_ALPHA_H + + +#include <llvm-c/Core.h> + +struct pipe_alpha_state; +union lp_type; +struct lp_build_mask_context; + + +void +lp_build_alpha_test(LLVMBuilderRef builder, + const struct pipe_alpha_state *state, + union lp_type type, + struct lp_build_mask_context *mask, + LLVMValueRef alpha, + LLVMValueRef ref); + + +#endif /* !LP_BLD_ALPHA_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_arit.c b/src/gallium/drivers/llvmpipe/lp_bld_arit.c new file mode 100644 index 00000000000..09a57ff33d5 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_arit.c @@ -0,0 +1,963 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * @file + * Helper + * + * LLVM IR doesn't support all basic arithmetic operations we care about (most + * notably min/max and saturated operations), and it is often necessary to + * resort machine-specific intrinsics directly. The functions here hide all + * these implementation details from the other modules. + * + * We also do simple expressions simplification here. Reasons are: + * - it is very easy given we have all necessary information readily available + * - LLVM optimization passes fail to simplify several vector expressions + * - We often know value constraints which the optimization passes have no way + * of knowing, such as when source arguments are known to be in [0, 1] range. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#include "util/u_memory.h" +#include "util/u_debug.h" +#include "util/u_string.h" + +#include "lp_bld_type.h" +#include "lp_bld_const.h" +#include "lp_bld_intr.h" +#include "lp_bld_logic.h" +#include "lp_bld_arit.h" + + +/** + * Generate min(a, b) + * No checks for special case values of a or b = 1 or 0 are done. + */ +static LLVMValueRef +lp_build_min_simple(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b) +{ + const union lp_type type = bld->type; + const char *intrinsic = NULL; + LLVMValueRef cond; + + /* TODO: optimize the constant case */ + +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) + if(type.width * type.length == 128) { + if(type.floating) { + if(type.width == 32) + intrinsic = "llvm.x86.sse.min.ps"; + if(type.width == 64) + intrinsic = "llvm.x86.sse2.min.pd"; + } + else { + if(type.width == 8 && !type.sign) + intrinsic = "llvm.x86.sse2.pminu.b"; + if(type.width == 8 && type.sign) + intrinsic = "llvm.x86.sse41.pminsb"; + if(type.width == 16 && !type.sign) + intrinsic = "llvm.x86.sse41.pminuw"; + if(type.width == 16 && type.sign) + intrinsic = "llvm.x86.sse2.pmins.w"; + if(type.width == 32 && !type.sign) + intrinsic = "llvm.x86.sse41.pminud"; + if(type.width == 32 && type.sign) + intrinsic = "llvm.x86.sse41.pminsd"; + } + } +#endif + + if(intrinsic) + return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b); + + cond = lp_build_cmp(bld, PIPE_FUNC_LESS, a, b); + return lp_build_select(bld, cond, a, b); +} + + +/** + * Generate max(a, b) + * No checks for special case values of a or b = 1 or 0 are done. + */ +static LLVMValueRef +lp_build_max_simple(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b) +{ + const union lp_type type = bld->type; + const char *intrinsic = NULL; + LLVMValueRef cond; + + /* TODO: optimize the constant case */ + +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) + if(type.width * type.length == 128) { + if(type.floating) { + if(type.width == 32) + intrinsic = "llvm.x86.sse.max.ps"; + if(type.width == 64) + intrinsic = "llvm.x86.sse2.max.pd"; + } + else { + if(type.width == 8 && !type.sign) + intrinsic = "llvm.x86.sse2.pmaxu.b"; + if(type.width == 8 && type.sign) + intrinsic = "llvm.x86.sse41.pmaxsb"; + if(type.width == 16 && !type.sign) + intrinsic = "llvm.x86.sse41.pmaxuw"; + if(type.width == 16 && type.sign) + intrinsic = "llvm.x86.sse2.pmaxs.w"; + if(type.width == 32 && !type.sign) + intrinsic = "llvm.x86.sse41.pmaxud"; + if(type.width == 32 && type.sign) + intrinsic = "llvm.x86.sse41.pmaxsd"; + } + } +#endif + + if(intrinsic) + return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b); + + cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, b); + return lp_build_select(bld, cond, a, b); +} + + +/** + * Generate 1 - a, or ~a depending on bld->type. + */ +LLVMValueRef +lp_build_comp(struct lp_build_context *bld, + LLVMValueRef a) +{ + const union lp_type type = bld->type; + + if(a == bld->one) + return bld->zero; + if(a == bld->zero) + return bld->one; + + if(type.norm && !type.floating && !type.fixed && !type.sign) { + if(LLVMIsConstant(a)) + return LLVMConstNot(a); + else + return LLVMBuildNot(bld->builder, a, ""); + } + + if(LLVMIsConstant(a)) + return LLVMConstSub(bld->one, a); + else + return LLVMBuildSub(bld->builder, bld->one, a, ""); +} + + +/** + * Generate a + b + */ +LLVMValueRef +lp_build_add(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b) +{ + const union lp_type type = bld->type; + LLVMValueRef res; + + if(a == bld->zero) + return b; + if(b == bld->zero) + return a; + if(a == bld->undef || b == bld->undef) + return bld->undef; + + if(bld->type.norm) { + const char *intrinsic = NULL; + + if(a == bld->one || b == bld->one) + return bld->one; + +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) + if(type.width * type.length == 128 && + !type.floating && !type.fixed) { + if(type.width == 8) + intrinsic = type.sign ? "llvm.x86.sse2.padds.b" : "llvm.x86.sse2.paddus.b"; + if(type.width == 16) + intrinsic = type.sign ? "llvm.x86.sse2.padds.w" : "llvm.x86.sse2.paddus.w"; + } +#endif + + if(intrinsic) + return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b); + } + + if(LLVMIsConstant(a) && LLVMIsConstant(b)) + res = LLVMConstAdd(a, b); + else + res = LLVMBuildAdd(bld->builder, a, b, ""); + + /* clamp to ceiling of 1.0 */ + if(bld->type.norm && (bld->type.floating || bld->type.fixed)) + res = lp_build_min_simple(bld, res, bld->one); + + /* XXX clamp to floor of -1 or 0??? */ + + return res; +} + + +/** + * Generate a - b + */ +LLVMValueRef +lp_build_sub(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b) +{ + const union lp_type type = bld->type; + LLVMValueRef res; + + if(b == bld->zero) + return a; + if(a == bld->undef || b == bld->undef) + return bld->undef; + if(a == b) + return bld->zero; + + if(bld->type.norm) { + const char *intrinsic = NULL; + + if(b == bld->one) + return bld->zero; + +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) + if(type.width * type.length == 128 && + !type.floating && !type.fixed) { + if(type.width == 8) + intrinsic = type.sign ? "llvm.x86.sse2.psubs.b" : "llvm.x86.sse2.psubus.b"; + if(type.width == 16) + intrinsic = type.sign ? "llvm.x86.sse2.psubs.w" : "llvm.x86.sse2.psubus.w"; + } +#endif + + if(intrinsic) + return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b); + } + + if(LLVMIsConstant(a) && LLVMIsConstant(b)) + res = LLVMConstSub(a, b); + else + res = LLVMBuildSub(bld->builder, a, b, ""); + + if(bld->type.norm && (bld->type.floating || bld->type.fixed)) + res = lp_build_max_simple(bld, res, bld->zero); + + return res; +} + + +/** + * Build shuffle vectors that match PUNPCKLxx and PUNPCKHxx instructions. + */ +static LLVMValueRef +lp_build_unpack_shuffle(unsigned n, unsigned lo_hi) +{ + LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; + unsigned i, j; + + assert(n <= LP_MAX_VECTOR_LENGTH); + assert(lo_hi < 2); + + for(i = 0, j = lo_hi*n/2; i < n; i += 2, ++j) { + elems[i + 0] = LLVMConstInt(LLVMInt32Type(), 0 + j, 0); + elems[i + 1] = LLVMConstInt(LLVMInt32Type(), n + j, 0); + } + + return LLVMConstVector(elems, n); +} + + +/** + * Build constant int vector of width 'n' and value 'c'. + */ +static LLVMValueRef +lp_build_const_vec(LLVMTypeRef type, unsigned n, long long c) +{ + LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; + unsigned i; + + assert(n <= LP_MAX_VECTOR_LENGTH); + + for(i = 0; i < n; ++i) + elems[i] = LLVMConstInt(type, c, 0); + + return LLVMConstVector(elems, n); +} + + +/** + * Normalized 8bit multiplication. + * + * - alpha plus one + * + * makes the following approximation to the division (Sree) + * + * a*b/255 ~= (a*(b + 1)) >> 256 + * + * which is the fastest method that satisfies the following OpenGL criteria + * + * 0*0 = 0 and 255*255 = 255 + * + * - geometric series + * + * takes the geometric series approximation to the division + * + * t/255 = (t >> 8) + (t >> 16) + (t >> 24) .. + * + * in this case just the first two terms to fit in 16bit arithmetic + * + * t/255 ~= (t + (t >> 8)) >> 8 + * + * note that just by itself it doesn't satisfies the OpenGL criteria, as + * 255*255 = 254, so the special case b = 255 must be accounted or roundoff + * must be used + * + * - geometric series plus rounding + * + * when using a geometric series division instead of truncating the result + * use roundoff in the approximation (Jim Blinn) + * + * t/255 ~= (t + (t >> 8) + 0x80) >> 8 + * + * achieving the exact results + * + * @sa Alvy Ray Smith, Image Compositing Fundamentals, Tech Memo 4, Aug 15, 1995, + * ftp://ftp.alvyray.com/Acrobat/4_Comp.pdf + * @sa Michael Herf, The "double blend trick", May 2000, + * http://www.stereopsis.com/doubleblend.html + */ +static LLVMValueRef +lp_build_mul_u8n(LLVMBuilderRef builder, + LLVMValueRef a, LLVMValueRef b) +{ + static LLVMValueRef c01 = NULL; + static LLVMValueRef c08 = NULL; + static LLVMValueRef c80 = NULL; + LLVMValueRef ab; + + if(!c01) c01 = lp_build_const_vec(LLVMInt16Type(), 8, 0x01); + if(!c08) c08 = lp_build_const_vec(LLVMInt16Type(), 8, 0x08); + if(!c80) c80 = lp_build_const_vec(LLVMInt16Type(), 8, 0x80); + +#if 0 + + /* a*b/255 ~= (a*(b + 1)) >> 256 */ + b = LLVMBuildAdd(builder, b, c01, ""); + ab = LLVMBuildMul(builder, a, b, ""); + +#else + + /* t/255 ~= (t + (t >> 8) + 0x80) >> 8 */ + ab = LLVMBuildMul(builder, a, b, ""); + ab = LLVMBuildAdd(builder, ab, LLVMBuildLShr(builder, ab, c08, ""), ""); + ab = LLVMBuildAdd(builder, ab, c80, ""); + +#endif + + ab = LLVMBuildLShr(builder, ab, c08, ""); + + return ab; +} + + +/** + * Generate a * b + */ +LLVMValueRef +lp_build_mul(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b) +{ + const union lp_type type = bld->type; + + if(a == bld->zero) + return bld->zero; + if(a == bld->one) + return b; + if(b == bld->zero) + return bld->zero; + if(b == bld->one) + return a; + if(a == bld->undef || b == bld->undef) + return bld->undef; + + if(!type.floating && !type.fixed && type.norm) { +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) + if(type.width == 8 && type.length == 16) { + LLVMTypeRef i16x8 = LLVMVectorType(LLVMInt16Type(), 8); + LLVMTypeRef i8x16 = LLVMVectorType(LLVMInt8Type(), 16); + static LLVMValueRef ml = NULL; + static LLVMValueRef mh = NULL; + LLVMValueRef al, ah, bl, bh; + LLVMValueRef abl, abh; + LLVMValueRef ab; + + if(!ml) ml = lp_build_unpack_shuffle(16, 0); + if(!mh) mh = lp_build_unpack_shuffle(16, 1); + + /* PUNPCKLBW, PUNPCKHBW */ + al = LLVMBuildShuffleVector(bld->builder, a, bld->zero, ml, ""); + bl = LLVMBuildShuffleVector(bld->builder, b, bld->zero, ml, ""); + ah = LLVMBuildShuffleVector(bld->builder, a, bld->zero, mh, ""); + bh = LLVMBuildShuffleVector(bld->builder, b, bld->zero, mh, ""); + + /* NOP */ + al = LLVMBuildBitCast(bld->builder, al, i16x8, ""); + bl = LLVMBuildBitCast(bld->builder, bl, i16x8, ""); + ah = LLVMBuildBitCast(bld->builder, ah, i16x8, ""); + bh = LLVMBuildBitCast(bld->builder, bh, i16x8, ""); + + /* PMULLW, PSRLW, PADDW */ + abl = lp_build_mul_u8n(bld->builder, al, bl); + abh = lp_build_mul_u8n(bld->builder, ah, bh); + + /* PACKUSWB */ + ab = lp_build_intrinsic_binary(bld->builder, "llvm.x86.sse2.packuswb.128" , i16x8, abl, abh); + + /* NOP */ + ab = LLVMBuildBitCast(bld->builder, ab, i8x16, ""); + + return ab; + } +#endif + + /* FIXME */ + assert(0); + } + + if(LLVMIsConstant(a) && LLVMIsConstant(b)) + return LLVMConstMul(a, b); + + return LLVMBuildMul(bld->builder, a, b, ""); +} + + +/** + * Generate a / b + */ +LLVMValueRef +lp_build_div(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b) +{ + const union lp_type type = bld->type; + + if(a == bld->zero) + return bld->zero; + if(a == bld->one) + return lp_build_rcp(bld, b); + if(b == bld->zero) + return bld->undef; + if(b == bld->one) + return a; + if(a == bld->undef || b == bld->undef) + return bld->undef; + + if(LLVMIsConstant(a) && LLVMIsConstant(b)) + return LLVMConstFDiv(a, b); + +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) + if(type.width == 32 && type.length == 4) + return lp_build_mul(bld, a, lp_build_rcp(bld, b)); +#endif + + return LLVMBuildFDiv(bld->builder, a, b, ""); +} + + +/** + * Generate min(a, b) + * Do checks for special cases. + */ +LLVMValueRef +lp_build_min(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b) +{ + if(a == bld->undef || b == bld->undef) + return bld->undef; + + if(a == b) + return a; + + if(bld->type.norm) { + if(a == bld->zero || b == bld->zero) + return bld->zero; + if(a == bld->one) + return b; + if(b == bld->one) + return a; + } + + return lp_build_min_simple(bld, a, b); +} + + +/** + * Generate max(a, b) + * Do checks for special cases. + */ +LLVMValueRef +lp_build_max(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b) +{ + if(a == bld->undef || b == bld->undef) + return bld->undef; + + if(a == b) + return a; + + if(bld->type.norm) { + if(a == bld->one || b == bld->one) + return bld->one; + if(a == bld->zero) + return b; + if(b == bld->zero) + return a; + } + + return lp_build_max_simple(bld, a, b); +} + + +/** + * Generate abs(a) + */ +LLVMValueRef +lp_build_abs(struct lp_build_context *bld, + LLVMValueRef a) +{ + const union lp_type type = bld->type; + + if(!type.sign) + return a; + + /* XXX: is this really necessary? */ +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) + if(!type.floating && type.width*type.length == 128) { + LLVMTypeRef vec_type = lp_build_vec_type(type); + if(type.width == 8) + return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.b.128", vec_type, a); + if(type.width == 16) + return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.w.128", vec_type, a); + if(type.width == 32) + return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.d.128", vec_type, a); + } +#endif + + return lp_build_max(bld, a, LLVMBuildNeg(bld->builder, a, "")); +} + + +LLVMValueRef +lp_build_sqrt(struct lp_build_context *bld, + LLVMValueRef a) +{ + const union lp_type type = bld->type; + LLVMTypeRef vec_type = lp_build_vec_type(type); + char intrinsic[32]; + + /* TODO: optimize the constant case */ + /* TODO: optimize the constant case */ + + assert(type.floating); + util_snprintf(intrinsic, sizeof intrinsic, "llvm.sqrt.v%uf%u", type.length, type.width); + + return lp_build_intrinsic_unary(bld->builder, intrinsic, vec_type, a); +} + + +LLVMValueRef +lp_build_rcp(struct lp_build_context *bld, + LLVMValueRef a) +{ + const union lp_type type = bld->type; + + if(a == bld->zero) + return bld->undef; + if(a == bld->one) + return bld->one; + if(a == bld->undef) + return bld->undef; + + assert(type.floating); + + if(LLVMIsConstant(a)) + return LLVMConstFDiv(bld->one, a); + + /* XXX: is this really necessary? */ +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) + if(type.width == 32 && type.length == 4) + return lp_build_intrinsic_unary(bld->builder, "llvm.x86.sse.rcp.ps", lp_build_vec_type(type), a); +#endif + + return LLVMBuildFDiv(bld->builder, bld->one, a, ""); +} + + +/** + * Generate 1/sqrt(a) + */ +LLVMValueRef +lp_build_rsqrt(struct lp_build_context *bld, + LLVMValueRef a) +{ + const union lp_type type = bld->type; + + assert(type.floating); + + /* XXX: is this really necessary? */ +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) + if(type.width == 32 && type.length == 4) + return lp_build_intrinsic_unary(bld->builder, "llvm.x86.sse.rsqrt.ps", lp_build_vec_type(type), a); +#endif + + return lp_build_rcp(bld, lp_build_sqrt(bld, a)); +} + + +/** + * Generate cos(a) + */ +LLVMValueRef +lp_build_cos(struct lp_build_context *bld, + LLVMValueRef a) +{ + const union lp_type type = bld->type; + LLVMTypeRef vec_type = lp_build_vec_type(type); + char intrinsic[32]; + + /* TODO: optimize the constant case */ + + assert(type.floating); + util_snprintf(intrinsic, sizeof intrinsic, "llvm.cos.v%uf%u", type.length, type.width); + + return lp_build_intrinsic_unary(bld->builder, intrinsic, vec_type, a); +} + + +/** + * Generate sin(a) + */ +LLVMValueRef +lp_build_sin(struct lp_build_context *bld, + LLVMValueRef a) +{ + const union lp_type type = bld->type; + LLVMTypeRef vec_type = lp_build_vec_type(type); + char intrinsic[32]; + + /* TODO: optimize the constant case */ + + assert(type.floating); + util_snprintf(intrinsic, sizeof intrinsic, "llvm.sin.v%uf%u", type.length, type.width); + + return lp_build_intrinsic_unary(bld->builder, intrinsic, vec_type, a); +} + + +/** + * Generate pow(x, y) + */ +LLVMValueRef +lp_build_pow(struct lp_build_context *bld, + LLVMValueRef x, + LLVMValueRef y) +{ + /* TODO: optimize the constant case */ + if(LLVMIsConstant(x) && LLVMIsConstant(y)) + debug_printf("%s: inefficient/imprecise constant arithmetic\n"); + + return lp_build_exp2(bld, lp_build_mul(bld, lp_build_log2(bld, x), y)); +} + + +/** + * Generate exp(x) + */ +LLVMValueRef +lp_build_exp(struct lp_build_context *bld, + LLVMValueRef x) +{ + /* log2(e) = 1/log(2) */ + LLVMValueRef log2e = lp_build_const_scalar(bld->type, 1.4426950408889634); + + return lp_build_mul(bld, log2e, lp_build_exp2(bld, x)); +} + + +/** + * Generate log(x) + */ +LLVMValueRef +lp_build_log(struct lp_build_context *bld, + LLVMValueRef x) +{ + /* log(2) */ + LLVMValueRef log2 = lp_build_const_scalar(bld->type, 1.4426950408889634); + + return lp_build_mul(bld, log2, lp_build_exp2(bld, x)); +} + + +#define EXP_POLY_DEGREE 3 +#define LOG_POLY_DEGREE 5 + + +/** + * Generate polynomial. + * Ex: x^2 * coeffs[0] + x * coeffs[1] + coeffs[2]. + */ +static LLVMValueRef +lp_build_polynomial(struct lp_build_context *bld, + LLVMValueRef x, + const double *coeffs, + unsigned num_coeffs) +{ + const union lp_type type = bld->type; + LLVMValueRef res = NULL; + unsigned i; + + /* TODO: optimize the constant case */ + if(LLVMIsConstant(x)) + debug_printf("%s: inefficient/imprecise constant arithmetic\n"); + + for (i = num_coeffs; i--; ) { + LLVMValueRef coeff = lp_build_const_scalar(type, coeffs[i]); + if(res) + res = lp_build_add(bld, coeff, lp_build_mul(bld, x, res)); + else + res = coeff; + } + + if(res) + return res; + else + return bld->undef; +} + + +/** + * Minimax polynomial fit of 2**x, in range [-0.5, 0.5[ + */ +const double lp_build_exp2_polynomial[] = { +#if EXP_POLY_DEGREE == 5 + 9.9999994e-1, 6.9315308e-1, 2.4015361e-1, 5.5826318e-2, 8.9893397e-3, 1.8775767e-3 +#elif EXP_POLY_DEGREE == 4 + 1.0000026, 6.9300383e-1, 2.4144275e-1, 5.2011464e-2, 1.3534167e-2 +#elif EXP_POLY_DEGREE == 3 + 9.9992520e-1, 6.9583356e-1, 2.2606716e-1, 7.8024521e-2 +#elif EXP_POLY_DEGREE == 2 + 1.0017247, 6.5763628e-1, 3.3718944e-1 +#else +#error +#endif +}; + + +void +lp_build_exp2_approx(struct lp_build_context *bld, + LLVMValueRef x, + LLVMValueRef *p_exp2_int_part, + LLVMValueRef *p_frac_part, + LLVMValueRef *p_exp2) +{ + const union lp_type type = bld->type; + LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); + LLVMValueRef ipart = NULL; + LLVMValueRef fpart = NULL; + LLVMValueRef expipart = NULL; + LLVMValueRef expfpart = NULL; + LLVMValueRef res = NULL; + + if(p_exp2_int_part || p_frac_part || p_exp2) { + /* TODO: optimize the constant case */ + if(LLVMIsConstant(x)) + debug_printf("%s: inefficient/imprecise constant arithmetic\n"); + + assert(type.floating && type.width == 32); + + x = lp_build_min(bld, x, lp_build_const_scalar(type, 129.0)); + x = lp_build_max(bld, x, lp_build_const_scalar(type, -126.99999)); + + /* ipart = int(x - 0.5) */ + ipart = LLVMBuildSub(bld->builder, x, lp_build_const_scalar(type, 0.5f), ""); + ipart = LLVMBuildFPToSI(bld->builder, ipart, int_vec_type, ""); + + /* fpart = x - ipart */ + fpart = LLVMBuildSIToFP(bld->builder, ipart, vec_type, ""); + fpart = LLVMBuildSub(bld->builder, x, fpart, ""); + } + + if(p_exp2_int_part || p_exp2) { + /* expipart = (float) (1 << ipart) */ + expipart = LLVMBuildAdd(bld->builder, ipart, lp_build_int_const_scalar(type, 127), ""); + expipart = LLVMBuildShl(bld->builder, expipart, lp_build_int_const_scalar(type, 23), ""); + expipart = LLVMBuildBitCast(bld->builder, expipart, vec_type, ""); + } + + if(p_exp2) { + expfpart = lp_build_polynomial(bld, fpart, lp_build_exp2_polynomial, + Elements(lp_build_exp2_polynomial)); + + res = LLVMBuildMul(bld->builder, expipart, expfpart, ""); + } + + if(p_exp2_int_part) + *p_exp2_int_part = expipart; + + if(p_frac_part) + *p_frac_part = fpart; + + if(p_exp2) + *p_exp2 = res; +} + + +LLVMValueRef +lp_build_exp2(struct lp_build_context *bld, + LLVMValueRef x) +{ + LLVMValueRef res; + lp_build_exp2_approx(bld, x, NULL, NULL, &res); + return res; +} + + +/** + * Minimax polynomial fit of log2(x)/(x - 1), for x in range [1, 2[ + * These coefficients can be generate with + * http://www.boost.org/doc/libs/1_36_0/libs/math/doc/sf_and_dist/html/math_toolkit/toolkit/internals2/minimax.html + */ +const double lp_build_log2_polynomial[] = { +#if LOG_POLY_DEGREE == 6 + 3.11578814719469302614, -3.32419399085241980044, 2.59883907202499966007, -1.23152682416275988241, 0.318212422185251071475, -0.0344359067839062357313 +#elif LOG_POLY_DEGREE == 5 + 2.8882704548164776201, -2.52074962577807006663, 1.48116647521213171641, -0.465725644288844778798, 0.0596515482674574969533 +#elif LOG_POLY_DEGREE == 4 + 2.61761038894603480148, -1.75647175389045657003, 0.688243882994381274313, -0.107254423828329604454 +#elif LOG_POLY_DEGREE == 3 + 2.28330284476918490682, -1.04913055217340124191, 0.204446009836232697516 +#else +#error +#endif +}; + + +/** + * See http://www.devmaster.net/forums/showthread.php?p=43580 + */ +void +lp_build_log2_approx(struct lp_build_context *bld, + LLVMValueRef x, + LLVMValueRef *p_exp, + LLVMValueRef *p_floor_log2, + LLVMValueRef *p_log2) +{ + const union lp_type type = bld->type; + LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); + + LLVMValueRef expmask = lp_build_int_const_scalar(type, 0x7f800000); + LLVMValueRef mantmask = lp_build_int_const_scalar(type, 0x007fffff); + LLVMValueRef one = LLVMConstBitCast(bld->one, int_vec_type); + + LLVMValueRef i = NULL; + LLVMValueRef exp = NULL; + LLVMValueRef mant = NULL; + LLVMValueRef logexp = NULL; + LLVMValueRef logmant = NULL; + LLVMValueRef res = NULL; + + if(p_exp || p_floor_log2 || p_log2) { + /* TODO: optimize the constant case */ + if(LLVMIsConstant(x)) + debug_printf("%s: inefficient/imprecise constant arithmetic\n"); + + assert(type.floating && type.width == 32); + + i = LLVMBuildBitCast(bld->builder, x, int_vec_type, ""); + + /* exp = (float) exponent(x) */ + exp = LLVMBuildAnd(bld->builder, i, expmask, ""); + } + + if(p_floor_log2 || p_log2) { + logexp = LLVMBuildLShr(bld->builder, exp, lp_build_int_const_scalar(type, 23), ""); + logexp = LLVMBuildSub(bld->builder, logexp, lp_build_int_const_scalar(type, 127), ""); + logexp = LLVMBuildSIToFP(bld->builder, logexp, vec_type, ""); + } + + if(p_log2) { + /* mant = (float) mantissa(x) */ + mant = LLVMBuildAnd(bld->builder, i, mantmask, ""); + mant = LLVMBuildOr(bld->builder, mant, one, ""); + mant = LLVMBuildSIToFP(bld->builder, mant, vec_type, ""); + + logmant = lp_build_polynomial(bld, mant, lp_build_log2_polynomial, + Elements(lp_build_log2_polynomial)); + + /* This effectively increases the polynomial degree by one, but ensures that log2(1) == 0*/ + logmant = LLVMBuildMul(bld->builder, logmant, LLVMBuildMul(bld->builder, mant, bld->one, ""), ""); + + res = LLVMBuildAdd(bld->builder, logmant, logexp, ""); + } + + if(p_exp) + *p_exp = exp; + + if(p_floor_log2) + *p_floor_log2 = logexp; + + if(p_log2) + *p_log2 = res; +} + + +LLVMValueRef +lp_build_log2(struct lp_build_context *bld, + LLVMValueRef x) +{ + LLVMValueRef res; + lp_build_log2_approx(bld, x, NULL, NULL, &res); + return res; +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_arit.h b/src/gallium/drivers/llvmpipe/lp_bld_arit.h new file mode 100644 index 00000000000..fc8cb25966e --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_arit.h @@ -0,0 +1,143 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Helper arithmetic functions. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#ifndef LP_BLD_ARIT_H +#define LP_BLD_ARIT_H + + +#include <llvm-c/Core.h> + + +union lp_type type; +struct lp_build_context; + + +/** + * Complement, i.e., 1 - a. + */ +LLVMValueRef +lp_build_comp(struct lp_build_context *bld, + LLVMValueRef a); + +LLVMValueRef +lp_build_add(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b); + +LLVMValueRef +lp_build_sub(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b); + +LLVMValueRef +lp_build_mul(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b); + +LLVMValueRef +lp_build_div(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b); + +LLVMValueRef +lp_build_min(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b); + +LLVMValueRef +lp_build_max(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b); + +LLVMValueRef +lp_build_abs(struct lp_build_context *bld, + LLVMValueRef a); + +LLVMValueRef +lp_build_sqrt(struct lp_build_context *bld, + LLVMValueRef a); + +LLVMValueRef +lp_build_rcp(struct lp_build_context *bld, + LLVMValueRef a); + +LLVMValueRef +lp_build_rsqrt(struct lp_build_context *bld, + LLVMValueRef a); + +LLVMValueRef +lp_build_cos(struct lp_build_context *bld, + LLVMValueRef a); + +LLVMValueRef +lp_build_sin(struct lp_build_context *bld, + LLVMValueRef a); + +LLVMValueRef +lp_build_pow(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b); + +LLVMValueRef +lp_build_exp(struct lp_build_context *bld, + LLVMValueRef a); + +LLVMValueRef +lp_build_log(struct lp_build_context *bld, + LLVMValueRef a); + +LLVMValueRef +lp_build_exp2(struct lp_build_context *bld, + LLVMValueRef a); + +LLVMValueRef +lp_build_log2(struct lp_build_context *bld, + LLVMValueRef a); + +void +lp_build_exp2_approx(struct lp_build_context *bld, + LLVMValueRef x, + LLVMValueRef *p_exp2_int_part, + LLVMValueRef *p_frac_part, + LLVMValueRef *p_exp2); + +void +lp_build_log2_approx(struct lp_build_context *bld, + LLVMValueRef x, + LLVMValueRef *p_exp, + LLVMValueRef *p_floor_log2, + LLVMValueRef *p_log2); + +#endif /* !LP_BLD_ARIT_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend.h b/src/gallium/drivers/llvmpipe/lp_bld_blend.h new file mode 100644 index 00000000000..d19e18846c2 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_blend.h @@ -0,0 +1,107 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef LP_BLD_BLEND_H +#define LP_BLD_BLEND_H + + +/** + * @file + * LLVM IR building helpers interfaces. + * + * We use LLVM-C bindings for now. They are not documented, but follow the C++ + * interfaces very closely, and appear to be complete enough for code + * genration. See + * http://npcontemplation.blogspot.com/2008/06/secret-of-llvm-c-bindings.html + * for a standalone example. + */ + +#include <llvm-c/Core.h> + +#include "pipe/p_format.h" + + +struct pipe_blend_state; +union lp_type; +struct lp_build_context; + + +/** + * Whether the blending function is commutative or not. + */ +boolean +lp_build_blend_func_commutative(unsigned func); + + +/** + * Whether the blending functions are the reverse of each other. + */ +boolean +lp_build_blend_func_reverse(unsigned rgb_func, unsigned alpha_func); + + +LLVMValueRef +lp_build_blend_func(struct lp_build_context *bld, + unsigned func, + LLVMValueRef term1, + LLVMValueRef term2); + + +LLVMValueRef +lp_build_blend_aos(LLVMBuilderRef builder, + const struct pipe_blend_state *blend, + union lp_type type, + LLVMValueRef src, + LLVMValueRef dst, + LLVMValueRef const_, + unsigned alpha_swizzle); + + +void +lp_build_blend_soa(LLVMBuilderRef builder, + const struct pipe_blend_state *blend, + union lp_type type, + LLVMValueRef src[4], + LLVMValueRef dst[4], + LLVMValueRef const_[4], + LLVMValueRef res[4]); + + +/** + * Apply a logic op. + * + * src/dst parameters are packed values. It should work regardless the inputs + * are scalars, or a vector. + */ +LLVMValueRef +lp_build_logicop(LLVMBuilderRef builder, + unsigned logicop_func, + LLVMValueRef src, + LLVMValueRef dst); + + +#endif /* !LP_BLD_BLEND_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c new file mode 100644 index 00000000000..c11a9398f87 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c @@ -0,0 +1,356 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * @file + * Blend LLVM IR generation -- AoS layout. + * + * AoS blending is in general much slower than SoA, but there are some cases + * where it might be faster. In particular, if a pixel is rendered only once + * then the overhead of tiling and untiling will dominate over the speedup that + * SoA gives. So we might want to detect such cases and fallback to AoS in the + * future, but for now this function is here for historical/benchmarking + * purposes. + * + * Run lp_blend_test after any change to this file. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#include "pipe/p_state.h" + +#include "lp_bld_type.h" +#include "lp_bld_const.h" +#include "lp_bld_arit.h" +#include "lp_bld_logic.h" +#include "lp_bld_swizzle.h" +#include "lp_bld_blend.h" +#include "lp_bld_debug.h" + + +/** + * We may the same values several times, so we keep them here to avoid + * recomputing them. Also reusing the values allows us to do simplifications + * that LLVM optimization passes wouldn't normally be able to do. + */ +struct lp_build_blend_aos_context +{ + struct lp_build_context base; + + LLVMValueRef src; + LLVMValueRef dst; + LLVMValueRef const_; + + LLVMValueRef inv_src; + LLVMValueRef inv_dst; + LLVMValueRef inv_const; + LLVMValueRef saturate; + + LLVMValueRef rgb_src_factor; + LLVMValueRef alpha_src_factor; + LLVMValueRef rgb_dst_factor; + LLVMValueRef alpha_dst_factor; +}; + + +static LLVMValueRef +lp_build_blend_factor_unswizzled(struct lp_build_blend_aos_context *bld, + unsigned factor, + boolean alpha) +{ + switch (factor) { + case PIPE_BLENDFACTOR_ZERO: + return bld->base.zero; + case PIPE_BLENDFACTOR_ONE: + return bld->base.one; + case PIPE_BLENDFACTOR_SRC_COLOR: + case PIPE_BLENDFACTOR_SRC_ALPHA: + return bld->src; + case PIPE_BLENDFACTOR_DST_COLOR: + case PIPE_BLENDFACTOR_DST_ALPHA: + return bld->dst; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + if(alpha) + return bld->base.one; + else { + if(!bld->inv_dst) + bld->inv_dst = lp_build_comp(&bld->base, bld->dst); + if(!bld->saturate) + bld->saturate = lp_build_min(&bld->base, bld->src, bld->inv_dst); + return bld->saturate; + } + case PIPE_BLENDFACTOR_CONST_COLOR: + case PIPE_BLENDFACTOR_CONST_ALPHA: + return bld->const_; + case PIPE_BLENDFACTOR_SRC1_COLOR: + case PIPE_BLENDFACTOR_SRC1_ALPHA: + /* TODO */ + assert(0); + return bld->base.zero; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + if(!bld->inv_src) + bld->inv_src = lp_build_comp(&bld->base, bld->src); + return bld->inv_src; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + if(!bld->inv_dst) + bld->inv_dst = lp_build_comp(&bld->base, bld->dst); + return bld->inv_dst; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + if(!bld->inv_const) + bld->inv_const = lp_build_comp(&bld->base, bld->const_); + return bld->inv_const; + case PIPE_BLENDFACTOR_INV_SRC1_COLOR: + case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: + /* TODO */ + assert(0); + return bld->base.zero; + default: + assert(0); + return bld->base.zero; + } +} + + +enum lp_build_blend_swizzle { + LP_BUILD_BLEND_SWIZZLE_RGBA = 0, + LP_BUILD_BLEND_SWIZZLE_AAAA = 1, +}; + + +/** + * How should we shuffle the base factor. + */ +static enum lp_build_blend_swizzle +lp_build_blend_factor_swizzle(unsigned factor) +{ + switch (factor) { + case PIPE_BLENDFACTOR_ONE: + case PIPE_BLENDFACTOR_ZERO: + case PIPE_BLENDFACTOR_SRC_COLOR: + case PIPE_BLENDFACTOR_DST_COLOR: + case PIPE_BLENDFACTOR_CONST_COLOR: + case PIPE_BLENDFACTOR_SRC1_COLOR: + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + case PIPE_BLENDFACTOR_INV_DST_COLOR: + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + case PIPE_BLENDFACTOR_INV_SRC1_COLOR: + return LP_BUILD_BLEND_SWIZZLE_RGBA; + case PIPE_BLENDFACTOR_SRC_ALPHA: + case PIPE_BLENDFACTOR_DST_ALPHA: + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + case PIPE_BLENDFACTOR_SRC1_ALPHA: + case PIPE_BLENDFACTOR_CONST_ALPHA: + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: + return LP_BUILD_BLEND_SWIZZLE_AAAA; + default: + assert(0); + return LP_BUILD_BLEND_SWIZZLE_RGBA; + } +} + + +static LLVMValueRef +lp_build_blend_swizzle(struct lp_build_blend_aos_context *bld, + LLVMValueRef rgb, + LLVMValueRef alpha, + enum lp_build_blend_swizzle rgb_swizzle, + unsigned alpha_swizzle) +{ + if(rgb == alpha) { + if(rgb_swizzle == LP_BUILD_BLEND_SWIZZLE_RGBA) + return rgb; + if(rgb_swizzle == LP_BUILD_BLEND_SWIZZLE_AAAA) + return lp_build_broadcast_aos(&bld->base, rgb, alpha_swizzle); + } + else { + if(rgb_swizzle == LP_BUILD_BLEND_SWIZZLE_RGBA) { + boolean cond[4] = {0, 0, 0, 0}; + cond[alpha_swizzle] = 1; + return lp_build_select_aos(&bld->base, alpha, rgb, cond); + } + if(rgb_swizzle == LP_BUILD_BLEND_SWIZZLE_AAAA) { + unsigned char swizzle[4]; + swizzle[0] = alpha_swizzle; + swizzle[1] = alpha_swizzle; + swizzle[2] = alpha_swizzle; + swizzle[3] = alpha_swizzle; + swizzle[alpha_swizzle] += 4; + return lp_build_swizzle2_aos(&bld->base, rgb, alpha, swizzle); + } + } + assert(0); + return bld->base.undef; +} + + +/** + * @sa http://www.opengl.org/sdk/docs/man/xhtml/glBlendFuncSeparate.xml + */ +static LLVMValueRef +lp_build_blend_factor(struct lp_build_blend_aos_context *bld, + LLVMValueRef factor1, + unsigned rgb_factor, + unsigned alpha_factor, + unsigned alpha_swizzle) +{ + LLVMValueRef rgb_factor_; + LLVMValueRef alpha_factor_; + LLVMValueRef factor2; + enum lp_build_blend_swizzle rgb_swizzle; + + rgb_factor_ = lp_build_blend_factor_unswizzled(bld, rgb_factor, FALSE); + alpha_factor_ = lp_build_blend_factor_unswizzled(bld, alpha_factor, TRUE); + + rgb_swizzle = lp_build_blend_factor_swizzle(rgb_factor); + + factor2 = lp_build_blend_swizzle(bld, rgb_factor_, alpha_factor_, rgb_swizzle, alpha_swizzle); + + return lp_build_mul(&bld->base, factor1, factor2); +} + + +boolean +lp_build_blend_func_commutative(unsigned func) +{ + switch (func) { + case PIPE_BLEND_ADD: + case PIPE_BLEND_MIN: + case PIPE_BLEND_MAX: + return TRUE; + case PIPE_BLEND_SUBTRACT: + case PIPE_BLEND_REVERSE_SUBTRACT: + return FALSE; + default: + assert(0); + return TRUE; + } +} + + +boolean +lp_build_blend_func_reverse(unsigned rgb_func, unsigned alpha_func) +{ + if(rgb_func == alpha_func) + return FALSE; + if(rgb_func == PIPE_BLEND_SUBTRACT && alpha_func == PIPE_BLEND_REVERSE_SUBTRACT) + return TRUE; + if(rgb_func == PIPE_BLEND_REVERSE_SUBTRACT && alpha_func == PIPE_BLEND_SUBTRACT) + return TRUE; + return FALSE; +} + + +/** + * @sa http://www.opengl.org/sdk/docs/man/xhtml/glBlendEquationSeparate.xml + */ +LLVMValueRef +lp_build_blend_func(struct lp_build_context *bld, + unsigned func, + LLVMValueRef term1, + LLVMValueRef term2) +{ + switch (func) { + case PIPE_BLEND_ADD: + return lp_build_add(bld, term1, term2); + break; + case PIPE_BLEND_SUBTRACT: + return lp_build_sub(bld, term1, term2); + case PIPE_BLEND_REVERSE_SUBTRACT: + return lp_build_sub(bld, term2, term1); + case PIPE_BLEND_MIN: + return lp_build_min(bld, term1, term2); + case PIPE_BLEND_MAX: + return lp_build_max(bld, term1, term2); + default: + assert(0); + return bld->zero; + } +} + + +LLVMValueRef +lp_build_blend_aos(LLVMBuilderRef builder, + const struct pipe_blend_state *blend, + union lp_type type, + LLVMValueRef src, + LLVMValueRef dst, + LLVMValueRef const_, + unsigned alpha_swizzle) +{ + struct lp_build_blend_aos_context bld; + LLVMValueRef src_term; + LLVMValueRef dst_term; + + /* FIXME */ + assert(blend->colormask == 0xf); + + if(!blend->blend_enable) + return src; + + /* It makes no sense to blend unless values are normalized */ + assert(type.norm); + + /* Setup build context */ + memset(&bld, 0, sizeof bld); + lp_build_context_init(&bld.base, builder, type); + bld.src = src; + bld.dst = dst; + bld.const_ = const_; + + /* TODO: There are still a few optimization opportunities here. For certain + * combinations it is possible to reorder the operations and therefore saving + * some instructions. */ + + src_term = lp_build_blend_factor(&bld, src, blend->rgb_src_factor, blend->alpha_src_factor, alpha_swizzle); + dst_term = lp_build_blend_factor(&bld, dst, blend->rgb_dst_factor, blend->alpha_dst_factor, alpha_swizzle); + + lp_build_name(src_term, "src_term"); + lp_build_name(dst_term, "dst_term"); + + if(blend->rgb_func == blend->alpha_func) { + return lp_build_blend_func(&bld.base, blend->rgb_func, src_term, dst_term); + } + else { + /* Seperate RGB / A functions */ + + LLVMValueRef rgb; + LLVMValueRef alpha; + + rgb = lp_build_blend_func(&bld.base, blend->rgb_func, src_term, dst_term); + alpha = lp_build_blend_func(&bld.base, blend->alpha_func, src_term, dst_term); + + return lp_build_blend_swizzle(&bld, rgb, alpha, LP_BUILD_BLEND_SWIZZLE_RGBA, alpha_swizzle); + } +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_logicop.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_logicop.c new file mode 100644 index 00000000000..88321f62a2c --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_logicop.c @@ -0,0 +1,108 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * @file + * Blend LLVM IR generation -- logic ops. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#include "pipe/p_state.h" + +#include "lp_bld_blend.h" + + +LLVMValueRef +lp_build_logicop(LLVMBuilderRef builder, + unsigned logicop_func, + LLVMValueRef src, + LLVMValueRef dst) +{ + LLVMTypeRef type; + LLVMValueRef res; + + type = LLVMTypeOf(src); + + switch (logicop_func) { + case PIPE_LOGICOP_CLEAR: + res = LLVMConstNull(type); + break; + case PIPE_LOGICOP_NOR: + res = LLVMBuildNot(builder, LLVMBuildOr(builder, src, dst, ""), ""); + break; + case PIPE_LOGICOP_AND_INVERTED: + res = LLVMBuildAnd(builder, LLVMBuildNot(builder, src, ""), dst, ""); + break; + case PIPE_LOGICOP_COPY_INVERTED: + res = LLVMBuildNot(builder, src, ""); + break; + case PIPE_LOGICOP_AND_REVERSE: + res = LLVMBuildAnd(builder, src, LLVMBuildNot(builder, dst, ""), ""); + break; + case PIPE_LOGICOP_INVERT: + res = LLVMBuildNot(builder, dst, ""); + break; + case PIPE_LOGICOP_XOR: + res = LLVMBuildXor(builder, src, dst, ""); + break; + case PIPE_LOGICOP_NAND: + res = LLVMBuildNot(builder, LLVMBuildAnd(builder, src, dst, ""), ""); + break; + case PIPE_LOGICOP_AND: + res = LLVMBuildAnd(builder, src, dst, ""); + break; + case PIPE_LOGICOP_EQUIV: + res = LLVMBuildNot(builder, LLVMBuildXor(builder, src, dst, ""), ""); + break; + case PIPE_LOGICOP_NOOP: + res = dst; + break; + case PIPE_LOGICOP_OR_INVERTED: + res = LLVMBuildOr(builder, LLVMBuildNot(builder, src, ""), dst, ""); + break; + case PIPE_LOGICOP_COPY: + res = src; + break; + case PIPE_LOGICOP_OR_REVERSE: + res = LLVMBuildOr(builder, src, LLVMBuildNot(builder, dst, ""), ""); + break; + case PIPE_LOGICOP_OR: + res = LLVMBuildOr(builder, src, dst, ""); + break; + case PIPE_LOGICOP_SET: + res = LLVMConstAllOnes(type); + break; + default: + assert(0); + res = src; + } + + return res; +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c new file mode 100644 index 00000000000..b92254a7d6f --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c @@ -0,0 +1,298 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * @file + * Blend LLVM IR generation -- SoA layout. + * + * Blending in SoA is much faster than AoS, especially when separate rgb/alpha + * factors/functions are used, since no channel masking/shuffling is necessary + * and we can achieve the full throughput of the SIMD operations. Furthermore + * the fragment shader output is also in SoA, so it fits nicely with the rest of + * the fragment pipeline. + * + * The drawback is that to be displayed the color buffer needs to be in AoS + * layout, so we need to tile/untile the color buffer before/after rendering. + * A color buffer like + * + * R11 G11 B11 A11 R12 G12 B12 A12 R13 G13 B13 A13 R14 G14 B14 A14 ... + * R21 G21 B21 A21 R22 G22 B22 A22 R23 G23 B23 A23 R24 G24 B24 A24 ... + * + * R31 G31 B31 A31 R32 G32 B32 A32 R33 G33 B33 A33 R34 G34 B34 A34 ... + * R41 G41 B41 A41 R42 G42 B42 A42 R43 G43 B43 A43 R44 G44 B44 A44 ... + * + * ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... + * + * will actually be stored in memory as + * + * R11 R12 R21 R22 R13 R14 R23 R24 ... G11 G12 G21 G22 G13 G14 G23 G24 ... B11 B12 B21 B22 B13 B14 B23 B24 ... A11 A12 A21 A22 A13 A14 A23 A24 ... + * R31 R32 R41 R42 R33 R34 R43 R44 ... G31 G32 G41 G42 G33 G34 G43 G44 ... B31 B32 B41 B42 B33 B34 B43 B44 ... A31 A32 A41 A42 A33 A34 A43 A44 ... + * ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... + * + * NOTE: Run lp_blend_test after any change to this file. + * + * You can also run lp_blend_test to obtain AoS vs SoA benchmarks. Invoking it + * as: + * + * lp_blend_test -o blend.tsv + * + * will generate a tab-seperated-file with the test results and performance + * measurements. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#include "pipe/p_state.h" + +#include "lp_bld_type.h" +#include "lp_bld_const.h" +#include "lp_bld_arit.h" +#include "lp_bld_blend.h" + + +/** + * We may the same values several times, so we keep them here to avoid + * recomputing them. Also reusing the values allows us to do simplifications + * that LLVM optimization passes wouldn't normally be able to do. + */ +struct lp_build_blend_soa_context +{ + struct lp_build_context base; + + LLVMValueRef src[4]; + LLVMValueRef dst[4]; + LLVMValueRef con[4]; + + LLVMValueRef inv_src[4]; + LLVMValueRef inv_dst[4]; + LLVMValueRef inv_con[4]; + + LLVMValueRef src_alpha_saturate; + + /** + * We store all factors in a table in order to eliminate redundant + * multiplications later. + */ + LLVMValueRef factor[2][2][4]; + + /** + * Table with all terms. + */ + LLVMValueRef term[2][4]; +}; + + +static LLVMValueRef +lp_build_blend_soa_factor(struct lp_build_blend_soa_context *bld, + unsigned factor, unsigned i) +{ + /* + * Compute src/first term RGB + */ + switch (factor) { + case PIPE_BLENDFACTOR_ONE: + return bld->base.one; + case PIPE_BLENDFACTOR_SRC_COLOR: + return bld->src[i]; + case PIPE_BLENDFACTOR_SRC_ALPHA: + return bld->src[3]; + case PIPE_BLENDFACTOR_DST_COLOR: + return bld->dst[i]; + case PIPE_BLENDFACTOR_DST_ALPHA: + return bld->dst[3]; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + if(i == 3) + return bld->base.one; + else { + if(!bld->inv_dst[3]) + bld->inv_dst[3] = lp_build_comp(&bld->base, bld->dst[3]); + if(!bld->src_alpha_saturate) + bld->src_alpha_saturate = lp_build_min(&bld->base, bld->src[3], bld->inv_dst[3]); + return bld->src_alpha_saturate; + } + case PIPE_BLENDFACTOR_CONST_COLOR: + return bld->con[i]; + case PIPE_BLENDFACTOR_CONST_ALPHA: + return bld->con[3]; + case PIPE_BLENDFACTOR_SRC1_COLOR: + /* TODO */ + assert(0); + return bld->base.zero; + case PIPE_BLENDFACTOR_SRC1_ALPHA: + /* TODO */ + assert(0); + return bld->base.zero; + case PIPE_BLENDFACTOR_ZERO: + return bld->base.zero; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + if(!bld->inv_src[i]) + bld->inv_src[i] = lp_build_comp(&bld->base, bld->src[i]); + return bld->inv_src[i]; + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + if(!bld->inv_src[3]) + bld->inv_src[3] = lp_build_comp(&bld->base, bld->src[3]); + return bld->inv_src[3]; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + if(!bld->inv_dst[i]) + bld->inv_dst[i] = lp_build_comp(&bld->base, bld->dst[i]); + return bld->inv_dst[i]; + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + if(!bld->inv_dst[3]) + bld->inv_dst[3] = lp_build_comp(&bld->base, bld->dst[3]); + return bld->inv_dst[3]; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + if(!bld->inv_con[i]) + bld->inv_con[i] = lp_build_comp(&bld->base, bld->con[i]); + return bld->inv_con[i]; + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + if(!bld->inv_con[3]) + bld->inv_con[3] = lp_build_comp(&bld->base, bld->con[3]); + return bld->inv_con[3]; + case PIPE_BLENDFACTOR_INV_SRC1_COLOR: + /* TODO */ + assert(0); + return bld->base.zero; + case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: + /* TODO */ + assert(0); + return bld->base.zero; + default: + assert(0); + return bld->base.zero; + } +} + + +/** + * Generate blend code in SOA mode. + * \param src src/fragment color + * \param dst dst/framebuffer color + * \param con constant blend color + * \param res the result/output + */ +void +lp_build_blend_soa(LLVMBuilderRef builder, + const struct pipe_blend_state *blend, + union lp_type type, + LLVMValueRef src[4], + LLVMValueRef dst[4], + LLVMValueRef con[4], + LLVMValueRef res[4]) +{ + struct lp_build_blend_soa_context bld; + unsigned i, j, k; + + /* Setup build context */ + memset(&bld, 0, sizeof bld); + lp_build_context_init(&bld.base, builder, type); + for (i = 0; i < 4; ++i) { + bld.src[i] = src[i]; + bld.dst[i] = dst[i]; + bld.con[i] = con[i]; + } + + for (i = 0; i < 4; ++i) { + if (blend->colormask & (1 << i)) { + if (blend->logicop_enable) { + if(!type.floating) { + res[i] = lp_build_logicop(builder, blend->logicop_func, src[i], dst[i]); + } + else + res[i] = dst[i]; + } + else if (blend->blend_enable) { + unsigned src_factor = i < 3 ? blend->rgb_src_factor : blend->alpha_src_factor; + unsigned dst_factor = i < 3 ? blend->rgb_dst_factor : blend->alpha_dst_factor; + unsigned func = i < 3 ? blend->rgb_func : blend->alpha_func; + boolean func_commutative = lp_build_blend_func_commutative(func); + + /* It makes no sense to blend unless values are normalized */ + assert(type.norm); + + /* + * Compute src/dst factors. + */ + + bld.factor[0][0][i] = src[i]; + bld.factor[0][1][i] = lp_build_blend_soa_factor(&bld, src_factor, i); + bld.factor[1][0][i] = dst[i]; + bld.factor[1][1][i] = lp_build_blend_soa_factor(&bld, dst_factor, i); + + /* + * Compute src/dst terms + */ + + for(k = 0; k < 2; ++k) { + /* See if this multiplication has been previously computed */ + for(j = 0; j < i; ++j) { + if((bld.factor[k][0][j] == bld.factor[k][0][i] && + bld.factor[k][1][j] == bld.factor[k][1][i]) || + (bld.factor[k][0][j] == bld.factor[k][1][i] && + bld.factor[k][1][j] == bld.factor[k][0][i])) + break; + } + + if(j < i) + bld.term[k][i] = bld.term[k][j]; + else + bld.term[k][i] = lp_build_mul(&bld.base, bld.factor[k][0][i], bld.factor[k][1][i]); + } + + /* + * Combine terms + */ + + /* See if this function has been previously applied */ + for(j = 0; j < i; ++j) { + unsigned prev_func = j < 3 ? blend->rgb_func : blend->alpha_func; + unsigned func_reverse = lp_build_blend_func_reverse(func, prev_func); + + if((!func_reverse && + bld.term[0][j] == bld.term[0][i] && + bld.term[1][j] == bld.term[1][i]) || + ((func_commutative || func_reverse) && + bld.term[0][j] == bld.term[1][i] && + bld.term[1][j] == bld.term[0][i])) + break; + } + + if(j < i) + res[i] = res[j]; + else + res[i] = lp_build_blend_func(&bld.base, func, bld.term[0][i], bld.term[1][i]); + } + else { + res[i] = src[i]; + } + } + else { + res[i] = dst[i]; + } + } +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_const.c b/src/gallium/drivers/llvmpipe/lp_bld_const.c new file mode 100644 index 00000000000..21487365eae --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_const.c @@ -0,0 +1,369 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * @file + * Helper functions for constant building. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#include <float.h> + +#include "util/u_debug.h" + +#include "lp_bld_type.h" +#include "lp_bld_const.h" + + +unsigned +lp_mantissa(union lp_type type) +{ + assert(type.floating); + + if(type.floating) { + switch(type.width) { + case 32: + return 23; + case 64: + return 53; + default: + assert(0); + return 0; + } + } + else { + if(type.sign) + return type.width - 1; + else + return type.width; + } +} + + +/** + * Shift of the unity. + * + * Same as lp_const_scale(), but in terms of shifts. + */ +unsigned +lp_const_shift(union lp_type type) +{ + if(type.floating) + return 0; + else if(type.fixed) + return type.width/2; + else if(type.norm) + return type.sign ? type.width - 1 : type.width; + else + return 0; +} + + +unsigned +lp_const_offset(union lp_type type) +{ + if(type.floating || type.fixed) + return 0; + else if(type.norm) + return 1; + else + return 0; +} + + +/** + * Scaling factor between the LLVM native value and its interpretation. + * + * This is 1.0 for all floating types and unnormalized integers, and something + * else for the fixed points types and normalized integers. + */ +double +lp_const_scale(union lp_type type) +{ + unsigned long long llscale; + double dscale; + + llscale = (unsigned long long)1 << lp_const_shift(type); + llscale -= lp_const_offset(type); + dscale = (double)llscale; + assert((unsigned long long)dscale == llscale); + + return dscale; +} + + +/** + * Minimum value representable by the type. + */ +double +lp_const_min(union lp_type type) +{ + unsigned bits; + + if(!type.sign) + return 0.0; + + if(type.norm) + return -1.0; + + if (type.floating) { + switch(type.width) { + case 32: + return -FLT_MAX; + case 64: + return -DBL_MAX; + default: + assert(0); + return 0.0; + } + } + + if(type.fixed) + /* FIXME: consider the fractional bits? */ + bits = type.width / 2 - 1; + else + bits = type.width - 1; + + return (double)-((long long)1 << bits); +} + + +/** + * Maximum value representable by the type. + */ +double +lp_const_max(union lp_type type) +{ + unsigned bits; + + if(type.norm) + return 1.0; + + if (type.floating) { + switch(type.width) { + case 32: + return FLT_MAX; + case 64: + return DBL_MAX; + default: + assert(0); + return 0.0; + } + } + + if(type.fixed) + bits = type.width / 2; + else + bits = type.width; + + if(type.sign) + bits -= 1; + + return (double)(((unsigned long long)1 << bits) - 1); +} + + +double +lp_const_eps(union lp_type type) +{ + if (type.floating) { + switch(type.width) { + case 32: + return FLT_EPSILON; + case 64: + return DBL_EPSILON; + default: + assert(0); + return 0.0; + } + } + else { + double scale = lp_const_scale(type); + return 1.0/scale; + } +} + + +LLVMValueRef +lp_build_undef(union lp_type type) +{ + LLVMTypeRef vec_type = lp_build_vec_type(type); + return LLVMGetUndef(vec_type); +} + + +LLVMValueRef +lp_build_zero(union lp_type type) +{ + LLVMTypeRef vec_type = lp_build_vec_type(type); + return LLVMConstNull(vec_type); +} + + +LLVMValueRef +lp_build_one(union lp_type type) +{ + LLVMTypeRef elem_type; + LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; + unsigned i; + + assert(type.length <= LP_MAX_VECTOR_LENGTH); + + elem_type = lp_build_elem_type(type); + + if(type.floating) + elems[0] = LLVMConstReal(elem_type, 1.0); + else if(type.fixed) + elems[0] = LLVMConstInt(elem_type, 1LL << (type.width/2), 0); + else if(!type.norm) + elems[0] = LLVMConstInt(elem_type, 1, 0); + else if(type.sign) + elems[0] = LLVMConstInt(elem_type, (1LL << (type.width - 1)) - 1, 0); + else { + /* special case' -- 1.0 for normalized types is more easily attained if + * we start with a vector consisting of all bits set */ + LLVMTypeRef vec_type = LLVMVectorType(elem_type, type.length); + LLVMValueRef vec = LLVMConstAllOnes(vec_type); + +#if 0 + if(type.sign) + /* TODO: Unfortunately this caused "Tried to create a shift operation + * on a non-integer type!" */ + vec = LLVMConstLShr(vec, lp_build_int_const_scalar(type, 1)); +#endif + + return vec; + } + + for(i = 1; i < type.length; ++i) + elems[i] = elems[0]; + + return LLVMConstVector(elems, type.length); +} + + +LLVMValueRef +lp_build_const_scalar(union lp_type type, + double val) +{ + LLVMTypeRef elem_type = lp_build_elem_type(type); + LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; + unsigned i; + + assert(type.length <= LP_MAX_VECTOR_LENGTH); + + if(type.floating) { + elems[0] = LLVMConstReal(elem_type, val); + } + else { + double dscale = lp_const_scale(type); + + elems[0] = LLVMConstInt(elem_type, val*dscale + 0.5, 0); + } + + for(i = 1; i < type.length; ++i) + elems[i] = elems[0]; + + return LLVMConstVector(elems, type.length); +} + + +LLVMValueRef +lp_build_int_const_scalar(union lp_type type, + long long val) +{ + LLVMTypeRef elem_type = lp_build_int_elem_type(type); + LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; + unsigned i; + + assert(type.length <= LP_MAX_VECTOR_LENGTH); + + for(i = 0; i < type.length; ++i) + elems[i] = LLVMConstInt(elem_type, val, type.sign ? 1 : 0); + + return LLVMConstVector(elems, type.length); +} + + +LLVMValueRef +lp_build_const_aos(union lp_type type, + double r, double g, double b, double a, + const unsigned char *swizzle) +{ + const unsigned char default_swizzle[4] = {0, 1, 2, 3}; + LLVMTypeRef elem_type; + LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; + unsigned i; + + assert(type.length % 4 == 0); + assert(type.length <= LP_MAX_VECTOR_LENGTH); + + elem_type = lp_build_elem_type(type); + + if(swizzle == NULL) + swizzle = default_swizzle; + + if(type.floating) { + elems[swizzle[0]] = LLVMConstReal(elem_type, r); + elems[swizzle[1]] = LLVMConstReal(elem_type, g); + elems[swizzle[2]] = LLVMConstReal(elem_type, b); + elems[swizzle[3]] = LLVMConstReal(elem_type, a); + } + else { + double dscale = lp_const_scale(type); + + elems[swizzle[0]] = LLVMConstInt(elem_type, r*dscale + 0.5, 0); + elems[swizzle[1]] = LLVMConstInt(elem_type, g*dscale + 0.5, 0); + elems[swizzle[2]] = LLVMConstInt(elem_type, b*dscale + 0.5, 0); + elems[swizzle[3]] = LLVMConstInt(elem_type, a*dscale + 0.5, 0); + } + + for(i = 4; i < type.length; ++i) + elems[i] = elems[i % 4]; + + return LLVMConstVector(elems, type.length); +} + + +LLVMValueRef +lp_build_const_mask_aos(union lp_type type, + boolean cond[4]) +{ + LLVMTypeRef elem_type = LLVMIntType(type.width); + LLVMValueRef masks[LP_MAX_VECTOR_LENGTH]; + unsigned i, j; + + assert(type.length <= LP_MAX_VECTOR_LENGTH); + + for(j = 0; j < type.length; j += 4) + for(i = 0; i < 4; ++i) + masks[j + i] = LLVMConstInt(elem_type, cond[i] ? ~0 : 0, 0); + + return LLVMConstVector(masks, type.length); +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_const.h b/src/gallium/drivers/llvmpipe/lp_bld_const.h new file mode 100644 index 00000000000..1934530ea3c --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_const.h @@ -0,0 +1,108 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Helper functions for constant building. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#ifndef LP_BLD_CONST_H +#define LP_BLD_CONST_H + + +#include <llvm-c/Core.h> + +#include <pipe/p_compiler.h> + + +union lp_type type; + + +unsigned +lp_mantissa(union lp_type type); + + +unsigned +lp_const_shift(union lp_type type); + + +unsigned +lp_const_offset(union lp_type type); + + +double +lp_const_scale(union lp_type type); + +double +lp_const_min(union lp_type type); + + +double +lp_const_max(union lp_type type); + + +double +lp_const_eps(union lp_type type); + + +LLVMValueRef +lp_build_undef(union lp_type type); + + +LLVMValueRef +lp_build_zero(union lp_type type); + + +LLVMValueRef +lp_build_one(union lp_type type); + + +LLVMValueRef +lp_build_const_scalar(union lp_type type, + double val); + + +LLVMValueRef +lp_build_int_const_scalar(union lp_type type, + long long val); + + +LLVMValueRef +lp_build_const_aos(union lp_type type, + double r, double g, double b, double a, + const unsigned char *swizzle); + + +LLVMValueRef +lp_build_const_mask_aos(union lp_type type, + boolean cond[4]); + + +#endif /* !LP_BLD_CONST_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_conv.c b/src/gallium/drivers/llvmpipe/lp_bld_conv.c new file mode 100644 index 00000000000..c8954c8a34f --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_conv.c @@ -0,0 +1,702 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * @file + * Helper functions for type conversions. + * + * We want to use the fastest type for a given computation whenever feasible. + * The other side of this is that we need to be able convert between several + * types accurately and efficiently. + * + * Conversion between types of different bit width is quite complex since a + * + * To remember there are a few invariants in type conversions: + * + * - register width must remain constant: + * + * src_type.width * src_type.length == dst_type.width * dst_type.length + * + * - total number of elements must remain constant: + * + * src_type.length * num_srcs == dst_type.length * num_dsts + * + * It is not always possible to do the conversion both accurately and + * efficiently, usually due to lack of adequate machine instructions. In these + * cases it is important not to cut shortcuts here and sacrifice accuracy, as + * there this functions can be used anywhere. In the future we might have a + * precision parameter which can gauge the accuracy vs efficiency compromise, + * but for now if the data conversion between two stages happens to be the + * bottleneck, then most likely should just avoid converting at all and run + * both stages with the same type. + * + * Make sure to run lp_test_conv unit test after any change to this file. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#include "util/u_debug.h" +#include "util/u_math.h" + +#include "lp_bld_type.h" +#include "lp_bld_const.h" +#include "lp_bld_intr.h" +#include "lp_bld_arit.h" +#include "lp_bld_conv.h" + + +/** + * Special case for converting clamped IEEE-754 floats to unsigned norms. + * + * The mathematical voodoo below may seem excessive but it is actually + * paramount we do it this way for several reasons. First, there is no single + * precision FP to unsigned integer conversion Intel SSE instruction. Second, + * secondly, even if there was, since the FP's mantissa takes only a fraction + * of register bits the typically scale and cast approach would require double + * precision for accurate results, and therefore half the throughput + * + * Although the result values can be scaled to an arbitrary bit width specified + * by dst_width, the actual result type will have the same width. + */ +LLVMValueRef +lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, + union lp_type src_type, + unsigned dst_width, + LLVMValueRef src) +{ + LLVMTypeRef int_vec_type = lp_build_int_vec_type(src_type); + LLVMValueRef res; + unsigned mantissa; + unsigned n; + unsigned long long ubound; + unsigned long long mask; + double scale; + double bias; + + assert(src_type.floating); + + mantissa = lp_mantissa(src_type); + + /* We cannot carry more bits than the mantissa */ + n = MIN2(mantissa, dst_width); + + /* This magic coefficients will make the desired result to appear in the + * lowest significant bits of the mantissa. + */ + ubound = ((unsigned long long)1 << n); + mask = ubound - 1; + scale = (double)mask/ubound; + bias = (double)((unsigned long long)1 << (mantissa - n)); + + res = LLVMBuildMul(builder, src, lp_build_const_scalar(src_type, scale), ""); + res = LLVMBuildAdd(builder, res, lp_build_const_scalar(src_type, bias), ""); + res = LLVMBuildBitCast(builder, res, int_vec_type, ""); + + if(dst_width > n) { + int shift = dst_width - n; + res = LLVMBuildShl(builder, res, lp_build_int_const_scalar(src_type, shift), ""); + + /* Fill in the empty lower bits for added precision? */ +#if 0 + { + LLVMValueRef msb; + msb = LLVMBuildLShr(builder, res, lp_build_int_const_scalar(src_type, dst_width - 1), ""); + msb = LLVMBuildShl(builder, msb, lp_build_int_const_scalar(src_type, shift), ""); + msb = LLVMBuildSub(builder, msb, lp_build_int_const_scalar(src_type, 1), ""); + res = LLVMBuildOr(builder, res, msb, ""); + } +#elif 0 + while(shift > 0) { + res = LLVMBuildOr(builder, res, LLVMBuildLShr(builder, res, lp_build_int_const_scalar(src_type, n), ""), ""); + shift -= n; + n *= 2; + } +#endif + } + else + res = LLVMBuildAnd(builder, res, lp_build_int_const_scalar(src_type, mask), ""); + + return res; +} + + +/** + * Inverse of lp_build_clamped_float_to_unsigned_norm above. + */ +LLVMValueRef +lp_build_unsigned_norm_to_float(LLVMBuilderRef builder, + unsigned src_width, + union lp_type dst_type, + LLVMValueRef src) +{ + LLVMTypeRef vec_type = lp_build_vec_type(dst_type); + LLVMTypeRef int_vec_type = lp_build_int_vec_type(dst_type); + LLVMValueRef bias_; + LLVMValueRef res; + unsigned mantissa; + unsigned n; + unsigned long long ubound; + unsigned long long mask; + double scale; + double bias; + + mantissa = lp_mantissa(dst_type); + + n = MIN2(mantissa, src_width); + + ubound = ((unsigned long long)1 << n); + mask = ubound - 1; + scale = (double)ubound/mask; + bias = (double)((unsigned long long)1 << (mantissa - n)); + + res = src; + + if(src_width > mantissa) { + int shift = src_width - mantissa; + res = LLVMBuildLShr(builder, res, lp_build_int_const_scalar(dst_type, shift), ""); + } + + bias_ = lp_build_const_scalar(dst_type, bias); + + res = LLVMBuildOr(builder, + res, + LLVMBuildBitCast(builder, bias_, int_vec_type, ""), ""); + + res = LLVMBuildBitCast(builder, res, vec_type, ""); + + res = LLVMBuildSub(builder, res, bias_, ""); + res = LLVMBuildMul(builder, res, lp_build_const_scalar(dst_type, scale), ""); + + return res; +} + + +/** + * Build shuffle vectors that match PUNPCKLxx and PUNPCKHxx instructions. + */ +static LLVMValueRef +lp_build_const_unpack_shuffle(unsigned n, unsigned lo_hi) +{ + LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; + unsigned i, j; + + assert(n <= LP_MAX_VECTOR_LENGTH); + assert(lo_hi < 2); + + /* TODO: cache results in a static table */ + + for(i = 0, j = lo_hi*n/2; i < n; i += 2, ++j) { + elems[i + 0] = LLVMConstInt(LLVMInt32Type(), 0 + j, 0); + elems[i + 1] = LLVMConstInt(LLVMInt32Type(), n + j, 0); + } + + return LLVMConstVector(elems, n); +} + + +/** + * Build shuffle vectors that match PACKxx instructions. + */ +static LLVMValueRef +lp_build_const_pack_shuffle(unsigned n) +{ + LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; + unsigned i; + + assert(n <= LP_MAX_VECTOR_LENGTH); + + /* TODO: cache results in a static table */ + + for(i = 0; i < n; ++i) + elems[i] = LLVMConstInt(LLVMInt32Type(), 2*i, 0); + + return LLVMConstVector(elems, n); +} + + +/** + * Expand the bit width. + * + * This will only change the number of bits the values are represented, not the + * values themselved. + */ +static void +lp_build_expand(LLVMBuilderRef builder, + union lp_type src_type, + union lp_type dst_type, + LLVMValueRef src, + LLVMValueRef *dst, unsigned num_dsts) +{ + unsigned num_tmps; + unsigned i; + + /* Register width must remain constant */ + assert(src_type.width * src_type.length == dst_type.width * dst_type.length); + + /* We must not loose or gain channels. Only precision */ + assert(src_type.length == dst_type.length * num_dsts); + + num_tmps = 1; + dst[0] = src; + + while(src_type.width < dst_type.width) { + union lp_type new_type = src_type; + LLVMTypeRef new_vec_type; + + new_type.width *= 2; + new_type.length /= 2; + new_vec_type = lp_build_vec_type(new_type); + + for(i = num_tmps; i--; ) { + LLVMValueRef zero; + LLVMValueRef shuffle_lo; + LLVMValueRef shuffle_hi; + LLVMValueRef lo; + LLVMValueRef hi; + + zero = lp_build_zero(src_type); + shuffle_lo = lp_build_const_unpack_shuffle(src_type.length, 0); + shuffle_hi = lp_build_const_unpack_shuffle(src_type.length, 1); + + /* PUNPCKLBW, PUNPCKHBW */ + lo = LLVMBuildShuffleVector(builder, dst[i], zero, shuffle_lo, ""); + hi = LLVMBuildShuffleVector(builder, dst[i], zero, shuffle_hi, ""); + + dst[2*i + 0] = LLVMBuildBitCast(builder, lo, new_vec_type, ""); + dst[2*i + 1] = LLVMBuildBitCast(builder, hi, new_vec_type, ""); + } + + src_type = new_type; + + num_tmps *= 2; + } + + assert(num_tmps == num_dsts); +} + + +/** + * Non-interleaved pack. + * + * This will move values as + * + * lo = __ l0 __ l1 __ l2 __.. __ ln + * hi = __ h0 __ h1 __ h2 __.. __ hn + * res = l0 l1 l2 .. ln h0 h1 h2 .. hn + * + * TODO: handle saturation consistently. + */ +static LLVMValueRef +lp_build_pack2(LLVMBuilderRef builder, + union lp_type src_type, + union lp_type dst_type, + boolean clamped, + LLVMValueRef lo, + LLVMValueRef hi) +{ + LLVMTypeRef src_vec_type = lp_build_vec_type(src_type); + LLVMTypeRef dst_vec_type = lp_build_vec_type(dst_type); + LLVMValueRef shuffle; + LLVMValueRef res; + + /* Register width must remain constant */ + assert(src_type.width * src_type.length == dst_type.width * dst_type.length); + + /* We must not loose or gain channels. Only precision */ + assert(src_type.length * 2 == dst_type.length); + + assert(!src_type.floating); + assert(!dst_type.floating); + +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) + if(src_type.width * src_type.length == 128) { + /* All X86 non-interleaved pack instructions all take signed inputs and + * saturate them, so saturate beforehand. */ + if(!src_type.sign && !clamped) { + struct lp_build_context bld; + unsigned dst_bits = dst_type.sign ? dst_type.width - 1 : dst_type.width; + LLVMValueRef dst_max = lp_build_int_const_scalar(src_type, ((unsigned long long)1 << dst_bits) - 1); + lp_build_context_init(&bld, builder, src_type); + lo = lp_build_min(&bld, lo, dst_max); + hi = lp_build_min(&bld, hi, dst_max); + } + + switch(src_type.width) { + case 32: + if(dst_type.sign) + res = lp_build_intrinsic_binary(builder, "llvm.x86.sse2.packssdw.128", src_vec_type, lo, hi); + else + /* PACKUSDW is the only instrinsic with a consistent signature */ + return lp_build_intrinsic_binary(builder, "llvm.x86.sse41.packusdw", dst_vec_type, lo, hi); + break; + + case 16: + if(dst_type.sign) + res = lp_build_intrinsic_binary(builder, "llvm.x86.sse2.packsswb.128", src_vec_type, lo, hi); + else + res = lp_build_intrinsic_binary(builder, "llvm.x86.sse2.packuswb.128", src_vec_type, lo, hi); + break; + + default: + assert(0); + return LLVMGetUndef(dst_vec_type); + break; + } + + res = LLVMBuildBitCast(builder, res, dst_vec_type, ""); + return res; + } +#endif + + lo = LLVMBuildBitCast(builder, lo, dst_vec_type, ""); + hi = LLVMBuildBitCast(builder, hi, dst_vec_type, ""); + + shuffle = lp_build_const_pack_shuffle(dst_type.length); + + res = LLVMBuildShuffleVector(builder, lo, hi, shuffle, ""); + + return res; +} + + +/** + * Truncate the bit width. + * + * TODO: Handle saturation consistently. + */ +static LLVMValueRef +lp_build_trunc(LLVMBuilderRef builder, + union lp_type src_type, + union lp_type dst_type, + boolean clamped, + const LLVMValueRef *src, unsigned num_srcs) +{ + LLVMValueRef tmp[LP_MAX_VECTOR_LENGTH]; + unsigned i; + + /* Register width must remain constant */ + assert(src_type.width * src_type.length == dst_type.width * dst_type.length); + + /* We must not loose or gain channels. Only precision */ + assert(src_type.length * num_srcs == dst_type.length); + + for(i = 0; i < num_srcs; ++i) + tmp[i] = src[i]; + + while(src_type.width > dst_type.width) { + union lp_type new_type = src_type; + + new_type.width /= 2; + new_type.length *= 2; + + /* Take in consideration the sign changes only in the last step */ + if(new_type.width == dst_type.width) + new_type.sign = dst_type.sign; + + num_srcs /= 2; + + for(i = 0; i < num_srcs; ++i) + tmp[i] = lp_build_pack2(builder, src_type, new_type, clamped, + tmp[2*i + 0], tmp[2*i + 1]); + + src_type = new_type; + } + + assert(num_srcs == 1); + + return tmp[0]; +} + + +/** + * Generic type conversion. + * + * TODO: Take a precision argument, or even better, add a new precision member + * to the lp_type union. + */ +void +lp_build_conv(LLVMBuilderRef builder, + union lp_type src_type, + union lp_type dst_type, + const LLVMValueRef *src, unsigned num_srcs, + LLVMValueRef *dst, unsigned num_dsts) +{ + union lp_type tmp_type; + LLVMValueRef tmp[LP_MAX_VECTOR_LENGTH]; + unsigned num_tmps; + unsigned i; + + /* Register width must remain constant */ + assert(src_type.width * src_type.length == dst_type.width * dst_type.length); + + /* We must not loose or gain channels. Only precision */ + assert(src_type.length * num_srcs == dst_type.length * num_dsts); + + assert(src_type.length <= LP_MAX_VECTOR_LENGTH); + assert(dst_type.length <= LP_MAX_VECTOR_LENGTH); + + tmp_type = src_type; + for(i = 0; i < num_srcs; ++i) + tmp[i] = src[i]; + num_tmps = num_srcs; + + /* + * Clamp if necessary + */ + + if(src_type.value != dst_type.value) { + struct lp_build_context bld; + double src_min = lp_const_min(src_type); + double dst_min = lp_const_min(dst_type); + double src_max = lp_const_max(src_type); + double dst_max = lp_const_max(dst_type); + LLVMValueRef thres; + + lp_build_context_init(&bld, builder, tmp_type); + + if(src_min < dst_min) { + if(dst_min == 0.0) + thres = bld.zero; + else + thres = lp_build_const_scalar(src_type, dst_min); + for(i = 0; i < num_tmps; ++i) + tmp[i] = lp_build_max(&bld, tmp[i], thres); + } + + if(src_max > dst_max) { + if(dst_max == 1.0) + thres = bld.one; + else + thres = lp_build_const_scalar(src_type, dst_max); + for(i = 0; i < num_tmps; ++i) + tmp[i] = lp_build_min(&bld, tmp[i], thres); + } + } + + /* + * Scale to the narrowest range + */ + + if(dst_type.floating) { + /* Nothing to do */ + } + else if(tmp_type.floating) { + if(!dst_type.fixed && !dst_type.sign && dst_type.norm) { + for(i = 0; i < num_tmps; ++i) { + tmp[i] = lp_build_clamped_float_to_unsigned_norm(builder, + tmp_type, + dst_type.width, + tmp[i]); + } + tmp_type.floating = FALSE; + } + else { + double dst_scale = lp_const_scale(dst_type); + LLVMTypeRef tmp_vec_type; + + if (dst_scale != 1.0) { + LLVMValueRef scale = lp_build_const_scalar(tmp_type, dst_scale); + for(i = 0; i < num_tmps; ++i) + tmp[i] = LLVMBuildMul(builder, tmp[i], scale, ""); + } + + /* Use an equally sized integer for intermediate computations */ + tmp_type.floating = FALSE; + tmp_vec_type = lp_build_vec_type(tmp_type); + for(i = 0; i < num_tmps; ++i) { +#if 0 + if(dst_type.sign) + tmp[i] = LLVMBuildFPToSI(builder, tmp[i], tmp_vec_type, ""); + else + tmp[i] = LLVMBuildFPToUI(builder, tmp[i], tmp_vec_type, ""); +#else + /* FIXME: there is no SSE counterpart for LLVMBuildFPToUI */ + tmp[i] = LLVMBuildFPToSI(builder, tmp[i], tmp_vec_type, ""); +#endif + } + } + } + else { + unsigned src_shift = lp_const_shift(src_type); + unsigned dst_shift = lp_const_shift(dst_type); + + /* FIXME: compensate different offsets too */ + if(src_shift > dst_shift) { + LLVMValueRef shift = lp_build_int_const_scalar(tmp_type, src_shift - dst_shift); + for(i = 0; i < num_tmps; ++i) + if(src_type.sign) + tmp[i] = LLVMBuildAShr(builder, tmp[i], shift, ""); + else + tmp[i] = LLVMBuildLShr(builder, tmp[i], shift, ""); + } + } + + /* + * Truncate or expand bit width + */ + + assert(!tmp_type.floating || tmp_type.width == dst_type.width); + + if(tmp_type.width > dst_type.width) { + assert(num_dsts == 1); + tmp[0] = lp_build_trunc(builder, tmp_type, dst_type, TRUE, tmp, num_tmps); + tmp_type.width = dst_type.width; + tmp_type.length = dst_type.length; + num_tmps = 1; + } + + if(tmp_type.width < dst_type.width) { + assert(num_tmps == 1); + lp_build_expand(builder, tmp_type, dst_type, tmp[0], tmp, num_dsts); + tmp_type.width = dst_type.width; + tmp_type.length = dst_type.length; + num_tmps = num_dsts; + } + + assert(tmp_type.width == dst_type.width); + assert(tmp_type.length == dst_type.length); + assert(num_tmps == num_dsts); + + /* + * Scale to the widest range + */ + + if(src_type.floating) { + /* Nothing to do */ + } + else if(!src_type.floating && dst_type.floating) { + if(!src_type.fixed && !src_type.sign && src_type.norm) { + for(i = 0; i < num_tmps; ++i) { + tmp[i] = lp_build_unsigned_norm_to_float(builder, + src_type.width, + dst_type, + tmp[i]); + } + tmp_type.floating = TRUE; + } + else { + double src_scale = lp_const_scale(src_type); + LLVMTypeRef tmp_vec_type; + + /* Use an equally sized integer for intermediate computations */ + tmp_type.floating = TRUE; + tmp_type.sign = TRUE; + tmp_vec_type = lp_build_vec_type(tmp_type); + for(i = 0; i < num_tmps; ++i) { +#if 0 + if(dst_type.sign) + tmp[i] = LLVMBuildSIToFP(builder, tmp[i], tmp_vec_type, ""); + else + tmp[i] = LLVMBuildUIToFP(builder, tmp[i], tmp_vec_type, ""); +#else + /* FIXME: there is no SSE counterpart for LLVMBuildUIToFP */ + tmp[i] = LLVMBuildSIToFP(builder, tmp[i], tmp_vec_type, ""); +#endif + } + + if (src_scale != 1.0) { + LLVMValueRef scale = lp_build_const_scalar(tmp_type, 1.0/src_scale); + for(i = 0; i < num_tmps; ++i) + tmp[i] = LLVMBuildMul(builder, tmp[i], scale, ""); + } + } + } + else { + unsigned src_shift = lp_const_shift(src_type); + unsigned dst_shift = lp_const_shift(dst_type); + + /* FIXME: compensate different offsets too */ + if(src_shift < dst_shift) { + LLVMValueRef shift = lp_build_int_const_scalar(tmp_type, dst_shift - src_shift); + for(i = 0; i < num_tmps; ++i) + tmp[i] = LLVMBuildShl(builder, tmp[i], shift, ""); + } + } + + for(i = 0; i < num_dsts; ++i) + dst[i] = tmp[i]; +} + + +/** + * Bit mask conversion. + * + * This will convert the integer masks that match the given types. + * + * The mask values should 0 or -1, i.e., all bits either set to zero or one. + * Any other value will likely cause in unpredictable results. + * + * This is basically a very trimmed down version of lp_build_conv. + */ +void +lp_build_conv_mask(LLVMBuilderRef builder, + union lp_type src_type, + union lp_type dst_type, + const LLVMValueRef *src, unsigned num_srcs, + LLVMValueRef *dst, unsigned num_dsts) +{ + /* Register width must remain constant */ + assert(src_type.width * src_type.length == dst_type.width * dst_type.length); + + /* We must not loose or gain channels. Only precision */ + assert(src_type.length * num_srcs == dst_type.length * num_dsts); + + /* + * Drop + * + * We assume all values are 0 or -1 + */ + + src_type.floating = FALSE; + src_type.fixed = FALSE; + src_type.sign = TRUE; + src_type.norm = FALSE; + + dst_type.floating = FALSE; + dst_type.fixed = FALSE; + dst_type.sign = TRUE; + dst_type.norm = FALSE; + + /* + * Truncate or expand bit width + */ + + if(src_type.width > dst_type.width) { + assert(num_dsts == 1); + dst[0] = lp_build_trunc(builder, src_type, dst_type, TRUE, src, num_srcs); + } + else if(src_type.width < dst_type.width) { + assert(num_srcs == 1); + lp_build_expand(builder, src_type, dst_type, src[0], dst, num_dsts); + } + else { + assert(num_srcs == num_dsts); + memcpy(dst, src, num_dsts * sizeof *dst); + } +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_conv.h b/src/gallium/drivers/llvmpipe/lp_bld_conv.h new file mode 100644 index 00000000000..05c1ef2a100 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_conv.h @@ -0,0 +1,73 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Helper functions for type conversions. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#ifndef LP_BLD_CONV_H +#define LP_BLD_CONV_H + + +#include <llvm-c/Core.h> + + +union lp_type type; + + +LLVMValueRef +lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, + union lp_type src_type, + unsigned dst_width, + LLVMValueRef src); + +LLVMValueRef +lp_build_unsigned_norm_to_float(LLVMBuilderRef builder, + unsigned src_width, + union lp_type dst_type, + LLVMValueRef src); + + +void +lp_build_conv(LLVMBuilderRef builder, + union lp_type src_type, + union lp_type dst_type, + const LLVMValueRef *srcs, unsigned num_srcs, + LLVMValueRef *dsts, unsigned num_dsts); + +void +lp_build_conv_mask(LLVMBuilderRef builder, + union lp_type src_type, + union lp_type dst_type, + const LLVMValueRef *src, unsigned num_srcs, + LLVMValueRef *dst, unsigned num_dsts); + +#endif /* !LP_BLD_CONV_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_debug.c b/src/gallium/drivers/llvmpipe/lp_bld_debug.c new file mode 100644 index 00000000000..30925b5f415 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_debug.c @@ -0,0 +1,108 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifdef HAVE_UDIS86 +#include <udis86.h> +#endif + +#include "util/u_debug.h" +#include "lp_bld_debug.h" + + +void +lp_disassemble(const void* func) +{ +#ifdef HAVE_UDIS86 + ud_t ud_obj; + uint64_t max_jmp_pc; + + ud_init(&ud_obj); + + ud_set_input_buffer(&ud_obj, (void*)func, 0xffff); + + max_jmp_pc = (uint64_t) (uintptr_t) func; + ud_set_pc(&ud_obj, max_jmp_pc); + +#ifdef PIPE_ARCH_X86 + ud_set_mode(&ud_obj, 32); +#endif +#ifdef PIPE_ARCH_X86_64 + ud_set_mode(&ud_obj, 64); +#endif + + ud_set_syntax(&ud_obj, UD_SYN_ATT); + + while (ud_disassemble(&ud_obj)) { + +#ifdef PIPE_ARCH_X86 + debug_printf("%08lx: ", (unsigned long)ud_insn_off(&ud_obj)); +#endif +#ifdef PIPE_ARCH_X86_64 + debug_printf("%016llx: ", (unsigned long long)ud_insn_off(&ud_obj)); +#endif + +#if 0 + debug_printf("%-16s ", ud_insn_hex(&ud_obj)); +#endif + + debug_printf("%s\n", ud_insn_asm(&ud_obj)); + + if(ud_obj.mnemonic != UD_Icall) { + unsigned i; + for(i = 0; i < 3; ++i) { + const struct ud_operand *op = &ud_obj.operand[i]; + if (op->type == UD_OP_JIMM){ + uint64_t pc = ud_obj.pc; + + switch (op->size) { + case 8: + pc += op->lval.sbyte; + break; + case 16: + pc += op->lval.sword; + break; + case 32: + pc += op->lval.sdword; + break; + default: + break; + } + if(pc > max_jmp_pc) + max_jmp_pc = pc; + } + } + } + + if (ud_insn_off(&ud_obj) >= max_jmp_pc && ud_obj.mnemonic == UD_Iret) + break; + } + debug_printf("\n"); +#else + (void)func; +#endif +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_debug.h b/src/gallium/drivers/llvmpipe/lp_bld_debug.h new file mode 100644 index 00000000000..ecdafef76d0 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_debug.h @@ -0,0 +1,60 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef LP_BLD_DEBUG_H +#define LP_BLD_DEBUG_H + + +#include <llvm-c/Core.h> + +#include "pipe/p_compiler.h" +#include "util/u_string.h" + + +static INLINE void +lp_build_name(LLVMValueRef val, const char *format, ...) +{ +#ifdef DEBUG + char name[32]; + va_list ap; + va_start(ap, format); + util_vsnprintf(name, sizeof name, format, ap); + va_end(ap); + LLVMSetValueName(val, name); +#else + (void)val; + (void)format; +#endif +} + + +void +lp_disassemble(const void* func); + + +#endif /* !LP_BLD_DEBUG_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_depth.c b/src/gallium/drivers/llvmpipe/lp_bld_depth.c new file mode 100644 index 00000000000..2cd6e6b9217 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_depth.c @@ -0,0 +1,214 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Depth/stencil testing to LLVM IR translation. + * + * To be done accurately/efficiently the depth/stencil test must be done with + * the same type/format of the depth/stencil buffer, which implies massaging + * the incoming depths to fit into place. Using a more straightforward + * type/format for depth/stencil values internally and only convert when + * flushing would avoid this, but it would most likely result in depth fighting + * artifacts. + * + * We are free to use a different pixel layout though. Since our basic + * processing unit is a quad (2x2 pixel block) we store the depth/stencil + * values tiled, a quad at time. That is, a depth buffer containing + * + * Z11 Z12 Z13 Z14 ... + * Z21 Z22 Z23 Z24 ... + * Z31 Z32 Z33 Z34 ... + * Z41 Z42 Z43 Z44 ... + * ... ... ... ... ... + * + * will actually be stored in memory as + * + * Z11 Z12 Z21 Z22 Z13 Z14 Z23 Z24 ... + * Z31 Z32 Z41 Z42 Z33 Z34 Z43 Z44 ... + * ... ... ... ... ... ... ... ... ... + * + * FIXME: Code generate stencil test + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#include "pipe/p_state.h" +#include "util/u_format.h" + +#include "lp_bld_type.h" +#include "lp_bld_const.h" +#include "lp_bld_logic.h" +#include "lp_bld_flow.h" +#include "lp_bld_debug.h" +#include "lp_bld_depth.h" + + +/** + * Return a type appropriate for depth/stencil testing. + */ +union lp_type +lp_depth_type(const struct util_format_description *format_desc, + unsigned length) +{ + union lp_type type; + unsigned swizzle; + + assert(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS); + assert(format_desc->block.width == 1); + assert(format_desc->block.height == 1); + + swizzle = format_desc->swizzle[0]; + assert(swizzle < 4); + + type.value = 0; + type.width = format_desc->block.bits; + + if(format_desc->channel[swizzle].type == UTIL_FORMAT_TYPE_FLOAT) { + type.floating = TRUE; + assert(swizzle = 0); + assert(format_desc->channel[swizzle].size == format_desc->block.bits); + } + else if(format_desc->channel[swizzle].type == UTIL_FORMAT_TYPE_UNSIGNED) { + assert(format_desc->block.bits <= 32); + if(format_desc->channel[swizzle].normalized) + type.norm = TRUE; + } + else + assert(0); + + assert(type.width <= length); + type.length = length / type.width; + + return type; +} + + +/** + * Depth test. + */ +void +lp_build_depth_test(LLVMBuilderRef builder, + const struct pipe_depth_state *state, + union lp_type type, + const struct util_format_description *format_desc, + struct lp_build_mask_context *mask, + LLVMValueRef src, + LLVMValueRef dst_ptr) +{ + struct lp_build_context bld; + unsigned z_swizzle; + LLVMValueRef dst; + LLVMValueRef z_bitmask = NULL; + LLVMValueRef test; + + if(!state->enabled) + return; + + assert(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS); + assert(format_desc->block.width == 1); + assert(format_desc->block.height == 1); + + z_swizzle = format_desc->swizzle[0]; + if(z_swizzle == UTIL_FORMAT_SWIZZLE_NONE) + return; + + /* Sanity checking */ + assert(z_swizzle < 4); + assert(format_desc->block.bits == type.width); + if(type.floating) { + assert(z_swizzle == 0); + assert(format_desc->channel[z_swizzle].type == UTIL_FORMAT_TYPE_FLOAT); + assert(format_desc->channel[z_swizzle].size == format_desc->block.bits); + } + else { + assert(format_desc->channel[z_swizzle].type == UTIL_FORMAT_TYPE_UNSIGNED); + assert(format_desc->channel[z_swizzle].normalized); + assert(!type.fixed); + assert(!type.sign); + assert(type.norm); + } + + /* Setup build context */ + lp_build_context_init(&bld, builder, type); + + dst = LLVMBuildLoad(builder, dst_ptr, ""); + + lp_build_name(dst, "zsbuf"); + + /* Align the source depth bits with the destination's, and mask out any + * stencil or padding bits from both */ + if(format_desc->channel[z_swizzle].size == format_desc->block.bits) { + assert(z_swizzle == 0); + /* nothing to do */ + } + else { + unsigned padding_left; + unsigned padding_right; + unsigned chan; + + assert(format_desc->layout == UTIL_FORMAT_LAYOUT_ARITH); + assert(format_desc->channel[z_swizzle].type == UTIL_FORMAT_TYPE_UNSIGNED); + assert(format_desc->channel[z_swizzle].size <= format_desc->block.bits); + assert(format_desc->channel[z_swizzle].normalized); + + padding_right = 0; + for(chan = 0; chan < z_swizzle; ++chan) + padding_right += format_desc->channel[chan].size; + padding_left = format_desc->block.bits - format_desc->channel[z_swizzle].size; + + if(padding_left || padding_right) { + const long long mask_left = ((long long)1 << (format_desc->block.bits - padding_left)) - 1; + const long long mask_right = ((long long)1 << (padding_right)) - 1; + z_bitmask = lp_build_int_const_scalar(type, mask_left & mask_right); + } + + if(padding_left) + src = LLVMBuildLShr(builder, src, lp_build_int_const_scalar(type, padding_left), ""); + if(padding_right) + src = LLVMBuildAnd(builder, src, z_bitmask, ""); + if(padding_left || padding_right) + dst = LLVMBuildAnd(builder, dst, z_bitmask, ""); + } + + lp_build_name(dst, "zsbuf.z"); + + test = lp_build_cmp(&bld, state->func, src, dst); + lp_build_mask_update(mask, test); + + if(state->writemask) { + if(z_bitmask) + z_bitmask = LLVMBuildAnd(builder, mask->value, z_bitmask, ""); + else + z_bitmask = mask->value; + + dst = lp_build_select(&bld, z_bitmask, src, dst); + LLVMBuildStore(builder, dst, dst_ptr); + } + + assert(!state->occlusion_count); +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_depth.h b/src/gallium/drivers/llvmpipe/lp_bld_depth.h new file mode 100644 index 00000000000..5d2e042fcc5 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_depth.h @@ -0,0 +1,63 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * Depth/stencil testing to LLVM IR translation. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#ifndef LP_BLD_DEPTH_H +#define LP_BLD_DEPTH_H + + +#include <llvm-c/Core.h> + + +struct pipe_depth_state; +struct util_format_description; +union lp_type; +struct lp_build_mask_context; + + +union lp_type +lp_depth_type(const struct util_format_description *format_desc, + unsigned length); + + +void +lp_build_depth_test(LLVMBuilderRef builder, + const struct pipe_depth_state *state, + union lp_type type, + const struct util_format_description *format_desc, + struct lp_build_mask_context *mask, + LLVMValueRef src, + LLVMValueRef dst_ptr); + + +#endif /* !LP_BLD_DEPTH_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_flow.c b/src/gallium/drivers/llvmpipe/lp_bld_flow.c new file mode 100644 index 00000000000..9d99e1a9d9f --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_flow.c @@ -0,0 +1,173 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * LLVM control flow build helpers. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#include "util/u_debug.h" + +#include "lp_bld_type.h" +#include "lp_bld_flow.h" + + +void +lp_build_mask_begin(struct lp_build_mask_context *mask, + LLVMBuilderRef builder, + union lp_type type, + LLVMValueRef value) +{ + memset(mask, 0, sizeof *mask); + + mask->builder = builder; + mask->reg_type = LLVMIntType(type.width * type.length); + mask->value = value; +} + + +void +lp_build_mask_update(struct lp_build_mask_context *mask, + LLVMValueRef value) +{ + + LLVMValueRef cond; + LLVMBasicBlockRef current_block; + LLVMBasicBlockRef next_block; + LLVMBasicBlockRef new_block; + + if(mask->value) + mask->value = LLVMBuildAnd(mask->builder, mask->value, value, ""); + else + mask->value = value; + + /* FIXME: disabled until we have proper control flow helpers */ +#if 0 + cond = LLVMBuildICmp(mask->builder, + LLVMIntEQ, + LLVMBuildBitCast(mask->builder, mask->value, mask->reg_type, ""), + LLVMConstNull(mask->reg_type), + ""); + + current_block = LLVMGetInsertBlock(mask->builder); + + if(!mask->skip_block) { + LLVMValueRef function = LLVMGetBasicBlockParent(current_block); + mask->skip_block = LLVMAppendBasicBlock(function, "skip"); + + mask->phi = LLVMBuildPhi(mask->builder, LLVMTypeOf(mask->value), ""); + } + + next_block = LLVMGetNextBasicBlock(current_block); + assert(next_block); + if(next_block) { + new_block = LLVMInsertBasicBlock(next_block, ""); + } + else { + LLVMValueRef function = LLVMGetBasicBlockParent(current_block); + new_block = LLVMAppendBasicBlock(function, ""); + } + + LLVMAddIncoming(mask->phi, &mask->value, ¤t_block, 1); + LLVMBuildCondBr(mask->builder, cond, mask->skip_block, new_block); + + LLVMPositionBuilderAtEnd(mask->builder, new_block); +#endif +} + + +LLVMValueRef +lp_build_mask_end(struct lp_build_mask_context *mask) +{ + if(mask->skip_block) { + LLVMBasicBlockRef current_block = LLVMGetInsertBlock(mask->builder); + + LLVMAddIncoming(mask->phi, &mask->value, ¤t_block, 1); + LLVMBuildBr(mask->builder, mask->skip_block); + + LLVMPositionBuilderAtEnd(mask->builder, mask->skip_block); + + mask->value = mask->phi; + mask->phi = NULL; + mask->skip_block = NULL; + } + + return mask->value; +} + + + +void +lp_build_loop_begin(LLVMBuilderRef builder, + LLVMValueRef start, + struct lp_build_loop_state *state) +{ + LLVMBasicBlockRef block = LLVMGetInsertBlock(builder); + LLVMValueRef function = LLVMGetBasicBlockParent(block); + + state->block = LLVMAppendBasicBlock(function, "loop"); + + LLVMBuildBr(builder, state->block); + + LLVMPositionBuilderAtEnd(builder, state->block); + + state->counter = LLVMBuildPhi(builder, LLVMTypeOf(start), ""); + + LLVMAddIncoming(state->counter, &start, &block, 1); + +} + + +void +lp_build_loop_end(LLVMBuilderRef builder, + LLVMValueRef end, + LLVMValueRef step, + struct lp_build_loop_state *state) +{ + LLVMBasicBlockRef block = LLVMGetInsertBlock(builder); + LLVMValueRef function = LLVMGetBasicBlockParent(block); + LLVMValueRef next; + LLVMValueRef cond; + LLVMBasicBlockRef after_block; + + if (!step) + step = LLVMConstInt(LLVMTypeOf(end), 1, 0); + + next = LLVMBuildAdd(builder, state->counter, step, ""); + + cond = LLVMBuildICmp(builder, LLVMIntNE, next, end, ""); + + after_block = LLVMAppendBasicBlock(function, ""); + + LLVMBuildCondBr(builder, cond, after_block, state->block); + + LLVMAddIncoming(state->counter, &next, &block, 1); + + LLVMPositionBuilderAtEnd(builder, after_block); +} + diff --git a/src/gallium/drivers/llvmpipe/lp_bld_flow.h b/src/gallium/drivers/llvmpipe/lp_bld_flow.h new file mode 100644 index 00000000000..1b634ff038d --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_flow.h @@ -0,0 +1,103 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * LLVM control flow build helpers. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#ifndef LP_BLD_FLOW_H +#define LP_BLD_FLOW_H + + +#include <llvm-c/Core.h> + + +union lp_type; + + +struct lp_build_mask_context +{ + LLVMBuilderRef builder; + + LLVMTypeRef reg_type; + + LLVMValueRef value; + + LLVMValueRef phi; + + LLVMBasicBlockRef skip_block; +}; + + +void +lp_build_mask_begin(struct lp_build_mask_context *mask, + LLVMBuilderRef builder, + union lp_type type, + LLVMValueRef value); + +/** + * Bitwise AND the mask with the given value, if a previous mask was set. + */ +void +lp_build_mask_update(struct lp_build_mask_context *mask, + LLVMValueRef value); + +LLVMValueRef +lp_build_mask_end(struct lp_build_mask_context *mask); + + +/** + * LLVM's IR doesn't represent for-loops directly. Furthermore it + * it requires creating code blocks, branches, phi variables, so it + * requires a fair amount of code. + * + * @sa http://www.llvm.org/docs/tutorial/LangImpl5.html#for + */ +struct lp_build_loop_state +{ + LLVMBasicBlockRef block; + LLVMValueRef counter; +}; + + +void +lp_build_loop_begin(LLVMBuilderRef builder, + LLVMValueRef start, + struct lp_build_loop_state *state); + + +void +lp_build_loop_end(LLVMBuilderRef builder, + LLVMValueRef end, + LLVMValueRef step, + struct lp_build_loop_state *state); + + + +#endif /* !LP_BLD_FLOW_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_format.h b/src/gallium/drivers/llvmpipe/lp_bld_format.h new file mode 100644 index 00000000000..01c8a752d18 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_format.h @@ -0,0 +1,101 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef LP_BLD_H +#define LP_BLD_H + + +/** + * @file + * LLVM IR building helpers interfaces. + * + * We use LLVM-C bindings for now. They are not documented, but follow the C++ + * interfaces very closely, and appear to be complete enough for code + * genration. See + * http://npcontemplation.blogspot.com/2008/06/secret-of-llvm-c-bindings.html + * for a standalone example. + */ + +#include <llvm-c/Core.h> + +#include "pipe/p_format.h" + + +union lp_type; + + +/** + * Unpack a pixel into its RGBA components. + * + * @param packed integer. + * + * @return RGBA in a 4 floats vector. + */ +LLVMValueRef +lp_build_unpack_rgba(LLVMBuilderRef builder, + enum pipe_format format, + LLVMValueRef packed); + + +/** + * Pack a pixel. + * + * @param rgba 4 float vector with the unpacked components. + */ +LLVMValueRef +lp_build_pack_rgba(LLVMBuilderRef builder, + enum pipe_format format, + LLVMValueRef rgba); + + +/** + * Load a pixel into its RGBA components. + * + * @param ptr value with the pointer to the packed pixel. Pointer type is + * irrelevant. + * + * @return RGBA in a 4 floats vector. + */ +LLVMValueRef +lp_build_load_rgba(LLVMBuilderRef builder, + enum pipe_format format, + LLVMValueRef ptr); + + +/** + * Store a pixel. + * + * @param rgba 4 float vector with the unpacked components. + */ +void +lp_build_store_rgba(LLVMBuilderRef builder, + enum pipe_format format, + LLVMValueRef ptr, + LLVMValueRef rgba); + + +#endif /* !LP_BLD_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_format_aos.c b/src/gallium/drivers/llvmpipe/lp_bld_format_aos.c new file mode 100644 index 00000000000..dcbc0076c7d --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_format_aos.c @@ -0,0 +1,303 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "util/u_format.h" + +#include "lp_bld_format.h" + + +LLVMValueRef +lp_build_unpack_rgba(LLVMBuilderRef builder, + enum pipe_format format, + LLVMValueRef packed) +{ + const struct util_format_description *desc; + LLVMTypeRef type; + LLVMValueRef shifted, casted, scaled, masked; + LLVMValueRef shifts[4]; + LLVMValueRef masks[4]; + LLVMValueRef scales[4]; + LLVMValueRef swizzles[4]; + LLVMValueRef aux[4]; + bool normalized; + int empty_channel; + unsigned shift; + unsigned i; + + desc = util_format_description(format); + + /* FIXME: Support more formats */ + assert(desc->layout == UTIL_FORMAT_LAYOUT_ARITH); + assert(desc->block.width == 1); + assert(desc->block.height == 1); + assert(desc->block.bits <= 32); + + type = LLVMIntType(desc->block.bits); + + /* Do the intermediate integer computations with 32bit integers since it + * matches floating point size */ + if (desc->block.bits < 32) + packed = LLVMBuildZExt(builder, packed, LLVMInt32Type(), ""); + + /* Broadcast the packed value to all four channels */ + packed = LLVMBuildInsertElement(builder, + LLVMGetUndef(LLVMVectorType(LLVMInt32Type(), 4)), + packed, + LLVMConstNull(LLVMInt32Type()), + ""); + packed = LLVMBuildShuffleVector(builder, + packed, + LLVMGetUndef(LLVMVectorType(LLVMInt32Type(), 4)), + LLVMConstNull(LLVMVectorType(LLVMInt32Type(), 4)), + ""); + + /* Initialize vector constants */ + normalized = FALSE; + empty_channel = -1; + shift = 0; + for (i = 0; i < 4; ++i) { + unsigned bits = desc->channel[i].size; + + if (desc->channel[i].type == UTIL_FORMAT_TYPE_VOID) { + shifts[i] = LLVMGetUndef(LLVMInt32Type()); + masks[i] = LLVMConstNull(LLVMInt32Type()); + scales[i] = LLVMConstNull(LLVMFloatType()); + empty_channel = i; + } + else { + unsigned mask = (1 << bits) - 1; + + assert(desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED); + assert(bits < 32); + + shifts[i] = LLVMConstInt(LLVMInt32Type(), shift, 0); + masks[i] = LLVMConstInt(LLVMInt32Type(), mask, 0); + + if (desc->channel[i].normalized) { + scales[i] = LLVMConstReal(LLVMFloatType(), 1.0/mask); + normalized = TRUE; + } + else + scales[i] = LLVMConstReal(LLVMFloatType(), 1.0); + } + + shift += bits; + } + + shifted = LLVMBuildLShr(builder, packed, LLVMConstVector(shifts, 4), ""); + masked = LLVMBuildAnd(builder, shifted, LLVMConstVector(masks, 4), ""); + // UIToFP can't be expressed in SSE2 + casted = LLVMBuildSIToFP(builder, masked, LLVMVectorType(LLVMFloatType(), 4), ""); + + if (normalized) + scaled = LLVMBuildMul(builder, casted, LLVMConstVector(scales, 4), ""); + else + scaled = casted; + + for (i = 0; i < 4; ++i) + aux[i] = LLVMGetUndef(LLVMFloatType()); + + for (i = 0; i < 4; ++i) { + enum util_format_swizzle swizzle = desc->swizzle[i]; + + switch (swizzle) { + case UTIL_FORMAT_SWIZZLE_X: + case UTIL_FORMAT_SWIZZLE_Y: + case UTIL_FORMAT_SWIZZLE_Z: + case UTIL_FORMAT_SWIZZLE_W: + swizzles[i] = LLVMConstInt(LLVMInt32Type(), swizzle, 0); + break; + case UTIL_FORMAT_SWIZZLE_0: + assert(empty_channel >= 0); + swizzles[i] = LLVMConstInt(LLVMInt32Type(), empty_channel, 0); + break; + case UTIL_FORMAT_SWIZZLE_1: + swizzles[i] = LLVMConstInt(LLVMInt32Type(), 4, 0); + aux[0] = LLVMConstReal(LLVMFloatType(), 1.0); + break; + case UTIL_FORMAT_SWIZZLE_NONE: + swizzles[i] = LLVMGetUndef(LLVMFloatType()); + assert(0); + break; + } + } + + return LLVMBuildShuffleVector(builder, scaled, LLVMConstVector(aux, 4), LLVMConstVector(swizzles, 4), ""); +} + + +LLVMValueRef +lp_build_pack_rgba(LLVMBuilderRef builder, + enum pipe_format format, + LLVMValueRef rgba) +{ + const struct util_format_description *desc; + LLVMTypeRef type; + LLVMValueRef packed = NULL; + LLVMValueRef swizzles[4]; + LLVMValueRef shifted, casted, scaled, unswizzled; + LLVMValueRef shifts[4]; + LLVMValueRef scales[4]; + bool normalized; + unsigned shift; + unsigned i, j; + + desc = util_format_description(format); + + assert(desc->layout == UTIL_FORMAT_LAYOUT_ARITH); + assert(desc->block.width == 1); + assert(desc->block.height == 1); + + type = LLVMIntType(desc->block.bits); + + /* Unswizzle the color components into the source vector. */ + for (i = 0; i < 4; ++i) { + for (j = 0; j < 4; ++j) { + if (desc->swizzle[j] == i) + break; + } + if (j < 4) + swizzles[i] = LLVMConstInt(LLVMInt32Type(), j, 0); + else + swizzles[i] = LLVMGetUndef(LLVMInt32Type()); + } + + unswizzled = LLVMBuildShuffleVector(builder, rgba, + LLVMGetUndef(LLVMVectorType(LLVMFloatType(), 4)), + LLVMConstVector(swizzles, 4), ""); + + normalized = FALSE; + shift = 0; + for (i = 0; i < 4; ++i) { + unsigned bits = desc->channel[i].size; + + if (desc->channel[i].type == UTIL_FORMAT_TYPE_VOID) { + shifts[i] = LLVMGetUndef(LLVMInt32Type()); + scales[i] = LLVMGetUndef(LLVMFloatType()); + } + else { + unsigned mask = (1 << bits) - 1; + + assert(desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED); + assert(bits < 32); + + shifts[i] = LLVMConstInt(LLVMInt32Type(), shift, 0); + + if (desc->channel[i].normalized) { + scales[i] = LLVMConstReal(LLVMFloatType(), mask); + normalized = TRUE; + } + else + scales[i] = LLVMConstReal(LLVMFloatType(), 1.0); + } + + shift += bits; + } + + if (normalized) + scaled = LLVMBuildMul(builder, unswizzled, LLVMConstVector(scales, 4), ""); + else + scaled = unswizzled; + + casted = LLVMBuildFPToSI(builder, scaled, LLVMVectorType(LLVMInt32Type(), 4), ""); + + shifted = LLVMBuildShl(builder, casted, LLVMConstVector(shifts, 4), ""); + + /* Bitwise or all components */ + for (i = 0; i < 4; ++i) { + if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED) { + LLVMValueRef component = LLVMBuildExtractElement(builder, shifted, LLVMConstInt(LLVMInt32Type(), i, 0), ""); + if (packed) + packed = LLVMBuildOr(builder, packed, component, ""); + else + packed = component; + } + } + + if (!packed) + packed = LLVMGetUndef(LLVMInt32Type()); + + if (desc->block.bits < 32) + packed = LLVMBuildTrunc(builder, packed, type, ""); + + return packed; +} + + +LLVMValueRef +lp_build_load_rgba(LLVMBuilderRef builder, + enum pipe_format format, + LLVMValueRef ptr) +{ + const struct util_format_description *desc; + LLVMTypeRef type; + LLVMValueRef packed; + + desc = util_format_description(format); + + /* FIXME: Support more formats */ + assert(desc->layout == UTIL_FORMAT_LAYOUT_ARITH); + assert(desc->block.width == 1); + assert(desc->block.height == 1); + assert(desc->block.bits <= 32); + + type = LLVMIntType(desc->block.bits); + + ptr = LLVMBuildBitCast(builder, ptr, LLVMPointerType(type, 0), ""); + + packed = LLVMBuildLoad(builder, ptr, ""); + + return lp_build_unpack_rgba(builder, format, packed); +} + + +void +lp_build_store_rgba(LLVMBuilderRef builder, + enum pipe_format format, + LLVMValueRef ptr, + LLVMValueRef rgba) +{ + const struct util_format_description *desc; + LLVMTypeRef type; + LLVMValueRef packed; + + desc = util_format_description(format); + + assert(desc->layout == UTIL_FORMAT_LAYOUT_ARITH); + assert(desc->block.width == 1); + assert(desc->block.height == 1); + + type = LLVMIntType(desc->block.bits); + + packed = lp_build_pack_rgba(builder, format, rgba); + + ptr = LLVMBuildBitCast(builder, ptr, LLVMPointerType(type, 0), ""); + + LLVMBuildStore(builder, packed, ptr); +} + diff --git a/src/gallium/drivers/llvmpipe/lp_bld_interp.c b/src/gallium/drivers/llvmpipe/lp_bld_interp.c new file mode 100644 index 00000000000..cfe20a0d75b --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_interp.c @@ -0,0 +1,377 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Position and shader input interpolation. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#include "pipe/p_shader_tokens.h" +#include "util/u_debug.h" +#include "util/u_memory.h" +#include "util/u_math.h" +#include "tgsi/tgsi_parse.h" +#include "lp_bld_debug.h" +#include "lp_bld_const.h" +#include "lp_bld_arit.h" +#include "lp_bld_swizzle.h" +#include "lp_bld_interp.h" + + +static void +attrib_name(LLVMValueRef val, unsigned attrib, unsigned chan, const char *suffix) +{ + if(attrib == 0) + lp_build_name(val, "pos.%c%s", "xyzw"[chan], suffix); + else + lp_build_name(val, "input%u.%c%s", attrib - 1, "xyzw"[chan], suffix); +} + + +static void +coeffs_init(struct lp_build_interp_soa_context *bld, + LLVMValueRef a0_ptr, + LLVMValueRef dadx_ptr, + LLVMValueRef dady_ptr) +{ + LLVMBuilderRef builder = bld->base.builder; + unsigned attrib; + unsigned chan; + + for(attrib = 0; attrib < bld->num_attribs; ++attrib) { + unsigned mask = bld->mask[attrib]; + unsigned mode = bld->mode[attrib]; + for(chan = 0; chan < NUM_CHANNELS; ++chan) { + if(mask & (1 << chan)) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), attrib*NUM_CHANNELS + chan, 0); + LLVMValueRef a0 = NULL; + LLVMValueRef dadx = NULL; + LLVMValueRef dady = NULL; + + switch( mode ) { + case TGSI_INTERPOLATE_PERSPECTIVE: + /* fall-through */ + + case TGSI_INTERPOLATE_LINEAR: + dadx = LLVMBuildLoad(builder, LLVMBuildGEP(builder, dadx_ptr, &index, 1, ""), ""); + dady = LLVMBuildLoad(builder, LLVMBuildGEP(builder, dady_ptr, &index, 1, ""), ""); + dadx = lp_build_broadcast_scalar(&bld->base, dadx); + dady = lp_build_broadcast_scalar(&bld->base, dady); + attrib_name(dadx, attrib, chan, ".dadx"); + attrib_name(dady, attrib, chan, ".dady"); + /* fall-through */ + + case TGSI_INTERPOLATE_CONSTANT: + a0 = LLVMBuildLoad(builder, LLVMBuildGEP(builder, a0_ptr, &index, 1, ""), ""); + a0 = lp_build_broadcast_scalar(&bld->base, a0); + attrib_name(a0, attrib, chan, ".dady"); + break; + + default: + assert(0); + break; + } + + bld->a0 [attrib][chan] = a0; + bld->dadx[attrib][chan] = dadx; + bld->dady[attrib][chan] = dady; + } + } + } +} + + +/** + * Small vector x scale multiplication optimization. + * + * TODO: Should be elsewhere. + */ +static LLVMValueRef +coeff_multiply(struct lp_build_interp_soa_context *bld, + LLVMValueRef coeff, + int step) +{ + LLVMValueRef factor; + + switch(step) { + case 0: + return bld->base.zero; + case 1: + return coeff; + case 2: + return lp_build_add(&bld->base, coeff, coeff); + default: + factor = lp_build_const_scalar(bld->base.type, (double)step); + return lp_build_mul(&bld->base, coeff, factor); + } +} + + +/** + * Multiply the dadx and dady with the xstep and ystep respectively. + */ +static void +coeffs_update(struct lp_build_interp_soa_context *bld) +{ + unsigned attrib; + unsigned chan; + + for(attrib = 0; attrib < bld->num_attribs; ++attrib) { + unsigned mask = bld->mask[attrib]; + unsigned mode = bld->mode[attrib]; + if (mode != TGSI_INTERPOLATE_CONSTANT) { + for(chan = 0; chan < NUM_CHANNELS; ++chan) { + if(mask & (1 << chan)) { + bld->dadx[attrib][chan] = coeff_multiply(bld, bld->dadx[attrib][chan], bld->xstep); + bld->dady[attrib][chan] = coeff_multiply(bld, bld->dady[attrib][chan], bld->ystep); + } + } + } + } +} + + +static void +attribs_init(struct lp_build_interp_soa_context *bld) +{ + LLVMValueRef x = bld->pos[0]; + LLVMValueRef y = bld->pos[1]; + LLVMValueRef oow = NULL; + unsigned attrib; + unsigned chan; + + for(attrib = 0; attrib < bld->num_attribs; ++attrib) { + unsigned mask = bld->mask[attrib]; + unsigned mode = bld->mode[attrib]; + for(chan = 0; chan < NUM_CHANNELS; ++chan) { + if(mask & (1 << chan)) { + LLVMValueRef a0 = bld->a0 [attrib][chan]; + LLVMValueRef dadx = bld->dadx[attrib][chan]; + LLVMValueRef dady = bld->dady[attrib][chan]; + LLVMValueRef res; + + res = a0; + + if (mode != TGSI_INTERPOLATE_CONSTANT) { + res = lp_build_add(&bld->base, res, lp_build_mul(&bld->base, x, dadx)); + res = lp_build_add(&bld->base, res, lp_build_mul(&bld->base, y, dady)); + } + + /* Keep the value of the attribue before perspective divide for faster updates */ + bld->attribs_pre[attrib][chan] = res; + + if (mode == TGSI_INTERPOLATE_PERSPECTIVE) { + LLVMValueRef w = bld->pos[3]; + assert(attrib != 0); + if(!oow) + oow = lp_build_rcp(&bld->base, w); + res = lp_build_mul(&bld->base, res, oow); + } + + attrib_name(res, attrib, chan, ""); + + bld->attribs[attrib][chan] = res; + } + } + } +} + + +static void +attribs_update(struct lp_build_interp_soa_context *bld) +{ + LLVMValueRef oow = NULL; + unsigned attrib; + unsigned chan; + + for(attrib = 0; attrib < bld->num_attribs; ++attrib) { + unsigned mask = bld->mask[attrib]; + unsigned mode = bld->mode[attrib]; + + if (mode != TGSI_INTERPOLATE_CONSTANT) { + for(chan = 0; chan < NUM_CHANNELS; ++chan) { + if(mask & (1 << chan)) { + LLVMValueRef dadx = bld->dadx[attrib][chan]; + LLVMValueRef dady = bld->dady[attrib][chan]; + LLVMValueRef res; + + res = bld->attribs_pre[attrib][chan]; + + if(bld->xstep) + res = lp_build_add(&bld->base, res, dadx); + + if(bld->ystep) + res = lp_build_add(&bld->base, res, dady); + + bld->attribs_pre[attrib][chan] = res; + + if (mode == TGSI_INTERPOLATE_PERSPECTIVE) { + LLVMValueRef w = bld->pos[3]; + assert(attrib != 0); + if(!oow) + oow = lp_build_rcp(&bld->base, w); + res = lp_build_mul(&bld->base, res, oow); + } + + attrib_name(res, attrib, chan, ""); + + bld->attribs[attrib][chan] = res; + } + } + } + } +} + + +/** + * Generate the position vectors. + * + * Parameter x0, y0 are the integer values with the quad upper left coordinates. + */ +static void +pos_init(struct lp_build_interp_soa_context *bld, + LLVMValueRef x0, + LLVMValueRef y0) +{ + lp_build_name(x0, "pos.x"); + lp_build_name(y0, "pos.y"); + + bld->attribs[0][0] = x0; + bld->attribs[0][1] = y0; +} + + +static void +pos_update(struct lp_build_interp_soa_context *bld) +{ + LLVMValueRef x = bld->attribs[0][0]; + LLVMValueRef y = bld->attribs[0][1]; + + if(bld->xstep) + x = lp_build_add(&bld->base, x, lp_build_const_scalar(bld->base.type, bld->xstep)); + + if(bld->ystep) + y = lp_build_add(&bld->base, y, lp_build_const_scalar(bld->base.type, bld->ystep)); + + lp_build_name(x, "pos.x"); + lp_build_name(y, "pos.y"); + + bld->attribs[0][0] = x; + bld->attribs[0][1] = y; +} + + +void +lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld, + const struct tgsi_token *tokens, + LLVMBuilderRef builder, + union lp_type type, + LLVMValueRef a0_ptr, + LLVMValueRef dadx_ptr, + LLVMValueRef dady_ptr, + LLVMValueRef x0, + LLVMValueRef y0, + int xstep, + int ystep) +{ + struct tgsi_parse_context parse; + struct tgsi_full_declaration *decl; + + memset(bld, 0, sizeof *bld); + + lp_build_context_init(&bld->base, builder, type); + + /* For convenience */ + bld->pos = bld->attribs[0]; + bld->inputs = (const LLVMValueRef (*)[NUM_CHANNELS]) bld->attribs[1]; + + /* Position */ + bld->num_attribs = 1; + bld->mask[0] = TGSI_WRITEMASK_ZW; + bld->mode[0] = TGSI_INTERPOLATE_LINEAR; + + /* Inputs */ + tgsi_parse_init( &parse, tokens ); + while( !tgsi_parse_end_of_tokens( &parse ) ) { + tgsi_parse_token( &parse ); + + switch( parse.FullToken.Token.Type ) { + case TGSI_TOKEN_TYPE_DECLARATION: + decl = &parse.FullToken.FullDeclaration; + if( decl->Declaration.File == TGSI_FILE_INPUT ) { + unsigned first, last, mask; + unsigned attrib; + + first = decl->DeclarationRange.First; + last = decl->DeclarationRange.Last; + mask = decl->Declaration.UsageMask; + + for( attrib = first; attrib <= last; ++attrib ) { + bld->mask[1 + attrib] = mask; + bld->mode[1 + attrib] = decl->Declaration.Interpolate; + } + + bld->num_attribs = MAX2(bld->num_attribs, 1 + last + 1); + } + break; + + case TGSI_TOKEN_TYPE_INSTRUCTION: + case TGSI_TOKEN_TYPE_IMMEDIATE: + break; + + default: + assert( 0 ); + } + } + tgsi_parse_free( &parse ); + + coeffs_init(bld, a0_ptr, dadx_ptr, dady_ptr); + + pos_init(bld, x0, y0); + + attribs_init(bld); + + bld->xstep = xstep; + bld->ystep = ystep; + + coeffs_update(bld); +} + + +/** + * Advance the position and inputs with the xstep and ystep. + */ +void +lp_build_interp_soa_update(struct lp_build_interp_soa_context *bld) +{ + pos_update(bld); + + attribs_update(bld); +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_interp.h b/src/gallium/drivers/llvmpipe/lp_bld_interp.h new file mode 100644 index 00000000000..9194f6233a7 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_interp.h @@ -0,0 +1,99 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Position and shader input interpolation. + * + * Special attention is given to the interpolation of side by side quads. + * Multiplications are made only for the first quad. Interpolation of + * inputs for posterior quads are done exclusively with additions, and + * perspective divide if necessary. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#ifndef LP_BLD_INTERP_H +#define LP_BLD_INTERP_H + + +#include <llvm-c/Core.h> + +#include "tgsi/tgsi_exec.h" + +#include "lp_bld_type.h" + + +struct tgsi_token; + + +struct lp_build_interp_soa_context +{ + struct lp_build_context base; + + unsigned num_attribs; + unsigned mask[1 + PIPE_MAX_SHADER_INPUTS]; + unsigned mode[1 + PIPE_MAX_SHADER_INPUTS]; + + LLVMValueRef a0 [1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]; + LLVMValueRef dadx[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]; + LLVMValueRef dady[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]; + + int xstep; + int ystep; + + /* Attribute values before perspective divide */ + LLVMValueRef attribs_pre[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]; + + LLVMValueRef attribs[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]; + + /* + * Convenience pointers. Callers may access this one. + */ + const LLVMValueRef *pos; + const LLVMValueRef (*inputs)[NUM_CHANNELS]; +}; + + +void +lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld, + const struct tgsi_token *tokens, + LLVMBuilderRef builder, + union lp_type type, + LLVMValueRef a0_ptr, + LLVMValueRef dadx_ptr, + LLVMValueRef dady_ptr, + LLVMValueRef x0, + LLVMValueRef y0, + int xstep, + int ystep); + +void +lp_build_interp_soa_update(struct lp_build_interp_soa_context *bld); + + +#endif /* LP_BLD_INTERP_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_intr.c b/src/gallium/drivers/llvmpipe/lp_bld_intr.c new file mode 100644 index 00000000000..9895749d568 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_intr.c @@ -0,0 +1,192 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * @file + * Helpers for emiting intrinsic calls. + * + * LLVM vanilla IR doesn't represent all basic arithmetic operations we care + * about, and it is often necessary to resort target-specific intrinsics for + * performance, convenience. + * + * Ideally we would like to stay away from target specific intrinsics and + * move all the instruction selection logic into upstream LLVM where it belongs. + * + * These functions are also used for calling C functions provided by us from + * generated LLVM code. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#include "util/u_debug.h" + +#include "lp_bld_intr.h" + + +LLVMValueRef +lp_declare_intrinsic(LLVMModuleRef module, + const char *name, + LLVMTypeRef ret_type, + LLVMTypeRef *arg_types, + unsigned num_args) +{ + LLVMTypeRef function_type; + LLVMValueRef function; + + assert(!LLVMGetNamedFunction(module, name)); + + function_type = LLVMFunctionType(ret_type, arg_types, num_args, 0); + function = LLVMAddFunction(module, name, function_type); + + LLVMSetFunctionCallConv(function, LLVMCCallConv); + LLVMSetLinkage(function, LLVMExternalLinkage); + + assert(LLVMIsDeclaration(function)); + + if(name[0] == 'l' && + name[1] == 'l' && + name[2] == 'v' && + name[3] == 'm' && + name[4] == '.') + assert(LLVMGetIntrinsicID(function)); + + return function; +} + + +LLVMValueRef +lp_build_intrinsic(LLVMBuilderRef builder, + const char *name, + LLVMTypeRef ret_type, + LLVMValueRef *args, + unsigned num_args) +{ + LLVMModuleRef module = LLVMGetGlobalParent(LLVMGetBasicBlockParent(LLVMGetInsertBlock(builder))); + LLVMValueRef function; + + function = LLVMGetNamedFunction(module, name); + if(!function) { + LLVMTypeRef arg_types[LP_MAX_FUNC_ARGS]; + unsigned i; + + assert(num_args <= LP_MAX_FUNC_ARGS); + + for(i = 0; i < num_args; ++i) { + assert(args[i]); + arg_types[i] = LLVMTypeOf(args[i]); + } + + function = lp_declare_intrinsic(module, name, ret_type, arg_types, num_args); + } + + return LLVMBuildCall(builder, function, args, num_args, ""); +} + + +LLVMValueRef +lp_build_intrinsic_unary(LLVMBuilderRef builder, + const char *name, + LLVMTypeRef ret_type, + LLVMValueRef a) +{ + return lp_build_intrinsic(builder, name, ret_type, &a, 1); +} + + +LLVMValueRef +lp_build_intrinsic_binary(LLVMBuilderRef builder, + const char *name, + LLVMTypeRef ret_type, + LLVMValueRef a, + LLVMValueRef b) +{ + LLVMValueRef args[2]; + + args[0] = a; + args[1] = b; + + return lp_build_intrinsic(builder, name, ret_type, args, 2); +} + + +LLVMValueRef +lp_build_intrinsic_map(LLVMBuilderRef builder, + const char *name, + LLVMTypeRef ret_type, + LLVMValueRef *args, + unsigned num_args) +{ + LLVMTypeRef ret_elem_type = LLVMGetElementType(ret_type); + unsigned n = LLVMGetVectorSize(ret_type); + unsigned i, j; + LLVMValueRef res; + + assert(num_args <= LP_MAX_FUNC_ARGS); + + res = LLVMGetUndef(ret_type); + for(i = 0; i < n; ++i) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef arg_elems[LP_MAX_FUNC_ARGS]; + LLVMValueRef res_elem; + for(j = 0; j < num_args; ++j) + arg_elems[j] = LLVMBuildExtractElement(builder, args[j], index, ""); + res_elem = lp_build_intrinsic(builder, name, ret_elem_type, arg_elems, num_args); + res = LLVMBuildInsertElement(builder, res, res_elem, index, ""); + } + + return res; +} + + +LLVMValueRef +lp_build_intrinsic_map_unary(LLVMBuilderRef builder, + const char *name, + LLVMTypeRef ret_type, + LLVMValueRef a) +{ + return lp_build_intrinsic_map(builder, name, ret_type, &a, 1); +} + + +LLVMValueRef +lp_build_intrinsic_map_binary(LLVMBuilderRef builder, + const char *name, + LLVMTypeRef ret_type, + LLVMValueRef a, + LLVMValueRef b) +{ + LLVMValueRef args[2]; + + args[0] = a; + args[1] = b; + + return lp_build_intrinsic_map(builder, name, ret_type, args, 2); +} + + diff --git a/src/gallium/drivers/llvmpipe/lp_bld_intr.h b/src/gallium/drivers/llvmpipe/lp_bld_intr.h new file mode 100644 index 00000000000..f813f27074b --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_intr.h @@ -0,0 +1,102 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Helper functions for calling intrinsics. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#ifndef LP_BLD_INTR_H +#define LP_BLD_INTR_H + + +#include <llvm-c/Core.h> + + +/** + * Max number of arguments in an intrinsic. + */ +#define LP_MAX_FUNC_ARGS 32 + + +LLVMValueRef +lp_declare_intrinsic(LLVMModuleRef module, + const char *name, + LLVMTypeRef ret_type, + LLVMTypeRef *arg_types, + unsigned num_args); + +LLVMValueRef +lp_build_intrinsic(LLVMBuilderRef builder, + const char *name, + LLVMTypeRef ret_type, + LLVMValueRef *args, + unsigned num_args); + + +LLVMValueRef +lp_build_intrinsic_unary(LLVMBuilderRef builder, + const char *name, + LLVMTypeRef ret_type, + LLVMValueRef a); + + +LLVMValueRef +lp_build_intrinsic_binary(LLVMBuilderRef builder, + const char *name, + LLVMTypeRef ret_type, + LLVMValueRef a, + LLVMValueRef b); + + +LLVMValueRef +lp_build_intrinsic_map(LLVMBuilderRef builder, + const char *name, + LLVMTypeRef ret_type, + LLVMValueRef *args, + unsigned num_args); + + +LLVMValueRef +lp_build_intrinsic_map_unary(LLVMBuilderRef builder, + const char *name, + LLVMTypeRef ret_type, + LLVMValueRef a); + + +LLVMValueRef +lp_build_intrinsic_map_binary(LLVMBuilderRef builder, + const char *name, + LLVMTypeRef ret_type, + LLVMValueRef a, + LLVMValueRef b); + + +#endif /* !LP_BLD_INTR_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_logic.c b/src/gallium/drivers/llvmpipe/lp_bld_logic.c new file mode 100644 index 00000000000..8631efd6c3e --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_logic.c @@ -0,0 +1,397 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Helper functions for logical operations. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#include "lp_bld_type.h" +#include "lp_bld_const.h" +#include "lp_bld_intr.h" +#include "lp_bld_logic.h" + + +LLVMValueRef +lp_build_cmp(struct lp_build_context *bld, + unsigned func, + LLVMValueRef a, + LLVMValueRef b) +{ + const union lp_type type = bld->type; + LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); + LLVMValueRef zeros = LLVMConstNull(int_vec_type); + LLVMValueRef ones = LLVMConstAllOnes(int_vec_type); + LLVMValueRef cond; + LLVMValueRef res; + unsigned i; + + if(func == PIPE_FUNC_NEVER) + return zeros; + if(func == PIPE_FUNC_ALWAYS) + return ones; + + /* TODO: optimize the constant case */ + + /* XXX: It is not clear if we should use the ordered or unordered operators */ + +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) + if(type.width * type.length == 128) { + if(type.floating) { + LLVMValueRef args[3]; + unsigned cc; + boolean swap; + + swap = FALSE; + switch(func) { + case PIPE_FUNC_EQUAL: + cc = 0; + break; + case PIPE_FUNC_NOTEQUAL: + cc = 4; + break; + case PIPE_FUNC_LESS: + cc = 1; + break; + case PIPE_FUNC_LEQUAL: + cc = 2; + break; + case PIPE_FUNC_GREATER: + cc = 1; + swap = TRUE; + break; + case PIPE_FUNC_GEQUAL: + cc = 2; + swap = TRUE; + break; + default: + assert(0); + return bld->undef; + } + + if(swap) { + args[0] = b; + args[1] = a; + } + else { + args[0] = a; + args[1] = b; + } + + args[2] = LLVMConstInt(LLVMInt8Type(), cc, 0); + res = lp_build_intrinsic(bld->builder, + "llvm.x86.sse.cmp.ps", + vec_type, + args, 3); + res = LLVMBuildBitCast(bld->builder, res, int_vec_type, ""); + return res; + } + else { + static const struct { + unsigned swap:1; + unsigned eq:1; + unsigned gt:1; + unsigned not:1; + } table[] = { + {0, 0, 0, 1}, /* PIPE_FUNC_NEVER */ + {1, 0, 1, 0}, /* PIPE_FUNC_LESS */ + {0, 1, 0, 0}, /* PIPE_FUNC_EQUAL */ + {0, 0, 1, 1}, /* PIPE_FUNC_LEQUAL */ + {0, 0, 1, 0}, /* PIPE_FUNC_GREATER */ + {0, 1, 0, 1}, /* PIPE_FUNC_NOTEQUAL */ + {1, 0, 1, 1}, /* PIPE_FUNC_GEQUAL */ + {0, 0, 0, 0} /* PIPE_FUNC_ALWAYS */ + }; + const char *pcmpeq; + const char *pcmpgt; + LLVMValueRef args[2]; + LLVMValueRef res; + + switch (type.width) { + case 8: + pcmpeq = "llvm.x86.sse2.pcmpeq.b"; + pcmpgt = "llvm.x86.sse2.pcmpgt.b"; + break; + case 16: + pcmpeq = "llvm.x86.sse2.pcmpeq.w"; + pcmpgt = "llvm.x86.sse2.pcmpgt.w"; + break; + case 32: + pcmpeq = "llvm.x86.sse2.pcmpeq.d"; + pcmpgt = "llvm.x86.sse2.pcmpgt.d"; + break; + default: + assert(0); + return bld->undef; + } + + /* There are no signed byte and unsigned word/dword comparison + * instructions. So flip the sign bit so that the results match. + */ + if(table[func].gt && + ((type.width == 8 && type.sign) || + (type.width != 8 && !type.sign))) { + LLVMValueRef msb = lp_build_int_const_scalar(type, (unsigned long long)1 << (type.width - 1)); + a = LLVMBuildXor(bld->builder, a, msb, ""); + b = LLVMBuildXor(bld->builder, b, msb, ""); + } + + if(table[func].swap) { + args[0] = b; + args[1] = a; + } + else { + args[0] = a; + args[1] = b; + } + + if(table[func].eq) + res = lp_build_intrinsic(bld->builder, pcmpeq, vec_type, args, 2); + else if (table[func].gt) + res = lp_build_intrinsic(bld->builder, pcmpgt, vec_type, args, 2); + else + res = LLVMConstNull(vec_type); + + if(table[func].not) + res = LLVMBuildNot(bld->builder, res, ""); + + return res; + } + } +#endif + + if(type.floating) { + LLVMRealPredicate op; + switch(func) { + case PIPE_FUNC_NEVER: + op = LLVMRealPredicateFalse; + break; + case PIPE_FUNC_ALWAYS: + op = LLVMRealPredicateTrue; + break; + case PIPE_FUNC_EQUAL: + op = LLVMRealUEQ; + break; + case PIPE_FUNC_NOTEQUAL: + op = LLVMRealUNE; + break; + case PIPE_FUNC_LESS: + op = LLVMRealULT; + break; + case PIPE_FUNC_LEQUAL: + op = LLVMRealULE; + break; + case PIPE_FUNC_GREATER: + op = LLVMRealUGT; + break; + case PIPE_FUNC_GEQUAL: + op = LLVMRealUGE; + break; + default: + assert(0); + return bld->undef; + } + +#if 0 + /* XXX: Although valid IR, no LLVM target currently support this */ + cond = LLVMBuildFCmp(bld->builder, op, a, b, ""); + res = LLVMBuildSelect(bld->builder, cond, ones, zeros, ""); +#else + debug_printf("%s: warning: using slow element-wise vector comparison\n", + __FUNCTION__); + res = LLVMGetUndef(int_vec_type); + for(i = 0; i < type.length; ++i) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + cond = LLVMBuildFCmp(bld->builder, op, + LLVMBuildExtractElement(bld->builder, a, index, ""), + LLVMBuildExtractElement(bld->builder, b, index, ""), + ""); + cond = LLVMBuildSelect(bld->builder, cond, + LLVMConstExtractElement(ones, index), + LLVMConstExtractElement(zeros, index), + ""); + res = LLVMBuildInsertElement(bld->builder, res, cond, index, ""); + } +#endif + } + else { + LLVMIntPredicate op; + switch(func) { + case PIPE_FUNC_EQUAL: + op = LLVMIntEQ; + break; + case PIPE_FUNC_NOTEQUAL: + op = LLVMIntNE; + break; + case PIPE_FUNC_LESS: + op = type.sign ? LLVMIntSLT : LLVMIntULT; + break; + case PIPE_FUNC_LEQUAL: + op = type.sign ? LLVMIntSLE : LLVMIntULE; + break; + case PIPE_FUNC_GREATER: + op = type.sign ? LLVMIntSGT : LLVMIntUGT; + break; + case PIPE_FUNC_GEQUAL: + op = type.sign ? LLVMIntSGE : LLVMIntUGE; + break; + default: + assert(0); + return bld->undef; + } + +#if 0 + /* XXX: Although valid IR, no LLVM target currently support this */ + cond = LLVMBuildICmp(bld->builder, op, a, b, ""); + res = LLVMBuildSelect(bld->builder, cond, ones, zeros, ""); +#else + debug_printf("%s: warning: using slow element-wise vector comparison\n", + __FUNCTION__); + res = LLVMGetUndef(int_vec_type); + for(i = 0; i < type.length; ++i) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + cond = LLVMBuildICmp(bld->builder, op, + LLVMBuildExtractElement(bld->builder, a, index, ""), + LLVMBuildExtractElement(bld->builder, b, index, ""), + ""); + cond = LLVMBuildSelect(bld->builder, cond, + LLVMConstExtractElement(ones, index), + LLVMConstExtractElement(zeros, index), + ""); + res = LLVMBuildInsertElement(bld->builder, res, cond, index, ""); + } +#endif + } + + return res; +} + + +LLVMValueRef +lp_build_select(struct lp_build_context *bld, + LLVMValueRef mask, + LLVMValueRef a, + LLVMValueRef b) +{ + union lp_type type = bld->type; + LLVMValueRef res; + + if(a == b) + return a; + + if(type.floating) { + LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); + a = LLVMBuildBitCast(bld->builder, a, int_vec_type, ""); + b = LLVMBuildBitCast(bld->builder, b, int_vec_type, ""); + } + + /* TODO: On SSE4 we could do this with a single instruction -- PBLENDVB */ + + a = LLVMBuildAnd(bld->builder, a, mask, ""); + + /* This often gets translated to PANDN, but sometimes the NOT is + * pre-computed and stored in another constant. The best strategy depends + * on available registers, so it is not a big deal -- hopefully LLVM does + * the right decision attending the rest of the program. + */ + b = LLVMBuildAnd(bld->builder, b, LLVMBuildNot(bld->builder, mask, ""), ""); + + res = LLVMBuildOr(bld->builder, a, b, ""); + + if(type.floating) { + LLVMTypeRef vec_type = lp_build_vec_type(type); + res = LLVMBuildBitCast(bld->builder, res, vec_type, ""); + } + + return res; +} + + +LLVMValueRef +lp_build_select_aos(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b, + boolean cond[4]) +{ + const union lp_type type = bld->type; + const unsigned n = type.length; + unsigned i, j; + + if(a == b) + return a; + if(cond[0] && cond[1] && cond[2] && cond[3]) + return a; + if(!cond[0] && !cond[1] && !cond[2] && !cond[3]) + return b; + if(a == bld->undef || b == bld->undef) + return bld->undef; + + /* + * There are three major ways of accomplishing this: + * - with a shuffle, + * - with a select, + * - or with a bit mask. + * + * Select isn't supported for vector types yet. + * The flip between these is empirical and might need to be. + */ + if (n <= 4) { + /* + * Shuffle. + */ + LLVMTypeRef elem_type = LLVMInt32Type(); + LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH]; + + for(j = 0; j < n; j += 4) + for(i = 0; i < 4; ++i) + shuffles[j + i] = LLVMConstInt(elem_type, (cond[i] ? 0 : n) + j + i, 0); + + return LLVMBuildShuffleVector(bld->builder, a, b, LLVMConstVector(shuffles, n), ""); + } +#if 0 + else if(0) { + /* FIXME: Unfortunately select of vectors do not work */ + /* Use a select */ + LLVMTypeRef elem_type = LLVMInt1Type(); + LLVMValueRef cond[LP_MAX_VECTOR_LENGTH]; + + for(j = 0; j < n; j += 4) + for(i = 0; i < 4; ++i) + cond[j + i] = LLVMConstInt(elem_type, cond[i] ? 1 : 0, 0); + + return LLVMBuildSelect(bld->builder, LLVMConstVector(cond, n), a, b, ""); + } +#endif + else { + LLVMValueRef mask = lp_build_const_mask_aos(type, cond); + return lp_build_select(bld, mask, a, b); + } +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_logic.h b/src/gallium/drivers/llvmpipe/lp_bld_logic.h new file mode 100644 index 00000000000..29b9e1c45b8 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_logic.h @@ -0,0 +1,72 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Helper functions for logical operations. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#ifndef LP_BLD_LOGIC_H +#define LP_BLD_LOGIC_H + + +#include <llvm-c/Core.h> + +#include "pipe/p_defines.h" /* For PIPE_FUNC_xxx */ + + +union lp_type type; +struct lp_build_context; + + +/** + * @param func is one of PIPE_FUNC_xxx + */ +LLVMValueRef +lp_build_cmp(struct lp_build_context *bld, + unsigned func, + LLVMValueRef a, + LLVMValueRef b); + + +LLVMValueRef +lp_build_select(struct lp_build_context *bld, + LLVMValueRef mask, + LLVMValueRef a, + LLVMValueRef b); + +LLVMValueRef +lp_build_select_aos(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b, + boolean cond[4]); + + +#endif /* !LP_BLD_LOGIC_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_struct.c b/src/gallium/drivers/llvmpipe/lp_bld_struct.c new file mode 100644 index 00000000000..14d2b10df9c --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_struct.c @@ -0,0 +1,59 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * @file + * Helper functions for manipulation structures. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#include "util/u_debug.h" +#include "util/u_memory.h" + +#include "lp_bld_debug.h" +#include "lp_bld_struct.h" + + +LLVMValueRef +lp_build_struct_get(LLVMBuilderRef builder, + LLVMValueRef ptr, + unsigned member, + const char *name) +{ + LLVMValueRef indices[2]; + LLVMValueRef member_ptr; + LLVMValueRef res; + indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0); + indices[1] = LLVMConstInt(LLVMInt32Type(), member, 0); + member_ptr = LLVMBuildGEP(builder, ptr, indices, Elements(indices), ""); + res = LLVMBuildLoad(builder, member_ptr, ""); + lp_build_name(res, "%s.%s", LLVMGetValueName(ptr), name); + return res; +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_struct.h b/src/gallium/drivers/llvmpipe/lp_bld_struct.h new file mode 100644 index 00000000000..cbefdc9f815 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_struct.h @@ -0,0 +1,63 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Helper functions for type conversions. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#ifndef LP_BLD_STRUCT_H +#define LP_BLD_STRUCT_H + + +#include <llvm-c/Core.h> +#include <llvm-c/Target.h> + +#include "util/u_debug.h" +#include "util/u_memory.h" + + +#define LP_CHECK_STRUCT_SIZE(_ctype, _ltarget, _ltype) \ + assert(LLVMABISizeOfType(_ltarget, _ltype) == \ + sizeof(_ctype)) + +#define LP_CHECK_MEMBER_OFFSET(_ctype, _cmember, _ltarget, _ltype, _lindex) \ + assert(LLVMOffsetOfElement(_ltarget, _ltype, _lindex) == \ + offsetof(_ctype, _cmember)) + + +LLVMValueRef +lp_build_struct_get(LLVMBuilderRef builder, + LLVMValueRef ptr, + unsigned member, + const char *name); + + +#endif /* !LP_BLD_STRUCT_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_swizzle.c b/src/gallium/drivers/llvmpipe/lp_bld_swizzle.c new file mode 100644 index 00000000000..ac7eed9379a --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_swizzle.c @@ -0,0 +1,238 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Helper functions for swizzling/shuffling. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#include "util/u_debug.h" + +#include "lp_bld_type.h" +#include "lp_bld_const.h" +#include "lp_bld_logic.h" +#include "lp_bld_swizzle.h" + + +LLVMValueRef +lp_build_broadcast(LLVMBuilderRef builder, + LLVMTypeRef vec_type, + LLVMValueRef scalar) +{ + const unsigned n = LLVMGetVectorSize(vec_type); + LLVMValueRef res; + unsigned i; + + res = LLVMGetUndef(vec_type); + for(i = 0; i < n; ++i) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + res = LLVMBuildInsertElement(builder, res, scalar, index, ""); + } + + return res; +} + + +LLVMValueRef +lp_build_broadcast_scalar(struct lp_build_context *bld, + LLVMValueRef scalar) +{ + const union lp_type type = bld->type; + LLVMValueRef res; + unsigned i; + + res = bld->undef; + for(i = 0; i < type.length; ++i) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + res = LLVMBuildInsertElement(bld->builder, res, scalar, index, ""); + } + + return res; +} + + +LLVMValueRef +lp_build_broadcast_aos(struct lp_build_context *bld, + LLVMValueRef a, + unsigned channel) +{ + const union lp_type type = bld->type; + const unsigned n = type.length; + unsigned i, j; + + if(a == bld->undef || a == bld->zero || a == bld->one) + return a; + + /* XXX: SSE3 has PSHUFB which should be better than bitmasks, but forcing + * using shuffles here actually causes worst results. More investigation is + * needed. */ + if (n <= 4) { + /* + * Shuffle. + */ + LLVMTypeRef elem_type = LLVMInt32Type(); + LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH]; + + for(j = 0; j < n; j += 4) + for(i = 0; i < 4; ++i) + shuffles[j + i] = LLVMConstInt(elem_type, j + channel, 0); + + return LLVMBuildShuffleVector(bld->builder, a, bld->undef, LLVMConstVector(shuffles, n), ""); + } + else { + /* + * Bit mask and recursive shifts + * + * XYZW XYZW .... XYZW <= input + * 0Y00 0Y00 .... 0Y00 + * YY00 YY00 .... YY00 + * YYYY YYYY .... YYYY <= output + */ + union lp_type type4 = type; + const char shifts[4][2] = { + { 1, 2}, + {-1, 2}, + { 1, -2}, + {-1, -2} + }; + boolean cond[4]; + unsigned i; + + memset(cond, 0, sizeof cond); + cond[channel] = 1; + + a = LLVMBuildAnd(bld->builder, a, lp_build_const_mask_aos(type, cond), ""); + + type4.width *= 4; + type4.length /= 4; + + a = LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(type4), ""); + + for(i = 0; i < 2; ++i) { + LLVMValueRef tmp = NULL; + int shift = shifts[channel][i]; + +#ifdef PIPE_ARCH_LITTLE_ENDIAN + shift = -shift; +#endif + + if(shift > 0) + tmp = LLVMBuildLShr(bld->builder, a, lp_build_int_const_scalar(type4, shift*type.width), ""); + if(shift < 0) + tmp = LLVMBuildShl(bld->builder, a, lp_build_int_const_scalar(type4, -shift*type.width), ""); + + assert(tmp); + if(tmp) + a = LLVMBuildOr(bld->builder, a, tmp, ""); + } + + return LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(type), ""); + } +} + + +LLVMValueRef +lp_build_swizzle1_aos(struct lp_build_context *bld, + LLVMValueRef a, + unsigned char swizzle[4]) +{ + const unsigned n = bld->type.length; + unsigned i, j; + + if(a == bld->undef || a == bld->zero || a == bld->one) + return a; + + if(swizzle[0] == swizzle[1] && swizzle[1] == swizzle[2] && swizzle[2] == swizzle[3]) + return lp_build_broadcast_aos(bld, a, swizzle[0]); + + { + /* + * Shuffle. + */ + LLVMTypeRef elem_type = LLVMInt32Type(); + LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH]; + + for(j = 0; j < n; j += 4) + for(i = 0; i < 4; ++i) + shuffles[j + i] = LLVMConstInt(elem_type, j + swizzle[i], 0); + + return LLVMBuildShuffleVector(bld->builder, a, bld->undef, LLVMConstVector(shuffles, n), ""); + } +} + + +LLVMValueRef +lp_build_swizzle2_aos(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b, + unsigned char swizzle[4]) +{ + const unsigned n = bld->type.length; + unsigned i, j; + + if(swizzle[0] < 4 && swizzle[1] < 4 && swizzle[2] < 4 && swizzle[3] < 4) + return lp_build_swizzle1_aos(bld, a, swizzle); + + if(a == b) { + swizzle[0] %= 4; + swizzle[1] %= 4; + swizzle[2] %= 4; + swizzle[3] %= 4; + return lp_build_swizzle1_aos(bld, a, swizzle); + } + + if(swizzle[0] % 4 == 0 && + swizzle[1] % 4 == 1 && + swizzle[2] % 4 == 2 && + swizzle[3] % 4 == 3) { + boolean cond[4]; + cond[0] = swizzle[0] / 4; + cond[1] = swizzle[1] / 4; + cond[2] = swizzle[2] / 4; + cond[3] = swizzle[3] / 4; + return lp_build_select_aos(bld, a, b, cond); + } + + { + /* + * Shuffle. + */ + LLVMTypeRef elem_type = LLVMInt32Type(); + LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH]; + + for(j = 0; j < n; j += 4) + for(i = 0; i < 4; ++i) + shuffles[j + i] = LLVMConstInt(elem_type, j + (swizzle[i] % 4) + (swizzle[i] / 4 * n), 0); + + return LLVMBuildShuffleVector(bld->builder, a, b, LLVMConstVector(shuffles, n), ""); + } +} + + diff --git a/src/gallium/drivers/llvmpipe/lp_bld_swizzle.h b/src/gallium/drivers/llvmpipe/lp_bld_swizzle.h new file mode 100644 index 00000000000..d7dd6a8a604 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_swizzle.h @@ -0,0 +1,91 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Helper functions for swizzling/shuffling. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#ifndef LP_BLD_SWIZZLE_H +#define LP_BLD_SWIZZLE_H + + +#include <llvm-c/Core.h> + + +union lp_type type; +struct lp_build_context; + + +LLVMValueRef +lp_build_broadcast(LLVMBuilderRef builder, + LLVMTypeRef vec_type, + LLVMValueRef scalar); + + +LLVMValueRef +lp_build_broadcast_scalar(struct lp_build_context *bld, + LLVMValueRef scalar); + + +/** + * Broadcast one channel of a vector composed of arrays of XYZW structures into + * all four channel. + */ +LLVMValueRef +lp_build_broadcast_aos(struct lp_build_context *bld, + LLVMValueRef a, + unsigned channel); + + +/** + * Swizzle a vector consisting of an array of XYZW structs. + * + * @param swizzle is the in [0,4[ range. + */ +LLVMValueRef +lp_build_swizzle1_aos(struct lp_build_context *bld, + LLVMValueRef a, + unsigned char swizzle[4]); + + +/** + * Swizzle two vector consisting of an array of XYZW structs. + * + * @param swizzle is the in [0,8[ range. Values in [4,8[ range refer to b. + */ +LLVMValueRef +lp_build_swizzle2_aos(struct lp_build_context *bld, + LLVMValueRef a, + LLVMValueRef b, + unsigned char swizzle[4]); + + +#endif /* !LP_BLD_SWIZZLE_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_tgsi.h b/src/gallium/drivers/llvmpipe/lp_bld_tgsi.h new file mode 100644 index 00000000000..912db24aecb --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_tgsi.h @@ -0,0 +1,69 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * TGSI to LLVM IR translation. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#ifndef LP_BLD_TGSI_H +#define LP_BLD_TGSI_H + +#include <llvm-c/Core.h> + + +struct tgsi_token; +union lp_type; +struct lp_build_context; +struct lp_build_mask_context; + + +typedef void +(*lp_emit_fetch_texel_soa_callback)( LLVMBuilderRef builder, + void *context, + unsigned unit, + unsigned num_coords, + const LLVMValueRef *coords, + LLVMValueRef lodbias, + LLVMValueRef *texel); + +void +lp_build_tgsi_soa(LLVMBuilderRef builder, + const struct tgsi_token *tokens, + union lp_type type, + struct lp_build_mask_context *mask, + LLVMValueRef consts_ptr, + const LLVMValueRef *pos, + const LLVMValueRef (*inputs)[4], + LLVMValueRef (*outputs)[4], + lp_emit_fetch_texel_soa_callback emit_fetch_texel, + void *emit_fetch_texel_context); + + +#endif /* LP_BLD_TGSI_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c new file mode 100644 index 00000000000..d4d18febec7 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c @@ -0,0 +1,1346 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * TGSI to LLVM IR translation -- SoA. + * + * @author Jose Fonseca <jfonseca@vmware.com> + * + * Based on tgsi_sse2.c code written by Michal Krol, Keith Whitwell, + * Brian Paul, and others. + */ + +#include "pipe/p_config.h" +#include "pipe/p_shader_tokens.h" +#include "util/u_debug.h" +#include "util/u_math.h" +#include "util/u_memory.h" +#include "tgsi/tgsi_info.h" +#include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_util.h" +#include "tgsi/tgsi_exec.h" +#include "lp_bld_type.h" +#include "lp_bld_const.h" +#include "lp_bld_intr.h" +#include "lp_bld_arit.h" +#include "lp_bld_logic.h" +#include "lp_bld_swizzle.h" +#include "lp_bld_flow.h" +#include "lp_bld_tgsi.h" +#include "lp_bld_debug.h" + + +#define LP_MAX_TEMPS 256 +#define LP_MAX_IMMEDIATES 256 + + +#define FOR_EACH_CHANNEL( CHAN )\ + for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++) + +#define IS_DST0_CHANNEL_ENABLED( INST, CHAN )\ + ((INST)->FullDstRegisters[0].DstRegister.WriteMask & (1 << (CHAN))) + +#define IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )\ + if (IS_DST0_CHANNEL_ENABLED( INST, CHAN )) + +#define FOR_EACH_DST0_ENABLED_CHANNEL( INST, CHAN )\ + FOR_EACH_CHANNEL( CHAN )\ + IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN ) + +#define CHAN_X 0 +#define CHAN_Y 1 +#define CHAN_Z 2 +#define CHAN_W 3 + + +struct lp_build_tgsi_soa_context +{ + struct lp_build_context base; + + LLVMValueRef consts_ptr; + const LLVMValueRef *pos; + const LLVMValueRef (*inputs)[NUM_CHANNELS]; + LLVMValueRef (*outputs)[NUM_CHANNELS]; + + lp_emit_fetch_texel_soa_callback emit_fetch_texel; + void *emit_fetch_texel_context; + + LLVMValueRef immediates[LP_MAX_IMMEDIATES][NUM_CHANNELS]; + LLVMValueRef temps[LP_MAX_TEMPS][NUM_CHANNELS]; + + struct lp_build_mask_context *mask; +}; + + +/** + * Register fetch. + */ +static LLVMValueRef +emit_fetch( + struct lp_build_tgsi_soa_context *bld, + const struct tgsi_full_instruction *inst, + unsigned index, + const unsigned chan_index ) +{ + const struct tgsi_full_src_register *reg = &inst->FullSrcRegisters[index]; + unsigned swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index ); + LLVMValueRef res; + + switch (swizzle) { + case TGSI_EXTSWIZZLE_X: + case TGSI_EXTSWIZZLE_Y: + case TGSI_EXTSWIZZLE_Z: + case TGSI_EXTSWIZZLE_W: + + switch (reg->SrcRegister.File) { + case TGSI_FILE_CONSTANT: { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), reg->SrcRegister.Index*4 + swizzle, 0); + LLVMValueRef scalar_ptr = LLVMBuildGEP(bld->base.builder, bld->consts_ptr, &index, 1, ""); + LLVMValueRef scalar = LLVMBuildLoad(bld->base.builder, scalar_ptr, ""); + res = lp_build_broadcast_scalar(&bld->base, scalar); + break; + } + + case TGSI_FILE_IMMEDIATE: + res = bld->immediates[reg->SrcRegister.Index][swizzle]; + assert(res); + break; + + case TGSI_FILE_INPUT: + res = bld->inputs[reg->SrcRegister.Index][swizzle]; + assert(res); + break; + + case TGSI_FILE_TEMPORARY: + res = bld->temps[reg->SrcRegister.Index][swizzle]; + if(!res) + return bld->base.undef; + break; + + default: + assert( 0 ); + return bld->base.undef; + } + break; + + case TGSI_EXTSWIZZLE_ZERO: + res = bld->base.zero; + break; + + case TGSI_EXTSWIZZLE_ONE: + res = bld->base.one; + break; + + default: + assert( 0 ); + return bld->base.undef; + } + + switch( tgsi_util_get_full_src_register_sign_mode( reg, chan_index ) ) { + case TGSI_UTIL_SIGN_CLEAR: + res = lp_build_abs( &bld->base, res ); + break; + + case TGSI_UTIL_SIGN_SET: + res = lp_build_abs( &bld->base, res ); + res = LLVMBuildNeg( bld->base.builder, res, "" ); + break; + + case TGSI_UTIL_SIGN_TOGGLE: + res = LLVMBuildNeg( bld->base.builder, res, "" ); + break; + + case TGSI_UTIL_SIGN_KEEP: + break; + } + + return res; +} + + +/** + * Register store. + */ +static void +emit_store( + struct lp_build_tgsi_soa_context *bld, + const struct tgsi_full_instruction *inst, + unsigned index, + unsigned chan_index, + LLVMValueRef value) +{ + const struct tgsi_full_dst_register *reg = &inst->FullDstRegisters[index]; + + switch( inst->Instruction.Saturate ) { + case TGSI_SAT_NONE: + break; + + case TGSI_SAT_ZERO_ONE: + value = lp_build_max(&bld->base, value, bld->base.zero); + value = lp_build_min(&bld->base, value, bld->base.one); + break; + + case TGSI_SAT_MINUS_PLUS_ONE: + value = lp_build_max(&bld->base, value, lp_build_const_scalar(bld->base.type, -1.0)); + value = lp_build_min(&bld->base, value, bld->base.one); + break; + + default: + assert(0); + } + + switch( reg->DstRegister.File ) { + case TGSI_FILE_OUTPUT: + bld->outputs[reg->DstRegister.Index][chan_index] = value; + break; + + case TGSI_FILE_TEMPORARY: + bld->temps[reg->DstRegister.Index][chan_index] = value; + break; + + case TGSI_FILE_ADDRESS: + /* FIXME */ + assert(0); + break; + + default: + assert( 0 ); + } +} + + +/** + * High-level instruction translators. + */ + +static void +emit_tex( struct lp_build_tgsi_soa_context *bld, + const struct tgsi_full_instruction *inst, + boolean apply_lodbias, + boolean projected) +{ + const uint unit = inst->FullSrcRegisters[1].SrcRegister.Index; + LLVMValueRef lodbias; + LLVMValueRef oow; + LLVMValueRef coords[3]; + LLVMValueRef texel[4]; + unsigned num_coords; + unsigned i; + + switch (inst->InstructionExtTexture.Texture) { + case TGSI_TEXTURE_1D: + num_coords = 1; + break; + case TGSI_TEXTURE_2D: + case TGSI_TEXTURE_RECT: + num_coords = 2; + break; + case TGSI_TEXTURE_SHADOW1D: + case TGSI_TEXTURE_SHADOW2D: + case TGSI_TEXTURE_SHADOWRECT: + case TGSI_TEXTURE_3D: + case TGSI_TEXTURE_CUBE: + num_coords = 3; + break; + default: + assert(0); + return; + } + + if(apply_lodbias) + lodbias = emit_fetch( bld, inst, 0, 3 ); + else + lodbias = bld->base.zero; + + if (projected) { + oow = emit_fetch( bld, inst, 0, 3 ); + oow = lp_build_rcp(&bld->base, oow); + } + + for (i = 0; i < num_coords; i++) { + coords[i] = emit_fetch( bld, inst, 0, i ); + if (projected) + coords[i] = lp_build_mul(&bld->base, coords[i], oow); + } + + bld->emit_fetch_texel(bld->base.builder, bld->emit_fetch_texel_context, + unit, num_coords, coords, lodbias, texel); + + FOR_EACH_DST0_ENABLED_CHANNEL( inst, i ) { + emit_store( bld, inst, 0, i, texel[i] ); + } +} + + +static void +emit_kil( + struct lp_build_tgsi_soa_context *bld, + const struct tgsi_full_instruction *inst ) +{ + const struct tgsi_full_src_register *reg = &inst->FullSrcRegisters[0]; + LLVMValueRef terms[NUM_CHANNELS]; + LLVMValueRef mask; + unsigned chan_index; + + memset(&terms, 0, sizeof terms); + + FOR_EACH_CHANNEL( chan_index ) { + unsigned swizzle; + + /* Unswizzle channel */ + swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index ); + + /* Note that we test if the value is less than zero, so 1.0 and 0.0 need + * not to be tested. */ + if(swizzle == TGSI_EXTSWIZZLE_ZERO || swizzle == TGSI_EXTSWIZZLE_ONE) + continue; + + /* Check if the component has not been already tested. */ + assert(swizzle < NUM_CHANNELS); + if( !terms[swizzle] ) + /* TODO: change the comparison operator instead of setting the sign */ + terms[swizzle] = emit_fetch(bld, inst, 0, chan_index ); + } + + mask = NULL; + FOR_EACH_CHANNEL( chan_index ) { + if(terms[chan_index]) { + LLVMValueRef chan_mask; + + chan_mask = lp_build_cmp(&bld->base, PIPE_FUNC_GEQUAL, terms[chan_index], bld->base.zero); + + if(mask) + mask = LLVMBuildAnd(bld->base.builder, mask, chan_mask, ""); + else + mask = chan_mask; + } + } + + if(mask) + lp_build_mask_update(bld->mask, mask); +} + + +static void +emit_kilp( + struct lp_build_tgsi_soa_context *bld ) +{ + /* XXX todo / fix me */ +} + + +/** + * Check if inst src/dest regs use indirect addressing into temporary + * register file. + */ +static boolean +indirect_temp_reference(const struct tgsi_full_instruction *inst) +{ + uint i; + for (i = 0; i < inst->Instruction.NumSrcRegs; i++) { + const struct tgsi_full_src_register *reg = &inst->FullSrcRegisters[i]; + if (reg->SrcRegister.File == TGSI_FILE_TEMPORARY && + reg->SrcRegister.Indirect) + return TRUE; + } + for (i = 0; i < inst->Instruction.NumDstRegs; i++) { + const struct tgsi_full_dst_register *reg = &inst->FullDstRegisters[i]; + if (reg->DstRegister.File == TGSI_FILE_TEMPORARY && + reg->DstRegister.Indirect) + return TRUE; + } + return FALSE; +} + + +static int +emit_instruction( + struct lp_build_tgsi_soa_context *bld, + struct tgsi_full_instruction *inst ) +{ + unsigned chan_index; + LLVMValueRef src0, src1, src2; + LLVMValueRef tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + LLVMValueRef dst0; + + /* we can't handle indirect addressing into temp register file yet */ + if (indirect_temp_reference(inst)) + return FALSE; + + switch (inst->Instruction.Opcode) { +#if 0 + case TGSI_OPCODE_ARL: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + tmp0 = emit_fetch( bld, inst, 0, chan_index ); + emit_flr(bld, 0, 0); + emit_f2it( bld, 0 ); + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; +#endif + + case TGSI_OPCODE_MOV: + case TGSI_OPCODE_SWZ: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + tmp0 = emit_fetch( bld, inst, 0, chan_index ); + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + + case TGSI_OPCODE_LIT: + if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) ) { + emit_store( bld, inst, 0, CHAN_X, bld->base.one); + } + if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) ) { + src0 = emit_fetch( bld, inst, 0, CHAN_X ); + dst0 = lp_build_max( &bld->base, src0, bld->base.zero); + emit_store( bld, inst, 0, CHAN_Y, dst0); + } + if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) ) { + /* XMM[1] = SrcReg[0].yyyy */ + tmp1 = emit_fetch( bld, inst, 0, CHAN_Y ); + /* XMM[1] = max(XMM[1], 0) */ + tmp1 = lp_build_max( &bld->base, tmp1, bld->base.zero); + /* XMM[2] = SrcReg[0].wwww */ + tmp2 = emit_fetch( bld, inst, 0, CHAN_W ); + tmp1 = lp_build_pow( &bld->base, tmp1, tmp2); + tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); + tmp2 = lp_build_cmp(&bld->base, PIPE_FUNC_GREATER, tmp0, bld->base.zero); + dst0 = lp_build_select(&bld->base, tmp2, tmp1, bld->base.zero); + emit_store( bld, inst, 0, CHAN_Z, dst0); + } + if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_W ) ) { + emit_store( bld, inst, 0, CHAN_W, bld->base.one); + } + break; + + case TGSI_OPCODE_RCP: + /* TGSI_OPCODE_RECIP */ + src0 = emit_fetch( bld, inst, 0, CHAN_X ); + dst0 = lp_build_rcp(&bld->base, src0); + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + emit_store( bld, inst, 0, chan_index, dst0 ); + } + break; + + case TGSI_OPCODE_RSQ: + /* TGSI_OPCODE_RECIPSQRT */ + src0 = emit_fetch( bld, inst, 0, CHAN_X ); + src0 = lp_build_abs(&bld->base, src0); + dst0 = lp_build_rsqrt(&bld->base, src0); + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + emit_store( bld, inst, 0, chan_index, dst0 ); + } + break; + + case TGSI_OPCODE_EXP: + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) || + IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) || + IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z )) { + LLVMValueRef *p_exp2_int_part = NULL; + LLVMValueRef *p_frac_part = NULL; + LLVMValueRef *p_exp2 = NULL; + + src0 = emit_fetch( bld, inst, 0, CHAN_X ); + + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X )) + p_exp2_int_part = &tmp0; + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y )) + p_frac_part = &tmp1; + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z )) + p_exp2 = &tmp2; + + lp_build_exp2_approx(&bld->base, src0, p_exp2_int_part, p_frac_part, p_exp2); + + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X )) + emit_store( bld, inst, 0, CHAN_X, tmp0); + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y )) + emit_store( bld, inst, 0, CHAN_Y, tmp1); + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z )) + emit_store( bld, inst, 0, CHAN_Z, tmp2); + } + /* dst.w = 1.0 */ + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_W )) { + tmp0 = bld->base.one; + emit_store( bld, inst, 0, CHAN_W, tmp0); + } + break; + + case TGSI_OPCODE_LOG: + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) || + IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) || + IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z )) { + LLVMValueRef *p_floor_log2; + LLVMValueRef *p_exp; + LLVMValueRef *p_log2; + + src0 = emit_fetch( bld, inst, 0, CHAN_X ); + src0 = lp_build_abs( &bld->base, src0 ); + + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X )) + p_floor_log2 = &tmp0; + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y )) + p_exp = &tmp1; + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z )) + p_log2 = &tmp2; + + lp_build_log2_approx(&bld->base, src0, p_exp, p_floor_log2, p_log2); + + /* dst.x = floor(lg2(abs(src.x))) */ + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X )) + emit_store( bld, inst, 0, CHAN_X, tmp0); + /* dst.y = abs(src)/ex2(floor(lg2(abs(src.x)))) */ + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y )) { + tmp1 = lp_build_div( &bld->base, src0, tmp1); + emit_store( bld, inst, 0, CHAN_Y, tmp1); + } + /* dst.z = lg2(abs(src.x)) */ + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z )) + emit_store( bld, inst, 0, CHAN_Z, tmp2); + } + /* dst.w = 1.0 */ + if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_W )) { + tmp0 = bld->base.one; + emit_store( bld, inst, 0, CHAN_W, tmp0); + } + break; + + case TGSI_OPCODE_MUL: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + src0 = emit_fetch( bld, inst, 0, chan_index ); + src1 = emit_fetch( bld, inst, 1, chan_index ); + dst0 = lp_build_mul(&bld->base, src0, src1); + emit_store( bld, inst, 0, chan_index, dst0); + } + break; + + case TGSI_OPCODE_ADD: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + src0 = emit_fetch( bld, inst, 0, chan_index ); + src1 = emit_fetch( bld, inst, 1, chan_index ); + dst0 = lp_build_add(&bld->base, src0, src1); + emit_store( bld, inst, 0, chan_index, dst0); + } + break; + + case TGSI_OPCODE_DP3: + /* TGSI_OPCODE_DOT3 */ + tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); + tmp1 = emit_fetch( bld, inst, 1, CHAN_X ); + tmp0 = lp_build_mul( &bld->base, tmp0, tmp1); + tmp1 = emit_fetch( bld, inst, 0, CHAN_Y ); + tmp2 = emit_fetch( bld, inst, 1, CHAN_Y ); + tmp1 = lp_build_mul( &bld->base, tmp1, tmp2); + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); + tmp1 = emit_fetch( bld, inst, 0, CHAN_Z ); + tmp2 = emit_fetch( bld, inst, 1, CHAN_Z ); + tmp1 = lp_build_mul( &bld->base, tmp1, tmp2); + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + + case TGSI_OPCODE_DP4: + /* TGSI_OPCODE_DOT4 */ + tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); + tmp1 = emit_fetch( bld, inst, 1, CHAN_X ); + tmp0 = lp_build_mul( &bld->base, tmp0, tmp1); + tmp1 = emit_fetch( bld, inst, 0, CHAN_Y ); + tmp2 = emit_fetch( bld, inst, 1, CHAN_Y ); + tmp1 = lp_build_mul( &bld->base, tmp1, tmp2); + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); + tmp1 = emit_fetch( bld, inst, 0, CHAN_Z ); + tmp2 = emit_fetch( bld, inst, 1, CHAN_Z ); + tmp1 = lp_build_mul( &bld->base, tmp1, tmp2); + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); + tmp1 = emit_fetch( bld, inst, 0, CHAN_W ); + tmp2 = emit_fetch( bld, inst, 1, CHAN_W ); + tmp1 = lp_build_mul( &bld->base, tmp1, tmp2); + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + + case TGSI_OPCODE_DST: + IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) { + tmp0 = bld->base.one; + emit_store( bld, inst, 0, CHAN_X, tmp0); + } + IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) { + tmp0 = emit_fetch( bld, inst, 0, CHAN_Y ); + tmp1 = emit_fetch( bld, inst, 1, CHAN_Y ); + tmp0 = lp_build_mul( &bld->base, tmp0, tmp1); + emit_store( bld, inst, 0, CHAN_Y, tmp0); + } + IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) { + tmp0 = emit_fetch( bld, inst, 0, CHAN_Z ); + emit_store( bld, inst, 0, CHAN_Z, tmp0); + } + IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_W ) { + tmp0 = emit_fetch( bld, inst, 1, CHAN_W ); + emit_store( bld, inst, 0, CHAN_W, tmp0); + } + break; + + case TGSI_OPCODE_MIN: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + src0 = emit_fetch( bld, inst, 0, chan_index ); + src1 = emit_fetch( bld, inst, 1, chan_index ); + dst0 = lp_build_min( &bld->base, src0, src1 ); + emit_store( bld, inst, 0, chan_index, dst0); + } + break; + + case TGSI_OPCODE_MAX: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + src0 = emit_fetch( bld, inst, 0, chan_index ); + src1 = emit_fetch( bld, inst, 1, chan_index ); + dst0 = lp_build_max( &bld->base, src0, src1 ); + emit_store( bld, inst, 0, chan_index, dst0); + } + break; + + case TGSI_OPCODE_SLT: + /* TGSI_OPCODE_SETLT */ + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + src0 = emit_fetch( bld, inst, 0, chan_index ); + src1 = emit_fetch( bld, inst, 1, chan_index ); + tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_LESS, src0, src1 ); + dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero ); + emit_store( bld, inst, 0, chan_index, dst0); + } + break; + + case TGSI_OPCODE_SGE: + /* TGSI_OPCODE_SETGE */ + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + src0 = emit_fetch( bld, inst, 0, chan_index ); + src1 = emit_fetch( bld, inst, 1, chan_index ); + tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_GEQUAL, src0, src1 ); + dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero ); + emit_store( bld, inst, 0, chan_index, dst0); + } + break; + + case TGSI_OPCODE_MAD: + /* TGSI_OPCODE_MADD */ + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + tmp0 = emit_fetch( bld, inst, 0, chan_index ); + tmp1 = emit_fetch( bld, inst, 1, chan_index ); + tmp2 = emit_fetch( bld, inst, 2, chan_index ); + tmp0 = lp_build_mul( &bld->base, tmp0, tmp1); + tmp0 = lp_build_add( &bld->base, tmp0, tmp2); + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + + case TGSI_OPCODE_SUB: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + tmp0 = emit_fetch( bld, inst, 0, chan_index ); + tmp1 = emit_fetch( bld, inst, 1, chan_index ); + tmp0 = lp_build_sub( &bld->base, tmp0, tmp1); + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + + case TGSI_OPCODE_LRP: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + src0 = emit_fetch( bld, inst, 0, chan_index ); + src1 = emit_fetch( bld, inst, 1, chan_index ); + src2 = emit_fetch( bld, inst, 2, chan_index ); + tmp0 = lp_build_sub( &bld->base, src1, src2 ); + tmp0 = lp_build_mul( &bld->base, src0, tmp0 ); + dst0 = lp_build_add( &bld->base, tmp0, src2 ); + emit_store( bld, inst, 0, chan_index, dst0 ); + } + break; + + case TGSI_OPCODE_CND: + return 0; + break; + + case TGSI_OPCODE_DP2A: + tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); /* xmm0 = src[0].x */ + tmp1 = emit_fetch( bld, inst, 1, CHAN_X ); /* xmm1 = src[1].x */ + tmp0 = lp_build_mul( &bld->base, tmp0, tmp1); /* xmm0 = xmm0 * xmm1 */ + tmp1 = emit_fetch( bld, inst, 0, CHAN_Y ); /* xmm1 = src[0].y */ + tmp2 = emit_fetch( bld, inst, 1, CHAN_Y ); /* xmm2 = src[1].y */ + tmp1 = lp_build_mul( &bld->base, tmp1, tmp2); /* xmm1 = xmm1 * xmm2 */ + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); /* xmm0 = xmm0 + xmm1 */ + tmp1 = emit_fetch( bld, inst, 2, CHAN_X ); /* xmm1 = src[2].x */ + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); /* xmm0 = xmm0 + xmm1 */ + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + emit_store( bld, inst, 0, chan_index, tmp0); /* dest[ch] = xmm0 */ + } + break; + +#if 0 + case TGSI_OPCODE_FRC: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + tmp0 = emit_fetch( bld, inst, 0, chan_index ); + emit_frc( bld, 0, 0 ); + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + + case TGSI_OPCODE_CLAMP: + return 0; + break; + + case TGSI_OPCODE_FLR: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + tmp0 = emit_fetch( bld, inst, 0, chan_index ); + emit_flr( bld, 0, 0 ); + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + + case TGSI_OPCODE_ROUND: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + tmp0 = emit_fetch( bld, inst, 0, chan_index ); + emit_rnd( bld, 0, 0 ); + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; +#endif + + case TGSI_OPCODE_EX2: { + tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); + tmp0 = lp_build_exp2( &bld->base, tmp0); + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + } + + case TGSI_OPCODE_LG2: + tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); + tmp0 = lp_build_log2( &bld->base, tmp0); + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + + case TGSI_OPCODE_POW: + src0 = emit_fetch( bld, inst, 0, CHAN_X ); + src1 = emit_fetch( bld, inst, 1, CHAN_X ); + dst0 = lp_build_pow( &bld->base, src0, src1 ); + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + emit_store( bld, inst, 0, chan_index, dst0 ); + } + break; + + case TGSI_OPCODE_XPD: + if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) || + IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) ) { + tmp1 = emit_fetch( bld, inst, 1, CHAN_Z ); + tmp3 = emit_fetch( bld, inst, 0, CHAN_Z ); + } + if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) || + IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) ) { + tmp0 = emit_fetch( bld, inst, 0, CHAN_Y ); + tmp4 = emit_fetch( bld, inst, 1, CHAN_Y ); + } + IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) { + tmp2 = tmp0; + tmp2 = lp_build_mul( &bld->base, tmp2, tmp1); + tmp5 = tmp3; + tmp5 = lp_build_mul( &bld->base, tmp5, tmp4); + tmp2 = lp_build_sub( &bld->base, tmp2, tmp5); + emit_store( bld, inst, 0, CHAN_X, tmp2); + } + if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) || + IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) ) { + tmp2 = emit_fetch( bld, inst, 1, CHAN_X ); + tmp5 = emit_fetch( bld, inst, 0, CHAN_X ); + } + IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) { + tmp3 = lp_build_mul( &bld->base, tmp3, tmp2); + tmp1 = lp_build_mul( &bld->base, tmp1, tmp5); + tmp3 = lp_build_sub( &bld->base, tmp3, tmp1); + emit_store( bld, inst, 0, CHAN_Y, tmp3); + } + IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) { + tmp5 = lp_build_mul( &bld->base, tmp5, tmp4); + tmp0 = lp_build_mul( &bld->base, tmp0, tmp2); + tmp5 = lp_build_sub( &bld->base, tmp5, tmp0); + emit_store( bld, inst, 0, CHAN_Z, tmp5); + } + IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_W ) { + tmp0 = bld->base.one; + emit_store( bld, inst, 0, CHAN_W, tmp0); + } + break; + + case TGSI_OPCODE_ABS: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + tmp0 = emit_fetch( bld, inst, 0, chan_index ); + tmp0 = lp_build_abs( &bld->base, tmp0 ) ; + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + + case TGSI_OPCODE_RCC: + return 0; + break; + + case TGSI_OPCODE_DPH: + tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); + tmp1 = emit_fetch( bld, inst, 1, CHAN_X ); + tmp0 = lp_build_mul( &bld->base, tmp0, tmp1); + tmp1 = emit_fetch( bld, inst, 0, CHAN_Y ); + tmp2 = emit_fetch( bld, inst, 1, CHAN_Y ); + tmp1 = lp_build_mul( &bld->base, tmp1, tmp2); + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); + tmp1 = emit_fetch( bld, inst, 0, CHAN_Z ); + tmp2 = emit_fetch( bld, inst, 1, CHAN_Z ); + tmp1 = lp_build_mul( &bld->base, tmp1, tmp2); + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); + tmp1 = emit_fetch( bld, inst, 1, CHAN_W ); + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + + case TGSI_OPCODE_COS: + tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); + tmp0 = lp_build_cos( &bld->base, tmp0 ); + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + + case TGSI_OPCODE_DDX: + return 0; + break; + + case TGSI_OPCODE_DDY: + return 0; + break; + +#if 0 + case TGSI_OPCODE_KILP: + /* predicated kill */ + emit_kilp( bld ); + return 0; /* XXX fix me */ + break; +#endif + + case TGSI_OPCODE_KIL: + /* conditional kill */ + emit_kil( bld, inst ); + break; + + case TGSI_OPCODE_PK2H: + return 0; + break; + + case TGSI_OPCODE_PK2US: + return 0; + break; + + case TGSI_OPCODE_PK4B: + return 0; + break; + + case TGSI_OPCODE_PK4UB: + return 0; + break; + + case TGSI_OPCODE_RFL: + return 0; + break; + + case TGSI_OPCODE_SEQ: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + src0 = emit_fetch( bld, inst, 0, chan_index ); + src1 = emit_fetch( bld, inst, 1, chan_index ); + tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_EQUAL, src0, src1 ); + dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero ); + emit_store( bld, inst, 0, chan_index, dst0); + } + break; + + case TGSI_OPCODE_SFL: + return 0; + break; + + case TGSI_OPCODE_SGT: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + src0 = emit_fetch( bld, inst, 0, chan_index ); + src1 = emit_fetch( bld, inst, 1, chan_index ); + tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_GREATER, src0, src1 ); + dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero ); + emit_store( bld, inst, 0, chan_index, dst0); + } + break; + + case TGSI_OPCODE_SIN: + tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); + tmp0 = lp_build_sin( &bld->base, tmp0 ); + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; + + case TGSI_OPCODE_SLE: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + src0 = emit_fetch( bld, inst, 0, chan_index ); + src1 = emit_fetch( bld, inst, 1, chan_index ); + tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_LEQUAL, src0, src1 ); + dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero ); + emit_store( bld, inst, 0, chan_index, dst0); + } + break; + + case TGSI_OPCODE_SNE: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + src0 = emit_fetch( bld, inst, 0, chan_index ); + src1 = emit_fetch( bld, inst, 1, chan_index ); + tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_NOTEQUAL, src0, src1 ); + dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero ); + emit_store( bld, inst, 0, chan_index, dst0); + } + break; + + case TGSI_OPCODE_STR: + return 0; + break; + + case TGSI_OPCODE_TEX: + emit_tex( bld, inst, FALSE, FALSE ); + break; + + case TGSI_OPCODE_TXD: + return 0; + break; + + case TGSI_OPCODE_UP2H: + return 0; + break; + + case TGSI_OPCODE_UP2US: + return 0; + break; + + case TGSI_OPCODE_UP4B: + return 0; + break; + + case TGSI_OPCODE_UP4UB: + return 0; + break; + + case TGSI_OPCODE_X2D: + return 0; + break; + + case TGSI_OPCODE_ARA: + return 0; + break; + +#if 0 + case TGSI_OPCODE_ARR: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + tmp0 = emit_fetch( bld, inst, 0, chan_index ); + emit_rnd( bld, 0, 0 ); + emit_f2it( bld, 0 ); + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; +#endif + + case TGSI_OPCODE_BRA: + return 0; + break; + + case TGSI_OPCODE_CAL: + return 0; + break; + +#if 0 + case TGSI_OPCODE_RET: + emit_ret( bld ); + break; +#endif + + case TGSI_OPCODE_END: + break; + +#if 0 + case TGSI_OPCODE_SSG: + /* TGSI_OPCODE_SGN */ + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + tmp0 = emit_fetch( bld, inst, 0, chan_index ); + emit_sgn( bld, 0, 0 ); + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; +#endif + + case TGSI_OPCODE_CMP: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + src0 = emit_fetch( bld, inst, 0, chan_index ); + src1 = emit_fetch( bld, inst, 1, chan_index ); + src2 = emit_fetch( bld, inst, 2, chan_index ); + tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_LESS, src0, bld->base.zero ); + dst0 = lp_build_select( &bld->base, tmp0, src1, src2); + emit_store( bld, inst, 0, chan_index, dst0); + } + break; + + case TGSI_OPCODE_SCS: + IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) { + tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); + tmp0 = lp_build_cos( &bld->base, tmp0 ); + emit_store( bld, inst, 0, CHAN_X, tmp0); + } + IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) { + tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); + tmp0 = lp_build_sin( &bld->base, tmp0 ); + emit_store( bld, inst, 0, CHAN_Y, tmp0); + } + IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) { + tmp0 = bld->base.zero; + emit_store( bld, inst, 0, CHAN_Z, tmp0); + } + IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_W ) { + tmp0 = bld->base.one; + emit_store( bld, inst, 0, CHAN_W, tmp0); + } + break; + + case TGSI_OPCODE_TXB: + emit_tex( bld, inst, TRUE, FALSE ); + break; + + case TGSI_OPCODE_NRM: + /* fall-through */ + case TGSI_OPCODE_NRM4: + /* 3 or 4-component normalization */ + { + uint dims = (inst->Instruction.Opcode == TGSI_OPCODE_NRM) ? 3 : 4; + + if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_X) || + IS_DST0_CHANNEL_ENABLED(inst, CHAN_Y) || + IS_DST0_CHANNEL_ENABLED(inst, CHAN_Z) || + (IS_DST0_CHANNEL_ENABLED(inst, CHAN_W) && dims == 4)) { + + /* NOTE: Cannot use xmm regs 2/3 here (see emit_rsqrt() above). */ + + /* xmm4 = src.x */ + /* xmm0 = src.x * src.x */ + tmp0 = emit_fetch(bld, inst, 0, CHAN_X); + if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_X)) { + tmp4 = tmp0; + } + tmp0 = lp_build_mul( &bld->base, tmp0, tmp0); + + /* xmm5 = src.y */ + /* xmm0 = xmm0 + src.y * src.y */ + tmp1 = emit_fetch(bld, inst, 0, CHAN_Y); + if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_Y)) { + tmp5 = tmp1; + } + tmp1 = lp_build_mul( &bld->base, tmp1, tmp1); + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); + + /* xmm6 = src.z */ + /* xmm0 = xmm0 + src.z * src.z */ + tmp1 = emit_fetch(bld, inst, 0, CHAN_Z); + if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_Z)) { + tmp6 = tmp1; + } + tmp1 = lp_build_mul( &bld->base, tmp1, tmp1); + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); + + if (dims == 4) { + /* xmm7 = src.w */ + /* xmm0 = xmm0 + src.w * src.w */ + tmp1 = emit_fetch(bld, inst, 0, CHAN_W); + if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_W)) { + tmp7 = tmp1; + } + tmp1 = lp_build_mul( &bld->base, tmp1, tmp1); + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); + } + + /* xmm1 = 1 / sqrt(xmm0) */ + tmp1 = lp_build_rsqrt( &bld->base, tmp0); + + /* dst.x = xmm1 * src.x */ + if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_X)) { + tmp4 = lp_build_mul( &bld->base, tmp4, tmp1); + emit_store(bld, inst, 0, CHAN_X, tmp4); + } + + /* dst.y = xmm1 * src.y */ + if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_Y)) { + tmp5 = lp_build_mul( &bld->base, tmp5, tmp1); + emit_store(bld, inst, 0, CHAN_Y, tmp5); + } + + /* dst.z = xmm1 * src.z */ + if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_Z)) { + tmp6 = lp_build_mul( &bld->base, tmp6, tmp1); + emit_store(bld, inst, 0, CHAN_Z, tmp6); + } + + /* dst.w = xmm1 * src.w */ + if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_X) && dims == 4) { + tmp7 = lp_build_mul( &bld->base, tmp7, tmp1); + emit_store(bld, inst, 0, CHAN_W, tmp7); + } + } + + /* dst0.w = 1.0 */ + if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_W) && dims == 3) { + tmp0 = bld->base.one; + emit_store(bld, inst, 0, CHAN_W, tmp0); + } + } + break; + + case TGSI_OPCODE_DIV: + return 0; + break; + + case TGSI_OPCODE_DP2: + tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); /* xmm0 = src[0].x */ + tmp1 = emit_fetch( bld, inst, 1, CHAN_X ); /* xmm1 = src[1].x */ + tmp0 = lp_build_mul( &bld->base, tmp0, tmp1); /* xmm0 = xmm0 * xmm1 */ + tmp1 = emit_fetch( bld, inst, 0, CHAN_Y ); /* xmm1 = src[0].y */ + tmp2 = emit_fetch( bld, inst, 1, CHAN_Y ); /* xmm2 = src[1].y */ + tmp1 = lp_build_mul( &bld->base, tmp1, tmp2); /* xmm1 = xmm1 * xmm2 */ + tmp0 = lp_build_add( &bld->base, tmp0, tmp1); /* xmm0 = xmm0 + xmm1 */ + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + emit_store( bld, inst, 0, chan_index, tmp0); /* dest[ch] = xmm0 */ + } + break; + + case TGSI_OPCODE_TXL: + emit_tex( bld, inst, TRUE, FALSE ); + break; + + case TGSI_OPCODE_TXP: + emit_tex( bld, inst, FALSE, TRUE ); + break; + + case TGSI_OPCODE_BRK: + return 0; + break; + + case TGSI_OPCODE_IF: + return 0; + break; + + case TGSI_OPCODE_BGNFOR: + return 0; + break; + + case TGSI_OPCODE_REP: + return 0; + break; + + case TGSI_OPCODE_ELSE: + return 0; + break; + + case TGSI_OPCODE_ENDIF: + return 0; + break; + + case TGSI_OPCODE_ENDFOR: + return 0; + break; + + case TGSI_OPCODE_ENDREP: + return 0; + break; + + case TGSI_OPCODE_PUSHA: + return 0; + break; + + case TGSI_OPCODE_POPA: + return 0; + break; + + case TGSI_OPCODE_CEIL: + return 0; + break; + + case TGSI_OPCODE_I2F: + return 0; + break; + + case TGSI_OPCODE_NOT: + return 0; + break; + +#if 0 + case TGSI_OPCODE_TRUNC: + FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + tmp0 = emit_fetch( bld, inst, 0, chan_index ); + emit_f2it( bld, 0 ); + emit_i2f( bld, 0 ); + emit_store( bld, inst, 0, chan_index, tmp0); + } + break; +#endif + + case TGSI_OPCODE_SHL: + return 0; + break; + + case TGSI_OPCODE_SHR: + return 0; + break; + + case TGSI_OPCODE_AND: + return 0; + break; + + case TGSI_OPCODE_OR: + return 0; + break; + + case TGSI_OPCODE_MOD: + return 0; + break; + + case TGSI_OPCODE_XOR: + return 0; + break; + + case TGSI_OPCODE_SAD: + return 0; + break; + + case TGSI_OPCODE_TXF: + return 0; + break; + + case TGSI_OPCODE_TXQ: + return 0; + break; + + case TGSI_OPCODE_CONT: + return 0; + break; + + case TGSI_OPCODE_EMIT: + return 0; + break; + + case TGSI_OPCODE_ENDPRIM: + return 0; + break; + + default: + return 0; + } + + return 1; +} + + +void +lp_build_tgsi_soa(LLVMBuilderRef builder, + const struct tgsi_token *tokens, + union lp_type type, + struct lp_build_mask_context *mask, + LLVMValueRef consts_ptr, + const LLVMValueRef *pos, + const LLVMValueRef (*inputs)[NUM_CHANNELS], + LLVMValueRef (*outputs)[NUM_CHANNELS], + lp_emit_fetch_texel_soa_callback emit_fetch_texel, + void *emit_fetch_texel_context) +{ + struct lp_build_tgsi_soa_context bld; + struct tgsi_parse_context parse; + uint num_immediates = 0; + unsigned i; + + /* Setup build context */ + memset(&bld, 0, sizeof bld); + lp_build_context_init(&bld.base, builder, type); + bld.mask = mask; + bld.pos = pos; + bld.inputs = inputs; + bld.outputs = outputs; + bld.consts_ptr = consts_ptr; + bld.emit_fetch_texel = emit_fetch_texel; + bld.emit_fetch_texel_context = emit_fetch_texel_context; + + tgsi_parse_init( &parse, tokens ); + + while( !tgsi_parse_end_of_tokens( &parse ) ) { + tgsi_parse_token( &parse ); + + switch( parse.FullToken.Token.Type ) { + case TGSI_TOKEN_TYPE_DECLARATION: + /* Input already interpolated */ + break; + + case TGSI_TOKEN_TYPE_INSTRUCTION: + if (!emit_instruction( &bld, &parse.FullToken.FullInstruction )) { + unsigned opcode = parse.FullToken.FullInstruction.Instruction.Opcode; + const struct tgsi_opcode_info *info = tgsi_get_opcode_info(opcode); + _debug_printf("warning: failed to translate tgsi opcode %s to LLVM\n", + info ? info->mnemonic : "<invalid>"); + } + break; + + case TGSI_TOKEN_TYPE_IMMEDIATE: + /* simply copy the immediate values into the next immediates[] slot */ + { + const uint size = parse.FullToken.FullImmediate.Immediate.NrTokens - 1; + assert(size <= 4); + assert(num_immediates < LP_MAX_IMMEDIATES); + for( i = 0; i < size; ++i ) + bld.immediates[num_immediates][i] = + lp_build_const_scalar(type, parse.FullToken.FullImmediate.u[i].Float); + for( i = size; i < 4; ++i ) + bld.immediates[num_immediates][i] = bld.base.undef; + num_immediates++; + } + break; + + default: + assert( 0 ); + } + } + + tgsi_parse_free( &parse ); +} + diff --git a/src/gallium/drivers/llvmpipe/lp_bld_type.c b/src/gallium/drivers/llvmpipe/lp_bld_type.c new file mode 100644 index 00000000000..8e0026fd973 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_type.c @@ -0,0 +1,170 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "util/u_debug.h" + +#include "lp_bld_type.h" +#include "lp_bld_const.h" + + +LLVMTypeRef +lp_build_elem_type(union lp_type type) +{ + if (type.floating) { + switch(type.width) { + case 32: + return LLVMFloatType(); + break; + case 64: + return LLVMDoubleType(); + break; + default: + assert(0); + return LLVMFloatType(); + } + } + else { + return LLVMIntType(type.width); + } +} + + +LLVMTypeRef +lp_build_vec_type(union lp_type type) +{ + LLVMTypeRef elem_type = lp_build_elem_type(type); + return LLVMVectorType(elem_type, type.length); +} + + +/** + * This function is a mirror of lp_build_elem_type() above. + * + * XXX: I'm not sure if it wouldn't be easier/efficient to just recreate the + * type and check for identity. + */ +boolean +lp_check_elem_type(union lp_type type, LLVMTypeRef elem_type) +{ + LLVMTypeKind elem_kind; + + assert(elem_type); + if(!elem_type) + return FALSE; + + elem_kind = LLVMGetTypeKind(elem_type); + + if (type.floating) { + switch(type.width) { + case 32: + if(elem_kind != LLVMFloatTypeKind) + return FALSE; + break; + case 64: + if(elem_kind != LLVMDoubleTypeKind) + return FALSE; + break; + default: + assert(0); + return FALSE; + } + } + else { + if(elem_kind != LLVMIntegerTypeKind) + return FALSE; + + if(LLVMGetIntTypeWidth(elem_type) != type.width) + return FALSE; + } + + return TRUE; +} + + +boolean +lp_check_vec_type(union lp_type type, LLVMTypeRef vec_type) +{ + LLVMTypeRef elem_type; + + assert(vec_type); + if(!vec_type) + return FALSE; + + if(LLVMGetTypeKind(vec_type) != LLVMVectorTypeKind) + return FALSE; + + if(LLVMGetVectorSize(vec_type) != type.length) + return FALSE; + + elem_type = LLVMGetElementType(vec_type); + + return lp_check_elem_type(type, elem_type); +} + + +boolean +lp_check_value(union lp_type type, LLVMValueRef val) +{ + LLVMTypeRef vec_type; + + assert(val); + if(!val) + return FALSE; + + vec_type = LLVMTypeOf(val); + + return lp_check_vec_type(type, vec_type); +} + + +LLVMTypeRef +lp_build_int_elem_type(union lp_type type) +{ + return LLVMIntType(type.width); +} + + +LLVMTypeRef +lp_build_int_vec_type(union lp_type type) +{ + LLVMTypeRef elem_type = lp_build_int_elem_type(type); + return LLVMVectorType(elem_type, type.length); +} + + +void +lp_build_context_init(struct lp_build_context *bld, + LLVMBuilderRef builder, + union lp_type type) +{ + bld->builder = builder; + bld->type = type; + bld->undef = lp_build_undef(type); + bld->zero = lp_build_zero(type); + bld->one = lp_build_one(type); +} diff --git a/src/gallium/drivers/llvmpipe/lp_bld_type.h b/src/gallium/drivers/llvmpipe/lp_bld_type.h new file mode 100644 index 00000000000..3ce566be641 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_bld_type.h @@ -0,0 +1,174 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Convenient representation of SIMD types. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#ifndef LP_BLD_TYPE_H +#define LP_BLD_TYPE_H + + +#include <llvm-c/Core.h> + +#include <pipe/p_compiler.h> + + +/** + * Several functions can only cope with vectors of length up to this value. + * You may need to increase that value if you want to represent bigger vectors. + */ +#define LP_MAX_VECTOR_LENGTH 16 + +#define LP_MAX_TYPE_WIDTH 64 + + +/** + * The LLVM type system can't conveniently express all the things we care about + * on the types used for intermediate computations, such as signed vs unsigned, + * normalized values, or fixed point. + */ +union lp_type { + struct { + /** + * Floating-point. Cannot be used with fixed. Integer numbers are + * represented by this zero. + */ + unsigned floating:1; + + /** + * Fixed-point. Cannot be used with floating. Integer numbers are + * represented by this zero. + */ + unsigned fixed:1; + + /** + * Whether it can represent negative values or not. + * + * If this is not set for floating point, it means that all values are + * assumed to be positive. + */ + unsigned sign:1; + + /** + * Whether values are normalized to fit [0, 1] interval, or [-1, 1] + * interval for signed types. + * + * For integer types it means the representable integer range should be + * interpreted as the interval above. + * + * For floating and fixed point formats it means the values should be + * clamped to the interval above. + */ + unsigned norm:1; + + /** + * Element width. + * + * For fixed point values, the fixed point is assumed to be at half the + * width. + */ + unsigned width:14; + + /** + * Vector length. + * + * width*length should be a power of two greater or equal to eight. + * + * @sa LP_MAX_VECTOR_LENGTH + */ + unsigned length:14; + }; + uint32_t value; +}; + + +/** + * We need most of the information here in order to correctly and efficiently + * translate an arithmetic operation into LLVM IR. Putting it here avoids the + * trouble of passing it as parameters. + */ +struct lp_build_context +{ + LLVMBuilderRef builder; + + /** + * This not only describes the input/output LLVM types, but also whether + * to normalize/clamp the results. + */ + union lp_type type; + + /** Same as lp_build_undef(type) */ + LLVMValueRef undef; + + /** Same as lp_build_zero(type) */ + LLVMValueRef zero; + + /** Same as lp_build_one(type) */ + LLVMValueRef one; +}; + + +LLVMTypeRef +lp_build_elem_type(union lp_type type); + + +LLVMTypeRef +lp_build_vec_type(union lp_type type); + + +boolean +lp_check_elem_type(union lp_type type, LLVMTypeRef elem_type); + + +boolean +lp_check_vec_type(union lp_type type, LLVMTypeRef vec_type); + + +boolean +lp_check_value(union lp_type type, LLVMValueRef val); + + +LLVMTypeRef +lp_build_int_elem_type(union lp_type type); + + +LLVMTypeRef +lp_build_int_vec_type(union lp_type type); + + +void +lp_build_context_init(struct lp_build_context *bld, + LLVMBuilderRef builder, + union lp_type type); + + +#endif /* !LP_BLD_TYPE_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_buffer.c b/src/gallium/drivers/llvmpipe/lp_buffer.c new file mode 100644 index 00000000000..66f1f8e1383 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_buffer.c @@ -0,0 +1,150 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "util/u_memory.h" +#include "util/u_math.h" + +#include "lp_winsys.h" +#include "lp_screen.h" +#include "lp_texture.h" +#include "lp_buffer.h" + + +static void * +llvmpipe_buffer_map(struct pipe_screen *screen, + struct pipe_buffer *buf, + unsigned flags) +{ + struct llvmpipe_buffer *llvmpipe_buf = llvmpipe_buffer(buf); + return llvmpipe_buf->data; +} + + +static void +llvmpipe_buffer_unmap(struct pipe_screen *screen, + struct pipe_buffer *buf) +{ +} + + +static void +llvmpipe_buffer_destroy(struct pipe_buffer *buf) +{ + struct llvmpipe_buffer *sbuf = llvmpipe_buffer(buf); + + if (!sbuf->userBuffer) + align_free(sbuf->data); + + FREE(sbuf); +} + + +static struct pipe_buffer * +llvmpipe_buffer_create(struct pipe_screen *screen, + unsigned alignment, + unsigned usage, + unsigned size) +{ + struct llvmpipe_buffer *buffer = CALLOC_STRUCT(llvmpipe_buffer); + + pipe_reference_init(&buffer->base.reference, 1); + buffer->base.screen = screen; + buffer->base.alignment = MAX2(alignment, 16); + buffer->base.usage = usage; + buffer->base.size = size; + + buffer->data = align_malloc(size, alignment); + + return &buffer->base; +} + + +/** + * Create buffer which wraps user-space data. + */ +static struct pipe_buffer * +llvmpipe_user_buffer_create(struct pipe_screen *screen, + void *ptr, + unsigned bytes) +{ + struct llvmpipe_buffer *buffer; + + buffer = CALLOC_STRUCT(llvmpipe_buffer); + if(!buffer) + return NULL; + + pipe_reference_init(&buffer->base.reference, 1); + buffer->base.screen = screen; + buffer->base.size = bytes; + buffer->userBuffer = TRUE; + buffer->data = ptr; + + return &buffer->base; +} + + +static void +llvmpipe_fence_reference(struct pipe_screen *screen, + struct pipe_fence_handle **ptr, + struct pipe_fence_handle *fence) +{ +} + + +static int +llvmpipe_fence_signalled(struct pipe_screen *screen, + struct pipe_fence_handle *fence, + unsigned flag) +{ + return 0; +} + + +static int +llvmpipe_fence_finish(struct pipe_screen *screen, + struct pipe_fence_handle *fence, + unsigned flag) +{ + return 0; +} + + +void +llvmpipe_init_screen_buffer_funcs(struct pipe_screen *screen) +{ + screen->buffer_create = llvmpipe_buffer_create; + screen->user_buffer_create = llvmpipe_user_buffer_create; + screen->buffer_map = llvmpipe_buffer_map; + screen->buffer_unmap = llvmpipe_buffer_unmap; + screen->buffer_destroy = llvmpipe_buffer_destroy; + + screen->fence_reference = llvmpipe_fence_reference; + screen->fence_signalled = llvmpipe_fence_signalled; + screen->fence_finish = llvmpipe_fence_finish; + +} diff --git a/src/gallium/drivers/llvmpipe/lp_buffer.h b/src/gallium/drivers/llvmpipe/lp_buffer.h new file mode 100644 index 00000000000..d6b8184a0b9 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_buffer.h @@ -0,0 +1,55 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef LP_BUFFER_H +#define LP_BUFFER_H + +#include "pipe/p_compiler.h" +#include "pipe/p_state.h" + + +struct llvmpipe_buffer +{ + struct pipe_buffer base; + boolean userBuffer; /** Is this a user-space buffer? */ + void *data; +}; + + +/** Cast wrapper */ +static INLINE struct llvmpipe_buffer * +llvmpipe_buffer( struct pipe_buffer *buf ) +{ + return (struct llvmpipe_buffer *)buf; +} + + +void +llvmpipe_init_screen_buffer_funcs(struct pipe_screen *screen); + + +#endif /* LP_BUFFER_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_clear.c b/src/gallium/drivers/llvmpipe/lp_clear.c new file mode 100644 index 00000000000..580cca5b463 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_clear.c @@ -0,0 +1,80 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Brian Paul + * Michel Dänzer + */ + + +#include "pipe/p_defines.h" +#include "util/u_pack_color.h" +#include "lp_clear.h" +#include "lp_context.h" +#include "lp_surface.h" +#include "lp_state.h" +#include "lp_tile_cache.h" + + +/** + * Clear the given buffers to the specified values. + * No masking, no scissor (clear entire buffer). + */ +void +llvmpipe_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba, + double depth, unsigned stencil) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + unsigned cv; + uint i; + + if (llvmpipe->no_rast) + return; + +#if 0 + llvmpipe_update_derived(llvmpipe); /* not needed?? */ +#endif + + if (buffers & PIPE_CLEAR_COLOR) { + for (i = 0; i < llvmpipe->framebuffer.nr_cbufs; i++) { + struct pipe_surface *ps = llvmpipe->framebuffer.cbufs[i]; + + util_pack_color(rgba, ps->format, &cv); + lp_tile_cache_clear(llvmpipe->cbuf_cache[i], rgba, cv); + } + } + + if (buffers & PIPE_CLEAR_DEPTHSTENCIL) { + struct pipe_surface *ps = llvmpipe->framebuffer.zsbuf; + + cv = util_pack_z_stencil(ps->format, depth, stencil); + + /* non-cached surface */ + pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, cv); + } +} diff --git a/src/gallium/drivers/llvmpipe/lp_clear.h b/src/gallium/drivers/llvmpipe/lp_clear.h new file mode 100644 index 00000000000..6d4ffccdf45 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_clear.h @@ -0,0 +1,43 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Brian Paul + */ + +#ifndef LP_CLEAR_H +#define LP_CLEAR_H + +#include "pipe/p_state.h" +struct pipe_context; + +extern void +llvmpipe_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba, + double depth, unsigned stencil); + + +#endif /* LP_CLEAR_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c new file mode 100644 index 00000000000..233d1df0e10 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_context.c @@ -0,0 +1,297 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * Copyright 2008 VMware, Inc. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "draw/draw_context.h" +#include "pipe/p_defines.h" +#include "util/u_math.h" +#include "util/u_memory.h" +#include "lp_clear.h" +#include "lp_context.h" +#include "lp_flush.h" +#include "lp_prim_setup.h" +#include "lp_prim_vbuf.h" +#include "lp_state.h" +#include "lp_surface.h" +#include "lp_tile_cache.h" +#include "lp_tex_cache.h" +#include "lp_texture.h" +#include "lp_winsys.h" +#include "lp_query.h" + + + +/** + * Map any drawing surfaces which aren't already mapped + */ +void +llvmpipe_map_transfers(struct llvmpipe_context *lp) +{ + struct pipe_screen *screen = lp->pipe.screen; + struct pipe_surface *zsbuf = lp->framebuffer.zsbuf; + unsigned i; + + for (i = 0; i < lp->framebuffer.nr_cbufs; i++) { + lp_tile_cache_map_transfers(lp->cbuf_cache[i]); + } + + if(zsbuf) { + if(!lp->zsbuf_transfer) + lp->zsbuf_transfer = screen->get_tex_transfer(screen, zsbuf->texture, + zsbuf->face, zsbuf->level, zsbuf->zslice, + PIPE_TRANSFER_READ_WRITE, + 0, 0, zsbuf->width, zsbuf->height); + if(lp->zsbuf_transfer && !lp->zsbuf_map) + lp->zsbuf_map = screen->transfer_map(screen, lp->zsbuf_transfer); + + } +} + + +/** + * Unmap any mapped drawing surfaces + */ +void +llvmpipe_unmap_transfers(struct llvmpipe_context *lp) +{ + uint i; + + for (i = 0; i < lp->framebuffer.nr_cbufs; i++) { + lp_tile_cache_unmap_transfers(lp->cbuf_cache[i]); + } + + if(lp->zsbuf_transfer) { + struct pipe_screen *screen = lp->pipe.screen; + + if(lp->zsbuf_map) { + screen->transfer_unmap(screen, lp->zsbuf_transfer); + lp->zsbuf_map = NULL; + } + } +} + + +static void llvmpipe_destroy( struct pipe_context *pipe ) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context( pipe ); + uint i; + + if (llvmpipe->draw) + draw_destroy( llvmpipe->draw ); + + for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) + lp_destroy_tile_cache(llvmpipe->cbuf_cache[i]); + + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) + lp_destroy_tex_tile_cache(llvmpipe->tex_cache[i]); + + for (i = 0; i < Elements(llvmpipe->constants); i++) { + if (llvmpipe->constants[i].buffer) { + pipe_buffer_reference(&llvmpipe->constants[i].buffer, NULL); + } + } + + align_free( llvmpipe ); +} + +static unsigned int +llvmpipe_is_texture_referenced( struct pipe_context *pipe, + struct pipe_texture *texture, + unsigned face, unsigned level) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context( pipe ); + unsigned i; + + if(llvmpipe->dirty_render_cache) { + for (i = 0; i < llvmpipe->framebuffer.nr_cbufs; i++) { + if(llvmpipe->framebuffer.cbufs[i] && + llvmpipe->framebuffer.cbufs[i]->texture == texture) + return PIPE_REFERENCED_FOR_WRITE; + } + if(llvmpipe->framebuffer.zsbuf && + llvmpipe->framebuffer.zsbuf->texture == texture) + return PIPE_REFERENCED_FOR_WRITE; + } + + /* FIXME: we also need to do the same for the texture cache */ + + return PIPE_UNREFERENCED; +} + +static unsigned int +llvmpipe_is_buffer_referenced( struct pipe_context *pipe, + struct pipe_buffer *buf) +{ + return PIPE_UNREFERENCED; +} + +struct pipe_context * +llvmpipe_create( struct pipe_screen *screen ) +{ + struct llvmpipe_context *llvmpipe; + uint i; + + llvmpipe = align_malloc(sizeof(struct llvmpipe_context), 16); + if (!llvmpipe) + return NULL; + + util_init_math(); + + memset(llvmpipe, 0, sizeof *llvmpipe); + + llvmpipe->pipe.winsys = screen->winsys; + llvmpipe->pipe.screen = screen; + llvmpipe->pipe.destroy = llvmpipe_destroy; + + /* state setters */ + llvmpipe->pipe.create_blend_state = llvmpipe_create_blend_state; + llvmpipe->pipe.bind_blend_state = llvmpipe_bind_blend_state; + llvmpipe->pipe.delete_blend_state = llvmpipe_delete_blend_state; + + llvmpipe->pipe.create_sampler_state = llvmpipe_create_sampler_state; + llvmpipe->pipe.bind_sampler_states = llvmpipe_bind_sampler_states; + llvmpipe->pipe.delete_sampler_state = llvmpipe_delete_sampler_state; + + llvmpipe->pipe.create_depth_stencil_alpha_state = llvmpipe_create_depth_stencil_state; + llvmpipe->pipe.bind_depth_stencil_alpha_state = llvmpipe_bind_depth_stencil_state; + llvmpipe->pipe.delete_depth_stencil_alpha_state = llvmpipe_delete_depth_stencil_state; + + llvmpipe->pipe.create_rasterizer_state = llvmpipe_create_rasterizer_state; + llvmpipe->pipe.bind_rasterizer_state = llvmpipe_bind_rasterizer_state; + llvmpipe->pipe.delete_rasterizer_state = llvmpipe_delete_rasterizer_state; + + llvmpipe->pipe.create_fs_state = llvmpipe_create_fs_state; + llvmpipe->pipe.bind_fs_state = llvmpipe_bind_fs_state; + llvmpipe->pipe.delete_fs_state = llvmpipe_delete_fs_state; + + llvmpipe->pipe.create_vs_state = llvmpipe_create_vs_state; + llvmpipe->pipe.bind_vs_state = llvmpipe_bind_vs_state; + llvmpipe->pipe.delete_vs_state = llvmpipe_delete_vs_state; + + llvmpipe->pipe.set_blend_color = llvmpipe_set_blend_color; + llvmpipe->pipe.set_clip_state = llvmpipe_set_clip_state; + llvmpipe->pipe.set_constant_buffer = llvmpipe_set_constant_buffer; + llvmpipe->pipe.set_framebuffer_state = llvmpipe_set_framebuffer_state; + llvmpipe->pipe.set_polygon_stipple = llvmpipe_set_polygon_stipple; + llvmpipe->pipe.set_scissor_state = llvmpipe_set_scissor_state; + llvmpipe->pipe.set_sampler_textures = llvmpipe_set_sampler_textures; + llvmpipe->pipe.set_viewport_state = llvmpipe_set_viewport_state; + + llvmpipe->pipe.set_vertex_buffers = llvmpipe_set_vertex_buffers; + llvmpipe->pipe.set_vertex_elements = llvmpipe_set_vertex_elements; + + llvmpipe->pipe.draw_arrays = llvmpipe_draw_arrays; + llvmpipe->pipe.draw_elements = llvmpipe_draw_elements; + llvmpipe->pipe.draw_range_elements = llvmpipe_draw_range_elements; + llvmpipe->pipe.set_edgeflags = llvmpipe_set_edgeflags; + + + llvmpipe->pipe.clear = llvmpipe_clear; + llvmpipe->pipe.flush = llvmpipe_flush; + + llvmpipe->pipe.is_texture_referenced = llvmpipe_is_texture_referenced; + llvmpipe->pipe.is_buffer_referenced = llvmpipe_is_buffer_referenced; + + llvmpipe_init_query_funcs( llvmpipe ); + llvmpipe_init_texture_funcs( llvmpipe ); + + /* + * Alloc caches for accessing drawing surfaces and textures. + */ + for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) + llvmpipe->cbuf_cache[i] = lp_create_tile_cache( screen ); + + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) + llvmpipe->tex_cache[i] = lp_create_tex_tile_cache( screen ); + + + /* vertex shader samplers */ + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + llvmpipe->tgsi.vert_samplers[i].base.get_samples = lp_get_samples; + llvmpipe->tgsi.vert_samplers[i].processor = TGSI_PROCESSOR_VERTEX; + llvmpipe->tgsi.vert_samplers[i].cache = llvmpipe->tex_cache[i]; + llvmpipe->tgsi.vert_samplers_list[i] = &llvmpipe->tgsi.vert_samplers[i]; + } + + /* fragment shader samplers */ + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + llvmpipe->tgsi.frag_samplers[i].base.get_samples = lp_get_samples; + llvmpipe->tgsi.frag_samplers[i].processor = TGSI_PROCESSOR_FRAGMENT; + llvmpipe->tgsi.frag_samplers[i].cache = llvmpipe->tex_cache[i]; + llvmpipe->tgsi.frag_samplers_list[i] = &llvmpipe->tgsi.frag_samplers[i]; + } + + /* + * Create drawing context and plug our rendering stage into it. + */ + llvmpipe->draw = draw_create(); + if (!llvmpipe->draw) + goto fail; + + draw_texture_samplers(llvmpipe->draw, + PIPE_MAX_SAMPLERS, + (struct tgsi_sampler **) + llvmpipe->tgsi.vert_samplers_list); + + llvmpipe->setup = lp_draw_render_stage(llvmpipe); + if (!llvmpipe->setup) + goto fail; + + if (debug_get_bool_option( "LP_NO_RAST", FALSE )) + llvmpipe->no_rast = TRUE; + + if (debug_get_bool_option( "LP_NO_VBUF", FALSE )) { + /* Deprecated path -- vbuf is the intended interface to the draw module: + */ + draw_set_rasterize_stage(llvmpipe->draw, llvmpipe->setup); + } + else { + lp_init_vbuf(llvmpipe); + } + + /* plug in AA line/point stages */ + draw_install_aaline_stage(llvmpipe->draw, &llvmpipe->pipe); + draw_install_aapoint_stage(llvmpipe->draw, &llvmpipe->pipe); + +#if USE_DRAW_STAGE_PSTIPPLE + /* Do polygon stipple w/ texture map + frag prog? */ + draw_install_pstipple_stage(llvmpipe->draw, &llvmpipe->pipe); +#endif + + lp_init_surface_functions(llvmpipe); + + return &llvmpipe->pipe; + + fail: + llvmpipe_destroy(&llvmpipe->pipe); + return NULL; +} + diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h new file mode 100644 index 00000000000..8d5a0d4f1fc --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_context.h @@ -0,0 +1,155 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef LP_CONTEXT_H +#define LP_CONTEXT_H + +#include "pipe/p_context.h" + +#include "draw/draw_vertex.h" + +#include "lp_tex_sample.h" +#include "lp_jit.h" + + +struct llvmpipe_vbuf_render; +struct draw_context; +struct draw_stage; +struct llvmpipe_tile_cache; +struct llvmpipe_tex_tile_cache; +struct lp_fragment_shader; +struct lp_vertex_shader; +struct lp_blend_state; + + +struct llvmpipe_context { + struct pipe_context pipe; /**< base class */ + + /** Constant state objects */ + const struct pipe_blend_state *blend; + const struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS]; + const struct pipe_depth_stencil_alpha_state *depth_stencil; + const struct pipe_rasterizer_state *rasterizer; + struct lp_fragment_shader *fs; + const struct lp_vertex_shader *vs; + + /** Other rendering state */ + struct pipe_blend_color blend_color[4][16]; + struct pipe_clip_state clip; + struct pipe_constant_buffer constants[PIPE_SHADER_TYPES]; + struct pipe_framebuffer_state framebuffer; + struct pipe_poly_stipple poly_stipple; + struct pipe_scissor_state scissor; + struct pipe_texture *texture[PIPE_MAX_SAMPLERS]; + struct pipe_viewport_state viewport; + struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS]; + struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS]; + + unsigned num_samplers; + unsigned num_textures; + unsigned num_vertex_elements; + unsigned num_vertex_buffers; + + unsigned dirty; /**< Mask of LP_NEW_x flags */ + + /* Counter for occlusion queries. Note this supports overlapping + * queries. + */ + uint64_t occlusion_count; + unsigned active_query_count; + + /** Mapped vertex buffers */ + ubyte *mapped_vbuffer[PIPE_MAX_ATTRIBS]; + + /** Mapped constant buffers */ + void *mapped_constants[PIPE_SHADER_TYPES]; + + /** Vertex format */ + struct vertex_info vertex_info; + struct vertex_info vertex_info_vbuf; + + /** Which vertex shader output slot contains point size */ + int psize_slot; + + /* The reduced version of the primitive supplied by the state + * tracker. + */ + unsigned reduced_api_prim; + + /* The reduced primitive after unfilled triangles, wide-line + * decomposition, etc, are taken into account. This is the + * primitive actually rasterized. + */ + unsigned reduced_prim; + + /** Derived from scissor and surface bounds: */ + struct pipe_scissor_state cliprect; + + unsigned line_stipple_counter; + + /** TGSI exec things */ + struct { + struct lp_shader_sampler vert_samplers[PIPE_MAX_SAMPLERS]; + struct lp_shader_sampler *vert_samplers_list[PIPE_MAX_SAMPLERS]; + struct lp_shader_sampler frag_samplers[PIPE_MAX_SAMPLERS]; + struct lp_shader_sampler *frag_samplers_list[PIPE_MAX_SAMPLERS]; + } tgsi; + + /** The primitive drawing context */ + struct draw_context *draw; + struct draw_stage *setup; + struct draw_stage *vbuf; + struct llvmpipe_vbuf_render *vbuf_render; + + boolean dirty_render_cache; + + struct llvmpipe_tile_cache *cbuf_cache[PIPE_MAX_COLOR_BUFS]; + + /* TODO: we shouldn't be using external interfaces internally like this */ + struct pipe_transfer *zsbuf_transfer; + uint8_t *zsbuf_map; + + unsigned tex_timestamp; + struct llvmpipe_tex_tile_cache *tex_cache[PIPE_MAX_SAMPLERS]; + + unsigned no_rast : 1; + + struct lp_jit_context jit_context; +}; + + +static INLINE struct llvmpipe_context * +llvmpipe_context( struct pipe_context *pipe ) +{ + return (struct llvmpipe_context *)pipe; +} + +#endif /* LP_CONTEXT_H */ + diff --git a/src/gallium/drivers/llvmpipe/lp_draw_arrays.c b/src/gallium/drivers/llvmpipe/lp_draw_arrays.c new file mode 100644 index 00000000000..89772e62d31 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_draw_arrays.c @@ -0,0 +1,192 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Brian Paul + * Keith Whitwell + */ + + +#include "pipe/p_defines.h" +#include "pipe/p_context.h" +#include "pipe/internal/p_winsys_screen.h" +#include "pipe/p_inlines.h" +#include "util/u_prim.h" + +#include "lp_buffer.h" +#include "lp_context.h" +#include "lp_state.h" + +#include "draw/draw_context.h" + + + +static void +llvmpipe_map_constant_buffers(struct llvmpipe_context *lp) +{ + struct pipe_screen *screen = lp->pipe.screen; + uint i, size; + + for (i = 0; i < PIPE_SHADER_TYPES; i++) { + if (lp->constants[i].buffer && lp->constants[i].buffer->size) + lp->mapped_constants[i] = screen->buffer_map(screen, lp->constants[i].buffer, + PIPE_BUFFER_USAGE_CPU_READ); + } + + if (lp->constants[PIPE_SHADER_VERTEX].buffer) + size = lp->constants[PIPE_SHADER_VERTEX].buffer->size; + else + size = 0; + + lp->jit_context.constants = lp->mapped_constants[PIPE_SHADER_FRAGMENT]; + + draw_set_mapped_constant_buffer(lp->draw, + lp->mapped_constants[PIPE_SHADER_VERTEX], + size); +} + + +static void +llvmpipe_unmap_constant_buffers(struct llvmpipe_context *lp) +{ + struct pipe_screen *screen = lp->pipe.screen; + uint i; + + /* really need to flush all prims since the vert/frag shaders const buffers + * are going away now. + */ + draw_flush(lp->draw); + + draw_set_mapped_constant_buffer(lp->draw, NULL, 0); + + lp->jit_context.constants = NULL; + + for (i = 0; i < 2; i++) { + if (lp->constants[i].buffer && lp->constants[i].buffer->size) + screen->buffer_unmap(screen, lp->constants[i].buffer); + lp->mapped_constants[i] = NULL; + } +} + + +boolean +llvmpipe_draw_arrays(struct pipe_context *pipe, unsigned mode, + unsigned start, unsigned count) +{ + return llvmpipe_draw_elements(pipe, NULL, 0, mode, start, count); +} + + +/** + * Draw vertex arrays, with optional indexing. + * Basically, map the vertex buffers (and drawing surfaces), then hand off + * the drawing to the 'draw' module. + */ +boolean +llvmpipe_draw_range_elements(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned min_index, + unsigned max_index, + unsigned mode, unsigned start, unsigned count) +{ + struct llvmpipe_context *lp = llvmpipe_context(pipe); + struct draw_context *draw = lp->draw; + unsigned i; + + lp->reduced_api_prim = u_reduced_prim(mode); + + if (lp->dirty) + llvmpipe_update_derived( lp ); + + llvmpipe_map_transfers(lp); + llvmpipe_map_constant_buffers(lp); + + /* + * Map vertex buffers + */ + for (i = 0; i < lp->num_vertex_buffers; i++) { + void *buf = llvmpipe_buffer(lp->vertex_buffer[i].buffer)->data; + draw_set_mapped_vertex_buffer(draw, i, buf); + } + + /* Map index buffer, if present */ + if (indexBuffer) { + void *mapped_indexes = llvmpipe_buffer(indexBuffer)->data; + draw_set_mapped_element_buffer_range(draw, indexSize, + min_index, + max_index, + mapped_indexes); + } + else { + /* no index/element buffer */ + draw_set_mapped_element_buffer_range(draw, 0, start, + start + count - 1, NULL); + } + + /* draw! */ + draw_arrays(draw, mode, start, count); + + /* + * unmap vertex/index buffers - will cause draw module to flush + */ + for (i = 0; i < lp->num_vertex_buffers; i++) { + draw_set_mapped_vertex_buffer(draw, i, NULL); + } + if (indexBuffer) { + draw_set_mapped_element_buffer(draw, 0, NULL); + } + + + /* Note: leave drawing surfaces mapped */ + llvmpipe_unmap_constant_buffers(lp); + + lp->dirty_render_cache = TRUE; + + return TRUE; +} + + +boolean +llvmpipe_draw_elements(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned mode, unsigned start, unsigned count) +{ + return llvmpipe_draw_range_elements( pipe, indexBuffer, + indexSize, + 0, 0xffffffff, + mode, start, count ); +} + + +void +llvmpipe_set_edgeflags(struct pipe_context *pipe, const unsigned *edgeflags) +{ + struct llvmpipe_context *lp = llvmpipe_context(pipe); + draw_set_edgeflags(lp->draw, edgeflags); +} diff --git a/src/gallium/drivers/llvmpipe/lp_flush.c b/src/gallium/drivers/llvmpipe/lp_flush.c new file mode 100644 index 00000000000..b5c1c95bb73 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_flush.c @@ -0,0 +1,98 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "pipe/p_defines.h" +#include "draw/draw_context.h" +#include "lp_flush.h" +#include "lp_context.h" +#include "lp_surface.h" +#include "lp_state.h" +#include "lp_tile_cache.h" +#include "lp_tex_cache.h" +#include "lp_winsys.h" + + +void +llvmpipe_flush( struct pipe_context *pipe, + unsigned flags, + struct pipe_fence_handle **fence ) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + uint i; + + draw_flush(llvmpipe->draw); + + if (flags & PIPE_FLUSH_SWAPBUFFERS) { + /* If this is a swapbuffers, just flush color buffers. + * + * The zbuffer changes are not discarded, but held in the cache + * in the hope that a later clear will wipe them out. + */ + for (i = 0; i < llvmpipe->framebuffer.nr_cbufs; i++) + if (llvmpipe->cbuf_cache[i]) + lp_flush_tile_cache(llvmpipe->cbuf_cache[i]); + + /* Need this call for hardware buffers before swapbuffers. + * + * there should probably be another/different flush-type function + * that's called before swapbuffers because we don't always want + * to unmap surfaces when flushing. + */ + llvmpipe_unmap_transfers(llvmpipe); + } + else if (flags & PIPE_FLUSH_RENDER_CACHE) { + for (i = 0; i < llvmpipe->framebuffer.nr_cbufs; i++) + if (llvmpipe->cbuf_cache[i]) + lp_flush_tile_cache(llvmpipe->cbuf_cache[i]); + + /* FIXME: untile zsbuf! */ + + llvmpipe->dirty_render_cache = FALSE; + } + + /* Enable to dump BMPs of the color/depth buffers each frame */ +#if 0 + if(flags & PIPE_FLUSH_FRAME) { + static unsigned frame_no = 1; + static char filename[256]; + util_snprintf(filename, sizeof(filename), "cbuf_%u.bmp", frame_no); + debug_dump_surface_bmp(filename, llvmpipe->framebuffer.cbufs[0]); + util_snprintf(filename, sizeof(filename), "zsbuf_%u.bmp", frame_no); + debug_dump_surface_bmp(filename, llvmpipe->framebuffer.zsbuf); + ++frame_no; + } +#endif + + if (fence) + *fence = NULL; +} + diff --git a/src/gallium/drivers/llvmpipe/lp_flush.h b/src/gallium/drivers/llvmpipe/lp_flush.h new file mode 100644 index 00000000000..10b2b525836 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_flush.h @@ -0,0 +1,37 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef LP_FLUSH_H +#define LP_FLUSH_H + +struct pipe_context; +struct pipe_fence_handle; + +void llvmpipe_flush(struct pipe_context *pipe, unsigned flags, + struct pipe_fence_handle **fence); + +#endif diff --git a/src/gallium/drivers/llvmpipe/lp_jit.c b/src/gallium/drivers/llvmpipe/lp_jit.c new file mode 100644 index 00000000000..d288460a1b8 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_jit.c @@ -0,0 +1,138 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * C - JIT interfaces + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#include <llvm-c/Transforms/Scalar.h> + +#include "util/u_memory.h" +#include "lp_screen.h" +#include "lp_bld_intr.h" +#include "lp_jit.h" + + +static void +lp_jit_init_globals(struct llvmpipe_screen *screen) +{ + /* struct lp_jit_context */ + { + LLVMTypeRef elem_types[4]; + LLVMTypeRef context_type; + + elem_types[0] = LLVMPointerType(LLVMFloatType(), 0); /* constants */ + elem_types[1] = LLVMPointerType(LLVMInt8Type(), 0); /* samplers */ + elem_types[2] = LLVMFloatType(); /* alpha_ref_value */ + elem_types[3] = LLVMPointerType(LLVMInt8Type(), 0); /* blend_color */ + + context_type = LLVMStructType(elem_types, Elements(elem_types), 0); + + LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, constants, + screen->target, context_type, 0); + LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, samplers, + screen->target, context_type, 1); + LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, alpha_ref_value, + screen->target, context_type, 2); + LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, blend_color, + screen->target, context_type, 3); + LP_CHECK_STRUCT_SIZE(struct lp_jit_context, + screen->target, context_type); + + LLVMAddTypeName(screen->module, "context", context_type); + + screen->context_ptr_type = LLVMPointerType(context_type, 0); + } + + /* fetch_texel + */ + { + LLVMTypeRef ret_type; + LLVMTypeRef arg_types[3]; + LLVMValueRef fetch_texel; + + ret_type = LLVMVoidType(); + arg_types[0] = LLVMPointerType(LLVMInt8Type(), 0); /* samplers */ + arg_types[1] = LLVMInt32Type(); /* unit */ + arg_types[2] = LLVMPointerType(LLVMVectorType(LLVMFloatType(), 4), 0); /* store */ + + fetch_texel = lp_declare_intrinsic(screen->module, "fetch_texel", + ret_type, arg_types, Elements(arg_types)); + + LLVMAddGlobalMapping(screen->engine, fetch_texel, lp_fetch_texel_soa); + } + +#ifdef DEBUG + LLVMDumpModule(screen->module); +#endif +} + + +void +lp_jit_screen_cleanup(struct llvmpipe_screen *screen) +{ + if(screen->engine) + LLVMDisposeExecutionEngine(screen->engine); + + if(screen->pass) + LLVMDisposePassManager(screen->pass); +} + + +void +lp_jit_screen_init(struct llvmpipe_screen *screen) +{ + char *error = NULL; + + screen->module = LLVMModuleCreateWithName("llvmpipe"); + + screen->provider = LLVMCreateModuleProviderForExistingModule(screen->module); + + if (LLVMCreateJITCompiler(&screen->engine, screen->provider, 1, &error)) { + fprintf(stderr, "%s\n", error); + LLVMDisposeMessage(error); + abort(); + } + + screen->target = LLVMGetExecutionEngineTargetData(screen->engine); + + screen->pass = LLVMCreateFunctionPassManager(screen->provider); + LLVMAddTargetData(screen->target, screen->pass); + /* These are the passes currently listed in llvm-c/Transforms/Scalar.h, + * but there are more on SVN. */ + LLVMAddConstantPropagationPass(screen->pass); + LLVMAddInstructionCombiningPass(screen->pass); + LLVMAddPromoteMemoryToRegisterPass(screen->pass); + LLVMAddGVNPass(screen->pass); + LLVMAddCFGSimplificationPass(screen->pass); + + lp_jit_init_globals(screen); +} diff --git a/src/gallium/drivers/llvmpipe/lp_jit.h b/src/gallium/drivers/llvmpipe/lp_jit.h new file mode 100644 index 00000000000..a7fb60f9f5c --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_jit.h @@ -0,0 +1,109 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * C - JIT interfaces + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#ifndef LP_JIT_H +#define LP_JIT_H + + +#include "lp_bld_struct.h" + + +struct tgsi_sampler; +struct llvmpipe_screen; + + +/** + * This structure is passed directly to the generated fragment shader. + * + * It contains the derived state. + * + * Changes here must be reflected in the lp_jit_context_* macros and + * lp_jit_init_types function. Changes to the ordering should be avoided. + * + * Only use types with a clear size and padding here, in particular prefer the + * stdint.h types to the basic integer types. + */ +struct lp_jit_context +{ + const float *constants; + + struct tgsi_sampler **samplers; + + /* TODO: alpha reference value */ + float alpha_ref_value; + + /* TODO: blend constant color */ + uint8_t *blend_color; +}; + + +#define lp_jit_context_constants(_builder, _ptr) \ + lp_build_struct_get(_builder, _ptr, 0, "constants") + +#define lp_jit_context_samplers(_builder, _ptr) \ + lp_build_struct_get(_builder, _ptr, 1, "samplers") + +#define lp_jit_context_alpha_ref_value(_builder, _ptr) \ + lp_build_struct_get(_builder, _ptr, 2, "alpha_ref_value") + +#define lp_jit_context_blend_color(_builder, _ptr) \ + lp_build_struct_get(_builder, _ptr, 3, "blend_color") + + +typedef void +(*lp_jit_frag_func)(struct lp_jit_context *context, + uint32_t x, + uint32_t y, + const void *a0, + const void *dadx, + const void *dady, + uint32_t *mask, + void *color, + void *depth); + +void PIPE_CDECL +lp_fetch_texel_soa( struct tgsi_sampler **samplers, + uint32_t unit, + float *store ); + + +void +lp_jit_screen_cleanup(struct llvmpipe_screen *screen); + + +void +lp_jit_screen_init(struct llvmpipe_screen *screen); + + +#endif /* LP_JIT_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_prim_setup.c b/src/gallium/drivers/llvmpipe/lp_prim_setup.c new file mode 100644 index 00000000000..b14f8fb99d9 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_prim_setup.c @@ -0,0 +1,190 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * \brief A draw stage that drives our triangle setup routines from + * within the draw pipeline. One of two ways to drive setup, the + * other being in lp_prim_vbuf.c. + * + * \author Keith Whitwell <keith@tungstengraphics.com> + * \author Brian Paul + */ + + +#include "lp_context.h" +#include "lp_setup.h" +#include "lp_state.h" +#include "lp_prim_setup.h" +#include "draw/draw_pipe.h" +#include "draw/draw_vertex.h" +#include "util/u_memory.h" + +/** + * Triangle setup info (derived from draw_stage). + * Also used for line drawing (taking some liberties). + */ +struct setup_stage { + struct draw_stage stage; /**< This must be first (base class) */ + + struct setup_context *setup; +}; + + + +/** + * Basically a cast wrapper. + */ +static INLINE struct setup_stage *setup_stage( struct draw_stage *stage ) +{ + return (struct setup_stage *)stage; +} + + +typedef const float (*cptrf4)[4]; + +static void +do_tri(struct draw_stage *stage, struct prim_header *prim) +{ + struct setup_stage *setup = setup_stage( stage ); + + llvmpipe_setup_tri( setup->setup, + (cptrf4)prim->v[0]->data, + (cptrf4)prim->v[1]->data, + (cptrf4)prim->v[2]->data ); +} + +static void +do_line(struct draw_stage *stage, struct prim_header *prim) +{ + struct setup_stage *setup = setup_stage( stage ); + + llvmpipe_setup_line( setup->setup, + (cptrf4)prim->v[0]->data, + (cptrf4)prim->v[1]->data ); +} + +static void +do_point(struct draw_stage *stage, struct prim_header *prim) +{ + struct setup_stage *setup = setup_stage( stage ); + + llvmpipe_setup_point( setup->setup, + (cptrf4)prim->v[0]->data ); +} + + + + +static void setup_begin( struct draw_stage *stage ) +{ + struct setup_stage *setup = setup_stage(stage); + + llvmpipe_setup_prepare( setup->setup ); + + stage->point = do_point; + stage->line = do_line; + stage->tri = do_tri; +} + + +static void setup_first_point( struct draw_stage *stage, + struct prim_header *header ) +{ + setup_begin(stage); + stage->point( stage, header ); +} + +static void setup_first_line( struct draw_stage *stage, + struct prim_header *header ) +{ + setup_begin(stage); + stage->line( stage, header ); +} + + +static void setup_first_tri( struct draw_stage *stage, + struct prim_header *header ) +{ + setup_begin(stage); + stage->tri( stage, header ); +} + + + +static void setup_flush( struct draw_stage *stage, + unsigned flags ) +{ + stage->point = setup_first_point; + stage->line = setup_first_line; + stage->tri = setup_first_tri; +} + + +static void reset_stipple_counter( struct draw_stage *stage ) +{ +} + + +static void render_destroy( struct draw_stage *stage ) +{ + struct setup_stage *ssetup = setup_stage(stage); + llvmpipe_setup_destroy_context(ssetup->setup); + FREE( stage ); +} + + +/** + * Create a new primitive setup/render stage. + */ +struct draw_stage *lp_draw_render_stage( struct llvmpipe_context *llvmpipe ) +{ + struct setup_stage *sstage = CALLOC_STRUCT(setup_stage); + + sstage->setup = llvmpipe_setup_create_context(llvmpipe); + sstage->stage.draw = llvmpipe->draw; + sstage->stage.point = setup_first_point; + sstage->stage.line = setup_first_line; + sstage->stage.tri = setup_first_tri; + sstage->stage.flush = setup_flush; + sstage->stage.reset_stipple_counter = reset_stipple_counter; + sstage->stage.destroy = render_destroy; + + return (struct draw_stage *)sstage; +} + +struct setup_context * +lp_draw_setup_context( struct draw_stage *stage ) +{ + struct setup_stage *ssetup = setup_stage(stage); + return ssetup->setup; +} + +void +lp_draw_flush( struct draw_stage *stage ) +{ + stage->flush( stage, 0 ); +} diff --git a/src/gallium/drivers/llvmpipe/lp_prim_setup.h b/src/gallium/drivers/llvmpipe/lp_prim_setup.h new file mode 100644 index 00000000000..da6cae63751 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_prim_setup.h @@ -0,0 +1,85 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef LP_PRIM_SETUP_H +#define LP_PRIM_SETUP_H + + +/** + * vbuf is a special stage to gather the stream of triangles, lines, points + * together and reconstruct vertex buffers for hardware upload. + * + * First attempt, work in progress. + * + * TODO: + * - separate out vertex buffer building and primitive emit, ie >1 draw per vb. + * - tell vbuf stage how to build hw vertices directly + * - pass vbuf stage a buffer pointer for direct emit to agp/vram. + * + * + * + * Vertices are just an array of floats, with all the attributes + * packed. We currently assume a layout like: + * + * attr[0][0..3] - window position + * attr[1..n][0..3] - remaining attributes. + * + * Attributes are assumed to be 4 floats wide but are packed so that + * all the enabled attributes run contiguously. + */ + + +struct draw_stage; +struct llvmpipe_context; + + +typedef void (*vbuf_draw_func)( struct pipe_context *pipe, + unsigned prim, + const ushort *elements, + unsigned nr_elements, + const void *vertex_buffer, + unsigned nr_vertices ); + + +extern struct draw_stage * +lp_draw_render_stage( struct llvmpipe_context *llvmpipe ); + +extern struct setup_context * +lp_draw_setup_context( struct draw_stage * ); + +extern void +lp_draw_flush( struct draw_stage * ); + + +extern struct draw_stage * +lp_draw_vbuf_stage( struct draw_context *draw_context, + struct pipe_context *pipe, + vbuf_draw_func draw ); + + +#endif /* LP_PRIM_SETUP_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_prim_vbuf.c b/src/gallium/drivers/llvmpipe/lp_prim_vbuf.c new file mode 100644 index 00000000000..c394dcb61d0 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_prim_vbuf.c @@ -0,0 +1,607 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Interface between 'draw' module's output and the llvmpipe rasterizer/setup + * code. When the 'draw' module has finished filling a vertex buffer, the + * draw_arrays() functions below will be called. Loop over the vertices and + * call the point/line/tri setup functions. + * + * Authors + * Brian Paul + */ + + +#include "lp_context.h" +#include "lp_state.h" +#include "lp_prim_vbuf.h" +#include "lp_prim_setup.h" +#include "lp_setup.h" +#include "draw/draw_context.h" +#include "draw/draw_vbuf.h" +#include "util/u_memory.h" +#include "util/u_prim.h" + + +#define LP_MAX_VBUF_INDEXES 1024 +#define LP_MAX_VBUF_SIZE 4096 + +typedef const float (*cptrf4)[4]; + +/** + * Subclass of vbuf_render. + */ +struct llvmpipe_vbuf_render +{ + struct vbuf_render base; + struct llvmpipe_context *llvmpipe; + uint prim; + uint vertex_size; + uint nr_vertices; + uint vertex_buffer_size; + void *vertex_buffer; +}; + + +/** cast wrapper */ +static struct llvmpipe_vbuf_render * +llvmpipe_vbuf_render(struct vbuf_render *vbr) +{ + return (struct llvmpipe_vbuf_render *) vbr; +} + + +static const struct vertex_info * +lp_vbuf_get_vertex_info(struct vbuf_render *vbr) +{ + struct llvmpipe_vbuf_render *cvbr = llvmpipe_vbuf_render(vbr); + return llvmpipe_get_vbuf_vertex_info(cvbr->llvmpipe); +} + + +static boolean +lp_vbuf_allocate_vertices(struct vbuf_render *vbr, + ushort vertex_size, ushort nr_vertices) +{ + struct llvmpipe_vbuf_render *cvbr = llvmpipe_vbuf_render(vbr); + unsigned size = vertex_size * nr_vertices; + + if (cvbr->vertex_buffer_size < size) { + align_free(cvbr->vertex_buffer); + cvbr->vertex_buffer = align_malloc(size, 16); + cvbr->vertex_buffer_size = size; + } + + cvbr->vertex_size = vertex_size; + cvbr->nr_vertices = nr_vertices; + + return cvbr->vertex_buffer != NULL; +} + +static void +lp_vbuf_release_vertices(struct vbuf_render *vbr) +{ +#if 0 + { + struct llvmpipe_vbuf_render *cvbr = llvmpipe_vbuf_render(vbr); + const struct vertex_info *info = + llvmpipe_get_vbuf_vertex_info(cvbr->llvmpipe); + const float *vtx = (const float *) cvbr->vertex_buffer; + uint i, j; + debug_printf("%s (vtx_size = %u, vtx_used = %u)\n", + __FUNCTION__, cvbr->vertex_size, cvbr->nr_vertices); + for (i = 0; i < cvbr->nr_vertices; i++) { + for (j = 0; j < info->num_attribs; j++) { + uint k; + switch (info->attrib[j].emit) { + case EMIT_4F: k = 4; break; + case EMIT_3F: k = 3; break; + case EMIT_2F: k = 2; break; + case EMIT_1F: k = 1; break; + default: assert(0); + } + debug_printf("Vert %u attr %u: ", i, j); + while (k-- > 0) { + debug_printf("%g ", vtx[0]); + vtx++; + } + debug_printf("\n"); + } + } + } +#endif + + /* keep the old allocation for next time */ +} + +static void * +lp_vbuf_map_vertices(struct vbuf_render *vbr) +{ + struct llvmpipe_vbuf_render *cvbr = llvmpipe_vbuf_render(vbr); + return cvbr->vertex_buffer; +} + +static void +lp_vbuf_unmap_vertices(struct vbuf_render *vbr, + ushort min_index, + ushort max_index ) +{ + struct llvmpipe_vbuf_render *cvbr = llvmpipe_vbuf_render(vbr); + assert( cvbr->vertex_buffer_size >= (max_index+1) * cvbr->vertex_size ); + /* do nothing */ +} + + +static boolean +lp_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim) +{ + struct llvmpipe_vbuf_render *cvbr = llvmpipe_vbuf_render(vbr); + + /* XXX: break this dependency - make setup_context live under + * llvmpipe, rename the old "setup" draw stage to something else. + */ + struct setup_context *setup_ctx = lp_draw_setup_context(cvbr->llvmpipe->setup); + + llvmpipe_setup_prepare( setup_ctx ); + + cvbr->llvmpipe->reduced_prim = u_reduced_prim(prim); + cvbr->prim = prim; + return TRUE; + +} + + +static INLINE cptrf4 get_vert( const void *vertex_buffer, + int index, + int stride ) +{ + return (cptrf4)((char *)vertex_buffer + index * stride); +} + + +/** + * draw elements / indexed primitives + */ +static void +lp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr) +{ + struct llvmpipe_vbuf_render *cvbr = llvmpipe_vbuf_render(vbr); + struct llvmpipe_context *llvmpipe = cvbr->llvmpipe; + const unsigned stride = llvmpipe->vertex_info_vbuf.size * sizeof(float); + const void *vertex_buffer = cvbr->vertex_buffer; + unsigned i; + + /* XXX: break this dependency - make setup_context live under + * llvmpipe, rename the old "setup" draw stage to something else. + */ + struct draw_stage *setup = llvmpipe->setup; + struct setup_context *setup_ctx = lp_draw_setup_context(setup); + + switch (cvbr->prim) { + case PIPE_PRIM_POINTS: + for (i = 0; i < nr; i++) { + llvmpipe_setup_point( setup_ctx, + get_vert(vertex_buffer, indices[i-0], stride) ); + } + break; + + case PIPE_PRIM_LINES: + for (i = 1; i < nr; i += 2) { + llvmpipe_setup_line( setup_ctx, + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } + break; + + case PIPE_PRIM_LINE_STRIP: + for (i = 1; i < nr; i ++) { + llvmpipe_setup_line( setup_ctx, + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } + break; + + case PIPE_PRIM_LINE_LOOP: + for (i = 1; i < nr; i ++) { + llvmpipe_setup_line( setup_ctx, + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } + if (nr) { + llvmpipe_setup_line( setup_ctx, + get_vert(vertex_buffer, indices[nr-1], stride), + get_vert(vertex_buffer, indices[0], stride) ); + } + break; + + case PIPE_PRIM_TRIANGLES: + if (llvmpipe->rasterizer->flatshade_first) { + for (i = 2; i < nr; i += 3) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride), + get_vert(vertex_buffer, indices[i-2], stride) ); + } + } + else { + for (i = 2; i < nr; i += 3) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-2], stride), + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } + } + break; + + case PIPE_PRIM_TRIANGLE_STRIP: + if (llvmpipe->rasterizer->flatshade_first) { + for (i = 2; i < nr; i += 1) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i+(i&1)-1], stride), + get_vert(vertex_buffer, indices[i-(i&1)], stride), + get_vert(vertex_buffer, indices[i-2], stride) ); + } + } + else { + for (i = 2; i < nr; i += 1) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i+(i&1)-2], stride), + get_vert(vertex_buffer, indices[i-(i&1)-1], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } + } + break; + + case PIPE_PRIM_TRIANGLE_FAN: + if (llvmpipe->rasterizer->flatshade_first) { + for (i = 2; i < nr; i += 1) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-0], stride), + get_vert(vertex_buffer, indices[0], stride), + get_vert(vertex_buffer, indices[i-1], stride) ); + } + } + else { + for (i = 2; i < nr; i += 1) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[0], stride), + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } + } + break; + + case PIPE_PRIM_QUADS: + if (llvmpipe->rasterizer->flatshade_first) { + for (i = 3; i < nr; i += 4) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-2], stride), + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-3], stride) ); + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride), + get_vert(vertex_buffer, indices[i-3], stride) ); + } + } + else { + for (i = 3; i < nr; i += 4) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-3], stride), + get_vert(vertex_buffer, indices[i-2], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-2], stride), + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } + } + break; + + case PIPE_PRIM_QUAD_STRIP: + if (llvmpipe->rasterizer->flatshade_first) { + for (i = 3; i < nr; i += 2) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-0], stride), + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-3], stride)); + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-2], stride), + get_vert(vertex_buffer, indices[i-0], stride), + get_vert(vertex_buffer, indices[i-3], stride) ); + } + } + else { + for (i = 3; i < nr; i += 2) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-3], stride), + get_vert(vertex_buffer, indices[i-2], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-3], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } + } + break; + + case PIPE_PRIM_POLYGON: + /* Almost same as tri fan but the _first_ vertex specifies the flat + * shading color. Note that the first polygon vertex is passed as + * the last triangle vertex here. + * flatshade_first state makes no difference. + */ + for (i = 2; i < nr; i += 1) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-0], stride), + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[0], stride) ); + } + break; + + default: + assert(0); + } + + /* XXX: why are we calling this??? If we had to call something, it + * would be a function in lp_setup.c: + */ + lp_draw_flush( setup ); +} + + +/** + * This function is hit when the draw module is working in pass-through mode. + * It's up to us to convert the vertex array into point/line/tri prims. + */ +static void +lp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr) +{ + struct llvmpipe_vbuf_render *cvbr = llvmpipe_vbuf_render(vbr); + struct llvmpipe_context *llvmpipe = cvbr->llvmpipe; + const unsigned stride = llvmpipe->vertex_info_vbuf.size * sizeof(float); + const void *vertex_buffer = + (void *) get_vert(cvbr->vertex_buffer, start, stride); + unsigned i; + + /* XXX: break this dependency - make setup_context live under + * llvmpipe, rename the old "setup" draw stage to something else. + */ + struct draw_stage *setup = llvmpipe->setup; + struct setup_context *setup_ctx = lp_draw_setup_context(setup); + + switch (cvbr->prim) { + case PIPE_PRIM_POINTS: + for (i = 0; i < nr; i++) { + llvmpipe_setup_point( setup_ctx, + get_vert(vertex_buffer, i-0, stride) ); + } + break; + + case PIPE_PRIM_LINES: + for (i = 1; i < nr; i += 2) { + llvmpipe_setup_line( setup_ctx, + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride) ); + } + break; + + case PIPE_PRIM_LINE_STRIP: + for (i = 1; i < nr; i ++) { + llvmpipe_setup_line( setup_ctx, + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride) ); + } + break; + + case PIPE_PRIM_LINE_LOOP: + for (i = 1; i < nr; i ++) { + llvmpipe_setup_line( setup_ctx, + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride) ); + } + if (nr) { + llvmpipe_setup_line( setup_ctx, + get_vert(vertex_buffer, nr-1, stride), + get_vert(vertex_buffer, 0, stride) ); + } + break; + + case PIPE_PRIM_TRIANGLES: + if (llvmpipe->rasterizer->flatshade_first) { + for (i = 2; i < nr; i += 3) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride), + get_vert(vertex_buffer, i-2, stride) ); + } + } + else { + for (i = 2; i < nr; i += 3) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i-2, stride), + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride) ); + } + } + break; + + case PIPE_PRIM_TRIANGLE_STRIP: + if (llvmpipe->rasterizer->flatshade_first) { + for (i = 2; i < nr; i++) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i+(i&1)-1, stride), + get_vert(vertex_buffer, i-(i&1), stride), + get_vert(vertex_buffer, i-2, stride) ); + } + } + else { + for (i = 2; i < nr; i++) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i+(i&1)-2, stride), + get_vert(vertex_buffer, i-(i&1)-1, stride), + get_vert(vertex_buffer, i-0, stride) ); + } + } + break; + + case PIPE_PRIM_TRIANGLE_FAN: + if (llvmpipe->rasterizer->flatshade_first) { + for (i = 2; i < nr; i += 1) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i-0, stride), + get_vert(vertex_buffer, 0, stride), + get_vert(vertex_buffer, i-1, stride) ); + } + } + else { + for (i = 2; i < nr; i += 1) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, 0, stride), + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride) ); + } + } + break; + + case PIPE_PRIM_QUADS: + if (llvmpipe->rasterizer->flatshade_first) { + for (i = 3; i < nr; i += 4) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i-2, stride), + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-3, stride) ); + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride), + get_vert(vertex_buffer, i-3, stride) ); + } + } + else { + for (i = 3; i < nr; i += 4) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i-3, stride), + get_vert(vertex_buffer, i-2, stride), + get_vert(vertex_buffer, i-0, stride) ); + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i-2, stride), + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride) ); + } + } + break; + + case PIPE_PRIM_QUAD_STRIP: + if (llvmpipe->rasterizer->flatshade_first) { + for (i = 3; i < nr; i += 2) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i-0, stride), + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-3, stride) ); + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i-2, stride), + get_vert(vertex_buffer, i-0, stride), + get_vert(vertex_buffer, i-3, stride) ); + } + } + else { + for (i = 3; i < nr; i += 2) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i-3, stride), + get_vert(vertex_buffer, i-2, stride), + get_vert(vertex_buffer, i-0, stride) ); + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-3, stride), + get_vert(vertex_buffer, i-0, stride) ); + } + } + break; + + case PIPE_PRIM_POLYGON: + /* Almost same as tri fan but the _first_ vertex specifies the flat + * shading color. Note that the first polygon vertex is passed as + * the last triangle vertex here. + * flatshade_first state makes no difference. + */ + for (i = 2; i < nr; i += 1) { + llvmpipe_setup_tri( setup_ctx, + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride), + get_vert(vertex_buffer, 0, stride) ); + } + break; + + default: + assert(0); + } +} + + + +static void +lp_vbuf_destroy(struct vbuf_render *vbr) +{ + struct llvmpipe_vbuf_render *cvbr = llvmpipe_vbuf_render(vbr); + cvbr->llvmpipe->vbuf_render = NULL; + FREE(cvbr); +} + + +/** + * Initialize the post-transform vertex buffer information for the given + * context. + */ +void +lp_init_vbuf(struct llvmpipe_context *lp) +{ + assert(lp->draw); + + lp->vbuf_render = CALLOC_STRUCT(llvmpipe_vbuf_render); + + lp->vbuf_render->base.max_indices = LP_MAX_VBUF_INDEXES; + lp->vbuf_render->base.max_vertex_buffer_bytes = LP_MAX_VBUF_SIZE; + + lp->vbuf_render->base.get_vertex_info = lp_vbuf_get_vertex_info; + lp->vbuf_render->base.allocate_vertices = lp_vbuf_allocate_vertices; + lp->vbuf_render->base.map_vertices = lp_vbuf_map_vertices; + lp->vbuf_render->base.unmap_vertices = lp_vbuf_unmap_vertices; + lp->vbuf_render->base.set_primitive = lp_vbuf_set_primitive; + lp->vbuf_render->base.draw = lp_vbuf_draw; + lp->vbuf_render->base.draw_arrays = lp_vbuf_draw_arrays; + lp->vbuf_render->base.release_vertices = lp_vbuf_release_vertices; + lp->vbuf_render->base.destroy = lp_vbuf_destroy; + + lp->vbuf_render->llvmpipe = lp; + + lp->vbuf = draw_vbuf_stage(lp->draw, &lp->vbuf_render->base); + + draw_set_rasterize_stage(lp->draw, lp->vbuf); + + draw_set_render(lp->draw, &lp->vbuf_render->base); +} diff --git a/src/gallium/drivers/llvmpipe/lp_prim_vbuf.h b/src/gallium/drivers/llvmpipe/lp_prim_vbuf.h new file mode 100644 index 00000000000..6c4e6063e6d --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_prim_vbuf.h @@ -0,0 +1,38 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef LP_VBUF_H +#define LP_VBUF_H + + +struct llvmpipe_context; + +extern void +lp_init_vbuf(struct llvmpipe_context *llvmpipe); + + +#endif /* LP_VBUF_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_quad.h b/src/gallium/drivers/llvmpipe/lp_quad.h new file mode 100644 index 00000000000..7eb05de77a1 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_quad.h @@ -0,0 +1,114 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef LP_QUAD_H +#define LP_QUAD_H + +#include "pipe/p_state.h" +#include "tgsi/tgsi_exec.h" + + +#define QUAD_PRIM_POINT 1 +#define QUAD_PRIM_LINE 2 +#define QUAD_PRIM_TRI 3 + + +/* The rasterizer generates 2x2 quads of fragment and feeds them to + * the current fp_machine (see below). + * Remember that Y=0=top with Y increasing down the window. + */ +#define QUAD_TOP_LEFT 0 +#define QUAD_TOP_RIGHT 1 +#define QUAD_BOTTOM_LEFT 2 +#define QUAD_BOTTOM_RIGHT 3 + +#define MASK_TOP_LEFT (1 << QUAD_TOP_LEFT) +#define MASK_TOP_RIGHT (1 << QUAD_TOP_RIGHT) +#define MASK_BOTTOM_LEFT (1 << QUAD_BOTTOM_LEFT) +#define MASK_BOTTOM_RIGHT (1 << QUAD_BOTTOM_RIGHT) +#define MASK_ALL 0xf + + +/** + * Quad stage inputs (pos, coverage, front/back face, etc) + */ +struct quad_header_input +{ + int x0, y0; /**< quad window pos, always even */ + float coverage[QUAD_SIZE]; /**< fragment coverage for antialiasing */ + unsigned facing:1; /**< Front (0) or back (1) facing? */ + unsigned prim:2; /**< QUAD_PRIM_POINT, LINE, TRI */ +}; + + +/** + * Quad stage inputs/outputs. + */ +struct quad_header_inout +{ + unsigned mask:4; +}; + + +/** + * Quad stage outputs (color & depth). + */ +struct quad_header_output +{ + /** colors in SOA format (rrrr, gggg, bbbb, aaaa) */ + float ALIGN16_ATTRIB color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS][QUAD_SIZE]; +}; + + +/** + * Input interpolation coefficients + */ +struct quad_interp_coef +{ + float ALIGN16_ATTRIB a0[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]; + float ALIGN16_ATTRIB dadx[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]; + float ALIGN16_ATTRIB dady[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]; +}; + + +/** + * Encodes everything we need to know about a 2x2 pixel block. Uses + * "Channel-Serial" or "SoA" layout. + */ +struct quad_header { + struct quad_header_input input; + struct quad_header_inout inout; + + /* Redundant/duplicated: + */ + const struct quad_interp_coef *coef; +}; + +#endif /* LP_QUAD_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_query.c b/src/gallium/drivers/llvmpipe/lp_query.c new file mode 100644 index 00000000000..5554285425d --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_query.c @@ -0,0 +1,111 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "draw/draw_context.h" +#include "pipe/p_defines.h" +#include "util/u_memory.h" +#include "lp_context.h" +#include "lp_query.h" +#include "lp_state.h" + +struct llvmpipe_query { + uint64_t start; + uint64_t end; +}; + + +static struct llvmpipe_query *llvmpipe_query( struct pipe_query *p ) +{ + return (struct llvmpipe_query *)p; +} + +static struct pipe_query * +llvmpipe_create_query(struct pipe_context *pipe, + unsigned type) +{ + assert(type == PIPE_QUERY_OCCLUSION_COUNTER); + return (struct pipe_query *)CALLOC_STRUCT( llvmpipe_query ); +} + + +static void +llvmpipe_destroy_query(struct pipe_context *pipe, struct pipe_query *q) +{ + FREE(q); +} + + +static void +llvmpipe_begin_query(struct pipe_context *pipe, struct pipe_query *q) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context( pipe ); + struct llvmpipe_query *sq = llvmpipe_query(q); + + sq->start = llvmpipe->occlusion_count; + llvmpipe->active_query_count++; + llvmpipe->dirty |= LP_NEW_QUERY; +} + + +static void +llvmpipe_end_query(struct pipe_context *pipe, struct pipe_query *q) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context( pipe ); + struct llvmpipe_query *sq = llvmpipe_query(q); + + llvmpipe->active_query_count--; + sq->end = llvmpipe->occlusion_count; + llvmpipe->dirty |= LP_NEW_QUERY; +} + + +static boolean +llvmpipe_get_query_result(struct pipe_context *pipe, + struct pipe_query *q, + boolean wait, + uint64_t *result ) +{ + struct llvmpipe_query *sq = llvmpipe_query(q); + *result = sq->end - sq->start; + return TRUE; +} + + +void llvmpipe_init_query_funcs(struct llvmpipe_context *llvmpipe ) +{ + llvmpipe->pipe.create_query = llvmpipe_create_query; + llvmpipe->pipe.destroy_query = llvmpipe_destroy_query; + llvmpipe->pipe.begin_query = llvmpipe_begin_query; + llvmpipe->pipe.end_query = llvmpipe_end_query; + llvmpipe->pipe.get_query_result = llvmpipe_get_query_result; +} + + diff --git a/src/gallium/drivers/llvmpipe/lp_query.h b/src/gallium/drivers/llvmpipe/lp_query.h new file mode 100644 index 00000000000..fa9fcd87139 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_query.h @@ -0,0 +1,39 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Keith Whitwell + */ + +#ifndef LP_QUERY_H +#define LP_QUERY_H + +struct llvmpipe_context; +extern void llvmpipe_init_query_funcs(struct llvmpipe_context * ); + + +#endif /* LP_QUERY_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c new file mode 100644 index 00000000000..125035771e5 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_screen.c @@ -0,0 +1,243 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "util/u_memory.h" +#include "util/u_simple_screen.h" +#include "pipe/internal/p_winsys_screen.h" +#include "pipe/p_defines.h" +#include "pipe/p_screen.h" + +#include "lp_texture.h" +#include "lp_buffer.h" +#include "lp_winsys.h" +#include "lp_jit.h" +#include "lp_screen.h" + + +static const char * +llvmpipe_get_vendor(struct pipe_screen *screen) +{ + return "VMware, Inc."; +} + + +static const char * +llvmpipe_get_name(struct pipe_screen *screen) +{ + return "llvmpipe"; +} + + +static int +llvmpipe_get_param(struct pipe_screen *screen, int param) +{ + switch (param) { + case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS: + return PIPE_MAX_SAMPLERS; + case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS: + return PIPE_MAX_SAMPLERS; + case PIPE_CAP_NPOT_TEXTURES: + return 1; + case PIPE_CAP_TWO_SIDED_STENCIL: + return 1; + case PIPE_CAP_GLSL: + return 1; + case PIPE_CAP_S3TC: + return 0; + case PIPE_CAP_ANISOTROPIC_FILTER: + return 0; + case PIPE_CAP_POINT_SPRITE: + return 1; + case PIPE_CAP_MAX_RENDER_TARGETS: + return PIPE_MAX_COLOR_BUFS; + case PIPE_CAP_OCCLUSION_QUERY: + return 1; + case PIPE_CAP_TEXTURE_MIRROR_CLAMP: + return 1; + case PIPE_CAP_TEXTURE_MIRROR_REPEAT: + return 1; + case PIPE_CAP_TEXTURE_SHADOW_MAP: + return 1; + case PIPE_CAP_MAX_TEXTURE_2D_LEVELS: + return 13; /* max 4Kx4K */ + case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: + return 8; /* max 128x128x128 */ + case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS: + return 13; /* max 4Kx4K */ + case PIPE_CAP_TGSI_CONT_SUPPORTED: + return 1; + case PIPE_CAP_BLEND_EQUATION_SEPARATE: + return 1; + default: + return 0; + } +} + + +static float +llvmpipe_get_paramf(struct pipe_screen *screen, int param) +{ + switch (param) { + case PIPE_CAP_MAX_LINE_WIDTH: + /* fall-through */ + case PIPE_CAP_MAX_LINE_WIDTH_AA: + return 255.0; /* arbitrary */ + case PIPE_CAP_MAX_POINT_WIDTH: + /* fall-through */ + case PIPE_CAP_MAX_POINT_WIDTH_AA: + return 255.0; /* arbitrary */ + case PIPE_CAP_MAX_TEXTURE_ANISOTROPY: + return 16.0; /* not actually signficant at this time */ + case PIPE_CAP_MAX_TEXTURE_LOD_BIAS: + return 16.0; /* arbitrary */ + default: + return 0; + } +} + + +/** + * Query format support for creating a texture, drawing surface, etc. + * \param format the format to test + * \param type one of PIPE_TEXTURE, PIPE_SURFACE + */ +static boolean +llvmpipe_is_format_supported( struct pipe_screen *_screen, + enum pipe_format format, + enum pipe_texture_target target, + unsigned tex_usage, + unsigned geom_flags ) +{ + struct llvmpipe_screen *screen = llvmpipe_screen(_screen); + struct llvmpipe_winsys *winsys = screen->winsys; + + assert(target == PIPE_TEXTURE_1D || + target == PIPE_TEXTURE_2D || + target == PIPE_TEXTURE_3D || + target == PIPE_TEXTURE_CUBE); + + if(format == PIPE_FORMAT_Z16_UNORM) + return FALSE; + if(format == PIPE_FORMAT_S8_UNORM) + return FALSE; + + switch(format) { + case PIPE_FORMAT_DXT1_RGB: + case PIPE_FORMAT_DXT1_RGBA: + case PIPE_FORMAT_DXT3_RGBA: + case PIPE_FORMAT_DXT5_RGBA: + return FALSE; + default: + break; + } + + if(tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET) + return winsys->is_displaytarget_format_supported(winsys, format); + + return TRUE; +} + + +static struct pipe_buffer * +llvmpipe_surface_buffer_create(struct pipe_screen *screen, + unsigned width, unsigned height, + enum pipe_format format, + unsigned tex_usage, + unsigned usage, + unsigned *stride) +{ + /* This function should never be used */ + assert(0); + return NULL; +} + + +static void +llvmpipe_flush_frontbuffer(struct pipe_screen *_screen, + struct pipe_surface *surface, + void *context_private) +{ + struct llvmpipe_screen *screen = llvmpipe_screen(_screen); + struct llvmpipe_winsys *winsys = screen->winsys; + struct llvmpipe_texture *texture = llvmpipe_texture(surface->texture); + + assert(texture->dt); + if (texture->dt) + winsys->displaytarget_display(winsys, texture->dt, context_private); +} + + +static void +llvmpipe_destroy_screen( struct pipe_screen *_screen ) +{ + struct llvmpipe_screen *screen = llvmpipe_screen(_screen); + + struct pipe_winsys *winsys = _screen->winsys; + + lp_jit_screen_cleanup(screen); + + if(winsys->destroy) + winsys->destroy(winsys); + + FREE(screen); +} + + + +/** + * Create a new pipe_screen object + * Note: we're not presently subclassing pipe_screen (no llvmpipe_screen). + */ +struct pipe_screen * +llvmpipe_create_screen(struct llvmpipe_winsys *winsys) +{ + struct llvmpipe_screen *screen = CALLOC_STRUCT(llvmpipe_screen); + + if (!screen) + return NULL; + + screen->winsys = winsys; + + screen->base.destroy = llvmpipe_destroy_screen; + + screen->base.get_name = llvmpipe_get_name; + screen->base.get_vendor = llvmpipe_get_vendor; + screen->base.get_param = llvmpipe_get_param; + screen->base.get_paramf = llvmpipe_get_paramf; + screen->base.is_format_supported = llvmpipe_is_format_supported; + + screen->base.surface_buffer_create = llvmpipe_surface_buffer_create; + screen->base.flush_frontbuffer = llvmpipe_flush_frontbuffer; + + llvmpipe_init_screen_texture_funcs(&screen->base); + llvmpipe_init_screen_buffer_funcs(&screen->base); + + lp_jit_screen_init(screen); + + return &screen->base; +} diff --git a/src/gallium/winsys/drm/nouveau/dri/nouveau_lock.c b/src/gallium/drivers/llvmpipe/lp_screen.h index 92f5bd09c9d..4a1b4d6f3e2 100644 --- a/src/gallium/winsys/drm/nouveau/dri/nouveau_lock.c +++ b/src/gallium/drivers/llvmpipe/lp_screen.h @@ -1,8 +1,9 @@ /************************************************************************** * - * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * Copyright 2009 VMware, Inc. + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. * All Rights Reserved. - * + * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the * "Software"), to deal in the Software without restriction, including @@ -25,49 +26,54 @@ * **************************************************************************/ -#include <pipe/p_thread.h> -#include "nouveau_context.h" -#include "nouveau_screen.h" -#include "nouveau_drmif.h" +/** + * @author Jose Fonseca <jfonseca@vmware.com> + * @author Keith Whitwell <keith@tungstengraphics.com> + */ -pipe_static_mutex(lockMutex); +#ifndef LP_SCREEN_H +#define LP_SCREEN_H -/* Lock the hardware and validate our state. - */ -void -LOCK_HARDWARE(struct nouveau_context *nv) -{ - struct nouveau_screen *nv_screen = nv->dri_screen->private; - struct nouveau_device *dev = nv_screen->device; - struct nouveau_device_priv *nvdev = nouveau_device(dev); - char __ret=0; - - assert(!nv->locked); - pipe_mutex_lock(lockMutex); - - DRM_CAS(nvdev->lock, nvdev->ctx, - (DRM_LOCK_HELD | nvdev->ctx), __ret); - - if (__ret) { - drmGetLock(nvdev->fd, nvdev->ctx, 0); - nouveau_contended_lock(nv); - } - nv->locked = 1; -} +#include <llvm-c/Core.h> +#include <llvm-c/Analysis.h> +#include <llvm-c/Target.h> +#include <llvm-c/ExecutionEngine.h> -/* Unlock the hardware using the global current context - */ -void -UNLOCK_HARDWARE(struct nouveau_context *nv) +#include "pipe/p_screen.h" +#include "pipe/p_defines.h" + + +struct llvmpipe_winsys; + + +struct llvmpipe_screen { - struct nouveau_screen *nv_screen = nv->dri_screen->private; - struct nouveau_device *dev = nv_screen->device; - struct nouveau_device_priv *nvdev = nouveau_device(dev); + struct pipe_screen base; - assert(nv->locked); - nv->locked = 0; + struct llvmpipe_winsys *winsys; + + LLVMModuleRef module; + LLVMExecutionEngineRef engine; + LLVMModuleProviderRef provider; + LLVMTargetDataRef target; + LLVMPassManagerRef pass; + + LLVMTypeRef context_ptr_type; + + /* Increments whenever textures are modified. Contexts can track + * this. + */ + unsigned timestamp; +}; + + + + +static INLINE struct llvmpipe_screen * +llvmpipe_screen( struct pipe_screen *pipe ) +{ + return (struct llvmpipe_screen *)pipe; +} - DRM_UNLOCK(nvdev->fd, nvdev->lock, nvdev->ctx); - pipe_mutex_unlock(lockMutex); -} +#endif /* LP_SCREEN_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_setup.c b/src/gallium/drivers/llvmpipe/lp_setup.c new file mode 100644 index 00000000000..d145f6d6bbc --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_setup.c @@ -0,0 +1,1482 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * \brief Primitive rasterization/rendering (points, lines, triangles) + * + * \author Keith Whitwell <keith@tungstengraphics.com> + * \author Brian Paul + */ + +#include "lp_context.h" +#include "lp_prim_setup.h" +#include "lp_quad.h" +#include "lp_setup.h" +#include "lp_state.h" +#include "draw/draw_context.h" +#include "draw/draw_private.h" +#include "draw/draw_vertex.h" +#include "pipe/p_shader_tokens.h" +#include "pipe/p_thread.h" +#include "util/u_math.h" +#include "util/u_memory.h" +#include "lp_tile_cache.h" +#include "lp_tile_soa.h" + + +#define DEBUG_VERTS 0 +#define DEBUG_FRAGS 0 + +/** + * Triangle edge info + */ +struct edge { + float dx; /**< X(v1) - X(v0), used only during setup */ + float dy; /**< Y(v1) - Y(v0), used only during setup */ + float dxdy; /**< dx/dy */ + float sx, sy; /**< first sample point coord */ + int lines; /**< number of lines on this edge */ +}; + + +#define MAX_QUADS 16 + + +/** + * Triangle setup info (derived from draw_stage). + * Also used for line drawing (taking some liberties). + */ +struct setup_context { + struct llvmpipe_context *llvmpipe; + + /* Vertices are just an array of floats making up each attribute in + * turn. Currently fixed at 4 floats, but should change in time. + * Codegen will help cope with this. + */ + const float (*vmax)[4]; + const float (*vmid)[4]; + const float (*vmin)[4]; + const float (*vprovoke)[4]; + + struct edge ebot; + struct edge etop; + struct edge emaj; + + float oneoverarea; + int facing; + + struct quad_header quad[MAX_QUADS]; + struct quad_header *quad_ptrs[MAX_QUADS]; + unsigned count; + + struct quad_interp_coef coef; + + struct { + int left[2]; /**< [0] = row0, [1] = row1 */ + int right[2]; + int y; + } span; + +#if DEBUG_FRAGS + uint numFragsEmitted; /**< per primitive */ + uint numFragsWritten; /**< per primitive */ +#endif + + unsigned winding; /* which winding to cull */ +}; + + + +/** + * Execute fragment shader for the four fragments in the quad. + */ +static void +shade_quads(struct llvmpipe_context *llvmpipe, + struct quad_header *quads[], + unsigned nr) +{ + struct lp_fragment_shader *fs = llvmpipe->fs; + struct quad_header *quad = quads[0]; + const unsigned x = quad->input.x0; + const unsigned y = quad->input.y0; + uint8_t *tile = lp_get_cached_tile(llvmpipe->cbuf_cache[0], x, y); + uint8_t *color; + void *depth; + uint32_t ALIGN16_ATTRIB mask[4][NUM_CHANNELS]; + unsigned chan_index; + unsigned q; + + assert(fs->current); + if(!fs->current) + return; + + /* Sanity checks */ + assert(nr * QUAD_SIZE == TILE_VECTOR_HEIGHT * TILE_VECTOR_WIDTH); + assert(x % TILE_VECTOR_WIDTH == 0); + assert(y % TILE_VECTOR_HEIGHT == 0); + for (q = 0; q < nr; ++q) { + assert(quads[q]->input.x0 == x + q*2); + assert(quads[q]->input.y0 == y); + } + + /* mask */ + for (q = 0; q < 4; ++q) + for (chan_index = 0; chan_index < NUM_CHANNELS; ++chan_index) + mask[q][chan_index] = quads[q]->inout.mask & (1 << chan_index) ? ~0 : 0; + + /* color buffer */ + color = &TILE_PIXEL(tile, x & (TILE_SIZE-1), y & (TILE_SIZE-1), 0); + + /* depth buffer */ + if(llvmpipe->zsbuf_map) { + assert((x % 2) == 0); + assert((y % 2) == 0); + depth = llvmpipe->zsbuf_map + + y*llvmpipe->zsbuf_transfer->stride + + 2*x*llvmpipe->zsbuf_transfer->block.size; + } + else + depth = NULL; + + /* TODO: blend color */ + + assert((((uintptr_t)mask) & 0xf) == 0); + assert((((uintptr_t)depth) & 0xf) == 0); + assert((((uintptr_t)color) & 0xf) == 0); + assert((((uintptr_t)llvmpipe->jit_context.blend_color) & 0xf) == 0); + + /* run shader */ + fs->current->jit_function( &llvmpipe->jit_context, + x, y, + quad->coef->a0, + quad->coef->dadx, + quad->coef->dady, + &mask[0][0], + color, + depth); +} + + + + +/** + * Do triangle cull test using tri determinant (sign indicates orientation) + * \return true if triangle is to be culled. + */ +static INLINE boolean +cull_tri(const struct setup_context *setup, float det) +{ + if (det != 0) { + /* if (det < 0 then Z points toward camera and triangle is + * counter-clockwise winding. + */ + unsigned winding = (det < 0) ? PIPE_WINDING_CCW : PIPE_WINDING_CW; + + if ((winding & setup->winding) == 0) + return FALSE; + } + + /* Culled: + */ + return TRUE; +} + + + +/** + * Clip setup->quad against the scissor/surface bounds. + */ +static INLINE void +quad_clip( struct setup_context *setup, struct quad_header *quad ) +{ + const struct pipe_scissor_state *cliprect = &setup->llvmpipe->cliprect; + const int minx = (int) cliprect->minx; + const int maxx = (int) cliprect->maxx; + const int miny = (int) cliprect->miny; + const int maxy = (int) cliprect->maxy; + + if (quad->input.x0 >= maxx || + quad->input.y0 >= maxy || + quad->input.x0 + 1 < minx || + quad->input.y0 + 1 < miny) { + /* totally clipped */ + quad->inout.mask = 0x0; + return; + } + if (quad->input.x0 < minx) + quad->inout.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT); + if (quad->input.y0 < miny) + quad->inout.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT); + if (quad->input.x0 == maxx - 1) + quad->inout.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT); + if (quad->input.y0 == maxy - 1) + quad->inout.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT); +} + + + +/** + * Given an X or Y coordinate, return the block/quad coordinate that it + * belongs to. + */ +static INLINE int block( int x ) +{ + return x & ~(2-1); +} + +static INLINE int block_x( int x ) +{ + return x & ~(TILE_VECTOR_WIDTH - 1); +} + + +/** + * Emit a quad (pass to next stage) with clipping. + */ +static INLINE void +clip_emit_quad( struct setup_context *setup, struct quad_header *quad ) +{ + quad_clip( setup, quad ); + + if (quad->inout.mask) { + struct llvmpipe_context *lp = setup->llvmpipe; + +#if 1 + /* XXX: The blender expects 4 quads. This is far from efficient, but + * until we codegenerate single-quad variants of the fragment pipeline + * we need this hack. */ + const unsigned nr_quads = TILE_VECTOR_HEIGHT*TILE_VECTOR_WIDTH/QUAD_SIZE; + struct quad_header quads[nr_quads]; + struct quad_header *quad_ptrs[nr_quads]; + int x0 = block_x(quad->input.x0); + unsigned i; + + for(i = 0; i < nr_quads; ++i) { + int x = x0 + 2*i; + if(x == quad->input.x0) + memcpy(&quads[i], quad, sizeof quads[i]); + else { + memset(&quads[i], 0, sizeof quads[i]); + quads[i].input.x0 = x; + quads[i].input.y0 = quad->input.y0; + quads[i].coef = quad->coef; + } + quad_ptrs[i] = &quads[i]; + } + + shade_quads( lp, quad_ptrs, nr_quads ); +#else + shade_quads( lp, &quad, 1 ); +#endif + } +} + + +/** + * Render a horizontal span of quads + */ +static void flush_spans( struct setup_context *setup ) +{ + const int step = TILE_VECTOR_WIDTH; + const int xleft0 = setup->span.left[0]; + const int xleft1 = setup->span.left[1]; + const int xright0 = setup->span.right[0]; + const int xright1 = setup->span.right[1]; + + + int minleft = block_x(MIN2(xleft0, xleft1)); + int maxright = MAX2(xright0, xright1); + int x; + + for (x = minleft; x < maxright; x += step) { + unsigned skip_left0 = CLAMP(xleft0 - x, 0, step); + unsigned skip_left1 = CLAMP(xleft1 - x, 0, step); + unsigned skip_right0 = CLAMP(x + step - xright0, 0, step); + unsigned skip_right1 = CLAMP(x + step - xright1, 0, step); + unsigned lx = x; + const unsigned nr_quads = TILE_VECTOR_HEIGHT*TILE_VECTOR_WIDTH/QUAD_SIZE; + unsigned q = 0; + + unsigned skipmask_left0 = (1U << skip_left0) - 1U; + unsigned skipmask_left1 = (1U << skip_left1) - 1U; + + /* These calculations fail when step == 32 and skip_right == 0. + */ + unsigned skipmask_right0 = ~0U << (unsigned)(step - skip_right0); + unsigned skipmask_right1 = ~0U << (unsigned)(step - skip_right1); + + unsigned mask0 = ~skipmask_left0 & ~skipmask_right0; + unsigned mask1 = ~skipmask_left1 & ~skipmask_right1; + + if (mask0 | mask1) { + for(q = 0; q < nr_quads; ++q) { + unsigned quadmask = (mask0 & 3) | ((mask1 & 3) << 2); + setup->quad[q].input.x0 = lx; + setup->quad[q].input.y0 = setup->span.y; + setup->quad[q].inout.mask = quadmask; + setup->quad_ptrs[q] = &setup->quad[q]; + mask0 >>= 2; + mask1 >>= 2; + lx += 2; + } + assert(!(mask0 | mask1)); + + shade_quads(setup->llvmpipe, setup->quad_ptrs, nr_quads ); + } + } + + + setup->span.y = 0; + setup->span.right[0] = 0; + setup->span.right[1] = 0; + setup->span.left[0] = 1000000; /* greater than right[0] */ + setup->span.left[1] = 1000000; /* greater than right[1] */ +} + + +#if DEBUG_VERTS +static void print_vertex(const struct setup_context *setup, + const float (*v)[4]) +{ + int i; + debug_printf(" Vertex: (%p)\n", v); + for (i = 0; i < setup->quad[0].nr_attrs; i++) { + debug_printf(" %d: %f %f %f %f\n", i, + v[i][0], v[i][1], v[i][2], v[i][3]); + if (util_is_inf_or_nan(v[i][0])) { + debug_printf(" NaN!\n"); + } + } +} +#endif + +/** + * Sort the vertices from top to bottom order, setting up the triangle + * edge fields (ebot, emaj, etop). + * \return FALSE if coords are inf/nan (cull the tri), TRUE otherwise + */ +static boolean setup_sort_vertices( struct setup_context *setup, + float det, + const float (*v0)[4], + const float (*v1)[4], + const float (*v2)[4] ) +{ + setup->vprovoke = v2; + + /* determine bottom to top order of vertices */ + { + float y0 = v0[0][1]; + float y1 = v1[0][1]; + float y2 = v2[0][1]; + if (y0 <= y1) { + if (y1 <= y2) { + /* y0<=y1<=y2 */ + setup->vmin = v0; + setup->vmid = v1; + setup->vmax = v2; + } + else if (y2 <= y0) { + /* y2<=y0<=y1 */ + setup->vmin = v2; + setup->vmid = v0; + setup->vmax = v1; + } + else { + /* y0<=y2<=y1 */ + setup->vmin = v0; + setup->vmid = v2; + setup->vmax = v1; + } + } + else { + if (y0 <= y2) { + /* y1<=y0<=y2 */ + setup->vmin = v1; + setup->vmid = v0; + setup->vmax = v2; + } + else if (y2 <= y1) { + /* y2<=y1<=y0 */ + setup->vmin = v2; + setup->vmid = v1; + setup->vmax = v0; + } + else { + /* y1<=y2<=y0 */ + setup->vmin = v1; + setup->vmid = v2; + setup->vmax = v0; + } + } + } + + setup->ebot.dx = setup->vmid[0][0] - setup->vmin[0][0]; + setup->ebot.dy = setup->vmid[0][1] - setup->vmin[0][1]; + setup->emaj.dx = setup->vmax[0][0] - setup->vmin[0][0]; + setup->emaj.dy = setup->vmax[0][1] - setup->vmin[0][1]; + setup->etop.dx = setup->vmax[0][0] - setup->vmid[0][0]; + setup->etop.dy = setup->vmax[0][1] - setup->vmid[0][1]; + + /* + * Compute triangle's area. Use 1/area to compute partial + * derivatives of attributes later. + * + * The area will be the same as prim->det, but the sign may be + * different depending on how the vertices get sorted above. + * + * To determine whether the primitive is front or back facing we + * use the prim->det value because its sign is correct. + */ + { + const float area = (setup->emaj.dx * setup->ebot.dy - + setup->ebot.dx * setup->emaj.dy); + + setup->oneoverarea = 1.0f / area; + + /* + debug_printf("%s one-over-area %f area %f det %f\n", + __FUNCTION__, setup->oneoverarea, area, det ); + */ + if (util_is_inf_or_nan(setup->oneoverarea)) + return FALSE; + } + + /* We need to know if this is a front or back-facing triangle for: + * - the GLSL gl_FrontFacing fragment attribute (bool) + * - two-sided stencil test + */ + setup->facing = + ((det > 0.0) ^ + (setup->llvmpipe->rasterizer->front_winding == PIPE_WINDING_CW)); + + return TRUE; +} + + +/** + * Compute a0, dadx and dady for a linearly interpolated coefficient, + * for a triangle. + */ +static void tri_pos_coeff( struct setup_context *setup, + uint vertSlot, unsigned i) +{ + float botda = setup->vmid[vertSlot][i] - setup->vmin[vertSlot][i]; + float majda = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i]; + float a = setup->ebot.dy * majda - botda * setup->emaj.dy; + float b = setup->emaj.dx * botda - majda * setup->ebot.dx; + float dadx = a * setup->oneoverarea; + float dady = b * setup->oneoverarea; + + assert(i <= 3); + + setup->coef.dadx[0][i] = dadx; + setup->coef.dady[0][i] = dady; + + /* calculate a0 as the value which would be sampled for the + * fragment at (0,0), taking into account that we want to sample at + * pixel centers, in other words (0.5, 0.5). + * + * this is neat but unfortunately not a good way to do things for + * triangles with very large values of dadx or dady as it will + * result in the subtraction and re-addition from a0 of a very + * large number, which means we'll end up loosing a lot of the + * fractional bits and precision from a0. the way to fix this is + * to define a0 as the sample at a pixel center somewhere near vmin + * instead - i'll switch to this later. + */ + setup->coef.a0[0][i] = (setup->vmin[vertSlot][i] - + (dadx * (setup->vmin[0][0] - 0.5f) + + dady * (setup->vmin[0][1] - 0.5f))); + + /* + debug_printf("attr[%d].%c: %f dx:%f dy:%f\n", + slot, "xyzw"[i], + setup->coef[slot].a0[i], + setup->coef[slot].dadx[i], + setup->coef[slot].dady[i]); + */ +} + + +/** + * Compute a0 for a constant-valued coefficient (GL_FLAT shading). + * The value value comes from vertex[slot][i]. + * The result will be put into setup->coef[slot].a0[i]. + * \param slot which attribute slot + * \param i which component of the slot (0..3) + */ +static void const_pos_coeff( struct setup_context *setup, + uint vertSlot, unsigned i) +{ + setup->coef.dadx[0][i] = 0; + setup->coef.dady[0][i] = 0; + + /* need provoking vertex info! + */ + setup->coef.a0[0][i] = setup->vprovoke[vertSlot][i]; +} + + +/** + * Compute a0 for a constant-valued coefficient (GL_FLAT shading). + * The value value comes from vertex[slot][i]. + * The result will be put into setup->coef[slot].a0[i]. + * \param slot which attribute slot + * \param i which component of the slot (0..3) + */ +static void const_coeff( struct setup_context *setup, + unsigned attrib, + uint vertSlot) +{ + unsigned i; + for (i = 0; i < NUM_CHANNELS; ++i) { + setup->coef.dadx[1 + attrib][i] = 0; + setup->coef.dady[1 + attrib][i] = 0; + + /* need provoking vertex info! + */ + setup->coef.a0[1 + attrib][i] = setup->vprovoke[vertSlot][i]; + } +} + + +/** + * Compute a0, dadx and dady for a linearly interpolated coefficient, + * for a triangle. + */ +static void tri_linear_coeff( struct setup_context *setup, + unsigned attrib, + uint vertSlot) +{ + unsigned i; + for (i = 0; i < NUM_CHANNELS; ++i) { + float botda = setup->vmid[vertSlot][i] - setup->vmin[vertSlot][i]; + float majda = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i]; + float a = setup->ebot.dy * majda - botda * setup->emaj.dy; + float b = setup->emaj.dx * botda - majda * setup->ebot.dx; + float dadx = a * setup->oneoverarea; + float dady = b * setup->oneoverarea; + + assert(i <= 3); + + setup->coef.dadx[1 + attrib][i] = dadx; + setup->coef.dady[1 + attrib][i] = dady; + + /* calculate a0 as the value which would be sampled for the + * fragment at (0,0), taking into account that we want to sample at + * pixel centers, in other words (0.5, 0.5). + * + * this is neat but unfortunately not a good way to do things for + * triangles with very large values of dadx or dady as it will + * result in the subtraction and re-addition from a0 of a very + * large number, which means we'll end up loosing a lot of the + * fractional bits and precision from a0. the way to fix this is + * to define a0 as the sample at a pixel center somewhere near vmin + * instead - i'll switch to this later. + */ + setup->coef.a0[1 + attrib][i] = (setup->vmin[vertSlot][i] - + (dadx * (setup->vmin[0][0] - 0.5f) + + dady * (setup->vmin[0][1] - 0.5f))); + + /* + debug_printf("attr[%d].%c: %f dx:%f dy:%f\n", + slot, "xyzw"[i], + setup->coef[slot].a0[i], + setup->coef[slot].dadx[i], + setup->coef[slot].dady[i]); + */ + } +} + + +/** + * Compute a0, dadx and dady for a perspective-corrected interpolant, + * for a triangle. + * We basically multiply the vertex value by 1/w before computing + * the plane coefficients (a0, dadx, dady). + * Later, when we compute the value at a particular fragment position we'll + * divide the interpolated value by the interpolated W at that fragment. + */ +static void tri_persp_coeff( struct setup_context *setup, + unsigned attrib, + uint vertSlot) +{ + unsigned i; + for (i = 0; i < NUM_CHANNELS; ++i) { + /* premultiply by 1/w (v[0][3] is always W): + */ + float mina = setup->vmin[vertSlot][i] * setup->vmin[0][3]; + float mida = setup->vmid[vertSlot][i] * setup->vmid[0][3]; + float maxa = setup->vmax[vertSlot][i] * setup->vmax[0][3]; + float botda = mida - mina; + float majda = maxa - mina; + float a = setup->ebot.dy * majda - botda * setup->emaj.dy; + float b = setup->emaj.dx * botda - majda * setup->ebot.dx; + float dadx = a * setup->oneoverarea; + float dady = b * setup->oneoverarea; + + /* + debug_printf("tri persp %d,%d: %f %f %f\n", vertSlot, i, + setup->vmin[vertSlot][i], + setup->vmid[vertSlot][i], + setup->vmax[vertSlot][i] + ); + */ + assert(i <= 3); + + setup->coef.dadx[1 + attrib][i] = dadx; + setup->coef.dady[1 + attrib][i] = dady; + setup->coef.a0[1 + attrib][i] = (mina - + (dadx * (setup->vmin[0][0] - 0.5f) + + dady * (setup->vmin[0][1] - 0.5f))); + } +} + + +/** + * Special coefficient setup for gl_FragCoord. + * X and Y are trivial, though Y has to be inverted for OpenGL. + * Z and W are copied from posCoef which should have already been computed. + * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask. + */ +static void +setup_fragcoord_coeff(struct setup_context *setup, uint slot) +{ + /*X*/ + setup->coef.a0[1 + slot][0] = 0; + setup->coef.dadx[1 + slot][0] = 1.0; + setup->coef.dady[1 + slot][0] = 0.0; + /*Y*/ + setup->coef.a0[1 + slot][1] = 0.0; + setup->coef.dadx[1 + slot][1] = 0.0; + setup->coef.dady[1 + slot][1] = 1.0; + /*Z*/ + setup->coef.a0[1 + slot][2] = setup->coef.a0[0][2]; + setup->coef.dadx[1 + slot][2] = setup->coef.dadx[0][2]; + setup->coef.dady[1 + slot][2] = setup->coef.dady[0][2]; + /*W*/ + setup->coef.a0[1 + slot][3] = setup->coef.a0[0][3]; + setup->coef.dadx[1 + slot][3] = setup->coef.dadx[0][3]; + setup->coef.dady[1 + slot][3] = setup->coef.dady[0][3]; +} + + + +/** + * Compute the setup->coef[] array dadx, dady, a0 values. + * Must be called after setup->vmin,vmid,vmax,vprovoke are initialized. + */ +static void setup_tri_coefficients( struct setup_context *setup ) +{ + struct llvmpipe_context *llvmpipe = setup->llvmpipe; + const struct lp_fragment_shader *lpfs = llvmpipe->fs; + const struct vertex_info *vinfo = llvmpipe_get_vertex_info(llvmpipe); + uint fragSlot; + + /* z and w are done by linear interpolation: + */ + tri_pos_coeff(setup, 0, 2); + tri_pos_coeff(setup, 0, 3); + + /* setup interpolation for all the remaining attributes: + */ + for (fragSlot = 0; fragSlot < lpfs->info.num_inputs; fragSlot++) { + const uint vertSlot = vinfo->attrib[fragSlot].src_index; + + switch (vinfo->attrib[fragSlot].interp_mode) { + case INTERP_CONSTANT: + const_coeff(setup, fragSlot, vertSlot); + break; + case INTERP_LINEAR: + tri_linear_coeff(setup, fragSlot, vertSlot); + break; + case INTERP_PERSPECTIVE: + tri_persp_coeff(setup, fragSlot, vertSlot); + break; + case INTERP_POS: + setup_fragcoord_coeff(setup, fragSlot); + break; + default: + assert(0); + } + + if (lpfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) { + setup->coef.a0[1 + fragSlot][0] = 1.0f - setup->facing; + setup->coef.dadx[1 + fragSlot][0] = 0.0; + setup->coef.dady[1 + fragSlot][0] = 0.0; + } + } +} + + + +static void setup_tri_edges( struct setup_context *setup ) +{ + float vmin_x = setup->vmin[0][0] + 0.5f; + float vmid_x = setup->vmid[0][0] + 0.5f; + + float vmin_y = setup->vmin[0][1] - 0.5f; + float vmid_y = setup->vmid[0][1] - 0.5f; + float vmax_y = setup->vmax[0][1] - 0.5f; + + setup->emaj.sy = ceilf(vmin_y); + setup->emaj.lines = (int) ceilf(vmax_y - setup->emaj.sy); + setup->emaj.dxdy = setup->emaj.dx / setup->emaj.dy; + setup->emaj.sx = vmin_x + (setup->emaj.sy - vmin_y) * setup->emaj.dxdy; + + setup->etop.sy = ceilf(vmid_y); + setup->etop.lines = (int) ceilf(vmax_y - setup->etop.sy); + setup->etop.dxdy = setup->etop.dx / setup->etop.dy; + setup->etop.sx = vmid_x + (setup->etop.sy - vmid_y) * setup->etop.dxdy; + + setup->ebot.sy = ceilf(vmin_y); + setup->ebot.lines = (int) ceilf(vmid_y - setup->ebot.sy); + setup->ebot.dxdy = setup->ebot.dx / setup->ebot.dy; + setup->ebot.sx = vmin_x + (setup->ebot.sy - vmin_y) * setup->ebot.dxdy; +} + + +/** + * Render the upper or lower half of a triangle. + * Scissoring/cliprect is applied here too. + */ +static void subtriangle( struct setup_context *setup, + struct edge *eleft, + struct edge *eright, + unsigned lines ) +{ + const struct pipe_scissor_state *cliprect = &setup->llvmpipe->cliprect; + const int minx = (int) cliprect->minx; + const int maxx = (int) cliprect->maxx; + const int miny = (int) cliprect->miny; + const int maxy = (int) cliprect->maxy; + int y, start_y, finish_y; + int sy = (int)eleft->sy; + + assert((int)eleft->sy == (int) eright->sy); + + /* clip top/bottom */ + start_y = sy; + if (start_y < miny) + start_y = miny; + + finish_y = sy + lines; + if (finish_y > maxy) + finish_y = maxy; + + start_y -= sy; + finish_y -= sy; + + /* + debug_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y); + */ + + for (y = start_y; y < finish_y; y++) { + + /* avoid accumulating adds as floats don't have the precision to + * accurately iterate large triangle edges that way. luckily we + * can just multiply these days. + * + * this is all drowned out by the attribute interpolation anyway. + */ + int left = (int)(eleft->sx + y * eleft->dxdy); + int right = (int)(eright->sx + y * eright->dxdy); + + /* clip left/right */ + if (left < minx) + left = minx; + if (right > maxx) + right = maxx; + + if (left < right) { + int _y = sy + y; + if (block(_y) != setup->span.y) { + flush_spans(setup); + setup->span.y = block(_y); + } + + setup->span.left[_y&1] = left; + setup->span.right[_y&1] = right; + } + } + + + /* save the values so that emaj can be restarted: + */ + eleft->sx += lines * eleft->dxdy; + eright->sx += lines * eright->dxdy; + eleft->sy += lines; + eright->sy += lines; +} + + +/** + * Recalculate prim's determinant. This is needed as we don't have + * get this information through the vbuf_render interface & we must + * calculate it here. + */ +static float +calc_det( const float (*v0)[4], + const float (*v1)[4], + const float (*v2)[4] ) +{ + /* edge vectors e = v0 - v2, f = v1 - v2 */ + const float ex = v0[0][0] - v2[0][0]; + const float ey = v0[0][1] - v2[0][1]; + const float fx = v1[0][0] - v2[0][0]; + const float fy = v1[0][1] - v2[0][1]; + + /* det = cross(e,f).z */ + return ex * fy - ey * fx; +} + + +/** + * Do setup for triangle rasterization, then render the triangle. + */ +void llvmpipe_setup_tri( struct setup_context *setup, + const float (*v0)[4], + const float (*v1)[4], + const float (*v2)[4] ) +{ + float det; + +#if DEBUG_VERTS + debug_printf("Setup triangle:\n"); + print_vertex(setup, v0); + print_vertex(setup, v1); + print_vertex(setup, v2); +#endif + + if (setup->llvmpipe->no_rast) + return; + + det = calc_det(v0, v1, v2); + /* + debug_printf("%s\n", __FUNCTION__ ); + */ + +#if DEBUG_FRAGS + setup->numFragsEmitted = 0; + setup->numFragsWritten = 0; +#endif + + if (cull_tri( setup, det )) + return; + + if (!setup_sort_vertices( setup, det, v0, v1, v2 )) + return; + setup_tri_coefficients( setup ); + setup_tri_edges( setup ); + + assert(setup->llvmpipe->reduced_prim == PIPE_PRIM_TRIANGLES); + + setup->span.y = 0; + setup->span.right[0] = 0; + setup->span.right[1] = 0; + /* setup->span.z_mode = tri_z_mode( setup->ctx ); */ + + /* init_constant_attribs( setup ); */ + + if (setup->oneoverarea < 0.0) { + /* emaj on left: + */ + subtriangle( setup, &setup->emaj, &setup->ebot, setup->ebot.lines ); + subtriangle( setup, &setup->emaj, &setup->etop, setup->etop.lines ); + } + else { + /* emaj on right: + */ + subtriangle( setup, &setup->ebot, &setup->emaj, setup->ebot.lines ); + subtriangle( setup, &setup->etop, &setup->emaj, setup->etop.lines ); + } + + flush_spans( setup ); + +#if DEBUG_FRAGS + printf("Tri: %u frags emitted, %u written\n", + setup->numFragsEmitted, + setup->numFragsWritten); +#endif +} + + + +/** + * Compute a0, dadx and dady for a linearly interpolated coefficient, + * for a line. + */ +static void +linear_pos_coeff(struct setup_context *setup, + uint vertSlot, uint i) +{ + const float da = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i]; + const float dadx = da * setup->emaj.dx * setup->oneoverarea; + const float dady = da * setup->emaj.dy * setup->oneoverarea; + setup->coef.dadx[0][i] = dadx; + setup->coef.dady[0][i] = dady; + setup->coef.a0[0][i] = (setup->vmin[vertSlot][i] - + (dadx * (setup->vmin[0][0] - 0.5f) + + dady * (setup->vmin[0][1] - 0.5f))); +} + + +/** + * Compute a0, dadx and dady for a linearly interpolated coefficient, + * for a line. + */ +static void +line_linear_coeff(struct setup_context *setup, + unsigned attrib, + uint vertSlot) +{ + unsigned i; + for (i = 0; i < NUM_CHANNELS; ++i) { + const float da = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i]; + const float dadx = da * setup->emaj.dx * setup->oneoverarea; + const float dady = da * setup->emaj.dy * setup->oneoverarea; + setup->coef.dadx[1 + attrib][i] = dadx; + setup->coef.dady[1 + attrib][i] = dady; + setup->coef.a0[1 + attrib][i] = (setup->vmin[vertSlot][i] - + (dadx * (setup->vmin[0][0] - 0.5f) + + dady * (setup->vmin[0][1] - 0.5f))); + } +} + + +/** + * Compute a0, dadx and dady for a perspective-corrected interpolant, + * for a line. + */ +static void +line_persp_coeff(struct setup_context *setup, + unsigned attrib, + uint vertSlot) +{ + unsigned i; + for (i = 0; i < NUM_CHANNELS; ++i) { + /* XXX double-check/verify this arithmetic */ + const float a0 = setup->vmin[vertSlot][i] * setup->vmin[0][3]; + const float a1 = setup->vmax[vertSlot][i] * setup->vmax[0][3]; + const float da = a1 - a0; + const float dadx = da * setup->emaj.dx * setup->oneoverarea; + const float dady = da * setup->emaj.dy * setup->oneoverarea; + setup->coef.dadx[1 + attrib][i] = dadx; + setup->coef.dady[1 + attrib][i] = dady; + setup->coef.a0[1 + attrib][i] = (setup->vmin[vertSlot][i] - + (dadx * (setup->vmin[0][0] - 0.5f) + + dady * (setup->vmin[0][1] - 0.5f))); + } +} + + +/** + * Compute the setup->coef[] array dadx, dady, a0 values. + * Must be called after setup->vmin,vmax are initialized. + */ +static INLINE boolean +setup_line_coefficients(struct setup_context *setup, + const float (*v0)[4], + const float (*v1)[4]) +{ + struct llvmpipe_context *llvmpipe = setup->llvmpipe; + const struct lp_fragment_shader *lpfs = llvmpipe->fs; + const struct vertex_info *vinfo = llvmpipe_get_vertex_info(llvmpipe); + uint fragSlot; + float area; + + /* use setup->vmin, vmax to point to vertices */ + if (llvmpipe->rasterizer->flatshade_first) + setup->vprovoke = v0; + else + setup->vprovoke = v1; + setup->vmin = v0; + setup->vmax = v1; + + setup->emaj.dx = setup->vmax[0][0] - setup->vmin[0][0]; + setup->emaj.dy = setup->vmax[0][1] - setup->vmin[0][1]; + + /* NOTE: this is not really area but something proportional to it */ + area = setup->emaj.dx * setup->emaj.dx + setup->emaj.dy * setup->emaj.dy; + if (area == 0.0f || util_is_inf_or_nan(area)) + return FALSE; + setup->oneoverarea = 1.0f / area; + + /* z and w are done by linear interpolation: + */ + linear_pos_coeff(setup, 0, 2); + linear_pos_coeff(setup, 0, 3); + + /* setup interpolation for all the remaining attributes: + */ + for (fragSlot = 0; fragSlot < lpfs->info.num_inputs; fragSlot++) { + const uint vertSlot = vinfo->attrib[fragSlot].src_index; + + switch (vinfo->attrib[fragSlot].interp_mode) { + case INTERP_CONSTANT: + const_coeff(setup, fragSlot, vertSlot); + break; + case INTERP_LINEAR: + line_linear_coeff(setup, fragSlot, vertSlot); + break; + case INTERP_PERSPECTIVE: + line_persp_coeff(setup, fragSlot, vertSlot); + break; + case INTERP_POS: + setup_fragcoord_coeff(setup, fragSlot); + break; + default: + assert(0); + } + + if (lpfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) { + setup->coef.a0[1 + fragSlot][0] = 1.0f - setup->facing; + setup->coef.dadx[1 + fragSlot][0] = 0.0; + setup->coef.dady[1 + fragSlot][0] = 0.0; + } + } + return TRUE; +} + + +/** + * Plot a pixel in a line segment. + */ +static INLINE void +plot(struct setup_context *setup, int x, int y) +{ + const int iy = y & 1; + const int ix = x & 1; + const int quadX = x - ix; + const int quadY = y - iy; + const int mask = (1 << ix) << (2 * iy); + + if (quadX != setup->quad[0].input.x0 || + quadY != setup->quad[0].input.y0) + { + /* flush prev quad, start new quad */ + + if (setup->quad[0].input.x0 != -1) + clip_emit_quad( setup, &setup->quad[0] ); + + setup->quad[0].input.x0 = quadX; + setup->quad[0].input.y0 = quadY; + setup->quad[0].inout.mask = 0x0; + } + + setup->quad[0].inout.mask |= mask; +} + + +/** + * Do setup for line rasterization, then render the line. + * Single-pixel width, no stipple, etc. We rely on the 'draw' module + * to handle stippling and wide lines. + */ +void +llvmpipe_setup_line(struct setup_context *setup, + const float (*v0)[4], + const float (*v1)[4]) +{ + int x0 = (int) v0[0][0]; + int x1 = (int) v1[0][0]; + int y0 = (int) v0[0][1]; + int y1 = (int) v1[0][1]; + int dx = x1 - x0; + int dy = y1 - y0; + int xstep, ystep; + +#if DEBUG_VERTS + debug_printf("Setup line:\n"); + print_vertex(setup, v0); + print_vertex(setup, v1); +#endif + + if (setup->llvmpipe->no_rast) + return; + + if (dx == 0 && dy == 0) + return; + + if (!setup_line_coefficients(setup, v0, v1)) + return; + + assert(v0[0][0] < 1.0e9); + assert(v0[0][1] < 1.0e9); + assert(v1[0][0] < 1.0e9); + assert(v1[0][1] < 1.0e9); + + if (dx < 0) { + dx = -dx; /* make positive */ + xstep = -1; + } + else { + xstep = 1; + } + + if (dy < 0) { + dy = -dy; /* make positive */ + ystep = -1; + } + else { + ystep = 1; + } + + assert(dx >= 0); + assert(dy >= 0); + assert(setup->llvmpipe->reduced_prim == PIPE_PRIM_LINES); + + setup->quad[0].input.x0 = setup->quad[0].input.y0 = -1; + setup->quad[0].inout.mask = 0x0; + + /* XXX temporary: set coverage to 1.0 so the line appears + * if AA mode happens to be enabled. + */ + setup->quad[0].input.coverage[0] = + setup->quad[0].input.coverage[1] = + setup->quad[0].input.coverage[2] = + setup->quad[0].input.coverage[3] = 1.0; + + if (dx > dy) { + /*** X-major line ***/ + int i; + const int errorInc = dy + dy; + int error = errorInc - dx; + const int errorDec = error - dx; + + for (i = 0; i < dx; i++) { + plot(setup, x0, y0); + + x0 += xstep; + if (error < 0) { + error += errorInc; + } + else { + error += errorDec; + y0 += ystep; + } + } + } + else { + /*** Y-major line ***/ + int i; + const int errorInc = dx + dx; + int error = errorInc - dy; + const int errorDec = error - dy; + + for (i = 0; i < dy; i++) { + plot(setup, x0, y0); + + y0 += ystep; + if (error < 0) { + error += errorInc; + } + else { + error += errorDec; + x0 += xstep; + } + } + } + + /* draw final quad */ + if (setup->quad[0].inout.mask) { + clip_emit_quad( setup, &setup->quad[0] ); + } +} + + +static void +point_persp_coeff(struct setup_context *setup, + const float (*vert)[4], + unsigned attrib, + uint vertSlot) +{ + unsigned i; + for(i = 0; i < NUM_CHANNELS; ++i) { + setup->coef.dadx[1 + attrib][i] = 0.0F; + setup->coef.dady[1 + attrib][i] = 0.0F; + setup->coef.a0[1 + attrib][i] = vert[vertSlot][i] * vert[0][3]; + } +} + + +/** + * Do setup for point rasterization, then render the point. + * Round or square points... + * XXX could optimize a lot for 1-pixel points. + */ +void +llvmpipe_setup_point( struct setup_context *setup, + const float (*v0)[4] ) +{ + struct llvmpipe_context *llvmpipe = setup->llvmpipe; + const struct lp_fragment_shader *lpfs = llvmpipe->fs; + const int sizeAttr = setup->llvmpipe->psize_slot; + const float size + = sizeAttr > 0 ? v0[sizeAttr][0] + : setup->llvmpipe->rasterizer->point_size; + const float halfSize = 0.5F * size; + const boolean round = (boolean) setup->llvmpipe->rasterizer->point_smooth; + const float x = v0[0][0]; /* Note: data[0] is always position */ + const float y = v0[0][1]; + const struct vertex_info *vinfo = llvmpipe_get_vertex_info(llvmpipe); + uint fragSlot; + +#if DEBUG_VERTS + debug_printf("Setup point:\n"); + print_vertex(setup, v0); +#endif + + if (llvmpipe->no_rast) + return; + + assert(setup->llvmpipe->reduced_prim == PIPE_PRIM_POINTS); + + /* For points, all interpolants are constant-valued. + * However, for point sprites, we'll need to setup texcoords appropriately. + * XXX: which coefficients are the texcoords??? + * We may do point sprites as textured quads... + * + * KW: We don't know which coefficients are texcoords - ultimately + * the choice of what interpolation mode to use for each attribute + * should be determined by the fragment program, using + * per-attribute declaration statements that include interpolation + * mode as a parameter. So either the fragment program will have + * to be adjusted for pointsprite vs normal point behaviour, or + * otherwise a special interpolation mode will have to be defined + * which matches the required behaviour for point sprites. But - + * the latter is not a feature of normal hardware, and as such + * probably should be ruled out on that basis. + */ + setup->vprovoke = v0; + + /* setup Z, W */ + const_pos_coeff(setup, 0, 2); + const_pos_coeff(setup, 0, 3); + + for (fragSlot = 0; fragSlot < lpfs->info.num_inputs; fragSlot++) { + const uint vertSlot = vinfo->attrib[fragSlot].src_index; + + switch (vinfo->attrib[fragSlot].interp_mode) { + case INTERP_CONSTANT: + /* fall-through */ + case INTERP_LINEAR: + const_coeff(setup, fragSlot, vertSlot); + break; + case INTERP_PERSPECTIVE: + point_persp_coeff(setup, setup->vprovoke, fragSlot, vertSlot); + break; + case INTERP_POS: + setup_fragcoord_coeff(setup, fragSlot); + break; + default: + assert(0); + } + + if (lpfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) { + setup->coef.a0[1 + fragSlot][0] = 1.0f - setup->facing; + setup->coef.dadx[1 + fragSlot][0] = 0.0; + setup->coef.dady[1 + fragSlot][0] = 0.0; + } + } + + + if (halfSize <= 0.5 && !round) { + /* special case for 1-pixel points */ + const int ix = ((int) x) & 1; + const int iy = ((int) y) & 1; + setup->quad[0].input.x0 = (int) x - ix; + setup->quad[0].input.y0 = (int) y - iy; + setup->quad[0].inout.mask = (1 << ix) << (2 * iy); + clip_emit_quad( setup, &setup->quad[0] ); + } + else { + if (round) { + /* rounded points */ + const int ixmin = block((int) (x - halfSize)); + const int ixmax = block((int) (x + halfSize)); + const int iymin = block((int) (y - halfSize)); + const int iymax = block((int) (y + halfSize)); + const float rmin = halfSize - 0.7071F; /* 0.7071 = sqrt(2)/2 */ + const float rmax = halfSize + 0.7071F; + const float rmin2 = MAX2(0.0F, rmin * rmin); + const float rmax2 = rmax * rmax; + const float cscale = 1.0F / (rmax2 - rmin2); + int ix, iy; + + for (iy = iymin; iy <= iymax; iy += 2) { + for (ix = ixmin; ix <= ixmax; ix += 2) { + float dx, dy, dist2, cover; + + setup->quad[0].inout.mask = 0x0; + + dx = (ix + 0.5f) - x; + dy = (iy + 0.5f) - y; + dist2 = dx * dx + dy * dy; + if (dist2 <= rmax2) { + cover = 1.0F - (dist2 - rmin2) * cscale; + setup->quad[0].input.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f); + setup->quad[0].inout.mask |= MASK_TOP_LEFT; + } + + dx = (ix + 1.5f) - x; + dy = (iy + 0.5f) - y; + dist2 = dx * dx + dy * dy; + if (dist2 <= rmax2) { + cover = 1.0F - (dist2 - rmin2) * cscale; + setup->quad[0].input.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f); + setup->quad[0].inout.mask |= MASK_TOP_RIGHT; + } + + dx = (ix + 0.5f) - x; + dy = (iy + 1.5f) - y; + dist2 = dx * dx + dy * dy; + if (dist2 <= rmax2) { + cover = 1.0F - (dist2 - rmin2) * cscale; + setup->quad[0].input.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f); + setup->quad[0].inout.mask |= MASK_BOTTOM_LEFT; + } + + dx = (ix + 1.5f) - x; + dy = (iy + 1.5f) - y; + dist2 = dx * dx + dy * dy; + if (dist2 <= rmax2) { + cover = 1.0F - (dist2 - rmin2) * cscale; + setup->quad[0].input.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f); + setup->quad[0].inout.mask |= MASK_BOTTOM_RIGHT; + } + + if (setup->quad[0].inout.mask) { + setup->quad[0].input.x0 = ix; + setup->quad[0].input.y0 = iy; + clip_emit_quad( setup, &setup->quad[0] ); + } + } + } + } + else { + /* square points */ + const int xmin = (int) (x + 0.75 - halfSize); + const int ymin = (int) (y + 0.25 - halfSize); + const int xmax = xmin + (int) size; + const int ymax = ymin + (int) size; + /* XXX could apply scissor to xmin,ymin,xmax,ymax now */ + const int ixmin = block(xmin); + const int ixmax = block(xmax - 1); + const int iymin = block(ymin); + const int iymax = block(ymax - 1); + int ix, iy; + + /* + debug_printf("(%f, %f) -> X:%d..%d Y:%d..%d\n", x, y, xmin, xmax,ymin,ymax); + */ + for (iy = iymin; iy <= iymax; iy += 2) { + uint rowMask = 0xf; + if (iy < ymin) { + /* above the top edge */ + rowMask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT); + } + if (iy + 1 >= ymax) { + /* below the bottom edge */ + rowMask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT); + } + + for (ix = ixmin; ix <= ixmax; ix += 2) { + uint mask = rowMask; + + if (ix < xmin) { + /* fragment is past left edge of point, turn off left bits */ + mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT); + } + if (ix + 1 >= xmax) { + /* past the right edge */ + mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT); + } + + setup->quad[0].inout.mask = mask; + setup->quad[0].input.x0 = ix; + setup->quad[0].input.y0 = iy; + clip_emit_quad( setup, &setup->quad[0] ); + } + } + } + } +} + +void llvmpipe_setup_prepare( struct setup_context *setup ) +{ + struct llvmpipe_context *lp = setup->llvmpipe; + + if (lp->dirty) { + llvmpipe_update_derived(lp); + } + + if (lp->reduced_api_prim == PIPE_PRIM_TRIANGLES && + lp->rasterizer->fill_cw == PIPE_POLYGON_MODE_FILL && + lp->rasterizer->fill_ccw == PIPE_POLYGON_MODE_FILL) { + /* we'll do culling */ + setup->winding = lp->rasterizer->cull_mode; + } + else { + /* 'draw' will do culling */ + setup->winding = PIPE_WINDING_NONE; + } +} + + + +void llvmpipe_setup_destroy_context( struct setup_context *setup ) +{ + align_free( setup ); +} + + +/** + * Create a new primitive setup/render stage. + */ +struct setup_context *llvmpipe_setup_create_context( struct llvmpipe_context *llvmpipe ) +{ + struct setup_context *setup; + unsigned i; + + setup = align_malloc(sizeof(struct setup_context), 16); + if (!setup) + return NULL; + + memset(setup, 0, sizeof *setup); + setup->llvmpipe = llvmpipe; + + for (i = 0; i < MAX_QUADS; i++) { + setup->quad[i].coef = &setup->coef; + } + + setup->span.left[0] = 1000000; /* greater than right[0] */ + setup->span.left[1] = 1000000; /* greater than right[1] */ + + return setup; +} + diff --git a/src/gallium/drivers/llvmpipe/lp_setup.h b/src/gallium/drivers/llvmpipe/lp_setup.h new file mode 100644 index 00000000000..89c43da0460 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_setup.h @@ -0,0 +1,53 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ +#ifndef LP_SETUP_H +#define LP_SETUP_H + +struct setup_context; +struct llvmpipe_context; + +void +llvmpipe_setup_tri( struct setup_context *setup, + const float (*v0)[4], + const float (*v1)[4], + const float (*v2)[4] ); + +void +llvmpipe_setup_line(struct setup_context *setup, + const float (*v0)[4], + const float (*v1)[4]); + +void +llvmpipe_setup_point( struct setup_context *setup, + const float (*v0)[4] ); + + +struct setup_context *llvmpipe_setup_create_context( struct llvmpipe_context *llvmpipe ); +void llvmpipe_setup_prepare( struct setup_context *setup ); +void llvmpipe_setup_destroy_context( struct setup_context *setup ); + +#endif diff --git a/src/gallium/drivers/llvmpipe/lp_state.h b/src/gallium/drivers/llvmpipe/lp_state.h new file mode 100644 index 00000000000..fb10329887d --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_state.h @@ -0,0 +1,225 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef LP_STATE_H +#define LP_STATE_H + +#include <llvm-c/Core.h> + +#include "pipe/p_state.h" +#include "tgsi/tgsi_scan.h" +#include "lp_jit.h" + + +#define LP_NEW_VIEWPORT 0x1 +#define LP_NEW_RASTERIZER 0x2 +#define LP_NEW_FS 0x4 +#define LP_NEW_BLEND 0x8 +#define LP_NEW_CLIP 0x10 +#define LP_NEW_SCISSOR 0x20 +#define LP_NEW_STIPPLE 0x40 +#define LP_NEW_FRAMEBUFFER 0x80 +#define LP_NEW_DEPTH_STENCIL_ALPHA 0x100 +#define LP_NEW_CONSTANTS 0x200 +#define LP_NEW_SAMPLER 0x400 +#define LP_NEW_TEXTURE 0x800 +#define LP_NEW_VERTEX 0x1000 +#define LP_NEW_VS 0x2000 +#define LP_NEW_QUERY 0x4000 + + +struct tgsi_sampler; +struct vertex_info; + + +struct lp_fragment_shader; + + +struct lp_fragment_shader_variant_key +{ + struct pipe_depth_state depth; + struct pipe_alpha_state alpha; + struct pipe_blend_state blend; +}; + + +struct lp_fragment_shader_variant +{ + struct lp_fragment_shader *shader; + + struct lp_fragment_shader_variant_key key; + + LLVMValueRef function; + + lp_jit_frag_func jit_function; + + struct lp_fragment_shader_variant *next; +}; + + +/** + * Subclass of pipe_shader_state (though it doesn't really need to be). + * + * This is starting to look an awful lot like a quad pipeline stage... + */ +struct lp_fragment_shader +{ + struct pipe_shader_state base; + + struct tgsi_shader_info info; + + struct lp_fragment_shader_variant *variants; + + struct lp_fragment_shader_variant *current; +}; + + +/** Subclass of pipe_shader_state */ +struct lp_vertex_shader { + struct pipe_shader_state shader; + struct draw_vertex_shader *draw_data; +}; + + + +void * +llvmpipe_create_blend_state(struct pipe_context *, + const struct pipe_blend_state *); +void llvmpipe_bind_blend_state(struct pipe_context *, + void *); +void llvmpipe_delete_blend_state(struct pipe_context *, + void *); + +void * +llvmpipe_create_sampler_state(struct pipe_context *, + const struct pipe_sampler_state *); +void llvmpipe_bind_sampler_states(struct pipe_context *, unsigned, void **); +void llvmpipe_delete_sampler_state(struct pipe_context *, void *); + +void * +llvmpipe_create_depth_stencil_state(struct pipe_context *, + const struct pipe_depth_stencil_alpha_state *); +void llvmpipe_bind_depth_stencil_state(struct pipe_context *, void *); +void llvmpipe_delete_depth_stencil_state(struct pipe_context *, void *); + +void * +llvmpipe_create_rasterizer_state(struct pipe_context *, + const struct pipe_rasterizer_state *); +void llvmpipe_bind_rasterizer_state(struct pipe_context *, void *); +void llvmpipe_delete_rasterizer_state(struct pipe_context *, void *); + +void llvmpipe_set_framebuffer_state( struct pipe_context *, + const struct pipe_framebuffer_state * ); + +void llvmpipe_set_blend_color( struct pipe_context *pipe, + const struct pipe_blend_color *blend_color ); + +void llvmpipe_set_clip_state( struct pipe_context *, + const struct pipe_clip_state * ); + +void llvmpipe_set_constant_buffer(struct pipe_context *, + uint shader, uint index, + const struct pipe_constant_buffer *buf); + +void *llvmpipe_create_fs_state(struct pipe_context *, + const struct pipe_shader_state *); +void llvmpipe_bind_fs_state(struct pipe_context *, void *); +void llvmpipe_delete_fs_state(struct pipe_context *, void *); +void *llvmpipe_create_vs_state(struct pipe_context *, + const struct pipe_shader_state *); +void llvmpipe_bind_vs_state(struct pipe_context *, void *); +void llvmpipe_delete_vs_state(struct pipe_context *, void *); + +void llvmpipe_set_polygon_stipple( struct pipe_context *, + const struct pipe_poly_stipple * ); + +void llvmpipe_set_scissor_state( struct pipe_context *, + const struct pipe_scissor_state * ); + +void llvmpipe_set_sampler_textures( struct pipe_context *, + unsigned num, + struct pipe_texture ** ); + +void llvmpipe_set_viewport_state( struct pipe_context *, + const struct pipe_viewport_state * ); + +void llvmpipe_set_vertex_elements(struct pipe_context *, + unsigned count, + const struct pipe_vertex_element *); + +void llvmpipe_set_vertex_buffers(struct pipe_context *, + unsigned count, + const struct pipe_vertex_buffer *); + +void llvmpipe_update_fs(struct llvmpipe_context *lp); + +void llvmpipe_update_derived( struct llvmpipe_context *llvmpipe ); + + +boolean llvmpipe_draw_arrays(struct pipe_context *pipe, unsigned mode, + unsigned start, unsigned count); + +boolean llvmpipe_draw_elements(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned mode, unsigned start, unsigned count); +boolean +llvmpipe_draw_range_elements(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned min_index, + unsigned max_index, + unsigned mode, unsigned start, unsigned count); + +void +llvmpipe_set_edgeflags(struct pipe_context *pipe, const unsigned *edgeflags); + + +void +llvmpipe_map_transfers(struct llvmpipe_context *lp); + +void +llvmpipe_unmap_transfers(struct llvmpipe_context *lp); + +void +llvmpipe_map_texture_surfaces(struct llvmpipe_context *lp); + +void +llvmpipe_unmap_texture_surfaces(struct llvmpipe_context *lp); + + +struct vertex_info * +llvmpipe_get_vertex_info(struct llvmpipe_context *llvmpipe); + +struct vertex_info * +llvmpipe_get_vbuf_vertex_info(struct llvmpipe_context *llvmpipe); + + +#endif diff --git a/src/gallium/drivers/llvmpipe/lp_state_blend.c b/src/gallium/drivers/llvmpipe/lp_state_blend.c new file mode 100644 index 00000000000..3f03bd00571 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_state_blend.c @@ -0,0 +1,114 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @author Jose Fonseca <jfonseca@vmware.com> + * @author Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "util/u_memory.h" +#include "util/u_math.h" +#include "util/u_debug_dump.h" +#include "lp_screen.h" +#include "lp_context.h" +#include "lp_state.h" + + +void * +llvmpipe_create_blend_state(struct pipe_context *pipe, + const struct pipe_blend_state *blend) +{ + return mem_dup(blend, sizeof(*blend)); +} + +void llvmpipe_bind_blend_state( struct pipe_context *pipe, + void *blend ) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + llvmpipe->blend = blend; + + llvmpipe->dirty |= LP_NEW_BLEND; +} + +void llvmpipe_delete_blend_state(struct pipe_context *pipe, + void *blend) +{ + FREE( blend ); +} + + +void llvmpipe_set_blend_color( struct pipe_context *pipe, + const struct pipe_blend_color *blend_color ) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + unsigned i, j; + + memcpy(&llvmpipe->blend_color, blend_color, sizeof *blend_color); + + if(!llvmpipe->jit_context.blend_color) + llvmpipe->jit_context.blend_color = align_malloc(4 * 16, 16); + for (i = 0; i < 4; ++i) { + uint8_t c = float_to_ubyte(blend_color->color[i]); + for (j = 0; j < 16; ++j) + llvmpipe->jit_context.blend_color[i*4 + j] = c; + } +} + + +/** XXX move someday? Or consolidate all these simple state setters + * into one file. + */ + + +void * +llvmpipe_create_depth_stencil_state(struct pipe_context *pipe, + const struct pipe_depth_stencil_alpha_state *depth_stencil) +{ + return mem_dup(depth_stencil, sizeof(*depth_stencil)); +} + +void +llvmpipe_bind_depth_stencil_state(struct pipe_context *pipe, + void *depth_stencil) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + llvmpipe->depth_stencil = (const struct pipe_depth_stencil_alpha_state *)depth_stencil; + + if(llvmpipe->depth_stencil) + llvmpipe->jit_context.alpha_ref_value = llvmpipe->depth_stencil->alpha.ref_value; + + llvmpipe->dirty |= LP_NEW_DEPTH_STENCIL_ALPHA; +} + +void +llvmpipe_delete_depth_stencil_state(struct pipe_context *pipe, void *depth) +{ + FREE( depth ); +} diff --git a/src/gallium/drivers/llvmpipe/lp_state_clip.c b/src/gallium/drivers/llvmpipe/lp_state_clip.c new file mode 100644 index 00000000000..df68f27acc9 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_state_clip.c @@ -0,0 +1,79 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ +#include "lp_context.h" +#include "lp_state.h" +#include "draw/draw_context.h" + + +void llvmpipe_set_clip_state( struct pipe_context *pipe, + const struct pipe_clip_state *clip ) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + /* pass the clip state to the draw module */ + draw_set_clip_state(llvmpipe->draw, clip); +} + + +void llvmpipe_set_viewport_state( struct pipe_context *pipe, + const struct pipe_viewport_state *viewport ) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + /* pass the viewport info to the draw module */ + draw_set_viewport_state(llvmpipe->draw, viewport); + + llvmpipe->viewport = *viewport; /* struct copy */ + llvmpipe->dirty |= LP_NEW_VIEWPORT; +} + + +void llvmpipe_set_scissor_state( struct pipe_context *pipe, + const struct pipe_scissor_state *scissor ) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + draw_flush(llvmpipe->draw); + + llvmpipe->scissor = *scissor; /* struct copy */ + llvmpipe->dirty |= LP_NEW_SCISSOR; +} + + +void llvmpipe_set_polygon_stipple( struct pipe_context *pipe, + const struct pipe_poly_stipple *stipple ) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + draw_flush(llvmpipe->draw); + + llvmpipe->poly_stipple = *stipple; /* struct copy */ + llvmpipe->dirty |= LP_NEW_STIPPLE; +} diff --git a/src/gallium/drivers/llvmpipe/lp_state_derived.c b/src/gallium/drivers/llvmpipe/lp_state_derived.c new file mode 100644 index 00000000000..6fbb057937e --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_state_derived.c @@ -0,0 +1,258 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "util/u_math.h" +#include "util/u_memory.h" +#include "pipe/p_shader_tokens.h" +#include "draw/draw_context.h" +#include "draw/draw_vertex.h" +#include "draw/draw_private.h" +#include "lp_context.h" +#include "lp_screen.h" +#include "lp_tex_cache.h" +#include "lp_state.h" + + +/** + * Mark the current vertex layout as "invalid". + * We'll validate the vertex layout later, when we start to actually + * render a point or line or tri. + */ +static void +invalidate_vertex_layout(struct llvmpipe_context *llvmpipe) +{ + llvmpipe->vertex_info.num_attribs = 0; +} + + +/** + * The vertex info describes how to convert the post-transformed vertices + * (simple float[][4]) used by the 'draw' module into vertices for + * rasterization. + * + * This function validates the vertex layout and returns a pointer to a + * vertex_info object. + */ +struct vertex_info * +llvmpipe_get_vertex_info(struct llvmpipe_context *llvmpipe) +{ + struct vertex_info *vinfo = &llvmpipe->vertex_info; + + if (vinfo->num_attribs == 0) { + /* compute vertex layout now */ + const struct lp_fragment_shader *lpfs = llvmpipe->fs; + const enum interp_mode colorInterp + = llvmpipe->rasterizer->flatshade ? INTERP_CONSTANT : INTERP_LINEAR; + uint i; + + if (llvmpipe->vbuf) { + /* if using the post-transform vertex buffer, tell draw_vbuf to + * simply emit the whole post-xform vertex as-is: + */ + struct vertex_info *vinfo_vbuf = &llvmpipe->vertex_info_vbuf; + const uint num = draw_num_vs_outputs(llvmpipe->draw); + uint i; + + /* No longer any need to try and emit draw vertex_header info. + */ + vinfo_vbuf->num_attribs = 0; + for (i = 0; i < num; i++) { + draw_emit_vertex_attr(vinfo_vbuf, EMIT_4F, INTERP_PERSPECTIVE, i); + } + draw_compute_vertex_size(vinfo_vbuf); + } + + /* + * Loop over fragment shader inputs, searching for the matching output + * from the vertex shader. + */ + vinfo->num_attribs = 0; + for (i = 0; i < lpfs->info.num_inputs; i++) { + int src; + switch (lpfs->info.input_semantic_name[i]) { + case TGSI_SEMANTIC_POSITION: + src = draw_find_vs_output(llvmpipe->draw, + TGSI_SEMANTIC_POSITION, 0); + draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_POS, src); + break; + + case TGSI_SEMANTIC_COLOR: + src = draw_find_vs_output(llvmpipe->draw, TGSI_SEMANTIC_COLOR, + lpfs->info.input_semantic_index[i]); + draw_emit_vertex_attr(vinfo, EMIT_4F, colorInterp, src); + break; + + case TGSI_SEMANTIC_FOG: + src = draw_find_vs_output(llvmpipe->draw, TGSI_SEMANTIC_FOG, 0); + draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src); + break; + + case TGSI_SEMANTIC_GENERIC: + case TGSI_SEMANTIC_FACE: + /* this includes texcoords and varying vars */ + src = draw_find_vs_output(llvmpipe->draw, TGSI_SEMANTIC_GENERIC, + lpfs->info.input_semantic_index[i]); + draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src); + break; + + default: + assert(0); + } + } + + llvmpipe->psize_slot = draw_find_vs_output(llvmpipe->draw, + TGSI_SEMANTIC_PSIZE, 0); + if (llvmpipe->psize_slot > 0) { + draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_CONSTANT, + llvmpipe->psize_slot); + } + + draw_compute_vertex_size(vinfo); + } + + return vinfo; +} + + +/** + * Called from vbuf module. + * + * Note that there's actually two different vertex layouts in llvmpipe. + * + * The normal one is computed in llvmpipe_get_vertex_info() above and is + * used by the point/line/tri "setup" code. + * + * The other one (this one) is only used by the vbuf module (which is + * not normally used by default but used in testing). For the vbuf module, + * we basically want to pass-through the draw module's vertex layout as-is. + * When the llvmpipe vbuf code begins drawing, the normal vertex layout + * will come into play again. + */ +struct vertex_info * +llvmpipe_get_vbuf_vertex_info(struct llvmpipe_context *llvmpipe) +{ + (void) llvmpipe_get_vertex_info(llvmpipe); + return &llvmpipe->vertex_info_vbuf; +} + + +/** + * Recompute cliprect from scissor bounds, scissor enable and surface size. + */ +static void +compute_cliprect(struct llvmpipe_context *lp) +{ + /* LP_NEW_FRAMEBUFFER + */ + uint surfWidth = lp->framebuffer.width; + uint surfHeight = lp->framebuffer.height; + + /* LP_NEW_RASTERIZER + */ + if (lp->rasterizer->scissor) { + + /* LP_NEW_SCISSOR + * + * clip to scissor rect: + */ + lp->cliprect.minx = MAX2(lp->scissor.minx, 0); + lp->cliprect.miny = MAX2(lp->scissor.miny, 0); + lp->cliprect.maxx = MIN2(lp->scissor.maxx, surfWidth); + lp->cliprect.maxy = MIN2(lp->scissor.maxy, surfHeight); + } + else { + /* clip to surface bounds */ + lp->cliprect.minx = 0; + lp->cliprect.miny = 0; + lp->cliprect.maxx = surfWidth; + lp->cliprect.maxy = surfHeight; + } +} + + +static void +update_tgsi_samplers( struct llvmpipe_context *llvmpipe ) +{ + unsigned i; + + /* vertex shader samplers */ + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + llvmpipe->tgsi.vert_samplers[i].sampler = llvmpipe->sampler[i]; + llvmpipe->tgsi.vert_samplers[i].texture = llvmpipe->texture[i]; + llvmpipe->tgsi.frag_samplers[i].base.get_samples = lp_get_samples; + } + + /* fragment shader samplers */ + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + llvmpipe->tgsi.frag_samplers[i].sampler = llvmpipe->sampler[i]; + llvmpipe->tgsi.frag_samplers[i].texture = llvmpipe->texture[i]; + llvmpipe->tgsi.frag_samplers[i].base.get_samples = lp_get_samples; + } + + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + lp_tex_tile_cache_validate_texture( llvmpipe->tex_cache[i] ); + } + + llvmpipe->jit_context.samplers = (struct tgsi_sampler **)llvmpipe->tgsi.frag_samplers_list; +} + +/* Hopefully this will remain quite simple, otherwise need to pull in + * something like the state tracker mechanism. + */ +void llvmpipe_update_derived( struct llvmpipe_context *llvmpipe ) +{ + struct llvmpipe_screen *lp_screen = llvmpipe_screen(llvmpipe->pipe.screen); + + /* Check for updated textures. + */ + if (llvmpipe->tex_timestamp != lp_screen->timestamp) { + llvmpipe->tex_timestamp = lp_screen->timestamp; + llvmpipe->dirty |= LP_NEW_TEXTURE; + } + + if (llvmpipe->dirty & (LP_NEW_SAMPLER | + LP_NEW_TEXTURE)) + update_tgsi_samplers( llvmpipe ); + + if (llvmpipe->dirty & (LP_NEW_RASTERIZER | + LP_NEW_FS | + LP_NEW_VS)) + invalidate_vertex_layout( llvmpipe ); + + if (llvmpipe->dirty & (LP_NEW_SCISSOR | + LP_NEW_RASTERIZER | + LP_NEW_FRAMEBUFFER)) + compute_cliprect(llvmpipe); + + if (llvmpipe->dirty & (LP_NEW_FS | + LP_NEW_BLEND | + LP_NEW_DEPTH_STENCIL_ALPHA)) + llvmpipe_update_fs( llvmpipe ); + + + llvmpipe->dirty = 0; +} diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c new file mode 100644 index 00000000000..94170bd7161 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c @@ -0,0 +1,804 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Code generate the whole fragment pipeline. + * + * The fragment pipeline consists of the following stages: + * - stipple (TBI) + * - early depth test + * - fragment shader + * - alpha test + * - depth/stencil test (stencil TBI) + * - blending + * + * This file has only the glue to assembly the fragment pipeline. The actual + * plumbing of converting Gallium state into LLVM IR is done elsewhere, in the + * lp_bld_*.[ch] files, and in a complete generic and reusable way. Here we + * muster the LLVM JIT execution engine to create a function that follows an + * established binary interface and that can be called from C directly. + * + * A big source of complexity here is that we often want to run different + * stages with different precisions and data types and precisions. For example, + * the fragment shader needs typically to be done in floats, but the + * depth/stencil test and blending is better done in the type that most closely + * matches the depth/stencil and color buffer respectively. + * + * Since the width of a SIMD vector register stays the same regardless of the + * element type, different types imply different number of elements, so we must + * code generate more instances of the stages with larger types to be able to + * feed/consume the stages with smaller types. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#include "pipe/p_defines.h" +#include "util/u_memory.h" +#include "util/u_format.h" +#include "util/u_debug_dump.h" +#include "pipe/internal/p_winsys_screen.h" +#include "pipe/p_shader_tokens.h" +#include "draw/draw_context.h" +#include "tgsi/tgsi_dump.h" +#include "tgsi/tgsi_scan.h" +#include "tgsi/tgsi_parse.h" +#include "lp_bld_type.h" +#include "lp_bld_const.h" +#include "lp_bld_conv.h" +#include "lp_bld_intr.h" +#include "lp_bld_logic.h" +#include "lp_bld_depth.h" +#include "lp_bld_interp.h" +#include "lp_bld_tgsi.h" +#include "lp_bld_alpha.h" +#include "lp_bld_blend.h" +#include "lp_bld_swizzle.h" +#include "lp_bld_flow.h" +#include "lp_bld_debug.h" +#include "lp_screen.h" +#include "lp_context.h" +#include "lp_state.h" +#include "lp_quad.h" + + +static const unsigned char quad_offset_x[4] = {0, 1, 0, 1}; +static const unsigned char quad_offset_y[4] = {0, 0, 1, 1}; + + +/* + * Derive from the quad's upper left scalar coordinates the coordinates for + * all other quad pixels + */ +static void +generate_pos0(LLVMBuilderRef builder, + LLVMValueRef x, + LLVMValueRef y, + LLVMValueRef *x0, + LLVMValueRef *y0) +{ + LLVMTypeRef int_elem_type = LLVMInt32Type(); + LLVMTypeRef int_vec_type = LLVMVectorType(int_elem_type, QUAD_SIZE); + LLVMTypeRef elem_type = LLVMFloatType(); + LLVMTypeRef vec_type = LLVMVectorType(elem_type, QUAD_SIZE); + LLVMValueRef x_offsets[QUAD_SIZE]; + LLVMValueRef y_offsets[QUAD_SIZE]; + unsigned i; + + x = lp_build_broadcast(builder, int_vec_type, x); + y = lp_build_broadcast(builder, int_vec_type, y); + + for(i = 0; i < QUAD_SIZE; ++i) { + x_offsets[i] = LLVMConstInt(int_elem_type, quad_offset_x[i], 0); + y_offsets[i] = LLVMConstInt(int_elem_type, quad_offset_y[i], 0); + } + + x = LLVMBuildAdd(builder, x, LLVMConstVector(x_offsets, QUAD_SIZE), ""); + y = LLVMBuildAdd(builder, y, LLVMConstVector(y_offsets, QUAD_SIZE), ""); + + *x0 = LLVMBuildSIToFP(builder, x, vec_type, ""); + *y0 = LLVMBuildSIToFP(builder, y, vec_type, ""); +} + + +/** + * Generate the depth test. + */ +static void +generate_depth(struct llvmpipe_context *lp, + LLVMBuilderRef builder, + const struct pipe_depth_state *state, + union lp_type src_type, + struct lp_build_mask_context *mask, + LLVMValueRef src, + LLVMValueRef dst_ptr) +{ + const struct util_format_description *format_desc; + union lp_type dst_type; + + if(!lp->framebuffer.zsbuf) + return; + + format_desc = util_format_description(lp->framebuffer.zsbuf->format); + assert(format_desc); + + /* Pick the depth type. */ + dst_type = lp_depth_type(format_desc, src_type.width*src_type.length); + + /* FIXME: Cope with a depth test type with a different bit width. */ + assert(dst_type.width == src_type.width); + assert(dst_type.length == src_type.length); + +#if 1 + src = lp_build_clamped_float_to_unsigned_norm(builder, + src_type, + dst_type.width, + src); +#else + lp_build_conv(builder, src_type, dst_type, &src, 1, &src, 1); +#endif + + lp_build_depth_test(builder, + state, + dst_type, + format_desc, + mask, + src, + dst_ptr); +} + + +struct build_fetch_texel_context +{ + LLVMValueRef context_ptr; + + LLVMValueRef samplers_ptr; + + /** Coords/texels store */ + LLVMValueRef store_ptr; +}; + + +void PIPE_CDECL +lp_fetch_texel_soa( struct tgsi_sampler **samplers, + uint32_t unit, + float *store ) +{ + struct tgsi_sampler *sampler = samplers[unit]; + +#if 0 + uint j; + + debug_printf("%s sampler: %p (%p) store: %p\n", + __FUNCTION__, + sampler, *sampler, + store ); + + debug_printf("lodbias %f\n", store[12]); + + for (j = 0; j < 4; j++) + debug_printf("sample %d texcoord %f %f\n", + j, + store[0+j], + store[4+j]); +#endif + + { + float rgba[NUM_CHANNELS][QUAD_SIZE]; + sampler->get_samples(sampler, + &store[0], + &store[4], + &store[8], + 0.0f, /*store[12], lodbias */ + rgba); + memcpy(store, rgba, sizeof rgba); + } + +#if 0 + for (j = 0; j < 4; j++) + debug_printf("sample %d result %f %f %f %f\n", + j, + store[0+j], + store[4+j], + store[8+j], + store[12+j]); +#endif +} + + +static void +emit_fetch_texel( LLVMBuilderRef builder, + void *context, + unsigned unit, + unsigned num_coords, + const LLVMValueRef *coords, + LLVMValueRef lodbias, + LLVMValueRef *texel) +{ + struct build_fetch_texel_context *bld = context; + LLVMTypeRef vec_type = LLVMTypeOf(coords[0]); + LLVMValueRef args[3]; + unsigned i; + + if(!bld->samplers_ptr) + bld->samplers_ptr = lp_jit_context_samplers(builder, bld->context_ptr); + + if(!bld->store_ptr) + bld->store_ptr = LLVMBuildArrayAlloca(builder, + vec_type, + LLVMConstInt(LLVMInt32Type(), 4, 0), + "texel_store"); + + for (i = 0; i < num_coords; i++) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef coord_ptr = LLVMBuildGEP(builder, bld->store_ptr, &index, 1, ""); + LLVMBuildStore(builder, coords[i], coord_ptr); + } + + args[0] = bld->samplers_ptr; + args[1] = LLVMConstInt(LLVMInt32Type(), unit, 0); + args[2] = bld->store_ptr; + + lp_build_intrinsic(builder, "fetch_texel", LLVMVoidType(), args, 3); + + for (i = 0; i < NUM_CHANNELS; ++i) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef texel_ptr = LLVMBuildGEP(builder, bld->store_ptr, &index, 1, ""); + texel[i] = LLVMBuildLoad(builder, texel_ptr, ""); + } +} + + +/** + * Generate the fragment shader, depth/stencil test, and alpha tests. + */ +static void +generate_fs(struct llvmpipe_context *lp, + struct lp_fragment_shader *shader, + const struct lp_fragment_shader_variant_key *key, + LLVMBuilderRef builder, + union lp_type type, + LLVMValueRef context_ptr, + unsigned i, + const struct lp_build_interp_soa_context *interp, + struct build_fetch_texel_context *sampler, + LLVMValueRef *pmask, + LLVMValueRef *color, + LLVMValueRef depth_ptr) +{ + const struct tgsi_token *tokens = shader->base.tokens; + LLVMTypeRef elem_type; + LLVMTypeRef vec_type; + LLVMTypeRef int_vec_type; + LLVMValueRef consts_ptr; + LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][NUM_CHANNELS]; + LLVMValueRef z = interp->pos[2]; + struct lp_build_mask_context mask; + boolean early_depth_test; + unsigned attrib; + unsigned chan; + + elem_type = lp_build_elem_type(type); + vec_type = lp_build_vec_type(type); + int_vec_type = lp_build_int_vec_type(type); + + consts_ptr = lp_jit_context_constants(builder, context_ptr); + + lp_build_mask_begin(&mask, builder, type, *pmask); + + early_depth_test = + lp->depth_stencil->depth.enabled && + lp->framebuffer.zsbuf && + !lp->depth_stencil->alpha.enabled && + !lp->fs->info.uses_kill && + !lp->fs->info.writes_z; + + if(early_depth_test) + generate_depth(lp, builder, &key->depth, + type, &mask, + z, depth_ptr); + + memset(outputs, 0, sizeof outputs); + + lp_build_tgsi_soa(builder, tokens, type, &mask, + consts_ptr, interp->pos, interp->inputs, + outputs, emit_fetch_texel, sampler); + + for (attrib = 0; attrib < shader->info.num_outputs; ++attrib) { + for(chan = 0; chan < NUM_CHANNELS; ++chan) { + if(outputs[attrib][chan]) { + lp_build_name(outputs[attrib][chan], "output%u.%u.%c", i, attrib, "xyzw"[chan]); + + switch (shader->info.output_semantic_name[attrib]) { + case TGSI_SEMANTIC_COLOR: + { + unsigned cbuf = shader->info.output_semantic_index[attrib]; + + lp_build_name(outputs[attrib][chan], "color%u.%u.%c", i, attrib, "rgba"[chan]); + + /* Alpha test */ + /* XXX: should the alpha reference value be passed separately? */ + if(cbuf == 0 && chan == 3) { + LLVMValueRef alpha = outputs[attrib][chan]; + LLVMValueRef alpha_ref_value; + alpha_ref_value = lp_jit_context_alpha_ref_value(builder, context_ptr); + alpha_ref_value = lp_build_broadcast(builder, vec_type, alpha_ref_value); + lp_build_alpha_test(builder, &key->alpha, type, + &mask, alpha, alpha_ref_value); + } + + if(cbuf == 0) + color[chan] = outputs[attrib][chan]; + + break; + } + + case TGSI_SEMANTIC_POSITION: + if(chan == 2) + z = outputs[attrib][chan]; + break; + } + } + } + } + + if(!early_depth_test) + generate_depth(lp, builder, &key->depth, + type, &mask, + z, depth_ptr); + + lp_build_mask_end(&mask); + + *pmask = mask.value; + +} + + +/** + * Generate color blending and color output. + */ +static void +generate_blend(const struct pipe_blend_state *blend, + LLVMBuilderRef builder, + union lp_type type, + LLVMValueRef context_ptr, + LLVMValueRef mask, + LLVMValueRef *src, + LLVMValueRef dst_ptr) +{ + struct lp_build_context bld; + LLVMTypeRef vec_type; + LLVMTypeRef int_vec_type; + LLVMValueRef const_ptr; + LLVMValueRef con[4]; + LLVMValueRef dst[4]; + LLVMValueRef res[4]; + unsigned chan; + + vec_type = lp_build_vec_type(type); + int_vec_type = lp_build_int_vec_type(type); + + lp_build_context_init(&bld, builder, type); + + const_ptr = lp_jit_context_blend_color(builder, context_ptr); + const_ptr = LLVMBuildBitCast(builder, const_ptr, + LLVMPointerType(vec_type, 0), ""); + + for(chan = 0; chan < 4; ++chan) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), chan, 0); + con[chan] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, const_ptr, &index, 1, ""), ""); + + dst[chan] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, dst_ptr, &index, 1, ""), ""); + + lp_build_name(con[chan], "con.%c", "rgba"[chan]); + lp_build_name(dst[chan], "dst.%c", "rgba"[chan]); + } + + lp_build_blend_soa(builder, blend, type, src, dst, con, res); + + for(chan = 0; chan < 4; ++chan) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), chan, 0); + lp_build_name(res[chan], "res.%c", "rgba"[chan]); + res[chan] = lp_build_select(&bld, mask, res[chan], dst[chan]); + LLVMBuildStore(builder, res[chan], LLVMBuildGEP(builder, dst_ptr, &index, 1, "")); + } +} + + +/** + * Generate the runtime callable function for the whole fragment pipeline. + */ +static struct lp_fragment_shader_variant * +generate_fragment(struct llvmpipe_context *lp, + struct lp_fragment_shader *shader, + const struct lp_fragment_shader_variant_key *key) +{ + struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen); + struct lp_fragment_shader_variant *variant; + union lp_type fs_type; + union lp_type blend_type; + LLVMTypeRef fs_elem_type; + LLVMTypeRef fs_vec_type; + LLVMTypeRef fs_int_vec_type; + LLVMTypeRef blend_vec_type; + LLVMTypeRef blend_int_vec_type; + LLVMTypeRef arg_types[9]; + LLVMTypeRef func_type; + LLVMValueRef context_ptr; + LLVMValueRef x; + LLVMValueRef y; + LLVMValueRef a0_ptr; + LLVMValueRef dadx_ptr; + LLVMValueRef dady_ptr; + LLVMValueRef mask_ptr; + LLVMValueRef color_ptr; + LLVMValueRef depth_ptr; + LLVMBasicBlockRef block; + LLVMBuilderRef builder; + LLVMValueRef x0; + LLVMValueRef y0; + struct build_fetch_texel_context sampler; + struct lp_build_interp_soa_context interp; + LLVMValueRef fs_mask[LP_MAX_VECTOR_LENGTH]; + LLVMValueRef fs_out_color[NUM_CHANNELS][LP_MAX_VECTOR_LENGTH]; + LLVMValueRef blend_mask; + LLVMValueRef blend_in_color[NUM_CHANNELS]; + unsigned num_fs; + unsigned i; + unsigned chan; + +#ifdef DEBUG + tgsi_dump(shader->base.tokens, 0); + if(key->depth.enabled) { + debug_printf("depth.func = %s\n", debug_dump_func(key->depth.func, TRUE)); + debug_printf("depth.writemask = %u\n", key->depth.writemask); + debug_printf("depth.occlusion_count = %u\n", key->depth.occlusion_count); + } + if(key->alpha.enabled) { + debug_printf("alpha.func = %s\n", debug_dump_func(key->alpha.func, TRUE)); + debug_printf("alpha.ref_value = %f\n", key->alpha.ref_value); + } + if(key->blend.logicop_enable) { + debug_printf("blend.logicop_func = %u\n", key->blend.logicop_func); + } + else if(key->blend.blend_enable) { + debug_printf("blend.rgb_func = %s\n", debug_dump_blend_func (key->blend.rgb_func, TRUE)); + debug_printf("rgb_src_factor = %s\n", debug_dump_blend_factor(key->blend.rgb_src_factor, TRUE)); + debug_printf("rgb_dst_factor = %s\n", debug_dump_blend_factor(key->blend.rgb_dst_factor, TRUE)); + debug_printf("alpha_func = %s\n", debug_dump_blend_func (key->blend.alpha_func, TRUE)); + debug_printf("alpha_src_factor = %s\n", debug_dump_blend_factor(key->blend.alpha_src_factor, TRUE)); + debug_printf("alpha_dst_factor = %s\n", debug_dump_blend_factor(key->blend.alpha_dst_factor, TRUE)); + } + debug_printf("blend.colormask = 0x%x\n", key->blend.colormask); +#endif + + variant = CALLOC_STRUCT(lp_fragment_shader_variant); + if(!variant) + return NULL; + + variant->shader = shader; + memcpy(&variant->key, key, sizeof *key); + + /* TODO: actually pick these based on the fs and color buffer + * characteristics. */ + + fs_type.value = 0; + fs_type.floating = TRUE; /* floating point values */ + fs_type.sign = TRUE; /* values are signed */ + fs_type.norm = FALSE; /* values are not limited to [0,1] or [-1,1] */ + fs_type.width = 32; /* 32-bit float */ + fs_type.length = 4; /* 4 element per vector */ + num_fs = 4; + + blend_type.value = 0; + blend_type.floating = FALSE; /* values are integers */ + blend_type.sign = FALSE; /* values are unsigned */ + blend_type.norm = TRUE; /* values are in [0,1] or [-1,1] */ + blend_type.width = 8; /* 8-bit ubyte values */ + blend_type.length = 16; /* 16 elements per vector */ + + /* + * Generate the function prototype. Any change here must be reflected in + * lp_jit.h's lp_jit_frag_func function pointer type, and vice-versa. + */ + + fs_elem_type = lp_build_elem_type(fs_type); + fs_vec_type = lp_build_vec_type(fs_type); + fs_int_vec_type = lp_build_int_vec_type(fs_type); + + blend_vec_type = lp_build_vec_type(blend_type); + blend_int_vec_type = lp_build_int_vec_type(blend_type); + + arg_types[0] = screen->context_ptr_type; /* context */ + arg_types[1] = LLVMInt32Type(); /* x */ + arg_types[2] = LLVMInt32Type(); /* y */ + arg_types[3] = LLVMPointerType(fs_elem_type, 0); /* a0 */ + arg_types[4] = LLVMPointerType(fs_elem_type, 0); /* dadx */ + arg_types[5] = LLVMPointerType(fs_elem_type, 0); /* dady */ + arg_types[6] = LLVMPointerType(fs_int_vec_type, 0); /* mask */ + arg_types[7] = LLVMPointerType(blend_vec_type, 0); /* color */ + arg_types[8] = LLVMPointerType(fs_int_vec_type, 0); /* depth */ + + func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0); + + variant->function = LLVMAddFunction(screen->module, "shader", func_type); + LLVMSetFunctionCallConv(variant->function, LLVMCCallConv); + for(i = 0; i < Elements(arg_types); ++i) + if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind) + LLVMAddAttribute(LLVMGetParam(variant->function, i), LLVMNoAliasAttribute); + + context_ptr = LLVMGetParam(variant->function, 0); + x = LLVMGetParam(variant->function, 1); + y = LLVMGetParam(variant->function, 2); + a0_ptr = LLVMGetParam(variant->function, 3); + dadx_ptr = LLVMGetParam(variant->function, 4); + dady_ptr = LLVMGetParam(variant->function, 5); + mask_ptr = LLVMGetParam(variant->function, 6); + color_ptr = LLVMGetParam(variant->function, 7); + depth_ptr = LLVMGetParam(variant->function, 8); + + lp_build_name(context_ptr, "context"); + lp_build_name(x, "x"); + lp_build_name(y, "y"); + lp_build_name(a0_ptr, "a0"); + lp_build_name(dadx_ptr, "dadx"); + lp_build_name(dady_ptr, "dady"); + lp_build_name(mask_ptr, "mask"); + lp_build_name(color_ptr, "color"); + lp_build_name(depth_ptr, "depth"); + + /* + * Function body + */ + + block = LLVMAppendBasicBlock(variant->function, "entry"); + builder = LLVMCreateBuilder(); + LLVMPositionBuilderAtEnd(builder, block); + + generate_pos0(builder, x, y, &x0, &y0); + + lp_build_interp_soa_init(&interp, shader->base.tokens, builder, fs_type, + a0_ptr, dadx_ptr, dady_ptr, + x0, y0, 2, 0); + + memset(&sampler, 0, sizeof sampler); + sampler.context_ptr = context_ptr; + + for(i = 0; i < num_fs; ++i) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef out_color[NUM_CHANNELS]; + LLVMValueRef depth_ptr_i; + + if(i != 0) + lp_build_interp_soa_update(&interp); + + fs_mask[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, mask_ptr, &index, 1, ""), ""); + depth_ptr_i = LLVMBuildGEP(builder, depth_ptr, &index, 1, ""); + + generate_fs(lp, shader, key, + builder, + fs_type, + context_ptr, + i, + &interp, + &sampler, + &fs_mask[i], + out_color, + depth_ptr_i); + + for(chan = 0; chan < NUM_CHANNELS; ++chan) + fs_out_color[chan][i] = out_color[chan]; + } + + /* + * Convert the fs's output color and mask to fit to the blending type. + */ + + for(chan = 0; chan < NUM_CHANNELS; ++chan) { + lp_build_conv(builder, fs_type, blend_type, + fs_out_color[chan], num_fs, + &blend_in_color[chan], 1); + lp_build_name(blend_in_color[chan], "color.%c", "rgba"[chan]); + + } + + lp_build_conv_mask(builder, fs_type, blend_type, + fs_mask, num_fs, + &blend_mask, 1); + + /* + * Blending. + */ + + generate_blend(&key->blend, + builder, + blend_type, + context_ptr, + blend_mask, + blend_in_color, + color_ptr); + + LLVMBuildRetVoid(builder); + + LLVMDisposeBuilder(builder); + + /* + * Translate the LLVM IR into machine code. + */ + + LLVMRunFunctionPassManager(screen->pass, variant->function); + +#ifdef DEBUG + LLVMDumpValue(variant->function); + debug_printf("\n"); +#endif + + if(LLVMVerifyFunction(variant->function, LLVMPrintMessageAction)) { + LLVMDumpValue(variant->function); + abort(); + } + + variant->jit_function = (lp_jit_frag_func)LLVMGetPointerToGlobal(screen->engine, variant->function); + +#ifdef DEBUG + lp_disassemble(variant->jit_function); +#endif + + variant->next = shader->variants; + shader->variants = variant; + + return variant; +} + + +void * +llvmpipe_create_fs_state(struct pipe_context *pipe, + const struct pipe_shader_state *templ) +{ + struct lp_fragment_shader *shader; + + shader = CALLOC_STRUCT(lp_fragment_shader); + if (!shader) + return NULL; + + /* get/save the summary info for this shader */ + tgsi_scan_shader(templ->tokens, &shader->info); + + /* we need to keep a local copy of the tokens */ + shader->base.tokens = tgsi_dup_tokens(templ->tokens); + + return shader; +} + + +void +llvmpipe_bind_fs_state(struct pipe_context *pipe, void *fs) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + llvmpipe->fs = (struct lp_fragment_shader *) fs; + + llvmpipe->dirty |= LP_NEW_FS; +} + + +void +llvmpipe_delete_fs_state(struct pipe_context *pipe, void *fs) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + struct llvmpipe_screen *screen = llvmpipe_screen(pipe->screen); + struct lp_fragment_shader *shader = fs; + struct lp_fragment_shader_variant *variant; + + assert(fs != llvmpipe->fs); + + variant = shader->variants; + while(variant) { + struct lp_fragment_shader_variant *next = variant->next; + + if(variant->function) { + if(variant->jit_function) + LLVMFreeMachineCodeForFunction(screen->engine, variant->function); + LLVMDeleteFunction(variant->function); + } + + FREE(variant); + + variant = next; + } + + FREE((void *) shader->base.tokens); + FREE(shader); +} + + + +void +llvmpipe_set_constant_buffer(struct pipe_context *pipe, + uint shader, uint index, + const struct pipe_constant_buffer *buf) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + assert(shader < PIPE_SHADER_TYPES); + assert(index == 0); + + /* note: reference counting */ + pipe_buffer_reference(&llvmpipe->constants[shader].buffer, + buf ? buf->buffer : NULL); + + llvmpipe->dirty |= LP_NEW_CONSTANTS; +} + + +/** + * We need to generate several variants of the fragment pipeline to match + * all the combinations of the contributing state atoms. + * + * TODO: there is actually no reason to tie this to context state -- the + * generated code could be cached globally in the screen. + */ +static void +make_variant_key(struct llvmpipe_context *lp, + struct lp_fragment_shader_variant_key *key) +{ + memset(key, 0, sizeof *key); + + memcpy(&key->depth, &lp->depth_stencil->depth, sizeof key->depth); + + key->alpha.enabled = lp->depth_stencil->alpha.enabled; + if(key->alpha.enabled) + key->alpha.func = lp->depth_stencil->alpha.func; + /* alpha.ref_value is passed in jit_context */ + + memcpy(&key->blend, lp->blend, sizeof key->blend); +} + + +void +llvmpipe_update_fs(struct llvmpipe_context *lp) +{ + struct lp_fragment_shader *shader = lp->fs; + struct lp_fragment_shader_variant_key key; + struct lp_fragment_shader_variant *variant; + + make_variant_key(lp, &key); + + variant = shader->variants; + while(variant) { + if(memcmp(&variant->key, &key, sizeof key) == 0) + break; + + variant = variant->next; + } + + if(!variant) + variant = generate_fragment(lp, shader, &key); + + shader->current = variant; +} diff --git a/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c b/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c new file mode 100644 index 00000000000..4561c6b8456 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c @@ -0,0 +1,62 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "pipe/p_defines.h" +#include "util/u_memory.h" +#include "lp_context.h" +#include "lp_state.h" +#include "draw/draw_context.h" + + + +void * +llvmpipe_create_rasterizer_state(struct pipe_context *pipe, + const struct pipe_rasterizer_state *rast) +{ + return mem_dup(rast, sizeof(*rast)); +} + +void llvmpipe_bind_rasterizer_state(struct pipe_context *pipe, + void *setup) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + /* pass-through to draw module */ + draw_set_rasterizer_state(llvmpipe->draw, setup); + + llvmpipe->rasterizer = (struct pipe_rasterizer_state *)setup; + + llvmpipe->dirty |= LP_NEW_RASTERIZER; +} + +void llvmpipe_delete_rasterizer_state(struct pipe_context *pipe, + void *rasterizer) +{ + FREE( rasterizer ); +} + + diff --git a/src/gallium/drivers/llvmpipe/lp_state_sampler.c b/src/gallium/drivers/llvmpipe/lp_state_sampler.c new file mode 100644 index 00000000000..4fef541b1e3 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_state_sampler.c @@ -0,0 +1,117 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: + * Brian Paul + */ + +#include "util/u_memory.h" + +#include "draw/draw_context.h" + +#include "lp_context.h" +#include "lp_context.h" +#include "lp_state.h" +#include "lp_texture.h" +#include "lp_tex_cache.h" +#include "draw/draw_context.h" + + + +void * +llvmpipe_create_sampler_state(struct pipe_context *pipe, + const struct pipe_sampler_state *sampler) +{ + return mem_dup(sampler, sizeof(*sampler)); +} + + +void +llvmpipe_bind_sampler_states(struct pipe_context *pipe, + unsigned num, void **sampler) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + unsigned i; + + assert(num <= PIPE_MAX_SAMPLERS); + + /* Check for no-op */ + if (num == llvmpipe->num_samplers && + !memcmp(llvmpipe->sampler, sampler, num * sizeof(void *))) + return; + + draw_flush(llvmpipe->draw); + + for (i = 0; i < num; ++i) + llvmpipe->sampler[i] = sampler[i]; + for (i = num; i < PIPE_MAX_SAMPLERS; ++i) + llvmpipe->sampler[i] = NULL; + + llvmpipe->num_samplers = num; + + llvmpipe->dirty |= LP_NEW_SAMPLER; +} + + +void +llvmpipe_set_sampler_textures(struct pipe_context *pipe, + unsigned num, struct pipe_texture **texture) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + uint i; + + assert(num <= PIPE_MAX_SAMPLERS); + + /* Check for no-op */ + if (num == llvmpipe->num_textures && + !memcmp(llvmpipe->texture, texture, num * sizeof(struct pipe_texture *))) + return; + + draw_flush(llvmpipe->draw); + + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + struct pipe_texture *tex = i < num ? texture[i] : NULL; + + pipe_texture_reference(&llvmpipe->texture[i], tex); + lp_tex_tile_cache_set_texture(llvmpipe->tex_cache[i], tex); + } + + llvmpipe->num_textures = num; + + llvmpipe->dirty |= LP_NEW_TEXTURE; +} + + +void +llvmpipe_delete_sampler_state(struct pipe_context *pipe, + void *sampler) +{ + FREE( sampler ); +} + + + diff --git a/src/gallium/drivers/llvmpipe/lp_state_surface.c b/src/gallium/drivers/llvmpipe/lp_state_surface.c new file mode 100644 index 00000000000..177a26b7b1f --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_state_surface.c @@ -0,0 +1,106 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "lp_context.h" +#include "lp_state.h" +#include "lp_surface.h" +#include "lp_tile_cache.h" + +#include "draw/draw_context.h" + + +/** + * XXX this might get moved someday + * Set the framebuffer surface info: color buffers, zbuffer, stencil buffer. + * Here, we flush the old surfaces and update the tile cache to point to the new + * surfaces. + */ +void +llvmpipe_set_framebuffer_state(struct pipe_context *pipe, + const struct pipe_framebuffer_state *fb) +{ + struct llvmpipe_context *lp = llvmpipe_context(pipe); + uint i; + + for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) { + /* check if changing cbuf */ + if (lp->framebuffer.cbufs[i] != fb->cbufs[i]) { + /* flush old */ + lp_flush_tile_cache(lp->cbuf_cache[i]); + + /* assign new */ + lp->framebuffer.cbufs[i] = fb->cbufs[i]; + + /* update cache */ + lp_tile_cache_set_surface(lp->cbuf_cache[i], fb->cbufs[i]); + } + } + + lp->framebuffer.nr_cbufs = fb->nr_cbufs; + + /* zbuf changing? */ + if (lp->framebuffer.zsbuf != fb->zsbuf) { + + if(lp->zsbuf_transfer) { + struct pipe_screen *screen = pipe->screen; + + if(lp->zsbuf_map) { + screen->transfer_unmap(screen, lp->zsbuf_transfer); + lp->zsbuf_map = NULL; + } + + screen->tex_transfer_destroy(lp->zsbuf_transfer); + lp->zsbuf_transfer = NULL; + } + + /* assign new */ + lp->framebuffer.zsbuf = fb->zsbuf; + + /* Tell draw module how deep the Z/depth buffer is */ + if (lp->framebuffer.zsbuf) { + int depth_bits; + double mrd; + depth_bits = pf_get_component_bits(lp->framebuffer.zsbuf->format, + PIPE_FORMAT_COMP_Z); + if (depth_bits > 16) { + mrd = 0.0000001; + } + else { + mrd = 0.00002; + } + draw_set_mrd(lp->draw, mrd); + } + } + + lp->framebuffer.width = fb->width; + lp->framebuffer.height = fb->height; + + lp->dirty |= LP_NEW_FRAMEBUFFER; +} diff --git a/src/gallium/drivers/llvmpipe/lp_state_vertex.c b/src/gallium/drivers/llvmpipe/lp_state_vertex.c new file mode 100644 index 00000000000..1a17631a4c6 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_state_vertex.c @@ -0,0 +1,73 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "lp_context.h" +#include "lp_state.h" +#include "lp_surface.h" + +#include "draw/draw_context.h" + + +void +llvmpipe_set_vertex_elements(struct pipe_context *pipe, + unsigned count, + const struct pipe_vertex_element *attribs) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + assert(count <= PIPE_MAX_ATTRIBS); + + memcpy(llvmpipe->vertex_element, attribs, + count * sizeof(struct pipe_vertex_element)); + llvmpipe->num_vertex_elements = count; + + llvmpipe->dirty |= LP_NEW_VERTEX; + + draw_set_vertex_elements(llvmpipe->draw, count, attribs); +} + + +void +llvmpipe_set_vertex_buffers(struct pipe_context *pipe, + unsigned count, + const struct pipe_vertex_buffer *buffers) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + assert(count <= PIPE_MAX_ATTRIBS); + + memcpy(llvmpipe->vertex_buffer, buffers, count * sizeof(buffers[0])); + llvmpipe->num_vertex_buffers = count; + + llvmpipe->dirty |= LP_NEW_VERTEX; + + draw_set_vertex_buffers(llvmpipe->draw, count, buffers); +} diff --git a/src/gallium/drivers/llvmpipe/lp_state_vs.c b/src/gallium/drivers/llvmpipe/lp_state_vs.c new file mode 100644 index 00000000000..15c30296144 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_state_vs.c @@ -0,0 +1,96 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "pipe/p_defines.h" +#include "tgsi/tgsi_parse.h" +#include "util/u_memory.h" +#include "draw/draw_context.h" + +#include "lp_context.h" +#include "lp_state.h" + + +void * +llvmpipe_create_vs_state(struct pipe_context *pipe, + const struct pipe_shader_state *templ) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + struct lp_vertex_shader *state; + + state = CALLOC_STRUCT(lp_vertex_shader); + if (state == NULL ) + goto fail; + + /* copy shader tokens, the ones passed in will go away. + */ + state->shader.tokens = tgsi_dup_tokens(templ->tokens); + if (state->shader.tokens == NULL) + goto fail; + + state->draw_data = draw_create_vertex_shader(llvmpipe->draw, templ); + if (state->draw_data == NULL) + goto fail; + + return state; + +fail: + if (state) { + FREE( (void *)state->shader.tokens ); + FREE( state->draw_data ); + FREE( state ); + } + return NULL; +} + + +void +llvmpipe_bind_vs_state(struct pipe_context *pipe, void *vs) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + llvmpipe->vs = (const struct lp_vertex_shader *)vs; + + draw_bind_vertex_shader(llvmpipe->draw, + (llvmpipe->vs ? llvmpipe->vs->draw_data : NULL)); + + llvmpipe->dirty |= LP_NEW_VS; +} + + +void +llvmpipe_delete_vs_state(struct pipe_context *pipe, void *vs) +{ + struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); + + struct lp_vertex_shader *state = + (struct lp_vertex_shader *)vs; + + draw_delete_vertex_shader(llvmpipe->draw, state->draw_data); + FREE( state ); +} diff --git a/src/gallium/drivers/llvmpipe/lp_surface.c b/src/gallium/drivers/llvmpipe/lp_surface.c new file mode 100644 index 00000000000..6110b0a193e --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_surface.c @@ -0,0 +1,50 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "util/u_rect.h" +#include "lp_context.h" +#include "lp_surface.h" + + +static void +lp_surface_copy(struct pipe_context *pipe, + struct pipe_surface *dest, unsigned destx, unsigned desty, + struct pipe_surface *src, unsigned srcx, unsigned srcy, + unsigned width, unsigned height) +{ + util_surface_copy(pipe, FALSE, + dest, destx, desty, + src, srcx, srcy, + width, height); +} + +void +lp_init_surface_functions(struct llvmpipe_context *lp) +{ + lp->pipe.surface_copy = lp_surface_copy; + lp->pipe.surface_fill = util_surface_fill; +} diff --git a/src/gallium/drivers/llvmpipe/lp_surface.h b/src/gallium/drivers/llvmpipe/lp_surface.h new file mode 100644 index 00000000000..4d78a53c4f5 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_surface.h @@ -0,0 +1,42 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef LP_SURFACE_H +#define LP_SURFACE_H + + +struct llvmpipe_context; + + +extern void +lp_init_surface_functions(struct llvmpipe_context *lp); + + +#endif /* LP_SURFACE_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_test.h b/src/gallium/drivers/llvmpipe/lp_test.h new file mode 100644 index 00000000000..69aaae26e0a --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_test.h @@ -0,0 +1,128 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Shared testing code. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#ifndef LP_TEST_H +#define LP_TEST_H + + +#include <stdlib.h> +#include <stdio.h> +#include <float.h> + +#include <llvm-c/Core.h> +#include <llvm-c/Analysis.h> +#include <llvm-c/ExecutionEngine.h> +#include <llvm-c/Target.h> +#include <llvm-c/BitWriter.h> +#include <llvm-c/Transforms/Scalar.h> + +#include "pipe/p_state.h" +#include "util/u_format.h" +#include "util/u_math.h" +#include "util/u_debug_dump.h" + +#include "lp_bld_type.h" + + +void +write_tsv_header(FILE *fp); + + +boolean +test_some(unsigned verbose, FILE *fp, unsigned long n); + + +boolean +test_all(unsigned verbose, FILE *fp); + + +static INLINE uint64_t +rdtsc(void) +{ +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) + uint32_t hi, lo; + __asm__ __volatile__ ("rdtsc" : "=a"(lo), "=d"(hi)); + return ((uint64_t)lo) | (((uint64_t)hi) << 32); +#else + return 0; +#endif +} + + +float +random_float(void); + + +void +dump_type(FILE *fp, union lp_type type); + + +double +read_elem(union lp_type type, const void *src, unsigned index); + + +void +write_elem(union lp_type type, void *dst, unsigned index, double src); + + +void +random_elem(union lp_type type, void *dst, unsigned index); + + +void +read_vec(union lp_type type, const void *src, double *dst); + + +void +write_vec(union lp_type type, void *dst, const double *src); + + +void +random_vec(union lp_type type, void *dst); + + +boolean +compare_vec_with_eps(union lp_type type, const void *res, const void *ref, double eps); + + +boolean +compare_vec(union lp_type type, const void *res, const void *ref); + + +void +dump_vec(FILE *fp, union lp_type type, const void *src); + + +#endif /* !LP_TEST_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_test_blend.c b/src/gallium/drivers/llvmpipe/lp_test_blend.c new file mode 100644 index 00000000000..8dfad468e3c --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_test_blend.c @@ -0,0 +1,881 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * @file + * Unit tests for blend LLVM IR generation + * + * @author Jose Fonseca <jfonseca@vmware.com> + * + * Blend computation code derived from code written by + * @author Brian Paul <brian@vmware.com> + */ + + +#include "lp_bld_type.h" +#include "lp_bld_arit.h" +#include "lp_bld_blend.h" +#include "lp_bld_debug.h" +#include "lp_test.h" + + +enum vector_mode +{ + AoS = 0, + SoA = 1 +}; + + +typedef void (*blend_test_ptr_t)(const void *src, const void *dst, const void *con, void *res); + + +void +write_tsv_header(FILE *fp) +{ + fprintf(fp, + "result\t" + "cycles_per_channel\t" + "mode\t" + "type\t" + "sep_func\t" + "sep_src_factor\t" + "sep_dst_factor\t" + "rgb_func\t" + "rgb_src_factor\t" + "rgb_dst_factor\t" + "alpha_func\t" + "alpha_src_factor\t" + "alpha_dst_factor\n"); + + fflush(fp); +} + + +static void +write_tsv_row(FILE *fp, + const struct pipe_blend_state *blend, + enum vector_mode mode, + union lp_type type, + double cycles, + boolean success) +{ + fprintf(fp, "%s\t", success ? "pass" : "fail"); + + if (mode == AoS) { + fprintf(fp, "%.1f\t", cycles / type.length); + fprintf(fp, "aos\t"); + } + + if (mode == SoA) { + fprintf(fp, "%.1f\t", cycles / (4 * type.length)); + fprintf(fp, "soa\t"); + } + + fprintf(fp, "%s%u%sx%u\t", + type.floating ? "f" : (type.fixed ? "h" : (type.sign ? "s" : "u")), + type.width, + type.norm ? "n" : "", + type.length); + + fprintf(fp, + "%s\t%s\t%s\t", + blend->rgb_func != blend->alpha_func ? "true" : "false", + blend->rgb_src_factor != blend->alpha_src_factor ? "true" : "false", + blend->rgb_dst_factor != blend->alpha_dst_factor ? "true" : "false"); + + fprintf(fp, + "%s\t%s\t%s\t%s\t%s\t%s\n", + debug_dump_blend_func(blend->rgb_func, TRUE), + debug_dump_blend_factor(blend->rgb_src_factor, TRUE), + debug_dump_blend_factor(blend->rgb_dst_factor, TRUE), + debug_dump_blend_func(blend->alpha_func, TRUE), + debug_dump_blend_factor(blend->alpha_src_factor, TRUE), + debug_dump_blend_factor(blend->alpha_dst_factor, TRUE)); + + fflush(fp); +} + + +static void +dump_blend_type(FILE *fp, + const struct pipe_blend_state *blend, + enum vector_mode mode, + union lp_type type) +{ + fprintf(fp, "%s", mode ? "soa" : "aos"); + + fprintf(fp, " type=%s%u%sx%u", + type.floating ? "f" : (type.fixed ? "h" : (type.sign ? "s" : "u")), + type.width, + type.norm ? "n" : "", + type.length); + + fprintf(fp, + " %s=%s %s=%s %s=%s %s=%s %s=%s %s=%s", + "rgb_func", debug_dump_blend_func(blend->rgb_func, TRUE), + "rgb_src_factor", debug_dump_blend_factor(blend->rgb_src_factor, TRUE), + "rgb_dst_factor", debug_dump_blend_factor(blend->rgb_dst_factor, TRUE), + "alpha_func", debug_dump_blend_func(blend->alpha_func, TRUE), + "alpha_src_factor", debug_dump_blend_factor(blend->alpha_src_factor, TRUE), + "alpha_dst_factor", debug_dump_blend_factor(blend->alpha_dst_factor, TRUE)); + + fprintf(fp, " ...\n"); + fflush(fp); +} + + +static LLVMValueRef +add_blend_test(LLVMModuleRef module, + const struct pipe_blend_state *blend, + enum vector_mode mode, + union lp_type type) +{ + LLVMTypeRef ret_type; + LLVMTypeRef vec_type; + LLVMTypeRef args[4]; + LLVMValueRef func; + LLVMValueRef src_ptr; + LLVMValueRef dst_ptr; + LLVMValueRef const_ptr; + LLVMValueRef res_ptr; + LLVMBasicBlockRef block; + LLVMBuilderRef builder; + + ret_type = LLVMInt64Type(); + vec_type = lp_build_vec_type(type); + + args[3] = args[2] = args[1] = args[0] = LLVMPointerType(vec_type, 0); + func = LLVMAddFunction(module, "test", LLVMFunctionType(LLVMVoidType(), args, 4, 0)); + LLVMSetFunctionCallConv(func, LLVMCCallConv); + src_ptr = LLVMGetParam(func, 0); + dst_ptr = LLVMGetParam(func, 1); + const_ptr = LLVMGetParam(func, 2); + res_ptr = LLVMGetParam(func, 3); + + block = LLVMAppendBasicBlock(func, "entry"); + builder = LLVMCreateBuilder(); + LLVMPositionBuilderAtEnd(builder, block); + + if (mode == AoS) { + LLVMValueRef src; + LLVMValueRef dst; + LLVMValueRef con; + LLVMValueRef res; + + src = LLVMBuildLoad(builder, src_ptr, "src"); + dst = LLVMBuildLoad(builder, dst_ptr, "dst"); + con = LLVMBuildLoad(builder, const_ptr, "const"); + + res = lp_build_blend_aos(builder, blend, type, src, dst, con, 3); + + lp_build_name(res, "res"); + + LLVMBuildStore(builder, res, res_ptr); + } + + if (mode == SoA) { + LLVMValueRef src[4]; + LLVMValueRef dst[4]; + LLVMValueRef con[4]; + LLVMValueRef res[4]; + unsigned i; + + for(i = 0; i < 4; ++i) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + src[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, src_ptr, &index, 1, ""), ""); + dst[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, dst_ptr, &index, 1, ""), ""); + con[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, const_ptr, &index, 1, ""), ""); + lp_build_name(src[i], "src.%c", "rgba"[i]); + lp_build_name(con[i], "con.%c", "rgba"[i]); + lp_build_name(dst[i], "dst.%c", "rgba"[i]); + } + + lp_build_blend_soa(builder, blend, type, src, dst, con, res); + + for(i = 0; i < 4; ++i) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + lp_build_name(res[i], "res.%c", "rgba"[i]); + LLVMBuildStore(builder, res[i], LLVMBuildGEP(builder, res_ptr, &index, 1, "")); + } + } + + LLVMBuildRetVoid(builder);; + + LLVMDisposeBuilder(builder); + return func; +} + + +/** Add and limit result to ceiling of 1.0 */ +#define ADD_SAT(R, A, B) \ +do { \ + R = (A) + (B); if (R > 1.0f) R = 1.0f; \ +} while (0) + +/** Subtract and limit result to floor of 0.0 */ +#define SUB_SAT(R, A, B) \ +do { \ + R = (A) - (B); if (R < 0.0f) R = 0.0f; \ +} while (0) + + +static void +compute_blend_ref_term(unsigned rgb_factor, + unsigned alpha_factor, + const double *factor, + const double *src, + const double *dst, + const double *con, + double *term) +{ + double temp; + + switch (rgb_factor) { + case PIPE_BLENDFACTOR_ONE: + term[0] = factor[0]; /* R */ + term[1] = factor[1]; /* G */ + term[2] = factor[2]; /* B */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + term[0] = factor[0] * src[0]; /* R */ + term[1] = factor[1] * src[1]; /* G */ + term[2] = factor[2] * src[2]; /* B */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA: + term[0] = factor[0] * src[3]; /* R */ + term[1] = factor[1] * src[3]; /* G */ + term[2] = factor[2] * src[3]; /* B */ + break; + case PIPE_BLENDFACTOR_DST_COLOR: + term[0] = factor[0] * dst[0]; /* R */ + term[1] = factor[1] * dst[1]; /* G */ + term[2] = factor[2] * dst[2]; /* B */ + break; + case PIPE_BLENDFACTOR_DST_ALPHA: + term[0] = factor[0] * dst[3]; /* R */ + term[1] = factor[1] * dst[3]; /* G */ + term[2] = factor[2] * dst[3]; /* B */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + temp = MIN2(src[3], 1.0f - dst[3]); + term[0] = factor[0] * temp; /* R */ + term[1] = factor[1] * temp; /* G */ + term[2] = factor[2] * temp; /* B */ + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + term[0] = factor[0] * con[0]; /* R */ + term[1] = factor[1] * con[1]; /* G */ + term[2] = factor[2] * con[2]; /* B */ + break; + case PIPE_BLENDFACTOR_CONST_ALPHA: + term[0] = factor[0] * con[3]; /* R */ + term[1] = factor[1] * con[3]; /* G */ + term[2] = factor[2] * con[3]; /* B */ + break; + case PIPE_BLENDFACTOR_SRC1_COLOR: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_SRC1_ALPHA: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_ZERO: + term[0] = 0.0f; /* R */ + term[1] = 0.0f; /* G */ + term[2] = 0.0f; /* B */ + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + term[0] = factor[0] * (1.0f - src[0]); /* R */ + term[1] = factor[1] * (1.0f - src[1]); /* G */ + term[2] = factor[2] * (1.0f - src[2]); /* B */ + break; + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + term[0] = factor[0] * (1.0f - src[3]); /* R */ + term[1] = factor[1] * (1.0f - src[3]); /* G */ + term[2] = factor[2] * (1.0f - src[3]); /* B */ + break; + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + term[0] = factor[0] * (1.0f - dst[3]); /* R */ + term[1] = factor[1] * (1.0f - dst[3]); /* G */ + term[2] = factor[2] * (1.0f - dst[3]); /* B */ + break; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + term[0] = factor[0] * (1.0f - dst[0]); /* R */ + term[1] = factor[1] * (1.0f - dst[1]); /* G */ + term[2] = factor[2] * (1.0f - dst[2]); /* B */ + break; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + term[0] = factor[0] * (1.0f - con[0]); /* R */ + term[1] = factor[1] * (1.0f - con[1]); /* G */ + term[2] = factor[2] * (1.0f - con[2]); /* B */ + break; + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + term[0] = factor[0] * (1.0f - con[3]); /* R */ + term[1] = factor[1] * (1.0f - con[3]); /* G */ + term[2] = factor[2] * (1.0f - con[3]); /* B */ + break; + case PIPE_BLENDFACTOR_INV_SRC1_COLOR: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: + assert(0); /* to do */ + break; + default: + assert(0); + } + + /* + * Compute src/first term A + */ + switch (alpha_factor) { + case PIPE_BLENDFACTOR_ONE: + term[3] = factor[3]; /* A */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + case PIPE_BLENDFACTOR_SRC_ALPHA: + term[3] = factor[3] * src[3]; /* A */ + break; + case PIPE_BLENDFACTOR_DST_COLOR: + case PIPE_BLENDFACTOR_DST_ALPHA: + term[3] = factor[3] * dst[3]; /* A */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + term[3] = src[3]; /* A */ + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + case PIPE_BLENDFACTOR_CONST_ALPHA: + term[3] = factor[3] * con[3]; /* A */ + break; + case PIPE_BLENDFACTOR_ZERO: + term[3] = 0.0f; /* A */ + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + term[3] = factor[3] * (1.0f - src[3]); /* A */ + break; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + term[3] = factor[3] * (1.0f - dst[3]); /* A */ + break; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + term[3] = factor[3] * (1.0f - con[3]); + break; + default: + assert(0); + } +} + + +static void +compute_blend_ref(const struct pipe_blend_state *blend, + const double *src, + const double *dst, + const double *con, + double *res) +{ + double src_term[4]; + double dst_term[4]; + + compute_blend_ref_term(blend->rgb_src_factor, blend->alpha_src_factor, src, src, dst, con, src_term); + compute_blend_ref_term(blend->rgb_dst_factor, blend->alpha_dst_factor, dst, src, dst, con, dst_term); + + /* + * Combine RGB terms + */ + switch (blend->rgb_func) { + case PIPE_BLEND_ADD: + ADD_SAT(res[0], src_term[0], dst_term[0]); /* R */ + ADD_SAT(res[1], src_term[1], dst_term[1]); /* G */ + ADD_SAT(res[2], src_term[2], dst_term[2]); /* B */ + break; + case PIPE_BLEND_SUBTRACT: + SUB_SAT(res[0], src_term[0], dst_term[0]); /* R */ + SUB_SAT(res[1], src_term[1], dst_term[1]); /* G */ + SUB_SAT(res[2], src_term[2], dst_term[2]); /* B */ + break; + case PIPE_BLEND_REVERSE_SUBTRACT: + SUB_SAT(res[0], dst_term[0], src_term[0]); /* R */ + SUB_SAT(res[1], dst_term[1], src_term[1]); /* G */ + SUB_SAT(res[2], dst_term[2], src_term[2]); /* B */ + break; + case PIPE_BLEND_MIN: + res[0] = MIN2(src_term[0], dst_term[0]); /* R */ + res[1] = MIN2(src_term[1], dst_term[1]); /* G */ + res[2] = MIN2(src_term[2], dst_term[2]); /* B */ + break; + case PIPE_BLEND_MAX: + res[0] = MAX2(src_term[0], dst_term[0]); /* R */ + res[1] = MAX2(src_term[1], dst_term[1]); /* G */ + res[2] = MAX2(src_term[2], dst_term[2]); /* B */ + break; + default: + assert(0); + } + + /* + * Combine A terms + */ + switch (blend->alpha_func) { + case PIPE_BLEND_ADD: + ADD_SAT(res[3], src_term[3], dst_term[3]); /* A */ + break; + case PIPE_BLEND_SUBTRACT: + SUB_SAT(res[3], src_term[3], dst_term[3]); /* A */ + break; + case PIPE_BLEND_REVERSE_SUBTRACT: + SUB_SAT(res[3], dst_term[3], src_term[3]); /* A */ + break; + case PIPE_BLEND_MIN: + res[3] = MIN2(src_term[3], dst_term[3]); /* A */ + break; + case PIPE_BLEND_MAX: + res[3] = MAX2(src_term[3], dst_term[3]); /* A */ + break; + default: + assert(0); + } +} + + +static boolean +test_one(unsigned verbose, + FILE *fp, + const struct pipe_blend_state *blend, + enum vector_mode mode, + union lp_type type) +{ + LLVMModuleRef module = NULL; + LLVMValueRef func = NULL; + LLVMExecutionEngineRef engine = NULL; + LLVMModuleProviderRef provider = NULL; + LLVMPassManagerRef pass = NULL; + char *error = NULL; + blend_test_ptr_t blend_test_ptr; + boolean success; + const unsigned n = 32; + int64_t cycles[n]; + double cycles_avg = 0.0; + unsigned i, j; + + if(verbose >= 1) + dump_blend_type(stdout, blend, mode, type); + + module = LLVMModuleCreateWithName("test"); + + func = add_blend_test(module, blend, mode, type); + + if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) { + LLVMDumpModule(module); + abort(); + } + LLVMDisposeMessage(error); + + provider = LLVMCreateModuleProviderForExistingModule(module); + if (LLVMCreateJITCompiler(&engine, provider, 1, &error)) { + if(verbose < 1) + dump_blend_type(stderr, blend, mode, type); + fprintf(stderr, "%s\n", error); + LLVMDisposeMessage(error); + abort(); + } + +#if 0 + pass = LLVMCreatePassManager(); + LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass); + /* These are the passes currently listed in llvm-c/Transforms/Scalar.h, + * but there are more on SVN. */ + LLVMAddConstantPropagationPass(pass); + LLVMAddInstructionCombiningPass(pass); + LLVMAddPromoteMemoryToRegisterPass(pass); + LLVMAddGVNPass(pass); + LLVMAddCFGSimplificationPass(pass); + LLVMRunPassManager(pass, module); +#else + (void)pass; +#endif + + if(verbose >= 2) + LLVMDumpModule(module); + + blend_test_ptr = (blend_test_ptr_t)LLVMGetPointerToGlobal(engine, func); + + if(verbose >= 2) + lp_disassemble(blend_test_ptr); + + success = TRUE; + for(i = 0; i < n && success; ++i) { + if(mode == AoS) { + uint8_t src[LP_MAX_VECTOR_LENGTH*LP_MAX_TYPE_WIDTH/8]; + uint8_t dst[LP_MAX_VECTOR_LENGTH*LP_MAX_TYPE_WIDTH/8]; + uint8_t con[LP_MAX_VECTOR_LENGTH*LP_MAX_TYPE_WIDTH/8]; + uint8_t res[LP_MAX_VECTOR_LENGTH*LP_MAX_TYPE_WIDTH/8]; + uint8_t ref[LP_MAX_VECTOR_LENGTH*LP_MAX_TYPE_WIDTH/8]; + int64_t start_counter = 0; + int64_t end_counter = 0; + + random_vec(type, src); + random_vec(type, dst); + random_vec(type, con); + + { + double fsrc[LP_MAX_VECTOR_LENGTH]; + double fdst[LP_MAX_VECTOR_LENGTH]; + double fcon[LP_MAX_VECTOR_LENGTH]; + double fref[LP_MAX_VECTOR_LENGTH]; + + read_vec(type, src, fsrc); + read_vec(type, dst, fdst); + read_vec(type, con, fcon); + + for(j = 0; j < type.length; j += 4) + compute_blend_ref(blend, fsrc + j, fdst + j, fcon + j, fref + j); + + write_vec(type, ref, fref); + } + + start_counter = rdtsc(); + blend_test_ptr(src, dst, con, res); + end_counter = rdtsc(); + + cycles[i] = end_counter - start_counter; + + if(!compare_vec(type, res, ref)) { + success = FALSE; + + if(verbose < 1) + dump_blend_type(stderr, blend, mode, type); + fprintf(stderr, "MISMATCH\n"); + + fprintf(stderr, " Src: "); + dump_vec(stderr, type, src); + fprintf(stderr, "\n"); + + fprintf(stderr, " Dst: "); + dump_vec(stderr, type, dst); + fprintf(stderr, "\n"); + + fprintf(stderr, " Con: "); + dump_vec(stderr, type, con); + fprintf(stderr, "\n"); + + fprintf(stderr, " Res: "); + dump_vec(stderr, type, res); + fprintf(stderr, "\n"); + + fprintf(stderr, " Ref: "); + dump_vec(stderr, type, ref); + fprintf(stderr, "\n"); + } + } + + if(mode == SoA) { + const unsigned stride = type.length*type.width/8; + uint8_t src[4*LP_MAX_VECTOR_LENGTH*LP_MAX_TYPE_WIDTH/8]; + uint8_t dst[4*LP_MAX_VECTOR_LENGTH*LP_MAX_TYPE_WIDTH/8]; + uint8_t con[4*LP_MAX_VECTOR_LENGTH*LP_MAX_TYPE_WIDTH/8]; + uint8_t res[4*LP_MAX_VECTOR_LENGTH*LP_MAX_TYPE_WIDTH/8]; + uint8_t ref[4*LP_MAX_VECTOR_LENGTH*LP_MAX_TYPE_WIDTH/8]; + int64_t start_counter = 0; + int64_t end_counter = 0; + boolean mismatch; + + for(j = 0; j < 4; ++j) { + random_vec(type, src + j*stride); + random_vec(type, dst + j*stride); + random_vec(type, con + j*stride); + } + + { + double fsrc[4]; + double fdst[4]; + double fcon[4]; + double fref[4]; + unsigned k; + + for(k = 0; k < type.length; ++k) { + for(j = 0; j < 4; ++j) { + fsrc[j] = read_elem(type, src + j*stride, k); + fdst[j] = read_elem(type, dst + j*stride, k); + fcon[j] = read_elem(type, con + j*stride, k); + } + + compute_blend_ref(blend, fsrc, fdst, fcon, fref); + + for(j = 0; j < 4; ++j) + write_elem(type, ref + j*stride, k, fref[j]); + } + } + + start_counter = rdtsc(); + blend_test_ptr(src, dst, con, res); + end_counter = rdtsc(); + + cycles[i] = end_counter - start_counter; + + mismatch = FALSE; + for (j = 0; j < 4; ++j) + if(!compare_vec(type, res + j*stride, ref + j*stride)) + mismatch = TRUE; + + if (mismatch) { + success = FALSE; + + if(verbose < 1) + dump_blend_type(stderr, blend, mode, type); + fprintf(stderr, "MISMATCH\n"); + for(j = 0; j < 4; ++j) { + char channel = "RGBA"[j]; + fprintf(stderr, " Src%c: ", channel); + dump_vec(stderr, type, src + j*stride); + fprintf(stderr, "\n"); + + fprintf(stderr, " Dst%c: ", channel); + dump_vec(stderr, type, dst + j*stride); + fprintf(stderr, "\n"); + + fprintf(stderr, " Con%c: ", channel); + dump_vec(stderr, type, con + j*stride); + fprintf(stderr, "\n"); + + fprintf(stderr, " Res%c: ", channel); + dump_vec(stderr, type, res + j*stride); + fprintf(stderr, "\n"); + + fprintf(stderr, " Ref%c: ", channel); + dump_vec(stderr, type, ref + j*stride); + fprintf(stderr, "\n"); + } + } + } + } + + /* + * Unfortunately the output of cycle counter is not very reliable as it comes + * -- sometimes we get outliers (due IRQs perhaps?) which are + * better removed to avoid random or biased data. + */ + { + double sum = 0.0, sum2 = 0.0; + double avg, std; + unsigned m; + + for(i = 0; i < n; ++i) { + sum += cycles[i]; + sum2 += cycles[i]*cycles[i]; + } + + avg = sum/n; + std = sqrtf((sum2 - n*avg*avg)/n); + + m = 0; + sum = 0.0; + for(i = 0; i < n; ++i) { + if(fabs(cycles[i] - avg) <= 4.0*std) { + sum += cycles[i]; + ++m; + } + } + + cycles_avg = sum/m; + + } + + if(fp) + write_tsv_row(fp, blend, mode, type, cycles_avg, success); + + if (!success) { + if(verbose < 2) + LLVMDumpModule(module); + LLVMWriteBitcodeToFile(module, "blend.bc"); + fprintf(stderr, "blend.bc written\n"); + fprintf(stderr, "Invoke as \"llc -o - blend.bc\"\n"); + abort(); + } + + LLVMFreeMachineCodeForFunction(engine, func); + + LLVMDisposeExecutionEngine(engine); + if(pass) + LLVMDisposePassManager(pass); + + return success; +} + + +const unsigned +blend_factors[] = { + PIPE_BLENDFACTOR_ZERO, + PIPE_BLENDFACTOR_ONE, + PIPE_BLENDFACTOR_SRC_COLOR, + PIPE_BLENDFACTOR_SRC_ALPHA, + PIPE_BLENDFACTOR_DST_COLOR, + PIPE_BLENDFACTOR_DST_ALPHA, + PIPE_BLENDFACTOR_CONST_COLOR, + PIPE_BLENDFACTOR_CONST_ALPHA, +#if 0 + PIPE_BLENDFACTOR_SRC1_COLOR, + PIPE_BLENDFACTOR_SRC1_ALPHA, +#endif + PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE, + PIPE_BLENDFACTOR_INV_SRC_COLOR, + PIPE_BLENDFACTOR_INV_SRC_ALPHA, + PIPE_BLENDFACTOR_INV_DST_COLOR, + PIPE_BLENDFACTOR_INV_DST_ALPHA, + PIPE_BLENDFACTOR_INV_CONST_COLOR, + PIPE_BLENDFACTOR_INV_CONST_ALPHA, +#if 0 + PIPE_BLENDFACTOR_INV_SRC1_COLOR, + PIPE_BLENDFACTOR_INV_SRC1_ALPHA, +#endif +}; + + +const unsigned +blend_funcs[] = { + PIPE_BLEND_ADD, + PIPE_BLEND_SUBTRACT, + PIPE_BLEND_REVERSE_SUBTRACT, + PIPE_BLEND_MIN, + PIPE_BLEND_MAX +}; + + +const union lp_type blend_types[] = { + /* float, fixed, sign, norm, width, len */ + {{ TRUE, FALSE, FALSE, TRUE, 32, 4 }}, /* f32 x 4 */ + {{ FALSE, FALSE, FALSE, TRUE, 8, 16 }}, /* u8n x 16 */ +}; + + +const unsigned num_funcs = sizeof(blend_funcs)/sizeof(blend_funcs[0]); +const unsigned num_factors = sizeof(blend_factors)/sizeof(blend_factors[0]); +const unsigned num_types = sizeof(blend_types)/sizeof(blend_types[0]); + + +boolean +test_all(unsigned verbose, FILE *fp) +{ + const unsigned *rgb_func; + const unsigned *rgb_src_factor; + const unsigned *rgb_dst_factor; + const unsigned *alpha_func; + const unsigned *alpha_src_factor; + const unsigned *alpha_dst_factor; + struct pipe_blend_state blend; + enum vector_mode mode; + const union lp_type *type; + bool success = TRUE; + + for(rgb_func = blend_funcs; rgb_func < &blend_funcs[num_funcs]; ++rgb_func) { + for(alpha_func = blend_funcs; alpha_func < &blend_funcs[num_funcs]; ++alpha_func) { + for(rgb_src_factor = blend_factors; rgb_src_factor < &blend_factors[num_factors]; ++rgb_src_factor) { + for(rgb_dst_factor = blend_factors; rgb_dst_factor <= rgb_src_factor; ++rgb_dst_factor) { + for(alpha_src_factor = blend_factors; alpha_src_factor < &blend_factors[num_factors]; ++alpha_src_factor) { + for(alpha_dst_factor = blend_factors; alpha_dst_factor <= alpha_src_factor; ++alpha_dst_factor) { + for(mode = 0; mode < 2; ++mode) { + for(type = blend_types; type < &blend_types[num_types]; ++type) { + + if(*rgb_dst_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || + *alpha_dst_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) + continue; + + memset(&blend, 0, sizeof blend); + blend.blend_enable = 1; + blend.rgb_func = *rgb_func; + blend.rgb_src_factor = *rgb_src_factor; + blend.rgb_dst_factor = *rgb_dst_factor; + blend.alpha_func = *alpha_func; + blend.alpha_src_factor = *alpha_src_factor; + blend.alpha_dst_factor = *alpha_dst_factor; + blend.colormask = PIPE_MASK_RGBA; + + if(!test_one(verbose, fp, &blend, mode, *type)) + success = FALSE; + + } + } + } + } + } + } + } + } + + return success; +} + + +boolean +test_some(unsigned verbose, FILE *fp, unsigned long n) +{ + const unsigned *rgb_func; + const unsigned *rgb_src_factor; + const unsigned *rgb_dst_factor; + const unsigned *alpha_func; + const unsigned *alpha_src_factor; + const unsigned *alpha_dst_factor; + struct pipe_blend_state blend; + enum vector_mode mode; + const union lp_type *type; + unsigned long i; + bool success = TRUE; + + for(i = 0; i < n; ++i) { + rgb_func = &blend_funcs[random() % num_funcs]; + alpha_func = &blend_funcs[random() % num_funcs]; + rgb_src_factor = &blend_factors[random() % num_factors]; + alpha_src_factor = &blend_factors[random() % num_factors]; + + do { + rgb_dst_factor = &blend_factors[random() % num_factors]; + } while(*rgb_dst_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE); + + do { + alpha_dst_factor = &blend_factors[random() % num_factors]; + } while(*alpha_dst_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE); + + mode = random() & 1; + + type = &blend_types[random() % num_types]; + + memset(&blend, 0, sizeof blend); + blend.blend_enable = 1; + blend.rgb_func = *rgb_func; + blend.rgb_src_factor = *rgb_src_factor; + blend.rgb_dst_factor = *rgb_dst_factor; + blend.alpha_func = *alpha_func; + blend.alpha_src_factor = *alpha_src_factor; + blend.alpha_dst_factor = *alpha_dst_factor; + blend.colormask = PIPE_MASK_RGBA; + + if(!test_one(verbose, fp, &blend, mode, *type)) + success = FALSE; + } + + return success; +} diff --git a/src/gallium/drivers/llvmpipe/lp_test_conv.c b/src/gallium/drivers/llvmpipe/lp_test_conv.c new file mode 100644 index 00000000000..e6489834af5 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_test_conv.c @@ -0,0 +1,427 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * @file + * Unit tests for type conversion. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#include "lp_bld_type.h" +#include "lp_bld_const.h" +#include "lp_bld_conv.h" +#include "lp_bld_debug.h" +#include "lp_test.h" + + +typedef void (*conv_test_ptr_t)(const void *src, const void *dst); + + +void +write_tsv_header(FILE *fp) +{ + fprintf(fp, + "result\t" + "cycles_per_channel\t" + "src_type\t" + "dst_type\n"); + + fflush(fp); +} + + +static void +write_tsv_row(FILE *fp, + union lp_type src_type, + union lp_type dst_type, + double cycles, + boolean success) +{ + fprintf(fp, "%s\t", success ? "pass" : "fail"); + + fprintf(fp, "%.1f\t", cycles / MAX2(src_type.length, dst_type.length)); + + dump_type(fp, src_type); + fprintf(fp, "\t"); + + dump_type(fp, dst_type); + fprintf(fp, "\n"); + + fflush(fp); +} + + +static void +dump_conv_types(FILE *fp, + union lp_type src_type, + union lp_type dst_type) +{ + fprintf(fp, "src_type="); + dump_type(fp, src_type); + + fprintf(fp, " dst_type="); + dump_type(fp, dst_type); + + fprintf(fp, " ...\n"); + fflush(fp); +} + + +static LLVMValueRef +add_conv_test(LLVMModuleRef module, + union lp_type src_type, unsigned num_srcs, + union lp_type dst_type, unsigned num_dsts) +{ + LLVMTypeRef args[2]; + LLVMValueRef func; + LLVMValueRef src_ptr; + LLVMValueRef dst_ptr; + LLVMBasicBlockRef block; + LLVMBuilderRef builder; + LLVMValueRef src[LP_MAX_VECTOR_LENGTH]; + LLVMValueRef dst[LP_MAX_VECTOR_LENGTH]; + unsigned i; + + args[0] = LLVMPointerType(lp_build_vec_type(src_type), 0); + args[1] = LLVMPointerType(lp_build_vec_type(dst_type), 0); + + func = LLVMAddFunction(module, "test", LLVMFunctionType(LLVMVoidType(), args, 2, 0)); + LLVMSetFunctionCallConv(func, LLVMCCallConv); + src_ptr = LLVMGetParam(func, 0); + dst_ptr = LLVMGetParam(func, 1); + + block = LLVMAppendBasicBlock(func, "entry"); + builder = LLVMCreateBuilder(); + LLVMPositionBuilderAtEnd(builder, block); + + for(i = 0; i < num_srcs; ++i) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef ptr = LLVMBuildGEP(builder, src_ptr, &index, 1, ""); + src[i] = LLVMBuildLoad(builder, ptr, ""); + } + + lp_build_conv(builder, src_type, dst_type, src, num_srcs, dst, num_dsts); + + for(i = 0; i < num_dsts; ++i) { + LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef ptr = LLVMBuildGEP(builder, dst_ptr, &index, 1, ""); + LLVMBuildStore(builder, dst[i], ptr); + } + + LLVMBuildRetVoid(builder);; + + LLVMDisposeBuilder(builder); + return func; +} + + +static boolean +test_one(unsigned verbose, + FILE *fp, + union lp_type src_type, + union lp_type dst_type) +{ + LLVMModuleRef module = NULL; + LLVMValueRef func = NULL; + LLVMExecutionEngineRef engine = NULL; + LLVMModuleProviderRef provider = NULL; + LLVMPassManagerRef pass = NULL; + char *error = NULL; + conv_test_ptr_t conv_test_ptr; + boolean success; + const unsigned n = 32; + int64_t cycles[n]; + double cycles_avg = 0.0; + unsigned num_srcs; + unsigned num_dsts; + double eps; + unsigned i, j; + + if(verbose >= 1) + dump_conv_types(stdout, src_type, dst_type); + + if(src_type.length > dst_type.length) { + num_srcs = 1; + num_dsts = src_type.length/dst_type.length; + } + else { + num_dsts = 1; + num_srcs = dst_type.length/src_type.length; + } + + assert(src_type.width * src_type.length == dst_type.width * dst_type.length); + + /* We must not loose or gain channels. Only precision */ + assert(src_type.length * num_srcs == dst_type.length * num_dsts); + + eps = MAX2(lp_const_eps(src_type), lp_const_eps(dst_type)); + + module = LLVMModuleCreateWithName("test"); + + func = add_conv_test(module, src_type, num_srcs, dst_type, num_dsts); + + if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) { + LLVMDumpModule(module); + abort(); + } + LLVMDisposeMessage(error); + + provider = LLVMCreateModuleProviderForExistingModule(module); + if (LLVMCreateJITCompiler(&engine, provider, 1, &error)) { + if(verbose < 1) + dump_conv_types(stderr, src_type, dst_type); + fprintf(stderr, "%s\n", error); + LLVMDisposeMessage(error); + abort(); + } + +#if 0 + pass = LLVMCreatePassManager(); + LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass); + /* These are the passes currently listed in llvm-c/Transforms/Scalar.h, + * but there are more on SVN. */ + LLVMAddConstantPropagationPass(pass); + LLVMAddInstructionCombiningPass(pass); + LLVMAddPromoteMemoryToRegisterPass(pass); + LLVMAddGVNPass(pass); + LLVMAddCFGSimplificationPass(pass); + LLVMRunPassManager(pass, module); +#else + (void)pass; +#endif + + if(verbose >= 2) + LLVMDumpModule(module); + + conv_test_ptr = (conv_test_ptr_t)LLVMGetPointerToGlobal(engine, func); + + if(verbose >= 2) + lp_disassemble(conv_test_ptr); + + success = TRUE; + for(i = 0; i < n && success; ++i) { + unsigned src_stride = src_type.length*src_type.width/8; + unsigned dst_stride = dst_type.length*dst_type.width/8; + uint8_t src[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH]; + uint8_t dst[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH]; + double fref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH]; + uint8_t ref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH]; + int64_t start_counter = 0; + int64_t end_counter = 0; + + for(j = 0; j < num_srcs; ++j) { + random_vec(src_type, src + j*src_stride); + read_vec(src_type, src + j*src_stride, fref + j*src_type.length); + } + + for(j = 0; j < num_dsts; ++j) { + write_vec(dst_type, ref + j*dst_stride, fref + j*dst_type.length); + } + + start_counter = rdtsc(); + conv_test_ptr(src, dst); + end_counter = rdtsc(); + + cycles[i] = end_counter - start_counter; + + for(j = 0; j < num_dsts; ++j) { + if(!compare_vec_with_eps(dst_type, dst + j*dst_stride, ref + j*dst_stride, eps)) + success = FALSE; + } + + if (!success) { + if(verbose < 1) + dump_conv_types(stderr, src_type, dst_type); + fprintf(stderr, "MISMATCH\n"); + + for(j = 0; j < num_srcs; ++j) { + fprintf(stderr, " Src%u: ", j); + dump_vec(stderr, src_type, src + j*src_stride); + fprintf(stderr, "\n"); + } + +#if 1 + fprintf(stderr, " Ref: "); + for(j = 0; j < src_type.length*num_srcs; ++j) + fprintf(stderr, " %f", fref[j]); + fprintf(stderr, "\n"); +#endif + + for(j = 0; j < num_dsts; ++j) { + fprintf(stderr, " Dst%u: ", j); + dump_vec(stderr, dst_type, dst + j*dst_stride); + fprintf(stderr, "\n"); + + fprintf(stderr, " Ref%u: ", j); + dump_vec(stderr, dst_type, ref + j*dst_stride); + fprintf(stderr, "\n"); + } + } + } + + /* + * Unfortunately the output of cycle counter is not very reliable as it comes + * -- sometimes we get outliers (due IRQs perhaps?) which are + * better removed to avoid random or biased data. + */ + { + double sum = 0.0, sum2 = 0.0; + double avg, std; + unsigned m; + + for(i = 0; i < n; ++i) { + sum += cycles[i]; + sum2 += cycles[i]*cycles[i]; + } + + avg = sum/n; + std = sqrtf((sum2 - n*avg*avg)/n); + + m = 0; + sum = 0.0; + for(i = 0; i < n; ++i) { + if(fabs(cycles[i] - avg) <= 4.0*std) { + sum += cycles[i]; + ++m; + } + } + + cycles_avg = sum/m; + + } + + if(fp) + write_tsv_row(fp, src_type, dst_type, cycles_avg, success); + + if (!success) { + static boolean firsttime = TRUE; + if(firsttime) { + if(verbose < 2) + LLVMDumpModule(module); + LLVMWriteBitcodeToFile(module, "conv.bc"); + fprintf(stderr, "conv.bc written\n"); + fprintf(stderr, "Invoke as \"llc -o - conv.bc\"\n"); + firsttime = FALSE; + //abort(); + } + } + + LLVMFreeMachineCodeForFunction(engine, func); + + LLVMDisposeExecutionEngine(engine); + if(pass) + LLVMDisposePassManager(pass); + + return success; +} + + +const union lp_type conv_types[] = { + /* float, fixed, sign, norm, width, len */ + + {{ TRUE, FALSE, TRUE, TRUE, 32, 4 }}, + {{ TRUE, FALSE, TRUE, FALSE, 32, 4 }}, + {{ TRUE, FALSE, FALSE, TRUE, 32, 4 }}, + {{ TRUE, FALSE, FALSE, FALSE, 32, 4 }}, + + /* TODO: test fixed formats too */ + + {{ FALSE, FALSE, TRUE, TRUE, 16, 8 }}, + {{ FALSE, FALSE, TRUE, FALSE, 16, 8 }}, + {{ FALSE, FALSE, FALSE, TRUE, 16, 8 }}, + {{ FALSE, FALSE, FALSE, FALSE, 16, 8 }}, + + {{ FALSE, FALSE, TRUE, TRUE, 32, 4 }}, + {{ FALSE, FALSE, TRUE, FALSE, 32, 4 }}, + {{ FALSE, FALSE, FALSE, TRUE, 32, 4 }}, + {{ FALSE, FALSE, FALSE, FALSE, 32, 4 }}, + + {{ FALSE, FALSE, TRUE, TRUE, 16, 8 }}, + {{ FALSE, FALSE, TRUE, FALSE, 16, 8 }}, + {{ FALSE, FALSE, FALSE, TRUE, 16, 8 }}, + {{ FALSE, FALSE, FALSE, FALSE, 16, 8 }}, + + {{ FALSE, FALSE, TRUE, TRUE, 8, 16 }}, + {{ FALSE, FALSE, TRUE, FALSE, 8, 16 }}, + {{ FALSE, FALSE, FALSE, TRUE, 8, 16 }}, + {{ FALSE, FALSE, FALSE, FALSE, 8, 16 }}, +}; + + +const unsigned num_types = sizeof(conv_types)/sizeof(conv_types[0]); + + +boolean +test_all(unsigned verbose, FILE *fp) +{ + const union lp_type *src_type; + const union lp_type *dst_type; + bool success = TRUE; + + for(src_type = conv_types; src_type < &conv_types[num_types]; ++src_type) { + for(dst_type = conv_types; dst_type < &conv_types[num_types]; ++dst_type) { + + if(src_type == dst_type) + continue; + + if(src_type->norm != dst_type->norm) + continue; + + if(!test_one(verbose, fp, *src_type, *dst_type)) + success = FALSE; + + } + } + + return success; +} + + +boolean +test_some(unsigned verbose, FILE *fp, unsigned long n) +{ + const union lp_type *src_type; + const union lp_type *dst_type; + unsigned long i; + bool success = TRUE; + + for(i = 0; i < n; ++i) { + src_type = &conv_types[random() % num_types]; + + do { + dst_type = &conv_types[random() % num_types]; + } while (src_type == dst_type || src_type->norm != dst_type->norm); + + if(!test_one(verbose, fp, *src_type, *dst_type)) + success = FALSE; + } + + return success; +} diff --git a/src/gallium/drivers/llvmpipe/lp_test_format.c b/src/gallium/drivers/llvmpipe/lp_test_format.c new file mode 100644 index 00000000000..1d192355eed --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_test_format.c @@ -0,0 +1,272 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include <stdlib.h> +#include <stdio.h> + +#include <llvm-c/Core.h> +#include <llvm-c/Analysis.h> +#include <llvm-c/ExecutionEngine.h> +#include <llvm-c/Target.h> +#include <llvm-c/Transforms/Scalar.h> + +#include "util/u_format.h" + +#include "lp_bld_flow.h" +#include "lp_bld_format.h" + + +struct pixel_test_case +{ + enum pipe_format format; + uint32_t packed; + double unpacked[4]; +}; + + +struct pixel_test_case test_cases[] = +{ + {PIPE_FORMAT_R5G6B5_UNORM, 0x0000, {0.0, 0.0, 0.0, 1.0}}, + {PIPE_FORMAT_R5G6B5_UNORM, 0x001f, {0.0, 0.0, 1.0, 1.0}}, + {PIPE_FORMAT_R5G6B5_UNORM, 0x07e0, {0.0, 1.0, 0.0, 1.0}}, + {PIPE_FORMAT_R5G6B5_UNORM, 0xf800, {1.0, 0.0, 0.0, 1.0}}, + {PIPE_FORMAT_R5G6B5_UNORM, 0xffff, {1.0, 1.0, 1.0, 1.0}}, + + {PIPE_FORMAT_A1R5G5B5_UNORM, 0x0000, {0.0, 0.0, 0.0, 0.0}}, + {PIPE_FORMAT_A1R5G5B5_UNORM, 0x001f, {0.0, 0.0, 1.0, 0.0}}, + {PIPE_FORMAT_A1R5G5B5_UNORM, 0x03e0, {0.0, 1.0, 0.0, 0.0}}, + {PIPE_FORMAT_A1R5G5B5_UNORM, 0x7c00, {1.0, 0.0, 0.0, 0.0}}, + {PIPE_FORMAT_A1R5G5B5_UNORM, 0x8000, {0.0, 0.0, 0.0, 1.0}}, + {PIPE_FORMAT_A1R5G5B5_UNORM, 0xffff, {1.0, 1.0, 1.0, 1.0}}, + + {PIPE_FORMAT_A8R8G8B8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}}, + {PIPE_FORMAT_A8R8G8B8_UNORM, 0x000000ff, {0.0, 0.0, 1.0, 0.0}}, + {PIPE_FORMAT_A8R8G8B8_UNORM, 0x0000ff00, {0.0, 1.0, 0.0, 0.0}}, + {PIPE_FORMAT_A8R8G8B8_UNORM, 0x00ff0000, {1.0, 0.0, 0.0, 0.0}}, + {PIPE_FORMAT_A8R8G8B8_UNORM, 0xff000000, {0.0, 0.0, 0.0, 1.0}}, + {PIPE_FORMAT_A8R8G8B8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}}, + +#if 0 + {PIPE_FORMAT_R8G8B8A8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}}, + {PIPE_FORMAT_R8G8B8A8_UNORM, 0x000000ff, {0.0, 0.0, 0.0, 1.0}}, + {PIPE_FORMAT_R8G8B8A8_UNORM, 0x0000ff00, {0.0, 0.0, 1.0, 0.0}}, + {PIPE_FORMAT_R8G8B8A8_UNORM, 0x00ff0000, {0.0, 1.0, 0.0, 0.0}}, + {PIPE_FORMAT_R8G8B8A8_UNORM, 0xff000000, {1.0, 0.0, 0.0, 0.0}}, + {PIPE_FORMAT_R8G8B8A8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}}, +#endif + + {PIPE_FORMAT_B8G8R8A8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}}, + {PIPE_FORMAT_B8G8R8A8_UNORM, 0x000000ff, {0.0, 0.0, 0.0, 1.0}}, + {PIPE_FORMAT_B8G8R8A8_UNORM, 0x0000ff00, {1.0, 0.0, 0.0, 0.0}}, + {PIPE_FORMAT_B8G8R8A8_UNORM, 0x00ff0000, {0.0, 1.0, 0.0, 0.0}}, + {PIPE_FORMAT_B8G8R8A8_UNORM, 0xff000000, {0.0, 0.0, 1.0, 0.0}}, + {PIPE_FORMAT_B8G8R8A8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}}, +}; + + +typedef void (*load_ptr_t)(const void *, float *); + + +static LLVMValueRef +add_load_rgba_test(LLVMModuleRef module, + enum pipe_format format) +{ + LLVMTypeRef args[2]; + LLVMValueRef func; + LLVMValueRef ptr; + LLVMValueRef rgba_ptr; + LLVMBasicBlockRef block; + LLVMBuilderRef builder; + LLVMValueRef rgba; + struct lp_build_loop_state loop; + + args[0] = LLVMPointerType(LLVMInt8Type(), 0); + args[1] = LLVMPointerType(LLVMVectorType(LLVMFloatType(), 4), 0); + + func = LLVMAddFunction(module, "load", LLVMFunctionType(LLVMVoidType(), args, 2, 0)); + LLVMSetFunctionCallConv(func, LLVMCCallConv); + ptr = LLVMGetParam(func, 0); + rgba_ptr = LLVMGetParam(func, 1); + + block = LLVMAppendBasicBlock(func, "entry"); + builder = LLVMCreateBuilder(); + LLVMPositionBuilderAtEnd(builder, block); + + lp_build_loop_begin(builder, LLVMConstInt(LLVMInt32Type(), 1, 0), &loop); + + rgba = lp_build_load_rgba(builder, format, ptr); + LLVMBuildStore(builder, rgba, rgba_ptr); + + lp_build_loop_end(builder, LLVMConstInt(LLVMInt32Type(), 4, 0), NULL, &loop); + + LLVMBuildRetVoid(builder); + + LLVMDisposeBuilder(builder); + return func; +} + + +typedef void (*store_ptr_t)(void *, const float *); + + +static LLVMValueRef +add_store_rgba_test(LLVMModuleRef module, + enum pipe_format format) +{ + LLVMTypeRef args[2]; + LLVMValueRef func; + LLVMValueRef ptr; + LLVMValueRef rgba_ptr; + LLVMBasicBlockRef block; + LLVMBuilderRef builder; + LLVMValueRef rgba; + + args[0] = LLVMPointerType(LLVMInt8Type(), 0); + args[1] = LLVMPointerType(LLVMVectorType(LLVMFloatType(), 4), 0); + + func = LLVMAddFunction(module, "store", LLVMFunctionType(LLVMVoidType(), args, 2, 0)); + LLVMSetFunctionCallConv(func, LLVMCCallConv); + ptr = LLVMGetParam(func, 0); + rgba_ptr = LLVMGetParam(func, 1); + + block = LLVMAppendBasicBlock(func, "entry"); + builder = LLVMCreateBuilder(); + LLVMPositionBuilderAtEnd(builder, block); + + rgba = LLVMBuildLoad(builder, rgba_ptr, ""); + + lp_build_store_rgba(builder, format, ptr, rgba); + + LLVMBuildRetVoid(builder); + + LLVMDisposeBuilder(builder); + return func; +} + + +static boolean +test_format(const struct pixel_test_case *test) +{ + LLVMModuleRef module = NULL; + LLVMValueRef load = NULL; + LLVMValueRef store = NULL; + LLVMExecutionEngineRef engine = NULL; + LLVMModuleProviderRef provider = NULL; + LLVMPassManagerRef pass = NULL; + char *error = NULL; + const struct util_format_description *desc; + load_ptr_t load_ptr; + store_ptr_t store_ptr; + float unpacked[4]; + unsigned packed; + boolean success; + unsigned i; + + desc = util_format_description(test->format); + fprintf(stderr, "%s\n", desc->name); + + module = LLVMModuleCreateWithName("test"); + + load = add_load_rgba_test(module, test->format); + store = add_store_rgba_test(module, test->format); + + if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) { + LLVMDumpModule(module); + abort(); + } + LLVMDisposeMessage(error); + + provider = LLVMCreateModuleProviderForExistingModule(module); + if (LLVMCreateJITCompiler(&engine, provider, 1, &error)) { + fprintf(stderr, "%s\n", error); + LLVMDisposeMessage(error); + abort(); + } + +#if 0 + pass = LLVMCreatePassManager(); + LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass); + /* These are the passes currently listed in llvm-c/Transforms/Scalar.h, + * but there are more on SVN. */ + LLVMAddConstantPropagationPass(pass); + LLVMAddInstructionCombiningPass(pass); + LLVMAddPromoteMemoryToRegisterPass(pass); + LLVMAddGVNPass(pass); + LLVMAddCFGSimplificationPass(pass); + LLVMRunPassManager(pass, module); +#else + (void)pass; +#endif + + load_ptr = (load_ptr_t) LLVMGetPointerToGlobal(engine, load); + store_ptr = (store_ptr_t)LLVMGetPointerToGlobal(engine, store); + + memset(unpacked, 0, sizeof unpacked); + packed = 0; + + load_ptr(&test->packed, unpacked); + store_ptr(&packed, unpacked); + + success = TRUE; + if(test->packed != packed) + success = FALSE; + for(i = 0; i < 4; ++i) + if(test->unpacked[i] != unpacked[i]) + success = FALSE; + + if (!success) { + printf("FAILED\n"); + printf(" Packed: %08x\n", test->packed); + printf(" %08x\n", packed); + printf(" Unpacked: %f %f %f %f\n", unpacked[0], unpacked[1], unpacked[2], unpacked[3]); + printf(" %f %f %f %f\n", test->unpacked[0], test->unpacked[1], test->unpacked[2], test->unpacked[3]); + LLVMDumpModule(module); + } + + LLVMFreeMachineCodeForFunction(engine, store); + LLVMFreeMachineCodeForFunction(engine, load); + + LLVMDisposeExecutionEngine(engine); + if(pass) + LLVMDisposePassManager(pass); + + return success; +} + + +int main(int argc, char **argv) +{ + unsigned i; + int ret; + + for (i = 0; i < sizeof(test_cases)/sizeof(test_cases[0]); ++i) + if(!test_format(&test_cases[i])) + ret = 1; + + return ret; +} diff --git a/src/gallium/drivers/llvmpipe/lp_test_main.c b/src/gallium/drivers/llvmpipe/lp_test_main.c new file mode 100644 index 00000000000..49213fb4f0b --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_test_main.c @@ -0,0 +1,384 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * @file + * Shared testing code. + * + * @author Jose Fonseca <jfonseca@vmware.com> + */ + + +#include "lp_bld_const.h" +#include "lp_test.h" + + +void +dump_type(FILE *fp, + union lp_type type) +{ + fprintf(fp, "%s%s%u%sx%u", + type.sign ? (type.floating || type.fixed ? "" : "s") : "u", + type.floating ? "f" : (type.fixed ? "h" : "i"), + type.width, + type.norm ? "n" : "", + type.length); +} + + +double +read_elem(union lp_type type, const void *src, unsigned index) +{ + double scale = lp_const_scale(type); + double value; + assert(index < type.length); + if (type.floating) { + switch(type.width) { + case 32: + value = *((const float *)src + index); + break; + case 64: + value = *((const double *)src + index); + break; + default: + assert(0); + return 0.0; + } + } + else { + if(type.sign) { + switch(type.width) { + case 8: + value = *((const int8_t *)src + index); + break; + case 16: + value = *((const int16_t *)src + index); + break; + case 32: + value = *((const int32_t *)src + index); + break; + case 64: + value = *((const int64_t *)src + index); + break; + default: + assert(0); + return 0.0; + } + } + else { + switch(type.width) { + case 8: + value = *((const uint8_t *)src + index); + break; + case 16: + value = *((const uint16_t *)src + index); + break; + case 32: + value = *((const uint32_t *)src + index); + break; + case 64: + value = *((const uint64_t *)src + index); + break; + default: + assert(0); + return 0.0; + } + } + } + return value/scale; +} + + +void +write_elem(union lp_type type, void *dst, unsigned index, double value) +{ + assert(index < type.length); + if(!type.sign && value < 0.0) + value = 0.0; + if(type.norm && value < -1.0) + value = -1.0; + if(type.norm && value > 1.0) + value = 1.0; + if (type.floating) { + switch(type.width) { + case 32: + *((float *)dst + index) = (float)(value); + break; + case 64: + *((double *)dst + index) = value; + break; + default: + assert(0); + } + } + else { + double scale = lp_const_scale(type); + value = round(value*scale); + if(type.sign) { + long long lvalue = (long long)value; + lvalue = MIN2(lvalue, ((long long)1 << (type.width - 1)) - 1); + switch(type.width) { + case 8: + *((int8_t *)dst + index) = (int8_t)lvalue; + break; + case 16: + *((int16_t *)dst + index) = (int16_t)lvalue; + break; + case 32: + *((int32_t *)dst + index) = (int32_t)lvalue; + break; + case 64: + *((int64_t *)dst + index) = (int64_t)lvalue; + break; + default: + assert(0); + } + } + else { + unsigned long long lvalue = (long long)value; + lvalue = MIN2(lvalue, ((unsigned long long)1 << type.width) - 1); + switch(type.width) { + case 8: + *((uint8_t *)dst + index) = (uint8_t)lvalue; + break; + case 16: + *((uint16_t *)dst + index) = (uint16_t)lvalue; + break; + case 32: + *((uint32_t *)dst + index) = (uint32_t)lvalue; + break; + case 64: + *((uint64_t *)dst + index) = (uint64_t)lvalue; + break; + default: + assert(0); + } + } + } +} + + +void +random_elem(union lp_type type, void *dst, unsigned index) +{ + double value; + assert(index < type.length); + value = (double)random()/(double)RAND_MAX; + if(!type.norm) { + unsigned long long mask; + if (type.floating) + mask = ~(unsigned long long)0; + else if (type.fixed) + mask = ((unsigned long long)1 << (type.width / 2)) - 1; + else if (type.sign) + mask = ((unsigned long long)1 << (type.width - 1)) - 1; + else + mask = ((unsigned long long)1 << type.width) - 1; + value += (double)(mask & random()); + } + if(!type.sign) + if(random() & 1) + value = -value; + write_elem(type, dst, index, value); +} + + +void +read_vec(union lp_type type, const void *src, double *dst) +{ + unsigned i; + for (i = 0; i < type.length; ++i) + dst[i] = read_elem(type, src, i); +} + + +void +write_vec(union lp_type type, void *dst, const double *src) +{ + unsigned i; + for (i = 0; i < type.length; ++i) + write_elem(type, dst, i, src[i]); +} + + +float +random_float(void) +{ + return (float)((double)random()/(double)RAND_MAX); +} + + +void +random_vec(union lp_type type, void *dst) +{ + unsigned i; + for (i = 0; i < type.length; ++i) + random_elem(type, dst, i); +} + + +boolean +compare_vec_with_eps(union lp_type type, const void *res, const void *ref, double eps) +{ + unsigned i; + for (i = 0; i < type.length; ++i) { + double res_elem = read_elem(type, res, i); + double ref_elem = read_elem(type, ref, i); + double delta = fabs(res_elem - ref_elem); + if(delta >= 2.0*eps) + return FALSE; + } + + return TRUE; +} + + +boolean +compare_vec(union lp_type type, const void *res, const void *ref) +{ + double eps = lp_const_eps(type); + return compare_vec_with_eps(type, res, ref, eps); +} + + +void +dump_vec(FILE *fp, union lp_type type, const void *src) +{ + unsigned i; + for (i = 0; i < type.length; ++i) { + if(i) + fprintf(fp, " "); + if (type.floating) { + double value; + switch(type.width) { + case 32: + value = *((const float *)src + i); + break; + case 64: + value = *((const double *)src + i); + break; + default: + assert(0); + value = 0.0; + } + fprintf(fp, "%f", value); + } + else { + if(type.sign && !type.norm) { + long long value; + const char *format; + switch(type.width) { + case 8: + value = *((const int8_t *)src + i); + format = "%3lli"; + break; + case 16: + value = *((const int16_t *)src + i); + format = "%5lli"; + break; + case 32: + value = *((const int32_t *)src + i); + format = "%10lli"; + break; + case 64: + value = *((const int64_t *)src + i); + format = "%20lli"; + break; + default: + assert(0); + value = 0.0; + format = "?"; + } + fprintf(fp, format, value); + } + else { + unsigned long long value; + const char *format; + switch(type.width) { + case 8: + value = *((const uint8_t *)src + i); + format = type.norm ? "%2x" : "%4llu"; + break; + case 16: + value = *((const uint16_t *)src + i); + format = type.norm ? "%4x" : "%6llx"; + break; + case 32: + value = *((const uint32_t *)src + i); + format = type.norm ? "%8x" : "%11llx"; + break; + case 64: + value = *((const uint64_t *)src + i); + format = type.norm ? "%16x" : "%21llx"; + break; + default: + assert(0); + value = 0.0; + format = "?"; + } + fprintf(fp, format, value); + } + } + } +} + + +int main(int argc, char **argv) +{ + unsigned verbose = 0; + FILE *fp = NULL; + unsigned long n = 1000; + unsigned i; + boolean success; + + for(i = 1; i < argc; ++i) { + if(strcmp(argv[i], "-v") == 0) + ++verbose; + else if(strcmp(argv[i], "-o") == 0) + fp = fopen(argv[++i], "wt"); + else + n = atoi(argv[i]); + } + + if(fp) { + /* Warm up the caches */ + test_some(0, NULL, 100); + + write_tsv_header(fp); + } + + if(n) + success = test_some(verbose, fp, n); + else + success = test_all(verbose, fp); + + if(fp) + fclose(fp); + + return success ? 0 : 1; +} diff --git a/src/gallium/drivers/llvmpipe/lp_tex_cache.c b/src/gallium/drivers/llvmpipe/lp_tex_cache.c new file mode 100644 index 00000000000..23a94b5b0d5 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_tex_cache.c @@ -0,0 +1,304 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Texture tile caching. + * + * Author: + * Brian Paul + */ + +#include "pipe/p_inlines.h" +#include "util/u_memory.h" +#include "util/u_tile.h" +#include "util/u_format.h" +#include "lp_context.h" +#include "lp_surface.h" +#include "lp_texture.h" +#include "lp_tex_cache.h" + + + +/** + * Return the position in the cache for the tile that contains win pos (x,y). + * We currently use a direct mapped cache so this is like a hack key. + * At some point we should investige something more sophisticated, like + * a LRU replacement policy. + */ +#define CACHE_POS(x, y) \ + (((x) + (y) * 5) % NUM_ENTRIES) + + + +/** + * Is the tile at (x,y) in cleared state? + */ +static INLINE uint +is_clear_flag_set(const uint *bitvec, union tex_tile_address addr) +{ + int pos, bit; + pos = addr.bits.y * (MAX_TEX_WIDTH / TEX_TILE_SIZE) + addr.bits.x; + assert(pos / 32 < (MAX_TEX_WIDTH / TEX_TILE_SIZE) * (MAX_TEX_HEIGHT / TEX_TILE_SIZE) / 32); + bit = bitvec[pos / 32] & (1 << (pos & 31)); + return bit; +} + + +/** + * Mark the tile at (x,y) as not cleared. + */ +static INLINE void +clear_clear_flag(uint *bitvec, union tex_tile_address addr) +{ + int pos; + pos = addr.bits.y * (MAX_TEX_WIDTH / TEX_TILE_SIZE) + addr.bits.x; + assert(pos / 32 < (MAX_TEX_WIDTH / TEX_TILE_SIZE) * (MAX_TEX_HEIGHT / TEX_TILE_SIZE) / 32); + bitvec[pos / 32] &= ~(1 << (pos & 31)); +} + + +struct llvmpipe_tex_tile_cache * +lp_create_tex_tile_cache( struct pipe_screen *screen ) +{ + struct llvmpipe_tex_tile_cache *tc; + uint pos; + + tc = CALLOC_STRUCT( llvmpipe_tex_tile_cache ); + if (tc) { + tc->screen = screen; + for (pos = 0; pos < NUM_ENTRIES; pos++) { + tc->entries[pos].addr.bits.invalid = 1; + } + tc->last_tile = &tc->entries[0]; /* any tile */ + } + return tc; +} + + +void +lp_destroy_tex_tile_cache(struct llvmpipe_tex_tile_cache *tc) +{ + struct pipe_screen *screen; + uint pos; + + for (pos = 0; pos < NUM_ENTRIES; pos++) { + /*assert(tc->entries[pos].x < 0);*/ + } + if (tc->transfer) { + screen = tc->transfer->texture->screen; + screen->tex_transfer_destroy(tc->transfer); + } + if (tc->tex_trans) { + screen = tc->tex_trans->texture->screen; + screen->tex_transfer_destroy(tc->tex_trans); + } + + FREE( tc ); +} + + +void +lp_tex_tile_cache_map_transfers(struct llvmpipe_tex_tile_cache *tc) +{ + if (tc->transfer && !tc->transfer_map) + tc->transfer_map = tc->screen->transfer_map(tc->screen, tc->transfer); + + if (tc->tex_trans && !tc->tex_trans_map) + tc->tex_trans_map = tc->screen->transfer_map(tc->screen, tc->tex_trans); +} + + +void +lp_tex_tile_cache_unmap_transfers(struct llvmpipe_tex_tile_cache *tc) +{ + if (tc->transfer_map) { + tc->screen->transfer_unmap(tc->screen, tc->transfer); + tc->transfer_map = NULL; + } + + if (tc->tex_trans_map) { + tc->screen->transfer_unmap(tc->screen, tc->tex_trans); + tc->tex_trans_map = NULL; + } +} + +void +lp_tex_tile_cache_validate_texture(struct llvmpipe_tex_tile_cache *tc) +{ + if (tc->texture) { + struct llvmpipe_texture *lpt = llvmpipe_texture(tc->texture); + if (lpt->timestamp != tc->timestamp) { + /* texture was modified, invalidate all cached tiles */ + uint i; + _debug_printf("INV %d %d\n", tc->timestamp, lpt->timestamp); + for (i = 0; i < NUM_ENTRIES; i++) { + tc->entries[i].addr.bits.invalid = 1; + } + + tc->timestamp = lpt->timestamp; + } + } +} + +/** + * Specify the texture to cache. + */ +void +lp_tex_tile_cache_set_texture(struct llvmpipe_tex_tile_cache *tc, + struct pipe_texture *texture) +{ + uint i; + + assert(!tc->transfer); + + if (tc->texture != texture) { + pipe_texture_reference(&tc->texture, texture); + + if (tc->tex_trans) { + struct pipe_screen *screen = tc->tex_trans->texture->screen; + + if (tc->tex_trans_map) { + screen->transfer_unmap(screen, tc->tex_trans); + tc->tex_trans_map = NULL; + } + + screen->tex_transfer_destroy(tc->tex_trans); + tc->tex_trans = NULL; + } + + /* mark as entries as invalid/empty */ + /* XXX we should try to avoid this when the teximage hasn't changed */ + for (i = 0; i < NUM_ENTRIES; i++) { + tc->entries[i].addr.bits.invalid = 1; + } + + tc->tex_face = -1; /* any invalid value here */ + } +} + + +/** + * Given the texture face, level, zslice, x and y values, compute + * the cache entry position/index where we'd hope to find the + * cached texture tile. + * This is basically a direct-map cache. + * XXX There's probably lots of ways in which we can improve this. + */ +static INLINE uint +tex_cache_pos( union tex_tile_address addr ) +{ + uint entry = (addr.bits.x + + addr.bits.y * 9 + + addr.bits.z * 3 + + addr.bits.face + + addr.bits.level * 7); + + return entry % NUM_ENTRIES; +} + +/** + * Similar to lp_get_cached_tile() but for textures. + * Tiles are read-only and indexed with more params. + */ +const struct llvmpipe_cached_tex_tile * +lp_find_cached_tex_tile(struct llvmpipe_tex_tile_cache *tc, + union tex_tile_address addr ) +{ + struct pipe_screen *screen = tc->screen; + struct llvmpipe_cached_tex_tile *tile; + + tile = tc->entries + tex_cache_pos( addr ); + + if (addr.value != tile->addr.value) { + + /* cache miss. Most misses are because we've invaldiated the + * texture cache previously -- most commonly on binding a new + * texture. Currently we effectively flush the cache on texture + * bind. + */ +#if 0 + _debug_printf("miss at %u: x=%d y=%d z=%d face=%d level=%d\n" + " tile %u: x=%d y=%d z=%d face=%d level=%d\n", + pos, x/TEX_TILE_SIZE, y/TEX_TILE_SIZE, z, face, level, + pos, tile->addr.bits.x, tile->addr.bits.y, tile->z, tile->face, tile->level); +#endif + + /* check if we need to get a new transfer */ + if (!tc->tex_trans || + tc->tex_face != addr.bits.face || + tc->tex_level != addr.bits.level || + tc->tex_z != addr.bits.z) { + /* get new transfer (view into texture) */ + + if (tc->tex_trans) { + if (tc->tex_trans_map) { + tc->screen->transfer_unmap(tc->screen, tc->tex_trans); + tc->tex_trans_map = NULL; + } + + screen->tex_transfer_destroy(tc->tex_trans); + tc->tex_trans = NULL; + } + + tc->tex_trans = + screen->get_tex_transfer(screen, tc->texture, + addr.bits.face, + addr.bits.level, + addr.bits.z, + PIPE_TRANSFER_READ, 0, 0, + tc->texture->width[addr.bits.level], + tc->texture->height[addr.bits.level]); + + tc->tex_trans_map = screen->transfer_map(screen, tc->tex_trans); + + tc->tex_face = addr.bits.face; + tc->tex_level = addr.bits.level; + tc->tex_z = addr.bits.z; + } + + { + unsigned x = addr.bits.x * TEX_TILE_SIZE; + unsigned y = addr.bits.y * TEX_TILE_SIZE; + unsigned w = TEX_TILE_SIZE; + unsigned h = TEX_TILE_SIZE; + + if (pipe_clip_tile(x, y, &w, &h, tc->tex_trans)) { + assert(0); + } + + util_format_read_4ub(tc->tex_trans->format, + (uint8_t *)tile->color, sizeof tile->color[0], + tc->tex_trans_map, tc->tex_trans->stride, + x, y, w, h); + } + + tile->addr = addr; + } + + tc->last_tile = tile; + return tile; +} diff --git a/src/gallium/drivers/llvmpipe/lp_tex_cache.h b/src/gallium/drivers/llvmpipe/lp_tex_cache.h new file mode 100644 index 00000000000..9fa6c368125 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_tex_cache.h @@ -0,0 +1,151 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef LP_TEX_CACHE_H +#define LP_TEX_CACHE_H + + +#include "pipe/p_compiler.h" + + +struct llvmpipe_context; +struct llvmpipe_tex_tile_cache; + + +/** + * Cache tile size (width and height). This needs to be a power of two. + */ +#define TEX_TILE_SIZE 64 + + +/* If we need to support > 4096, just expand this to be a 64 bit + * union, or consider tiling in Z as well. + */ +union tex_tile_address { + struct { + unsigned x:6; /* 4096 / TEX_TILE_SIZE */ + unsigned y:6; /* 4096 / TEX_TILE_SIZE */ + unsigned z:12; /* 4096 -- z not tiled */ + unsigned face:3; + unsigned level:4; + unsigned invalid:1; + } bits; + unsigned value; +}; + + +struct llvmpipe_cached_tex_tile +{ + union tex_tile_address addr; + uint8_t color[TEX_TILE_SIZE][TEX_TILE_SIZE][4]; +}; + +#define NUM_ENTRIES 50 + + +/** XXX move these */ +#define MAX_TEX_WIDTH 2048 +#define MAX_TEX_HEIGHT 2048 + + +struct llvmpipe_tex_tile_cache +{ + struct pipe_screen *screen; + struct pipe_surface *surface; /**< the surface we're caching */ + struct pipe_transfer *transfer; + void *transfer_map; + + struct pipe_texture *texture; /**< if caching a texture */ + unsigned timestamp; + + struct llvmpipe_cached_tex_tile entries[NUM_ENTRIES]; + + struct pipe_transfer *tex_trans; + void *tex_trans_map; + int tex_face, tex_level, tex_z; + + struct llvmpipe_cached_tex_tile *last_tile; /**< most recently retrieved tile */ +}; + + +extern struct llvmpipe_tex_tile_cache * +lp_create_tex_tile_cache( struct pipe_screen *screen ); + +extern void +lp_destroy_tex_tile_cache(struct llvmpipe_tex_tile_cache *tc); + +extern void +lp_tex_tile_cache_map_transfers(struct llvmpipe_tex_tile_cache *tc); + +extern void +lp_tex_tile_cache_unmap_transfers(struct llvmpipe_tex_tile_cache *tc); + +extern void +lp_tex_tile_cache_set_texture(struct llvmpipe_tex_tile_cache *tc, + struct pipe_texture *texture); + +void +lp_tex_tile_cache_validate_texture(struct llvmpipe_tex_tile_cache *tc); + +extern const struct llvmpipe_cached_tex_tile * +lp_find_cached_tex_tile(struct llvmpipe_tex_tile_cache *tc, + union tex_tile_address addr ); + +static INLINE const union tex_tile_address +tex_tile_address( unsigned x, + unsigned y, + unsigned z, + unsigned face, + unsigned level ) +{ + union tex_tile_address addr; + + addr.value = 0; + addr.bits.x = x / TEX_TILE_SIZE; + addr.bits.y = y / TEX_TILE_SIZE; + addr.bits.z = z; + addr.bits.face = face; + addr.bits.level = level; + + return addr; +} + +/* Quickly retrieve tile if it matches last lookup. + */ +static INLINE const struct llvmpipe_cached_tex_tile * +lp_get_cached_tex_tile(struct llvmpipe_tex_tile_cache *tc, + union tex_tile_address addr ) +{ + if (tc->last_tile->addr.value == addr.value) + return tc->last_tile; + + return lp_find_cached_tex_tile( tc, addr ); +} + + +#endif /* LP_TEX_CACHE_H */ + diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample.c b/src/gallium/drivers/llvmpipe/lp_tex_sample.c new file mode 100644 index 00000000000..94eb6dad5af --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_tex_sample.c @@ -0,0 +1,1580 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * Copyright 2008 VMware, Inc. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Texture sampling + * + * Authors: + * Brian Paul + */ + +#include "lp_context.h" +#include "lp_quad.h" +#include "lp_surface.h" +#include "lp_texture.h" +#include "lp_tex_sample.h" +#include "lp_tex_cache.h" +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_shader_tokens.h" +#include "util/u_math.h" +#include "util/u_memory.h" + + + +/* + * Note, the FRAC macro has to work perfectly. Otherwise you'll sometimes + * see 1-pixel bands of improperly weighted linear-filtered textures. + * The tests/texwrap.c demo is a good test. + * Also note, FRAC(x) doesn't truly return the fractional part of x for x < 0. + * Instead, if x < 0 then FRAC(x) = 1 - true_frac(x). + */ +#define FRAC(f) ((f) - util_ifloor(f)) + + +/** + * Linear interpolation macro + */ +static INLINE float +lerp(float a, float v0, float v1) +{ + return v0 + a * (v1 - v0); +} + + +/** + * Do 2D/biliner interpolation of float values. + * v00, v10, v01 and v11 are typically four texture samples in a square/box. + * a and b are the horizontal and vertical interpolants. + * It's important that this function is inlined when compiled with + * optimization! If we find that's not true on some systems, convert + * to a macro. + */ +static INLINE float +lerp_2d(float a, float b, + float v00, float v10, float v01, float v11) +{ + const float temp0 = lerp(a, v00, v10); + const float temp1 = lerp(a, v01, v11); + return lerp(b, temp0, temp1); +} + + +/** + * As above, but 3D interpolation of 8 values. + */ +static INLINE float +lerp_3d(float a, float b, float c, + float v000, float v100, float v010, float v110, + float v001, float v101, float v011, float v111) +{ + const float temp0 = lerp_2d(a, b, v000, v100, v010, v110); + const float temp1 = lerp_2d(a, b, v001, v101, v011, v111); + return lerp(c, temp0, temp1); +} + + + +/** + * If A is a signed integer, A % B doesn't give the right value for A < 0 + * (in terms of texture repeat). Just casting to unsigned fixes that. + */ +#define REMAINDER(A, B) ((unsigned) (A) % (unsigned) (B)) + + +/** + * Apply texture coord wrapping mode and return integer texture indexes + * for a vector of four texcoords (S or T or P). + * \param wrapMode PIPE_TEX_WRAP_x + * \param s the incoming texcoords + * \param size the texture image size + * \param icoord returns the integer texcoords + * \return integer texture index + */ +static INLINE void +nearest_texcoord_4(unsigned wrapMode, const float s[4], unsigned size, + int icoord[4]) +{ + uint ch; + switch (wrapMode) { + case PIPE_TEX_WRAP_REPEAT: + /* s limited to [0,1) */ + /* i limited to [0,size-1] */ + for (ch = 0; ch < 4; ch++) { + int i = util_ifloor(s[ch] * size); + icoord[ch] = REMAINDER(i, size); + } + return; + case PIPE_TEX_WRAP_CLAMP: + /* s limited to [0,1] */ + /* i limited to [0,size-1] */ + for (ch = 0; ch < 4; ch++) { + if (s[ch] <= 0.0F) + icoord[ch] = 0; + else if (s[ch] >= 1.0F) + icoord[ch] = size - 1; + else + icoord[ch] = util_ifloor(s[ch] * size); + } + return; + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: + { + /* s limited to [min,max] */ + /* i limited to [0, size-1] */ + const float min = 1.0F / (2.0F * size); + const float max = 1.0F - min; + for (ch = 0; ch < 4; ch++) { + if (s[ch] < min) + icoord[ch] = 0; + else if (s[ch] > max) + icoord[ch] = size - 1; + else + icoord[ch] = util_ifloor(s[ch] * size); + } + } + return; + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: + { + /* s limited to [min,max] */ + /* i limited to [-1, size] */ + const float min = -1.0F / (2.0F * size); + const float max = 1.0F - min; + for (ch = 0; ch < 4; ch++) { + if (s[ch] <= min) + icoord[ch] = -1; + else if (s[ch] >= max) + icoord[ch] = size; + else + icoord[ch] = util_ifloor(s[ch] * size); + } + } + return; + case PIPE_TEX_WRAP_MIRROR_REPEAT: + { + const float min = 1.0F / (2.0F * size); + const float max = 1.0F - min; + for (ch = 0; ch < 4; ch++) { + const int flr = util_ifloor(s[ch]); + float u; + if (flr & 1) + u = 1.0F - (s[ch] - (float) flr); + else + u = s[ch] - (float) flr; + if (u < min) + icoord[ch] = 0; + else if (u > max) + icoord[ch] = size - 1; + else + icoord[ch] = util_ifloor(u * size); + } + } + return; + case PIPE_TEX_WRAP_MIRROR_CLAMP: + for (ch = 0; ch < 4; ch++) { + /* s limited to [0,1] */ + /* i limited to [0,size-1] */ + const float u = fabsf(s[ch]); + if (u <= 0.0F) + icoord[ch] = 0; + else if (u >= 1.0F) + icoord[ch] = size - 1; + else + icoord[ch] = util_ifloor(u * size); + } + return; + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: + { + /* s limited to [min,max] */ + /* i limited to [0, size-1] */ + const float min = 1.0F / (2.0F * size); + const float max = 1.0F - min; + for (ch = 0; ch < 4; ch++) { + const float u = fabsf(s[ch]); + if (u < min) + icoord[ch] = 0; + else if (u > max) + icoord[ch] = size - 1; + else + icoord[ch] = util_ifloor(u * size); + } + } + return; + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: + { + /* s limited to [min,max] */ + /* i limited to [0, size-1] */ + const float min = -1.0F / (2.0F * size); + const float max = 1.0F - min; + for (ch = 0; ch < 4; ch++) { + const float u = fabsf(s[ch]); + if (u < min) + icoord[ch] = -1; + else if (u > max) + icoord[ch] = size; + else + icoord[ch] = util_ifloor(u * size); + } + } + return; + default: + assert(0); + } +} + + +/** + * Used to compute texel locations for linear sampling for four texcoords. + * \param wrapMode PIPE_TEX_WRAP_x + * \param s the texcoords + * \param size the texture image size + * \param icoord0 returns first texture indexes + * \param icoord1 returns second texture indexes (usually icoord0 + 1) + * \param w returns blend factor/weight between texture indexes + * \param icoord returns the computed integer texture coords + */ +static INLINE void +linear_texcoord_4(unsigned wrapMode, const float s[4], unsigned size, + int icoord0[4], int icoord1[4], float w[4]) +{ + uint ch; + + switch (wrapMode) { + case PIPE_TEX_WRAP_REPEAT: + for (ch = 0; ch < 4; ch++) { + float u = s[ch] * size - 0.5F; + icoord0[ch] = REMAINDER(util_ifloor(u), size); + icoord1[ch] = REMAINDER(icoord0[ch] + 1, size); + w[ch] = FRAC(u); + } + break;; + case PIPE_TEX_WRAP_CLAMP: + for (ch = 0; ch < 4; ch++) { + float u = CLAMP(s[ch], 0.0F, 1.0F); + u = u * size - 0.5f; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + w[ch] = FRAC(u); + } + break;; + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: + for (ch = 0; ch < 4; ch++) { + float u = CLAMP(s[ch], 0.0F, 1.0F); + u = u * size - 0.5f; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + if (icoord0[ch] < 0) + icoord0[ch] = 0; + if (icoord1[ch] >= (int) size) + icoord1[ch] = size - 1; + w[ch] = FRAC(u); + } + break;; + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: + { + const float min = -1.0F / (2.0F * size); + const float max = 1.0F - min; + for (ch = 0; ch < 4; ch++) { + float u = CLAMP(s[ch], min, max); + u = u * size - 0.5f; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + w[ch] = FRAC(u); + } + } + break;; + case PIPE_TEX_WRAP_MIRROR_REPEAT: + for (ch = 0; ch < 4; ch++) { + const int flr = util_ifloor(s[ch]); + float u; + if (flr & 1) + u = 1.0F - (s[ch] - (float) flr); + else + u = s[ch] - (float) flr; + u = u * size - 0.5F; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + if (icoord0[ch] < 0) + icoord0[ch] = 0; + if (icoord1[ch] >= (int) size) + icoord1[ch] = size - 1; + w[ch] = FRAC(u); + } + break;; + case PIPE_TEX_WRAP_MIRROR_CLAMP: + for (ch = 0; ch < 4; ch++) { + float u = fabsf(s[ch]); + if (u >= 1.0F) + u = (float) size; + else + u *= size; + u -= 0.5F; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + w[ch] = FRAC(u); + } + break;; + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: + for (ch = 0; ch < 4; ch++) { + float u = fabsf(s[ch]); + if (u >= 1.0F) + u = (float) size; + else + u *= size; + u -= 0.5F; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + if (icoord0[ch] < 0) + icoord0[ch] = 0; + if (icoord1[ch] >= (int) size) + icoord1[ch] = size - 1; + w[ch] = FRAC(u); + } + break;; + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: + { + const float min = -1.0F / (2.0F * size); + const float max = 1.0F - min; + for (ch = 0; ch < 4; ch++) { + float u = fabsf(s[ch]); + if (u <= min) + u = min * size; + else if (u >= max) + u = max * size; + else + u *= size; + u -= 0.5F; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + w[ch] = FRAC(u); + } + } + break;; + default: + assert(0); + } +} + + +/** + * For RECT textures / unnormalized texcoords + * Only a subset of wrap modes supported. + */ +static INLINE void +nearest_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size, + int icoord[4]) +{ + uint ch; + switch (wrapMode) { + case PIPE_TEX_WRAP_CLAMP: + for (ch = 0; ch < 4; ch++) { + int i = util_ifloor(s[ch]); + icoord[ch]= CLAMP(i, 0, (int) size-1); + } + return; + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: + /* fall-through */ + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: + for (ch = 0; ch < 4; ch++) { + icoord[ch]= util_ifloor( CLAMP(s[ch], 0.5F, (float) size - 0.5F) ); + } + return; + default: + assert(0); + } +} + + +/** + * For RECT textures / unnormalized texcoords. + * Only a subset of wrap modes supported. + */ +static INLINE void +linear_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size, + int icoord0[4], int icoord1[4], float w[4]) +{ + uint ch; + switch (wrapMode) { + case PIPE_TEX_WRAP_CLAMP: + for (ch = 0; ch < 4; ch++) { + /* Not exactly what the spec says, but it matches NVIDIA output */ + float u = CLAMP(s[ch] - 0.5F, 0.0f, (float) size - 1.0f); + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + w[ch] = FRAC(u); + } + return; + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: + /* fall-through */ + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: + for (ch = 0; ch < 4; ch++) { + float u = CLAMP(s[ch], 0.5F, (float) size - 0.5F); + u -= 0.5F; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + if (icoord1[ch] > (int) size - 1) + icoord1[ch] = size - 1; + w[ch] = FRAC(u); + } + break; + default: + assert(0); + } +} + + +static unsigned +choose_cube_face(float rx, float ry, float rz, float *newS, float *newT) +{ + /* + major axis + direction target sc tc ma + ---------- ------------------------------- --- --- --- + +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx + -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx + +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry + -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry + +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz + -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz + */ + const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz); + unsigned face; + float sc, tc, ma; + + if (arx > ary && arx > arz) { + if (rx >= 0.0F) { + face = PIPE_TEX_FACE_POS_X; + sc = -rz; + tc = -ry; + ma = arx; + } + else { + face = PIPE_TEX_FACE_NEG_X; + sc = rz; + tc = -ry; + ma = arx; + } + } + else if (ary > arx && ary > arz) { + if (ry >= 0.0F) { + face = PIPE_TEX_FACE_POS_Y; + sc = rx; + tc = rz; + ma = ary; + } + else { + face = PIPE_TEX_FACE_NEG_Y; + sc = rx; + tc = -rz; + ma = ary; + } + } + else { + if (rz > 0.0F) { + face = PIPE_TEX_FACE_POS_Z; + sc = rx; + tc = -ry; + ma = arz; + } + else { + face = PIPE_TEX_FACE_NEG_Z; + sc = -rx; + tc = -ry; + ma = arz; + } + } + + *newS = ( sc / ma + 1.0F ) * 0.5F; + *newT = ( tc / ma + 1.0F ) * 0.5F; + + return face; +} + + +/** + * Examine the quad's texture coordinates to compute the partial + * derivatives w.r.t X and Y, then compute lambda (level of detail). + * + * This is only done for fragment shaders, not vertex shaders. + */ +static float +compute_lambda(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias) +{ + const struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; + float rho, lambda; + + if (samp->processor == TGSI_PROCESSOR_VERTEX) + return lodbias; + + assert(sampler->normalized_coords); + + assert(s); + { + float dsdx = s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]; + float dsdy = s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]; + dsdx = fabsf(dsdx); + dsdy = fabsf(dsdy); + rho = MAX2(dsdx, dsdy) * texture->width[0]; + } + if (t) { + float dtdx = t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]; + float dtdy = t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]; + float max; + dtdx = fabsf(dtdx); + dtdy = fabsf(dtdy); + max = MAX2(dtdx, dtdy) * texture->height[0]; + rho = MAX2(rho, max); + } + if (p) { + float dpdx = p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]; + float dpdy = p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]; + float max; + dpdx = fabsf(dpdx); + dpdy = fabsf(dpdy); + max = MAX2(dpdx, dpdy) * texture->depth[0]; + rho = MAX2(rho, max); + } + + lambda = util_fast_log2(rho); + lambda += lodbias + sampler->lod_bias; + lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod); + + return lambda; +} + + +/** + * Do several things here: + * 1. Compute lambda from the texcoords, if needed + * 2. Determine if we're minifying or magnifying + * 3. If minifying, choose mipmap levels + * 4. Return image filter to use within mipmap images + * \param level0 Returns first mipmap level to sample from + * \param level1 Returns second mipmap level to sample from + * \param levelBlend Returns blend factor between levels, in [0,1] + * \param imgFilter Returns either the min or mag filter, depending on lambda + */ +static void +choose_mipmap_levels(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + unsigned *level0, unsigned *level1, float *levelBlend, + unsigned *imgFilter) +{ + const struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; + + if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) { + /* no mipmap selection needed */ + *level0 = *level1 = CLAMP((int) sampler->min_lod, + 0, (int) texture->last_level); + + if (sampler->min_img_filter != sampler->mag_img_filter) { + /* non-mipmapped texture, but still need to determine if doing + * minification or magnification. + */ + float lambda = compute_lambda(tgsi_sampler, s, t, p, lodbias); + if (lambda <= 0.0) { + *imgFilter = sampler->mag_img_filter; + } + else { + *imgFilter = sampler->min_img_filter; + } + } + else { + *imgFilter = sampler->mag_img_filter; + } + } + else { + float lambda = compute_lambda(tgsi_sampler, s, t, p, lodbias); + + if (lambda <= 0.0) { /* XXX threshold depends on the filter */ + /* magnifying */ + *imgFilter = sampler->mag_img_filter; + *level0 = *level1 = 0; + } + else { + /* minifying */ + *imgFilter = sampler->min_img_filter; + + /* choose mipmap level(s) and compute the blend factor between them */ + if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NEAREST) { + /* Nearest mipmap level */ + const int lvl = (int) (lambda + 0.5); + *level0 = + *level1 = CLAMP(lvl, 0, (int) texture->last_level); + } + else { + /* Linear interpolation between mipmap levels */ + const int lvl = (int) lambda; + *level0 = CLAMP(lvl, 0, (int) texture->last_level); + *level1 = CLAMP(lvl + 1, 0, (int) texture->last_level); + *levelBlend = FRAC(lambda); /* blending weight between levels */ + } + } + } +} + + +/** + * Get a texel from a texture, using the texture tile cache. + * + * \param face the cube face in 0..5 + * \param level the mipmap level + * \param x the x coord of texel within 2D image + * \param y the y coord of texel within 2D image + * \param z which slice of a 3D texture + * \param rgba the quad to put the texel/color into + * \param j which element of the rgba quad to write to + * + * XXX maybe move this into lp_tile_cache.c and merge with the + * lp_get_cached_tile_tex() function. Also, get 4 texels instead of 1... + */ +static void +get_texel_quad_2d(const struct tgsi_sampler *tgsi_sampler, + unsigned face, unsigned level, int x, int y, + const uint8_t *out[4]) +{ + const struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + + const struct llvmpipe_cached_tex_tile *tile + = lp_get_cached_tex_tile(samp->cache, + tex_tile_address(x, y, 0, face, level)); + + y %= TEX_TILE_SIZE; + x %= TEX_TILE_SIZE; + + out[0] = &tile->color[y ][x ][0]; + out[1] = &tile->color[y ][x+1][0]; + out[2] = &tile->color[y+1][x ][0]; + out[3] = &tile->color[y+1][x+1][0]; +} + +static INLINE const uint8_t * +get_texel_2d_ptr(const struct tgsi_sampler *tgsi_sampler, + unsigned face, unsigned level, int x, int y) +{ + const struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + + const struct llvmpipe_cached_tex_tile *tile + = lp_get_cached_tex_tile(samp->cache, + tex_tile_address(x, y, 0, face, level)); + + y %= TEX_TILE_SIZE; + x %= TEX_TILE_SIZE; + + return &tile->color[y][x][0]; +} + + +static void +get_texel_quad_2d_mt(const struct tgsi_sampler *tgsi_sampler, + unsigned face, unsigned level, + int x0, int y0, + int x1, int y1, + const uint8_t *out[4]) +{ + unsigned i; + + for (i = 0; i < 4; i++) { + unsigned tx = (i & 1) ? x1 : x0; + unsigned ty = (i >> 1) ? y1 : y0; + + out[i] = get_texel_2d_ptr( tgsi_sampler, face, level, tx, ty ); + } +} + +static void +get_texel(const struct tgsi_sampler *tgsi_sampler, + unsigned face, unsigned level, int x, int y, int z, + float rgba[NUM_CHANNELS][QUAD_SIZE], unsigned j) +{ + const struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; + + if (x < 0 || x >= (int) texture->width[level] || + y < 0 || y >= (int) texture->height[level] || + z < 0 || z >= (int) texture->depth[level]) { + rgba[0][j] = sampler->border_color[0]; + rgba[1][j] = sampler->border_color[1]; + rgba[2][j] = sampler->border_color[2]; + rgba[3][j] = sampler->border_color[3]; + } + else { + const unsigned tx = x % TEX_TILE_SIZE; + const unsigned ty = y % TEX_TILE_SIZE; + const struct llvmpipe_cached_tex_tile *tile; + + tile = lp_get_cached_tex_tile(samp->cache, + tex_tile_address(x, y, z, face, level)); + + rgba[0][j] = ubyte_to_float(tile->color[ty][tx][0]); + rgba[1][j] = ubyte_to_float(tile->color[ty][tx][1]); + rgba[2][j] = ubyte_to_float(tile->color[ty][tx][2]); + rgba[3][j] = ubyte_to_float(tile->color[ty][tx][3]); + if (0) + { + debug_printf("Get texel %f %f %f %f from %s\n", + rgba[0][j], rgba[1][j], rgba[2][j], rgba[3][j], + pf_name(texture->format)); + } + } +} + + +/** + * Compare texcoord 'p' (aka R) against texture value 'rgba[0]' + * When we sampled the depth texture, the depth value was put into all + * RGBA channels. We look at the red channel here. + * \param rgba quad of (depth) texel values + * \param p texture 'P' components for four pixels in quad + * \param j which pixel in the quad to test [0..3] + */ +static INLINE void +shadow_compare(const struct pipe_sampler_state *sampler, + float rgba[NUM_CHANNELS][QUAD_SIZE], + const float p[QUAD_SIZE], + uint j) +{ + int k; + switch (sampler->compare_func) { + case PIPE_FUNC_LESS: + k = p[j] < rgba[0][j]; + break; + case PIPE_FUNC_LEQUAL: + k = p[j] <= rgba[0][j]; + break; + case PIPE_FUNC_GREATER: + k = p[j] > rgba[0][j]; + break; + case PIPE_FUNC_GEQUAL: + k = p[j] >= rgba[0][j]; + break; + case PIPE_FUNC_EQUAL: + k = p[j] == rgba[0][j]; + break; + case PIPE_FUNC_NOTEQUAL: + k = p[j] != rgba[0][j]; + break; + case PIPE_FUNC_ALWAYS: + k = 1; + break; + case PIPE_FUNC_NEVER: + k = 0; + break; + default: + k = 0; + assert(0); + break; + } + + /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */ + rgba[0][j] = rgba[1][j] = rgba[2][j] = (float) k; + rgba[3][j] = 1.0F; +} + + +/** + * As above, but do four z/texture comparisons. + */ +static INLINE void +shadow_compare4(const struct pipe_sampler_state *sampler, + float rgba[NUM_CHANNELS][QUAD_SIZE], + const float p[QUAD_SIZE]) +{ + int j, k0, k1, k2, k3; + float val; + + /* compare four texcoords vs. four texture samples */ + switch (sampler->compare_func) { + case PIPE_FUNC_LESS: + k0 = p[0] < rgba[0][0]; + k1 = p[1] < rgba[0][1]; + k2 = p[2] < rgba[0][2]; + k3 = p[3] < rgba[0][3]; + break; + case PIPE_FUNC_LEQUAL: + k0 = p[0] <= rgba[0][0]; + k1 = p[1] <= rgba[0][1]; + k2 = p[2] <= rgba[0][2]; + k3 = p[3] <= rgba[0][3]; + break; + case PIPE_FUNC_GREATER: + k0 = p[0] > rgba[0][0]; + k1 = p[1] > rgba[0][1]; + k2 = p[2] > rgba[0][2]; + k3 = p[3] > rgba[0][3]; + break; + case PIPE_FUNC_GEQUAL: + k0 = p[0] >= rgba[0][0]; + k1 = p[1] >= rgba[0][1]; + k2 = p[2] >= rgba[0][2]; + k3 = p[3] >= rgba[0][3]; + break; + case PIPE_FUNC_EQUAL: + k0 = p[0] == rgba[0][0]; + k1 = p[1] == rgba[0][1]; + k2 = p[2] == rgba[0][2]; + k3 = p[3] == rgba[0][3]; + break; + case PIPE_FUNC_NOTEQUAL: + k0 = p[0] != rgba[0][0]; + k1 = p[1] != rgba[0][1]; + k2 = p[2] != rgba[0][2]; + k3 = p[3] != rgba[0][3]; + break; + case PIPE_FUNC_ALWAYS: + k0 = k1 = k2 = k3 = 1; + break; + case PIPE_FUNC_NEVER: + k0 = k1 = k2 = k3 = 0; + break; + default: + k0 = k1 = k2 = k3 = 0; + assert(0); + break; + } + + /* convert four pass/fail values to an intensity in [0,1] */ + val = 0.25F * (k0 + k1 + k2 + k3); + + /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */ + for (j = 0; j < 4; j++) { + rgba[0][j] = rgba[1][j] = rgba[2][j] = val; + rgba[3][j] = 1.0F; + } +} + + + +static void +lp_get_samples_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + const struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + unsigned j; + unsigned level = samp->level; + unsigned xpot = 1 << (samp->xpot - level); + unsigned ypot = 1 << (samp->ypot - level); + unsigned xmax = (xpot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, xpot) - 1; */ + unsigned ymax = (ypot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, ypot) - 1; */ + + for (j = 0; j < QUAD_SIZE; j++) { + int c; + + float u = s[j] * xpot - 0.5F; + float v = t[j] * ypot - 0.5F; + + int uflr = util_ifloor(u); + int vflr = util_ifloor(v); + + float xw = u - (float)uflr; + float yw = v - (float)vflr; + + int x0 = uflr & (xpot - 1); + int y0 = vflr & (ypot - 1); + + const uint8_t *tx[4]; + + + /* Can we fetch all four at once: + */ + if (x0 < xmax && y0 < ymax) + { + get_texel_quad_2d(tgsi_sampler, 0, level, x0, y0, tx); + } + else + { + unsigned x1 = (x0 + 1) & (xpot - 1); + unsigned y1 = (y0 + 1) & (ypot - 1); + get_texel_quad_2d_mt(tgsi_sampler, 0, level, + x0, y0, x1, y1, tx); + } + + + /* interpolate R, G, B, A */ + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp_2d(xw, yw, + ubyte_to_float(tx[0][c]), ubyte_to_float(tx[1][c]), + ubyte_to_float(tx[2][c]), ubyte_to_float(tx[3][c])); + } + } +} + + +static void +lp_get_samples_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + const struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + unsigned j; + unsigned level = samp->level; + unsigned xpot = 1 << (samp->xpot - level); + unsigned ypot = 1 << (samp->ypot - level); + + for (j = 0; j < QUAD_SIZE; j++) { + int c; + + float u = s[j] * xpot; + float v = t[j] * ypot; + + int uflr = util_ifloor(u); + int vflr = util_ifloor(v); + + int x0 = uflr & (xpot - 1); + int y0 = vflr & (ypot - 1); + + const uint8_t *out = get_texel_2d_ptr(tgsi_sampler, 0, level, x0, y0); + + for (c = 0; c < 4; c++) { + rgba[c][j] = ubyte_to_float(out[c]); + } + } +} + + +static void +lp_get_samples_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + const struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + unsigned j; + unsigned level = samp->level; + unsigned xpot = 1 << (samp->xpot - level); + unsigned ypot = 1 << (samp->ypot - level); + + for (j = 0; j < QUAD_SIZE; j++) { + int c; + + float u = s[j] * xpot; + float v = t[j] * ypot; + + int x0, y0; + const uint8_t *out; + + x0 = util_ifloor(u); + if (x0 < 0) + x0 = 0; + else if (x0 > xpot - 1) + x0 = xpot - 1; + + y0 = util_ifloor(v); + if (y0 < 0) + y0 = 0; + else if (y0 > ypot - 1) + y0 = ypot - 1; + + out = get_texel_2d_ptr(tgsi_sampler, 0, level, x0, y0); + + for (c = 0; c < 4; c++) { + rgba[c][j] = ubyte_to_float(out[c]); + } + } +} + + +static void +lp_get_samples_2d_linear_mip_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + int level0; + float lambda; + + lambda = compute_lambda(tgsi_sampler, s, t, p, lodbias); + level0 = (int)lambda; + + if (lambda < 0.0) { + samp->level = 0; + lp_get_samples_2d_linear_repeat_POT( tgsi_sampler, + s, t, p, 0, rgba ); + } + else if (level0 >= texture->last_level) { + samp->level = texture->last_level; + lp_get_samples_2d_linear_repeat_POT( tgsi_sampler, + s, t, p, 0, rgba ); + } + else { + float levelBlend = lambda - level0; + float rgba0[4][4]; + float rgba1[4][4]; + int c,j; + + samp->level = level0; + lp_get_samples_2d_linear_repeat_POT( tgsi_sampler, + s, t, p, 0, rgba0 ); + + samp->level = level0+1; + lp_get_samples_2d_linear_repeat_POT( tgsi_sampler, + s, t, p, 0, rgba1 ); + + for (j = 0; j < QUAD_SIZE; j++) { + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]); + } + } + } +} + +/** + * Common code for sampling 1D/2D/cube textures. + * Could probably extend for 3D... + */ +static void +lp_get_samples_2d_common(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE], + const unsigned faces[4]) +{ + const struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; + unsigned level0, level1, j, imgFilter; + int width, height; + float levelBlend; + + choose_mipmap_levels(tgsi_sampler, s, t, p, + lodbias, + &level0, &level1, &levelBlend, &imgFilter); + + assert(sampler->normalized_coords); + + width = texture->width[level0]; + height = texture->height[level0]; + + assert(width > 0); + + switch (imgFilter) { + case PIPE_TEX_FILTER_NEAREST: + { + int x[4], y[4]; + nearest_texcoord_4(sampler->wrap_s, s, width, x); + nearest_texcoord_4(sampler->wrap_t, t, height, y); + + for (j = 0; j < QUAD_SIZE; j++) { + get_texel(tgsi_sampler, faces[j], level0, x[j], y[j], 0, rgba, j); + if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { + shadow_compare(sampler, rgba, p, j); + } + + if (level0 != level1) { + /* get texels from second mipmap level and blend */ + float rgba2[4][4]; + unsigned c; + x[j] /= 2; + y[j] /= 2; + get_texel(tgsi_sampler, faces[j], level1, x[j], y[j], 0, + rgba2, j); + if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){ + shadow_compare(sampler, rgba2, p, j); + } + + for (c = 0; c < NUM_CHANNELS; c++) { + rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]); + } + } + } + } + break; + case PIPE_TEX_FILTER_LINEAR: + case PIPE_TEX_FILTER_ANISO: + { + int x0[4], y0[4], x1[4], y1[4]; + float xw[4], yw[4]; /* weights */ + + linear_texcoord_4(sampler->wrap_s, s, width, x0, x1, xw); + linear_texcoord_4(sampler->wrap_t, t, height, y0, y1, yw); + + for (j = 0; j < QUAD_SIZE; j++) { + float tx[4][4]; /* texels */ + int c; + get_texel(tgsi_sampler, faces[j], level0, x0[j], y0[j], 0, tx, 0); + get_texel(tgsi_sampler, faces[j], level0, x1[j], y0[j], 0, tx, 1); + get_texel(tgsi_sampler, faces[j], level0, x0[j], y1[j], 0, tx, 2); + get_texel(tgsi_sampler, faces[j], level0, x1[j], y1[j], 0, tx, 3); + if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { + shadow_compare4(sampler, tx, p); + } + + /* interpolate R, G, B, A */ + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp_2d(xw[j], yw[j], + tx[c][0], tx[c][1], + tx[c][2], tx[c][3]); + } + + if (level0 != level1) { + /* get texels from second mipmap level and blend */ + float rgba2[4][4]; + + /* XXX: This is incorrect -- will often end up with (x0 + * == x1 && y0 == y1), meaning that we fetch the same + * texel four times and linearly interpolate between + * identical values. The correct approach would be to + * call linear_texcoord again for the second level. + */ + x0[j] /= 2; + y0[j] /= 2; + x1[j] /= 2; + y1[j] /= 2; + get_texel(tgsi_sampler, faces[j], level1, x0[j], y0[j], 0, tx, 0); + get_texel(tgsi_sampler, faces[j], level1, x1[j], y0[j], 0, tx, 1); + get_texel(tgsi_sampler, faces[j], level1, x0[j], y1[j], 0, tx, 2); + get_texel(tgsi_sampler, faces[j], level1, x1[j], y1[j], 0, tx, 3); + if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){ + shadow_compare4(sampler, tx, p); + } + + /* interpolate R, G, B, A */ + for (c = 0; c < 4; c++) { + rgba2[c][j] = lerp_2d(xw[j], yw[j], + tx[c][0], tx[c][1], tx[c][2], tx[c][3]); + } + + for (c = 0; c < NUM_CHANNELS; c++) { + rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]); + } + } + } + } + break; + default: + assert(0); + } +} + + +static INLINE void +lp_get_samples_1d(struct tgsi_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + static const unsigned faces[4] = {0, 0, 0, 0}; + static const float tzero[4] = {0, 0, 0, 0}; + lp_get_samples_2d_common(sampler, s, tzero, NULL, + lodbias, rgba, faces); +} + + +static INLINE void +lp_get_samples_2d(struct tgsi_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + static const unsigned faces[4] = {0, 0, 0, 0}; + lp_get_samples_2d_common(sampler, s, t, p, + lodbias, rgba, faces); +} + + +static INLINE void +lp_get_samples_3d(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + const struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; + /* get/map pipe_surfaces corresponding to 3D tex slices */ + unsigned level0, level1, j, imgFilter; + int width, height, depth; + float levelBlend; + const uint face = 0; + + choose_mipmap_levels(tgsi_sampler, s, t, p, + lodbias, + &level0, &level1, &levelBlend, &imgFilter); + + assert(sampler->normalized_coords); + + width = texture->width[level0]; + height = texture->height[level0]; + depth = texture->depth[level0]; + + assert(width > 0); + assert(height > 0); + assert(depth > 0); + + switch (imgFilter) { + case PIPE_TEX_FILTER_NEAREST: + { + int x[4], y[4], z[4]; + nearest_texcoord_4(sampler->wrap_s, s, width, x); + nearest_texcoord_4(sampler->wrap_t, t, height, y); + nearest_texcoord_4(sampler->wrap_r, p, depth, z); + for (j = 0; j < QUAD_SIZE; j++) { + get_texel(tgsi_sampler, face, level0, x[j], y[j], z[j], rgba, j); + if (level0 != level1) { + /* get texels from second mipmap level and blend */ + float rgba2[4][4]; + unsigned c; + x[j] /= 2; + y[j] /= 2; + z[j] /= 2; + get_texel(tgsi_sampler, face, level1, x[j], y[j], z[j], rgba2, j); + for (c = 0; c < NUM_CHANNELS; c++) { + rgba[c][j] = lerp(levelBlend, rgba2[c][j], rgba[c][j]); + } + } + } + } + break; + case PIPE_TEX_FILTER_LINEAR: + case PIPE_TEX_FILTER_ANISO: + { + int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4]; + float xw[4], yw[4], zw[4]; /* interpolation weights */ + linear_texcoord_4(sampler->wrap_s, s, width, x0, x1, xw); + linear_texcoord_4(sampler->wrap_t, t, height, y0, y1, yw); + linear_texcoord_4(sampler->wrap_r, p, depth, z0, z1, zw); + + for (j = 0; j < QUAD_SIZE; j++) { + int c; + float tx0[4][4], tx1[4][4]; + get_texel(tgsi_sampler, face, level0, x0[j], y0[j], z0[j], tx0, 0); + get_texel(tgsi_sampler, face, level0, x1[j], y0[j], z0[j], tx0, 1); + get_texel(tgsi_sampler, face, level0, x0[j], y1[j], z0[j], tx0, 2); + get_texel(tgsi_sampler, face, level0, x1[j], y1[j], z0[j], tx0, 3); + get_texel(tgsi_sampler, face, level0, x0[j], y0[j], z1[j], tx1, 0); + get_texel(tgsi_sampler, face, level0, x1[j], y0[j], z1[j], tx1, 1); + get_texel(tgsi_sampler, face, level0, x0[j], y1[j], z1[j], tx1, 2); + get_texel(tgsi_sampler, face, level0, x1[j], y1[j], z1[j], tx1, 3); + + /* interpolate R, G, B, A */ + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j], + tx0[c][0], tx0[c][1], + tx0[c][2], tx0[c][3], + tx1[c][0], tx1[c][1], + tx1[c][2], tx1[c][3]); + } + + if (level0 != level1) { + /* get texels from second mipmap level and blend */ + float rgba2[4][4]; + x0[j] /= 2; + y0[j] /= 2; + z0[j] /= 2; + x1[j] /= 2; + y1[j] /= 2; + z1[j] /= 2; + get_texel(tgsi_sampler, face, level1, x0[j], y0[j], z0[j], tx0, 0); + get_texel(tgsi_sampler, face, level1, x1[j], y0[j], z0[j], tx0, 1); + get_texel(tgsi_sampler, face, level1, x0[j], y1[j], z0[j], tx0, 2); + get_texel(tgsi_sampler, face, level1, x1[j], y1[j], z0[j], tx0, 3); + get_texel(tgsi_sampler, face, level1, x0[j], y0[j], z1[j], tx1, 0); + get_texel(tgsi_sampler, face, level1, x1[j], y0[j], z1[j], tx1, 1); + get_texel(tgsi_sampler, face, level1, x0[j], y1[j], z1[j], tx1, 2); + get_texel(tgsi_sampler, face, level1, x1[j], y1[j], z1[j], tx1, 3); + + /* interpolate R, G, B, A */ + for (c = 0; c < 4; c++) { + rgba2[c][j] = lerp_3d(xw[j], yw[j], zw[j], + tx0[c][0], tx0[c][1], + tx0[c][2], tx0[c][3], + tx1[c][0], tx1[c][1], + tx1[c][2], tx1[c][3]); + } + + /* blend mipmap levels */ + for (c = 0; c < NUM_CHANNELS; c++) { + rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]); + } + } + } + } + break; + default: + assert(0); + } +} + + +static void +lp_get_samples_cube(struct tgsi_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + unsigned faces[QUAD_SIZE], j; + float ssss[4], tttt[4]; + for (j = 0; j < QUAD_SIZE; j++) { + faces[j] = choose_cube_face(s[j], t[j], p[j], ssss + j, tttt + j); + } + lp_get_samples_2d_common(sampler, ssss, tttt, NULL, + lodbias, rgba, faces); +} + + +static void +lp_get_samples_rect(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + const struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; + const uint face = 0; + unsigned level0, level1, j, imgFilter; + int width, height; + float levelBlend; + + choose_mipmap_levels(tgsi_sampler, s, t, p, + lodbias, + &level0, &level1, &levelBlend, &imgFilter); + + /* texture RECTS cannot be mipmapped */ + assert(level0 == level1); + + width = texture->width[level0]; + height = texture->height[level0]; + + assert(width > 0); + + switch (imgFilter) { + case PIPE_TEX_FILTER_NEAREST: + { + int x[4], y[4]; + nearest_texcoord_unnorm_4(sampler->wrap_s, s, width, x); + nearest_texcoord_unnorm_4(sampler->wrap_t, t, height, y); + for (j = 0; j < QUAD_SIZE; j++) { + get_texel(tgsi_sampler, face, level0, x[j], y[j], 0, rgba, j); + if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { + shadow_compare(sampler, rgba, p, j); + } + } + } + break; + case PIPE_TEX_FILTER_LINEAR: + case PIPE_TEX_FILTER_ANISO: + { + int x0[4], y0[4], x1[4], y1[4]; + float xw[4], yw[4]; /* weights */ + linear_texcoord_unnorm_4(sampler->wrap_s, s, width, x0, x1, xw); + linear_texcoord_unnorm_4(sampler->wrap_t, t, height, y0, y1, yw); + for (j = 0; j < QUAD_SIZE; j++) { + float tx[4][4]; /* texels */ + int c; + get_texel(tgsi_sampler, face, level0, x0[j], y0[j], 0, tx, 0); + get_texel(tgsi_sampler, face, level0, x1[j], y0[j], 0, tx, 1); + get_texel(tgsi_sampler, face, level0, x0[j], y1[j], 0, tx, 2); + get_texel(tgsi_sampler, face, level0, x1[j], y1[j], 0, tx, 3); + if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { + shadow_compare4(sampler, tx, p); + } + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp_2d(xw[j], yw[j], + tx[c][0], tx[c][1], tx[c][2], tx[c][3]); + } + } + } + break; + default: + assert(0); + } +} + + +/** + * Error condition handler + */ +static INLINE void +lp_get_samples_null(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + int i,j; + + for (i = 0; i < 4; i++) + for (j = 0; j < 4; j++) + rgba[i][j] = 1.0; +} + +/** + * Called via tgsi_sampler::get_samples() when using a sampler for the + * first time. Determine the actual sampler function, link it in and + * call it. + */ +void +lp_get_samples(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + struct lp_shader_sampler *samp = lp_shader_sampler(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; + + /* Default to the 'undefined' case: + */ + tgsi_sampler->get_samples = lp_get_samples_null; + + if (!texture) { + assert(0); /* is this legal?? */ + goto out; + } + + if (!sampler->normalized_coords) { + assert (texture->target == PIPE_TEXTURE_2D); + tgsi_sampler->get_samples = lp_get_samples_rect; + goto out; + } + + switch (texture->target) { + case PIPE_TEXTURE_1D: + tgsi_sampler->get_samples = lp_get_samples_1d; + break; + case PIPE_TEXTURE_2D: + tgsi_sampler->get_samples = lp_get_samples_2d; + break; + case PIPE_TEXTURE_3D: + tgsi_sampler->get_samples = lp_get_samples_3d; + break; + case PIPE_TEXTURE_CUBE: + tgsi_sampler->get_samples = lp_get_samples_cube; + break; + default: + assert(0); + break; + } + + /* Do this elsewhere: + */ + samp->xpot = util_unsigned_logbase2( samp->texture->width[0] ); + samp->ypot = util_unsigned_logbase2( samp->texture->height[0] ); + + /* Try to hook in a faster sampler. Ultimately we'll have to + * code-generate these. Luckily most of this looks like it is + * orthogonal state within the sampler. + */ + if (texture->target == PIPE_TEXTURE_2D && + sampler->min_img_filter == sampler->mag_img_filter && + sampler->wrap_s == sampler->wrap_t && + sampler->compare_mode == FALSE && + sampler->normalized_coords) + { + if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) { + samp->level = CLAMP((int) sampler->min_lod, + 0, (int) texture->last_level); + + if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) { + switch (sampler->min_img_filter) { + case PIPE_TEX_FILTER_NEAREST: + tgsi_sampler->get_samples = lp_get_samples_2d_nearest_repeat_POT; + break; + case PIPE_TEX_FILTER_LINEAR: + tgsi_sampler->get_samples = lp_get_samples_2d_linear_repeat_POT; + break; + default: + break; + } + } + else if (sampler->wrap_s == PIPE_TEX_WRAP_CLAMP) { + switch (sampler->min_img_filter) { + case PIPE_TEX_FILTER_NEAREST: + tgsi_sampler->get_samples = lp_get_samples_2d_nearest_clamp_POT; + break; + default: + break; + } + } + } + else if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_LINEAR) { + if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) { + switch (sampler->min_img_filter) { + case PIPE_TEX_FILTER_LINEAR: + tgsi_sampler->get_samples = lp_get_samples_2d_linear_mip_linear_repeat_POT; + break; + default: + break; + } + } + } + } + else if (0) { + _debug_printf("target %d/%d min_mip %d/%d min_img %d/%d wrap %d/%d compare %d/%d norm %d/%d\n", + texture->target, PIPE_TEXTURE_2D, + sampler->min_mip_filter, PIPE_TEX_MIPFILTER_NONE, + sampler->min_img_filter, sampler->mag_img_filter, + sampler->wrap_s, sampler->wrap_t, + sampler->compare_mode, FALSE, + sampler->normalized_coords, TRUE); + } + +out: + tgsi_sampler->get_samples( tgsi_sampler, s, t, p, lodbias, rgba ); +} + diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample.h b/src/gallium/drivers/llvmpipe/lp_tex_sample.h new file mode 100644 index 00000000000..628ec3f1efd --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_tex_sample.h @@ -0,0 +1,78 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef LP_TEX_SAMPLE_H +#define LP_TEX_SAMPLE_H + + +#include "tgsi/tgsi_exec.h" + + +struct llvmpipe_tex_tile_cache; + + +/** + * Subclass of tgsi_sampler + */ +struct lp_shader_sampler +{ + struct tgsi_sampler base; /**< base class */ + + unsigned processor; + + /* For lp_get_samples_2d_linear_POT: + */ + unsigned xpot; + unsigned ypot; + unsigned level; + + const struct pipe_texture *texture; + const struct pipe_sampler_state *sampler; + + struct llvmpipe_tex_tile_cache *cache; +}; + + + +static INLINE struct lp_shader_sampler * +lp_shader_sampler(const struct tgsi_sampler *sampler) +{ + return (struct lp_shader_sampler *) sampler; +} + + + +extern void +lp_get_samples(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]); + + +#endif /* LP_TEX_SAMPLE_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_texture.c b/src/gallium/drivers/llvmpipe/lp_texture.c new file mode 100644 index 00000000000..724d4378336 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_texture.c @@ -0,0 +1,429 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + * Michel Dänzer <michel@tungstengraphics.com> + */ + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_inlines.h" +#include "pipe/internal/p_winsys_screen.h" +#include "util/u_math.h" +#include "util/u_memory.h" + +#include "lp_context.h" +#include "lp_state.h" +#include "lp_texture.h" +#include "lp_tex_cache.h" +#include "lp_screen.h" +#include "lp_winsys.h" + + +/* Simple, maximally packed layout. + */ + + +/* Conventional allocation path for non-display textures: + */ +static boolean +llvmpipe_texture_layout(struct llvmpipe_screen *screen, + struct llvmpipe_texture * lpt) +{ + struct pipe_texture *pt = &lpt->base; + unsigned level; + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + unsigned depth = pt->depth[0]; + + unsigned buffer_size = 0; + + pf_get_block(lpt->base.format, &lpt->base.block); + + for (level = 0; level <= pt->last_level; level++) { + pt->width[level] = width; + pt->height[level] = height; + pt->depth[level] = depth; + pt->nblocksx[level] = pf_get_nblocksx(&pt->block, width); + pt->nblocksy[level] = pf_get_nblocksy(&pt->block, height); + lpt->stride[level] = align(pt->nblocksx[level]*pt->block.size, 16); + + lpt->level_offset[level] = buffer_size; + + buffer_size += (pt->nblocksy[level] * + ((pt->target == PIPE_TEXTURE_CUBE) ? 6 : depth) * + lpt->stride[level]); + + width = minify(width); + height = minify(height); + depth = minify(depth); + } + + lpt->data = align_malloc(buffer_size, 16); + + return lpt->data != NULL; +} + +static boolean +llvmpipe_displaytarget_layout(struct llvmpipe_screen *screen, + struct llvmpipe_texture * lpt) +{ + struct llvmpipe_winsys *winsys = screen->winsys; + + pf_get_block(lpt->base.format, &lpt->base.block); + lpt->base.nblocksx[0] = pf_get_nblocksx(&lpt->base.block, lpt->base.width[0]); + lpt->base.nblocksy[0] = pf_get_nblocksy(&lpt->base.block, lpt->base.height[0]); + + lpt->dt = winsys->displaytarget_create(winsys, + lpt->base.format, + lpt->base.width[0], + lpt->base.height[0], + 16, + &lpt->stride[0] ); + + return lpt->dt != NULL; +} + + + + + +static struct pipe_texture * +llvmpipe_texture_create(struct pipe_screen *_screen, + const struct pipe_texture *templat) +{ + struct llvmpipe_screen *screen = llvmpipe_screen(_screen); + struct llvmpipe_texture *lpt = CALLOC_STRUCT(llvmpipe_texture); + if (!lpt) + return NULL; + + lpt->base = *templat; + pipe_reference_init(&lpt->base.reference, 1); + lpt->base.screen = &screen->base; + + /* XXX: The xlib state tracker is brain-dead and will request + * PIPE_FORMAT_Z16_UNORM no matter how much we tell it we don't support it. + */ + if(lpt->base.format == PIPE_FORMAT_Z16_UNORM) + lpt->base.format = PIPE_FORMAT_Z32_UNORM; + + if (lpt->base.tex_usage & (PIPE_TEXTURE_USAGE_DISPLAY_TARGET | + PIPE_TEXTURE_USAGE_PRIMARY)) { + if (!llvmpipe_displaytarget_layout(screen, lpt)) + goto fail; + } + else { + if (!llvmpipe_texture_layout(screen, lpt)) + goto fail; + } + + return &lpt->base; + + fail: + FREE(lpt); + return NULL; +} + + +static struct pipe_texture * +llvmpipe_texture_blanket(struct pipe_screen * screen, + const struct pipe_texture *base, + const unsigned *stride, + struct pipe_buffer *buffer) +{ + /* FIXME */ +#if 0 + struct llvmpipe_texture *lpt; + assert(screen); + + /* Only supports one type */ + if (base->target != PIPE_TEXTURE_2D || + base->last_level != 0 || + base->depth[0] != 1) { + return NULL; + } + + lpt = CALLOC_STRUCT(llvmpipe_texture); + if (!lpt) + return NULL; + + lpt->base = *base; + pipe_reference_init(&lpt->base.reference, 1); + lpt->base.screen = screen; + lpt->base.nblocksx[0] = pf_get_nblocksx(&lpt->base.block, lpt->base.width[0]); + lpt->base.nblocksy[0] = pf_get_nblocksy(&lpt->base.block, lpt->base.height[0]); + lpt->stride[0] = stride[0]; + + pipe_buffer_reference(&lpt->buffer, buffer); + + return &lpt->base; +#else + return NULL; +#endif +} + + +static void +llvmpipe_texture_destroy(struct pipe_texture *pt) +{ + struct llvmpipe_screen *screen = llvmpipe_screen(pt->screen); + struct llvmpipe_texture *lpt = llvmpipe_texture(pt); + + if(lpt->dt) { + struct llvmpipe_winsys *winsys = screen->winsys; + winsys->displaytarget_destroy(winsys, lpt->dt); + } + else + align_free(lpt->data); + + FREE(lpt); +} + + +static struct pipe_surface * +llvmpipe_get_tex_surface(struct pipe_screen *screen, + struct pipe_texture *pt, + unsigned face, unsigned level, unsigned zslice, + unsigned usage) +{ + struct llvmpipe_texture *lpt = llvmpipe_texture(pt); + struct pipe_surface *ps; + + assert(level <= pt->last_level); + + ps = CALLOC_STRUCT(pipe_surface); + if (ps) { + pipe_reference_init(&ps->reference, 1); + pipe_texture_reference(&ps->texture, pt); + ps->format = pt->format; + ps->width = pt->width[level]; + ps->height = pt->height[level]; + ps->offset = lpt->level_offset[level]; + ps->usage = usage; + + /* Because we are llvmpipe, anything that the state tracker + * thought was going to be done with the GPU will actually get + * done with the CPU. Let's adjust the flags to take that into + * account. + */ + if (ps->usage & PIPE_BUFFER_USAGE_GPU_WRITE) { + /* GPU_WRITE means "render" and that can involve reads (blending) */ + ps->usage |= PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_CPU_READ; + } + + if (ps->usage & PIPE_BUFFER_USAGE_GPU_READ) + ps->usage |= PIPE_BUFFER_USAGE_CPU_READ; + + if (ps->usage & (PIPE_BUFFER_USAGE_CPU_WRITE | + PIPE_BUFFER_USAGE_GPU_WRITE)) { + /* Mark the surface as dirty. The tile cache will look for this. */ + lpt->timestamp++; + llvmpipe_screen(screen)->timestamp++; + } + + ps->face = face; + ps->level = level; + ps->zslice = zslice; + + if (pt->target == PIPE_TEXTURE_CUBE) { + ps->offset += face * pt->nblocksy[level] * lpt->stride[level]; + } + else if (pt->target == PIPE_TEXTURE_3D) { + ps->offset += zslice * pt->nblocksy[level] * lpt->stride[level]; + } + else { + assert(face == 0); + assert(zslice == 0); + } + } + return ps; +} + + +static void +llvmpipe_tex_surface_destroy(struct pipe_surface *surf) +{ + /* Effectively do the texture_update work here - if texture images + * needed post-processing to put them into hardware layout, this is + * where it would happen. For llvmpipe, nothing to do. + */ + assert(surf->texture); + pipe_texture_reference(&surf->texture, NULL); + FREE(surf); +} + + +static struct pipe_transfer * +llvmpipe_get_tex_transfer(struct pipe_screen *screen, + struct pipe_texture *texture, + unsigned face, unsigned level, unsigned zslice, + enum pipe_transfer_usage usage, + unsigned x, unsigned y, unsigned w, unsigned h) +{ + struct llvmpipe_texture *lptex = llvmpipe_texture(texture); + struct llvmpipe_transfer *lpt; + + assert(texture); + assert(level <= texture->last_level); + + lpt = CALLOC_STRUCT(llvmpipe_transfer); + if (lpt) { + struct pipe_transfer *pt = &lpt->base; + pipe_texture_reference(&pt->texture, texture); + pt->format = texture->format; + pt->block = texture->block; + pt->x = x; + pt->y = y; + pt->width = w; + pt->height = h; + pt->nblocksx = texture->nblocksx[level]; + pt->nblocksy = texture->nblocksy[level]; + pt->stride = lptex->stride[level]; + pt->usage = usage; + pt->face = face; + pt->level = level; + pt->zslice = zslice; + + lpt->offset = lptex->level_offset[level]; + + if (texture->target == PIPE_TEXTURE_CUBE) { + lpt->offset += face * pt->nblocksy * pt->stride; + } + else if (texture->target == PIPE_TEXTURE_3D) { + lpt->offset += zslice * pt->nblocksy * pt->stride; + } + else { + assert(face == 0); + assert(zslice == 0); + } + return pt; + } + return NULL; +} + + +static void +llvmpipe_tex_transfer_destroy(struct pipe_transfer *transfer) +{ + /* Effectively do the texture_update work here - if texture images + * needed post-processing to put them into hardware layout, this is + * where it would happen. For llvmpipe, nothing to do. + */ + assert (transfer->texture); + pipe_texture_reference(&transfer->texture, NULL); + FREE(transfer); +} + + +static void * +llvmpipe_transfer_map( struct pipe_screen *_screen, + struct pipe_transfer *transfer ) +{ + struct llvmpipe_screen *screen = llvmpipe_screen(_screen); + ubyte *map, *xfer_map; + struct llvmpipe_texture *lpt; + + assert(transfer->texture); + lpt = llvmpipe_texture(transfer->texture); + + if(lpt->dt) { + struct llvmpipe_winsys *winsys = screen->winsys; + unsigned flags = 0; + + if (transfer->usage != PIPE_TRANSFER_READ) { + flags |= PIPE_BUFFER_USAGE_CPU_WRITE; + } + + if (transfer->usage != PIPE_TRANSFER_WRITE) { + flags |= PIPE_BUFFER_USAGE_CPU_READ; + } + + map = winsys->displaytarget_map(winsys, lpt->dt, flags); + if (map == NULL) + return NULL; + } + else + map = lpt->data; + + /* May want to different things here depending on read/write nature + * of the map: + */ + if (transfer->texture && transfer->usage != PIPE_TRANSFER_READ) + { + /* Do something to notify sharing contexts of a texture change. + * In llvmpipe, that would mean flushing the texture cache. + */ + screen->timestamp++; + } + + xfer_map = map + llvmpipe_transfer(transfer)->offset + + transfer->y / transfer->block.height * transfer->stride + + transfer->x / transfer->block.width * transfer->block.size; + /*printf("map = %p xfer map = %p\n", map, xfer_map);*/ + return xfer_map; +} + + +static void +llvmpipe_transfer_unmap(struct pipe_screen *_screen, + struct pipe_transfer *transfer) +{ + struct llvmpipe_screen *screen = llvmpipe_screen(_screen); + struct llvmpipe_texture *lpt; + + assert(transfer->texture); + lpt = llvmpipe_texture(transfer->texture); + + if(lpt->dt) { + struct llvmpipe_winsys *winsys = screen->winsys; + winsys->displaytarget_unmap(winsys, lpt->dt); + } +} + + +void +llvmpipe_init_texture_funcs(struct llvmpipe_context *lp) +{ +} + + +void +llvmpipe_init_screen_texture_funcs(struct pipe_screen *screen) +{ + screen->texture_create = llvmpipe_texture_create; + screen->texture_blanket = llvmpipe_texture_blanket; + screen->texture_destroy = llvmpipe_texture_destroy; + + screen->get_tex_surface = llvmpipe_get_tex_surface; + screen->tex_surface_destroy = llvmpipe_tex_surface_destroy; + + screen->get_tex_transfer = llvmpipe_get_tex_transfer; + screen->tex_transfer_destroy = llvmpipe_tex_transfer_destroy; + screen->transfer_map = llvmpipe_transfer_map; + screen->transfer_unmap = llvmpipe_transfer_unmap; +} diff --git a/src/gallium/drivers/llvmpipe/lp_texture.h b/src/gallium/drivers/llvmpipe/lp_texture.h new file mode 100644 index 00000000000..00a20763e43 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_texture.h @@ -0,0 +1,90 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef LP_TEXTURE_H +#define LP_TEXTURE_H + + +#include "pipe/p_state.h" + + +struct pipe_context; +struct pipe_screen; +struct llvmpipe_context; +struct llvmpipe_displaytarget; + +struct llvmpipe_texture +{ + struct pipe_texture base; + + unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS]; + unsigned stride[PIPE_MAX_TEXTURE_LEVELS]; + + /** + * Display target, for textures with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET + * usage. + */ + struct llvmpipe_displaytarget *dt; + + /** + * Malloc'ed data for regular textures, or a mapping to dt above. + */ + void *data; + + unsigned timestamp; +}; + +struct llvmpipe_transfer +{ + struct pipe_transfer base; + + unsigned long offset; +}; + + +/** cast wrappers */ +static INLINE struct llvmpipe_texture * +llvmpipe_texture(struct pipe_texture *pt) +{ + return (struct llvmpipe_texture *) pt; +} + +static INLINE struct llvmpipe_transfer * +llvmpipe_transfer(struct pipe_transfer *pt) +{ + return (struct llvmpipe_transfer *) pt; +} + + +extern void +llvmpipe_init_texture_funcs( struct llvmpipe_context *llvmpipe ); + +extern void +llvmpipe_init_screen_texture_funcs(struct pipe_screen *screen); + + +#endif /* LP_TEXTURE */ diff --git a/src/gallium/drivers/llvmpipe/lp_tile_cache.c b/src/gallium/drivers/llvmpipe/lp_tile_cache.c new file mode 100644 index 00000000000..143afec3d35 --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_tile_cache.c @@ -0,0 +1,299 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Texture tile caching. + * + * Author: + * Brian Paul + */ + +#include "pipe/p_inlines.h" +#include "util/u_memory.h" +#include "util/u_math.h" +#include "util/u_tile.h" +#include "util/u_rect.h" +#include "lp_context.h" +#include "lp_surface.h" +#include "lp_texture.h" +#include "lp_tile_soa.h" +#include "lp_tile_cache.h" + + +struct llvmpipe_tile_cache * +lp_create_tile_cache( struct pipe_screen *screen ) +{ + struct llvmpipe_tile_cache *tc; + + tc = CALLOC_STRUCT( llvmpipe_tile_cache ); + if(!tc) + return NULL; + + tc->screen = screen; + + return tc; +} + + +void +lp_destroy_tile_cache(struct llvmpipe_tile_cache *tc) +{ + struct pipe_screen *screen; + unsigned x, y; + + for (y = 0; y < MAX_HEIGHT; y += TILE_SIZE) { + for (x = 0; x < MAX_WIDTH; x += TILE_SIZE) { + struct llvmpipe_cached_tile *tile = &tc->entries[y/TILE_SIZE][x/TILE_SIZE]; + + if(tile->color) + align_free(tile->color); + } + } + + if (tc->transfer) { + screen = tc->transfer->texture->screen; + screen->tex_transfer_destroy(tc->transfer); + } + + FREE( tc ); +} + + +/** + * Specify the surface to cache. + */ +void +lp_tile_cache_set_surface(struct llvmpipe_tile_cache *tc, + struct pipe_surface *ps) +{ + if (tc->transfer) { + struct pipe_screen *screen = tc->transfer->texture->screen; + + if (ps == tc->surface) + return; + + if (tc->transfer_map) { + screen->transfer_unmap(screen, tc->transfer); + tc->transfer_map = NULL; + } + + screen->tex_transfer_destroy(tc->transfer); + tc->transfer = NULL; + } + + tc->surface = ps; + + if (ps) { + struct pipe_screen *screen = ps->texture->screen; + unsigned x, y; + + tc->transfer = screen->get_tex_transfer(screen, ps->texture, ps->face, + ps->level, ps->zslice, + PIPE_TRANSFER_READ_WRITE, + 0, 0, ps->width, ps->height); + + for (y = 0; y < ps->height; y += TILE_SIZE) { + for (x = 0; x < ps->width; x += TILE_SIZE) { + struct llvmpipe_cached_tile *tile = &tc->entries[y/TILE_SIZE][x/TILE_SIZE]; + + tile->status = LP_TILE_STATUS_UNDEFINED; + + if(!tile->color) + tile->color = align_malloc( TILE_SIZE*TILE_SIZE*NUM_CHANNELS, 16 ); + } + } + } +} + + +/** + * Return the transfer being cached. + */ +struct pipe_surface * +lp_tile_cache_get_surface(struct llvmpipe_tile_cache *tc) +{ + return tc->surface; +} + + +void +lp_tile_cache_map_transfers(struct llvmpipe_tile_cache *tc) +{ + if (tc->transfer && !tc->transfer_map) + tc->transfer_map = tc->screen->transfer_map(tc->screen, tc->transfer); +} + + +void +lp_tile_cache_unmap_transfers(struct llvmpipe_tile_cache *tc) +{ + if (tc->transfer_map) { + tc->screen->transfer_unmap(tc->screen, tc->transfer); + tc->transfer_map = NULL; + } +} + + +/** + * Set a tile to a solid color. + */ +static void +clear_tile(struct llvmpipe_cached_tile *tile, + uint8_t clear_color[4]) +{ + if (clear_color[0] == clear_color[1] && + clear_color[1] == clear_color[2] && + clear_color[2] == clear_color[3]) { + memset(tile->color, clear_color[0], TILE_SIZE * TILE_SIZE * 4); + } + else { + uint x, y, chan; + for (y = 0; y < TILE_SIZE; y++) + for (x = 0; x < TILE_SIZE; x++) + for (chan = 0; chan < 4; ++chan) + TILE_PIXEL(tile->color, x, y, chan) = clear_color[chan]; + } +} + + +/** + * Flush the tile cache: write all dirty tiles back to the transfer. + * any tiles "flagged" as cleared will be "really" cleared. + */ +void +lp_flush_tile_cache(struct llvmpipe_tile_cache *tc) +{ + struct pipe_transfer *pt = tc->transfer; + unsigned x, y; + + if(!pt) + return; + + /* push the tile to all positions marked as clear */ + for (y = 0; y < pt->height; y += TILE_SIZE) { + for (x = 0; x < pt->width; x += TILE_SIZE) { + struct llvmpipe_cached_tile *tile = &tc->entries[y/TILE_SIZE][x/TILE_SIZE]; + + switch(tile->status) { + case LP_TILE_STATUS_UNDEFINED: + break; + + case LP_TILE_STATUS_CLEAR: { + /** + * Actually clear the tiles which were flagged as being in a clear state. + */ + + struct pipe_screen *screen = pt->texture->screen; + unsigned tw = TILE_SIZE; + unsigned th = TILE_SIZE; + void *dst; + + if (pipe_clip_tile(x, y, &tw, &th, pt)) + continue; + + dst = screen->transfer_map(screen, pt); + assert(dst); + if(!dst) + continue; + + util_fill_rect(dst, &pt->block, pt->stride, + x, y, tw, th, + tc->clear_val); + + screen->transfer_unmap(screen, pt); + break; + } + + case LP_TILE_STATUS_DEFINED: + lp_put_tile_rgba_soa(pt, x, y, tile->color); + break; + } + } + } +} + + +/** + * Get a tile from the cache. + * \param x, y position of tile, in pixels + */ +void * +lp_get_cached_tile(struct llvmpipe_tile_cache *tc, + unsigned x, unsigned y ) +{ + struct llvmpipe_cached_tile *tile = &tc->entries[y/TILE_SIZE][x/TILE_SIZE]; + struct pipe_transfer *pt = tc->transfer; + + switch(tile->status) { + case LP_TILE_STATUS_CLEAR: + /* don't get tile from framebuffer, just clear it */ + clear_tile(tile, tc->clear_color); + tile->status = LP_TILE_STATUS_DEFINED; + break; + + case LP_TILE_STATUS_UNDEFINED: + /* get new tile data from transfer */ + lp_get_tile_rgba_soa(pt, x, y, tile->color); + tile->status = LP_TILE_STATUS_DEFINED; + break; + + case LP_TILE_STATUS_DEFINED: + /* nothing to do */ + break; + } + + return tile->color; +} + + +/** + * When a whole surface is being cleared to a value we can avoid + * fetching tiles above. + * Save the color and set a 'clearflag' for each tile of the screen. + */ +void +lp_tile_cache_clear(struct llvmpipe_tile_cache *tc, const float *rgba, + uint clearValue) +{ + struct pipe_transfer *pt = tc->transfer; + const unsigned w = pt->width; + const unsigned h = pt->height; + unsigned x, y, chan; + + for(chan = 0; chan < 4; ++chan) + tc->clear_color[chan] = float_to_ubyte(rgba[chan]); + + tc->clear_val = clearValue; + + /* push the tile to all positions marked as clear */ + for (y = 0; y < h; y += TILE_SIZE) { + for (x = 0; x < w; x += TILE_SIZE) { + struct llvmpipe_cached_tile *tile = &tc->entries[y/TILE_SIZE][x/TILE_SIZE]; + tile->status = LP_TILE_STATUS_CLEAR; + } + } +} diff --git a/src/gallium/drivers/llvmpipe/lp_tile_cache.h b/src/gallium/drivers/llvmpipe/lp_tile_cache.h new file mode 100644 index 00000000000..6d8ba5ece7a --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_tile_cache.h @@ -0,0 +1,106 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef LP_TILE_CACHE_H +#define LP_TILE_CACHE_H + + +#include "pipe/p_compiler.h" +#include "lp_tile_soa.h" + + +enum llvmpipe_tile_status +{ + LP_TILE_STATUS_UNDEFINED = 0, + LP_TILE_STATUS_CLEAR = 1, + LP_TILE_STATUS_DEFINED = 2 +}; + + +struct llvmpipe_cached_tile +{ + enum llvmpipe_tile_status status; + + /** color in SOA format */ + uint8_t *color; +}; + + +/** XXX move these */ +#define MAX_WIDTH 2048 +#define MAX_HEIGHT 2048 + + +struct llvmpipe_tile_cache +{ + struct pipe_screen *screen; + struct pipe_surface *surface; /**< the surface we're caching */ + struct pipe_transfer *transfer; + void *transfer_map; + + struct llvmpipe_cached_tile entries[MAX_WIDTH/TILE_SIZE][MAX_HEIGHT/TILE_SIZE]; + + uint8_t clear_color[4]; /**< for color bufs */ + uint clear_val; /**< for z+stencil, or packed color clear value */ + + struct llvmpipe_cached_tile *last_tile; /**< most recently retrieved tile */ +}; + + +extern struct llvmpipe_tile_cache * +lp_create_tile_cache( struct pipe_screen *screen ); + +extern void +lp_destroy_tile_cache(struct llvmpipe_tile_cache *tc); + +extern void +lp_tile_cache_set_surface(struct llvmpipe_tile_cache *tc, + struct pipe_surface *lps); + +extern struct pipe_surface * +lp_tile_cache_get_surface(struct llvmpipe_tile_cache *tc); + +extern void +lp_tile_cache_map_transfers(struct llvmpipe_tile_cache *tc); + +extern void +lp_tile_cache_unmap_transfers(struct llvmpipe_tile_cache *tc); + +extern void +lp_flush_tile_cache(struct llvmpipe_tile_cache *tc); + +extern void +lp_tile_cache_clear(struct llvmpipe_tile_cache *tc, const float *rgba, + uint clearValue); + +extern void * +lp_get_cached_tile(struct llvmpipe_tile_cache *tc, + unsigned x, unsigned y ); + + +#endif /* LP_TILE_CACHE_H */ + diff --git a/src/gallium/drivers/llvmpipe/lp_tile_soa.c b/src/gallium/drivers/llvmpipe/lp_tile_soa.c new file mode 100644 index 00000000000..4e4ccb31ccd --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_tile_soa.c @@ -0,0 +1,931 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * RGBA/float tile get/put functions. + * Usable both by drivers and state trackers. + */ + + +#include "pipe/p_defines.h" +#include "pipe/p_inlines.h" + +#include "util/u_math.h" +#include "util/u_memory.h" +#include "util/u_rect.h" +#include "util/u_tile.h" +#include "lp_tile_cache.h" +#include "lp_tile_soa.h" + + +const unsigned char +tile_offset[TILE_VECTOR_HEIGHT][TILE_VECTOR_WIDTH] = { + { 0, 1, 4, 5, 8, 9, 12, 13}, + { 2, 3, 6, 7, 10, 11, 14, 15} +}; + + + +/*** PIPE_FORMAT_A8R8G8B8_UNORM ***/ + +static void +a8r8g8b8_get_tile_rgba(const unsigned *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + const unsigned pixel = *src++; + TILE_PIXEL(p, j, i, 0) = (pixel >> 16) & 0xff; + TILE_PIXEL(p, j, i, 1) = (pixel >> 8) & 0xff; + TILE_PIXEL(p, j, i, 2) = (pixel >> 0) & 0xff; + TILE_PIXEL(p, j, i, 3) = (pixel >> 24) & 0xff; + } + } +} + + +static void +a8r8g8b8_put_tile_rgba(unsigned *dst, + unsigned w, unsigned h, + const uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + unsigned r, g, b, a; + r = TILE_PIXEL(p, j, i, 0); + g = TILE_PIXEL(p, j, i, 1); + b = TILE_PIXEL(p, j, i, 2); + a = TILE_PIXEL(p, j, i, 3); + *dst++ = (a << 24) | (r << 16) | (g << 8) | b; + } + } +} + + +/*** PIPE_FORMAT_A8R8G8B8_UNORM ***/ + +static void +x8r8g8b8_get_tile_rgba(const unsigned *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + const unsigned pixel = *src++; + TILE_PIXEL(p, j, i, 0) = (pixel >> 16) & 0xff; + TILE_PIXEL(p, j, i, 1) = (pixel >> 8) & 0xff; + TILE_PIXEL(p, j, i, 2) = (pixel >> 0) & 0xff; + TILE_PIXEL(p, j, i, 3) = 0xff; + } + } +} + + +static void +x8r8g8b8_put_tile_rgba(unsigned *dst, + unsigned w, unsigned h, + const uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + unsigned r, g, b; + r = TILE_PIXEL(p, j, i, 0); + g = TILE_PIXEL(p, j, i, 1); + b = TILE_PIXEL(p, j, i, 2); + *dst++ = (0xff << 24) | (r << 16) | (g << 8) | b; + } + } +} + + +/*** PIPE_FORMAT_B8G8R8A8_UNORM ***/ + +static void +b8g8r8a8_get_tile_rgba(const unsigned *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + const unsigned pixel = *src++; + TILE_PIXEL(p, j, i, 0) = (pixel >> 8) & 0xff; + TILE_PIXEL(p, j, i, 1) = (pixel >> 16) & 0xff; + TILE_PIXEL(p, j, i, 2) = (pixel >> 24) & 0xff; + TILE_PIXEL(p, j, i, 3) = (pixel >> 0) & 0xff; + } + } +} + + +static void +b8g8r8a8_put_tile_rgba(unsigned *dst, + unsigned w, unsigned h, + const uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + unsigned r, g, b, a; + r = TILE_PIXEL(p, j, i, 0); + g = TILE_PIXEL(p, j, i, 1); + b = TILE_PIXEL(p, j, i, 2); + a = TILE_PIXEL(p, j, i, 3); + *dst++ = (b << 24) | (g << 16) | (r << 8) | a; + } + } +} + + +/*** PIPE_FORMAT_A1R5G5B5_UNORM ***/ + +static void +a1r5g5b5_get_tile_rgba(const ushort *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + const ushort pixel = *src++; + TILE_PIXEL(p, j, i, 0) = ((pixel >> 10) & 0x1f) * 255 / 31; + TILE_PIXEL(p, j, i, 1) = ((pixel >> 5) & 0x1f) * 255 / 31; + TILE_PIXEL(p, j, i, 2) = ((pixel ) & 0x1f) * 255 / 31; + TILE_PIXEL(p, j, i, 3) = ((pixel >> 15) ) * 255; + } + } +} + + +static void +a1r5g5b5_put_tile_rgba(ushort *dst, + unsigned w, unsigned h, + const uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + unsigned r, g, b, a; + r = TILE_PIXEL(p, j, i, 0); + g = TILE_PIXEL(p, j, i, 1); + b = TILE_PIXEL(p, j, i, 2); + a = TILE_PIXEL(p, j, i, 3); + r = r >> 3; /* 5 bits */ + g = g >> 3; /* 5 bits */ + b = b >> 3; /* 5 bits */ + a = a >> 7; /* 1 bit */ + *dst++ = (a << 15) | (r << 10) | (g << 5) | b; + } + } +} + + +/*** PIPE_FORMAT_A4R4G4B4_UNORM ***/ + +static void +a4r4g4b4_get_tile_rgba(const ushort *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + const ushort pixel = *src++; + TILE_PIXEL(p, j, i, 0) = ((pixel >> 8) & 0xf) * 255 / 15; + TILE_PIXEL(p, j, i, 1) = ((pixel >> 4) & 0xf) * 255 / 15; + TILE_PIXEL(p, j, i, 2) = ((pixel ) & 0xf) * 255 / 15; + TILE_PIXEL(p, j, i, 3) = ((pixel >> 12) ) * 255 / 15; + } + } +} + + +static void +a4r4g4b4_put_tile_rgba(ushort *dst, + unsigned w, unsigned h, + const uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + unsigned r, g, b, a; + r = TILE_PIXEL(p, j, i, 0); + g = TILE_PIXEL(p, j, i, 1); + b = TILE_PIXEL(p, j, i, 2); + a = TILE_PIXEL(p, j, i, 3); + r >>= 4; + g >>= 4; + b >>= 4; + a >>= 4; + *dst++ = (a << 12) | (r << 16) | (g << 4) | b; + } + } +} + + +/*** PIPE_FORMAT_R5G6B5_UNORM ***/ + +static void +r5g6b5_get_tile_rgba(const ushort *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + const ushort pixel = *src++; + TILE_PIXEL(p, j, i, 0) = ((pixel >> 11) & 0x1f) * 255 / 31; + TILE_PIXEL(p, j, i, 1) = ((pixel >> 5) & 0x3f) * 255 / 63; + TILE_PIXEL(p, j, i, 2) = ((pixel ) & 0x1f) * 255 / 31; + TILE_PIXEL(p, j, i, 3) = 255; + } + } +} + + +static void +r5g6b5_put_tile_rgba(ushort *dst, + unsigned w, unsigned h, + const uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + uint r = (uint) TILE_PIXEL(p, j, i, 0) * 31 / 255; + uint g = (uint) TILE_PIXEL(p, j, i, 1) * 63 / 255; + uint b = (uint) TILE_PIXEL(p, j, i, 2) * 31 / 255; + *dst++ = (r << 11) | (g << 5) | (b); + } + } +} + + + +/*** PIPE_FORMAT_Z16_UNORM ***/ + +/** + * Return each Z value as four floats in [0,1]. + */ +static void +z16_get_tile_rgba(const ushort *src, + unsigned w, unsigned h, + uint8_t *p) +{ + const float scale = 1.0f / 65535.0f; + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + TILE_PIXEL(p, j, i, 0) = + TILE_PIXEL(p, j, i, 1) = + TILE_PIXEL(p, j, i, 2) = + TILE_PIXEL(p, j, i, 3) = *src++ * scale; + } + } +} + + + + +/*** PIPE_FORMAT_L8_UNORM ***/ + +static void +l8_get_tile_rgba(const ubyte *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++, src++) { + TILE_PIXEL(p, j, i, 0) = + TILE_PIXEL(p, j, i, 1) = + TILE_PIXEL(p, j, i, 2) = *src; + TILE_PIXEL(p, j, i, 3) = 255; + } + } +} + + +static void +l8_put_tile_rgba(ubyte *dst, + unsigned w, unsigned h, + const uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + unsigned r; + r = TILE_PIXEL(p, j, i, 0); + *dst++ = (ubyte) r; + } + } +} + + + +/*** PIPE_FORMAT_A8_UNORM ***/ + +static void +a8_get_tile_rgba(const ubyte *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++, src++) { + TILE_PIXEL(p, j, i, 0) = + TILE_PIXEL(p, j, i, 1) = + TILE_PIXEL(p, j, i, 2) = 0; + TILE_PIXEL(p, j, i, 3) = *src; + } + } +} + + +static void +a8_put_tile_rgba(ubyte *dst, + unsigned w, unsigned h, + const uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + unsigned a; + a = TILE_PIXEL(p, j, i, 3); + *dst++ = (ubyte) a; + } + } +} + + + +/*** PIPE_FORMAT_R16_SNORM ***/ + +static void +r16_get_tile_rgba(const short *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++, src++) { + TILE_PIXEL(p, j, i, 0) = MAX2(src[0] >> 7, 0); + TILE_PIXEL(p, j, i, 1) = + TILE_PIXEL(p, j, i, 2) = 0; + TILE_PIXEL(p, j, i, 3) = 255; + } + } +} + + +static void +r16_put_tile_rgba(short *dst, + unsigned w, unsigned h, + const uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++, dst++) { + dst[0] = TILE_PIXEL(p, j, i, 0) << 7; + } + } +} + + +/*** PIPE_FORMAT_R16G16B16A16_SNORM ***/ + +static void +r16g16b16a16_get_tile_rgba(const short *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++, src += 4) { + TILE_PIXEL(p, j, i, 0) = src[0] >> 8; + TILE_PIXEL(p, j, i, 1) = src[1] >> 8; + TILE_PIXEL(p, j, i, 2) = src[2] >> 8; + TILE_PIXEL(p, j, i, 3) = src[3] >> 8; + } + } +} + + +static void +r16g16b16a16_put_tile_rgba(short *dst, + unsigned w, unsigned h, + const uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++, dst += 4) { + dst[0] = TILE_PIXEL(p, j, i, 0) << 8; + dst[1] = TILE_PIXEL(p, j, i, 1) << 8; + dst[2] = TILE_PIXEL(p, j, i, 2) << 8; + dst[3] = TILE_PIXEL(p, j, i, 3) << 8; + } + } +} + + + +/*** PIPE_FORMAT_I8_UNORM ***/ + +static void +i8_get_tile_rgba(const ubyte *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++, src++) { + TILE_PIXEL(p, j, i, 0) = + TILE_PIXEL(p, j, i, 1) = + TILE_PIXEL(p, j, i, 2) = + TILE_PIXEL(p, j, i, 3) = *src; + } + } +} + + +static void +i8_put_tile_rgba(ubyte *dst, + unsigned w, unsigned h, + const uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + unsigned r; + r = TILE_PIXEL(p, j, i, 0); + *dst++ = (ubyte) r; + } + } +} + + +/*** PIPE_FORMAT_A8L8_UNORM ***/ + +static void +a8l8_get_tile_rgba(const ushort *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + ushort ra = *src++; + TILE_PIXEL(p, j, i, 0) = + TILE_PIXEL(p, j, i, 1) = + TILE_PIXEL(p, j, i, 2) = ra & 0xff; + TILE_PIXEL(p, j, i, 3) = ra >> 8; + } + } +} + + +static void +a8l8_put_tile_rgba(ushort *dst, + unsigned w, unsigned h, + const uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + unsigned r, a; + r = TILE_PIXEL(p, j, i, 0); + a = TILE_PIXEL(p, j, i, 3); + *dst++ = (a << 8) | r; + } + } +} + + + + +/*** PIPE_FORMAT_Z32_UNORM ***/ + +/** + * Return each Z value as four floats in [0,1]. + */ +static void +z32_get_tile_rgba(const unsigned *src, + unsigned w, unsigned h, + uint8_t *p) +{ + const double scale = 1.0 / (double) 0xffffffff; + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + TILE_PIXEL(p, j, i, 0) = + TILE_PIXEL(p, j, i, 1) = + TILE_PIXEL(p, j, i, 2) = + TILE_PIXEL(p, j, i, 3) = (float) (*src++ * scale); + } + } +} + + +/*** PIPE_FORMAT_S8Z24_UNORM ***/ + +/** + * Return Z component as four float in [0,1]. Stencil part ignored. + */ +static void +s8z24_get_tile_rgba(const unsigned *src, + unsigned w, unsigned h, + uint8_t *p) +{ + const double scale = 1.0 / ((1 << 24) - 1); + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + TILE_PIXEL(p, j, i, 0) = + TILE_PIXEL(p, j, i, 1) = + TILE_PIXEL(p, j, i, 2) = + TILE_PIXEL(p, j, i, 3) = (float) (scale * (*src++ & 0xffffff)); + } + } +} + + +/*** PIPE_FORMAT_Z24S8_UNORM ***/ + +/** + * Return Z component as four float in [0,1]. Stencil part ignored. + */ +static void +z24s8_get_tile_rgba(const unsigned *src, + unsigned w, unsigned h, + uint8_t *p) +{ + const double scale = 1.0 / ((1 << 24) - 1); + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + TILE_PIXEL(p, j, i, 0) = + TILE_PIXEL(p, j, i, 1) = + TILE_PIXEL(p, j, i, 2) = + TILE_PIXEL(p, j, i, 3) = (float) (scale * (*src++ >> 8)); + } + } +} + + +/*** PIPE_FORMAT_Z32_FLOAT ***/ + +/** + * Return each Z value as four floats in [0,1]. + */ +static void +z32f_get_tile_rgba(const float *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + TILE_PIXEL(p, j, i, 0) = + TILE_PIXEL(p, j, i, 1) = + TILE_PIXEL(p, j, i, 2) = + TILE_PIXEL(p, j, i, 3) = *src++; + } + } +} + + +/*** PIPE_FORMAT_YCBCR / PIPE_FORMAT_YCBCR_REV ***/ + +/** + * Convert YCbCr (or YCrCb) to RGBA. + */ +static void +ycbcr_get_tile_rgba(const ushort *src, + unsigned w, unsigned h, + uint8_t *p, + boolean rev) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + /* do two texels at a time */ + for (j = 0; j < (w & ~1); j += 2, src += 2) { + const ushort t0 = src[0]; + const ushort t1 = src[1]; + const ubyte y0 = (t0 >> 8) & 0xff; /* luminance */ + const ubyte y1 = (t1 >> 8) & 0xff; /* luminance */ + ubyte cb, cr; + float r, g, b; + + if (rev) { + cb = t1 & 0xff; /* chroma U */ + cr = t0 & 0xff; /* chroma V */ + } + else { + cb = t0 & 0xff; /* chroma U */ + cr = t1 & 0xff; /* chroma V */ + } + + /* even pixel: y0,cr,cb */ + r = 1.164f * (y0-16) + 1.596f * (cr-128); + g = 1.164f * (y0-16) - 0.813f * (cr-128) - 0.391f * (cb-128); + b = 1.164f * (y0-16) + 2.018f * (cb-128); + TILE_PIXEL(p, j, i, 0) = r; + TILE_PIXEL(p, j, i, 1) = g; + TILE_PIXEL(p, j, i, 2) = b; + TILE_PIXEL(p, j, i, 3) = 255; + + /* odd pixel: use y1,cr,cb */ + r = 1.164f * (y1-16) + 1.596f * (cr-128); + g = 1.164f * (y1-16) - 0.813f * (cr-128) - 0.391f * (cb-128); + b = 1.164f * (y1-16) + 2.018f * (cb-128); + TILE_PIXEL(p, j + 1, i, 0) = r; + TILE_PIXEL(p, j + 1, i, 1) = g; + TILE_PIXEL(p, j + 1, i, 2) = b; + TILE_PIXEL(p, j + 1, i, 3) = 255; + } + /* do the last texel */ + if (w & 1) { + const ushort t0 = src[0]; + const ushort t1 = src[1]; + const ubyte y0 = (t0 >> 8) & 0xff; /* luminance */ + ubyte cb, cr; + float r, g, b; + + if (rev) { + cb = t1 & 0xff; /* chroma U */ + cr = t0 & 0xff; /* chroma V */ + } + else { + cb = t0 & 0xff; /* chroma U */ + cr = t1 & 0xff; /* chroma V */ + } + + /* even pixel: y0,cr,cb */ + r = 1.164f * (y0-16) + 1.596f * (cr-128); + g = 1.164f * (y0-16) - 0.813f * (cr-128) - 0.391f * (cb-128); + b = 1.164f * (y0-16) + 2.018f * (cb-128); + TILE_PIXEL(p, j, i, 0) = r; + TILE_PIXEL(p, j, i, 1) = g; + TILE_PIXEL(p, j, i, 2) = b; + TILE_PIXEL(p, j, i, 3) = 255; + } + } +} + + +static void +fake_get_tile_rgba(const ushort *src, + unsigned w, unsigned h, + uint8_t *p) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + TILE_PIXEL(p, j, i, 0) = + TILE_PIXEL(p, j, i, 1) = + TILE_PIXEL(p, j, i, 2) = + TILE_PIXEL(p, j, i, 3) = (i ^ j) & 1 ? 255 : 0; + } + } +} + + +static void +lp_tile_raw_to_rgba_soa(enum pipe_format format, + void *src, + uint w, uint h, + uint8_t *p) +{ + switch (format) { + case PIPE_FORMAT_A8R8G8B8_UNORM: + a8r8g8b8_get_tile_rgba((unsigned *) src, w, h, p); + break; + case PIPE_FORMAT_X8R8G8B8_UNORM: + x8r8g8b8_get_tile_rgba((unsigned *) src, w, h, p); + break; + case PIPE_FORMAT_B8G8R8A8_UNORM: + b8g8r8a8_get_tile_rgba((unsigned *) src, w, h, p); + break; + case PIPE_FORMAT_A1R5G5B5_UNORM: + a1r5g5b5_get_tile_rgba((ushort *) src, w, h, p); + break; + case PIPE_FORMAT_A4R4G4B4_UNORM: + a4r4g4b4_get_tile_rgba((ushort *) src, w, h, p); + break; + case PIPE_FORMAT_R5G6B5_UNORM: + r5g6b5_get_tile_rgba((ushort *) src, w, h, p); + break; + case PIPE_FORMAT_L8_UNORM: + l8_get_tile_rgba((ubyte *) src, w, h, p); + break; + case PIPE_FORMAT_A8_UNORM: + a8_get_tile_rgba((ubyte *) src, w, h, p); + break; + case PIPE_FORMAT_I8_UNORM: + i8_get_tile_rgba((ubyte *) src, w, h, p); + break; + case PIPE_FORMAT_A8L8_UNORM: + a8l8_get_tile_rgba((ushort *) src, w, h, p); + break; + case PIPE_FORMAT_R16_SNORM: + r16_get_tile_rgba((short *) src, w, h, p); + break; + case PIPE_FORMAT_R16G16B16A16_SNORM: + r16g16b16a16_get_tile_rgba((short *) src, w, h, p); + break; + case PIPE_FORMAT_Z16_UNORM: + z16_get_tile_rgba((ushort *) src, w, h, p); + break; + case PIPE_FORMAT_Z32_UNORM: + z32_get_tile_rgba((unsigned *) src, w, h, p); + break; + case PIPE_FORMAT_S8Z24_UNORM: + case PIPE_FORMAT_X8Z24_UNORM: + s8z24_get_tile_rgba((unsigned *) src, w, h, p); + break; + case PIPE_FORMAT_Z24S8_UNORM: + case PIPE_FORMAT_Z24X8_UNORM: + z24s8_get_tile_rgba((unsigned *) src, w, h, p); + break; + case PIPE_FORMAT_Z32_FLOAT: + z32f_get_tile_rgba((float *) src, w, h, p); + break; + case PIPE_FORMAT_YCBCR: + ycbcr_get_tile_rgba((ushort *) src, w, h, p, FALSE); + break; + case PIPE_FORMAT_YCBCR_REV: + ycbcr_get_tile_rgba((ushort *) src, w, h, p, TRUE); + break; + default: + debug_printf("%s: unsupported format %s\n", __FUNCTION__, pf_name(format)); + fake_get_tile_rgba(src, w, h, p); + } +} + + +void +lp_get_tile_rgba_soa(struct pipe_transfer *pt, + uint x, uint y, + uint8_t *p) +{ + uint w = TILE_SIZE, h = TILE_SIZE; + void *packed; + + if (pipe_clip_tile(x, y, &w, &h, pt)) + return; + + packed = MALLOC(pf_get_nblocks(&pt->block, w, h) * pt->block.size); + + if (!packed) + return; + + if(pt->format == PIPE_FORMAT_YCBCR || pt->format == PIPE_FORMAT_YCBCR_REV) + assert((x & 1) == 0); + + pipe_get_tile_raw(pt, x, y, w, h, packed, 0); + + lp_tile_raw_to_rgba_soa(pt->format, packed, w, h, p); + + FREE(packed); +} + + +void +lp_put_tile_rgba_soa(struct pipe_transfer *pt, + uint x, uint y, + const uint8_t *p) +{ + uint w = TILE_SIZE, h = TILE_SIZE; + void *packed; + + if (pipe_clip_tile(x, y, &w, &h, pt)) + return; + + packed = MALLOC(pf_get_nblocks(&pt->block, w, h) * pt->block.size); + + if (!packed) + return; + + switch (pt->format) { + case PIPE_FORMAT_A8R8G8B8_UNORM: + a8r8g8b8_put_tile_rgba((unsigned *) packed, w, h, p); + break; + case PIPE_FORMAT_X8R8G8B8_UNORM: + x8r8g8b8_put_tile_rgba((unsigned *) packed, w, h, p); + break; + case PIPE_FORMAT_B8G8R8A8_UNORM: + b8g8r8a8_put_tile_rgba((unsigned *) packed, w, h, p); + break; + case PIPE_FORMAT_A1R5G5B5_UNORM: + a1r5g5b5_put_tile_rgba((ushort *) packed, w, h, p); + break; + case PIPE_FORMAT_R5G6B5_UNORM: + r5g6b5_put_tile_rgba((ushort *) packed, w, h, p); + break; + case PIPE_FORMAT_R8G8B8A8_UNORM: + assert(0); + break; + case PIPE_FORMAT_A4R4G4B4_UNORM: + a4r4g4b4_put_tile_rgba((ushort *) packed, w, h, p); + break; + case PIPE_FORMAT_L8_UNORM: + l8_put_tile_rgba((ubyte *) packed, w, h, p); + break; + case PIPE_FORMAT_A8_UNORM: + a8_put_tile_rgba((ubyte *) packed, w, h, p); + break; + case PIPE_FORMAT_I8_UNORM: + i8_put_tile_rgba((ubyte *) packed, w, h, p); + break; + case PIPE_FORMAT_A8L8_UNORM: + a8l8_put_tile_rgba((ushort *) packed, w, h, p); + break; + case PIPE_FORMAT_R16_SNORM: + r16_put_tile_rgba((short *) packed, w, h, p); + break; + case PIPE_FORMAT_R16G16B16A16_SNORM: + r16g16b16a16_put_tile_rgba((short *) packed, w, h, p); + break; + case PIPE_FORMAT_Z16_UNORM: + /*z16_put_tile_rgba((ushort *) packed, w, h, p);*/ + break; + case PIPE_FORMAT_Z32_UNORM: + /*z32_put_tile_rgba((unsigned *) packed, w, h, p);*/ + break; + case PIPE_FORMAT_S8Z24_UNORM: + case PIPE_FORMAT_X8Z24_UNORM: + /*s8z24_put_tile_rgba((unsigned *) packed, w, h, p);*/ + break; + case PIPE_FORMAT_Z24S8_UNORM: + case PIPE_FORMAT_Z24X8_UNORM: + /*z24s8_put_tile_rgba((unsigned *) packed, w, h, p);*/ + break; + default: + debug_printf("%s: unsupported format %s\n", __FUNCTION__, pf_name(pt->format)); + } + + pipe_put_tile_raw(pt, x, y, w, h, packed, 0); + + FREE(packed); +} + + diff --git a/src/gallium/drivers/llvmpipe/lp_tile_soa.h b/src/gallium/drivers/llvmpipe/lp_tile_soa.h new file mode 100644 index 00000000000..3d8c703b73d --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_tile_soa.h @@ -0,0 +1,81 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef LP_TILE_SOA_H +#define LP_TILE_SOA_H + +#include "pipe/p_compiler.h" +#include "tgsi/tgsi_exec.h" // for NUM_CHANNELS + + +#ifdef __cplusplus +extern "C" { +#endif + + +struct pipe_transfer; + + +/** + * Cache tile size (width and height). This needs to be a power of two. + */ +#define TILE_SIZE 64 + + +#define TILE_VECTOR_HEIGHT 2 +#define TILE_VECTOR_WIDTH 8 + +extern const unsigned char +tile_offset[TILE_VECTOR_HEIGHT][TILE_VECTOR_WIDTH]; + +#define TILE_C_STRIDE (TILE_VECTOR_HEIGHT*TILE_VECTOR_WIDTH) +#define TILE_X_STRIDE (NUM_CHANNELS*TILE_C_STRIDE) +#define TILE_Y_STRIDE (TILE_VECTOR_HEIGHT*TILE_SIZE*NUM_CHANNELS) + +#define TILE_PIXEL(_p, _x, _y, _c) \ + ((_p)[((_y)/TILE_VECTOR_HEIGHT)*TILE_Y_STRIDE + \ + ((_x)/TILE_VECTOR_WIDTH)*TILE_X_STRIDE + \ + (_c)*TILE_C_STRIDE + \ + tile_offset[(_y) % TILE_VECTOR_HEIGHT][(_x) % TILE_VECTOR_WIDTH]]) + + +void +lp_get_tile_rgba_soa(struct pipe_transfer *pt, + uint x, uint y, + uint8_t *p); + +void +lp_put_tile_rgba_soa(struct pipe_transfer *pt, + uint x, uint y, + const uint8_t *p); + + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/src/gallium/drivers/llvmpipe/lp_winsys.h b/src/gallium/drivers/llvmpipe/lp_winsys.h new file mode 100644 index 00000000000..595481c2cbc --- /dev/null +++ b/src/gallium/drivers/llvmpipe/lp_winsys.h @@ -0,0 +1,128 @@ +/************************************************************************** + * + * Copyright 2007-2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * llvmpipe public interface. + */ + + +#ifndef LP_WINSYS_H +#define LP_WINSYS_H + + +#include "pipe/p_compiler.h" // for boolean +#include "pipe/p_format.h" + + +#ifdef __cplusplus +extern "C" { +#endif + + +struct pipe_screen; +struct pipe_context; + + +/** + * Opaque pointer. + */ +struct llvmpipe_displaytarget; + + +/** + * This is the interface that llvmpipe expects any window system + * hosting it to implement. + * + * llvmpipe is for the most part a self sufficient driver. The only thing it + * does not know is how to display a surface. + */ +struct llvmpipe_winsys +{ + void + (*destroy)( struct llvmpipe_winsys *ws ); + + boolean + (*is_displaytarget_format_supported)( struct llvmpipe_winsys *ws, + enum pipe_format format ); + + /** + * Allocate storage for a render target. + * + * Often surfaces which are meant to be blitted to the front screen (i.e., + * display targets) must be allocated with special characteristics, memory + * pools, or obtained directly from the windowing system. + * + * This callback is invoked by the pipe_screen when creating a texture marked + * with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET flag to get the underlying + * storage. + */ + struct llvmpipe_displaytarget * + (*displaytarget_create)( struct llvmpipe_winsys *ws, + enum pipe_format format, + unsigned width, unsigned height, + unsigned alignment, + unsigned *stride ); + + void * + (*displaytarget_map)( struct llvmpipe_winsys *ws, + struct llvmpipe_displaytarget *dt, + unsigned flags ); + + void + (*displaytarget_unmap)( struct llvmpipe_winsys *ws, + struct llvmpipe_displaytarget *dt ); + + /** + * @sa pipe_screen:flush_frontbuffer. + * + * This call will likely become asynchronous eventually. + */ + void + (*displaytarget_display)( struct llvmpipe_winsys *ws, + struct llvmpipe_displaytarget *dt, + void *context_private ); + + void + (*displaytarget_destroy)( struct llvmpipe_winsys *ws, + struct llvmpipe_displaytarget *dt ); +}; + + +struct pipe_context * +llvmpipe_create( struct pipe_screen * ); + + +struct pipe_screen * +llvmpipe_create_screen( struct llvmpipe_winsys * ); + + +#ifdef __cplusplus +} +#endif + +#endif /* LP_WINSYS_H */ diff --git a/src/gallium/drivers/llvmpipe/sp2lp.sh b/src/gallium/drivers/llvmpipe/sp2lp.sh new file mode 100755 index 00000000000..c45a81ce3cf --- /dev/null +++ b/src/gallium/drivers/llvmpipe/sp2lp.sh @@ -0,0 +1,34 @@ +#!/bin/sh +# +# Port changes from softpipe to llvmpipe. Invoke as +# +# sp2lp.sh <commit> +# +# Note that this will only affect llvmpipe -- you still need to actually +# cherry-pick/merge the softpipe changes themselves if they affect directories +# outside src/gallium/drivers/softpipe + +git format-patch \ + --keep-subject \ + --relative=src/gallium/drivers/softpipe \ + --src-prefix=a/src/gallium/drivers/llvmpipe/ \ + --dst-prefix=b/src/gallium/drivers/llvmpipe/ \ + --stdout "$1^1..$1" \ +| sed \ + -e 's/\<softpipe\>/llvmpipe/g' \ + -e 's/\<sp\>/lp/g' \ + -e 's/\<softpipe_/llvmpipe_/g' \ + -e 's/\<sp_/lp_/g' \ + -e 's/\<SP_/LP_/g' \ + -e 's/\<SOFTPIPE_/LLVMPIPE_/g' \ + -e 's/\<spt\>/lpt/g' \ + -e 's/\<sps\>/lps/g' \ + -e 's/\<spfs\>/lpfs/g' \ + -e 's/\<sptex\>/lptex/g' \ + -e 's/\<setup_\(point\|line\|tri\)\>/llvmpipe_\0/g' \ + -e 's/\<llvmpipe_cached_tile\>/llvmpipe_cached_tex_tile/g' \ + -e 's/_get_cached_tile_tex\>/_get_cached_tex_tile/g' \ + -e 's/\<TILE_SIZE\>/TEX_TILE_SIZE/g' \ + -e 's/\<tile_address\>/tex_tile_address/g' \ + -e 's/\<tile->data\.color\>/tile->color/g' \ +| patch -p1 diff --git a/src/gallium/drivers/nouveau/Makefile b/src/gallium/drivers/nouveau/Makefile new file mode 100644 index 00000000000..dbe8a6e7bf5 --- /dev/null +++ b/src/gallium/drivers/nouveau/Makefile @@ -0,0 +1,8 @@ +TOP = ../../../.. +include $(TOP)/configs/current + +LIBNAME = nouveau + +C_SOURCES = nouveau_screen.c + +include ../../Makefile.template diff --git a/src/gallium/drivers/nouveau/nouveau_push.h b/src/gallium/drivers/nouveau/nouveau_push.h index 54ef1c1291e..9c235080a55 100644 --- a/src/gallium/drivers/nouveau/nouveau_push.h +++ b/src/gallium/drivers/nouveau/nouveau_push.h @@ -9,13 +9,13 @@ #define OUT_RING(data) do { \ NOUVEAU_PUSH_CONTEXT(pc); \ - (*pc->nvws->channel->pushbuf->cur++) = (data); \ + (*pc->base.channel->pushbuf->cur++) = (data); \ } while(0) #define OUT_RINGp(src,size) do { \ NOUVEAU_PUSH_CONTEXT(pc); \ - memcpy(pc->nvws->channel->pushbuf->cur, (src), (size) * 4); \ - pc->nvws->channel->pushbuf->cur += (size); \ + memcpy(pc->base.channel->pushbuf->cur, (src), (size) * 4); \ + pc->base.channel->pushbuf->cur += (size); \ } while(0) #define OUT_RINGf(data) do { \ @@ -26,25 +26,35 @@ #define BEGIN_RING(obj,mthd,size) do { \ NOUVEAU_PUSH_CONTEXT(pc); \ - if (pc->nvws->channel->pushbuf->remaining < ((size) + 1)) \ - pc->nvws->push_flush(pc->nvws, ((size) + 1), NULL); \ + struct nouveau_channel *chan = pc->base.channel; \ + if (chan->pushbuf->remaining < ((size) + 1)) \ + nouveau_pushbuf_flush(chan, ((size) + 1)); \ OUT_RING((pc->obj->subc << 13) | ((size) << 18) | (mthd)); \ - pc->nvws->channel->pushbuf->remaining -= ((size) + 1); \ + chan->pushbuf->remaining -= ((size) + 1); \ } while(0) #define BEGIN_RING_NI(obj,mthd,size) do { \ BEGIN_RING(obj, (mthd) | 0x40000000, (size)); \ } while(0) +static inline void +DO_FIRE_RING(struct nouveau_channel *chan, struct pipe_fence_handle **fence) +{ + nouveau_pushbuf_flush(chan, 0); + if (fence) + *fence = NULL; +} + #define FIRE_RING(fence) do { \ NOUVEAU_PUSH_CONTEXT(pc); \ - pc->nvws->push_flush(pc->nvws, 0, fence); \ + DO_FIRE_RING(pc->base.channel, fence); \ } while(0) #define OUT_RELOC(bo,data,flags,vor,tor) do { \ NOUVEAU_PUSH_CONTEXT(pc); \ - pc->nvws->push_reloc(pc->nvws, pc->nvws->channel->pushbuf->cur++, \ - (bo), (data), (flags), (vor), (tor)); \ + struct nouveau_channel *chan = pc->base.channel; \ + nouveau_pushbuf_emit_reloc(chan, chan->pushbuf->cur++, nouveau_bo(bo), \ + (data), 0, (flags), (vor), (tor)); \ } while(0) /* Raw data + flags depending on FB/TT buffer */ @@ -55,8 +65,8 @@ /* FB/TT object handle */ #define OUT_RELOCo(bo,flags) do { \ OUT_RELOC((bo), 0, (flags) | NOUVEAU_BO_OR, \ - pc->nvws->channel->vram->handle, \ - pc->nvws->channel->gart->handle); \ + pc->base.channel->vram->handle, \ + pc->base.channel->gart->handle); \ } while(0) /* Low 32-bits of offset */ @@ -72,11 +82,12 @@ /* A reloc which'll recombine into a NV_DMA_METHOD packet header */ #define OUT_RELOCm(bo, flags, obj, mthd, size) do { \ NOUVEAU_PUSH_CONTEXT(pc); \ - if (pc->nvws->channel->pushbuf->remaining < ((size) + 1)) \ - pc->nvws->push_flush(pc->nvws->channel, ((size) + 1), NULL); \ + struct nouveau_channel *chan = pc->base.channel; \ + if (chan->pushbuf->remaining < ((size) + 1)) \ + nouveau_pushbuf_flush(chan, ((size) + 1)); \ OUT_RELOCd((bo), (pc->obj->subc << 13) | ((size) << 18) | (mthd), \ (flags), 0, 0); \ - pc->nvws->channel->pushbuf->remaining -= ((size) + 1); \ + chan->pushbuf->remaining -= ((size) + 1); \ } while(0) #endif diff --git a/src/gallium/drivers/nouveau/nouveau_screen.c b/src/gallium/drivers/nouveau/nouveau_screen.c new file mode 100644 index 00000000000..e4cf91c005c --- /dev/null +++ b/src/gallium/drivers/nouveau/nouveau_screen.c @@ -0,0 +1,243 @@ +#include <pipe/p_defines.h> +#include <pipe/p_screen.h> +#include <pipe/p_state.h> + +#include <util/u_memory.h> + +#include <errno.h> + +#include "nouveau/nouveau_bo.h" +#include "nouveau_winsys.h" +#include "nouveau_screen.h" + +static const char * +nouveau_screen_get_name(struct pipe_screen *pscreen) +{ + struct nouveau_device *dev = nouveau_screen(pscreen)->device; + static char buffer[128]; + + snprintf(buffer, sizeof(buffer), "NV%02X", dev->chipset); + return buffer; +} + +static const char * +nouveau_screen_get_vendor(struct pipe_screen *pscreen) +{ + return "nouveau"; +} + +static struct pipe_buffer * +nouveau_screen_bo_skel(struct pipe_screen *pscreen, struct nouveau_bo *bo, + unsigned alignment, unsigned usage, unsigned size) +{ + struct pipe_buffer *pb; + + pb = CALLOC(1, sizeof(struct pipe_buffer)+sizeof(struct nouveau_bo *)); + if (!pb) { + nouveau_bo_ref(NULL, &bo); + return NULL; + } + + pipe_reference_init(&pb->reference, 1); + pb->screen = pscreen; + pb->alignment = alignment; + pb->usage = usage; + pb->size = size; + *(struct nouveau_bo **)(pb + 1) = bo; + return pb; +} + +static struct pipe_buffer * +nouveau_screen_bo_new(struct pipe_screen *pscreen, unsigned alignment, + unsigned usage, unsigned size) +{ + struct nouveau_device *dev = nouveau_screen(pscreen)->device; + struct nouveau_bo *bo = NULL; + uint32_t flags = NOUVEAU_BO_MAP; + int ret; + + if (usage & NOUVEAU_BUFFER_USAGE_TRANSFER) + flags |= NOUVEAU_BO_GART; + else + if (usage & PIPE_BUFFER_USAGE_VERTEX) { + if (pscreen->get_param(pscreen, NOUVEAU_CAP_HW_VTXBUF)) + flags |= NOUVEAU_BO_GART; + } else + if (usage & PIPE_BUFFER_USAGE_INDEX) { + if (pscreen->get_param(pscreen, NOUVEAU_CAP_HW_IDXBUF)) + flags |= NOUVEAU_BO_GART; + } + + if (usage & PIPE_BUFFER_USAGE_PIXEL) { + if (usage & NOUVEAU_BUFFER_USAGE_TEXTURE) + flags |= NOUVEAU_BO_GART; + if (!(usage & PIPE_BUFFER_USAGE_CPU_READ_WRITE)) + flags |= NOUVEAU_BO_VRAM; + + if (dev->chipset == 0x50 || dev->chipset >= 0x80) { + flags |= NOUVEAU_BO_TILED; + if (usage & NOUVEAU_BUFFER_USAGE_ZETA) + flags |= NOUVEAU_BO_ZTILE; + } + } + + ret = nouveau_bo_new(dev, flags, alignment, size, &bo); + if (ret) + return NULL; + + return nouveau_screen_bo_skel(pscreen, bo, alignment, usage, size); +} + +static struct pipe_buffer * +nouveau_screen_bo_user(struct pipe_screen *pscreen, void *ptr, unsigned bytes) +{ + struct nouveau_device *dev = nouveau_screen(pscreen)->device; + struct nouveau_bo *bo = NULL; + int ret; + + ret = nouveau_bo_user(dev, ptr, bytes, &bo); + if (ret) + return NULL; + + return nouveau_screen_bo_skel(pscreen, bo, 0, 0, bytes); +} + +static inline uint32_t +nouveau_screen_map_flags(unsigned pipe) +{ + uint32_t flags = 0; + + if (pipe & PIPE_BUFFER_USAGE_CPU_READ) + flags |= NOUVEAU_BO_RD; + if (pipe & PIPE_BUFFER_USAGE_CPU_WRITE) + flags |= NOUVEAU_BO_WR; + if (pipe & PIPE_BUFFER_USAGE_DISCARD) + flags |= NOUVEAU_BO_INVAL; + if (pipe & PIPE_BUFFER_USAGE_DONTBLOCK) + flags |= NOUVEAU_BO_NOWAIT; + else + if (pipe & 0 /*PIPE_BUFFER_USAGE_UNSYNCHRONIZED*/) + flags |= NOUVEAU_BO_NOSYNC; + + return flags; +} + +static void * +nouveau_screen_bo_map(struct pipe_screen *pscreen, struct pipe_buffer *pb, + unsigned usage) +{ + struct nouveau_bo *bo = nouveau_bo(pb); + int ret; + + ret = nouveau_bo_map(bo, nouveau_screen_map_flags(usage)); + if (ret) { + debug_printf("map failed: %d\n", ret); + return NULL; + } + + return bo->map; +} + +static void * +nouveau_screen_bo_map_range(struct pipe_screen *pscreen, struct pipe_buffer *pb, + unsigned offset, unsigned length, unsigned usage) +{ + struct nouveau_bo *bo = nouveau_bo(pb); + uint32_t flags = nouveau_screen_map_flags(usage); + int ret; + + ret = nouveau_bo_map_range(bo, offset, length, flags); + if (ret) { + if (!(flags & NOUVEAU_BO_NOWAIT) || ret != -EBUSY) + debug_printf("map_range failed: %d\n", ret); + return NULL; + } + + return (char *)bo->map - offset; /* why gallium? why? */ +} + +static void +nouveau_screen_bo_map_flush(struct pipe_screen *pscreen, struct pipe_buffer *pb, + unsigned offset, unsigned length) +{ + struct nouveau_bo *bo = nouveau_bo(pb); + + nouveau_bo_map_flush(bo, offset, length); +} + +static void +nouveau_screen_bo_unmap(struct pipe_screen *pscreen, struct pipe_buffer *pb) +{ + struct nouveau_bo *bo = nouveau_bo(pb); + + nouveau_bo_unmap(bo); +} + +static void +nouveau_screen_bo_del(struct pipe_buffer *pb) +{ + struct nouveau_bo *bo = nouveau_bo(pb); + + nouveau_bo_ref(NULL, &bo); + FREE(pb); +} + +static void +nouveau_screen_fence_ref(struct pipe_screen *pscreen, + struct pipe_fence_handle **ptr, + struct pipe_fence_handle *pfence) +{ + *ptr = pfence; +} + +static int +nouveau_screen_fence_signalled(struct pipe_screen *screen, + struct pipe_fence_handle *pfence, + unsigned flags) +{ + return 0; +} + +static int +nouveau_screen_fence_finish(struct pipe_screen *screen, + struct pipe_fence_handle *pfence, + unsigned flags) +{ + return 0; +} + +int +nouveau_screen_init(struct nouveau_screen *screen, struct nouveau_device *dev) +{ + struct pipe_screen *pscreen = &screen->base; + int ret; + + ret = nouveau_channel_alloc(dev, 0xbeef0201, 0xbeef0202, + &screen->channel); + if (ret) + return ret; + screen->device = dev; + + pscreen->get_name = nouveau_screen_get_name; + pscreen->get_vendor = nouveau_screen_get_vendor; + + pscreen->buffer_create = nouveau_screen_bo_new; + pscreen->user_buffer_create = nouveau_screen_bo_user; + pscreen->buffer_map = nouveau_screen_bo_map; + pscreen->buffer_map_range = nouveau_screen_bo_map_range; + pscreen->buffer_flush_mapped_range = nouveau_screen_bo_map_flush; + pscreen->buffer_unmap = nouveau_screen_bo_unmap; + pscreen->buffer_destroy = nouveau_screen_bo_del; + + pscreen->fence_reference = nouveau_screen_fence_ref; + pscreen->fence_signalled = nouveau_screen_fence_signalled; + pscreen->fence_finish = nouveau_screen_fence_finish; + + return 0; +} + +void +nouveau_screen_fini(struct nouveau_screen *screen) +{ +} + diff --git a/src/gallium/drivers/nouveau/nouveau_screen.h b/src/gallium/drivers/nouveau/nouveau_screen.h new file mode 100644 index 00000000000..ebfc67ad1c1 --- /dev/null +++ b/src/gallium/drivers/nouveau/nouveau_screen.h @@ -0,0 +1,36 @@ +#ifndef __NOUVEAU_SCREEN_H__ +#define __NOUVEAU_SCREEN_H__ + +struct nouveau_screen { + struct pipe_screen base; + struct nouveau_device *device; + struct nouveau_channel *channel; +}; + +static inline struct nouveau_screen * +nouveau_screen(struct pipe_screen *pscreen) +{ + return (struct nouveau_screen *)pscreen; +} + +static inline struct nouveau_bo * +nouveau_bo(struct pipe_buffer *pb) +{ + return pb ? *(struct nouveau_bo **)(pb + 1) : NULL; +} + +int nouveau_screen_init(struct nouveau_screen *, struct nouveau_device *); +void nouveau_screen_fini(struct nouveau_screen *); + +struct nouveau_miptree { + struct pipe_texture base; + struct nouveau_bo *bo; +}; + +static inline struct nouveau_miptree * +nouveau_miptree(struct pipe_texture *pt) +{ + return (struct nouveau_miptree *)pt; +} + +#endif diff --git a/src/gallium/drivers/nouveau/nouveau_stateobj.h b/src/gallium/drivers/nouveau/nouveau_stateobj.h index a54820e8512..b595405357f 100644 --- a/src/gallium/drivers/nouveau/nouveau_stateobj.h +++ b/src/gallium/drivers/nouveau/nouveau_stateobj.h @@ -4,7 +4,7 @@ #include "util/u_debug.h" struct nouveau_stateobj_reloc { - struct pipe_buffer *bo; + struct nouveau_bo *bo; unsigned offset; unsigned packet; @@ -51,7 +51,7 @@ so_ref(struct nouveau_stateobj *ref, struct nouveau_stateobj **pso) if (pipe_reference((struct pipe_reference**)pso, &ref->reference)) { free(so->push); for (i = 0; i < so->cur_reloc; i++) - pipe_buffer_reference(&so->reloc[i].bo, NULL); + nouveau_bo_ref(NULL, &so->reloc[i].bo); free(so->reloc); free(so); } @@ -81,13 +81,13 @@ so_method(struct nouveau_stateobj *so, struct nouveau_grobj *gr, } static INLINE void -so_reloc(struct nouveau_stateobj *so, struct pipe_buffer *bo, +so_reloc(struct nouveau_stateobj *so, struct nouveau_bo *bo, unsigned data, unsigned flags, unsigned vor, unsigned tor) { struct nouveau_stateobj_reloc *r = &so->reloc[so->cur_reloc++]; r->bo = NULL; - pipe_buffer_reference(&r->bo, bo); + nouveau_bo_ref(bo, &r->bo); r->offset = so->cur - so->push; r->packet = so->cur_packet; r->data = data; @@ -107,50 +107,52 @@ so_dump(struct nouveau_stateobj *so) } static INLINE void -so_emit(struct nouveau_winsys *nvws, struct nouveau_stateobj *so) +so_emit(struct nouveau_channel *chan, struct nouveau_stateobj *so) { - struct nouveau_pushbuf *pb = nvws->channel->pushbuf; + struct nouveau_pushbuf *pb = chan->pushbuf; unsigned nr, i; nr = so->cur - so->push; if (pb->remaining < nr) - nvws->push_flush(nvws, nr, NULL); + nouveau_pushbuf_flush(chan, nr); pb->remaining -= nr; memcpy(pb->cur, so->push, nr * 4); for (i = 0; i < so->cur_reloc; i++) { struct nouveau_stateobj_reloc *r = &so->reloc[i]; - nvws->push_reloc(nvws, pb->cur + r->offset, r->bo, - r->data, r->flags, r->vor, r->tor); + nouveau_pushbuf_emit_reloc(chan, pb->cur + r->offset, + r->bo, r->data, 0, r->flags, + r->vor, r->tor); } pb->cur += nr; } static INLINE void -so_emit_reloc_markers(struct nouveau_winsys *nvws, struct nouveau_stateobj *so) +so_emit_reloc_markers(struct nouveau_channel *chan, struct nouveau_stateobj *so) { - struct nouveau_pushbuf *pb = nvws->channel->pushbuf; + struct nouveau_pushbuf *pb = chan->pushbuf; unsigned i; if (!so) return; i = so->cur_reloc << 1; - if (nvws->channel->pushbuf->remaining < i) - nvws->push_flush(nvws, i, NULL); - nvws->channel->pushbuf->remaining -= i; + if (pb->remaining < i) + nouveau_pushbuf_flush(chan, i); + pb->remaining -= i; for (i = 0; i < so->cur_reloc; i++) { struct nouveau_stateobj_reloc *r = &so->reloc[i]; - nvws->push_reloc(nvws, pb->cur++, r->bo, r->packet, - (r->flags & (NOUVEAU_BO_VRAM | - NOUVEAU_BO_GART | - NOUVEAU_BO_RDWR)) | - NOUVEAU_BO_DUMMY, 0, 0); - nvws->push_reloc(nvws, pb->cur++, r->bo, r->data, - r->flags | NOUVEAU_BO_DUMMY, r->vor, r->tor); + nouveau_pushbuf_emit_reloc(chan, pb->cur++, r->bo, r->packet, 0, + (r->flags & (NOUVEAU_BO_VRAM | + NOUVEAU_BO_GART | + NOUVEAU_BO_RDWR)) | + NOUVEAU_BO_DUMMY, 0, 0); + nouveau_pushbuf_emit_reloc(chan, pb->cur++, r->bo, r->data, 0, + r->flags | NOUVEAU_BO_DUMMY, + r->vor, r->tor); } } diff --git a/src/gallium/drivers/nouveau/nouveau_winsys.h b/src/gallium/drivers/nouveau/nouveau_winsys.h index ff7dd1c51c8..42c77e5e778 100644 --- a/src/gallium/drivers/nouveau/nouveau_winsys.h +++ b/src/gallium/drivers/nouveau/nouveau_winsys.h @@ -23,77 +23,38 @@ #define NOUVEAU_BUFFER_USAGE_ZETA (1 << 17) #define NOUVEAU_BUFFER_USAGE_TRANSFER (1 << 18) -struct nouveau_winsys { - struct pipe_winsys *ws; - - struct nouveau_channel *channel; - - int (*res_init)(struct nouveau_resource **heap, unsigned start, - unsigned size); - int (*res_alloc)(struct nouveau_resource *heap, int size, void *priv, - struct nouveau_resource **); - void (*res_free)(struct nouveau_resource **); - - int (*push_reloc)(struct nouveau_winsys *, void *ptr, - struct pipe_buffer *, uint32_t data, - uint32_t flags, uint32_t vor, uint32_t tor); - int (*push_flush)(struct nouveau_winsys *, unsigned size, - struct pipe_fence_handle **fence); - - int (*grobj_alloc)(struct nouveau_winsys *, int grclass, - struct nouveau_grobj **); - void (*grobj_free)(struct nouveau_grobj **); - - int (*notifier_alloc)(struct nouveau_winsys *, int count, - struct nouveau_notifier **); - void (*notifier_free)(struct nouveau_notifier **); - void (*notifier_reset)(struct nouveau_notifier *, int id); - uint32_t (*notifier_status)(struct nouveau_notifier *, int id); - uint32_t (*notifier_retval)(struct nouveau_notifier *, int id); - int (*notifier_wait)(struct nouveau_notifier *, int id, - int status, double timeout); - - int (*surface_copy)(struct nouveau_winsys *, struct pipe_surface *, - unsigned, unsigned, struct pipe_surface *, - unsigned, unsigned, unsigned, unsigned); - int (*surface_fill)(struct nouveau_winsys *, struct pipe_surface *, - unsigned, unsigned, unsigned, unsigned, unsigned); - - struct nouveau_bo *(*get_bo)(struct pipe_buffer *); -}; - extern struct pipe_screen * -nv04_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *); +nv04_screen_create(struct pipe_winsys *ws, struct nouveau_device *); extern struct pipe_context * nv04_create(struct pipe_screen *, unsigned pctx_id); extern struct pipe_screen * -nv10_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *); +nv10_screen_create(struct pipe_winsys *ws, struct nouveau_device *); extern struct pipe_context * nv10_create(struct pipe_screen *, unsigned pctx_id); extern struct pipe_screen * -nv20_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *); +nv20_screen_create(struct pipe_winsys *ws, struct nouveau_device *); extern struct pipe_context * nv20_create(struct pipe_screen *, unsigned pctx_id); extern struct pipe_screen * -nv30_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *); +nv30_screen_create(struct pipe_winsys *ws, struct nouveau_device *); extern struct pipe_context * nv30_create(struct pipe_screen *, unsigned pctx_id); extern struct pipe_screen * -nv40_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *); +nv40_screen_create(struct pipe_winsys *ws, struct nouveau_device *); extern struct pipe_context * nv40_create(struct pipe_screen *, unsigned pctx_id); extern struct pipe_screen * -nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *); +nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *); extern struct pipe_context * nv50_create(struct pipe_screen *, unsigned pctx_id); diff --git a/src/gallium/drivers/nv04/nv04_miptree.c b/src/gallium/drivers/nv04/nv04_miptree.c index 4da833c25e8..93f752faec9 100644 --- a/src/gallium/drivers/nv04/nv04_miptree.c +++ b/src/gallium/drivers/nv04/nv04_miptree.c @@ -31,7 +31,8 @@ nv04_miptree_layout(struct nv04_miptree *nv04mt) for (l = 0; l <= pt->last_level; l++) { - nv04mt->level[l].image_offset = offset; + nv04mt->level[l].image_offset = + CALLOC(nr_faces, sizeof(unsigned)); offset += nv04mt->level[l].pitch * pt->height[l]; } diff --git a/src/gallium/drivers/nv04/nv04_screen.c b/src/gallium/drivers/nv04/nv04_screen.c index f9f6d974264..ff2febb668e 100644 --- a/src/gallium/drivers/nv04/nv04_screen.c +++ b/src/gallium/drivers/nv04/nv04_screen.c @@ -1,27 +1,9 @@ #include "pipe/p_screen.h" #include "pipe/p_inlines.h" -#include "util/u_simple_screen.h" #include "nv04_context.h" #include "nv04_screen.h" -static const char * -nv04_screen_get_name(struct pipe_screen *screen) -{ - struct nv04_screen *nv04screen = nv04_screen(screen); - struct nouveau_device *dev = nv04screen->nvws->channel->device; - static char buffer[128]; - - snprintf(buffer, sizeof(buffer), "NV%02X", dev->chipset); - return buffer; -} - -static const char * -nv04_screen_get_vendor(struct pipe_screen *screen) -{ - return "nouveau"; -} - static int nv04_screen_get_param(struct pipe_screen *screen, int param) { @@ -58,6 +40,10 @@ nv04_screen_get_param(struct pipe_screen *screen, int param) return 0; case PIPE_CAP_TEXTURE_MIRROR_REPEAT: return 1; + case PIPE_CAP_TGSI_CONT_SUPPORTED: + return 0; + case PIPE_CAP_BLEND_EQUATION_SEPARATE: + return 0; case NOUVEAU_CAP_HW_VTXBUF: case NOUVEAU_CAP_HW_IDXBUF: return 0; @@ -97,6 +83,13 @@ nv04_screen_is_format_supported(struct pipe_screen *screen, switch (format) { case PIPE_FORMAT_A8R8G8B8_UNORM: case PIPE_FORMAT_R5G6B5_UNORM: + return TRUE; + default: + break; + } + } else + if (tex_usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL) { + switch (format) { case PIPE_FORMAT_Z16_UNORM: return TRUE; default: @@ -123,10 +116,9 @@ static void nv04_screen_destroy(struct pipe_screen *pscreen) { struct nv04_screen *screen = nv04_screen(pscreen); - struct nouveau_winsys *nvws = screen->nvws; - nvws->notifier_free(&screen->sync); - nvws->grobj_free(&screen->fahrenheit); + nouveau_notifier_free(&screen->sync); + nouveau_grobj_free(&screen->fahrenheit); nv04_surface_2d_takedown(&screen->eng2d); FREE(pscreen); @@ -141,21 +133,38 @@ nv04_surface_buffer(struct pipe_surface *surf) } struct pipe_screen * -nv04_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) +nv04_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) { struct nv04_screen *screen = CALLOC_STRUCT(nv04_screen); + struct nouveau_channel *chan; + struct pipe_screen *pscreen; unsigned fahrenheit_class = 0, sub3d_class = 0; - unsigned chipset = nvws->channel->device->chipset; int ret; if (!screen) return NULL; - screen->nvws = nvws; + pscreen = &screen->base.base; + + ret = nouveau_screen_init(&screen->base, dev); + if (ret) { + nv04_screen_destroy(pscreen); + return NULL; + } + chan = screen->base.channel; + + pscreen->winsys = ws; + pscreen->destroy = nv04_screen_destroy; + pscreen->get_param = nv04_screen_get_param; + pscreen->get_paramf = nv04_screen_get_paramf; + pscreen->is_format_supported = nv04_screen_is_format_supported; - if (chipset>=0x20) { + nv04_screen_init_miptree_functions(pscreen); + nv04_screen_init_transfer_functions(pscreen); + + if (dev->chipset >= 0x20) { fahrenheit_class = 0; sub3d_class = 0; - } else if (chipset>=0x10) { + } else if (dev->chipset >= 0x10) { fahrenheit_class = NV10_DX5_TEXTURED_TRIANGLE; sub3d_class = NV10_CONTEXT_SURFACES_3D; } else { @@ -164,50 +173,40 @@ nv04_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) } if (!fahrenheit_class) { - NOUVEAU_ERR("Unknown nv04 chipset: nv%02x\n", chipset); + NOUVEAU_ERR("Unknown nv04 chipset: nv%02x\n", dev->chipset); return NULL; } - /* 2D engine setup */ - screen->eng2d = nv04_surface_2d_init(nvws); - screen->eng2d->buf = nv04_surface_buffer; - /* 3D object */ - ret = nvws->grobj_alloc(nvws, fahrenheit_class, &screen->fahrenheit); + ret = nouveau_grobj_alloc(chan, 0xbeef0001, fahrenheit_class, + &screen->fahrenheit); if (ret) { NOUVEAU_ERR("Error creating 3D object: %d\n", ret); return NULL; } + BIND_RING(chan, screen->fahrenheit, 7); /* 3D surface object */ - ret = nvws->grobj_alloc(nvws, sub3d_class, &screen->context_surfaces_3d); + ret = nouveau_grobj_alloc(chan, 0xbeef0002, sub3d_class, + &screen->context_surfaces_3d); if (ret) { NOUVEAU_ERR("Error creating 3D surface object: %d\n", ret); return NULL; } + BIND_RING(chan, screen->context_surfaces_3d, 6); + + /* 2D engine setup */ + screen->eng2d = nv04_surface_2d_init(&screen->base); + screen->eng2d->buf = nv04_surface_buffer; /* Notifier for sync purposes */ - ret = nvws->notifier_alloc(nvws, 1, &screen->sync); + ret = nouveau_notifier_alloc(chan, 0xbeef0301, 1, &screen->sync); if (ret) { NOUVEAU_ERR("Error creating notifier object: %d\n", ret); - nv04_screen_destroy(&screen->pipe); + nv04_screen_destroy(pscreen); return NULL; } - screen->pipe.winsys = ws; - screen->pipe.destroy = nv04_screen_destroy; - - screen->pipe.get_name = nv04_screen_get_name; - screen->pipe.get_vendor = nv04_screen_get_vendor; - screen->pipe.get_param = nv04_screen_get_param; - screen->pipe.get_paramf = nv04_screen_get_paramf; - - screen->pipe.is_format_supported = nv04_screen_is_format_supported; - - nv04_screen_init_miptree_functions(&screen->pipe); - nv04_screen_init_transfer_functions(&screen->pipe); - u_simple_screen_init(&screen->pipe); - - return &screen->pipe; + return pscreen; } diff --git a/src/gallium/drivers/nv04/nv04_screen.h b/src/gallium/drivers/nv04/nv04_screen.h index ee6fb6db44a..11466b9442c 100644 --- a/src/gallium/drivers/nv04/nv04_screen.h +++ b/src/gallium/drivers/nv04/nv04_screen.h @@ -1,11 +1,11 @@ #ifndef __NV04_SCREEN_H__ #define __NV04_SCREEN_H__ -#include "pipe/p_screen.h" +#include "nouveau/nouveau_screen.h" #include "nv04_surface_2d.h" struct nv04_screen { - struct pipe_screen pipe; + struct nouveau_screen base; struct nouveau_winsys *nvws; unsigned chipset; diff --git a/src/gallium/drivers/nv04/nv04_state.c b/src/gallium/drivers/nv04/nv04_state.c index 87c635f962a..d356ebd8b36 100644 --- a/src/gallium/drivers/nv04/nv04_state.c +++ b/src/gallium/drivers/nv04/nv04_state.c @@ -2,6 +2,7 @@ #include "pipe/p_state.h" #include "pipe/p_defines.h" #include "pipe/p_shader_tokens.h" +#include "pipe/p_inlines.h" #include "tgsi/tgsi_parse.h" @@ -334,7 +335,7 @@ nv04_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, const struct pipe_constant_buffer *buf ) { struct nv04_context *nv04 = nv04_context(pipe); - struct pipe_winsys *ws = pipe->winsys; + struct pipe_screen *pscreen = pipe->screen; assert(shader < PIPE_SHADER_TYPES); assert(index == 0); @@ -342,12 +343,12 @@ nv04_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, if (buf) { void *mapped; if (buf->buffer && buf->buffer->size && - (mapped = ws->buffer_map(ws, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ))) + (mapped = pipe_buffer_map(pscreen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ))) { memcpy(nv04->constbuf[shader], mapped, buf->buffer->size); nv04->constbuf_nr[shader] = buf->buffer->size / (4 * sizeof(float)); - ws->buffer_unmap(ws, buf->buffer); + pipe_buffer_unmap(pscreen, buf->buffer); } } } diff --git a/src/gallium/drivers/nv04/nv04_state.h b/src/gallium/drivers/nv04/nv04_state.h index 0d51439e3ff..399f750dbe7 100644 --- a/src/gallium/drivers/nv04/nv04_state.h +++ b/src/gallium/drivers/nv04/nv04_state.h @@ -37,7 +37,7 @@ struct nv04_miptree { struct { uint pitch; - uint image_offset; + uint *image_offset; } level[PIPE_MAX_TEXTURE_LEVELS]; }; diff --git a/src/gallium/drivers/nv04/nv04_surface_2d.c b/src/gallium/drivers/nv04/nv04_surface_2d.c index f3a8d7efeed..f88e138c79d 100644 --- a/src/gallium/drivers/nv04/nv04_surface_2d.c +++ b/src/gallium/drivers/nv04/nv04_surface_2d.c @@ -4,6 +4,7 @@ #include "nouveau/nouveau_winsys.h" #include "nouveau/nouveau_util.h" +#include "nouveau/nouveau_screen.h" #include "nv04_surface_2d.h" static INLINE int @@ -14,11 +15,13 @@ nv04_surface_format(enum pipe_format format) return NV04_CONTEXT_SURFACES_2D_FORMAT_Y8; case PIPE_FORMAT_R16_SNORM: case PIPE_FORMAT_R5G6B5_UNORM: + case PIPE_FORMAT_Z16_UNORM: return NV04_CONTEXT_SURFACES_2D_FORMAT_R5G6B5; case PIPE_FORMAT_X8R8G8B8_UNORM: case PIPE_FORMAT_A8R8G8B8_UNORM: return NV04_CONTEXT_SURFACES_2D_FORMAT_A8R8G8B8; case PIPE_FORMAT_Z24S8_UNORM: + case PIPE_FORMAT_Z24X8_UNORM: return NV04_CONTEXT_SURFACES_2D_FORMAT_Y32; default: return -1; @@ -32,9 +35,11 @@ nv04_rect_format(enum pipe_format format) case PIPE_FORMAT_A8_UNORM: return NV04_GDI_RECTANGLE_TEXT_COLOR_FORMAT_A8R8G8B8; case PIPE_FORMAT_R5G6B5_UNORM: + case PIPE_FORMAT_Z16_UNORM: return NV04_GDI_RECTANGLE_TEXT_COLOR_FORMAT_A16R5G6B5; case PIPE_FORMAT_A8R8G8B8_UNORM: case PIPE_FORMAT_Z24S8_UNORM: + case PIPE_FORMAT_Z24X8_UNORM: return NV04_GDI_RECTANGLE_TEXT_COLOR_FORMAT_A8R8G8B8; default: return -1; @@ -96,11 +101,11 @@ nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx, struct pipe_surface *src, int sx, int sy, int w, int h) { - struct nouveau_channel *chan = ctx->nvws->channel; + struct nouveau_channel *chan = ctx->swzsurf->channel; struct nouveau_grobj *swzsurf = ctx->swzsurf; struct nouveau_grobj *sifm = ctx->sifm; - struct nouveau_bo *src_bo = ctx->nvws->get_bo(ctx->buf(src)); - struct nouveau_bo *dst_bo = ctx->nvws->get_bo(ctx->buf(dst)); + struct nouveau_bo *src_bo = nouveau_bo(ctx->buf(src)); + struct nouveau_bo *dst_bo = nouveau_bo(ctx->buf(dst)); const unsigned src_pitch = ((struct nv04_surface *)src)->pitch; const unsigned max_w = 1024; const unsigned max_h = 1024; @@ -109,10 +114,10 @@ nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx, unsigned cx; unsigned cy; - /* POT or GTFO */ - assert(!(w & (w - 1)) && !(h & (h - 1))); +#if 0 /* That's the way she likes it */ assert(src_pitch == ((struct nv04_surface *)dst)->pitch); +#endif BEGIN_RING(chan, swzsurf, NV04_SWIZZLED_SURFACE_DMA_IMAGE, 1); OUT_RELOCo(chan, dst_bo, @@ -132,7 +137,7 @@ nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx, for (cy = 0; cy < h; cy += sub_h) { for (cx = 0; cx < w; cx += sub_w) { BEGIN_RING(chan, swzsurf, NV04_SWIZZLED_SURFACE_OFFSET, 1); - OUT_RELOCl(chan, dst_bo, dst->offset + nv04_swizzle_bits(cx, cy) * + OUT_RELOCl(chan, dst_bo, dst->offset + nv04_swizzle_bits(cx+dx, cy+dy) * dst->texture->block.size, NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_WR); @@ -152,8 +157,8 @@ nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx, OUT_RING (chan, src_pitch | NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_ORIGIN_CENTER | NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_FILTER_POINT_SAMPLE); - OUT_RELOCl(chan, src_bo, src->offset + cy * src_pitch + - cx * src->texture->block.size, NOUVEAU_BO_GART | + OUT_RELOCl(chan, src_bo, src->offset + (cy+sy) * src_pitch + + (cx+sx) * src->texture->block.size, NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_RD); OUT_RING (chan, 0); } @@ -167,10 +172,10 @@ nv04_surface_copy_m2mf(struct nv04_surface_2d *ctx, struct pipe_surface *dst, int dx, int dy, struct pipe_surface *src, int sx, int sy, int w, int h) { - struct nouveau_channel *chan = ctx->nvws->channel; + struct nouveau_channel *chan = ctx->m2mf->channel; struct nouveau_grobj *m2mf = ctx->m2mf; - struct nouveau_bo *src_bo = ctx->nvws->get_bo(ctx->buf(src)); - struct nouveau_bo *dst_bo = ctx->nvws->get_bo(ctx->buf(dst)); + struct nouveau_bo *src_bo = nouveau_bo(ctx->buf(src)); + struct nouveau_bo *dst_bo = nouveau_bo(ctx->buf(dst)); unsigned src_pitch = ((struct nv04_surface *)src)->pitch; unsigned dst_pitch = ((struct nv04_surface *)dst)->pitch; unsigned dst_offset = dst->offset + dy * dst_pitch + @@ -209,15 +214,52 @@ nv04_surface_copy_m2mf(struct nv04_surface_2d *ctx, } static int +nv04_surface_copy_m2mf_swizzle(struct nv04_surface_2d *ctx, + struct pipe_surface *dst, int dx, int dy, + struct pipe_surface *src, int sx, int sy) +{ + struct nouveau_channel *chan = ctx->m2mf->channel; + struct nouveau_grobj *m2mf = ctx->m2mf; + struct nouveau_bo *src_bo = nouveau_bo(ctx->buf(src)); + struct nouveau_bo *dst_bo = nouveau_bo(ctx->buf(dst)); + unsigned src_pitch = ((struct nv04_surface *)src)->pitch; + unsigned dst_pitch = ((struct nv04_surface *)dst)->pitch; + unsigned dst_offset = dst->offset + nv04_swizzle_bits(dx, dy) * + dst->texture->block.size; + unsigned src_offset = src->offset + sy * src_pitch + + sx * src->texture->block.size; + + BEGIN_RING(chan, m2mf, NV04_MEMORY_TO_MEMORY_FORMAT_DMA_BUFFER_IN, 2); + OUT_RELOCo(chan, src_bo, + NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_RD); + OUT_RELOCo(chan, dst_bo, + NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_WR); + + BEGIN_RING(chan, m2mf, NV04_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN, 8); + OUT_RELOCl(chan, src_bo, src_offset, + NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD); + OUT_RELOCl(chan, dst_bo, dst_offset, + NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_WR); + OUT_RING (chan, src_pitch); + OUT_RING (chan, dst_pitch); + OUT_RING (chan, 1 * src->texture->block.size); + OUT_RING (chan, 1); + OUT_RING (chan, 0x0101); + OUT_RING (chan, 0); + + return 0; +} + +static int nv04_surface_copy_blit(struct nv04_surface_2d *ctx, struct pipe_surface *dst, int dx, int dy, struct pipe_surface *src, int sx, int sy, int w, int h) { - struct nouveau_channel *chan = ctx->nvws->channel; + struct nouveau_channel *chan = ctx->surf2d->channel; struct nouveau_grobj *surf2d = ctx->surf2d; struct nouveau_grobj *blit = ctx->blit; - struct nouveau_bo *src_bo = ctx->nvws->get_bo(ctx->buf(src)); - struct nouveau_bo *dst_bo = ctx->nvws->get_bo(ctx->buf(dst)); + struct nouveau_bo *src_bo = nouveau_bo(ctx->buf(src)); + struct nouveau_bo *dst_bo = nouveau_bo(ctx->buf(dst)); unsigned src_pitch = ((struct nv04_surface *)src)->pitch; unsigned dst_pitch = ((struct nv04_surface *)dst)->pitch; int format; @@ -257,8 +299,59 @@ nv04_surface_copy(struct nv04_surface_2d *ctx, struct pipe_surface *dst, assert(src->format == dst->format); /* Setup transfer to swizzle the texture to vram if needed */ - if (src_linear && !dst_linear && w > 1 && h > 1) { - nv04_surface_copy_swizzle(ctx, dst, dx, dy, src, sx, sy, w, h); + if (src_linear && !dst_linear) { + int x,y; + + if ((w>1) && (h>1)) { + int potWidth = 1<<log2i(w); + int potHeight = 1<<log2i(h); + int remainWidth = w-potWidth; + int remainHeight = h-potHeight; + int squareDim = (potWidth>potHeight ? potHeight : potWidth); + + /* top left is always POT, but we can only swizzle squares */ + for (y=0; y<potHeight; y+=squareDim) { + for (x=0; x<potWidth; x+= squareDim) { + nv04_surface_copy_swizzle(ctx, dst, dx+x, dy+y, + src, sx+x, sy+y, + squareDim, squareDim); + } + } + + /* top right */ + if (remainWidth>0) { + nv04_surface_copy(ctx, dst, dx+potWidth, dy, + src, sx+potWidth, sy, + remainWidth, potHeight); + } + + /* bottom left */ + if (remainHeight>0) { + nv04_surface_copy(ctx, dst, dx, dy+potHeight, + src, sx, sy+potHeight, + potWidth, remainHeight); + } + + /* bottom right */ + if ((remainWidth>0) && (remainHeight>0)) { + nv04_surface_copy(ctx, dst, dx+potWidth, dy+potHeight, + src, sx+potWidth, sy+potHeight, + remainWidth, remainHeight); + } + } else if (w==1) { + /* We have a column to copy to a swizzled texture */ + for (y=0; y<h; y++) { + nv04_surface_copy_m2mf_swizzle(ctx, dst, dx, dy+y, + src, sx, sy+y); + } + } else if (h==1) { + /* We have a row to copy to a swizzled texture */ + for (x=0; x<w; x++) { + nv04_surface_copy_m2mf_swizzle(ctx, dst, dx+x, dy, + src, sx+x, sy); + } + } + return; } @@ -266,8 +359,7 @@ nv04_surface_copy(struct nv04_surface_2d *ctx, struct pipe_surface *dst, * to NV_MEMORY_TO_MEMORY_FORMAT in this case. */ if ((src->offset & 63) || (dst->offset & 63) || - (src_pitch & 63) || (dst_pitch & 63) || - debug_get_bool_option("NOUVEAU_NO_COPYBLIT", FALSE)) { + (src_pitch & 63) || (dst_pitch & 63)) { nv04_surface_copy_m2mf(ctx, dst, dx, dy, src, sx, sy, w, h); return; } @@ -279,10 +371,10 @@ static void nv04_surface_fill(struct nv04_surface_2d *ctx, struct pipe_surface *dst, int dx, int dy, int w, int h, unsigned value) { - struct nouveau_channel *chan = ctx->nvws->channel; + struct nouveau_channel *chan = ctx->surf2d->channel; struct nouveau_grobj *surf2d = ctx->surf2d; struct nouveau_grobj *rect = ctx->rect; - struct nouveau_bo *dst_bo = ctx->nvws->get_bo(ctx->buf(dst)); + struct nouveau_bo *dst_bo = nouveau_bo(ctx->buf(dst)); unsigned dst_pitch = ((struct nv04_surface *)dst)->pitch; int cs2d_format, gdirect_format; @@ -334,10 +426,10 @@ nv04_surface_2d_takedown(struct nv04_surface_2d **pctx) } struct nv04_surface_2d * -nv04_surface_2d_init(struct nouveau_winsys *nvws) +nv04_surface_2d_init(struct nouveau_screen *screen) { struct nv04_surface_2d *ctx = CALLOC_STRUCT(nv04_surface_2d); - struct nouveau_channel *chan = nvws->channel; + struct nouveau_channel *chan = screen->channel; unsigned handle = 0x88000000, class; int ret; @@ -460,7 +552,6 @@ nv04_surface_2d_init(struct nouveau_winsys *nvws) return NULL; } - ctx->nvws = nvws; ctx->copy = nv04_surface_copy; ctx->fill = nv04_surface_fill; return ctx; diff --git a/src/gallium/drivers/nv04/nv04_surface_2d.h b/src/gallium/drivers/nv04/nv04_surface_2d.h index 82ce7189c8b..02b3f56ba8b 100644 --- a/src/gallium/drivers/nv04/nv04_surface_2d.h +++ b/src/gallium/drivers/nv04/nv04_surface_2d.h @@ -7,7 +7,6 @@ struct nv04_surface { }; struct nv04_surface_2d { - struct nouveau_winsys *nvws; struct nouveau_notifier *ntfy; struct nouveau_grobj *surf2d; struct nouveau_grobj *swzsurf; @@ -26,7 +25,7 @@ struct nv04_surface_2d { }; struct nv04_surface_2d * -nv04_surface_2d_init(struct nouveau_winsys *nvws); +nv04_surface_2d_init(struct nouveau_screen *screen); void nv04_surface_2d_takedown(struct nv04_surface_2d **); diff --git a/src/gallium/drivers/nv04/nv04_vbo.c b/src/gallium/drivers/nv04/nv04_vbo.c index d21a0e34f71..e3167814f2b 100644 --- a/src/gallium/drivers/nv04/nv04_vbo.c +++ b/src/gallium/drivers/nv04/nv04_vbo.c @@ -1,6 +1,7 @@ #include "draw/draw_context.h" #include "pipe/p_context.h" #include "pipe/p_state.h" +#include "pipe/p_inlines.h" #include "nv04_context.h" #include "nv04_state.h" @@ -13,6 +14,7 @@ boolean nv04_draw_elements( struct pipe_context *pipe, unsigned indexSize, unsigned prim, unsigned start, unsigned count) { + struct pipe_screen *pscreen = pipe->screen; struct nv04_context *nv04 = nv04_context( pipe ); struct draw_context *draw = nv04->draw; unsigned i; @@ -25,17 +27,17 @@ boolean nv04_draw_elements( struct pipe_context *pipe, for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { if (nv04->vtxbuf[i].buffer) { void *buf - = pipe->winsys->buffer_map(pipe->winsys, - nv04->vtxbuf[i].buffer, - PIPE_BUFFER_USAGE_CPU_READ); + = pipe_buffer_map(pscreen, + nv04->vtxbuf[i].buffer, + PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_vertex_buffer(draw, i, buf); } } /* Map index buffer, if present */ if (indexBuffer) { void *mapped_indexes - = pipe->winsys->buffer_map(pipe->winsys, indexBuffer, - PIPE_BUFFER_USAGE_CPU_READ); + = pipe_buffer_map(pscreen, indexBuffer, + PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes); } else { @@ -55,12 +57,12 @@ boolean nv04_draw_elements( struct pipe_context *pipe, */ for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { if (nv04->vtxbuf[i].buffer) { - pipe->winsys->buffer_unmap(pipe->winsys, nv04->vtxbuf[i].buffer); + pipe_buffer_unmap(pscreen, nv04->vtxbuf[i].buffer); draw_set_mapped_vertex_buffer(draw, i, NULL); } } if (indexBuffer) { - pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer); + pipe_buffer_unmap(pscreen, indexBuffer); draw_set_mapped_element_buffer(draw, 0, NULL); } diff --git a/src/gallium/drivers/nv10/nv10_context.c b/src/gallium/drivers/nv10/nv10_context.c index 3da8d2f568f..a127b134ecd 100644 --- a/src/gallium/drivers/nv10/nv10_context.c +++ b/src/gallium/drivers/nv10/nv10_context.c @@ -30,18 +30,18 @@ nv10_destroy(struct pipe_context *pipe) static void nv10_init_hwctx(struct nv10_context *nv10) { struct nv10_screen *screen = nv10->screen; - struct nouveau_winsys *nvws = screen->nvws; + struct nouveau_channel *chan = screen->base.channel; int i; float projectionmatrix[16]; BEGIN_RING(celsius, NV10TCL_DMA_NOTIFY, 1); OUT_RING (screen->sync->handle); BEGIN_RING(celsius, NV10TCL_DMA_IN_MEMORY0, 2); - OUT_RING (nvws->channel->vram->handle); - OUT_RING (nvws->channel->gart->handle); + OUT_RING (chan->vram->handle); + OUT_RING (chan->gart->handle); BEGIN_RING(celsius, NV10TCL_DMA_IN_MEMORY2, 2); - OUT_RING (nvws->channel->vram->handle); - OUT_RING (nvws->channel->vram->handle); + OUT_RING (chan->vram->handle); + OUT_RING (chan->vram->handle); BEGIN_RING(celsius, NV10TCL_NOP, 1); OUT_RING (0); diff --git a/src/gallium/drivers/nv10/nv10_prim_vbuf.c b/src/gallium/drivers/nv10/nv10_prim_vbuf.c index 089c236302a..1806d5f8ccc 100644 --- a/src/gallium/drivers/nv10/nv10_prim_vbuf.c +++ b/src/gallium/drivers/nv10/nv10_prim_vbuf.c @@ -40,7 +40,6 @@ #include "util/u_debug.h" #include "pipe/p_inlines.h" -#include "pipe/internal/p_winsys_screen.h" #include "nv10_context.h" #include "nv10_state.h" @@ -124,11 +123,10 @@ nv10_vbuf_render_map_vertices( struct vbuf_render *render ) { struct nv10_vbuf_render *nv10_render = nv10_vbuf_render(render); struct nv10_context *nv10 = nv10_render->nv10; - struct pipe_winsys *winsys = nv10->pipe.winsys; + struct pipe_screen *pscreen = nv10->pipe.screen; - return winsys->buffer_map(winsys, - nv10_render->buffer, - PIPE_BUFFER_USAGE_CPU_WRITE); + return pipe_buffer_map(pscreen, nv10_render->buffer, + PIPE_BUFFER_USAGE_CPU_WRITE); } static void @@ -138,10 +136,10 @@ nv10_vbuf_render_unmap_vertices( struct vbuf_render *render, { struct nv10_vbuf_render *nv10_render = nv10_vbuf_render(render); struct nv10_context *nv10 = nv10_render->nv10; - struct pipe_winsys *winsys = nv10->pipe.winsys; + struct pipe_screen *pscreen = nv10->pipe.screen; assert(!nv10_render->buffer); - winsys->buffer_unmap(winsys, nv10_render->buffer); + pipe_buffer_unmap(pscreen, nv10_render->buffer); } static boolean @@ -202,8 +200,6 @@ static void nv10_vbuf_render_release_vertices( struct vbuf_render *render ) { struct nv10_vbuf_render *nv10_render = nv10_vbuf_render(render); - struct nv10_context *nv10 = nv10_render->nv10; - struct pipe_screen *pscreen = &nv10->screen->pipe; assert(nv10_render->buffer); pipe_buffer_reference(&nv10_render->buffer, NULL); diff --git a/src/gallium/drivers/nv10/nv10_screen.c b/src/gallium/drivers/nv10/nv10_screen.c index 6532a93c7ba..4469b22d91a 100644 --- a/src/gallium/drivers/nv10/nv10_screen.c +++ b/src/gallium/drivers/nv10/nv10_screen.c @@ -1,26 +1,8 @@ #include "pipe/p_screen.h" -#include "util/u_simple_screen.h" #include "nv10_context.h" #include "nv10_screen.h" -static const char * -nv10_screen_get_name(struct pipe_screen *screen) -{ - struct nv10_screen *nv10screen = nv10_screen(screen); - struct nouveau_device *dev = nv10screen->nvws->channel->device; - static char buffer[128]; - - snprintf(buffer, sizeof(buffer), "NV%02X", dev->chipset); - return buffer; -} - -static const char * -nv10_screen_get_vendor(struct pipe_screen *screen) -{ - return "nouveau"; -} - static int nv10_screen_get_param(struct pipe_screen *screen, int param) { @@ -53,6 +35,10 @@ nv10_screen_get_param(struct pipe_screen *screen, int param) return 12; case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS: return 0; + case PIPE_CAP_TGSI_CONT_SUPPORTED: + return 0; + case PIPE_CAP_BLEND_EQUATION_SEPARATE: + return 0; case NOUVEAU_CAP_HW_VTXBUF: case NOUVEAU_CAP_HW_IDXBUF: return 0; @@ -92,7 +78,15 @@ nv10_screen_is_format_supported(struct pipe_screen *screen, switch (format) { case PIPE_FORMAT_A8R8G8B8_UNORM: case PIPE_FORMAT_R5G6B5_UNORM: + return TRUE; + default: + break; + } + } else + if (tex_usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL) { + switch (format) { case PIPE_FORMAT_Z24S8_UNORM: + case PIPE_FORMAT_Z24X8_UNORM: case PIPE_FORMAT_Z16_UNORM: return TRUE; default: @@ -120,10 +114,9 @@ static void nv10_screen_destroy(struct pipe_screen *pscreen) { struct nv10_screen *screen = nv10_screen(pscreen); - struct nouveau_winsys *nvws = screen->nvws; - nvws->notifier_free(&screen->sync); - nvws->grobj_free(&screen->celsius); + nouveau_notifier_free(&screen->sync); + nouveau_grobj_free(&screen->celsius); FREE(pscreen); } @@ -137,64 +130,69 @@ nv10_surface_buffer(struct pipe_surface *surf) } struct pipe_screen * -nv10_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) +nv10_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) { struct nv10_screen *screen = CALLOC_STRUCT(nv10_screen); + struct nouveau_channel *chan; + struct pipe_screen *pscreen; unsigned celsius_class; - unsigned chipset = nvws->channel->device->chipset; int ret; if (!screen) return NULL; - screen->nvws = nvws; + pscreen = &screen->base.base; - /* 2D engine setup */ - screen->eng2d = nv04_surface_2d_init(nvws); - screen->eng2d->buf = nv10_surface_buffer; + ret = nouveau_screen_init(&screen->base, dev); + if (ret) { + nv10_screen_destroy(pscreen); + return NULL; + } + chan = screen->base.channel; + + pscreen->winsys = ws; + pscreen->destroy = nv10_screen_destroy; + pscreen->get_param = nv10_screen_get_param; + pscreen->get_paramf = nv10_screen_get_paramf; + pscreen->is_format_supported = nv10_screen_is_format_supported; + + nv10_screen_init_miptree_functions(pscreen); + nv10_screen_init_transfer_functions(pscreen); /* 3D object */ - if (chipset>=0x20) - celsius_class=NV11TCL; - else if (chipset>=0x17) - celsius_class=NV17TCL; - else if (chipset>=0x11) - celsius_class=NV11TCL; + if (dev->chipset >= 0x20) + celsius_class = NV11TCL; + else if (dev->chipset >= 0x17) + celsius_class = NV17TCL; + else if (dev->chipset >= 0x11) + celsius_class = NV11TCL; else - celsius_class=NV10TCL; + celsius_class = NV10TCL; if (!celsius_class) { - NOUVEAU_ERR("Unknown nv1x chipset: nv%02x\n", chipset); + NOUVEAU_ERR("Unknown nv1x chipset: nv%02x\n", dev->chipset); return NULL; } - ret = nvws->grobj_alloc(nvws, celsius_class, &screen->celsius); + ret = nouveau_grobj_alloc(chan, 0xbeef0001, celsius_class, + &screen->celsius); if (ret) { NOUVEAU_ERR("Error creating 3D object: %d\n", ret); return FALSE; } + BIND_RING(chan, screen->celsius, 7); + + /* 2D engine setup */ + screen->eng2d = nv04_surface_2d_init(&screen->base); + screen->eng2d->buf = nv10_surface_buffer; /* Notifier for sync purposes */ - ret = nvws->notifier_alloc(nvws, 1, &screen->sync); + ret = nouveau_notifier_alloc(chan, 0xbeef0301, 1, &screen->sync); if (ret) { NOUVEAU_ERR("Error creating notifier object: %d\n", ret); - nv10_screen_destroy(&screen->pipe); + nv10_screen_destroy(pscreen); return NULL; } - screen->pipe.winsys = ws; - screen->pipe.destroy = nv10_screen_destroy; - - screen->pipe.get_name = nv10_screen_get_name; - screen->pipe.get_vendor = nv10_screen_get_vendor; - screen->pipe.get_param = nv10_screen_get_param; - screen->pipe.get_paramf = nv10_screen_get_paramf; - - screen->pipe.is_format_supported = nv10_screen_is_format_supported; - - nv10_screen_init_miptree_functions(&screen->pipe); - nv10_screen_init_transfer_functions(&screen->pipe); - u_simple_screen_init(&screen->pipe); - - return &screen->pipe; + return pscreen; } diff --git a/src/gallium/drivers/nv10/nv10_screen.h b/src/gallium/drivers/nv10/nv10_screen.h index ad829ee3fd4..86b6d8def54 100644 --- a/src/gallium/drivers/nv10/nv10_screen.h +++ b/src/gallium/drivers/nv10/nv10_screen.h @@ -1,11 +1,11 @@ #ifndef __NV10_SCREEN_H__ #define __NV10_SCREEN_H__ -#include "pipe/p_screen.h" +#include "nouveau/nouveau_screen.h" #include "nv04/nv04_surface_2d.h" struct nv10_screen { - struct pipe_screen pipe; + struct nouveau_screen base; struct nouveau_winsys *nvws; diff --git a/src/gallium/drivers/nv10/nv10_state.c b/src/gallium/drivers/nv10/nv10_state.c index 119af66dfd0..9b38219b996 100644 --- a/src/gallium/drivers/nv10/nv10_state.c +++ b/src/gallium/drivers/nv10/nv10_state.c @@ -2,6 +2,7 @@ #include "pipe/p_state.h" #include "pipe/p_defines.h" #include "pipe/p_shader_tokens.h" +#include "pipe/p_inlines.h" #include "tgsi/tgsi_parse.h" @@ -460,7 +461,7 @@ nv10_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, const struct pipe_constant_buffer *buf ) { struct nv10_context *nv10 = nv10_context(pipe); - struct pipe_winsys *ws = pipe->winsys; + struct pipe_screen *pscreen = pipe->screen; assert(shader < PIPE_SHADER_TYPES); assert(index == 0); @@ -468,12 +469,12 @@ nv10_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, if (buf) { void *mapped; if (buf->buffer && buf->buffer->size && - (mapped = ws->buffer_map(ws, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ))) + (mapped = pipe_buffer_map(pscreen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ))) { memcpy(nv10->constbuf[shader], mapped, buf->buffer->size); nv10->constbuf_nr[shader] = buf->buffer->size / (4 * sizeof(float)); - ws->buffer_unmap(ws, buf->buffer); + pipe_buffer_unmap(pscreen, buf->buffer); } } } diff --git a/src/gallium/drivers/nv10/nv10_vbo.c b/src/gallium/drivers/nv10/nv10_vbo.c index d0e788ac036..441a4f75f3f 100644 --- a/src/gallium/drivers/nv10/nv10_vbo.c +++ b/src/gallium/drivers/nv10/nv10_vbo.c @@ -1,6 +1,7 @@ #include "draw/draw_context.h" #include "pipe/p_context.h" #include "pipe/p_state.h" +#include "pipe/p_inlines.h" #include "nv10_context.h" #include "nv10_state.h" @@ -15,6 +16,7 @@ boolean nv10_draw_elements( struct pipe_context *pipe, { struct nv10_context *nv10 = nv10_context( pipe ); struct draw_context *draw = nv10->draw; + struct pipe_screen *pscreen = pipe->screen; unsigned i; nv10_emit_hw_state(nv10); @@ -24,9 +26,8 @@ boolean nv10_draw_elements( struct pipe_context *pipe, */ for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { if (nv10->vtxbuf[i].buffer) { - void *buf - = pipe->winsys->buffer_map(pipe->winsys, - nv10->vtxbuf[i].buffer, + void *buf = + pipe_buffer_map(pscreen, nv10->vtxbuf[i].buffer, PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_vertex_buffer(draw, i, buf); } @@ -34,8 +35,8 @@ boolean nv10_draw_elements( struct pipe_context *pipe, /* Map index buffer, if present */ if (indexBuffer) { void *mapped_indexes - = pipe->winsys->buffer_map(pipe->winsys, indexBuffer, - PIPE_BUFFER_USAGE_CPU_READ); + = pipe_buffer_map(pscreen, indexBuffer, + PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes); } else { @@ -55,12 +56,12 @@ boolean nv10_draw_elements( struct pipe_context *pipe, */ for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { if (nv10->vtxbuf[i].buffer) { - pipe->winsys->buffer_unmap(pipe->winsys, nv10->vtxbuf[i].buffer); + pipe_buffer_unmap(pscreen, nv10->vtxbuf[i].buffer); draw_set_mapped_vertex_buffer(draw, i, NULL); } } if (indexBuffer) { - pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer); + pipe_buffer_unmap(pscreen, indexBuffer); draw_set_mapped_element_buffer(draw, 0, NULL); } diff --git a/src/gallium/drivers/nv20/nv20_context.c b/src/gallium/drivers/nv20/nv20_context.c index cbc41707d54..b32d0d83ba0 100644 --- a/src/gallium/drivers/nv20/nv20_context.c +++ b/src/gallium/drivers/nv20/nv20_context.c @@ -30,7 +30,7 @@ nv20_destroy(struct pipe_context *pipe) static void nv20_init_hwctx(struct nv20_context *nv20) { struct nv20_screen *screen = nv20->screen; - struct nouveau_winsys *nvws = screen->nvws; + struct nouveau_channel *chan = screen->base.channel; int i; float projectionmatrix[16]; const boolean is_nv25tcl = (nv20->screen->kelvin->grclass == NV25TCL); @@ -38,11 +38,11 @@ static void nv20_init_hwctx(struct nv20_context *nv20) BEGIN_RING(kelvin, NV20TCL_DMA_NOTIFY, 1); OUT_RING (screen->sync->handle); BEGIN_RING(kelvin, NV20TCL_DMA_TEXTURE0, 2); - OUT_RING (nvws->channel->vram->handle); - OUT_RING (nvws->channel->gart->handle); /* TEXTURE1 */ + OUT_RING (chan->vram->handle); + OUT_RING (chan->gart->handle); /* TEXTURE1 */ BEGIN_RING(kelvin, NV20TCL_DMA_COLOR, 2); - OUT_RING (nvws->channel->vram->handle); - OUT_RING (nvws->channel->vram->handle); /* ZETA */ + OUT_RING (chan->vram->handle); + OUT_RING (chan->vram->handle); /* ZETA */ BEGIN_RING(kelvin, NV20TCL_DMA_QUERY, 1); OUT_RING (0); /* renouveau: beef0351, unique */ @@ -99,9 +99,9 @@ static void nv20_init_hwctx(struct nv20_context *nv20) OUT_RING (3); BEGIN_RING(kelvin, NV25TCL_DMA_IN_MEMORY9, 1); - OUT_RING (nvws->channel->vram->handle); + OUT_RING (chan->vram->handle); BEGIN_RING(kelvin, NV25TCL_DMA_IN_MEMORY8, 1); - OUT_RING (nvws->channel->vram->handle); + OUT_RING (chan->vram->handle); } BEGIN_RING(kelvin, NV20TCL_DMA_FENCE, 1); OUT_RING (0); /* renouveau: beef1e10 */ diff --git a/src/gallium/drivers/nv20/nv20_prim_vbuf.c b/src/gallium/drivers/nv20/nv20_prim_vbuf.c index 8aa342cd2db..ddfcdb8057a 100644 --- a/src/gallium/drivers/nv20/nv20_prim_vbuf.c +++ b/src/gallium/drivers/nv20/nv20_prim_vbuf.c @@ -152,12 +152,11 @@ static void * nv20_vbuf_render_map_vertices( struct vbuf_render *render ) { struct nv20_vbuf_render *nv20_render = nv20_vbuf_render(render); - struct pipe_winsys *winsys = nv20_render->nv20->pipe.winsys; + struct pipe_screen *pscreen = nv20_render->nv20->pipe.screen; if (nv20_render->pbuffer) { - return winsys->buffer_map(winsys, - nv20_render->pbuffer, - PIPE_BUFFER_USAGE_CPU_WRITE); + return pipe_buffer_map(pscreen, nv20_render->pbuffer, + PIPE_BUFFER_USAGE_CPU_WRITE); } else if (nv20_render->mbuffer) { return nv20_render->mbuffer; } else @@ -173,10 +172,10 @@ nv20_vbuf_render_unmap_vertices( struct vbuf_render *render, ushort max_index ) { struct nv20_vbuf_render *nv20_render = nv20_vbuf_render(render); - struct pipe_winsys *winsys = nv20_render->nv20->pipe.winsys; + struct pipe_screen *pscreen = nv20_render->nv20->pipe.screen; if (nv20_render->pbuffer) - winsys->buffer_unmap(winsys, nv20_render->pbuffer); + pipe_buffer_unmap(pscreen, nv20_render->pbuffer); } static boolean @@ -358,7 +357,6 @@ nv20_vbuf_render_release_vertices( struct vbuf_render *render ) { struct nv20_vbuf_render *nv20_render = nv20_vbuf_render(render); struct nv20_context *nv20 = nv20_render->nv20; - struct pipe_screen *pscreen = &nv20->screen->pipe; if (nv20_render->pbuffer) { pipe_buffer_reference(&nv20_render->pbuffer, NULL); diff --git a/src/gallium/drivers/nv20/nv20_screen.c b/src/gallium/drivers/nv20/nv20_screen.c index 7760ae27c0e..e6924ad71eb 100644 --- a/src/gallium/drivers/nv20/nv20_screen.c +++ b/src/gallium/drivers/nv20/nv20_screen.c @@ -1,26 +1,8 @@ #include "pipe/p_screen.h" -#include "util/u_simple_screen.h" #include "nv20_context.h" #include "nv20_screen.h" -static const char * -nv20_screen_get_name(struct pipe_screen *screen) -{ - struct nv20_screen *nv20screen = nv20_screen(screen); - struct nouveau_device *dev = nv20screen->nvws->channel->device; - static char buffer[128]; - - snprintf(buffer, sizeof(buffer), "NV%02X", dev->chipset); - return buffer; -} - -static const char * -nv20_screen_get_vendor(struct pipe_screen *screen) -{ - return "nouveau"; -} - static int nv20_screen_get_param(struct pipe_screen *screen, int param) { @@ -53,6 +35,10 @@ nv20_screen_get_param(struct pipe_screen *screen, int param) return 12; case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS: return 0; + case PIPE_CAP_TGSI_CONT_SUPPORTED: + return 0; + case PIPE_CAP_BLEND_EQUATION_SEPARATE: + return 0; case NOUVEAU_CAP_HW_VTXBUF: case NOUVEAU_CAP_HW_IDXBUF: return 0; @@ -92,7 +78,15 @@ nv20_screen_is_format_supported(struct pipe_screen *screen, switch (format) { case PIPE_FORMAT_A8R8G8B8_UNORM: case PIPE_FORMAT_R5G6B5_UNORM: + return TRUE; + default: + break; + } + } else + if (tex_usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL) { + switch (format) { case PIPE_FORMAT_Z24S8_UNORM: + case PIPE_FORMAT_Z24X8_UNORM: case PIPE_FORMAT_Z16_UNORM: return TRUE; default: @@ -120,10 +114,9 @@ static void nv20_screen_destroy(struct pipe_screen *pscreen) { struct nv20_screen *screen = nv20_screen(pscreen); - struct nouveau_winsys *nvws = screen->nvws; - nvws->notifier_free(&screen->sync); - nvws->grobj_free(&screen->kelvin); + nouveau_notifier_free(&screen->sync); + nouveau_grobj_free(&screen->kelvin); FREE(pscreen); } @@ -137,60 +130,65 @@ nv20_surface_buffer(struct pipe_surface *surf) } struct pipe_screen * -nv20_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) +nv20_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) { struct nv20_screen *screen = CALLOC_STRUCT(nv20_screen); + struct nouveau_channel *chan; + struct pipe_screen *pscreen; unsigned kelvin_class = 0; - unsigned chipset = nvws->channel->device->chipset; int ret; if (!screen) return NULL; - screen->nvws = nvws; + pscreen = &screen->base.base; - /* 2D engine setup */ - screen->eng2d = nv04_surface_2d_init(nvws); - screen->eng2d->buf = nv20_surface_buffer; + ret = nouveau_screen_init(&screen->base, dev); + if (ret) { + nv20_screen_destroy(pscreen); + return NULL; + } + chan = screen->base.channel; + + pscreen->winsys = ws; + pscreen->destroy = nv20_screen_destroy; + pscreen->get_param = nv20_screen_get_param; + pscreen->get_paramf = nv20_screen_get_paramf; + pscreen->is_format_supported = nv20_screen_is_format_supported; + + nv20_screen_init_miptree_functions(pscreen); + nv20_screen_init_transfer_functions(pscreen); /* 3D object */ - if (chipset >= 0x25) + if (dev->chipset >= 0x25) kelvin_class = NV25TCL; - else if (chipset >= 0x20) + else if (dev->chipset >= 0x20) kelvin_class = NV20TCL; - if (!kelvin_class || chipset >= 0x30) { - NOUVEAU_ERR("Unknown nv2x chipset: nv%02x\n", chipset); + if (!kelvin_class || dev->chipset >= 0x30) { + NOUVEAU_ERR("Unknown nv2x chipset: nv%02x\n", dev->chipset); return NULL; } - ret = nvws->grobj_alloc(nvws, kelvin_class, &screen->kelvin); + ret = nouveau_grobj_alloc(chan, 0xbeef0097, kelvin_class, + &screen->kelvin); if (ret) { NOUVEAU_ERR("Error creating 3D object: %d\n", ret); return FALSE; } + BIND_RING(chan, screen->kelvin, 7); + + /* 2D engine setup */ + screen->eng2d = nv04_surface_2d_init(&screen->base); + screen->eng2d->buf = nv20_surface_buffer; /* Notifier for sync purposes */ - ret = nvws->notifier_alloc(nvws, 1, &screen->sync); + ret = nouveau_notifier_alloc(chan, 0xbeef0301, 1, &screen->sync); if (ret) { NOUVEAU_ERR("Error creating notifier object: %d\n", ret); - nv20_screen_destroy(&screen->pipe); + nv20_screen_destroy(pscreen); return NULL; } - screen->pipe.winsys = ws; - screen->pipe.destroy = nv20_screen_destroy; - - screen->pipe.get_name = nv20_screen_get_name; - screen->pipe.get_vendor = nv20_screen_get_vendor; - screen->pipe.get_param = nv20_screen_get_param; - screen->pipe.get_paramf = nv20_screen_get_paramf; - - screen->pipe.is_format_supported = nv20_screen_is_format_supported; - - nv20_screen_init_miptree_functions(&screen->pipe); - nv20_screen_init_transfer_functions(&screen->pipe); - u_simple_screen_init(&screen->pipe); - - return &screen->pipe; + return pscreen; } diff --git a/src/gallium/drivers/nv20/nv20_screen.h b/src/gallium/drivers/nv20/nv20_screen.h index d9fce2bced8..fc7bb050334 100644 --- a/src/gallium/drivers/nv20/nv20_screen.h +++ b/src/gallium/drivers/nv20/nv20_screen.h @@ -1,11 +1,11 @@ #ifndef __NV20_SCREEN_H__ #define __NV20_SCREEN_H__ -#include "pipe/p_screen.h" +#include "nouveau/nouveau_screen.h" #include "nv04/nv04_surface_2d.h" struct nv20_screen { - struct pipe_screen pipe; + struct nouveau_screen base; struct nouveau_winsys *nvws; diff --git a/src/gallium/drivers/nv20/nv20_state.c b/src/gallium/drivers/nv20/nv20_state.c index ecec4f49a0c..ed4084980f2 100644 --- a/src/gallium/drivers/nv20/nv20_state.c +++ b/src/gallium/drivers/nv20/nv20_state.c @@ -2,6 +2,7 @@ #include "pipe/p_state.h" #include "pipe/p_defines.h" #include "pipe/p_shader_tokens.h" +#include "pipe/p_inlines.h" #include "tgsi/tgsi_parse.h" @@ -453,7 +454,7 @@ nv20_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, const struct pipe_constant_buffer *buf ) { struct nv20_context *nv20 = nv20_context(pipe); - struct pipe_winsys *ws = pipe->winsys; + struct pipe_screen *pscreen = pipe->screen; assert(shader < PIPE_SHADER_TYPES); assert(index == 0); @@ -461,12 +462,12 @@ nv20_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, if (buf) { void *mapped; if (buf->buffer && buf->buffer->size && - (mapped = ws->buffer_map(ws, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ))) + (mapped = pipe_buffer_map(pscreen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ))) { memcpy(nv20->constbuf[shader], mapped, buf->buffer->size); nv20->constbuf_nr[shader] = buf->buffer->size / (4 * sizeof(float)); - ws->buffer_unmap(ws, buf->buffer); + pipe_buffer_unmap(pscreen, buf->buffer); } } } diff --git a/src/gallium/drivers/nv20/nv20_vbo.c b/src/gallium/drivers/nv20/nv20_vbo.c index 24d8f4bef09..84d7db6c5e2 100644 --- a/src/gallium/drivers/nv20/nv20_vbo.c +++ b/src/gallium/drivers/nv20/nv20_vbo.c @@ -1,6 +1,7 @@ #include "draw/draw_context.h" #include "pipe/p_context.h" #include "pipe/p_state.h" +#include "pipe/p_inlines.h" #include "nv20_context.h" #include "nv20_state.h" @@ -13,6 +14,7 @@ boolean nv20_draw_elements( struct pipe_context *pipe, unsigned indexSize, unsigned prim, unsigned start, unsigned count) { + struct pipe_screen *pscreen = pipe->screen; struct nv20_context *nv20 = nv20_context( pipe ); struct draw_context *draw = nv20->draw; unsigned i; @@ -25,17 +27,17 @@ boolean nv20_draw_elements( struct pipe_context *pipe, for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { if (nv20->vtxbuf[i].buffer) { void *buf - = pipe->winsys->buffer_map(pipe->winsys, - nv20->vtxbuf[i].buffer, - PIPE_BUFFER_USAGE_CPU_READ); + = pipe_buffer_map(pscreen, + nv20->vtxbuf[i].buffer, + PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_vertex_buffer(draw, i, buf); } } /* Map index buffer, if present */ if (indexBuffer) { void *mapped_indexes - = pipe->winsys->buffer_map(pipe->winsys, indexBuffer, - PIPE_BUFFER_USAGE_CPU_READ); + = pipe_buffer_map(pscreen, indexBuffer, + PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes); } else { @@ -55,12 +57,12 @@ boolean nv20_draw_elements( struct pipe_context *pipe, */ for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { if (nv20->vtxbuf[i].buffer) { - pipe->winsys->buffer_unmap(pipe->winsys, nv20->vtxbuf[i].buffer); + pipe_buffer_unmap(pscreen, nv20->vtxbuf[i].buffer); draw_set_mapped_vertex_buffer(draw, i, NULL); } } if (indexBuffer) { - pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer); + pipe_buffer_unmap(pscreen, indexBuffer); draw_set_mapped_element_buffer(draw, 0, NULL); } diff --git a/src/gallium/drivers/nv20/nv20_vertprog.c b/src/gallium/drivers/nv20/nv20_vertprog.c index 5db0e807ff5..388245ecb04 100644 --- a/src/gallium/drivers/nv20/nv20_vertprog.c +++ b/src/gallium/drivers/nv20/nv20_vertprog.c @@ -1,6 +1,7 @@ #include "pipe/p_context.h" #include "pipe/p_defines.h" #include "pipe/p_state.h" +#include "pipe/p_inlines.h" #include "pipe/p_shader_tokens.h" #include "tgsi/tgsi_parse.h" @@ -616,10 +617,10 @@ nv20_vertprog_translate(struct nv20_context *nv20, assert(imm->Immediate.NrTokens == 4 + 1); vpc->imm[vpc->nr_imm++] = constant(vpc, -1, - imm->u.ImmediateFloat32[0].Float, - imm->u.ImmediateFloat32[1].Float, - imm->u.ImmediateFloat32[2].Float, - imm->u.ImmediateFloat32[3].Float); + imm->u[0].Float, + imm->u[1].Float, + imm->u[2].Float, + imm->u[3].Float); } break; case TGSI_TOKEN_TYPE_INSTRUCTION: @@ -645,8 +646,8 @@ out_err: static boolean nv20_vertprog_validate(struct nv20_context *nv20) { + struct pipe_screen *pscreen = nv20->pipe.screen; struct nouveau_winsys *nvws = nv20->nvws; - struct pipe_winsys *ws = nv20->pipe.winsys; struct nouveau_grobj *rankine = nv20->screen->rankine; struct nv20_vertex_program *vp; struct pipe_buffer *constbuf; @@ -749,8 +750,8 @@ nv20_vertprog_validate(struct nv20_context *nv20) float *map = NULL; if (constbuf) { - map = ws->buffer_map(ws, constbuf, - PIPE_BUFFER_USAGE_CPU_READ); + map = pipe_buffer_map(pscreen, constbuf, + PIPE_BUFFER_USAGE_CPU_READ); } for (i = 0; i < vp->nr_consts; i++) { @@ -770,9 +771,8 @@ nv20_vertprog_validate(struct nv20_context *nv20) OUT_RINGp ((uint32_t *)vpd->value, 4); } - if (constbuf) { - ws->buffer_unmap(ws, constbuf); - } + if (constbuf) + pipe_buffer_unmap(pscreen, constbuf); } /* Upload vtxprog */ diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index bdfe1425d24..a48ba9782b3 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -1,6 +1,7 @@ #include "pipe/p_context.h" #include "pipe/p_defines.h" #include "pipe/p_state.h" +#include "pipe/p_inlines.h" #include "pipe/p_shader_tokens.h" #include "tgsi/tgsi_parse.h" @@ -703,10 +704,10 @@ nv30_fragprog_prepare(struct nv30_fpc *fpc) assert(imm->Immediate.DataType == TGSI_IMM_FLOAT32); assert(fpc->nr_imm < MAX_IMM); - vals[0] = imm->u.ImmediateFloat32[0].Float; - vals[1] = imm->u.ImmediateFloat32[1].Float; - vals[2] = imm->u.ImmediateFloat32[2].Float; - vals[3] = imm->u.ImmediateFloat32[3].Float; + vals[0] = imm->u[0].Float; + vals[1] = imm->u[1].Float; + vals[2] = imm->u[2].Float; + vals[3] = imm->u[3].Float; fpc->imm[fpc->nr_imm++] = constant(fpc, -1, vals); } break; @@ -798,12 +799,12 @@ static void nv30_fragprog_upload(struct nv30_context *nv30, struct nv30_fragment_program *fp) { - struct pipe_winsys *ws = nv30->pipe.winsys; + struct pipe_screen *pscreen = nv30->pipe.screen; const uint32_t le = 1; uint32_t *map; int i; - map = ws->buffer_map(ws, fp->buffer, PIPE_BUFFER_USAGE_CPU_WRITE); + map = pipe_buffer_map(pscreen, fp->buffer, PIPE_BUFFER_USAGE_CPU_WRITE); #if 0 for (i = 0; i < fp->insn_len; i++) { @@ -825,7 +826,7 @@ nv30_fragprog_upload(struct nv30_context *nv30, } } - ws->buffer_unmap(ws, fp->buffer); + pipe_buffer_unmap(pscreen, fp->buffer); } static boolean @@ -834,8 +835,7 @@ nv30_fragprog_validate(struct nv30_context *nv30) struct nv30_fragment_program *fp = nv30->fragprog; struct pipe_buffer *constbuf = nv30->constbuf[PIPE_SHADER_FRAGMENT]; - struct pipe_screen *screen = nv30->pipe.screen; - struct pipe_winsys *ws = nv30->pipe.winsys; + struct pipe_screen *pscreen = nv30->pipe.screen; struct nouveau_stateobj *so; boolean new_consts = FALSE; int i; @@ -850,14 +850,15 @@ nv30_fragprog_validate(struct nv30_context *nv30) return FALSE; } - fp->buffer = screen->buffer_create(screen, 0x100, 0, fp->insn_len * 4); + fp->buffer = pscreen->buffer_create(pscreen, 0x100, 0, fp->insn_len * 4); nv30_fragprog_upload(nv30, fp); so = so_new(8, 1); so_method(so, nv30->screen->rankine, NV34TCL_FP_ACTIVE_PROGRAM, 1); - so_reloc (so, fp->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | - NOUVEAU_BO_RD | NOUVEAU_BO_LOW | NOUVEAU_BO_OR, - NV34TCL_FP_ACTIVE_PROGRAM_DMA0, NV34TCL_FP_ACTIVE_PROGRAM_DMA1); + so_reloc (so, nouveau_bo(fp->buffer), 0, NOUVEAU_BO_VRAM | + NOUVEAU_BO_GART | NOUVEAU_BO_RD | NOUVEAU_BO_LOW | + NOUVEAU_BO_OR, NV34TCL_FP_ACTIVE_PROGRAM_DMA0, + NV34TCL_FP_ACTIVE_PROGRAM_DMA1); so_method(so, nv30->screen->rankine, NV34TCL_FP_CONTROL, 1); so_data (so, fp->fp_control); so_method(so, nv30->screen->rankine, NV34TCL_FP_REG_CONTROL, 1); @@ -871,7 +872,8 @@ update_constants: if (fp->nr_consts) { float *map; - map = ws->buffer_map(ws, constbuf, PIPE_BUFFER_USAGE_CPU_READ); + map = pipe_buffer_map(pscreen, constbuf, + PIPE_BUFFER_USAGE_CPU_READ); for (i = 0; i < fp->nr_consts; i++) { struct nv30_fragment_program_data *fpd = &fp->consts[i]; uint32_t *p = &fp->insn[fpd->offset]; @@ -882,7 +884,7 @@ update_constants: memcpy(p, cb, 4 * sizeof(float)); new_consts = TRUE; } - ws->buffer_unmap(ws, constbuf); + pipe_buffer_unmap(pscreen, constbuf); if (new_consts) nv30_fragprog_upload(nv30, fp); diff --git a/src/gallium/drivers/nv30/nv30_fragtex.c b/src/gallium/drivers/nv30/nv30_fragtex.c index 8b6ab992d16..822e1d8defe 100644 --- a/src/gallium/drivers/nv30/nv30_fragtex.c +++ b/src/gallium/drivers/nv30/nv30_fragtex.c @@ -61,6 +61,7 @@ nv30_fragtex_build(struct nv30_context *nv30, int unit) struct nv30_sampler_state *ps = nv30->tex_sampler[unit]; struct nv30_miptree *nv30mt = nv30->tex_miptree[unit]; struct pipe_texture *pt = &nv30mt->base; + struct nouveau_bo *bo = nouveau_bo(nv30mt->buffer); struct nv30_texture_format *tf; struct nouveau_stateobj *so; uint32_t txf, txs , txp; @@ -106,9 +107,9 @@ nv30_fragtex_build(struct nv30_context *nv30, int unit) so = so_new(16, 2); so_method(so, nv30->screen->rankine, NV34TCL_TX_OFFSET(unit), 8); - so_reloc (so, nv30mt->buffer, 0, tex_flags | NOUVEAU_BO_LOW, 0, 0); - so_reloc (so, nv30mt->buffer, txf, tex_flags | NOUVEAU_BO_OR, - NV34TCL_TX_FORMAT_DMA0, NV34TCL_TX_FORMAT_DMA1); + so_reloc (so, bo, 0, tex_flags | NOUVEAU_BO_LOW, 0, 0); + so_reloc (so, bo, txf, tex_flags | NOUVEAU_BO_OR, + NV34TCL_TX_FORMAT_DMA0, NV34TCL_TX_FORMAT_DMA1); so_data (so, ps->wrap); so_data (so, NV34TCL_TX_ENABLE_ENABLE | ps->en); so_data (so, txs); diff --git a/src/gallium/drivers/nv30/nv30_query.c b/src/gallium/drivers/nv30/nv30_query.c index 2f974cf5c40..1d1c8a484e1 100644 --- a/src/gallium/drivers/nv30/nv30_query.c +++ b/src/gallium/drivers/nv30/nv30_query.c @@ -29,11 +29,10 @@ nv30_query_create(struct pipe_context *pipe, unsigned query_type) static void nv30_query_destroy(struct pipe_context *pipe, struct pipe_query *pq) { - struct nv30_context *nv30 = nv30_context(pipe); struct nv30_query *q = nv30_query(pq); if (q->object) - nv30->nvws->res_free(&q->object); + nouveau_resource_free(&q->object); FREE(q); } @@ -54,9 +53,9 @@ nv30_query_begin(struct pipe_context *pipe, struct pipe_query *pq) pipe->get_query_result(pipe, pq, 1, &tmp); } - if (nv30->nvws->res_alloc(nv30->screen->query_heap, 1, NULL, &q->object)) + if (nouveau_resource_alloc(nv30->screen->query_heap, 1, NULL, &q->object)) assert(0); - nv30->nvws->notifier_reset(nv30->screen->query, q->object->start); + nouveau_notifier_reset(nv30->screen->query, q->object->start); BEGIN_RING(rankine, NV34TCL_QUERY_RESET, 1); OUT_RING (1); @@ -84,27 +83,27 @@ nv30_query_result(struct pipe_context *pipe, struct pipe_query *pq, { struct nv30_context *nv30 = nv30_context(pipe); struct nv30_query *q = nv30_query(pq); - struct nouveau_winsys *nvws = nv30->nvws; assert(q->object && q->type == PIPE_QUERY_OCCLUSION_COUNTER); if (!q->ready) { unsigned status; - status = nvws->notifier_status(nv30->screen->query, - q->object->start); + status = nouveau_notifier_status(nv30->screen->query, + q->object->start); if (status != NV_NOTIFY_STATE_STATUS_COMPLETED) { if (wait == FALSE) return FALSE; - nvws->notifier_wait(nv30->screen->query, q->object->start, - NV_NOTIFY_STATE_STATUS_COMPLETED, - 0); + + nouveau_notifier_wait_status(nv30->screen->query, + q->object->start, + NV_NOTIFY_STATE_STATUS_COMPLETED, 0); } - q->result = nvws->notifier_retval(nv30->screen->query, - q->object->start); + q->result = nouveau_notifier_return_val(nv30->screen->query, + q->object->start); q->ready = TRUE; - nvws->res_free(&q->object); + nouveau_resource_free(&q->object); } *result = q->result; diff --git a/src/gallium/drivers/nv30/nv30_screen.c b/src/gallium/drivers/nv30/nv30_screen.c index d395c5e1b7f..f8285e4455f 100644 --- a/src/gallium/drivers/nv30/nv30_screen.c +++ b/src/gallium/drivers/nv30/nv30_screen.c @@ -1,5 +1,7 @@ #include "pipe/p_screen.h" -#include "util/u_simple_screen.h" +#include "pipe/p_state.h" + +#include "nouveau/nouveau_screen.h" #include "nv30_context.h" #include "nv30_screen.h" @@ -8,23 +10,6 @@ #define NV34TCL_CHIPSET_3X_MASK 0x00000010 #define NV35TCL_CHIPSET_3X_MASK 0x000001e0 -static const char * -nv30_screen_get_name(struct pipe_screen *pscreen) -{ - struct nv30_screen *screen = nv30_screen(pscreen); - struct nouveau_device *dev = screen->nvws->channel->device; - static char buffer[128]; - - snprintf(buffer, sizeof(buffer), "NV%02X", dev->chipset); - return buffer; -} - -static const char * -nv30_screen_get_vendor(struct pipe_screen *pscreen) -{ - return "nouveau"; -} - static int nv30_screen_get_param(struct pipe_screen *pscreen, int param) { @@ -61,6 +46,10 @@ nv30_screen_get_param(struct pipe_screen *pscreen, int param) return 1; case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS: return 0; + case PIPE_CAP_TGSI_CONT_SUPPORTED: + return 0; + case PIPE_CAP_BLEND_EQUATION_SEPARATE: + return 0; case NOUVEAU_CAP_HW_VTXBUF: case NOUVEAU_CAP_HW_IDXBUF: return 1; @@ -100,7 +89,15 @@ nv30_screen_surface_format_supported(struct pipe_screen *pscreen, switch (format) { case PIPE_FORMAT_A8R8G8B8_UNORM: case PIPE_FORMAT_R5G6B5_UNORM: + return TRUE; + default: + break; + } + } else + if (tex_usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL) { + switch (format) { case PIPE_FORMAT_Z24S8_UNORM: + case PIPE_FORMAT_Z24X8_UNORM: case PIPE_FORMAT_Z16_UNORM: return TRUE; default: @@ -139,45 +136,57 @@ static void nv30_screen_destroy(struct pipe_screen *pscreen) { struct nv30_screen *screen = nv30_screen(pscreen); - struct nouveau_winsys *nvws = screen->nvws; - nvws->res_free(&screen->vp_exec_heap); - nvws->res_free(&screen->vp_data_heap); - nvws->res_free(&screen->query_heap); - nvws->notifier_free(&screen->query); - nvws->notifier_free(&screen->sync); - nvws->grobj_free(&screen->rankine); + nouveau_resource_free(&screen->vp_exec_heap); + nouveau_resource_free(&screen->vp_data_heap); + nouveau_resource_free(&screen->query_heap); + nouveau_notifier_free(&screen->query); + nouveau_notifier_free(&screen->sync); + nouveau_grobj_free(&screen->rankine); FREE(pscreen); } struct pipe_screen * -nv30_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) +nv30_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) { struct nv30_screen *screen = CALLOC_STRUCT(nv30_screen); + struct nouveau_channel *chan; + struct pipe_screen *pscreen; struct nouveau_stateobj *so; unsigned rankine_class = 0; - unsigned chipset = nvws->channel->device->chipset; int ret, i; if (!screen) return NULL; - screen->nvws = nvws; + pscreen = &screen->base.base; - /* 2D engine setup */ - screen->eng2d = nv04_surface_2d_init(nvws); - screen->eng2d->buf = nv30_surface_buffer; + ret = nouveau_screen_init(&screen->base, dev); + if (ret) { + nv30_screen_destroy(pscreen); + return NULL; + } + chan = screen->base.channel; + + pscreen->winsys = ws; + pscreen->destroy = nv30_screen_destroy; + pscreen->get_param = nv30_screen_get_param; + pscreen->get_paramf = nv30_screen_get_paramf; + pscreen->is_format_supported = nv30_screen_surface_format_supported; + + nv30_screen_init_miptree_functions(pscreen); + nv30_screen_init_transfer_functions(pscreen); /* 3D object */ - switch (chipset & 0xf0) { + switch (dev->chipset & 0xf0) { case 0x30: - if (NV30TCL_CHIPSET_3X_MASK & (1 << (chipset & 0x0f))) + if (NV30TCL_CHIPSET_3X_MASK & (1 << (dev->chipset & 0x0f))) rankine_class = 0x0397; else - if (NV34TCL_CHIPSET_3X_MASK & (1 << (chipset & 0x0f))) + if (NV34TCL_CHIPSET_3X_MASK & (1 << (dev->chipset & 0x0f))) rankine_class = 0x0697; else - if (NV35TCL_CHIPSET_3X_MASK & (1 << (chipset & 0x0f))) + if (NV35TCL_CHIPSET_3X_MASK & (1 << (dev->chipset & 0x0f))) rankine_class = 0x0497; break; default: @@ -185,43 +194,49 @@ nv30_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) } if (!rankine_class) { - NOUVEAU_ERR("Unknown nv3x chipset: nv%02x\n", chipset); + NOUVEAU_ERR("Unknown nv3x chipset: nv%02x\n", dev->chipset); return NULL; } - ret = nvws->grobj_alloc(nvws, rankine_class, &screen->rankine); + ret = nouveau_grobj_alloc(chan, 0xbeef3097, rankine_class, + &screen->rankine); if (ret) { NOUVEAU_ERR("Error creating 3D object: %d\n", ret); return FALSE; } + BIND_RING(chan, screen->rankine, 7); + + /* 2D engine setup */ + screen->eng2d = nv04_surface_2d_init(&screen->base); + screen->eng2d->buf = nv30_surface_buffer; /* Notifier for sync purposes */ - ret = nvws->notifier_alloc(nvws, 1, &screen->sync); + ret = nouveau_notifier_alloc(chan, 0xbeef0301, 1, &screen->sync); if (ret) { NOUVEAU_ERR("Error creating notifier object: %d\n", ret); - nv30_screen_destroy(&screen->pipe); + nv30_screen_destroy(pscreen); return NULL; } /* Query objects */ - ret = nvws->notifier_alloc(nvws, 32, &screen->query); + ret = nouveau_notifier_alloc(chan, 0xbeef0302, 32, &screen->query); if (ret) { NOUVEAU_ERR("Error initialising query objects: %d\n", ret); - nv30_screen_destroy(&screen->pipe); + nv30_screen_destroy(pscreen); return NULL; } - ret = nvws->res_init(&screen->query_heap, 0, 32); + ret = nouveau_resource_init(&screen->query_heap, 0, 32); if (ret) { NOUVEAU_ERR("Error initialising query object heap: %d\n", ret); - nv30_screen_destroy(&screen->pipe); + nv30_screen_destroy(pscreen); return NULL; } /* Vtxprog resources */ - if (nvws->res_init(&screen->vp_exec_heap, 0, 256) || - nvws->res_init(&screen->vp_data_heap, 0, 256)) { - nv30_screen_destroy(&screen->pipe); + if (nouveau_resource_init(&screen->vp_exec_heap, 0, 256) || + nouveau_resource_init(&screen->vp_data_heap, 0, 256)) { + nv30_screen_destroy(pscreen); return NULL; } @@ -230,23 +245,23 @@ nv30_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) so_method(so, screen->rankine, NV34TCL_DMA_NOTIFY, 1); so_data (so, screen->sync->handle); so_method(so, screen->rankine, NV34TCL_DMA_TEXTURE0, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->gart->handle); + so_data (so, chan->vram->handle); + so_data (so, chan->gart->handle); so_method(so, screen->rankine, NV34TCL_DMA_COLOR1, 1); - so_data (so, nvws->channel->vram->handle); + so_data (so, chan->vram->handle); so_method(so, screen->rankine, NV34TCL_DMA_COLOR0, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->vram->handle); + so_data (so, chan->vram->handle); + so_data (so, chan->vram->handle); so_method(so, screen->rankine, NV34TCL_DMA_VTXBUF0, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->gart->handle); + so_data (so, chan->vram->handle); + so_data (so, chan->gart->handle); /* so_method(so, screen->rankine, NV34TCL_DMA_FENCE, 2); so_data (so, 0); so_data (so, screen->query->handle);*/ so_method(so, screen->rankine, NV34TCL_DMA_IN_MEMORY7, 1); - so_data (so, nvws->channel->vram->handle); + so_data (so, chan->vram->handle); so_method(so, screen->rankine, NV34TCL_DMA_IN_MEMORY8, 1); - so_data (so, nvws->channel->vram->handle); + so_data (so, chan->vram->handle); for (i=1; i<8; i++) { so_method(so, screen->rankine, NV34TCL_VIEWPORT_CLIP_HORIZ(i), 1); @@ -301,23 +316,9 @@ nv30_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) so_method(so, screen->rankine, 0x1e94, 1); so_data (so, 0x13); - so_emit(nvws, so); + so_emit(chan, so); so_ref(NULL, &so); - nvws->push_flush(nvws, 0, NULL); - - screen->pipe.winsys = ws; - screen->pipe.destroy = nv30_screen_destroy; - - screen->pipe.get_name = nv30_screen_get_name; - screen->pipe.get_vendor = nv30_screen_get_vendor; - screen->pipe.get_param = nv30_screen_get_param; - screen->pipe.get_paramf = nv30_screen_get_paramf; - - screen->pipe.is_format_supported = nv30_screen_surface_format_supported; - - nv30_screen_init_miptree_functions(&screen->pipe); - nv30_screen_init_transfer_functions(&screen->pipe); - u_simple_screen_init(&screen->pipe); + nouveau_pushbuf_flush(chan, 0); - return &screen->pipe; + return pscreen; } diff --git a/src/gallium/drivers/nv30/nv30_screen.h b/src/gallium/drivers/nv30/nv30_screen.h index 8e368839759..5fbd998b539 100644 --- a/src/gallium/drivers/nv30/nv30_screen.h +++ b/src/gallium/drivers/nv30/nv30_screen.h @@ -1,11 +1,12 @@ #ifndef __NV30_SCREEN_H__ #define __NV30_SCREEN_H__ -#include "pipe/p_screen.h" +#include "nouveau/nouveau_screen.h" + #include "nv04/nv04_surface_2d.h" struct nv30_screen { - struct pipe_screen pipe; + struct nouveau_screen base; struct nouveau_winsys *nvws; diff --git a/src/gallium/drivers/nv30/nv30_state_emit.c b/src/gallium/drivers/nv30/nv30_state_emit.c index c18be20a327..621b8846c8e 100644 --- a/src/gallium/drivers/nv30/nv30_state_emit.c +++ b/src/gallium/drivers/nv30/nv30_state_emit.c @@ -38,6 +38,7 @@ nv30_state_do_validate(struct nv30_context *nv30, void nv30_state_emit(struct nv30_context *nv30) { + struct nouveau_channel *chan = nv30->screen->base.channel; struct nv30_state *state = &nv30->state; struct nv30_screen *screen = nv30->screen; unsigned i, samplers; @@ -57,23 +58,23 @@ nv30_state_emit(struct nv30_context *nv30) continue; so_ref (state->hw[i], &nv30->screen->state[i]); if (state->hw[i]) - so_emit(nv30->nvws, nv30->screen->state[i]); + so_emit(chan, nv30->screen->state[i]); states &= ~(1ULL << i); } state->dirty = 0; - so_emit_reloc_markers(nv30->nvws, state->hw[NV30_STATE_FB]); + so_emit_reloc_markers(chan, state->hw[NV30_STATE_FB]); for (i = 0, samplers = state->fp_samplers; i < 16 && samplers; i++) { if (!(samplers & (1 << i))) continue; - so_emit_reloc_markers(nv30->nvws, + so_emit_reloc_markers(chan, state->hw[NV30_STATE_FRAGTEX0+i]); samplers &= ~(1ULL << i); } - so_emit_reloc_markers(nv30->nvws, state->hw[NV30_STATE_FRAGPROG]); + so_emit_reloc_markers(chan, state->hw[NV30_STATE_FRAGPROG]); if (state->hw[NV30_STATE_VTXBUF] /*&& nv30->render_mode == HW*/) - so_emit_reloc_markers(nv30->nvws, state->hw[NV30_STATE_VTXBUF]); + so_emit_reloc_markers(chan, state->hw[NV30_STATE_VTXBUF]); } boolean diff --git a/src/gallium/drivers/nv30/nv30_state_fb.c b/src/gallium/drivers/nv30/nv30_state_fb.c index fdc1cade905..44b6a74715a 100644 --- a/src/gallium/drivers/nv30/nv30_state_fb.c +++ b/src/gallium/drivers/nv30/nv30_state_fb.c @@ -5,6 +5,8 @@ static boolean nv30_state_framebuffer_validate(struct nv30_context *nv30) { struct pipe_framebuffer_state *fb = &nv30->framebuffer; + struct nouveau_channel *chan = nv30->screen->base.channel; + struct nouveau_grobj *rankine = nv30->screen->rankine; struct nv04_surface *rt[2], *zeta = NULL; uint32_t rt_enable, rt_format; int i, colour_format = 0, zeta_format = 0; @@ -63,6 +65,7 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30) rt_format |= NV34TCL_RT_FORMAT_ZETA_Z16; break; case PIPE_FORMAT_Z24S8_UNORM: + case PIPE_FORMAT_Z24X8_UNORM: case 0: rt_format |= NV34TCL_RT_FORMAT_ZETA_Z24S8; break; @@ -79,56 +82,53 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30) } nv30mt = (struct nv30_miptree *)rt[0]->base.texture; - so_method(so, nv30->screen->rankine, NV34TCL_DMA_COLOR0, 1); - so_reloc (so, nv30mt->buffer, 0, rt_flags | NOUVEAU_BO_OR, - nv30->nvws->channel->vram->handle, - nv30->nvws->channel->gart->handle); - so_method(so, nv30->screen->rankine, NV34TCL_COLOR0_PITCH, 2); + so_method(so, rankine, NV34TCL_DMA_COLOR0, 1); + so_reloc (so, nouveau_bo(nv30mt->buffer), 0, rt_flags | NOUVEAU_BO_OR, + chan->vram->handle, chan->gart->handle); + so_method(so, rankine, NV34TCL_COLOR0_PITCH, 2); so_data (so, pitch); - so_reloc (so, nv30mt->buffer, rt[0]->base.offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); + so_reloc (so, nouveau_bo(nv30mt->buffer), rt[0]->base.offset, + rt_flags | NOUVEAU_BO_LOW, 0, 0); } if (rt_enable & NV34TCL_RT_ENABLE_COLOR1) { nv30mt = (struct nv30_miptree *)rt[1]->base.texture; - so_method(so, nv30->screen->rankine, NV34TCL_DMA_COLOR1, 1); - so_reloc (so, nv30mt->buffer, 0, rt_flags | NOUVEAU_BO_OR, - nv30->nvws->channel->vram->handle, - nv30->nvws->channel->gart->handle); - so_method(so, nv30->screen->rankine, NV34TCL_COLOR1_OFFSET, 2); - so_reloc (so, nv30mt->buffer, rt[1]->base.offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); + so_method(so, rankine, NV34TCL_DMA_COLOR1, 1); + so_reloc (so, nouveau_bo(nv30mt->buffer), 0, rt_flags | NOUVEAU_BO_OR, + chan->vram->handle, chan->gart->handle); + so_method(so, rankine, NV34TCL_COLOR1_OFFSET, 2); + so_reloc (so, nouveau_bo(nv30mt->buffer), rt[1]->base.offset, + rt_flags | NOUVEAU_BO_LOW, 0, 0); so_data (so, rt[1]->pitch); } if (zeta_format) { nv30mt = (struct nv30_miptree *)zeta->base.texture; - so_method(so, nv30->screen->rankine, NV34TCL_DMA_ZETA, 1); - so_reloc (so, nv30mt->buffer, 0, rt_flags | NOUVEAU_BO_OR, - nv30->nvws->channel->vram->handle, - nv30->nvws->channel->gart->handle); - so_method(so, nv30->screen->rankine, NV34TCL_ZETA_OFFSET, 1); - so_reloc (so, nv30mt->buffer, zeta->base.offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); + so_method(so, rankine, NV34TCL_DMA_ZETA, 1); + so_reloc (so, nouveau_bo(nv30mt->buffer), 0, rt_flags | NOUVEAU_BO_OR, + chan->vram->handle, chan->gart->handle); + so_method(so, rankine, NV34TCL_ZETA_OFFSET, 1); + so_reloc (so, nouveau_bo(nv30mt->buffer), zeta->base.offset, + rt_flags | NOUVEAU_BO_LOW, 0, 0); /* TODO: allocate LMA depth buffer */ } - so_method(so, nv30->screen->rankine, NV34TCL_RT_ENABLE, 1); + so_method(so, rankine, NV34TCL_RT_ENABLE, 1); so_data (so, rt_enable); - so_method(so, nv30->screen->rankine, NV34TCL_RT_HORIZ, 3); + so_method(so, rankine, NV34TCL_RT_HORIZ, 3); so_data (so, (w << 16) | 0); so_data (so, (h << 16) | 0); so_data (so, rt_format); - so_method(so, nv30->screen->rankine, NV34TCL_VIEWPORT_HORIZ, 2); + so_method(so, rankine, NV34TCL_VIEWPORT_HORIZ, 2); so_data (so, (w << 16) | 0); so_data (so, (h << 16) | 0); - so_method(so, nv30->screen->rankine, NV34TCL_VIEWPORT_CLIP_HORIZ(0), 2); + so_method(so, rankine, NV34TCL_VIEWPORT_CLIP_HORIZ(0), 2); so_data (so, ((w - 1) << 16) | 0); so_data (so, ((h - 1) << 16) | 0); - so_method(so, nv30->screen->rankine, 0x1d88, 1); + so_method(so, rankine, 0x1d88, 1); so_data (so, (1 << 12) | h); /* Wonder why this is needed, context should all be set to zero on init */ - so_method(so, nv30->screen->rankine, NV34TCL_VIEWPORT_TX_ORIGIN, 1); + so_method(so, rankine, NV34TCL_VIEWPORT_TX_ORIGIN, 1); so_data (so, 0); so_ref(so, &nv30->state.hw[NV30_STATE_FB]); diff --git a/src/gallium/drivers/nv30/nv30_vbo.c b/src/gallium/drivers/nv30/nv30_vbo.c index 990a8763824..189656ec817 100644 --- a/src/gallium/drivers/nv30/nv30_vbo.c +++ b/src/gallium/drivers/nv30/nv30_vbo.c @@ -1,5 +1,6 @@ #include "pipe/p_context.h" #include "pipe/p_state.h" +#include "pipe/p_inlines.h" #include "nv30_context.h" #include "nv30_state.h" @@ -70,7 +71,7 @@ static boolean nv30_vbo_set_idxbuf(struct nv30_context *nv30, struct pipe_buffer *ib, unsigned ib_size) { - struct pipe_screen *pscreen = &nv30->screen->pipe; + struct pipe_screen *pscreen = &nv30->screen->base.base; unsigned type; if (!ib) { @@ -108,7 +109,7 @@ nv30_vbo_static_attrib(struct nv30_context *nv30, struct nouveau_stateobj *so, int attrib, struct pipe_vertex_element *ve, struct pipe_vertex_buffer *vb) { - struct pipe_winsys *ws = nv30->pipe.winsys; + struct pipe_screen *pscreen = nv30->pipe.screen; struct nouveau_grobj *rankine = nv30->screen->rankine; unsigned type, ncomp; void *map; @@ -116,7 +117,7 @@ nv30_vbo_static_attrib(struct nv30_context *nv30, struct nouveau_stateobj *so, if (nv30_vbo_format_to_hw(ve->src_format, &type, &ncomp)) return FALSE; - map = ws->buffer_map(ws, vb->buffer, PIPE_BUFFER_USAGE_CPU_READ); + map = pipe_buffer_map(pscreen, vb->buffer, PIPE_BUFFER_USAGE_CPU_READ); map += vb->buffer_offset + ve->src_offset; switch (type) { @@ -148,18 +149,17 @@ nv30_vbo_static_attrib(struct nv30_context *nv30, struct nouveau_stateobj *so, so_data (so, fui(v[0])); break; default: - ws->buffer_unmap(ws, vb->buffer); + pipe_buffer_unmap(pscreen, vb->buffer); return FALSE; } } break; default: - ws->buffer_unmap(ws, vb->buffer); + pipe_buffer_unmap(pscreen, vb->buffer); return FALSE; } - ws->buffer_unmap(ws, vb->buffer); - + pipe_buffer_unmap(pscreen, vb->buffer); return TRUE; } @@ -168,7 +168,7 @@ nv30_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { struct nv30_context *nv30 = nv30_context(pipe); - struct nouveau_channel *chan = nv30->nvws->channel; + struct nouveau_channel *chan = nv30->screen->base.channel; unsigned restart = 0; nv30_vbo_set_idxbuf(nv30, NULL, 0); @@ -228,7 +228,7 @@ static INLINE void nv30_draw_elements_u08(struct nv30_context *nv30, void *ib, unsigned mode, unsigned start, unsigned count) { - struct nouveau_channel *chan = nv30->nvws->channel; + struct nouveau_channel *chan = nv30->screen->base.channel; while (count) { uint8_t *elts = (uint8_t *)ib + start; @@ -277,7 +277,7 @@ static INLINE void nv30_draw_elements_u16(struct nv30_context *nv30, void *ib, unsigned mode, unsigned start, unsigned count) { - struct nouveau_channel *chan = nv30->nvws->channel; + struct nouveau_channel *chan = nv30->screen->base.channel; while (count) { uint16_t *elts = (uint16_t *)ib + start; @@ -326,7 +326,7 @@ static INLINE void nv30_draw_elements_u32(struct nv30_context *nv30, void *ib, unsigned mode, unsigned start, unsigned count) { - struct nouveau_channel *chan = nv30->nvws->channel; + struct nouveau_channel *chan = nv30->screen->base.channel; while (count) { uint32_t *elts = (uint32_t *)ib + start; @@ -368,10 +368,10 @@ nv30_draw_elements_inline(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { struct nv30_context *nv30 = nv30_context(pipe); - struct pipe_winsys *ws = pipe->winsys; + struct pipe_screen *pscreen = pipe->screen; void *map; - map = ws->buffer_map(ws, ib, PIPE_BUFFER_USAGE_CPU_READ); + map = pipe_buffer_map(pscreen, ib, PIPE_BUFFER_USAGE_CPU_READ); if (!ib) { NOUVEAU_ERR("failed mapping ib\n"); return FALSE; @@ -392,7 +392,7 @@ nv30_draw_elements_inline(struct pipe_context *pipe, break; } - ws->buffer_unmap(ws, ib); + pipe_buffer_unmap(pscreen, ib); return TRUE; } @@ -401,7 +401,7 @@ nv30_draw_elements_vbo(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { struct nv30_context *nv30 = nv30_context(pipe); - struct nouveau_channel *chan = nv30->nvws->channel; + struct nouveau_channel *chan = nv30->screen->base.channel; unsigned restart = 0; while (count) { @@ -521,18 +521,20 @@ nv30_vbo_validate(struct nv30_context *nv30) return FALSE; } - so_reloc(vtxbuf, vb->buffer, vb->buffer_offset + ve->src_offset, - vb_flags | NOUVEAU_BO_LOW | NOUVEAU_BO_OR, - 0, NV34TCL_VTXBUF_ADDRESS_DMA1); + so_reloc(vtxbuf, nouveau_bo(vb->buffer), vb->buffer_offset + + ve->src_offset, vb_flags | NOUVEAU_BO_LOW | + NOUVEAU_BO_OR, 0, NV34TCL_VTXBUF_ADDRESS_DMA1); so_data (vtxfmt, ((vb->stride << NV34TCL_VTXFMT_STRIDE_SHIFT) | (ncomp << NV34TCL_VTXFMT_SIZE_SHIFT) | type)); } if (ib) { + struct nouveau_bo *bo = nouveau_bo(ib); + so_method(vtxbuf, rankine, NV34TCL_IDXBUF_ADDRESS, 2); - so_reloc (vtxbuf, ib, 0, vb_flags | NOUVEAU_BO_LOW, 0, 0); - so_reloc (vtxbuf, ib, ib_format, vb_flags | NOUVEAU_BO_OR, - 0, NV34TCL_IDXBUF_FORMAT_DMA1); + so_reloc (vtxbuf, bo, 0, vb_flags | NOUVEAU_BO_LOW, 0, 0); + so_reloc (vtxbuf, bo, ib_format, vb_flags | NOUVEAU_BO_OR, + 0, NV34TCL_IDXBUF_FORMAT_DMA1); } so_method(vtxbuf, rankine, 0x1710, 1); diff --git a/src/gallium/drivers/nv30/nv30_vertprog.c b/src/gallium/drivers/nv30/nv30_vertprog.c index eaf543b8f79..14a5c0260d0 100644 --- a/src/gallium/drivers/nv30/nv30_vertprog.c +++ b/src/gallium/drivers/nv30/nv30_vertprog.c @@ -1,6 +1,7 @@ #include "pipe/p_context.h" #include "pipe/p_defines.h" #include "pipe/p_state.h" +#include "pipe/p_inlines.h" #include "pipe/p_shader_tokens.h" #include "tgsi/tgsi_parse.h" @@ -616,10 +617,10 @@ nv30_vertprog_translate(struct nv30_context *nv30, assert(imm->Immediate.NrTokens == 4 + 1); vpc->imm[vpc->nr_imm++] = constant(vpc, -1, - imm->u.ImmediateFloat32[0].Float, - imm->u.ImmediateFloat32[1].Float, - imm->u.ImmediateFloat32[2].Float, - imm->u.ImmediateFloat32[3].Float); + imm->u[0].Float, + imm->u[1].Float, + imm->u[2].Float, + imm->u[3].Float); } break; case TGSI_TOKEN_TYPE_INSTRUCTION: @@ -645,8 +646,7 @@ out_err: static boolean nv30_vertprog_validate(struct nv30_context *nv30) { - struct nouveau_winsys *nvws = nv30->nvws; - struct pipe_winsys *ws = nv30->pipe.winsys; + struct pipe_screen *pscreen = nv30->pipe.screen; struct nouveau_grobj *rankine = nv30->screen->rankine; struct nv30_vertex_program *vp; struct pipe_buffer *constbuf; @@ -669,15 +669,15 @@ nv30_vertprog_validate(struct nv30_context *nv30) struct nouveau_stateobj *so; uint vplen = vp->nr_insns; - if (nvws->res_alloc(heap, vplen, vp, &vp->exec)) { + if (nouveau_resource_alloc(heap, vplen, vp, &vp->exec)) { while (heap->next && heap->size < vplen) { struct nv30_vertex_program *evict; evict = heap->next->priv; - nvws->res_free(&evict->exec); + nouveau_resource_free(&evict->exec); } - if (nvws->res_alloc(heap, vplen, vp, &vp->exec)) + if (nouveau_resource_alloc(heap, vplen, vp, &vp->exec)) assert(0); } @@ -694,15 +694,16 @@ nv30_vertprog_validate(struct nv30_context *nv30) if (vp->nr_consts && !vp->data) { struct nouveau_resource *heap = nv30->screen->vp_data_heap; - if (nvws->res_alloc(heap, vp->nr_consts, vp, &vp->data)) { + if (nouveau_resource_alloc(heap, vp->nr_consts, vp, &vp->data)) { while (heap->next && heap->size < vp->nr_consts) { struct nv30_vertex_program *evict; evict = heap->next->priv; - nvws->res_free(&evict->data); + nouveau_resource_free(&evict->data); } - if (nvws->res_alloc(heap, vp->nr_consts, vp, &vp->data)) + if (nouveau_resource_alloc(heap, vp->nr_consts, vp, + &vp->data)) assert(0); } @@ -750,8 +751,8 @@ nv30_vertprog_validate(struct nv30_context *nv30) float *map = NULL; if (constbuf) { - map = ws->buffer_map(ws, constbuf, - PIPE_BUFFER_USAGE_CPU_READ); + map = pipe_buffer_map(pscreen, constbuf, + PIPE_BUFFER_USAGE_CPU_READ); } for (i = 0; i < vp->nr_consts; i++) { @@ -771,9 +772,8 @@ nv30_vertprog_validate(struct nv30_context *nv30) OUT_RINGp ((uint32_t *)vpd->value, 4); } - if (constbuf) { - ws->buffer_unmap(ws, constbuf); - } + if (constbuf) + pipe_buffer_unmap(pscreen, constbuf); } /* Upload vtxprog */ @@ -804,8 +804,6 @@ nv30_vertprog_validate(struct nv30_context *nv30) void nv30_vertprog_destroy(struct nv30_context *nv30, struct nv30_vertex_program *vp) { - struct nouveau_winsys *nvws = nv30->screen->nvws; - vp->translated = FALSE; if (vp->nr_insns) { @@ -820,9 +818,9 @@ nv30_vertprog_destroy(struct nv30_context *nv30, struct nv30_vertex_program *vp) vp->nr_consts = 0; } - nvws->res_free(&vp->exec); + nouveau_resource_free(&vp->exec); vp->exec_start = 0; - nvws->res_free(&vp->data); + nouveau_resource_free(&vp->data); vp->data_start = 0; vp->data_start_min = 0; diff --git a/src/gallium/drivers/nv40/nv40_draw.c b/src/gallium/drivers/nv40/nv40_draw.c index c83ff91d7e3..b2f19ecb699 100644 --- a/src/gallium/drivers/nv40/nv40_draw.c +++ b/src/gallium/drivers/nv40/nv40_draw.c @@ -1,4 +1,5 @@ #include "pipe/p_shader_tokens.h" +#include "pipe/p_inlines.h" #include "util/u_pack_color.h" @@ -81,7 +82,7 @@ nv40_render_prim(struct draw_stage *stage, struct prim_header *prim, { struct nv40_render_stage *rs = nv40_render_stage(stage); struct nv40_context *nv40 = rs->nv40; - struct nouveau_pushbuf *pb = nv40->nvws->channel->pushbuf; + struct nouveau_pushbuf *pb = nv40->screen->base.channel->pushbuf; unsigned i; /* Ensure there's room for 4xfloat32 + potentially 3 begin/end */ @@ -231,7 +232,7 @@ nv40_draw_elements_swtnl(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { struct nv40_context *nv40 = nv40_context(pipe); - struct pipe_winsys *ws = pipe->winsys; + struct pipe_screen *pscreen = pipe->screen; unsigned i; void *map; @@ -241,13 +242,14 @@ nv40_draw_elements_swtnl(struct pipe_context *pipe, nv40_state_emit(nv40); for (i = 0; i < nv40->vtxbuf_nr; i++) { - map = ws->buffer_map(ws, nv40->vtxbuf[i].buffer, + map = pipe_buffer_map(pscreen, nv40->vtxbuf[i].buffer, PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_vertex_buffer(nv40->draw, i, map); } if (idxbuf) { - map = ws->buffer_map(ws, idxbuf, PIPE_BUFFER_USAGE_CPU_READ); + map = pipe_buffer_map(pscreen, idxbuf, + PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_element_buffer(nv40->draw, idxbuf_size, map); } else { draw_set_mapped_element_buffer(nv40->draw, 0, NULL); @@ -256,21 +258,22 @@ nv40_draw_elements_swtnl(struct pipe_context *pipe, if (nv40->constbuf[PIPE_SHADER_VERTEX]) { const unsigned nr = nv40->constbuf_nr[PIPE_SHADER_VERTEX]; - map = ws->buffer_map(ws, nv40->constbuf[PIPE_SHADER_VERTEX], - PIPE_BUFFER_USAGE_CPU_READ); + map = pipe_buffer_map(pscreen, + nv40->constbuf[PIPE_SHADER_VERTEX], + PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_constant_buffer(nv40->draw, map, nr); } draw_arrays(nv40->draw, mode, start, count); for (i = 0; i < nv40->vtxbuf_nr; i++) - ws->buffer_unmap(ws, nv40->vtxbuf[i].buffer); + pipe_buffer_unmap(pscreen, nv40->vtxbuf[i].buffer); if (idxbuf) - ws->buffer_unmap(ws, idxbuf); + pipe_buffer_unmap(pscreen, idxbuf); if (nv40->constbuf[PIPE_SHADER_VERTEX]) - ws->buffer_unmap(ws, nv40->constbuf[PIPE_SHADER_VERTEX]); + pipe_buffer_unmap(pscreen, nv40->constbuf[PIPE_SHADER_VERTEX]); draw_flush(nv40->draw); pipe->flush(pipe, 0, NULL); diff --git a/src/gallium/drivers/nv40/nv40_fragprog.c b/src/gallium/drivers/nv40/nv40_fragprog.c index 16e40889ec3..32d9ed1a7f8 100644 --- a/src/gallium/drivers/nv40/nv40_fragprog.c +++ b/src/gallium/drivers/nv40/nv40_fragprog.c @@ -1,6 +1,7 @@ #include "pipe/p_context.h" #include "pipe/p_defines.h" #include "pipe/p_state.h" +#include "pipe/p_inlines.h" #include "pipe/p_shader_tokens.h" #include "tgsi/tgsi_parse.h" @@ -789,10 +790,10 @@ nv40_fragprog_prepare(struct nv40_fpc *fpc) assert(imm->Immediate.DataType == TGSI_IMM_FLOAT32); assert(fpc->nr_imm < MAX_IMM); - vals[0] = imm->u.ImmediateFloat32[0].Float; - vals[1] = imm->u.ImmediateFloat32[1].Float; - vals[2] = imm->u.ImmediateFloat32[2].Float; - vals[3] = imm->u.ImmediateFloat32[3].Float; + vals[0] = imm->u[0].Float; + vals[1] = imm->u[1].Float; + vals[2] = imm->u[2].Float; + vals[3] = imm->u[3].Float; fpc->imm[fpc->nr_imm++] = constant(fpc, -1, vals); } break; @@ -881,12 +882,12 @@ static void nv40_fragprog_upload(struct nv40_context *nv40, struct nv40_fragment_program *fp) { - struct pipe_winsys *ws = nv40->pipe.winsys; + struct pipe_screen *pscreen = nv40->pipe.screen; const uint32_t le = 1; uint32_t *map; int i; - map = ws->buffer_map(ws, fp->buffer, PIPE_BUFFER_USAGE_CPU_WRITE); + map = pipe_buffer_map(pscreen, fp->buffer, PIPE_BUFFER_USAGE_CPU_WRITE); #if 0 for (i = 0; i < fp->insn_len; i++) { @@ -908,7 +909,7 @@ nv40_fragprog_upload(struct nv40_context *nv40, } } - ws->buffer_unmap(ws, fp->buffer); + pipe_buffer_unmap(pscreen, fp->buffer); } static boolean @@ -917,8 +918,7 @@ nv40_fragprog_validate(struct nv40_context *nv40) struct nv40_fragment_program *fp = nv40->fragprog; struct pipe_buffer *constbuf = nv40->constbuf[PIPE_SHADER_FRAGMENT]; - struct pipe_screen *screen = nv40->pipe.screen; - struct pipe_winsys *ws = nv40->pipe.winsys; + struct pipe_screen *pscreen = nv40->pipe.screen; struct nouveau_stateobj *so; boolean new_consts = FALSE; int i; @@ -933,14 +933,15 @@ nv40_fragprog_validate(struct nv40_context *nv40) return FALSE; } - fp->buffer = screen->buffer_create(screen, 0x100, 0, fp->insn_len * 4); + fp->buffer = pscreen->buffer_create(pscreen, 0x100, 0, fp->insn_len * 4); nv40_fragprog_upload(nv40, fp); so = so_new(4, 1); so_method(so, nv40->screen->curie, NV40TCL_FP_ADDRESS, 1); - so_reloc (so, fp->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | - NOUVEAU_BO_RD | NOUVEAU_BO_LOW | NOUVEAU_BO_OR, - NV40TCL_FP_ADDRESS_DMA0, NV40TCL_FP_ADDRESS_DMA1); + so_reloc (so, nouveau_bo(fp->buffer), 0, NOUVEAU_BO_VRAM | + NOUVEAU_BO_GART | NOUVEAU_BO_RD | NOUVEAU_BO_LOW | + NOUVEAU_BO_OR, NV40TCL_FP_ADDRESS_DMA0, + NV40TCL_FP_ADDRESS_DMA1); so_method(so, nv40->screen->curie, NV40TCL_FP_CONTROL, 1); so_data (so, fp->fp_control); so_ref(so, &fp->so); @@ -950,7 +951,8 @@ update_constants: if (fp->nr_consts) { float *map; - map = ws->buffer_map(ws, constbuf, PIPE_BUFFER_USAGE_CPU_READ); + map = pipe_buffer_map(pscreen, constbuf, + PIPE_BUFFER_USAGE_CPU_READ); for (i = 0; i < fp->nr_consts; i++) { struct nv40_fragment_program_data *fpd = &fp->consts[i]; uint32_t *p = &fp->insn[fpd->offset]; @@ -961,7 +963,7 @@ update_constants: memcpy(p, cb, 4 * sizeof(float)); new_consts = TRUE; } - ws->buffer_unmap(ws, constbuf); + pipe_buffer_unmap(pscreen, constbuf); if (new_consts) nv40_fragprog_upload(nv40, fp); diff --git a/src/gallium/drivers/nv40/nv40_fragtex.c b/src/gallium/drivers/nv40/nv40_fragtex.c index eb3002dc053..f6cdf31dfee 100644 --- a/src/gallium/drivers/nv40/nv40_fragtex.c +++ b/src/gallium/drivers/nv40/nv40_fragtex.c @@ -62,6 +62,7 @@ nv40_fragtex_build(struct nv40_context *nv40, int unit) { struct nv40_sampler_state *ps = nv40->tex_sampler[unit]; struct nv40_miptree *nv40mt = nv40->tex_miptree[unit]; + struct nouveau_bo *bo = nouveau_bo(nv40mt->buffer); struct pipe_texture *pt = &nv40mt->base; struct nv40_texture_format *tf; struct nouveau_stateobj *so; @@ -108,9 +109,9 @@ nv40_fragtex_build(struct nv40_context *nv40, int unit) so = so_new(16, 2); so_method(so, nv40->screen->curie, NV40TCL_TEX_OFFSET(unit), 8); - so_reloc (so, nv40mt->buffer, 0, tex_flags | NOUVEAU_BO_LOW, 0, 0); - so_reloc (so, nv40mt->buffer, txf, tex_flags | NOUVEAU_BO_OR, - NV40TCL_TEX_FORMAT_DMA0, NV40TCL_TEX_FORMAT_DMA1); + so_reloc (so, bo, 0, tex_flags | NOUVEAU_BO_LOW, 0, 0); + so_reloc (so, bo, txf, tex_flags | NOUVEAU_BO_OR, + NV40TCL_TEX_FORMAT_DMA0, NV40TCL_TEX_FORMAT_DMA1); so_data (so, ps->wrap); so_data (so, NV40TCL_TEX_ENABLE_ENABLE | ps->en); so_data (so, txs); diff --git a/src/gallium/drivers/nv40/nv40_query.c b/src/gallium/drivers/nv40/nv40_query.c index 9b9a43f49df..7874aedd428 100644 --- a/src/gallium/drivers/nv40/nv40_query.c +++ b/src/gallium/drivers/nv40/nv40_query.c @@ -29,11 +29,10 @@ nv40_query_create(struct pipe_context *pipe, unsigned query_type) static void nv40_query_destroy(struct pipe_context *pipe, struct pipe_query *pq) { - struct nv40_context *nv40 = nv40_context(pipe); struct nv40_query *q = nv40_query(pq); if (q->object) - nv40->nvws->res_free(&q->object); + nouveau_resource_free(&q->object); FREE(q); } @@ -54,9 +53,9 @@ nv40_query_begin(struct pipe_context *pipe, struct pipe_query *pq) pipe->get_query_result(pipe, pq, 1, &tmp); } - if (nv40->nvws->res_alloc(nv40->screen->query_heap, 1, NULL, &q->object)) + if (nouveau_resource_alloc(nv40->screen->query_heap, 1, NULL, &q->object)) assert(0); - nv40->nvws->notifier_reset(nv40->screen->query, q->object->start); + nouveau_notifier_reset(nv40->screen->query, q->object->start); BEGIN_RING(curie, NV40TCL_QUERY_RESET, 1); OUT_RING (1); @@ -84,27 +83,27 @@ nv40_query_result(struct pipe_context *pipe, struct pipe_query *pq, { struct nv40_context *nv40 = nv40_context(pipe); struct nv40_query *q = nv40_query(pq); - struct nouveau_winsys *nvws = nv40->nvws; assert(q->object && q->type == PIPE_QUERY_OCCLUSION_COUNTER); if (!q->ready) { unsigned status; - status = nvws->notifier_status(nv40->screen->query, - q->object->start); + status = nouveau_notifier_status(nv40->screen->query, + q->object->start); if (status != NV_NOTIFY_STATE_STATUS_COMPLETED) { if (wait == FALSE) return FALSE; - nvws->notifier_wait(nv40->screen->query, q->object->start, - NV_NOTIFY_STATE_STATUS_COMPLETED, - 0); + nouveau_notifier_wait_status(nv40->screen->query, + q->object->start, + NV_NOTIFY_STATE_STATUS_COMPLETED, + 0); } - q->result = nvws->notifier_retval(nv40->screen->query, - q->object->start); + q->result = nouveau_notifier_return_val(nv40->screen->query, + q->object->start); q->ready = TRUE; - nvws->res_free(&q->object); + nouveau_resource_free(&q->object); } *result = q->result; diff --git a/src/gallium/drivers/nv40/nv40_screen.c b/src/gallium/drivers/nv40/nv40_screen.c index 0d4baefaea3..5d2a4216c5a 100644 --- a/src/gallium/drivers/nv40/nv40_screen.c +++ b/src/gallium/drivers/nv40/nv40_screen.c @@ -1,5 +1,4 @@ #include "pipe/p_screen.h" -#include "util/u_simple_screen.h" #include "nv40_context.h" #include "nv40_screen.h" @@ -8,23 +7,6 @@ #define NV4X_GRCLASS4497_CHIPSETS 0x00005450 #define NV6X_GRCLASS4497_CHIPSETS 0x00000088 -static const char * -nv40_screen_get_name(struct pipe_screen *pscreen) -{ - struct nv40_screen *screen = nv40_screen(pscreen); - struct nouveau_device *dev = screen->nvws->channel->device; - static char buffer[128]; - - snprintf(buffer, sizeof(buffer), "NV%02X", dev->chipset); - return buffer; -} - -static const char * -nv40_screen_get_vendor(struct pipe_screen *pscreen) -{ - return "nouveau"; -} - static int nv40_screen_get_param(struct pipe_screen *pscreen, int param) { @@ -62,6 +44,10 @@ nv40_screen_get_param(struct pipe_screen *pscreen, int param) return 1; case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS: return 0; /* We have 4 - but unsupported currently */ + case PIPE_CAP_TGSI_CONT_SUPPORTED: + return 0; + case PIPE_CAP_BLEND_EQUATION_SEPARATE: + return 1; case NOUVEAU_CAP_HW_VTXBUF: return 1; case NOUVEAU_CAP_HW_IDXBUF: @@ -104,7 +90,15 @@ nv40_screen_surface_format_supported(struct pipe_screen *pscreen, switch (format) { case PIPE_FORMAT_A8R8G8B8_UNORM: case PIPE_FORMAT_R5G6B5_UNORM: + return TRUE; + default: + break; + } + } else + if (tex_usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL) { + switch (format) { case PIPE_FORMAT_Z24S8_UNORM: + case PIPE_FORMAT_Z24X8_UNORM: case PIPE_FORMAT_Z16_UNORM: return TRUE; default: @@ -148,88 +142,107 @@ static void nv40_screen_destroy(struct pipe_screen *pscreen) { struct nv40_screen *screen = nv40_screen(pscreen); - struct nouveau_winsys *nvws = screen->nvws; - nvws->res_free(&screen->vp_exec_heap); - nvws->res_free(&screen->vp_data_heap); - nvws->res_free(&screen->query_heap); - nvws->notifier_free(&screen->query); - nvws->notifier_free(&screen->sync); - nvws->grobj_free(&screen->curie); + nouveau_resource_free(&screen->vp_exec_heap); + nouveau_resource_free(&screen->vp_data_heap); + nouveau_resource_free(&screen->query_heap); + nouveau_notifier_free(&screen->query); + nouveau_notifier_free(&screen->sync); + nouveau_grobj_free(&screen->curie); + + nouveau_screen_fini(&screen->base); FREE(pscreen); } struct pipe_screen * -nv40_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) +nv40_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) { struct nv40_screen *screen = CALLOC_STRUCT(nv40_screen); + struct nouveau_channel *chan; + struct pipe_screen *pscreen; struct nouveau_stateobj *so; unsigned curie_class = 0; - unsigned chipset = nvws->channel->device->chipset; int ret; if (!screen) return NULL; - screen->nvws = nvws; + pscreen = &screen->base.base; - /* 2D engine setup */ - screen->eng2d = nv04_surface_2d_init(nvws); - screen->eng2d->buf = nv40_surface_buffer; + ret = nouveau_screen_init(&screen->base, dev); + if (ret) { + nv40_screen_destroy(pscreen); + return NULL; + } + chan = screen->base.channel; + + pscreen->winsys = ws; + pscreen->destroy = nv40_screen_destroy; + pscreen->get_param = nv40_screen_get_param; + pscreen->get_paramf = nv40_screen_get_paramf; + pscreen->is_format_supported = nv40_screen_surface_format_supported; + + nv40_screen_init_miptree_functions(pscreen); + nv40_screen_init_transfer_functions(pscreen); /* 3D object */ - switch (chipset & 0xf0) { + switch (dev->chipset & 0xf0) { case 0x40: - if (NV4X_GRCLASS4097_CHIPSETS & (1 << (chipset & 0x0f))) + if (NV4X_GRCLASS4097_CHIPSETS & (1 << (dev->chipset & 0x0f))) curie_class = NV40TCL; else - if (NV4X_GRCLASS4497_CHIPSETS & (1 << (chipset & 0x0f))) + if (NV4X_GRCLASS4497_CHIPSETS & (1 << (dev->chipset & 0x0f))) curie_class = NV44TCL; break; case 0x60: - if (NV6X_GRCLASS4497_CHIPSETS & (1 << (chipset & 0x0f))) + if (NV6X_GRCLASS4497_CHIPSETS & (1 << (dev->chipset & 0x0f))) curie_class = NV44TCL; break; } if (!curie_class) { - NOUVEAU_ERR("Unknown nv4x chipset: nv%02x\n", chipset); + NOUVEAU_ERR("Unknown nv4x chipset: nv%02x\n", dev->chipset); return NULL; } - ret = nvws->grobj_alloc(nvws, curie_class, &screen->curie); + ret = nouveau_grobj_alloc(chan, 0xbeef3097, curie_class, &screen->curie); if (ret) { NOUVEAU_ERR("Error creating 3D object: %d\n", ret); return FALSE; } + BIND_RING(chan, screen->curie, 7); + + /* 2D engine setup */ + screen->eng2d = nv04_surface_2d_init(&screen->base); + screen->eng2d->buf = nv40_surface_buffer; /* Notifier for sync purposes */ - ret = nvws->notifier_alloc(nvws, 1, &screen->sync); + ret = nouveau_notifier_alloc(chan, 0xbeef0301, 1, &screen->sync); if (ret) { NOUVEAU_ERR("Error creating notifier object: %d\n", ret); - nv40_screen_destroy(&screen->pipe); + nv40_screen_destroy(pscreen); return NULL; } /* Query objects */ - ret = nvws->notifier_alloc(nvws, 32, &screen->query); + ret = nouveau_notifier_alloc(chan, 0xbeef0302, 32, &screen->query); if (ret) { NOUVEAU_ERR("Error initialising query objects: %d\n", ret); - nv40_screen_destroy(&screen->pipe); + nv40_screen_destroy(pscreen); return NULL; } - ret = nvws->res_init(&screen->query_heap, 0, 32); + nouveau_resource_init(&screen->query_heap, 0, 32); if (ret) { NOUVEAU_ERR("Error initialising query object heap: %d\n", ret); - nv40_screen_destroy(&screen->pipe); + nv40_screen_destroy(pscreen); return NULL; } /* Vtxprog resources */ - if (nvws->res_init(&screen->vp_exec_heap, 0, 512) || - nvws->res_init(&screen->vp_data_heap, 0, 256)) { - nv40_screen_destroy(&screen->pipe); + if (nouveau_resource_init(&screen->vp_exec_heap, 0, 512) || + nouveau_resource_init(&screen->vp_data_heap, 0, 256)) { + nv40_screen_destroy(pscreen); return NULL; } @@ -238,25 +251,25 @@ nv40_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) so_method(so, screen->curie, NV40TCL_DMA_NOTIFY, 1); so_data (so, screen->sync->handle); so_method(so, screen->curie, NV40TCL_DMA_TEXTURE0, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->gart->handle); + so_data (so, chan->vram->handle); + so_data (so, chan->gart->handle); so_method(so, screen->curie, NV40TCL_DMA_COLOR1, 1); - so_data (so, nvws->channel->vram->handle); + so_data (so, chan->vram->handle); so_method(so, screen->curie, NV40TCL_DMA_COLOR0, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->vram->handle); + so_data (so, chan->vram->handle); + so_data (so, chan->vram->handle); so_method(so, screen->curie, NV40TCL_DMA_VTXBUF0, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->gart->handle); + so_data (so, chan->vram->handle); + so_data (so, chan->gart->handle); so_method(so, screen->curie, NV40TCL_DMA_FENCE, 2); so_data (so, 0); so_data (so, screen->query->handle); so_method(so, screen->curie, NV40TCL_DMA_UNK01AC, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->vram->handle); + so_data (so, chan->vram->handle); + so_data (so, chan->vram->handle); so_method(so, screen->curie, NV40TCL_DMA_COLOR2, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->vram->handle); + so_data (so, chan->vram->handle); + so_data (so, chan->vram->handle); so_method(so, screen->curie, 0x1ea4, 3); so_data (so, 0x00000010); @@ -281,24 +294,10 @@ nv40_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) so_method(so, screen->curie, 0x1e94, 1); so_data (so, 0x00000001); - so_emit(nvws, so); + so_emit(chan, so); so_ref(NULL, &so); - nvws->push_flush(nvws, 0, NULL); - - screen->pipe.winsys = ws; - screen->pipe.destroy = nv40_screen_destroy; - - screen->pipe.get_name = nv40_screen_get_name; - screen->pipe.get_vendor = nv40_screen_get_vendor; - screen->pipe.get_param = nv40_screen_get_param; - screen->pipe.get_paramf = nv40_screen_get_paramf; - - screen->pipe.is_format_supported = nv40_screen_surface_format_supported; - - nv40_screen_init_miptree_functions(&screen->pipe); - nv40_screen_init_transfer_functions(&screen->pipe); - u_simple_screen_init(&screen->pipe); + nouveau_pushbuf_flush(chan, 0); - return &screen->pipe; + return pscreen; } diff --git a/src/gallium/drivers/nv40/nv40_screen.h b/src/gallium/drivers/nv40/nv40_screen.h index 7b503bd207d..57b4c8fc46c 100644 --- a/src/gallium/drivers/nv40/nv40_screen.h +++ b/src/gallium/drivers/nv40/nv40_screen.h @@ -1,11 +1,11 @@ #ifndef __NV40_SCREEN_H__ #define __NV40_SCREEN_H__ -#include "pipe/p_screen.h" +#include "nouveau/nouveau_screen.h" #include "nv04/nv04_surface_2d.h" struct nv40_screen { - struct pipe_screen pipe; + struct nouveau_screen base; struct nouveau_winsys *nvws; diff --git a/src/gallium/drivers/nv40/nv40_state_emit.c b/src/gallium/drivers/nv40/nv40_state_emit.c index 10aae298328..198692965dc 100644 --- a/src/gallium/drivers/nv40/nv40_state_emit.c +++ b/src/gallium/drivers/nv40/nv40_state_emit.c @@ -54,6 +54,7 @@ nv40_state_do_validate(struct nv40_context *nv40, void nv40_state_emit(struct nv40_context *nv40) { + struct nouveau_channel *chan = nv40->screen->base.channel; struct nv40_state *state = &nv40->state; struct nv40_screen *screen = nv40->screen; unsigned i, samplers; @@ -73,7 +74,7 @@ nv40_state_emit(struct nv40_context *nv40) continue; so_ref (state->hw[i], &nv40->screen->state[i]); if (state->hw[i]) - so_emit(nv40->nvws, nv40->screen->state[i]); + so_emit(chan, nv40->screen->state[i]); states &= ~(1ULL << i); } @@ -87,17 +88,17 @@ nv40_state_emit(struct nv40_context *nv40) state->dirty = 0; - so_emit_reloc_markers(nv40->nvws, state->hw[NV40_STATE_FB]); + so_emit_reloc_markers(chan, state->hw[NV40_STATE_FB]); for (i = 0, samplers = state->fp_samplers; i < 16 && samplers; i++) { if (!(samplers & (1 << i))) continue; - so_emit_reloc_markers(nv40->nvws, + so_emit_reloc_markers(chan, state->hw[NV40_STATE_FRAGTEX0+i]); samplers &= ~(1ULL << i); } - so_emit_reloc_markers(nv40->nvws, state->hw[NV40_STATE_FRAGPROG]); + so_emit_reloc_markers(chan, state->hw[NV40_STATE_FRAGPROG]); if (state->hw[NV40_STATE_VTXBUF] && nv40->render_mode == HW) - so_emit_reloc_markers(nv40->nvws, state->hw[NV40_STATE_VTXBUF]); + so_emit_reloc_markers(chan, state->hw[NV40_STATE_VTXBUF]); } boolean diff --git a/src/gallium/drivers/nv40/nv40_state_fb.c b/src/gallium/drivers/nv40/nv40_state_fb.c index be618a306bf..c2f739157ad 100644 --- a/src/gallium/drivers/nv40/nv40_state_fb.c +++ b/src/gallium/drivers/nv40/nv40_state_fb.c @@ -2,15 +2,19 @@ #include "nouveau/nouveau_util.h" static struct pipe_buffer * -nv40_surface_buffer(struct pipe_surface *surface) +nv40_do_surface_buffer(struct pipe_surface *surface) { struct nv40_miptree *mt = (struct nv40_miptree *)surface->texture; return mt->buffer; } +#define nv40_surface_buffer(ps) nouveau_bo(nv40_do_surface_buffer(ps)) + static boolean nv40_state_framebuffer_validate(struct nv40_context *nv40) { + struct nouveau_channel *chan = nv40->screen->base.channel; + struct nouveau_grobj *curie = nv40->screen->curie; struct pipe_framebuffer_state *fb = &nv40->framebuffer; struct nv04_surface *rt[4], *zeta; uint32_t rt_enable, rt_format; @@ -69,6 +73,7 @@ nv40_state_framebuffer_validate(struct nv40_context *nv40) rt_format |= NV40TCL_RT_FORMAT_ZETA_Z16; break; case PIPE_FORMAT_Z24S8_UNORM: + case PIPE_FORMAT_Z24X8_UNORM: case 0: rt_format |= NV40TCL_RT_FORMAT_ZETA_Z24S8; break; @@ -77,76 +82,80 @@ nv40_state_framebuffer_validate(struct nv40_context *nv40) } if (rt_enable & NV40TCL_RT_ENABLE_COLOR0) { - so_method(so, nv40->screen->curie, NV40TCL_DMA_COLOR0, 1); - so_reloc (so, nv40_surface_buffer(&rt[0]->base), 0, rt_flags | NOUVEAU_BO_OR, - nv40->nvws->channel->vram->handle, - nv40->nvws->channel->gart->handle); - so_method(so, nv40->screen->curie, NV40TCL_COLOR0_PITCH, 2); + so_method(so, curie, NV40TCL_DMA_COLOR0, 1); + so_reloc (so, nv40_surface_buffer(&rt[0]->base), 0, + rt_flags | NOUVEAU_BO_OR, + chan->vram->handle, chan->gart->handle); + so_method(so, curie, NV40TCL_COLOR0_PITCH, 2); so_data (so, rt[0]->pitch); - so_reloc (so, nv40_surface_buffer(&rt[0]->base), rt[0]->base.offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); + so_reloc (so, nv40_surface_buffer(&rt[0]->base), + rt[0]->base.offset, rt_flags | NOUVEAU_BO_LOW, + 0, 0); } if (rt_enable & NV40TCL_RT_ENABLE_COLOR1) { - so_method(so, nv40->screen->curie, NV40TCL_DMA_COLOR1, 1); - so_reloc (so, nv40_surface_buffer(&rt[1]->base), 0, rt_flags | NOUVEAU_BO_OR, - nv40->nvws->channel->vram->handle, - nv40->nvws->channel->gart->handle); - so_method(so, nv40->screen->curie, NV40TCL_COLOR1_OFFSET, 2); - so_reloc (so, nv40_surface_buffer(&rt[1]->base), rt[1]->base.offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); + so_method(so, curie, NV40TCL_DMA_COLOR1, 1); + so_reloc (so, nv40_surface_buffer(&rt[1]->base), 0, + rt_flags | NOUVEAU_BO_OR, + chan->vram->handle, chan->gart->handle); + so_method(so, curie, NV40TCL_COLOR1_OFFSET, 2); + so_reloc (so, nv40_surface_buffer(&rt[1]->base), + rt[1]->base.offset, rt_flags | NOUVEAU_BO_LOW, + 0, 0); so_data (so, rt[1]->pitch); } if (rt_enable & NV40TCL_RT_ENABLE_COLOR2) { - so_method(so, nv40->screen->curie, NV40TCL_DMA_COLOR2, 1); - so_reloc (so, nv40_surface_buffer(&rt[2]->base), 0, rt_flags | NOUVEAU_BO_OR, - nv40->nvws->channel->vram->handle, - nv40->nvws->channel->gart->handle); - so_method(so, nv40->screen->curie, NV40TCL_COLOR2_OFFSET, 1); - so_reloc (so, nv40_surface_buffer(&rt[2]->base), rt[2]->base.offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); - so_method(so, nv40->screen->curie, NV40TCL_COLOR2_PITCH, 1); + so_method(so, curie, NV40TCL_DMA_COLOR2, 1); + so_reloc (so, nv40_surface_buffer(&rt[2]->base), 0, + rt_flags | NOUVEAU_BO_OR, + chan->vram->handle, chan->gart->handle); + so_method(so, curie, NV40TCL_COLOR2_OFFSET, 1); + so_reloc (so, nv40_surface_buffer(&rt[2]->base), + rt[2]->base.offset, rt_flags | NOUVEAU_BO_LOW, + 0, 0); + so_method(so, curie, NV40TCL_COLOR2_PITCH, 1); so_data (so, rt[2]->pitch); } if (rt_enable & NV40TCL_RT_ENABLE_COLOR3) { - so_method(so, nv40->screen->curie, NV40TCL_DMA_COLOR3, 1); - so_reloc (so, nv40_surface_buffer(&rt[3]->base), 0, rt_flags | NOUVEAU_BO_OR, - nv40->nvws->channel->vram->handle, - nv40->nvws->channel->gart->handle); - so_method(so, nv40->screen->curie, NV40TCL_COLOR3_OFFSET, 1); - so_reloc (so, nv40_surface_buffer(&rt[3]->base), rt[3]->base.offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); - so_method(so, nv40->screen->curie, NV40TCL_COLOR3_PITCH, 1); + so_method(so, curie, NV40TCL_DMA_COLOR3, 1); + so_reloc (so, nv40_surface_buffer(&rt[3]->base), 0, + rt_flags | NOUVEAU_BO_OR, + chan->vram->handle, chan->gart->handle); + so_method(so, curie, NV40TCL_COLOR3_OFFSET, 1); + so_reloc (so, nv40_surface_buffer(&rt[3]->base), + rt[3]->base.offset, rt_flags | NOUVEAU_BO_LOW, + 0, 0); + so_method(so, curie, NV40TCL_COLOR3_PITCH, 1); so_data (so, rt[3]->pitch); } if (zeta_format) { - so_method(so, nv40->screen->curie, NV40TCL_DMA_ZETA, 1); - so_reloc (so, nv40_surface_buffer(&zeta->base), 0, rt_flags | NOUVEAU_BO_OR, - nv40->nvws->channel->vram->handle, - nv40->nvws->channel->gart->handle); - so_method(so, nv40->screen->curie, NV40TCL_ZETA_OFFSET, 1); - so_reloc (so, nv40_surface_buffer(&zeta->base), zeta->base.offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); - so_method(so, nv40->screen->curie, NV40TCL_ZETA_PITCH, 1); + so_method(so, curie, NV40TCL_DMA_ZETA, 1); + so_reloc (so, nv40_surface_buffer(&zeta->base), 0, + rt_flags | NOUVEAU_BO_OR, + chan->vram->handle, chan->gart->handle); + so_method(so, curie, NV40TCL_ZETA_OFFSET, 1); + so_reloc (so, nv40_surface_buffer(&zeta->base), + zeta->base.offset, rt_flags | NOUVEAU_BO_LOW, 0, 0); + so_method(so, curie, NV40TCL_ZETA_PITCH, 1); so_data (so, zeta->pitch); } - so_method(so, nv40->screen->curie, NV40TCL_RT_ENABLE, 1); + so_method(so, curie, NV40TCL_RT_ENABLE, 1); so_data (so, rt_enable); - so_method(so, nv40->screen->curie, NV40TCL_RT_HORIZ, 3); + so_method(so, curie, NV40TCL_RT_HORIZ, 3); so_data (so, (w << 16) | 0); so_data (so, (h << 16) | 0); so_data (so, rt_format); - so_method(so, nv40->screen->curie, NV40TCL_VIEWPORT_HORIZ, 2); + so_method(so, curie, NV40TCL_VIEWPORT_HORIZ, 2); so_data (so, (w << 16) | 0); so_data (so, (h << 16) | 0); - so_method(so, nv40->screen->curie, NV40TCL_VIEWPORT_CLIP_HORIZ(0), 2); + so_method(so, curie, NV40TCL_VIEWPORT_CLIP_HORIZ(0), 2); so_data (so, ((w - 1) << 16) | 0); so_data (so, ((h - 1) << 16) | 0); - so_method(so, nv40->screen->curie, 0x1d88, 1); + so_method(so, curie, 0x1d88, 1); so_data (so, (1 << 12) | h); so_ref(so, &nv40->state.hw[NV40_STATE_FB]); diff --git a/src/gallium/drivers/nv40/nv40_surface.c b/src/gallium/drivers/nv40/nv40_surface.c index 1a849da32d7..a596547974a 100644 --- a/src/gallium/drivers/nv40/nv40_surface.c +++ b/src/gallium/drivers/nv40/nv40_surface.c @@ -26,12 +26,13 @@ * **************************************************************************/ -#include "nv40_context.h" #include "pipe/p_defines.h" -#include "pipe/internal/p_winsys_screen.h" #include "pipe/p_inlines.h" + #include "util/u_tile.h" +#include "nv40_context.h" + static void nv40_surface_copy(struct pipe_context *pipe, struct pipe_surface *dest, unsigned destx, unsigned desty, diff --git a/src/gallium/drivers/nv40/nv40_transfer.c b/src/gallium/drivers/nv40/nv40_transfer.c index ce45055fe82..6d92ac3db9c 100644 --- a/src/gallium/drivers/nv40/nv40_transfer.c +++ b/src/gallium/drivers/nv40/nv40_transfer.c @@ -136,7 +136,7 @@ nv40_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt, } static void -nv40_transfer_del(struct pipe_screen *pscreen, struct pipe_transfer *ptx) +nv40_transfer_del(struct pipe_transfer *ptx) { struct nv40_transfer *tx = (struct nv40_transfer *)ptx; diff --git a/src/gallium/drivers/nv40/nv40_vbo.c b/src/gallium/drivers/nv40/nv40_vbo.c index f3518b2e4fe..b2753b8e2e0 100644 --- a/src/gallium/drivers/nv40/nv40_vbo.c +++ b/src/gallium/drivers/nv40/nv40_vbo.c @@ -1,5 +1,6 @@ #include "pipe/p_context.h" #include "pipe/p_state.h" +#include "pipe/p_inlines.h" #include "nv40_context.h" #include "nv40_state.h" @@ -70,7 +71,7 @@ static boolean nv40_vbo_set_idxbuf(struct nv40_context *nv40, struct pipe_buffer *ib, unsigned ib_size) { - struct pipe_screen *pscreen = &nv40->screen->pipe; + struct pipe_screen *pscreen = &nv40->screen->base.base; unsigned type; if (!ib) { @@ -108,7 +109,7 @@ nv40_vbo_static_attrib(struct nv40_context *nv40, struct nouveau_stateobj *so, int attrib, struct pipe_vertex_element *ve, struct pipe_vertex_buffer *vb) { - struct pipe_winsys *ws = nv40->pipe.winsys; + struct pipe_screen *pscreen = nv40->pipe.screen; struct nouveau_grobj *curie = nv40->screen->curie; unsigned type, ncomp; void *map; @@ -116,7 +117,7 @@ nv40_vbo_static_attrib(struct nv40_context *nv40, struct nouveau_stateobj *so, if (nv40_vbo_format_to_hw(ve->src_format, &type, &ncomp)) return FALSE; - map = ws->buffer_map(ws, vb->buffer, PIPE_BUFFER_USAGE_CPU_READ); + map = pipe_buffer_map(pscreen, vb->buffer, PIPE_BUFFER_USAGE_CPU_READ); map += vb->buffer_offset + ve->src_offset; switch (type) { @@ -148,17 +149,17 @@ nv40_vbo_static_attrib(struct nv40_context *nv40, struct nouveau_stateobj *so, so_data (so, fui(v[0])); break; default: - ws->buffer_unmap(ws, vb->buffer); + pipe_buffer_unmap(pscreen, vb->buffer); return FALSE; } } break; default: - ws->buffer_unmap(ws, vb->buffer); + pipe_buffer_unmap(pscreen, vb->buffer); return FALSE; } - ws->buffer_unmap(ws, vb->buffer); + pipe_buffer_unmap(pscreen, vb->buffer); return TRUE; } @@ -168,7 +169,7 @@ nv40_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { struct nv40_context *nv40 = nv40_context(pipe); - struct nouveau_channel *chan = nv40->nvws->channel; + struct nouveau_channel *chan = nv40->screen->base.channel; unsigned restart; nv40_vbo_set_idxbuf(nv40, NULL, 0); @@ -227,7 +228,7 @@ static INLINE void nv40_draw_elements_u08(struct nv40_context *nv40, void *ib, unsigned mode, unsigned start, unsigned count) { - struct nouveau_channel *chan = nv40->nvws->channel; + struct nouveau_channel *chan = nv40->screen->base.channel; while (count) { uint8_t *elts = (uint8_t *)ib + start; @@ -276,7 +277,7 @@ static INLINE void nv40_draw_elements_u16(struct nv40_context *nv40, void *ib, unsigned mode, unsigned start, unsigned count) { - struct nouveau_channel *chan = nv40->nvws->channel; + struct nouveau_channel *chan = nv40->screen->base.channel; while (count) { uint16_t *elts = (uint16_t *)ib + start; @@ -325,7 +326,7 @@ static INLINE void nv40_draw_elements_u32(struct nv40_context *nv40, void *ib, unsigned mode, unsigned start, unsigned count) { - struct nouveau_channel *chan = nv40->nvws->channel; + struct nouveau_channel *chan = nv40->screen->base.channel; while (count) { uint32_t *elts = (uint32_t *)ib + start; @@ -367,10 +368,10 @@ nv40_draw_elements_inline(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { struct nv40_context *nv40 = nv40_context(pipe); - struct pipe_winsys *ws = pipe->winsys; + struct pipe_screen *pscreen = pipe->screen; void *map; - map = ws->buffer_map(ws, ib, PIPE_BUFFER_USAGE_CPU_READ); + map = pipe_buffer_map(pscreen, ib, PIPE_BUFFER_USAGE_CPU_READ); if (!ib) { NOUVEAU_ERR("failed mapping ib\n"); return FALSE; @@ -391,7 +392,7 @@ nv40_draw_elements_inline(struct pipe_context *pipe, break; } - ws->buffer_unmap(ws, ib); + pipe_buffer_unmap(pscreen, ib); return TRUE; } @@ -400,7 +401,7 @@ nv40_draw_elements_vbo(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { struct nv40_context *nv40 = nv40_context(pipe); - struct nouveau_channel *chan = nv40->nvws->channel; + struct nouveau_channel *chan = nv40->screen->base.channel; unsigned restart; while (count) { @@ -519,17 +520,20 @@ nv40_vbo_validate(struct nv40_context *nv40) return FALSE; } - so_reloc(vtxbuf, vb->buffer, vb->buffer_offset + ve->src_offset, - vb_flags | NOUVEAU_BO_LOW | NOUVEAU_BO_OR, - 0, NV40TCL_VTXBUF_ADDRESS_DMA1); + so_reloc(vtxbuf, nouveau_bo(vb->buffer), + vb->buffer_offset + ve->src_offset, + vb_flags | NOUVEAU_BO_LOW | NOUVEAU_BO_OR, + 0, NV40TCL_VTXBUF_ADDRESS_DMA1); so_data (vtxfmt, ((vb->stride << NV40TCL_VTXFMT_STRIDE_SHIFT) | (ncomp << NV40TCL_VTXFMT_SIZE_SHIFT) | type)); } if (ib) { + struct nouveau_bo *bo = nouveau_bo(ib); + so_method(vtxbuf, curie, NV40TCL_IDXBUF_ADDRESS, 2); - so_reloc (vtxbuf, ib, 0, vb_flags | NOUVEAU_BO_LOW, 0, 0); - so_reloc (vtxbuf, ib, ib_format, vb_flags | NOUVEAU_BO_OR, + so_reloc (vtxbuf, bo, 0, vb_flags | NOUVEAU_BO_LOW, 0, 0); + so_reloc (vtxbuf, bo, ib_format, vb_flags | NOUVEAU_BO_OR, 0, NV40TCL_IDXBUF_FORMAT_DMA1); } diff --git a/src/gallium/drivers/nv40/nv40_vertprog.c b/src/gallium/drivers/nv40/nv40_vertprog.c index 7df9a4d3264..0382dbba8f6 100644 --- a/src/gallium/drivers/nv40/nv40_vertprog.c +++ b/src/gallium/drivers/nv40/nv40_vertprog.c @@ -1,6 +1,7 @@ #include "pipe/p_context.h" #include "pipe/p_defines.h" #include "pipe/p_state.h" +#include "pipe/p_inlines.h" #include "pipe/p_shader_tokens.h" #include "tgsi/tgsi_parse.h" @@ -787,10 +788,10 @@ nv40_vertprog_translate(struct nv40_context *nv40, assert(imm->Immediate.NrTokens == 4 + 1); vpc->imm[vpc->nr_imm++] = constant(vpc, -1, - imm->u.ImmediateFloat32[0].Float, - imm->u.ImmediateFloat32[1].Float, - imm->u.ImmediateFloat32[2].Float, - imm->u.ImmediateFloat32[3].Float); + imm->u[0].Float, + imm->u[1].Float, + imm->u[2].Float, + imm->u[3].Float); } break; case TGSI_TOKEN_TYPE_INSTRUCTION: @@ -855,8 +856,7 @@ out_err: static boolean nv40_vertprog_validate(struct nv40_context *nv40) { - struct nouveau_winsys *nvws = nv40->nvws; - struct pipe_winsys *ws = nv40->pipe.winsys; + struct pipe_screen *pscreen = nv40->pipe.screen; struct nouveau_grobj *curie = nv40->screen->curie; struct nv40_vertex_program *vp; struct pipe_buffer *constbuf; @@ -895,15 +895,15 @@ check_gpu_resources: struct nouveau_stateobj *so; uint vplen = vp->nr_insns; - if (nvws->res_alloc(heap, vplen, vp, &vp->exec)) { + if (nouveau_resource_alloc(heap, vplen, vp, &vp->exec)) { while (heap->next && heap->size < vplen) { struct nv40_vertex_program *evict; evict = heap->next->priv; - nvws->res_free(&evict->exec); + nouveau_resource_free(&evict->exec); } - if (nvws->res_alloc(heap, vplen, vp, &vp->exec)) + if (nouveau_resource_alloc(heap, vplen, vp, &vp->exec)) assert(0); } @@ -925,15 +925,15 @@ check_gpu_resources: if (vp->nr_consts && !vp->data) { struct nouveau_resource *heap = nv40->screen->vp_data_heap; - if (nvws->res_alloc(heap, vp->nr_consts, vp, &vp->data)) { + if (nouveau_resource_alloc(heap, vp->nr_consts, vp, &vp->data)) { while (heap->next && heap->size < vp->nr_consts) { struct nv40_vertex_program *evict; evict = heap->next->priv; - nvws->res_free(&evict->data); + nouveau_resource_free(&evict->data); } - if (nvws->res_alloc(heap, vp->nr_consts, vp, &vp->data)) + if (nouveau_resource_alloc(heap, vp->nr_consts, vp, &vp->data)) assert(0); } @@ -981,8 +981,8 @@ check_gpu_resources: float *map = NULL; if (constbuf) { - map = ws->buffer_map(ws, constbuf, - PIPE_BUFFER_USAGE_CPU_READ); + map = pipe_buffer_map(pscreen, constbuf, + PIPE_BUFFER_USAGE_CPU_READ); } for (i = 0; i < vp->nr_consts; i++) { @@ -1003,7 +1003,7 @@ check_gpu_resources: } if (constbuf) - ws->buffer_unmap(ws, constbuf); + pscreen->buffer_unmap(pscreen, constbuf); } /* Upload vtxprog */ @@ -1035,8 +1035,6 @@ check_gpu_resources: void nv40_vertprog_destroy(struct nv40_context *nv40, struct nv40_vertex_program *vp) { - struct nouveau_winsys *nvws = nv40->screen->nvws; - vp->translated = FALSE; if (vp->nr_insns) { @@ -1051,9 +1049,9 @@ nv40_vertprog_destroy(struct nv40_context *nv40, struct nv40_vertex_program *vp) vp->nr_consts = 0; } - nvws->res_free(&vp->exec); + nouveau_resource_free(&vp->exec); vp->exec_start = 0; - nvws->res_free(&vp->data); + nouveau_resource_free(&vp->data); vp->data_start = 0; vp->data_start_min = 0; diff --git a/src/gallium/drivers/nv50/nv50_clear.c b/src/gallium/drivers/nv50/nv50_clear.c index 33427a15a56..e0b2d2880b0 100644 --- a/src/gallium/drivers/nv50/nv50_clear.c +++ b/src/gallium/drivers/nv50/nv50_clear.c @@ -31,7 +31,7 @@ nv50_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba, double depth, unsigned stencil) { struct nv50_context *nv50 = nv50_context(pipe); - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->base.channel; struct nouveau_grobj *tesla = nv50->screen->tesla; struct pipe_framebuffer_state *fb = &nv50->framebuffer; unsigned mode = 0, i; diff --git a/src/gallium/drivers/nv50/nv50_context.c b/src/gallium/drivers/nv50/nv50_context.c index a511f655c19..6e8f4f9750d 100644 --- a/src/gallium/drivers/nv50/nv50_context.c +++ b/src/gallium/drivers/nv50/nv50_context.c @@ -31,15 +31,23 @@ static void nv50_flush(struct pipe_context *pipe, unsigned flags, struct pipe_fence_handle **fence) { - struct nv50_context *nv50 = (struct nv50_context *)pipe; - - FIRE_RING(nv50->screen->nvws->channel); + struct nv50_context *nv50 = nv50_context(pipe); + struct nouveau_channel *chan = nv50->screen->base.channel; + struct nouveau_grobj *eng2d = nv50->screen->eng2d; + + /* We need this in the ddx for reliable composite, not sure what we're + * actually flushing. We generate all our own flushes with flags = 0. */ + WAIT_RING(chan, 3); + BEGIN_RING(chan, eng2d, 0x0110, 1); + OUT_RING (chan, 0); + + FIRE_RING(chan); } static void nv50_destroy(struct pipe_context *pipe) { - struct nv50_context *nv50 = (struct nv50_context *)pipe; + struct nv50_context *nv50 = nv50_context(pipe); draw_destroy(nv50->draw); FREE(nv50); @@ -112,5 +120,3 @@ nv50_create(struct pipe_screen *pscreen, unsigned pctx_id) return &nv50->pipe; } - - diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h index 7b67a754397..1e9e8e49bfb 100644 --- a/src/gallium/drivers/nv50/nv50_context.h +++ b/src/gallium/drivers/nv50/nv50_context.h @@ -63,14 +63,19 @@ struct nv50_rasterizer_stateobj { struct nouveau_stateobj *so; }; +struct nv50_sampler_stateobj { + bool normalized; + unsigned tsc[8]; +}; + struct nv50_miptree_level { int *image_offset; unsigned pitch; + unsigned tile_mode; }; struct nv50_miptree { - struct pipe_texture base; - struct pipe_buffer *buffer; + struct nouveau_miptree base; struct nv50_miptree_level level[PIPE_MAX_TEXTURE_LEVELS]; int image_nr; @@ -93,13 +98,6 @@ nv50_surface(struct pipe_surface *pt) return (struct nv50_surface *)pt; } -static INLINE struct pipe_buffer * -nv50_surface_buffer(struct pipe_surface *surface) -{ - struct nv50_miptree *mt = (struct nv50_miptree *)surface->texture; - return mt->buffer; -} - struct nv50_state { unsigned dirty; @@ -115,10 +113,12 @@ struct nv50_state { unsigned viewport_bypass; struct nouveau_stateobj *tsc_upload; struct nouveau_stateobj *tic_upload; + unsigned miptree_nr; struct nouveau_stateobj *vertprog; struct nouveau_stateobj *fragprog; struct nouveau_stateobj *vtxfmt; struct nouveau_stateobj *vtxbuf; + struct nouveau_stateobj *vtxattr; }; struct nv50_context { @@ -147,7 +147,7 @@ struct nv50_context { unsigned vtxbuf_nr; struct pipe_vertex_element vtxelt[PIPE_MAX_ATTRIBS]; unsigned vtxelt_nr; - unsigned *sampler[PIPE_MAX_SAMPLERS]; + struct nv50_sampler_stateobj *sampler[PIPE_MAX_SAMPLERS]; unsigned sampler_nr; struct nv50_miptree *miptree[PIPE_MAX_SAMPLERS]; unsigned miptree_nr; diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c index f79a7ca86c9..03b9243b828 100644 --- a/src/gallium/drivers/nv50/nv50_miptree.c +++ b/src/gallium/drivers/nv50/nv50_miptree.c @@ -29,26 +29,41 @@ static struct pipe_texture * nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp) { + struct nouveau_device *dev = nouveau_screen(pscreen)->device; struct nv50_miptree *mt = CALLOC_STRUCT(nv50_miptree); - struct pipe_texture *pt = &mt->base; - unsigned usage, width = tmp->width[0], height = tmp->height[0]; + struct pipe_texture *pt = &mt->base.base; + unsigned width = tmp->width[0], height = tmp->height[0]; unsigned depth = tmp->depth[0]; - int i, l; + uint32_t tile_mode, tile_flags, tile_h; + int ret, i, l; - mt->base = *tmp; - pipe_reference_init(&mt->base.reference, 1); - mt->base.screen = pscreen; + *pt = *tmp; + pipe_reference_init(&pt->reference, 1); + pt->screen = pscreen; - usage = PIPE_BUFFER_USAGE_PIXEL; switch (pt->format) { + case PIPE_FORMAT_Z32_FLOAT: + tile_flags = 0x4800; + break; case PIPE_FORMAT_Z24S8_UNORM: - case PIPE_FORMAT_Z16_UNORM: - usage |= NOUVEAU_BUFFER_USAGE_ZETA; + tile_flags = 0x1800; + break; + case PIPE_FORMAT_X8Z24_UNORM: + case PIPE_FORMAT_S8Z24_UNORM: + tile_flags = 0x2800; break; default: + tile_flags = 0x7000; break; } + if (pt->height[0] > 32) tile_mode = 4; + else if (pt->height[0] > 16) tile_mode = 3; + else if (pt->height[0] > 8) tile_mode = 2; + else if (pt->height[0] > 4) tile_mode = 1; + else tile_mode = 0; + tile_h = 1 << (tile_mode + 2); + switch (pt->target) { case PIPE_TEXTURE_3D: mt->image_nr = pt->depth[0]; @@ -72,20 +87,27 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp) lvl->image_offset = CALLOC(mt->image_nr, sizeof(int)); lvl->pitch = align(pt->width[l] * pt->block.size, 64); + lvl->tile_mode = tile_mode; width = MAX2(1, width >> 1); height = MAX2(1, height >> 1); depth = MAX2(1, depth >> 1); + + if (tile_mode && height <= (tile_h >> 1)) { + tile_mode--; + tile_h >>= 1; + } } for (i = 0; i < mt->image_nr; i++) { for (l = 0; l <= pt->last_level; l++) { struct nv50_miptree_level *lvl = &mt->level[l]; int size; + tile_h = 1 << (lvl->tile_mode + 2); size = align(pt->width[l], 8) * pt->block.size; size = align(size, 64); - size *= align(pt->height[l], 8) * pt->block.size; + size *= align(pt->height[l], tile_h); lvl->image_offset[i] = mt->total_size; @@ -93,19 +115,22 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp) } } - mt->buffer = pscreen->buffer_create(pscreen, 256, usage, mt->total_size); - if (!mt->buffer) { + ret = nouveau_bo_new_tile(dev, NOUVEAU_BO_VRAM, 256, mt->total_size, + mt->level[0].tile_mode, tile_flags, + &mt->base.bo); + if (ret) { FREE(mt); return NULL; } - return &mt->base; + return pt; } static struct pipe_texture * nv50_miptree_blanket(struct pipe_screen *pscreen, const struct pipe_texture *pt, const unsigned *stride, struct pipe_buffer *pb) { + struct nouveau_bo *bo = nouveau_bo(pb); struct nv50_miptree *mt; /* Only supports 2D, non-mipmapped textures for the moment */ @@ -117,15 +142,15 @@ nv50_miptree_blanket(struct pipe_screen *pscreen, const struct pipe_texture *pt, if (!mt) return NULL; - mt->base = *pt; - pipe_reference_init(&mt->base.reference, 1); - mt->base.screen = pscreen; + mt->base.base = *pt; + pipe_reference_init(&mt->base.base.reference, 1); + mt->base.base.screen = pscreen; mt->image_nr = 1; mt->level[0].pitch = *stride; mt->level[0].image_offset = CALLOC(1, sizeof(unsigned)); - pipe_buffer_reference(&mt->buffer, pb); - return &mt->base; + nouveau_bo_ref(bo, &mt->base.bo); + return &mt->base.base; } static void @@ -133,8 +158,8 @@ nv50_miptree_destroy(struct pipe_texture *pt) { struct nv50_miptree *mt = nv50_miptree(pt); - pipe_buffer_reference(&mt->buffer, NULL); - FREE(mt); + nouveau_bo_ref(NULL, &mt->base.bo); + FREE(mt); } static struct pipe_surface * @@ -177,8 +202,8 @@ nv50_miptree_surface_del(struct pipe_surface *ps) { struct nv50_surface *s = nv50_surface(ps); - pipe_texture_reference(&ps->texture, NULL); - FREE(s); + pipe_texture_reference(&ps->texture, NULL); + FREE(s); } void diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c index 2d15868ae84..4a838529de7 100644 --- a/src/gallium/drivers/nv50/nv50_program.c +++ b/src/gallium/drivers/nv50/nv50_program.c @@ -85,6 +85,9 @@ struct nv50_reg { int hw; int neg; + + int rhw; /* result hw for FP outputs, or interpolant index */ + int acc; /* instruction where this reg is last read (first insn == 1) */ }; struct nv50_pc { @@ -108,12 +111,23 @@ struct nv50_pc { struct nv50_reg *temp_temp[16]; unsigned temp_temp_nr; + + unsigned interp_mode[32]; + /* perspective interpolation registers */ + struct nv50_reg *iv_p; + struct nv50_reg *iv_c; + + /* current instruction and total number of insns */ + unsigned insn_cur; + unsigned insn_nr; + + boolean allow32; }; static void alloc_reg(struct nv50_pc *pc, struct nv50_reg *reg) { - int i; + int i = 0; if (reg->type == P_RESULT) { if (pc->p->cfg.high_result < (reg->hw + 1)) @@ -131,7 +145,22 @@ alloc_reg(struct nv50_pc *pc, struct nv50_reg *reg) return; } - for (i = 0; i < NV50_SU_MAX_TEMP; i++) { + if (reg->rhw != -1) { + /* try to allocate temporary with index rhw first */ + if (!(pc->r_temp[reg->rhw])) { + pc->r_temp[reg->rhw] = reg; + reg->hw = reg->rhw; + if (pc->p->cfg.high_temp < (reg->rhw + 1)) + pc->p->cfg.high_temp = reg->rhw + 1; + return; + } + /* make sure we don't get things like $r0 needs to go + * in $r1 and $r1 in $r0 + */ + i = pc->result_nr * 4; + } + + for (; i < NV50_SU_MAX_TEMP; i++) { if (!(pc->r_temp[i])) { pc->r_temp[i] = reg; reg->hw = i; @@ -159,6 +188,7 @@ alloc_temp(struct nv50_pc *pc, struct nv50_reg *dst) r->type = P_TEMP; r->index = -1; r->hw = i; + r->rhw = -1; pc->r_temp[i] = r; return r; } @@ -168,6 +198,38 @@ alloc_temp(struct nv50_pc *pc, struct nv50_reg *dst) return NULL; } +/* Assign the hw of the discarded temporary register src + * to the tgsi register dst and free src. + */ +static void +assimilate_temp(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) +{ + assert(src->index == -1 && src->hw != -1); + + if (dst->hw != -1) + pc->r_temp[dst->hw] = NULL; + pc->r_temp[src->hw] = dst; + dst->hw = src->hw; + + FREE(src); +} + +/* release the hardware resource held by r */ +static void +release_hw(struct nv50_pc *pc, struct nv50_reg *r) +{ + assert(r->type == P_TEMP); + if (r->hw == -1) + return; + + assert(pc->r_temp[r->hw] == r); + pc->r_temp[r->hw] = NULL; + + r->acc = 0; + if (r->index == -1) + FREE(r); +} + static void free_temp(struct nv50_pc *pc, struct nv50_reg *r) { @@ -189,7 +251,7 @@ alloc_temp4(struct nv50_pc *pc, struct nv50_reg *dst[4], int idx) if (pc->r_temp[idx] || pc->r_temp[idx + 1] || pc->r_temp[idx + 2] || pc->r_temp[idx + 3]) - return alloc_temp4(pc, dst, idx + 1); + return alloc_temp4(pc, dst, idx + 4); for (i = 0; i < 4; i++) { dst[i] = CALLOC_STRUCT(nv50_reg); @@ -234,7 +296,7 @@ kill_temp_temp(struct nv50_pc *pc) static int ctor_immd(struct nv50_pc *pc, float x, float y, float z, float w) { - pc->immd_buf = REALLOC(pc->immd_buf, (pc->immd_nr * r * sizeof(float)), + pc->immd_buf = REALLOC(pc->immd_buf, (pc->immd_nr * 4 * sizeof(float)), (pc->immd_nr + 1) * 4 * sizeof(float)); pc->immd_buf[(pc->immd_nr * 4) + 0] = x; pc->immd_buf[(pc->immd_nr * 4) + 1] = y; @@ -250,7 +312,13 @@ alloc_immd(struct nv50_pc *pc, float f) struct nv50_reg *r = CALLOC_STRUCT(nv50_reg); unsigned hw; - hw = ctor_immd(pc, f, 0, 0, 0) * 4; + for (hw = 0; hw < pc->immd_nr * 4; hw++) + if (pc->immd_buf[hw] == f) + break; + + if (hw == pc->immd_nr * 4) + hw = ctor_immd(pc, f, -f, 0.5 * f, 0) * 4; + r->type = P_IMMD; r->hw = hw; r->index = -1; @@ -341,7 +409,8 @@ set_dst(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_program_exec *e) static INLINE void set_immd(struct nv50_pc *pc, struct nv50_reg *imm, struct nv50_program_exec *e) { - unsigned val = fui(pc->immd_buf[imm->hw]); /* XXX */ + float f = pc->immd_buf[imm->hw]; + unsigned val = fui(imm->neg ? -f : f); set_long(pc, e); /*XXX: can't be predicated - bits overlap.. catch cases where both @@ -354,20 +423,35 @@ set_immd(struct nv50_pc *pc, struct nv50_reg *imm, struct nv50_program_exec *e) e->inst[1] |= (val >> 6) << 2; } + +#define INTERP_LINEAR 0 +#define INTERP_FLAT 1 +#define INTERP_PERSPECTIVE 2 +#define INTERP_CENTROID 4 + +/* interpolant index has been stored in dst->rhw */ static void -emit_interp(struct nv50_pc *pc, struct nv50_reg *dst, - struct nv50_reg *src, struct nv50_reg *iv) +emit_interp(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *iv, + unsigned mode) { + assert(dst->rhw != -1); struct nv50_program_exec *e = exec(pc); e->inst[0] |= 0x80000000; set_dst(pc, dst, e); - alloc_reg(pc, src); - e->inst[0] |= (src->hw << 16); - if (iv) { - e->inst[0] |= (1 << 25); - alloc_reg(pc, iv); - e->inst[0] |= (iv->hw << 9); + e->inst[0] |= (dst->rhw << 16); + + if (mode & INTERP_FLAT) { + e->inst[0] |= (1 << 8); + } else { + if (mode & INTERP_PERSPECTIVE) { + e->inst[0] |= (1 << 25); + alloc_reg(pc, iv); + e->inst[0] |= (iv->hw << 9); + } + + if (mode & INTERP_CENTROID) + e->inst[0] |= (1 << 24); } emit(pc, e); @@ -378,22 +462,12 @@ set_data(struct nv50_pc *pc, struct nv50_reg *src, unsigned m, unsigned s, struct nv50_program_exec *e) { set_long(pc, e); -#if 1 - e->inst[1] |= (1 << 22); -#else - if (src->type == P_IMMD) { - e->inst[1] |= (NV50_CB_PMISC << 22); - } else { - if (pc->p->type == PIPE_SHADER_VERTEX) - e->inst[1] |= (NV50_CB_PVP << 22); - else - e->inst[1] |= (NV50_CB_PFP << 22); - } -#endif e->param.index = src->hw; e->param.shift = s; e->param.mask = m << (s % 32); + + e->inst[1] |= (((src->type == P_IMMD) ? 0 : 1) << 22); } static void @@ -405,12 +479,11 @@ emit_mov(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) set_dst(pc, dst, e); - if (0 && dst->type != P_RESULT && src->type == P_IMMD) { + if (pc->allow32 && dst->type != P_RESULT && src->type == P_IMMD) { set_immd(pc, src, e); /*XXX: 32-bit, but steals part of "half" reg space - need to * catch and handle this case if/when we do half-regs */ - e->inst[0] |= 0x00008000; } else if (src->type == P_IMMD || src->type == P_CONST) { set_long(pc, e); @@ -426,18 +499,25 @@ emit_mov(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) e->inst[0] |= (src->hw << 9); } - /* We really should support "half" instructions here at some point, - * but I don't feel confident enough about them yet. - */ - set_long(pc, e); if (is_long(e) && !is_immd(e)) { e->inst[1] |= 0x04000000; /* 32-bit */ - e->inst[1] |= 0x0003c000; /* "subsubop" 0xf == mov */ - } + e->inst[1] |= 0x0000c000; /* "subsubop" 0x3 */ + if (!(e->inst[1] & 0x20000000)) + e->inst[1] |= 0x00030000; /* "subsubop" 0xf */ + } else + e->inst[0] |= 0x00008000; emit(pc, e); } +static INLINE void +emit_mov_immdval(struct nv50_pc *pc, struct nv50_reg *dst, float f) +{ + struct nv50_reg *imm = alloc_immd(pc, f); + emit_mov(pc, dst, imm); + FREE(imm); +} + static boolean check_swap_src_0_1(struct nv50_pc *pc, struct nv50_reg **s0, struct nv50_reg **s1) @@ -541,12 +621,26 @@ emit_mul(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0, struct nv50_program_exec *e = exec(pc); e->inst[0] |= 0xc0000000; - set_long(pc, e); + + if (!pc->allow32) + set_long(pc, e); check_swap_src_0_1(pc, &src0, &src1); set_dst(pc, dst, e); set_src_0(pc, src0, e); - set_src_1(pc, src1, e); + if (src1->type == P_IMMD && !is_long(e)) { + if (src0->neg) + e->inst[0] |= 0x00008000; + set_immd(pc, src1, e); + } else { + set_src_1(pc, src1, e); + if (src0->neg ^ src1->neg) { + if (is_long(e)) + e->inst[1] |= 0x08000000; + else + e->inst[0] |= 0x00008000; + } + } emit(pc, e); } @@ -560,11 +654,20 @@ emit_add(struct nv50_pc *pc, struct nv50_reg *dst, e->inst[0] |= 0xb0000000; check_swap_src_0_1(pc, &src0, &src1); + + if (!pc->allow32 || src0->neg || src1->neg) { + set_long(pc, e); + e->inst[1] |= (src0->neg << 26) | (src1->neg << 27); + } + set_dst(pc, dst, e); set_src_0(pc, src0, e); - if (is_long(e)) + if (src1->type == P_CONST || src1->type == P_ATTR || is_long(e)) set_src_2(pc, src1, e); else + if (src1->type == P_IMMD) + set_immd(pc, src1, e); + else set_src_1(pc, src1, e); emit(pc, e); @@ -588,25 +691,13 @@ emit_minmax(struct nv50_pc *pc, unsigned sub, struct nv50_reg *dst, emit(pc, e); } -static void +static INLINE void emit_sub(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0, struct nv50_reg *src1) { - struct nv50_program_exec *e = exec(pc); - - e->inst[0] |= 0xb0000000; - - set_long(pc, e); - if (check_swap_src_0_1(pc, &src0, &src1)) - e->inst[1] |= 0x04000000; - else - e->inst[1] |= 0x08000000; - - set_dst(pc, dst, e); - set_src_0(pc, src0, e); - set_src_2(pc, src1, e); - - emit(pc, e); + src1->neg ^= 1; + emit_add(pc, dst, src0, src1); + src1->neg ^= 1; } static void @@ -623,26 +714,21 @@ emit_mad(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0, set_src_1(pc, src1, e); set_src_2(pc, src2, e); + if (src0->neg ^ src1->neg) + e->inst[1] |= 0x04000000; + if (src2->neg) + e->inst[1] |= 0x08000000; + emit(pc, e); } -static void +static INLINE void emit_msb(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0, struct nv50_reg *src1, struct nv50_reg *src2) { - struct nv50_program_exec *e = exec(pc); - - e->inst[0] |= 0xe0000000; - set_long(pc, e); - e->inst[1] |= 0x08000000; /* src0 * src1 - src2 */ - - check_swap_src_0_1(pc, &src0, &src1); - set_dst(pc, dst, e); - set_src_0(pc, src0, e); - set_src_1(pc, src1, e); - set_src_2(pc, src2, e); - - emit(pc, e); + src2->neg ^= 1; + emit_mad(pc, dst, src0, src1, src2); + src2->neg ^= 1; } static void @@ -693,6 +779,48 @@ emit_precossin(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) emit(pc, e); } +#define CVTOP_RN 0x01 +#define CVTOP_FLOOR 0x03 +#define CVTOP_CEIL 0x05 +#define CVTOP_TRUNC 0x07 +#define CVTOP_SAT 0x08 +#define CVTOP_ABS 0x10 + +#define CVT_F32_F32 0xc4 +#define CVT_F32_S32 0x44 +#define CVT_F32_U32 0x64 +#define CVT_S32_F32 0x8c +#define CVT_S32_S32 0x0c +#define CVT_F32_F32_ROP 0xcc + +static void +emit_cvt(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src, + int wp, unsigned cop, unsigned fmt) +{ + struct nv50_program_exec *e; + + e = exec(pc); + set_long(pc, e); + + e->inst[0] |= 0xa0000000; + e->inst[1] |= 0x00004000; + e->inst[1] |= (cop << 16); + e->inst[1] |= (fmt << 24); + set_src_0(pc, src, e); + + if (wp >= 0) + set_pred_wr(pc, 1, wp, e); + + if (dst) + set_dst(pc, dst, e); + else { + e->inst[0] |= 0x000001fc; + e->inst[1] |= 0x00000008; + } + + emit(pc, e); +} + static void emit_set(struct nv50_pc *pc, unsigned c_op, struct nv50_reg *dst, struct nv50_reg *src0, struct nv50_reg *src1) @@ -736,22 +864,10 @@ emit_set(struct nv50_pc *pc, unsigned c_op, struct nv50_reg *dst, free_temp(pc, dst); } -static void +static INLINE void emit_flr(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) { - struct nv50_program_exec *e = exec(pc); - - e->inst[0] = 0xa0000000; /* cvt */ - set_long(pc, e); - e->inst[1] |= (6 << 29); /* cvt */ - e->inst[1] |= 0x08000000; /* integer mode */ - e->inst[1] |= 0x04000000; /* 32 bit */ - e->inst[1] |= ((0x1 << 3)) << 14; /* .rn */ - e->inst[1] |= (1 << 14); /* src .f32 */ - set_dst(pc, dst, e); - set_src_0(pc, src, e); - - emit(pc, e); + emit_cvt(pc, dst, src, -1, CVTOP_FLOOR, CVT_F32_F32_ROP); } static void @@ -768,21 +884,10 @@ emit_pow(struct nv50_pc *pc, struct nv50_reg *dst, free_temp(pc, temp); } -static void +static INLINE void emit_abs(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) { - struct nv50_program_exec *e = exec(pc); - - e->inst[0] = 0xa0000000; /* cvt */ - set_long(pc, e); - e->inst[1] |= (6 << 29); /* cvt */ - e->inst[1] |= 0x04000000; /* 32 bit */ - e->inst[1] |= (1 << 14); /* src .f32 */ - e->inst[1] |= ((1 << 6) << 14); /* .abs */ - set_dst(pc, dst, e); - set_src_0(pc, src, e); - - emit(pc, e); + emit_cvt(pc, dst, src, -1, CVTOP_ABS, CVT_F32_F32); } static void @@ -794,18 +899,12 @@ emit_lit(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask, struct nv50_reg *neg128 = alloc_immd(pc, -127.999999); struct nv50_reg *pos128 = alloc_immd(pc, 127.999999); struct nv50_reg *tmp[4]; + boolean allow32 = pc->allow32; - if (mask & (1 << 0)) - emit_mov(pc, dst[0], one); - - if (mask & (1 << 3)) - emit_mov(pc, dst[3], one); + pc->allow32 = FALSE; if (mask & (3 << 1)) { - if (mask & (1 << 1)) - tmp[0] = dst[1]; - else - tmp[0] = temp_temp(pc); + tmp[0] = alloc_temp(pc, NULL); emit_minmax(pc, 4, tmp[0], src[0], zero); } @@ -823,6 +922,26 @@ emit_lit(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask, emit_mov(pc, dst[2], zero); set_pred(pc, 3, 0, pc->p->exec_tail); } + + if (mask & (1 << 1)) + assimilate_temp(pc, dst[1], tmp[0]); + else + if (mask & (1 << 2)) + free_temp(pc, tmp[0]); + + pc->allow32 = allow32; + + /* do this last, in case src[i,j] == dst[0,3] */ + if (mask & (1 << 0)) + emit_mov(pc, dst[0], one); + + if (mask & (1 << 3)) + emit_mov(pc, dst[3], one); + + FREE(pos128); + FREE(neg128); + FREE(zero); + FREE(one); } static void @@ -853,6 +972,8 @@ emit_kil(struct nv50_pc *pc, struct nv50_reg *src) e->inst[1] = 0xc4014788; set_src_0(pc, src, e); set_pred_wr(pc, 1, r_pred, e); + if (src->neg) + e->inst[1] |= 0x20000000; emit(pc, e); /* This is probably KILP */ @@ -863,6 +984,181 @@ emit_kil(struct nv50_pc *pc, struct nv50_reg *src) emit(pc, e); } +static void +emit_tex(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask, + struct nv50_reg **src, unsigned unit, unsigned type, boolean proj) +{ + struct nv50_reg *temp, *t[4]; + struct nv50_program_exec *e; + + unsigned c, mode, dim; + + switch (type) { + case TGSI_TEXTURE_1D: + dim = 1; + break; + case TGSI_TEXTURE_UNKNOWN: + case TGSI_TEXTURE_2D: + case TGSI_TEXTURE_SHADOW1D: /* XXX: x, z */ + case TGSI_TEXTURE_RECT: + dim = 2; + break; + case TGSI_TEXTURE_3D: + case TGSI_TEXTURE_CUBE: + case TGSI_TEXTURE_SHADOW2D: + case TGSI_TEXTURE_SHADOWRECT: /* XXX */ + dim = 3; + break; + default: + assert(0); + break; + } + + /* some cards need t[0]'s hw index to be a multiple of 4 */ + alloc_temp4(pc, t, 0); + + if (proj) { + if (src[0]->type == P_TEMP && src[0]->rhw != -1) { + mode = pc->interp_mode[src[0]->index]; + + t[3]->rhw = src[3]->rhw; + emit_interp(pc, t[3], NULL, (mode & INTERP_CENTROID)); + emit_flop(pc, 0, t[3], t[3]); + + for (c = 0; c < dim; c++) { + t[c]->rhw = src[c]->rhw; + emit_interp(pc, t[c], t[3], + (mode | INTERP_PERSPECTIVE)); + } + } else { + emit_flop(pc, 0, t[3], src[3]); + for (c = 0; c < dim; c++) + emit_mul(pc, t[c], src[c], t[3]); + + /* XXX: for some reason the blob sometimes uses MAD: + * emit_mad(pc, t[c], src[0][c], t[3], t[3]) + * pc->p->exec_tail->inst[1] |= 0x080fc000; + */ + } + } else { + if (type == TGSI_TEXTURE_CUBE) { + temp = temp_temp(pc); + emit_minmax(pc, 4, temp, src[0], src[1]); + emit_minmax(pc, 4, temp, temp, src[2]); + emit_flop(pc, 0, temp, temp); + for (c = 0; c < 3; c++) + emit_mul(pc, t[c], src[c], temp); + } else { + for (c = 0; c < dim; c++) + emit_mov(pc, t[c], src[c]); + } + } + + e = exec(pc); + set_long(pc, e); + e->inst[0] |= 0xf0000000; + e->inst[1] |= 0x00000004; + set_dst(pc, t[0], e); + e->inst[0] |= (unit << 9); + + if (dim == 2) + e->inst[0] |= 0x00400000; + else + if (dim == 3) + e->inst[0] |= 0x00800000; + + e->inst[0] |= (mask & 0x3) << 25; + e->inst[1] |= (mask & 0xc) << 12; + + emit(pc, e); + +#if 1 + if (mask & 1) emit_mov(pc, dst[0], t[0]); + if (mask & 2) emit_mov(pc, dst[1], t[1]); + if (mask & 4) emit_mov(pc, dst[2], t[2]); + if (mask & 8) emit_mov(pc, dst[3], t[3]); + + free_temp4(pc, t); +#else + /* XXX: if p.e. MUL is used directly after TEX, it would still use + * the texture coordinates, not the fetched values: latency ? */ + + for (c = 0; c < 4; c++) { + if (mask & (1 << c)) + assimilate_temp(pc, dst[c], t[c]); + else + free_temp(pc, t[c]); + } +#endif +} + +static void +convert_to_long(struct nv50_pc *pc, struct nv50_program_exec *e) +{ + unsigned q = 0, m = ~0; + + assert(!is_long(e)); + + switch (e->inst[0] >> 28) { + case 0x1: + /* MOV */ + q = 0x0403c000; + m = 0xffff7fff; + break; + case 0x8: + /* INTERP (move centroid, perspective and flat bits) */ + m = ~0x03000100; + q = (e->inst[0] & (3 << 24)) >> (24 - 16); + q |= (e->inst[0] & (1 << 8)) << (18 - 8); + break; + case 0x9: + /* RCP */ + break; + case 0xB: + /* ADD */ + m = ~(127 << 16); + q = ((e->inst[0] & (~m)) >> 2); + break; + case 0xC: + /* MUL */ + m = ~0x00008000; + q = ((e->inst[0] & (~m)) << 12); + break; + case 0xE: + /* MAD (if src2 == dst) */ + q = ((e->inst[0] & 0x1fc) << 12); + break; + default: + assert(0); + break; + } + + set_long(pc, e); + pc->p->exec_size++; + + e->inst[0] &= m; + e->inst[1] |= q; +} + +static boolean +negate_supported(const struct tgsi_full_instruction *insn, int i) +{ + switch (insn->Instruction.Opcode) { + case TGSI_OPCODE_DP3: + case TGSI_OPCODE_DP4: + case TGSI_OPCODE_MUL: + case TGSI_OPCODE_KIL: + case TGSI_OPCODE_ADD: + case TGSI_OPCODE_SUB: + case TGSI_OPCODE_MAD: + return TRUE; + case TGSI_OPCODE_POW: + return (i == 1) ? TRUE : FALSE; + default: + return FALSE; + } +} + static struct nv50_reg * tgsi_dst(struct nv50_pc *pc, int c, const struct tgsi_full_dst_register *dst) { @@ -881,11 +1177,14 @@ tgsi_dst(struct nv50_pc *pc, int c, const struct tgsi_full_dst_register *dst) } static struct nv50_reg * -tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src) +tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src, + boolean neg) { struct nv50_reg *r = NULL; struct nv50_reg *temp; - unsigned c; + unsigned sgn, c; + + sgn = tgsi_util_get_full_src_register_sign_mode(src, chan); c = tgsi_util_get_full_src_register_extswizzle(src, chan); switch (c) { @@ -915,16 +1214,17 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src) break; case TGSI_EXTSWIZZLE_ZERO: r = alloc_immd(pc, 0.0); - break; + return r; case TGSI_EXTSWIZZLE_ONE: - r = alloc_immd(pc, 1.0); - break; + if (sgn == TGSI_UTIL_SIGN_TOGGLE || sgn == TGSI_UTIL_SIGN_SET) + return alloc_immd(pc, -1.0); + return alloc_immd(pc, 1.0); default: assert(0); break; } - switch (tgsi_util_get_full_src_register_sign_mode(src, chan)) { + switch (sgn) { case TGSI_UTIL_SIGN_KEEP: break; case TGSI_UTIL_SIGN_CLEAR: @@ -933,14 +1233,21 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src) r = temp; break; case TGSI_UTIL_SIGN_TOGGLE: - temp = temp_temp(pc); - emit_neg(pc, temp, r); - r = temp; + if (neg) + r->neg = 1; + else { + temp = temp_temp(pc); + emit_neg(pc, temp, r); + r = temp; + } break; case TGSI_UTIL_SIGN_SET: temp = temp_temp(pc); emit_abs(pc, temp, r); - emit_neg(pc, temp, r); + if (neg) + temp->neg = 1; + else + emit_neg(pc, temp, temp); r = temp; break; default: @@ -951,12 +1258,40 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src) return r; } +/* returns TRUE if instruction can overwrite sources before they're read */ +static boolean +direct2dest_op(const struct tgsi_full_instruction *insn) +{ + if (insn->Instruction.Saturate) + return FALSE; + + switch (insn->Instruction.Opcode) { + case TGSI_OPCODE_COS: + case TGSI_OPCODE_DP3: + case TGSI_OPCODE_DP4: + case TGSI_OPCODE_DPH: + case TGSI_OPCODE_KIL: + case TGSI_OPCODE_LIT: + case TGSI_OPCODE_POW: + case TGSI_OPCODE_RCP: + case TGSI_OPCODE_RSQ: + case TGSI_OPCODE_SCS: + case TGSI_OPCODE_SIN: + case TGSI_OPCODE_TEX: + case TGSI_OPCODE_TXP: + return FALSE; + default: + return TRUE; + } +} + static boolean nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) { const struct tgsi_full_instruction *inst = &tok->FullInstruction; struct nv50_reg *rdst[4], *dst[4], *src[3][4], *temp; unsigned mask, sat, unit; + boolean assimilate = FALSE; int i, c; mask = inst->FullDstRegisters[0].DstRegister.WriteMask; @@ -967,6 +1302,10 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) dst[c] = tgsi_dst(pc, c, &inst->FullDstRegisters[0]); else dst[c] = NULL; + rdst[c] = NULL; + src[0][c] = NULL; + src[1][c] = NULL; + src[2][c] = NULL; } for (i = 0; i < inst->Instruction.NumSrcRegs; i++) { @@ -976,7 +1315,8 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) unit = fs->SrcRegister.Index; for (c = 0; c < 4; c++) - src[i][c] = tgsi_src(pc, c, fs); + src[i][c] = tgsi_src(pc, c, fs, + negate_supported(inst, i)); } if (sat) { @@ -984,6 +1324,25 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) rdst[c] = dst[c]; dst[c] = temp_temp(pc); } + } else + if (direct2dest_op(inst)) { + for (c = 0; c < 4; c++) { + if (!dst[c] || dst[c]->type != P_TEMP) + continue; + + for (i = c + 1; i < 4; i++) { + if (dst[c] == src[0][i] || + dst[c] == src[1][i] || + dst[c] == src[2][i]) + break; + } + if (i == 4) + continue; + + assimilate = TRUE; + rdst[c] = dst[c]; + dst[c] = alloc_temp(pc, NULL); + } } switch (inst->Instruction.Opcode) { @@ -1002,7 +1361,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_COS: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_precossin(pc, temp, src[0][0]); emit_flop(pc, 5, temp, temp); for (c = 0; c < 4; c++) { @@ -1012,7 +1371,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_DP3: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_mul(pc, temp, src[0][0], src[1][0]); emit_mad(pc, temp, src[0][1], src[1][1], temp); emit_mad(pc, temp, src[0][2], src[1][2], temp); @@ -1021,10 +1380,9 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) continue; emit_mov(pc, dst[c], temp); } - free_temp(pc, temp); break; case TGSI_OPCODE_DP4: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_mul(pc, temp, src[0][0], src[1][0]); emit_mad(pc, temp, src[0][1], src[1][1], temp); emit_mad(pc, temp, src[0][2], src[1][2], temp); @@ -1034,10 +1392,9 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) continue; emit_mov(pc, dst[c], temp); } - free_temp(pc, temp); break; case TGSI_OPCODE_DPH: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_mul(pc, temp, src[0][0], src[1][0]); emit_mad(pc, temp, src[0][1], src[1][1], temp); emit_mad(pc, temp, src[0][2], src[1][2], temp); @@ -1047,7 +1404,6 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) continue; emit_mov(pc, dst[c], temp); } - free_temp(pc, temp); break; case TGSI_OPCODE_DST: { @@ -1064,7 +1420,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_EX2: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_preex2(pc, temp, src[0][0]); emit_flop(pc, 6, temp, temp); for (c = 0; c < 4; c++) { @@ -1072,7 +1428,6 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) continue; emit_mov(pc, dst[c], temp); } - free_temp(pc, temp); break; case TGSI_OPCODE_FLR: for (c = 0; c < 4; c++) { @@ -1082,26 +1437,26 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_FRC: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; emit_flr(pc, temp, src[0][c]); emit_sub(pc, dst[c], src[0][c], temp); } - free_temp(pc, temp); break; case TGSI_OPCODE_KIL: emit_kil(pc, src[0][0]); emit_kil(pc, src[0][1]); emit_kil(pc, src[0][2]); emit_kil(pc, src[0][3]); + pc->p->cfg.fp.regs[2] |= 0x00100000; break; case TGSI_OPCODE_LIT: emit_lit(pc, &dst[0], mask, &src[0][0]); break; case TGSI_OPCODE_LG2: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_flop(pc, 3, temp, src[0][0]); for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) @@ -1110,15 +1465,12 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_LRP: + temp = temp_temp(pc); for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; - /*XXX: we can do better than this */ - temp = alloc_temp(pc, NULL); - emit_neg(pc, temp, src[0][c]); - emit_mad(pc, temp, temp, src[2][c], src[2][c]); - emit_mad(pc, dst[c], src[0][c], src[1][c], temp); - free_temp(pc, temp); + emit_sub(pc, temp, src[1][c], src[2][c]); + emit_mad(pc, dst[c], temp, src[0][c], src[2][c]); } break; case TGSI_OPCODE_MAD: @@ -1143,6 +1495,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_MOV: + case TGSI_OPCODE_SWZ: for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; @@ -1157,36 +1510,39 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_POW: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_pow(pc, temp, src[0][0], src[1][0]); for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; emit_mov(pc, dst[c], temp); } - free_temp(pc, temp); break; case TGSI_OPCODE_RCP: - for (c = 0; c < 4; c++) { + for (c = 3; c >= 0; c--) { if (!(mask & (1 << c))) continue; emit_flop(pc, 0, dst[c], src[0][0]); } break; case TGSI_OPCODE_RSQ: - for (c = 0; c < 4; c++) { + for (c = 3; c >= 0; c--) { if (!(mask & (1 << c))) continue; emit_flop(pc, 2, dst[c], src[0][0]); } break; case TGSI_OPCODE_SCS: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_precossin(pc, temp, src[0][0]); if (mask & (1 << 0)) emit_flop(pc, 5, dst[0], temp); if (mask & (1 << 1)) emit_flop(pc, 4, dst[1], temp); + if (mask & (1 << 2)) + emit_mov_immdval(pc, dst[2], 0.0); + if (mask & (1 << 3)) + emit_mov_immdval(pc, dst[3], 1.0); break; case TGSI_OPCODE_SGE: for (c = 0; c < 4; c++) { @@ -1196,7 +1552,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_SIN: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_precossin(pc, temp, src[0][0]); emit_flop(pc, 4, temp, temp); for (c = 0; c < 4; c++) { @@ -1220,33 +1576,15 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_TEX: + emit_tex(pc, dst, mask, src[0], unit, + inst->InstructionExtTexture.Texture, FALSE); + break; case TGSI_OPCODE_TXP: - { - struct nv50_reg *t[4]; - struct nv50_program_exec *e; - - alloc_temp4(pc, t, 0); - emit_mov(pc, t[0], src[0][0]); - emit_mov(pc, t[1], src[0][1]); - - e = exec(pc); - e->inst[0] = 0xf6400000; - e->inst[0] |= (unit << 9); - set_long(pc, e); - e->inst[1] |= 0x0000c004; - set_dst(pc, t[0], e); - emit(pc, e); - - if (mask & (1 << 0)) emit_mov(pc, dst[0], t[0]); - if (mask & (1 << 1)) emit_mov(pc, dst[1], t[1]); - if (mask & (1 << 2)) emit_mov(pc, dst[2], t[2]); - if (mask & (1 << 3)) emit_mov(pc, dst[3], t[3]); - - free_temp4(pc, t); - } + emit_tex(pc, dst, mask, src[0], unit, + inst->InstructionExtTexture.Texture, TRUE); break; case TGSI_OPCODE_XPD: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); if (mask & (1 << 0)) { emit_mul(pc, temp, src[0][2], src[1][1]); emit_msb(pc, dst[0], src[0][1], src[1][2], temp); @@ -1259,7 +1597,8 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) emit_mul(pc, temp, src[0][1], src[1][0]); emit_msb(pc, dst[2], src[0][0], src[1][1], temp); } - free_temp(pc, temp); + if (mask & (1 << 3)) + emit_mov_immdval(pc, dst[3], 1.0); break; case TGSI_OPCODE_END: break; @@ -1270,21 +1609,26 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) if (sat) { for (c = 0; c < 4; c++) { - struct nv50_program_exec *e; - if (!(mask & (1 << c))) continue; - e = exec(pc); + emit_cvt(pc, rdst[c], dst[c], -1, CVTOP_SAT, + CVT_F32_F32); + } + } else if (assimilate) { + for (c = 0; c < 4; c++) + if (rdst[c]) + assimilate_temp(pc, rdst[c], dst[c]); + } - e->inst[0] = 0xa0000000; /* cvt */ - set_long(pc, e); - e->inst[1] |= (6 << 29); /* cvt */ - e->inst[1] |= 0x04000000; /* 32 bit */ - e->inst[1] |= (1 << 14); /* src .f32 */ - e->inst[1] |= ((1 << 5) << 14); /* .sat */ - set_dst(pc, rdst[c], e); - set_src_0(pc, dst[c], e); - emit(pc, e); + for (i = 0; i < inst->Instruction.NumSrcRegs; i++) { + for (c = 0; c < 4; c++) { + if (!src[i][c]) + continue; + if (src[i][c]->index == -1 && src[i][c]->type == P_IMMD) + FREE(src[i][c]); + else + if (src[i][c]->acc == pc->insn_cur) + release_hw(pc, src[i][c]); } } @@ -1292,12 +1636,169 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) return TRUE; } +/* Adjust a bitmask that indicates what components of a source are used, + * we use this in tx_prep so we only load interpolants that are needed. + */ +static void +insn_adjust_mask(const struct tgsi_full_instruction *insn, unsigned *mask) +{ + const struct tgsi_instruction_ext_texture *tex; + + switch (insn->Instruction.Opcode) { + case TGSI_OPCODE_DP3: + *mask = 0x7; + break; + case TGSI_OPCODE_DP4: + case TGSI_OPCODE_DPH: + *mask = 0xF; + break; + case TGSI_OPCODE_LIT: + *mask = 0xB; + break; + case TGSI_OPCODE_RCP: + case TGSI_OPCODE_RSQ: + *mask = 0x1; + break; + case TGSI_OPCODE_TEX: + case TGSI_OPCODE_TXP: + assert(insn->Instruction.Extended); + tex = &insn->InstructionExtTexture; + + *mask = 0x7; + if (tex->Texture == TGSI_TEXTURE_1D) + *mask = 0x1; + else + if (tex->Texture == TGSI_TEXTURE_2D) + *mask = 0x3; + + if (insn->Instruction.Opcode == TGSI_OPCODE_TXP) + *mask |= 0x8; + break; + default: + break; + } +} + +static void +prep_inspect_insn(struct nv50_pc *pc, const union tgsi_full_token *tok, + unsigned *r_usage[2]) +{ + const struct tgsi_full_instruction *insn; + const struct tgsi_full_src_register *src; + const struct tgsi_dst_register *dst; + + unsigned i, c, k, n, mask, *acc_p; + + insn = &tok->FullInstruction; + dst = &insn->FullDstRegisters[0].DstRegister; + mask = dst->WriteMask; + + if (!r_usage[0]) + r_usage[0] = CALLOC(pc->temp_nr * 4, sizeof(unsigned)); + if (!r_usage[1]) + r_usage[1] = CALLOC(pc->attr_nr * 4, sizeof(unsigned)); + + if (dst->File == TGSI_FILE_TEMPORARY) { + for (c = 0; c < 4; c++) { + if (!(mask & (1 << c))) + continue; + r_usage[0][dst->Index * 4 + c] = pc->insn_nr; + } + } + + for (i = 0; i < insn->Instruction.NumSrcRegs; i++) { + src = &insn->FullSrcRegisters[i]; + + switch (src->SrcRegister.File) { + case TGSI_FILE_TEMPORARY: + acc_p = r_usage[0]; + break; + case TGSI_FILE_INPUT: + acc_p = r_usage[1]; + break; + default: + continue; + } + + insn_adjust_mask(insn, &mask); + + for (c = 0; c < 4; c++) { + if (!(mask & (1 << c))) + continue; + + k = tgsi_util_get_full_src_register_extswizzle(src, c); + switch (k) { + case TGSI_EXTSWIZZLE_X: + case TGSI_EXTSWIZZLE_Y: + case TGSI_EXTSWIZZLE_Z: + case TGSI_EXTSWIZZLE_W: + n = src->SrcRegister.Index * 4 + k; + acc_p[n] = pc->insn_nr; + break; + default: + break; + } + } + } +} + +static unsigned +load_fp_attrib(struct nv50_pc *pc, int i, unsigned *acc, int *mid, + int *aid, int *p_oid) +{ + struct nv50_reg *iv; + int oid, c, n; + unsigned mask = 0; + + iv = (pc->interp_mode[i] & INTERP_CENTROID) ? pc->iv_c : pc->iv_p; + + for (c = 0, n = i * 4; c < 4; c++, n++) { + oid = (*p_oid)++; + pc->attr[n].type = P_TEMP; + pc->attr[n].index = i; + + if (pc->attr[n].acc == acc[n]) + continue; + mask |= (1 << c); + + pc->attr[n].acc = acc[n]; + pc->attr[n].rhw = pc->attr[n].hw = -1; + alloc_reg(pc, &pc->attr[n]); + + pc->attr[n].rhw = (*aid)++; + emit_interp(pc, &pc->attr[n], iv, pc->interp_mode[i]); + + pc->p->cfg.fp.map[(*mid) / 4] |= oid << (8 * ((*mid) % 4)); + (*mid)++; + pc->p->cfg.fp.regs[1] += 0x00010001; + } + + return mask; +} + static boolean nv50_program_tx_prep(struct nv50_pc *pc) { struct tgsi_parse_context p; boolean ret = FALSE; unsigned i, c; + unsigned fcol, bcol, fcrd, depr; + + /* count (centroid) perspective interpolations */ + unsigned centroid_loads = 0; + unsigned perspect_loads = 0; + + /* track register access for temps and attrs */ + unsigned *r_usage[2]; + r_usage[0] = NULL; + r_usage[1] = NULL; + + depr = fcol = bcol = fcrd = 0xffff; + + if (pc->p->type == PIPE_SHADER_FRAGMENT) { + pc->p->cfg.fp.regs[0] = 0x01000404; + pc->p->cfg.fp.regs[1] = 0x00000400; + } tgsi_parse_init(&p, pc->p->pipe.tokens); while (!tgsi_parse_end_of_tokens(&p)) { @@ -1310,18 +1811,19 @@ nv50_program_tx_prep(struct nv50_pc *pc) const struct tgsi_full_immediate *imm = &p.FullToken.FullImmediate; - ctor_immd(pc, imm->u.ImmediateFloat32[0].Float, - imm->u.ImmediateFloat32[1].Float, - imm->u.ImmediateFloat32[2].Float, - imm->u.ImmediateFloat32[3].Float); + ctor_immd(pc, imm->u[0].Float, + imm->u[1].Float, + imm->u[2].Float, + imm->u[3].Float); } break; case TGSI_TOKEN_TYPE_DECLARATION: { const struct tgsi_full_declaration *d; - unsigned last; + unsigned last, first, mode; d = &p.FullToken.FullDeclaration; + first = d->DeclarationRange.First; last = d->DeclarationRange.Last; switch (d->Declaration.File) { @@ -1332,10 +1834,69 @@ nv50_program_tx_prep(struct nv50_pc *pc) case TGSI_FILE_OUTPUT: if (pc->result_nr < (last + 1)) pc->result_nr = last + 1; + + if (!d->Declaration.Semantic) + break; + + switch (d->Semantic.SemanticName) { + case TGSI_SEMANTIC_POSITION: + depr = first; + pc->p->cfg.fp.regs[2] |= 0x00000100; + pc->p->cfg.fp.regs[3] |= 0x00000011; + break; + default: + break; + } + break; case TGSI_FILE_INPUT: + { if (pc->attr_nr < (last + 1)) pc->attr_nr = last + 1; + + if (pc->p->type != PIPE_SHADER_FRAGMENT) + break; + + switch (d->Declaration.Interpolate) { + case TGSI_INTERPOLATE_CONSTANT: + mode = INTERP_FLAT; + break; + case TGSI_INTERPOLATE_PERSPECTIVE: + mode = INTERP_PERSPECTIVE; + break; + default: + mode = INTERP_LINEAR; + break; + } + + if (d->Declaration.Semantic) { + switch (d->Semantic.SemanticName) { + case TGSI_SEMANTIC_POSITION: + fcrd = first; + break; + case TGSI_SEMANTIC_COLOR: + fcol = first; + mode = INTERP_PERSPECTIVE; + break; + case TGSI_SEMANTIC_BCOLOR: + bcol = first; + mode = INTERP_PERSPECTIVE; + break; + } + } + + if (d->Declaration.Centroid) { + mode |= INTERP_CENTROID; + if (mode & INTERP_PERSPECTIVE) + centroid_loads++; + } else + if (mode & INTERP_PERSPECTIVE) + perspect_loads++; + + assert(last < 32); + for (i = first; i <= last; i++) + pc->interp_mode[i] = mode; + } break; case TGSI_FILE_CONSTANT: if (pc->param_nr < (last + 1)) @@ -1351,6 +1912,8 @@ nv50_program_tx_prep(struct nv50_pc *pc) } break; case TGSI_TOKEN_TYPE_INSTRUCTION: + pc->insn_nr++; + prep_inspect_insn(pc, tok, r_usage); break; default: break; @@ -1366,56 +1929,95 @@ nv50_program_tx_prep(struct nv50_pc *pc) for (c = 0; c < 4; c++) { pc->temp[i*4+c].type = P_TEMP; pc->temp[i*4+c].hw = -1; + pc->temp[i*4+c].rhw = -1; pc->temp[i*4+c].index = i; + pc->temp[i*4+c].acc = r_usage[0][i*4+c]; } } } if (pc->attr_nr) { - struct nv50_reg *iv = NULL; - int aid = 0; + int oid = 4, mid = 4, aid = 0; + /* oid = VP output id + * aid = FP attribute/interpolant id + * mid = VP output mapping field ID + */ pc->attr = CALLOC(pc->attr_nr * 4, sizeof(struct nv50_reg)); if (!pc->attr) goto out_err; if (pc->p->type == PIPE_SHADER_FRAGMENT) { - iv = alloc_temp(pc, NULL); - emit_interp(pc, iv, iv, NULL); - emit_flop(pc, 0, iv, iv); - aid++; - } + /* position should be loaded first */ + if (fcrd != 0xffff) { + unsigned mask; + mid = 0; + mask = load_fp_attrib(pc, fcrd, r_usage[1], + &mid, &aid, &oid); + oid = 0; + pc->p->cfg.fp.regs[1] |= (mask << 24); + pc->p->cfg.fp.map[0] = 0x04040404 * fcrd; + } + pc->p->cfg.fp.map[0] += 0x03020100; - for (i = 0; i < pc->attr_nr; i++) { - struct nv50_reg *a = &pc->attr[i*4]; + /* should do MAD fcrd.xy, fcrd, SOME_CONST, fcrd */ - for (c = 0; c < 4; c++) { - if (pc->p->type == PIPE_SHADER_FRAGMENT) { - struct nv50_reg *at = - alloc_temp(pc, NULL); - pc->attr[i*4+c].type = at->type; - pc->attr[i*4+c].hw = at->hw; - pc->attr[i*4+c].index = at->index; + if (perspect_loads) { + pc->iv_p = alloc_temp(pc, NULL); + + if (!(pc->p->cfg.fp.regs[1] & 0x08000000)) { + pc->p->cfg.fp.regs[1] |= 0x08000000; + pc->iv_p->rhw = aid++; + emit_interp(pc, pc->iv_p, NULL, + INTERP_LINEAR); + emit_flop(pc, 0, pc->iv_p, pc->iv_p); } else { - pc->p->cfg.vp.attr[aid/32] |= - (1 << (aid % 32)); - pc->attr[i*4+c].type = P_ATTR; - pc->attr[i*4+c].hw = aid++; - pc->attr[i*4+c].index = i; + pc->iv_p->rhw = aid - 1; + emit_flop(pc, 0, pc->iv_p, + &pc->attr[fcrd * 4 + 3]); } } - if (pc->p->type != PIPE_SHADER_FRAGMENT) - continue; + if (centroid_loads) { + pc->iv_c = alloc_temp(pc, NULL); + pc->iv_c->rhw = pc->iv_p ? aid - 1 : aid++; + emit_interp(pc, pc->iv_c, NULL, + INTERP_CENTROID); + emit_flop(pc, 0, pc->iv_c, pc->iv_c); + pc->p->cfg.fp.regs[1] |= 0x08000000; + } - emit_interp(pc, &a[0], &a[0], iv); - emit_interp(pc, &a[1], &a[1], iv); - emit_interp(pc, &a[2], &a[2], iv); - emit_interp(pc, &a[3], &a[3], iv); - } + for (c = 0; c < 4; c++) { + /* I don't know what these values do, but + * let's set them like the blob does: + */ + if (fcol != 0xffff && r_usage[1][fcol * 4 + c]) + pc->p->cfg.fp.regs[0] += 0x00010000; + if (bcol != 0xffff && r_usage[1][bcol * 4 + c]) + pc->p->cfg.fp.regs[0] += 0x00010000; + } + + for (i = 0; i < pc->attr_nr; i++) + load_fp_attrib(pc, i, r_usage[1], + &mid, &aid, &oid); - if (iv) - free_temp(pc, iv); + if (pc->iv_p) + free_temp(pc, pc->iv_p); + if (pc->iv_c) + free_temp(pc, pc->iv_c); + + pc->p->cfg.fp.high_map = (mid / 4); + pc->p->cfg.fp.high_map += ((mid % 4) ? 1 : 0); + } else { + /* vertex program */ + for (i = 0; i < pc->attr_nr * 4; i++) { + pc->p->cfg.vp.attr[aid / 32] |= + (1 << (aid % 32)); + pc->attr[i].type = P_ATTR; + pc->attr[i].hw = aid++; + pc->attr[i].index = i / 4; + } + } } if (pc->result_nr) { @@ -1430,12 +2032,20 @@ nv50_program_tx_prep(struct nv50_pc *pc) if (pc->p->type == PIPE_SHADER_FRAGMENT) { pc->result[i*4+c].type = P_TEMP; pc->result[i*4+c].hw = -1; + pc->result[i*4+c].rhw = (i == depr) ? + -1 : rid++; } else { pc->result[i*4+c].type = P_RESULT; pc->result[i*4+c].hw = rid++; } pc->result[i*4+c].index = i; } + + if (pc->p->type == PIPE_SHADER_FRAGMENT && + depr != 0xffff) { + pc->result[depr * 4 + 2].rhw = + (pc->result_nr - 1) * 4; + } } } @@ -1456,7 +2066,7 @@ nv50_program_tx_prep(struct nv50_pc *pc) } if (pc->immd_nr) { - int rid = pc->param_nr * 4; + int rid = 0; pc->immd = CALLOC(pc->immd_nr * 4, sizeof(struct nv50_reg)); if (!pc->immd) @@ -1473,15 +2083,38 @@ nv50_program_tx_prep(struct nv50_pc *pc) ret = TRUE; out_err: + if (r_usage[0]) + FREE(r_usage[0]); + if (r_usage[1]) + FREE(r_usage[1]); + tgsi_parse_free(&p); return ret; } +static void +free_nv50_pc(struct nv50_pc *pc) +{ + if (pc->immd) + FREE(pc->immd); + if (pc->param) + FREE(pc->param); + if (pc->result) + FREE(pc->result); + if (pc->attr) + FREE(pc->attr); + if (pc->temp) + FREE(pc->temp); + + FREE(pc); +} + static boolean nv50_program_tx(struct nv50_program *p) { struct tgsi_parse_context parse; struct nv50_pc *pc; + unsigned k; boolean ret; pc = CALLOC_STRUCT(nv50_pc); @@ -1498,10 +2131,16 @@ nv50_program_tx(struct nv50_program *p) while (!tgsi_parse_end_of_tokens(&parse)) { const union tgsi_full_token *tok = &parse.FullToken; + /* don't allow half insn/immd on first and last instruction */ + pc->allow32 = TRUE; + if (pc->insn_cur == 0 || pc->insn_cur + 2 == pc->insn_nr) + pc->allow32 = FALSE; + tgsi_parse_token(&parse); switch (tok->Token.Type) { case TGSI_TOKEN_TYPE_INSTRUCTION: + ++pc->insn_cur; ret = nv50_program_tx_insn(pc, tok); if (ret == FALSE) goto out_err; @@ -1515,8 +2154,40 @@ nv50_program_tx(struct nv50_program *p) struct nv50_reg out; out.type = P_TEMP; - for (out.hw = 0; out.hw < pc->result_nr * 4; out.hw++) - emit_mov(pc, &out, &pc->result[out.hw]); + for (k = 0; k < pc->result_nr * 4; k++) { + if (pc->result[k].rhw == -1) + continue; + if (pc->result[k].hw != pc->result[k].rhw) { + out.hw = pc->result[k].rhw; + emit_mov(pc, &out, &pc->result[k]); + } + if (pc->p->cfg.high_result < (pc->result[k].rhw + 1)) + pc->p->cfg.high_result = pc->result[k].rhw + 1; + } + } + + /* look for single half instructions and make them long */ + struct nv50_program_exec *e, *e_prev; + + for (k = 0, e = pc->p->exec_head, e_prev = NULL; e; e = e->next) { + if (!is_long(e)) + k++; + + if (!e->next || is_long(e->next)) { + if (k & 1) + convert_to_long(pc, e); + k = 0; + } + + if (e->next) + e_prev = e; + } + + if (!is_long(pc->p->exec_tail)) { + /* this may occur if moving FP results */ + assert(e_prev && !is_long(e_prev)); + convert_to_long(pc, e_prev); + convert_to_long(pc, pc->p->exec_tail); } assert(is_long(pc->p->exec_tail) && !is_immd(pc->p->exec_head)); @@ -1530,6 +2201,7 @@ out_err: tgsi_parse_free(&parse); out_cleanup: + free_nv50_pc(pc); return ret; } @@ -1543,17 +2215,17 @@ nv50_program_validate(struct nv50_context *nv50, struct nv50_program *p) static void nv50_program_upload_data(struct nv50_context *nv50, float *map, - unsigned start, unsigned count) + unsigned start, unsigned count, unsigned cbuf) { - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->base.channel; struct nouveau_grobj *tesla = nv50->screen->tesla; while (count) { unsigned nr = count > 2047 ? 2047 : count; - BEGIN_RING(chan, tesla, 0x00000f00, 1); - OUT_RING (chan, (NV50_CB_PMISC << 0) | (start << 8)); - BEGIN_RING(chan, tesla, 0x40000f04, nr); + BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 1); + OUT_RING (chan, (cbuf << 0) | (start << 8)); + BEGIN_RING(chan, tesla, NV50TCL_CB_DATA(0) | 0x40000000, nr); OUT_RINGp (chan, map, nr); map += nr; @@ -1565,70 +2237,93 @@ nv50_program_upload_data(struct nv50_context *nv50, float *map, static void nv50_program_validate_data(struct nv50_context *nv50, struct nv50_program *p) { - struct nouveau_winsys *nvws = nv50->screen->nvws; - struct pipe_winsys *ws = nv50->pipe.winsys; - unsigned nr = p->param_nr + p->immd_nr; + struct pipe_screen *pscreen = nv50->pipe.screen; - if (!p->data && nr) { - struct nouveau_resource *heap = nv50->screen->vp_data_heap; + if (!p->data[0] && p->immd_nr) { + struct nouveau_resource *heap = nv50->screen->immd_heap[0]; - if (nvws->res_alloc(heap, nr, p, &p->data)) { - while (heap->next && heap->size < nr) { + if (nouveau_resource_alloc(heap, p->immd_nr, p, &p->data[0])) { + while (heap->next && heap->size < p->immd_nr) { struct nv50_program *evict = heap->next->priv; - nvws->res_free(&evict->data); + nouveau_resource_free(&evict->data[0]); } - if (nvws->res_alloc(heap, nr, p, &p->data)) + if (nouveau_resource_alloc(heap, p->immd_nr, p, + &p->data[0])) assert(0); } + + /* immediates only need to be uploaded again when freed */ + nv50_program_upload_data(nv50, p->immd, p->data[0]->start, + p->immd_nr, NV50_CB_PMISC); } - if (p->param_nr) { - float *map = ws->buffer_map(ws, nv50->constbuf[p->type], - PIPE_BUFFER_USAGE_CPU_READ); - nv50_program_upload_data(nv50, map, p->data->start, - p->param_nr); - ws->buffer_unmap(ws, nv50->constbuf[p->type]); + if (!p->data[1] && p->param_nr) { + struct nouveau_resource *heap = + nv50->screen->parm_heap[p->type]; + + if (nouveau_resource_alloc(heap, p->param_nr, p, &p->data[1])) { + while (heap->next && heap->size < p->param_nr) { + struct nv50_program *evict = heap->next->priv; + nouveau_resource_free(&evict->data[1]); + } + + if (nouveau_resource_alloc(heap, p->param_nr, p, + &p->data[1])) + assert(0); + } } - if (p->immd_nr) { - nv50_program_upload_data(nv50, p->immd, - p->data->start + p->param_nr, - p->immd_nr); + if (p->param_nr) { + unsigned cbuf = NV50_CB_PVP; + float *map = pipe_buffer_map(pscreen, nv50->constbuf[p->type], + PIPE_BUFFER_USAGE_CPU_READ); + if (p->type == PIPE_SHADER_FRAGMENT) + cbuf = NV50_CB_PFP; + nv50_program_upload_data(nv50, map, p->data[1]->start, + p->param_nr, cbuf); + pipe_buffer_unmap(pscreen, nv50->constbuf[p->type]); } } static void nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p) { - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->base.channel; struct nouveau_grobj *tesla = nv50->screen->tesla; - struct pipe_screen *screen = nv50->pipe.screen; struct nv50_program_exec *e; struct nouveau_stateobj *so; const unsigned flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_WR; unsigned start, count, *up, *ptr; boolean upload = FALSE; - if (!p->buffer) { - p->buffer = screen->buffer_create(screen, 0x100, 0, p->exec_size * 4); + if (!p->bo) { + nouveau_bo_new(chan->device, NOUVEAU_BO_VRAM, 0x100, + p->exec_size * 4, &p->bo); upload = TRUE; } - if (p->data && p->data->start != p->data_start) { + if ((p->data[0] && p->data[0]->start != p->data_start[0]) || + (p->data[1] && p->data[1]->start != p->data_start[1])) { for (e = p->exec_head; e; e = e->next) { - unsigned ei, ci; + unsigned ei, ci, bs; if (e->param.index < 0) continue; + bs = (e->inst[1] >> 22) & 0x07; + assert(bs < 2); ei = e->param.shift >> 5; - ci = e->param.index + p->data->start; + ci = e->param.index + p->data[bs]->start; e->inst[ei] &= ~e->param.mask; e->inst[ei] |= (ci << e->param.shift); } - p->data_start = p->data->start; + if (p->data[0]) + p->data_start[0] = p->data[0]->start; + if (p->data[1]) + p->data_start[1] = p->data[1]->start; + upload = TRUE; } @@ -1637,13 +2332,11 @@ nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p) #ifdef NV50_PROGRAM_DUMP NOUVEAU_ERR("-------\n"); - up = ptr = MALLOC(p->exec_size * 4); for (e = p->exec_head; e; e = e->next) { NOUVEAU_ERR("0x%08x\n", e->inst[0]); if (is_long(e)) NOUVEAU_ERR("0x%08x\n", e->inst[1]); } - #endif up = ptr = MALLOC(p->exec_size * 4); @@ -1654,28 +2347,28 @@ nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p) } so = so_new(4,2); - so_method(so, nv50->screen->tesla, 0x1280, 3); - so_reloc (so, p->buffer, 0, flags | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, p->buffer, 0, flags | NOUVEAU_BO_LOW, 0, 0); + so_method(so, nv50->screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); + so_reloc (so, p->bo, 0, flags | NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, p->bo, 0, flags | NOUVEAU_BO_LOW, 0, 0); so_data (so, (NV50_CB_PUPLOAD << 16) | 0x0800); //(p->exec_size * 4)); start = 0; count = p->exec_size; while (count) { - struct nouveau_winsys *nvws = nv50->screen->nvws; + struct nouveau_channel *chan = nv50->screen->base.channel; unsigned nr; - so_emit(nvws, so); + so_emit(chan, so); nr = MIN2(count, 2047); - nr = MIN2(nvws->channel->pushbuf->remaining, nr); - if (nvws->channel->pushbuf->remaining < (nr + 3)) { + nr = MIN2(chan->pushbuf->remaining, nr); + if (chan->pushbuf->remaining < (nr + 3)) { FIRE_RING(chan); continue; } - BEGIN_RING(chan, tesla, 0x0f00, 1); + BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 1); OUT_RING (chan, (start << 8) | NV50_CB_PUPLOAD); - BEGIN_RING(chan, tesla, 0x40000f04, nr); + BEGIN_RING(chan, tesla, NV50TCL_CB_DATA(0) | 0x40000000, nr); OUT_RINGp (chan, up + start, nr); start += nr; @@ -1704,19 +2397,19 @@ nv50_vertprog_validate(struct nv50_context *nv50) so = so_new(13, 2); so_method(so, tesla, NV50TCL_VP_ADDRESS_HIGH, 2); - so_reloc (so, p->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | - NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, p->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | - NOUVEAU_BO_LOW, 0, 0); - so_method(so, tesla, 0x1650, 2); + so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | + NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | + NOUVEAU_BO_LOW, 0, 0); + so_method(so, tesla, NV50TCL_VP_ATTR_EN_0, 2); so_data (so, p->cfg.vp.attr[0]); so_data (so, p->cfg.vp.attr[1]); - so_method(so, tesla, 0x16b8, 1); + so_method(so, tesla, NV50TCL_VP_REG_ALLOC_RESULT, 1); so_data (so, p->cfg.high_result); - so_method(so, tesla, 0x16ac, 2); + so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 2); so_data (so, p->cfg.high_result); //8); so_data (so, p->cfg.high_temp); - so_method(so, tesla, 0x140c, 1); + so_method(so, tesla, NV50TCL_VP_START_ID, 1); so_data (so, 0); /* program start offset */ so_ref(so, &nv50->state.vertprog); so_ref(NULL, &so); @@ -1728,6 +2421,7 @@ nv50_fragprog_validate(struct nv50_context *nv50) struct nouveau_grobj *tesla = nv50->screen->tesla; struct nv50_program *p = nv50->fragprog; struct nouveau_stateobj *so; + unsigned i; if (!p->translated) { nv50_program_validate(nv50, p); @@ -1740,23 +2434,28 @@ nv50_fragprog_validate(struct nv50_context *nv50) so = so_new(64, 2); so_method(so, tesla, NV50TCL_FP_ADDRESS_HIGH, 2); - so_reloc (so, p->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | - NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, p->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | - NOUVEAU_BO_LOW, 0, 0); - so_method(so, tesla, 0x1904, 4); - so_data (so, 0x00040404); /* p: 0x01000404 */ + so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | + NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | + NOUVEAU_BO_LOW, 0, 0); + so_method(so, tesla, NV50TCL_MAP_SEMANTIC_0, 4); + so_data (so, p->cfg.fp.regs[0]); /* 0x01000404 / 0x00040404 */ so_data (so, 0x00000004); so_data (so, 0x00000000); so_data (so, 0x00000000); - so_method(so, tesla, 0x16bc, 3); /*XXX: fixme */ - so_data (so, 0x03020100); - so_data (so, 0x07060504); - so_data (so, 0x0b0a0908); - so_method(so, tesla, 0x1988, 2); - so_data (so, 0x08080408); //0x08040404); /* p: 0x0f000401 */ + so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), p->cfg.fp.high_map); + for (i = 0; i < p->cfg.fp.high_map; i++) + so_data(so, p->cfg.fp.map[i]); + so_method(so, tesla, NV50TCL_FP_INTERPOLANT_CTRL, 2); + so_data (so, p->cfg.fp.regs[1]); /* 0x08040404 / 0x0f000401 */ so_data (so, p->cfg.high_temp); - so_method(so, tesla, 0x1414, 1); + so_method(so, tesla, NV50TCL_FP_RESULT_COUNT, 1); + so_data (so, p->cfg.high_result); + so_method(so, tesla, NV50TCL_FP_CTRL_UNK19A8, 1); + so_data (so, p->cfg.fp.regs[2]); + so_method(so, tesla, NV50TCL_FP_CTRL_UNK196C, 1); + so_data (so, p->cfg.fp.regs[3]); + so_method(so, tesla, NV50TCL_FP_START_ID, 1); so_data (so, 0); /* program start offset */ so_ref(so, &nv50->state.fragprog); so_ref(NULL, &so); @@ -1765,8 +2464,6 @@ nv50_fragprog_validate(struct nv50_context *nv50) void nv50_program_destroy(struct nv50_context *nv50, struct nv50_program *p) { - struct pipe_screen *pscreen = nv50->pipe.screen; - while (p->exec_head) { struct nv50_program_exec *e = p->exec_head; @@ -1776,11 +2473,10 @@ nv50_program_destroy(struct nv50_context *nv50, struct nv50_program *p) p->exec_tail = NULL; p->exec_size = 0; - if (p->buffer) - pipe_buffer_reference(&p->buffer, NULL); + nouveau_bo_ref(NULL, &p->bo); - nv50->screen->nvws->res_free(&p->data); + nouveau_resource_free(&p->data[0]); + nouveau_resource_free(&p->data[1]); p->translated = 0; } - diff --git a/src/gallium/drivers/nv50/nv50_program.h b/src/gallium/drivers/nv50/nv50_program.h index 78deed6a384..096e0476aab 100644 --- a/src/gallium/drivers/nv50/nv50_program.h +++ b/src/gallium/drivers/nv50/nv50_program.h @@ -24,10 +24,10 @@ struct nv50_program { struct nv50_program_exec *exec_head; struct nv50_program_exec *exec_tail; unsigned exec_size; - struct nouveau_resource *data; - unsigned data_start; + struct nouveau_resource *data[2]; + unsigned data_start[2]; - struct pipe_buffer *buffer; + struct nouveau_bo *bo; float *immd; unsigned immd_nr; @@ -39,6 +39,11 @@ struct nv50_program { struct { unsigned attr[2]; } vp; + struct { + unsigned regs[4]; + unsigned map[5]; + unsigned high_map; + } fp; } cfg; }; diff --git a/src/gallium/drivers/nv50/nv50_query.c b/src/gallium/drivers/nv50/nv50_query.c index 35cebdbdc32..5305c93d59a 100644 --- a/src/gallium/drivers/nv50/nv50_query.c +++ b/src/gallium/drivers/nv50/nv50_query.c @@ -26,7 +26,7 @@ #include "nv50_context.h" struct nv50_query { - struct pipe_buffer *buffer; + struct nouveau_bo *bo; unsigned type; boolean ready; uint64_t result; @@ -41,14 +41,16 @@ nv50_query(struct pipe_query *pipe) static struct pipe_query * nv50_query_create(struct pipe_context *pipe, unsigned type) { - struct pipe_screen *screen = pipe->screen; + struct nouveau_device *dev = nouveau_screen(pipe->screen)->device; struct nv50_query *q = CALLOC_STRUCT(nv50_query); + int ret; assert (q->type == PIPE_QUERY_OCCLUSION_COUNTER); q->type = type; - q->buffer = screen->buffer_create(screen, 256, 0, 16); - if (!q->buffer) { + ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM | NOUVEAU_BO_MAP, 256, + 16, &q->bo); + if (ret) { FREE(q); return NULL; } @@ -62,7 +64,7 @@ nv50_query_destroy(struct pipe_context *pipe, struct pipe_query *pq) struct nv50_query *q = nv50_query(pq); if (q) { - pipe_buffer_reference(&q->buffer, NULL); + nouveau_bo_ref(NULL, &q->bo); FREE(q); } } @@ -71,7 +73,7 @@ static void nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq) { struct nv50_context *nv50 = nv50_context(pipe); - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->base.channel; struct nouveau_grobj *tesla = nv50->screen->tesla; struct nv50_query *q = nv50_query(pq); @@ -87,15 +89,14 @@ static void nv50_query_end(struct pipe_context *pipe, struct pipe_query *pq) { struct nv50_context *nv50 = nv50_context(pipe); - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->base.channel; struct nouveau_grobj *tesla = nv50->screen->tesla; struct nv50_query *q = nv50_query(pq); - struct nouveau_bo *bo = nv50->screen->nvws->get_bo(q->buffer); WAIT_RING (chan, 5); - BEGIN_RING(chan, tesla, 0x1b00, 4); - OUT_RELOCh(chan, bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR); - OUT_RELOCl(chan, bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR); + BEGIN_RING(chan, tesla, NV50TCL_QUERY_ADDRESS_HIGH, 4); + OUT_RELOCh(chan, q->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR); + OUT_RELOCl(chan, q->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR); OUT_RING (chan, 0x00000000); OUT_RING (chan, 0x0100f002); FIRE_RING (chan); @@ -105,19 +106,17 @@ static boolean nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq, boolean wait, uint64_t *result) { - struct pipe_winsys *ws = pipe->winsys; struct nv50_query *q = nv50_query(pq); - - /*XXX: Want to be able to return FALSE here instead of blocking - * until the result is available.. - */ + int ret; if (!q->ready) { - uint32_t *map = ws->buffer_map(ws, q->buffer, - PIPE_BUFFER_USAGE_CPU_READ); - q->result = map[1]; + ret = nouveau_bo_map(q->bo, NOUVEAU_BO_RD | + wait ? 0 : NOUVEAU_BO_NOWAIT); + if (ret) + return false; + q->result = ((uint32_t *)q->bo->map)[1]; q->ready = TRUE; - ws->buffer_unmap(ws, q->buffer); + nouveau_bo_unmap(q->bo); } *result = q->result; diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c index 29805645948..c7f80a22037 100644 --- a/src/gallium/drivers/nv50/nv50_screen.c +++ b/src/gallium/drivers/nv50/nv50_screen.c @@ -22,8 +22,6 @@ #include "pipe/p_screen.h" -#include "util/u_simple_screen.h" - #include "nv50_context.h" #include "nv50_screen.h" @@ -39,8 +37,17 @@ nv50_screen_is_format_supported(struct pipe_screen *pscreen, switch (format) { case PIPE_FORMAT_A8R8G8B8_UNORM: case PIPE_FORMAT_R5G6B5_UNORM: + return TRUE; + default: + break; + } + } else + if (tex_usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL) { + switch (format) { + case PIPE_FORMAT_Z32_FLOAT: case PIPE_FORMAT_Z24S8_UNORM: - case PIPE_FORMAT_Z16_UNORM: + case PIPE_FORMAT_X8Z24_UNORM: + case PIPE_FORMAT_S8Z24_UNORM: return TRUE; default: break; @@ -68,23 +75,6 @@ nv50_screen_is_format_supported(struct pipe_screen *pscreen, return FALSE; } -static const char * -nv50_screen_get_name(struct pipe_screen *pscreen) -{ - struct nv50_screen *screen = nv50_screen(pscreen); - struct nouveau_device *dev = screen->nvws->channel->device; - static char buffer[128]; - - snprintf(buffer, sizeof(buffer), "NV%02X", dev->chipset); - return buffer; -} - -static const char * -nv50_screen_get_vendor(struct pipe_screen *pscreen) -{ - return "nouveau"; -} - static int nv50_screen_get_param(struct pipe_screen *pscreen, int param) { @@ -120,6 +110,10 @@ nv50_screen_get_param(struct pipe_screen *pscreen, int param) return 1; case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS: return 0; + case PIPE_CAP_TGSI_CONT_SUPPORTED: + return 0; + case PIPE_CAP_BLEND_EQUATION_SEPARATE: + return 1; case NOUVEAU_CAP_HW_VTXBUF: return 1; case NOUVEAU_CAP_HW_IDXBUF: @@ -153,116 +147,131 @@ nv50_screen_get_paramf(struct pipe_screen *pscreen, int param) static void nv50_screen_destroy(struct pipe_screen *pscreen) { - FREE(pscreen); + struct nv50_screen *screen = nv50_screen(pscreen); + + nouveau_notifier_free(&screen->sync); + nouveau_grobj_free(&screen->tesla); + nouveau_grobj_free(&screen->eng2d); + nouveau_grobj_free(&screen->m2mf); + nouveau_screen_fini(&screen->base); + FREE(screen); } struct pipe_screen * -nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) +nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) { struct nv50_screen *screen = CALLOC_STRUCT(nv50_screen); + struct nouveau_channel *chan; + struct pipe_screen *pscreen; struct nouveau_stateobj *so; - unsigned tesla_class = 0, ret; - unsigned chipset = nvws->channel->device->chipset; - int i; + unsigned chipset = dev->chipset; + unsigned tesla_class = 0; + int ret, i; if (!screen) return NULL; - screen->nvws = nvws; + pscreen = &screen->base.base; + + ret = nouveau_screen_init(&screen->base, dev); + if (ret) { + nv50_screen_destroy(pscreen); + return NULL; + } + chan = screen->base.channel; + + pscreen->winsys = ws; + pscreen->destroy = nv50_screen_destroy; + pscreen->get_param = nv50_screen_get_param; + pscreen->get_paramf = nv50_screen_get_paramf; + pscreen->is_format_supported = nv50_screen_is_format_supported; + + nv50_screen_init_miptree_functions(pscreen); + nv50_transfer_init_screen_functions(pscreen); /* DMA engine object */ - ret = nvws->grobj_alloc(nvws, 0x5039, &screen->m2mf); + ret = nouveau_grobj_alloc(chan, 0xbeef5039, + NV50_MEMORY_TO_MEMORY_FORMAT, &screen->m2mf); if (ret) { NOUVEAU_ERR("Error creating M2MF object: %d\n", ret); - nv50_screen_destroy(&screen->pipe); + nv50_screen_destroy(pscreen); return NULL; } + BIND_RING(chan, screen->m2mf, 1); /* 2D object */ - ret = nvws->grobj_alloc(nvws, NV50_2D, &screen->eng2d); + ret = nouveau_grobj_alloc(chan, 0xbeef502d, NV50_2D, &screen->eng2d); if (ret) { NOUVEAU_ERR("Error creating 2D object: %d\n", ret); - nv50_screen_destroy(&screen->pipe); + nv50_screen_destroy(pscreen); return NULL; } + BIND_RING(chan, screen->eng2d, 2); /* 3D object */ switch (chipset & 0xf0) { case 0x50: - tesla_class = 0x5097; + tesla_class = NV50TCL; break; case 0x80: case 0x90: - tesla_class = 0x8297; + /* this stupid name should be corrected. */ + tesla_class = NV54TCL; break; case 0xa0: - tesla_class = 0x8397; + tesla_class = NVA0TCL; break; default: NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", chipset); - nv50_screen_destroy(&screen->pipe); + nv50_screen_destroy(pscreen); return NULL; } if (tesla_class == 0) { NOUVEAU_ERR("Unknown G8x chipset: NV%02x\n", chipset); - nv50_screen_destroy(&screen->pipe); + nv50_screen_destroy(pscreen); return NULL; } - ret = nvws->grobj_alloc(nvws, tesla_class, &screen->tesla); + ret = nouveau_grobj_alloc(chan, 0xbeef5097, tesla_class, + &screen->tesla); if (ret) { NOUVEAU_ERR("Error creating 3D object: %d\n", ret); - nv50_screen_destroy(&screen->pipe); + nv50_screen_destroy(pscreen); return NULL; } + BIND_RING(chan, screen->tesla, 3); /* Sync notifier */ - ret = nvws->notifier_alloc(nvws, 1, &screen->sync); + ret = nouveau_notifier_alloc(chan, 0xbeef0301, 1, &screen->sync); if (ret) { NOUVEAU_ERR("Error creating notifier object: %d\n", ret); - nv50_screen_destroy(&screen->pipe); + nv50_screen_destroy(pscreen); return NULL; } - /* Setup the pipe */ - screen->pipe.winsys = ws; - - screen->pipe.destroy = nv50_screen_destroy; - - screen->pipe.get_name = nv50_screen_get_name; - screen->pipe.get_vendor = nv50_screen_get_vendor; - screen->pipe.get_param = nv50_screen_get_param; - screen->pipe.get_paramf = nv50_screen_get_paramf; - - screen->pipe.is_format_supported = nv50_screen_is_format_supported; - - nv50_screen_init_miptree_functions(&screen->pipe); - nv50_transfer_init_screen_functions(&screen->pipe); - u_simple_screen_init(&screen->pipe); - /* Static M2MF init */ so = so_new(32, 0); - so_method(so, screen->m2mf, 0x0180, 3); + so_method(so, screen->m2mf, NV04_MEMORY_TO_MEMORY_FORMAT_DMA_NOTIFY, 3); so_data (so, screen->sync->handle); - so_data (so, screen->nvws->channel->vram->handle); - so_data (so, screen->nvws->channel->vram->handle); - so_emit(nvws, so); + so_data (so, chan->vram->handle); + so_data (so, chan->vram->handle); + so_emit(chan, so); so_ref (NULL, &so); /* Static 2D init */ so = so_new(64, 0); so_method(so, screen->eng2d, NV50_2D_DMA_NOTIFY, 4); so_data (so, screen->sync->handle); - so_data (so, screen->nvws->channel->vram->handle); - so_data (so, screen->nvws->channel->vram->handle); - so_data (so, screen->nvws->channel->vram->handle); + so_data (so, chan->vram->handle); + so_data (so, chan->vram->handle); + so_data (so, chan->vram->handle); so_method(so, screen->eng2d, NV50_2D_OPERATION, 1); so_data (so, NV50_2D_OPERATION_SRCCOPY); so_method(so, screen->eng2d, 0x0290, 1); so_data (so, 0); so_method(so, screen->eng2d, 0x0888, 1); so_data (so, 1); - so_emit(nvws, so); + so_emit(chan, so); so_ref(NULL, &so); /* Static tesla init */ @@ -275,63 +284,123 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) so_method(so, screen->tesla, NV50TCL_DMA_UNK0(0), NV50TCL_DMA_UNK0__SIZE); for (i = 0; i < NV50TCL_DMA_UNK0__SIZE; i++) - so_data(so, nvws->channel->vram->handle); + so_data(so, chan->vram->handle); so_method(so, screen->tesla, NV50TCL_DMA_UNK1(0), NV50TCL_DMA_UNK1__SIZE); for (i = 0; i < NV50TCL_DMA_UNK1__SIZE; i++) - so_data(so, nvws->channel->vram->handle); + so_data(so, chan->vram->handle); so_method(so, screen->tesla, 0x121c, 1); so_data (so, 1); so_method(so, screen->tesla, 0x13bc, 1); so_data (so, 0x54); + /* origin is top left (set to 1 for bottom left) */ so_method(so, screen->tesla, 0x13ac, 1); - so_data (so, 1); - so_method(so, screen->tesla, 0x16b8, 1); + so_data (so, 0); + so_method(so, screen->tesla, NV50TCL_VP_REG_ALLOC_RESULT, 1); so_data (so, 8); - /* Shared constant buffer */ - screen->constbuf = screen->pipe.buffer_create(&screen->pipe, 0, 0, 128 * 4 * 4); - if (nvws->res_init(&screen->vp_data_heap, 0, 128)) { - NOUVEAU_ERR("Error initialising constant buffer\n"); - nv50_screen_destroy(&screen->pipe); + /* constant buffers for immediates and VP/FP parameters */ + ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, 128*4*4, + &screen->constbuf_misc[0]); + if (ret) { + nv50_screen_destroy(pscreen); return NULL; } - so_method(so, screen->tesla, 0x1280, 3); - so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM | + for (i = 0; i < 2; i++) { + ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, 128*4*4, + &screen->constbuf_parm[i]); + if (ret) { + nv50_screen_destroy(pscreen); + return NULL; + } + } + + if (nouveau_resource_init(&screen->immd_heap[0], 0, 128) || + nouveau_resource_init(&screen->parm_heap[0], 0, 128) || + nouveau_resource_init(&screen->parm_heap[1], 0, 128)) + { + NOUVEAU_ERR("Error initialising constant buffers.\n"); + nv50_screen_destroy(pscreen); + return NULL; + } + + /* + // map constant buffers: + // B = buffer ID (maybe more than 1 byte) + // N = CB index used in shader instruction + // P = program type (0 = VP, 2 = GP, 3 = FP) + so_method(so, screen->tesla, 0x1694, 1); + so_data (so, 0x000BBNP1); + */ + + so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); + so_reloc (so, screen->constbuf_misc[0], 0, NOUVEAU_BO_VRAM | + NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, screen->constbuf_misc[0], 0, NOUVEAU_BO_VRAM | + NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); + so_data (so, (NV50_CB_PMISC << 16) | 0x00000800); + so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); + so_data (so, 0x00000001 | (NV50_CB_PMISC << 12)); + so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); + so_data (so, 0x00000031 | (NV50_CB_PMISC << 12)); + + so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); + so_reloc (so, screen->constbuf_parm[0], 0, NOUVEAU_BO_VRAM | + NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, screen->constbuf_parm[0], 0, NOUVEAU_BO_VRAM | + NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); + so_data (so, (NV50_CB_PVP << 16) | 0x00000800); + so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); + so_data (so, 0x00000101 | (NV50_CB_PVP << 12)); + + so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); + so_reloc (so, screen->constbuf_parm[1], 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM | + so_reloc (so, screen->constbuf_parm[1], 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); - so_data (so, (NV50_CB_PMISC << 16) | 0x00001000); + so_data (so, (NV50_CB_PFP << 16) | 0x00000800); + so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); + so_data (so, 0x00000131 | (NV50_CB_PFP << 12)); /* Texture sampler/image unit setup - we abuse the constant buffer * upload mechanism for the moment to upload data to the tex config * blocks. At some point we *may* want to go the NVIDIA way of doing * things? */ - screen->tic = screen->pipe.buffer_create(&screen->pipe, 0, 0, 32 * 8 * 4); - so_method(so, screen->tesla, 0x1280, 3); + ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, 32*8*4, &screen->tic); + if (ret) { + nv50_screen_destroy(pscreen); + return NULL; + } + + so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); so_data (so, (NV50_CB_TIC << 16) | 0x0800); - so_method(so, screen->tesla, 0x1574, 3); + so_method(so, screen->tesla, NV50TCL_TIC_ADDRESS_HIGH, 3); so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); so_data (so, 0x00000800); - screen->tsc = screen->pipe.buffer_create(&screen->pipe, 0, 0, 32 * 8 * 4); - so_method(so, screen->tesla, 0x1280, 3); + ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, 32*8*4, &screen->tsc); + if (ret) { + nv50_screen_destroy(pscreen); + return NULL; + } + + so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); so_data (so, (NV50_CB_TSC << 16) | 0x0800); - so_method(so, screen->tesla, 0x155c, 3); + so_method(so, screen->tesla, NV50TCL_TSC_ADDRESS_HIGH, 3); so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM | @@ -341,7 +410,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) /* Vertex array limits - max them out */ for (i = 0; i < 16; i++) { - so_method(so, screen->tesla, 0x1080 + (i * 8), 2); + so_method(so, screen->tesla, NV50TCL_UNK1080_OFFSET_HIGH(i), 2); so_data (so, 0x000000ff); so_data (so, 0xffffffff); } @@ -352,14 +421,16 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) so_method(so, screen->tesla, 0x1234, 1); so_data (so, 1); - so_method(so, screen->tesla, 0x1458, 1); + + /* activate first scissor rectangle */ + so_method(so, screen->tesla, NV50TCL_SCISSOR_ENABLE, 1); so_data (so, 1); - so_emit(nvws, so); + so_emit(chan, so); so_ref (so, &screen->static_init); so_ref (NULL, &so); - nvws->push_flush(nvws, 0, NULL); + nouveau_pushbuf_flush(chan, 0); - return &screen->pipe; + return pscreen; } diff --git a/src/gallium/drivers/nv50/nv50_screen.h b/src/gallium/drivers/nv50/nv50_screen.h index db567aaac89..61e24a5b571 100644 --- a/src/gallium/drivers/nv50/nv50_screen.h +++ b/src/gallium/drivers/nv50/nv50_screen.h @@ -1,10 +1,10 @@ #ifndef __NV50_SCREEN_H__ #define __NV50_SCREEN_H__ -#include "pipe/p_screen.h" +#include "nouveau/nouveau_screen.h" struct nv50_screen { - struct pipe_screen pipe; + struct nouveau_screen base; struct nouveau_winsys *nvws; @@ -15,11 +15,14 @@ struct nv50_screen { struct nouveau_grobj *m2mf; struct nouveau_notifier *sync; - struct pipe_buffer *constbuf; - struct nouveau_resource *vp_data_heap; + struct nouveau_bo *constbuf_misc[1]; + struct nouveau_bo *constbuf_parm[2]; - struct pipe_buffer *tic; - struct pipe_buffer *tsc; + struct nouveau_resource *immd_heap[1]; + struct nouveau_resource *parm_heap[2]; + + struct nouveau_bo *tic; + struct nouveau_bo *tsc; struct nouveau_stateobj *static_init; }; diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c index ba852194cdd..4283808ed93 100644 --- a/src/gallium/drivers/nv50/nv50_state.c +++ b/src/gallium/drivers/nv50/nv50_state.c @@ -136,9 +136,11 @@ static void * nv50_sampler_state_create(struct pipe_context *pipe, const struct pipe_sampler_state *cso) { - unsigned *tsc = CALLOC(8, sizeof(unsigned)); + struct nv50_sampler_stateobj *sso = CALLOC(1, sizeof(*sso)); + unsigned *tsc = sso->tsc; + float limit; - tsc[0] = (0x00024000 | + tsc[0] = (0x00026000 | (wrap_mode(cso->wrap_s) << 0) | (wrap_mode(cso->wrap_t) << 3) | (wrap_mode(cso->wrap_r) << 6)); @@ -202,7 +204,19 @@ nv50_sampler_state_create(struct pipe_context *pipe, tsc[0] |= (nvgl_comparison_op(cso->compare_func) & 0x7); } - return (void *)tsc; + limit = CLAMP(cso->lod_bias, -16.0, 15.0); + tsc[1] |= ((int)(limit * 256.0) & 0x1fff) << 12; + + tsc[2] |= ((int)CLAMP(cso->max_lod, 0.0, 15.0) << 20) | + ((int)CLAMP(cso->min_lod, 0.0, 15.0) << 8); + + tsc[4] = fui(cso->border_color[0]); + tsc[5] = fui(cso->border_color[1]); + tsc[6] = fui(cso->border_color[2]); + tsc[7] = fui(cso->border_color[3]); + + sso->normalized = cso->normalized_coords; + return (void *)sso; } static void @@ -259,6 +273,8 @@ nv50_rasterizer_state_create(struct pipe_context *pipe, so_method(so, tesla, NV50TCL_SHADE_MODEL, 1); so_data (so, cso->flatshade ? NV50TCL_SHADE_MODEL_FLAT : NV50TCL_SHADE_MODEL_SMOOTH); + so_method(so, tesla, 0x1684, 1); + so_data (so, cso->flatshade_first ? 0 : 1); so_method(so, tesla, NV50TCL_LINE_WIDTH, 1); so_data (so, fui(cso->line_width)); @@ -395,35 +411,35 @@ nv50_depth_stencil_alpha_state_create(struct pipe_context *pipe, so_data (so, 0); } - /*XXX: yes, I know they're backwards.. header needs fixing */ + /* XXX: keep hex values until header is updated (names reversed) */ if (cso->stencil[0].enabled) { - so_method(so, tesla, NV50TCL_STENCIL_BACK_ENABLE, 5); + so_method(so, tesla, 0x1380, 8); so_data (so, 1); so_data (so, nvgl_stencil_op(cso->stencil[0].fail_op)); so_data (so, nvgl_stencil_op(cso->stencil[0].zfail_op)); so_data (so, nvgl_stencil_op(cso->stencil[0].zpass_op)); so_data (so, nvgl_comparison_op(cso->stencil[0].func)); - so_method(so, tesla, NV50TCL_STENCIL_BACK_FUNC_REF, 3); so_data (so, cso->stencil[0].ref_value); so_data (so, cso->stencil[0].writemask); so_data (so, cso->stencil[0].valuemask); } else { - so_method(so, tesla, NV50TCL_STENCIL_BACK_ENABLE, 1); + so_method(so, tesla, 0x1380, 1); so_data (so, 0); } if (cso->stencil[1].enabled) { - so_method(so, tesla, NV50TCL_STENCIL_FRONT_ENABLE, 8); + so_method(so, tesla, 0x1594, 5); so_data (so, 1); so_data (so, nvgl_stencil_op(cso->stencil[1].fail_op)); so_data (so, nvgl_stencil_op(cso->stencil[1].zfail_op)); so_data (so, nvgl_stencil_op(cso->stencil[1].zpass_op)); so_data (so, nvgl_comparison_op(cso->stencil[1].func)); + so_method(so, tesla, 0x0f54, 3); so_data (so, cso->stencil[1].ref_value); so_data (so, cso->stencil[1].writemask); so_data (so, cso->stencil[1].valuemask); } else { - so_method(so, tesla, NV50TCL_STENCIL_FRONT_ENABLE, 1); + so_method(so, tesla, 0x1594, 1); so_data (so, 0); } diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c index c13d3de1cb9..344c2cf6dde 100644 --- a/src/gallium/drivers/nv50/nv50_state_validate.c +++ b/src/gallium/drivers/nv50/nv50_state_validate.c @@ -32,6 +32,9 @@ nv50_state_validate_fb(struct nv50_context *nv50) unsigned i, w, h, gw = 0; for (i = 0; i < fb->nr_cbufs; i++) { + struct pipe_texture *pt = fb->cbufs[i]->texture; + struct nouveau_bo *bo = nv50_miptree(pt)->base.bo; + if (!gw) { w = fb->cbufs[i]->width; h = fb->cbufs[i]->height; @@ -46,26 +49,24 @@ nv50_state_validate_fb(struct nv50_context *nv50) so_data (so, fb->cbufs[i]->height); so_method(so, tesla, NV50TCL_RT_ADDRESS_HIGH(i), 5); - so_reloc (so, nv50_surface_buffer(fb->cbufs[i]), fb->cbufs[i]->offset, - NOUVEAU_BO_VRAM | NOUVEAU_BO_HIGH | - NOUVEAU_BO_RDWR, 0, 0); - so_reloc (so, nv50_surface_buffer(fb->cbufs[i]), fb->cbufs[i]->offset, - NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | - NOUVEAU_BO_RDWR, 0, 0); + so_reloc (so, bo, fb->cbufs[i]->offset, NOUVEAU_BO_VRAM | + NOUVEAU_BO_HIGH | NOUVEAU_BO_RDWR, 0, 0); + so_reloc (so, bo, fb->cbufs[i]->offset, NOUVEAU_BO_VRAM | + NOUVEAU_BO_LOW | NOUVEAU_BO_RDWR, 0, 0); switch (fb->cbufs[i]->format) { case PIPE_FORMAT_A8R8G8B8_UNORM: - so_data(so, 0xcf); + so_data(so, NV50TCL_RT_FORMAT_A8R8G8B8_UNORM); break; case PIPE_FORMAT_R5G6B5_UNORM: - so_data(so, 0xe8); + so_data(so, NV50TCL_RT_FORMAT_R5G6B5_UNORM); break; default: NOUVEAU_ERR("AIIII unknown format %s\n", pf_name(fb->cbufs[i]->format)); - so_data(so, 0xe6); + so_data(so, NV50TCL_RT_FORMAT_X8R8G8B8_UNORM); break; } - so_data(so, 0x00000000); + so_data(so, bo->tile_mode << 4); so_data(so, 0x00000000); so_method(so, tesla, 0x1224, 1); @@ -73,6 +74,9 @@ nv50_state_validate_fb(struct nv50_context *nv50) } if (fb->zsbuf) { + struct pipe_texture *pt = fb->zsbuf->texture; + struct nouveau_bo *bo = nv50_miptree(pt)->base.bo; + if (!gw) { w = fb->zsbuf->width; h = fb->zsbuf->height; @@ -83,31 +87,35 @@ nv50_state_validate_fb(struct nv50_context *nv50) } so_method(so, tesla, NV50TCL_ZETA_ADDRESS_HIGH, 5); - so_reloc (so, nv50_surface_buffer(fb->zsbuf), fb->zsbuf->offset, - NOUVEAU_BO_VRAM | NOUVEAU_BO_HIGH | - NOUVEAU_BO_RDWR, 0, 0); - so_reloc (so, nv50_surface_buffer(fb->zsbuf), fb->zsbuf->offset, - NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | - NOUVEAU_BO_RDWR, 0, 0); + so_reloc (so, bo, fb->zsbuf->offset, NOUVEAU_BO_VRAM | + NOUVEAU_BO_HIGH | NOUVEAU_BO_RDWR, 0, 0); + so_reloc (so, bo, fb->zsbuf->offset, NOUVEAU_BO_VRAM | + NOUVEAU_BO_LOW | NOUVEAU_BO_RDWR, 0, 0); switch (fb->zsbuf->format) { + case PIPE_FORMAT_Z32_FLOAT: + so_data(so, NV50TCL_ZETA_FORMAT_Z32_FLOAT); + break; case PIPE_FORMAT_Z24S8_UNORM: - so_data(so, 0x16); + so_data(so, NV50TCL_ZETA_FORMAT_Z24S8_UNORM); + break; + case PIPE_FORMAT_X8Z24_UNORM: + so_data(so, NV50TCL_ZETA_FORMAT_X8Z24_UNORM); break; - case PIPE_FORMAT_Z16_UNORM: - so_data(so, 0x15); + case PIPE_FORMAT_S8Z24_UNORM: + so_data(so, NV50TCL_ZETA_FORMAT_S8Z24_UNORM); break; default: NOUVEAU_ERR("AIIII unknown format %s\n", pf_name(fb->zsbuf->format)); - so_data(so, 0x16); + so_data(so, NV50TCL_ZETA_FORMAT_S8Z24_UNORM); break; } - so_data(so, 0x00000000); + so_data(so, bo->tile_mode << 4); so_data(so, 0x00000000); so_method(so, tesla, 0x1538, 1); so_data (so, 1); - so_method(so, tesla, 0x1228, 3); + so_method(so, tesla, NV50TCL_ZETA_HORIZ, 3); so_data (so, fb->zsbuf->width); so_data (so, fb->zsbuf->height); so_data (so, 0x00010001); @@ -116,12 +124,18 @@ nv50_state_validate_fb(struct nv50_context *nv50) so_method(so, tesla, NV50TCL_VIEWPORT_HORIZ, 2); so_data (so, w << 16); so_data (so, h << 16); - so_method(so, tesla, 0x0e04, 2); + /* set window lower left corner */ + so_method(so, tesla, NV50TCL_WINDOW_LEFT, 2); + so_data (so, 0); + so_data (so, 0); + /* set screen scissor rectangle */ + so_method(so, tesla, NV50TCL_SCREEN_SCISSOR_HORIZ, 2); so_data (so, w << 16); so_data (so, h << 16); - so_method(so, tesla, 0xdf8, 2); - so_data (so, 0); - so_data (so, h); + + /* we set scissors to framebuffer size when they're 'turned off' */ + nv50->dirty |= NV50_NEW_SCISSOR; + so_ref(NULL, &nv50->state.scissor); so_ref(so, &nv50->state.fb); so_ref(NULL, &so); @@ -131,48 +145,75 @@ static void nv50_state_emit(struct nv50_context *nv50) { struct nv50_screen *screen = nv50->screen; - struct nouveau_winsys *nvws = screen->nvws; + struct nouveau_channel *chan = screen->base.channel; if (nv50->pctx_id != screen->cur_pctx) { - nv50->state.dirty |= 0xffffffff; + if (nv50->state.fb) + nv50->state.dirty |= NV50_NEW_FRAMEBUFFER; + if (nv50->state.blend) + nv50->state.dirty |= NV50_NEW_BLEND; + if (nv50->state.zsa) + nv50->state.dirty |= NV50_NEW_ZSA; + if (nv50->state.vertprog) + nv50->state.dirty |= NV50_NEW_VERTPROG; + if (nv50->state.fragprog) + nv50->state.dirty |= NV50_NEW_FRAGPROG; + if (nv50->state.rast) + nv50->state.dirty |= NV50_NEW_RASTERIZER; + if (nv50->state.blend_colour) + nv50->state.dirty |= NV50_NEW_BLEND_COLOUR; + if (nv50->state.stipple) + nv50->state.dirty |= NV50_NEW_STIPPLE; + if (nv50->state.scissor) + nv50->state.dirty |= NV50_NEW_SCISSOR; + if (nv50->state.viewport) + nv50->state.dirty |= NV50_NEW_VIEWPORT; + if (nv50->state.tsc_upload) + nv50->state.dirty |= NV50_NEW_SAMPLER; + if (nv50->state.tic_upload) + nv50->state.dirty |= NV50_NEW_TEXTURE; + if (nv50->state.vtxfmt && nv50->state.vtxbuf) + nv50->state.dirty |= NV50_NEW_ARRAYS; screen->cur_pctx = nv50->pctx_id; } if (nv50->state.dirty & NV50_NEW_FRAMEBUFFER) - so_emit(nvws, nv50->state.fb); + so_emit(chan, nv50->state.fb); if (nv50->state.dirty & NV50_NEW_BLEND) - so_emit(nvws, nv50->state.blend); + so_emit(chan, nv50->state.blend); if (nv50->state.dirty & NV50_NEW_ZSA) - so_emit(nvws, nv50->state.zsa); + so_emit(chan, nv50->state.zsa); if (nv50->state.dirty & NV50_NEW_VERTPROG) - so_emit(nvws, nv50->state.vertprog); + so_emit(chan, nv50->state.vertprog); if (nv50->state.dirty & NV50_NEW_FRAGPROG) - so_emit(nvws, nv50->state.fragprog); + so_emit(chan, nv50->state.fragprog); if (nv50->state.dirty & NV50_NEW_RASTERIZER) - so_emit(nvws, nv50->state.rast); + so_emit(chan, nv50->state.rast); if (nv50->state.dirty & NV50_NEW_BLEND_COLOUR) - so_emit(nvws, nv50->state.blend_colour); + so_emit(chan, nv50->state.blend_colour); if (nv50->state.dirty & NV50_NEW_STIPPLE) - so_emit(nvws, nv50->state.stipple); + so_emit(chan, nv50->state.stipple); if (nv50->state.dirty & NV50_NEW_SCISSOR) - so_emit(nvws, nv50->state.scissor); + so_emit(chan, nv50->state.scissor); if (nv50->state.dirty & NV50_NEW_VIEWPORT) - so_emit(nvws, nv50->state.viewport); + so_emit(chan, nv50->state.viewport); if (nv50->state.dirty & NV50_NEW_SAMPLER) - so_emit(nvws, nv50->state.tsc_upload); + so_emit(chan, nv50->state.tsc_upload); if (nv50->state.dirty & NV50_NEW_TEXTURE) - so_emit(nvws, nv50->state.tic_upload); + so_emit(chan, nv50->state.tic_upload); if (nv50->state.dirty & NV50_NEW_ARRAYS) { - so_emit(nvws, nv50->state.vtxfmt); - so_emit(nvws, nv50->state.vtxbuf); + so_emit(chan, nv50->state.vtxfmt); + so_emit(chan, nv50->state.vtxbuf); + if (nv50->state.vtxattr) + so_emit(chan, nv50->state.vtxattr); } nv50->state.dirty = 0; - so_emit_reloc_markers(nvws, nv50->state.fb); - so_emit_reloc_markers(nvws, nv50->state.vertprog); - so_emit_reloc_markers(nvws, nv50->state.fragprog); - so_emit_reloc_markers(nvws, nv50->state.vtxbuf); - so_emit_reloc_markers(nvws, nv50->screen->static_init); + so_emit_reloc_markers(chan, nv50->state.fb); + so_emit_reloc_markers(chan, nv50->state.vertprog); + so_emit_reloc_markers(chan, nv50->state.fragprog); + so_emit_reloc_markers(chan, nv50->state.vtxbuf); + so_emit_reloc_markers(chan, nv50->screen->static_init); } boolean @@ -230,13 +271,13 @@ nv50_state_validate(struct nv50_context *nv50) nv50->state.scissor_enabled = rast->scissor; so = so_new(3, 0); - so_method(so, tesla, 0x0ff4, 2); + so_method(so, tesla, NV50TCL_SCISSOR_HORIZ, 2); if (nv50->state.scissor_enabled) { - so_data(so, ((s->maxx - s->minx) << 16) | s->minx); - so_data(so, ((s->maxy - s->miny) << 16) | s->miny); + so_data(so, (s->maxx << 16) | s->minx); + so_data(so, (s->maxy << 16) | s->miny); } else { - so_data(so, (8192 << 16)); - so_data(so, (8192 << 16)); + so_data(so, (nv50->framebuffer.width << 16)); + so_data(so, (nv50->framebuffer.height << 16)); } so_ref(so, &nv50->state.scissor); so_ref(NULL, &so); @@ -260,20 +301,22 @@ scissor_uptodate: so = so_new(12, 0); if (!bypass) { - so_method(so, tesla, NV50TCL_VIEWPORT_UNK1(0), 3); + so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE(0), 3); so_data (so, fui(nv50->viewport.translate[0])); so_data (so, fui(nv50->viewport.translate[1])); so_data (so, fui(nv50->viewport.translate[2])); - so_method(so, tesla, NV50TCL_VIEWPORT_UNK0(0), 3); + so_method(so, tesla, NV50TCL_VIEWPORT_SCALE(0), 3); so_data (so, fui(nv50->viewport.scale[0])); - so_data (so, fui(-nv50->viewport.scale[1])); + so_data (so, fui(nv50->viewport.scale[1])); so_data (so, fui(nv50->viewport.scale[2])); - so_method(so, tesla, 0x192c, 1); + + so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1); so_data (so, 1); + /* no idea what 0f90 does */ so_method(so, tesla, 0x0f90, 1); so_data (so, 0); } else { - so_method(so, tesla, 0x192c, 1); + so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1); so_data (so, 0); so_method(so, tesla, 0x0f90, 1); so_data (so, 1); @@ -289,16 +332,17 @@ viewport_uptodate: int i; so = so_new(nv50->sampler_nr * 8 + 3, 0); - so_method(so, tesla, 0x0f00, 1); + so_method(so, tesla, NV50TCL_CB_ADDR, 1); so_data (so, NV50_CB_TSC); - so_method(so, tesla, 0x40000f04, nv50->sampler_nr * 8); + so_method(so, tesla, NV50TCL_CB_DATA(0) | 0x40000000, + nv50->sampler_nr * 8); for (i = 0; i < nv50->sampler_nr; i++) - so_datap (so, nv50->sampler[i], 8); + so_datap (so, nv50->sampler[i]->tsc, 8); so_ref(so, &nv50->state.tsc_upload); so_ref(NULL, &so); } - if (nv50->dirty & NV50_NEW_TEXTURE) + if (nv50->dirty & (NV50_NEW_TEXTURE | NV50_NEW_SAMPLER)) nv50_tex_validate(nv50); if (nv50->dirty & NV50_NEW_ARRAYS) diff --git a/src/gallium/drivers/nv50/nv50_surface.c b/src/gallium/drivers/nv50/nv50_surface.c index 0cc5168144d..b266324f58d 100644 --- a/src/gallium/drivers/nv50/nv50_surface.c +++ b/src/gallium/drivers/nv50/nv50_surface.c @@ -35,14 +35,13 @@ nv50_format(enum pipe_format format) { switch (format) { case PIPE_FORMAT_A8R8G8B8_UNORM: - case PIPE_FORMAT_Z24S8_UNORM: - return NV50_2D_DST_FORMAT_32BPP; + return NV50_2D_DST_FORMAT_A8R8G8B8_UNORM; case PIPE_FORMAT_X8R8G8B8_UNORM: - return NV50_2D_DST_FORMAT_24BPP; + return NV50_2D_DST_FORMAT_X8R8G8B8_UNORM; case PIPE_FORMAT_R5G6B5_UNORM: - return NV50_2D_DST_FORMAT_16BPP; + return NV50_2D_DST_FORMAT_R5G6B5_UNORM; case PIPE_FORMAT_A8_UNORM: - return NV50_2D_DST_FORMAT_8BPP; + return NV50_2D_DST_FORMAT_R8_UNORM; default: return -1; } @@ -52,21 +51,17 @@ static int nv50_surface_set(struct nv50_screen *screen, struct pipe_surface *ps, int dst) { struct nv50_miptree *mt = nv50_miptree(ps->texture); - struct nouveau_channel *chan = screen->nvws->channel; + struct nouveau_channel *chan = screen->eng2d->channel; struct nouveau_grobj *eng2d = screen->eng2d; - struct nouveau_bo *bo; + struct nouveau_bo *bo = nv50_miptree(ps->texture)->base.bo; int format, mthd = dst ? NV50_2D_DST_FORMAT : NV50_2D_SRC_FORMAT; int flags = NOUVEAU_BO_VRAM | (dst ? NOUVEAU_BO_WR : NOUVEAU_BO_RD); - - bo = screen->nvws->get_bo(nv50_miptree(ps->texture)->buffer); - if (!bo) - return 1; format = nv50_format(ps->format); if (format < 0) return 1; - if (!bo->tiled) { + if (!bo->tile_flags) { BEGIN_RING(chan, eng2d, mthd, 2); OUT_RING (chan, format); OUT_RING (chan, 1); @@ -80,7 +75,7 @@ nv50_surface_set(struct nv50_screen *screen, struct pipe_surface *ps, int dst) BEGIN_RING(chan, eng2d, mthd, 5); OUT_RING (chan, format); OUT_RING (chan, 0); - OUT_RING (chan, 0); + OUT_RING (chan, bo->tile_mode << 4); OUT_RING (chan, 1); OUT_RING (chan, 0); BEGIN_RING(chan, eng2d, mthd + 0x18, 4); @@ -108,7 +103,7 @@ nv50_surface_do_copy(struct nv50_screen *screen, struct pipe_surface *dst, int dx, int dy, struct pipe_surface *src, int sx, int sy, int w, int h) { - struct nouveau_channel *chan = screen->nvws->channel; + struct nouveau_channel *chan = screen->eng2d->channel; struct nouveau_grobj *eng2d = screen->eng2d; int ret; @@ -149,7 +144,7 @@ nv50_surface_copy(struct pipe_context *pipe, struct pipe_surface *src, unsigned srcx, unsigned srcy, unsigned width, unsigned height) { - struct nv50_context *nv50 = (struct nv50_context *)pipe; + struct nv50_context *nv50 = nv50_context(pipe); struct nv50_screen *screen = nv50->screen; assert(src->format == dest->format); @@ -163,9 +158,9 @@ nv50_surface_fill(struct pipe_context *pipe, struct pipe_surface *dest, unsigned destx, unsigned desty, unsigned width, unsigned height, unsigned value) { - struct nv50_context *nv50 = (struct nv50_context *)pipe; + struct nv50_context *nv50 = nv50_context(pipe); struct nv50_screen *screen = nv50->screen; - struct nouveau_channel *chan = screen->nvws->channel; + struct nouveau_channel *chan = screen->eng2d->channel; struct nouveau_grobj *eng2d = screen->eng2d; int format, ret; diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c index 223c8a3a456..033cb50c115 100644 --- a/src/gallium/drivers/nv50/nv50_tex.c +++ b/src/gallium/drivers/nv50/nv50_tex.c @@ -26,35 +26,36 @@ #include "nouveau/nouveau_stateobj.h" static int -nv50_tex_construct(struct nouveau_stateobj *so, struct nv50_miptree *mt) +nv50_tex_construct(struct nv50_context *nv50, struct nouveau_stateobj *so, + struct nv50_miptree *mt, int unit) { - switch (mt->base.format) { + switch (mt->base.base.format) { case PIPE_FORMAT_A8R8G8B8_UNORM: so_data(so, NV50TIC_0_0_MAPA_C3 | NV50TIC_0_0_TYPEA_UNORM | - NV50TIC_0_0_MAPR_C0 | NV50TIC_0_0_TYPER_UNORM | + NV50TIC_0_0_MAPR_C2 | NV50TIC_0_0_TYPER_UNORM | NV50TIC_0_0_MAPG_C1 | NV50TIC_0_0_TYPEG_UNORM | - NV50TIC_0_0_MAPB_C2 | NV50TIC_0_0_TYPEB_UNORM | + NV50TIC_0_0_MAPB_C0 | NV50TIC_0_0_TYPEB_UNORM | NV50TIC_0_0_FMT_8_8_8_8); break; case PIPE_FORMAT_A1R5G5B5_UNORM: so_data(so, NV50TIC_0_0_MAPA_C3 | NV50TIC_0_0_TYPEA_UNORM | - NV50TIC_0_0_MAPR_C0 | NV50TIC_0_0_TYPER_UNORM | + NV50TIC_0_0_MAPR_C2 | NV50TIC_0_0_TYPER_UNORM | NV50TIC_0_0_MAPG_C1 | NV50TIC_0_0_TYPEG_UNORM | - NV50TIC_0_0_MAPB_C2 | NV50TIC_0_0_TYPEB_UNORM | + NV50TIC_0_0_MAPB_C0 | NV50TIC_0_0_TYPEB_UNORM | NV50TIC_0_0_FMT_1_5_5_5); break; case PIPE_FORMAT_A4R4G4B4_UNORM: so_data(so, NV50TIC_0_0_MAPA_C3 | NV50TIC_0_0_TYPEA_UNORM | - NV50TIC_0_0_MAPR_C0 | NV50TIC_0_0_TYPER_UNORM | + NV50TIC_0_0_MAPR_C2 | NV50TIC_0_0_TYPER_UNORM | NV50TIC_0_0_MAPG_C1 | NV50TIC_0_0_TYPEG_UNORM | - NV50TIC_0_0_MAPB_C2 | NV50TIC_0_0_TYPEB_UNORM | + NV50TIC_0_0_MAPB_C0 | NV50TIC_0_0_TYPEB_UNORM | NV50TIC_0_0_FMT_4_4_4_4); break; case PIPE_FORMAT_R5G6B5_UNORM: so_data(so, NV50TIC_0_0_MAPA_ONE | NV50TIC_0_0_TYPEA_UNORM | - NV50TIC_0_0_MAPR_C0 | NV50TIC_0_0_TYPER_UNORM | + NV50TIC_0_0_MAPR_C2 | NV50TIC_0_0_TYPER_UNORM | NV50TIC_0_0_MAPG_C1 | NV50TIC_0_0_TYPEG_UNORM | - NV50TIC_0_0_MAPB_C2 | NV50TIC_0_0_TYPEB_UNORM | + NV50TIC_0_0_MAPB_C0 | NV50TIC_0_0_TYPEB_UNORM | NV50TIC_0_0_FMT_5_6_5); break; case PIPE_FORMAT_L8_UNORM: @@ -117,15 +118,18 @@ nv50_tex_construct(struct nouveau_stateobj *so, struct nv50_miptree *mt) return 1; } - so_reloc(so, mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | + so_reloc(so, mt->base.bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | NOUVEAU_BO_RD, 0, 0); - so_data (so, 0xd0005000); + if (nv50->sampler[unit]->normalized) + so_data (so, 0xd0005000 | mt->base.bo->tile_mode << 22); + else + so_data (so, 0x5001d000 | mt->base.bo->tile_mode << 22); so_data (so, 0x00300000); - so_data (so, mt->base.width[0]); - so_data (so, (mt->base.depth[0] << 16) | mt->base.height[0]); + so_data (so, mt->base.base.width[0]); + so_data (so, (mt->base.base.last_level << 28) | + (mt->base.base.depth[0] << 16) | mt->base.base.height[0]); so_data (so, 0x03000000); - so_reloc(so, mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_HIGH | - NOUVEAU_BO_RD, 0, 0); + so_data (so, mt->base.base.last_level << 4); return 0; } @@ -135,23 +139,38 @@ nv50_tex_validate(struct nv50_context *nv50) { struct nouveau_grobj *tesla = nv50->screen->tesla; struct nouveau_stateobj *so; - int unit; + int unit, push; - so = so_new(nv50->miptree_nr * 8 + 3, nv50->miptree_nr * 2); - so_method(so, tesla, 0x0f00, 1); + push = nv50->miptree_nr * 9 + 2; + push += MAX2(nv50->miptree_nr, nv50->state.miptree_nr) * 2; + + so = so_new(push, nv50->miptree_nr * 2); + so_method(so, tesla, NV50TCL_CB_ADDR, 1); so_data (so, NV50_CB_TIC); - so_method(so, tesla, 0x40000f04, nv50->miptree_nr * 8); for (unit = 0; unit < nv50->miptree_nr; unit++) { struct nv50_miptree *mt = nv50->miptree[unit]; - if (nv50_tex_construct(so, mt)) { + so_method(so, tesla, NV50TCL_CB_DATA(0) | 0x40000000, 8); + if (nv50_tex_construct(nv50, so, mt, unit)) { NOUVEAU_ERR("failed tex validate\n"); so_ref(NULL, &so); return; } + + so_method(so, tesla, NV50TCL_SET_SAMPLER_TEX, 1); + so_data (so, (unit << NV50TCL_SET_SAMPLER_TEX_TIC_SHIFT) | + (unit << NV50TCL_SET_SAMPLER_TEX_SAMPLER_SHIFT) | + NV50TCL_SET_SAMPLER_TEX_VALID); + } + + for (; unit < nv50->state.miptree_nr; unit++) { + so_method(so, tesla, NV50TCL_SET_SAMPLER_TEX, 1); + so_data (so, + (unit << NV50TCL_SET_SAMPLER_TEX_SAMPLER_SHIFT) | 0); } so_ref(so, &nv50->state.tic_upload); so_ref(NULL, &so); + nv50->state.miptree_nr = nv50->miptree_nr; } diff --git a/src/gallium/drivers/nv50/nv50_texture.h b/src/gallium/drivers/nv50/nv50_texture.h index aca622c73b1..207fb039f70 100644 --- a/src/gallium/drivers/nv50/nv50_texture.h +++ b/src/gallium/drivers/nv50/nv50_texture.h @@ -14,13 +14,13 @@ #define NV50TIC_0_0_MAPA_C2 0x20000000 #define NV50TIC_0_0_MAPA_C3 0x28000000 #define NV50TIC_0_0_MAPA_ONE 0x38000000 -#define NV50TIC_0_0_MAPR_MASK 0x07000000 -#define NV50TIC_0_0_MAPR_ZERO 0x00000000 -#define NV50TIC_0_0_MAPR_C0 0x02000000 -#define NV50TIC_0_0_MAPR_C1 0x03000000 -#define NV50TIC_0_0_MAPR_C2 0x04000000 -#define NV50TIC_0_0_MAPR_C3 0x05000000 -#define NV50TIC_0_0_MAPR_ONE 0x07000000 +#define NV50TIC_0_0_MAPB_MASK 0x07000000 +#define NV50TIC_0_0_MAPB_ZERO 0x00000000 +#define NV50TIC_0_0_MAPB_C0 0x02000000 +#define NV50TIC_0_0_MAPB_C1 0x03000000 +#define NV50TIC_0_0_MAPB_C2 0x04000000 +#define NV50TIC_0_0_MAPB_C3 0x05000000 +#define NV50TIC_0_0_MAPB_ONE 0x07000000 #define NV50TIC_0_0_MAPG_MASK 0x00e00000 #define NV50TIC_0_0_MAPG_ZERO 0x00000000 #define NV50TIC_0_0_MAPG_C0 0x00400000 @@ -28,31 +28,49 @@ #define NV50TIC_0_0_MAPG_C2 0x00800000 #define NV50TIC_0_0_MAPG_C3 0x00a00000 #define NV50TIC_0_0_MAPG_ONE 0x00e00000 -#define NV50TIC_0_0_MAPB_MASK 0x001c0000 -#define NV50TIC_0_0_MAPB_ZERO 0x00000000 -#define NV50TIC_0_0_MAPB_C0 0x00080000 -#define NV50TIC_0_0_MAPB_C1 0x000c0000 -#define NV50TIC_0_0_MAPB_C2 0x00100000 -#define NV50TIC_0_0_MAPB_C3 0x00140000 -#define NV50TIC_0_0_MAPB_ONE 0x001c0000 +#define NV50TIC_0_0_MAPR_MASK 0x001c0000 +#define NV50TIC_0_0_MAPR_ZERO 0x00000000 +#define NV50TIC_0_0_MAPR_C0 0x00080000 +#define NV50TIC_0_0_MAPR_C1 0x000c0000 +#define NV50TIC_0_0_MAPR_C2 0x00100000 +#define NV50TIC_0_0_MAPR_C3 0x00140000 +#define NV50TIC_0_0_MAPR_ONE 0x001c0000 #define NV50TIC_0_0_TYPEA_MASK 0x00038000 #define NV50TIC_0_0_TYPEA_UNORM 0x00010000 -#define NV50TIC_0_0_TYPER_MASK 0x00007000 -#define NV50TIC_0_0_TYPER_UNORM 0x00002000 +#define NV50TIC_0_0_TYPEA_SNORM 0x00008000 +#define NV50TIC_0_0_TYPEA_FLOAT 0x00038000 +#define NV50TIC_0_0_TYPEB_MASK 0x00007000 +#define NV50TIC_0_0_TYPEB_UNORM 0x00002000 +#define NV50TIC_0_0_TYPEB_SNORM 0x00001000 +#define NV50TIC_0_0_TYPEB_FLOAT 0x00007000 #define NV50TIC_0_0_TYPEG_MASK 0x00000e00 #define NV50TIC_0_0_TYPEG_UNORM 0x00000400 -#define NV50TIC_0_0_TYPEB_MASK 0x000001c0 -#define NV50TIC_0_0_TYPEB_UNORM 0x00000080 -#define NV50TIC_0_0_FMT_MASK 0x0000003c +#define NV50TIC_0_0_TYPEG_SNORM 0x00000200 +#define NV50TIC_0_0_TYPEG_FLOAT 0x00000e00 +#define NV50TIC_0_0_TYPER_MASK 0x000001c0 +#define NV50TIC_0_0_TYPER_UNORM 0x00000080 +#define NV50TIC_0_0_TYPER_SNORM 0x00000040 +#define NV50TIC_0_0_TYPER_FLOAT 0x000001c0 +#define NV50TIC_0_0_FMT_MASK 0x0000003f +#define NV50TIC_0_0_FMT_32_32_32_32 0x00000001 +#define NV50TIC_0_0_FMT_16_16_16_16 0x00000003 +#define NV50TIC_0_0_FMT_32_32 0x00000004 #define NV50TIC_0_0_FMT_8_8_8_8 0x00000008 +#define NV50TIC_0_0_FMT_2_10_10_10 0x00000009 +#define NV50TIC_0_0_FMT_32 0x0000000f #define NV50TIC_0_0_FMT_4_4_4_4 0x00000012 -#define NV50TIC_0_0_FMT_1_5_5_5 0x00000013 +/* #define NV50TIC_0_0_FMT_1_5_5_5 0x00000013 */ +#define NV50TIC_0_0_FMT_1_5_5_5 0x00000014 #define NV50TIC_0_0_FMT_5_6_5 0x00000015 #define NV50TIC_0_0_FMT_8_8 0x00000018 +#define NV50TIC_0_0_FMT_16 0x0000001b #define NV50TIC_0_0_FMT_8 0x0000001d +#define NV50TIC_0_0_FMT_10_11_11 0x00000021 #define NV50TIC_0_0_FMT_DXT1 0x00000024 #define NV50TIC_0_0_FMT_DXT3 0x00000025 #define NV50TIC_0_0_FMT_DXT5 0x00000026 +#define NV50TIC_0_0_FMT_RGTC1 0x00000027 +#define NV50TIC_0_0_FMT_RGTC2 0x00000028 #define NV50TIC_0_1_OFFSET_LOW_MASK 0xffffffff #define NV50TIC_0_1_OFFSET_LOW_SHIFT 0 @@ -102,6 +120,7 @@ #define NV50TSC_1_0_WRAPR_MIRROR_CLAMP_TO_EDGE 0x00000140 #define NV50TSC_1_0_WRAPR_MIRROR_CLAMP_TO_BORDER 0x00000180 #define NV50TSC_1_0_WRAPR_MIRROR_CLAMP 0x000001c0 +#define NV50TSC_1_0_MAX_ANISOTROPY_MASK 0x00700000 #define NV50TSC_1_1_MAGF_MASK 0x00000003 #define NV50TSC_1_1_MAGF_NEAREST 0x00000001 @@ -113,17 +132,19 @@ #define NV50TSC_1_1_MIPF_NONE 0x00000040 #define NV50TSC_1_1_MIPF_NEAREST 0x00000080 #define NV50TSC_1_1_MIPF_LINEAR 0x000000c0 +#define NV50TSC_1_1_LOD_BIAS_MASK 0x01fff000 -#define NV50TSC_1_2_UNKNOWN_MASK 0xffffffff +#define NV50TSC_1_2_MIN_LOD_MASK 0x00000f00 +#define NV50TSC_1_2_MAX_LOD_MASK 0x00f00000 #define NV50TSC_1_3_UNKNOWN_MASK 0xffffffff -#define NV50TSC_1_4_UNKNOWN_MASK 0xffffffff +#define NV50TSC_1_4_BORDER_COLOR_RED_MASK 0xffffffff -#define NV50TSC_1_5_UNKNOWN_MASK 0xffffffff +#define NV50TSC_1_5_BORDER_COLOR_GREEN_MASK 0xffffffff -#define NV50TSC_1_6_UNKNOWN_MASK 0xffffffff +#define NV50TSC_1_6_BORDER_COLOR_BLUE_MASK 0xffffffff -#define NV50TSC_1_7_UNKNOWN_MASK 0xffffffff +#define NV50TSC_1_7_BORDER_COLOR_ALPHA_MASK 0xffffffff #endif diff --git a/src/gallium/drivers/nv50/nv50_transfer.c b/src/gallium/drivers/nv50/nv50_transfer.c index 747195b4f63..e9c3562194b 100644 --- a/src/gallium/drivers/nv50/nv50_transfer.c +++ b/src/gallium/drivers/nv50/nv50_transfer.c @@ -6,8 +6,9 @@ struct nv50_transfer { struct pipe_transfer base; - struct pipe_buffer *buffer; - struct nv50_miptree_level *level; + struct nouveau_bo *bo; + unsigned level_offset; + unsigned level_tiling; int level_pitch; int level_width; int level_height; @@ -16,51 +17,57 @@ struct nv50_transfer { }; static void -nv50_transfer_rect_m2mf(struct pipe_screen *pscreen, struct pipe_buffer *src, - int src_pitch, int sx, int sy, int sw, int sh, - struct pipe_buffer *dst, int dst_pitch, int dx, int dy, - int dw, int dh, int cpp, int width, int height, +nv50_transfer_rect_m2mf(struct pipe_screen *pscreen, + struct nouveau_bo *src_bo, unsigned src_offset, + int src_pitch, unsigned src_tile_mode, + int sx, int sy, int sw, int sh, + struct nouveau_bo *dst_bo, unsigned dst_offset, + int dst_pitch, unsigned dst_tile_mode, + int dx, int dy, int dw, int dh, + int cpp, int width, int height, unsigned src_reloc, unsigned dst_reloc) { struct nv50_screen *screen = nv50_screen(pscreen); - struct nouveau_winsys *nvws = screen->nvws; - struct nouveau_channel *chan = nvws->channel; + struct nouveau_channel *chan = screen->m2mf->channel; struct nouveau_grobj *m2mf = screen->m2mf; - struct nouveau_bo *src_bo = nvws->get_bo(src); - struct nouveau_bo *dst_bo = nvws->get_bo(dst); - unsigned src_offset = 0, dst_offset = 0; src_reloc |= NOUVEAU_BO_RD; dst_reloc |= NOUVEAU_BO_WR; WAIT_RING (chan, 14); - if (!src_bo->tiled) { - BEGIN_RING(chan, m2mf, 0x0200, 1); + if (!src_bo->tile_flags) { + BEGIN_RING(chan, m2mf, + NV50_MEMORY_TO_MEMORY_FORMAT_LINEAR_IN, 1); OUT_RING (chan, 1); - BEGIN_RING(chan, m2mf, 0x0314, 1); + BEGIN_RING(chan, m2mf, + NV50_MEMORY_TO_MEMORY_FORMAT_PITCH_IN, 1); OUT_RING (chan, src_pitch); - src_offset = (sy * src_pitch) + (sx * cpp); + src_offset += (sy * src_pitch) + (sx * cpp); } else { - BEGIN_RING(chan, m2mf, 0x0200, 6); - OUT_RING (chan, 0); + BEGIN_RING(chan, m2mf, + NV50_MEMORY_TO_MEMORY_FORMAT_LINEAR_IN, 6); OUT_RING (chan, 0); + OUT_RING (chan, src_tile_mode << 4); OUT_RING (chan, sw * cpp); OUT_RING (chan, sh); OUT_RING (chan, 1); OUT_RING (chan, 0); } - if (!dst_bo->tiled) { - BEGIN_RING(chan, m2mf, 0x021c, 1); + if (!dst_bo->tile_flags) { + BEGIN_RING(chan, m2mf, + NV50_MEMORY_TO_MEMORY_FORMAT_LINEAR_OUT, 1); OUT_RING (chan, 1); - BEGIN_RING(chan, m2mf, 0x0318, 1); + BEGIN_RING(chan, m2mf, + NV50_MEMORY_TO_MEMORY_FORMAT_PITCH_OUT, 1); OUT_RING (chan, dst_pitch); - dst_offset = (dy * dst_pitch) + (dx * cpp); + dst_offset += (dy * dst_pitch) + (dx * cpp); } else { - BEGIN_RING(chan, m2mf, 0x021c, 6); - OUT_RING (chan, 0); + BEGIN_RING(chan, m2mf, + NV50_MEMORY_TO_MEMORY_FORMAT_LINEAR_OUT, 6); OUT_RING (chan, 0); + OUT_RING (chan, dst_tile_mode << 4); OUT_RING (chan, dw * cpp); OUT_RING (chan, dh); OUT_RING (chan, 1); @@ -71,25 +78,30 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen, struct pipe_buffer *src, int line_count = height > 2047 ? 2047 : height; WAIT_RING (chan, 15); - BEGIN_RING(chan, m2mf, 0x0238, 2); + BEGIN_RING(chan, m2mf, + NV50_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN_HIGH, 2); OUT_RELOCh(chan, src_bo, src_offset, src_reloc); OUT_RELOCh(chan, dst_bo, dst_offset, dst_reloc); - BEGIN_RING(chan, m2mf, 0x030c, 2); + BEGIN_RING(chan, m2mf, + NV50_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN, 2); OUT_RELOCl(chan, src_bo, src_offset, src_reloc); OUT_RELOCl(chan, dst_bo, dst_offset, dst_reloc); - if (src_bo->tiled) { - BEGIN_RING(chan, m2mf, 0x0218, 1); - OUT_RING (chan, (dy << 16) | sx); + if (src_bo->tile_flags) { + BEGIN_RING(chan, m2mf, + NV50_MEMORY_TO_MEMORY_FORMAT_TILING_POSITION_IN, 1); + OUT_RING (chan, (sy << 16) | sx); } else { src_offset += (line_count * src_pitch); } - if (dst_bo->tiled) { - BEGIN_RING(chan, m2mf, 0x0234, 1); - OUT_RING (chan, (sy << 16) | dx); + if (dst_bo->tile_flags) { + BEGIN_RING(chan, m2mf, + NV50_MEMORY_TO_MEMORY_FORMAT_TILING_POSITION_OUT, 1); + OUT_RING (chan, (dy << 16) | dx); } else { dst_offset += (line_count * dst_pitch); } - BEGIN_RING(chan, m2mf, 0x031c, 4); + BEGIN_RING(chan, m2mf, + NV50_MEMORY_TO_MEMORY_FORMAT_LINE_LENGTH_IN, 4); OUT_RING (chan, width * cpp); OUT_RING (chan, line_count); OUT_RING (chan, 0x00000101); @@ -108,10 +120,12 @@ nv50_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt, enum pipe_transfer_usage usage, unsigned x, unsigned y, unsigned w, unsigned h) { + struct nouveau_device *dev = nouveau_screen(pscreen)->device; struct nv50_miptree *mt = nv50_miptree(pt); struct nv50_miptree_level *lvl = &mt->level[level]; struct nv50_transfer *tx; unsigned image = 0; + int ret; if (pt->target == PIPE_TEXTURE_CUBE) image = face; @@ -133,20 +147,27 @@ nv50_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt, tx->base.stride = (w * pt->block.size); tx->base.usage = usage; - tx->level = lvl; tx->level_pitch = lvl->pitch; - tx->level_width = mt->base.width[level]; - tx->level_height = mt->base.height[level]; + tx->level_width = mt->base.base.width[level]; + tx->level_height = mt->base.base.height[level]; + tx->level_offset = lvl->image_offset[image]; + tx->level_tiling = lvl->tile_mode; tx->level_x = x; tx->level_y = y; - tx->buffer = - pipe_buffer_create(pscreen, 0, NOUVEAU_BUFFER_USAGE_TRANSFER, - w * tx->base.block.size * h); + ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, + w * pt->block.size * h, &tx->bo); + if (ret) { + FREE(tx); + return NULL; + } if (usage != PIPE_TRANSFER_WRITE) { - nv50_transfer_rect_m2mf(pscreen, mt->buffer, tx->level_pitch, - x, y, tx->level_width, tx->level_height, - tx->buffer, tx->base.stride, 0, 0, + nv50_transfer_rect_m2mf(pscreen, mt->base.bo, tx->level_offset, + tx->level_pitch, tx->level_tiling, + x, y, + tx->level_width, tx->level_height, + tx->bo, 0, tx->base.stride, + tx->bo->tile_mode, 0, 0, tx->base.width, tx->base.height, tx->base.block.size, w, h, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART, @@ -164,17 +185,20 @@ nv50_transfer_del(struct pipe_transfer *ptx) if (ptx->usage != PIPE_TRANSFER_READ) { struct pipe_screen *pscreen = ptx->texture->screen; - nv50_transfer_rect_m2mf(pscreen, tx->buffer, tx->base.stride, - 0, 0, tx->base.width, tx->base.height, - mt->buffer, tx->level_pitch, + nv50_transfer_rect_m2mf(pscreen, tx->bo, 0, tx->base.stride, + tx->bo->tile_mode, 0, 0, + tx->base.width, tx->base.height, + mt->base.bo, tx->level_offset, + tx->level_pitch, tx->level_tiling, tx->level_x, tx->level_y, tx->level_width, tx->level_height, tx->base.block.size, tx->base.width, - tx->base.height, NOUVEAU_BO_GART, - NOUVEAU_BO_VRAM | NOUVEAU_BO_GART); + tx->base.height, + NOUVEAU_BO_GART, NOUVEAU_BO_VRAM | + NOUVEAU_BO_GART); } - pipe_buffer_reference(&tx->buffer, NULL); + nouveau_bo_ref(NULL, &tx->bo); pipe_texture_reference(&ptx->texture, NULL); FREE(ptx); } @@ -184,13 +208,17 @@ nv50_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx) { struct nv50_transfer *tx = (struct nv50_transfer *)ptx; unsigned flags = 0; + int ret; if (ptx->usage & PIPE_TRANSFER_WRITE) - flags |= PIPE_BUFFER_USAGE_CPU_WRITE; + flags |= NOUVEAU_BO_WR; if (ptx->usage & PIPE_TRANSFER_READ) - flags |= PIPE_BUFFER_USAGE_CPU_READ; + flags |= NOUVEAU_BO_RD; - return pipe_buffer_map(pscreen, tx->buffer, flags); + ret = nouveau_bo_map(tx->bo, flags); + if (ret) + return NULL; + return tx->bo->map; } static void @@ -198,7 +226,7 @@ nv50_transfer_unmap(struct pipe_screen *pscreen, struct pipe_transfer *ptx) { struct nv50_transfer *tx = (struct nv50_transfer *)ptx; - pipe_buffer_unmap(pscreen, tx->buffer); + nouveau_bo_unmap(tx->bo); } void diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c index 0749c906914..eeed148c7b9 100644 --- a/src/gallium/drivers/nv50/nv50_vbo.c +++ b/src/gallium/drivers/nv50/nv50_vbo.c @@ -22,6 +22,7 @@ #include "pipe/p_context.h" #include "pipe/p_state.h" +#include "pipe/p_inlines.h" #include "nv50_context.h" @@ -48,12 +49,88 @@ nv50_prim(unsigned mode) return NV50TCL_VERTEX_BEGIN_POINTS; } +static INLINE uint32_t +nv50_vbo_type_to_hw(unsigned type) +{ + switch (type) { + case PIPE_FORMAT_TYPE_FLOAT: + return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_FLOAT; + case PIPE_FORMAT_TYPE_UNORM: + return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_UNORM; + case PIPE_FORMAT_TYPE_SNORM: + return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SNORM; + case PIPE_FORMAT_TYPE_USCALED: + return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_USCALED; + case PIPE_FORMAT_TYPE_SSCALED: + return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SSCALED; + /* + case PIPE_FORMAT_TYPE_UINT: + return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_UINT; + case PIPE_FORMAT_TYPE_SINT: + return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SINT; */ + default: + return 0; + } +} + +static INLINE uint32_t +nv50_vbo_size_to_hw(unsigned size, unsigned nr_c) +{ + static const uint32_t hw_values[] = { + 0, 0, 0, 0, + NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_8, + NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_8_8, + NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_8_8_8, + NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_8_8_8_8, + NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_16, + NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_16_16, + NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_16_16_16, + NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_16_16_16_16, + 0, 0, 0, 0, + NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_32, + NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_32_32, + NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_32_32_32, + NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_32_32_32_32 }; + + /* we'd also have R11G11B10 and R10G10B10A2 */ + + assert(nr_c > 0 && nr_c <= 4); + + if (size > 32) + return 0; + size >>= (3 - 2); + + return hw_values[size + (nr_c - 1)]; +} + +static INLINE uint32_t +nv50_vbo_vtxelt_to_hw(struct pipe_vertex_element *ve) +{ + uint32_t hw_type, hw_size; + enum pipe_format pf = ve->src_format; + unsigned size = pf_size_x(pf) << pf_exp2(pf); + + hw_type = nv50_vbo_type_to_hw(pf_type(pf)); + hw_size = nv50_vbo_size_to_hw(size, ve->nr_components); + + if (!hw_type || !hw_size) { + NOUVEAU_ERR("unsupported vbo format: %s\n", pf_name(pf)); + abort(); + return 0x24e80000; + } + + if (pf_swizzle_x(pf) == 2) /* BGRA */ + hw_size |= (1 << 31); /* no real swizzle bits :-( */ + + return (hw_type | hw_size); +} + boolean nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { struct nv50_context *nv50 = nv50_context(pipe); - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->tesla->channel; struct nouveau_grobj *tesla = nv50->screen->tesla; nv50_state_validate(nv50); @@ -83,7 +160,7 @@ static INLINE void nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map, unsigned start, unsigned count) { - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->tesla->channel; struct nouveau_grobj *tesla = nv50->screen->tesla; map += start; @@ -101,7 +178,7 @@ nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map, BEGIN_RING(chan, tesla, 0x400015f0, nr >> 1); for (i = 0; i < nr; i += 2) - OUT_RING (chan, (map[1] << 16) | map[0]); + OUT_RING (chan, (map[i + 1] << 16) | map[i]); count -= nr; map += nr; @@ -112,7 +189,7 @@ static INLINE void nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map, unsigned start, unsigned count) { - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->tesla->channel; struct nouveau_grobj *tesla = nv50->screen->tesla; map += start; @@ -130,7 +207,7 @@ nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map, BEGIN_RING(chan, tesla, 0x400015f0, nr >> 1); for (i = 0; i < nr; i += 2) - OUT_RING (chan, (map[1] << 16) | map[0]); + OUT_RING (chan, (map[i + 1] << 16) | map[i]); count -= nr; map += nr; @@ -138,10 +215,10 @@ nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map, } static INLINE void -nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint8_t *map, +nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint32_t *map, unsigned start, unsigned count) { - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->tesla->channel; struct nouveau_grobj *tesla = nv50->screen->tesla; map += start; @@ -163,10 +240,12 @@ nv50_draw_elements(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { struct nv50_context *nv50 = nv50_context(pipe); - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->tesla->channel; struct nouveau_grobj *tesla = nv50->screen->tesla; - struct pipe_winsys *ws = pipe->winsys; - void *map = ws->buffer_map(ws, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ); + struct pipe_screen *pscreen = pipe->screen; + void *map; + + map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ); nv50_state_validate(nv50); @@ -193,64 +272,129 @@ nv50_draw_elements(struct pipe_context *pipe, BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1); OUT_RING (chan, 0); + pipe_buffer_unmap(pscreen, indexBuffer); pipe->flush(pipe, 0, NULL); return TRUE; } +static INLINE boolean +nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib, + struct nouveau_stateobj **pso, + struct pipe_vertex_element *ve, + struct pipe_vertex_buffer *vb) + +{ + struct nouveau_stateobj *so; + struct nouveau_grobj *tesla = nv50->screen->tesla; + struct nouveau_bo *bo = nouveau_bo(vb->buffer); + float *v; + int ret; + enum pipe_format pf = ve->src_format; + + if ((pf_type(pf) != PIPE_FORMAT_TYPE_FLOAT) || + (pf_size_x(pf) << pf_exp2(pf)) != 32) + return FALSE; + + ret = nouveau_bo_map(bo, NOUVEAU_BO_RD); + if (ret) + return FALSE; + v = (float *)(bo->map + (vb->buffer_offset + ve->src_offset)); + + so = *pso; + if (!so) + *pso = so = so_new(nv50->vtxelt_nr * 5, 0); + + switch (ve->nr_components) { + case 4: + so_method(so, tesla, NV50TCL_VTX_ATTR_4F_X(attrib), 4); + so_data (so, fui(v[0])); + so_data (so, fui(v[1])); + so_data (so, fui(v[2])); + so_data (so, fui(v[3])); + break; + case 3: + so_method(so, tesla, NV50TCL_VTX_ATTR_3F_X(attrib), 3); + so_data (so, fui(v[0])); + so_data (so, fui(v[1])); + so_data (so, fui(v[2])); + break; + case 2: + so_method(so, tesla, NV50TCL_VTX_ATTR_2F_X(attrib), 2); + so_data (so, fui(v[0])); + so_data (so, fui(v[1])); + break; + case 1: + so_method(so, tesla, NV50TCL_VTX_ATTR_1F(attrib), 1); + so_data (so, fui(v[0])); + break; + default: + nouveau_bo_unmap(bo); + return FALSE; + } + + nouveau_bo_unmap(bo); + return TRUE; +} + void nv50_vbo_validate(struct nv50_context *nv50) { struct nouveau_grobj *tesla = nv50->screen->tesla; - struct nouveau_stateobj *vtxbuf, *vtxfmt; - int i; + struct nouveau_stateobj *vtxbuf, *vtxfmt, *vtxattr; + unsigned i; + + /* don't validate if Gallium took away our buffers */ + if (nv50->vtxbuf_nr == 0) + return; - vtxbuf = so_new(nv50->vtxelt_nr * 4, nv50->vtxelt_nr * 2); + vtxattr = NULL; + vtxbuf = so_new(nv50->vtxelt_nr * 7, nv50->vtxelt_nr * 4); vtxfmt = so_new(nv50->vtxelt_nr + 1, 0); - so_method(vtxfmt, tesla, 0x1ac0, nv50->vtxelt_nr); + so_method(vtxfmt, tesla, NV50TCL_VERTEX_ARRAY_ATTRIB(0), + nv50->vtxelt_nr); for (i = 0; i < nv50->vtxelt_nr; i++) { struct pipe_vertex_element *ve = &nv50->vtxelt[i]; struct pipe_vertex_buffer *vb = &nv50->vtxbuf[ve->vertex_buffer_index]; + struct nouveau_bo *bo = nouveau_bo(vb->buffer); + uint32_t hw = nv50_vbo_vtxelt_to_hw(ve); - switch (ve->src_format) { - case PIPE_FORMAT_R32G32B32A32_FLOAT: - so_data(vtxfmt, 0x7e080000 | i); - break; - case PIPE_FORMAT_R32G32B32_FLOAT: - so_data(vtxfmt, 0x7e100000 | i); - break; - case PIPE_FORMAT_R32G32_FLOAT: - so_data(vtxfmt, 0x7e200000 | i); - break; - case PIPE_FORMAT_R32_FLOAT: - so_data(vtxfmt, 0x7e900000 | i); - break; - case PIPE_FORMAT_R8G8B8A8_UNORM: - so_data(vtxfmt, 0x24500000 | i); - break; - default: - { - NOUVEAU_ERR("invalid vbo format %s\n", - pf_name(ve->src_format)); - assert(0); - return; - } + if (!vb->stride && + nv50_vbo_static_attrib(nv50, i, &vtxattr, ve, vb)) { + so_data(vtxfmt, hw | (1 << 4)); + + so_method(vtxbuf, tesla, + NV50TCL_VERTEX_ARRAY_FORMAT(i), 1); + so_data (vtxbuf, 0); + continue; } + so_data(vtxfmt, hw | i); - so_method(vtxbuf, tesla, 0x900 + (i * 16), 3); + so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_FORMAT(i), 3); so_data (vtxbuf, 0x20000000 | vb->stride); - so_reloc (vtxbuf, vb->buffer, vb->buffer_offset + + so_reloc (vtxbuf, bo, vb->buffer_offset + ve->src_offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (vtxbuf, vb->buffer, vb->buffer_offset + + so_reloc (vtxbuf, bo, vb->buffer_offset + ve->src_offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); + + /* vertex array limits */ + so_method(vtxbuf, tesla, 0x1080 + (i * 8), 2); + so_reloc (vtxbuf, bo, vb->buffer->size - 1, + NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD | + NOUVEAU_BO_HIGH, 0, 0); + so_reloc (vtxbuf, bo, vb->buffer->size - 1, + NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD | + NOUVEAU_BO_LOW, 0, 0); } so_ref (vtxfmt, &nv50->state.vtxfmt); so_ref (vtxbuf, &nv50->state.vtxbuf); + so_ref (vtxattr, &nv50->state.vtxattr); so_ref (NULL, &vtxbuf); so_ref (NULL, &vtxfmt); + so_ref (NULL, &vtxattr); } diff --git a/src/gallium/drivers/r300/Makefile b/src/gallium/drivers/r300/Makefile index e44f9b9dfc5..d7a2c8c462c 100644 --- a/src/gallium/drivers/r300/Makefile +++ b/src/gallium/drivers/r300/Makefile @@ -4,21 +4,38 @@ include $(TOP)/configs/current LIBNAME = r300 C_SOURCES = \ + r3xx_fs.c \ + r5xx_fs.c \ r300_chipset.c \ r300_clear.c \ r300_context.c \ - r300_debug.c \ r300_emit.c \ r300_flush.c \ + r300_fs.c \ r300_query.c \ r300_render.c \ r300_screen.c \ r300_state.c \ r300_state_derived.c \ r300_state_invariant.c \ - r300_state_shader.c \ - r300_state_tcl.c \ + r300_vs.c \ r300_surface.c \ - r300_texture.c + r300_texture.c \ + r300_tgsi_to_rc.c + +LIBRARY_INCLUDES = \ + -I$(TOP)/src/mesa/drivers/dri/r300/compiler \ + -I$(TOP)/src/mesa \ + -I$(TOP)/include + +COMPILER_ARCHIVE = $(TOP)/src/mesa/drivers/dri/r300/compiler/libr300compiler.a + +EXTRA_OBJECTS = \ + $(COMPILER_ARCHIVE) include ../../Makefile.template + +.PHONY : $(COMPILER_ARCHIVE) + +$(COMPILER_ARCHIVE): + cd $(TOP)/src/mesa/drivers/dri/r300/compiler; make diff --git a/src/gallium/drivers/r300/SConscript b/src/gallium/drivers/r300/SConscript index 182ed2d459a..493d7b28bc3 100644 --- a/src/gallium/drivers/r300/SConscript +++ b/src/gallium/drivers/r300/SConscript @@ -5,20 +5,22 @@ env = env.Clone() r300 = env.ConvenienceLibrary( target = 'r300', source = [ + 'r3xx_fs.c', + 'r5xx_fs.c', 'r300_chipset.c', 'r300_clear.c', 'r300_context.c', 'r300_debug.c', 'r300_emit.c', 'r300_flush.c', + 'r300_fs.c', 'r300_query.c', 'r300_render.c', 'r300_screen.c', 'r300_state.c', 'r300_state_derived.c', 'r300_state_invariant.c', - 'r300_state_shader.c', - 'r300_state_tcl.c', + 'r300_vs.c', 'r300_surface.c', 'r300_texture.c', ]) diff --git a/src/gallium/drivers/r300/r300_chipset.c b/src/gallium/drivers/r300/r300_chipset.c index 9d95ad918c2..d138866d33c 100644 --- a/src/gallium/drivers/r300/r300_chipset.c +++ b/src/gallium/drivers/r300/r300_chipset.c @@ -30,9 +30,10 @@ void r300_parse_chipset(struct r300_capabilities* caps) { /* Reasonable defaults */ + caps->num_vert_fpus = 4; caps->has_tcl = getenv("RADEON_NO_TCL") ? FALSE : TRUE; caps->is_r500 = FALSE; - caps->num_vert_fpus = 4; + caps->high_second_pipe = FALSE; /* Note: These are not ordered by PCI ID. I leave that task to GCC, @@ -41,6 +42,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) switch (caps->pci_id) { case 0x4144: caps->family = CHIP_FAMILY_R300; + caps->high_second_pipe = TRUE; break; case 0x4145: @@ -51,6 +53,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x4E46: case 0x4E47: caps->family = CHIP_FAMILY_R300; + caps->high_second_pipe = TRUE; break; case 0x4150: @@ -67,6 +70,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x4E54: case 0x4E56: caps->family = CHIP_FAMILY_RV350; + caps->high_second_pipe = TRUE; break; case 0x4148: @@ -77,10 +81,12 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x4E49: case 0x4E4B: caps->family = CHIP_FAMILY_R350; + caps->high_second_pipe = TRUE; break; case 0x4E4A: caps->family = CHIP_FAMILY_R360; + caps->high_second_pipe = TRUE; break; case 0x5460: @@ -92,6 +98,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x5B64: case 0x5B65: caps->family = CHIP_FAMILY_RV370; + caps->high_second_pipe = TRUE; break; case 0x3150: @@ -100,6 +107,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x3E50: case 0x3E54: caps->family = CHIP_FAMILY_RV380; + caps->high_second_pipe = TRUE; break; case 0x4A48: @@ -150,6 +158,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->num_vert_fpus = 6; break; + case 0x4B48: case 0x4B49: case 0x4B4A: case 0x4B4B: @@ -349,7 +358,4 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->pci_id); break; } - - /* XXX SW TCL is broken so no forcing it off right now - caps->has_tcl = FALSE; */ } diff --git a/src/gallium/drivers/r300/r300_chipset.h b/src/gallium/drivers/r300/r300_chipset.h index 21eebeae600..322d4a57e41 100644 --- a/src/gallium/drivers/r300/r300_chipset.h +++ b/src/gallium/drivers/r300/r300_chipset.h @@ -44,6 +44,8 @@ struct r300_capabilities { * - Blend color is split across two registers * - Universal Shader (US) block used for fragment shaders */ boolean is_r500; + /* Whether or not the second pixel pipe is accessed with the high bit */ + boolean high_second_pipe; }; /* Enumerations for legibility and telling which card we're running on. */ diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c index 6bdf544a05c..da67bc29b89 100644 --- a/src/gallium/drivers/r300/r300_context.c +++ b/src/gallium/drivers/r300/r300_context.c @@ -34,10 +34,6 @@ static boolean r300_draw_range_elements(struct pipe_context* pipe, struct r300_context* r300 = r300_context(pipe); int i; - if (r300->dirty_state) { - r300_emit_dirty_state(r300); - } - for (i = 0; i < r300->vertex_buffer_count; i++) { void* buf = pipe_buffer_map(pipe->screen, r300->vertex_buffers[i].buffer, @@ -56,7 +52,7 @@ static boolean r300_draw_range_elements(struct pipe_context* pipe, draw_set_mapped_constant_buffer(r300->draw, r300->shader_constants[PIPE_SHADER_VERTEX].constants, - r300->shader_constants[PIPE_SHADER_VERTEX].user_count * + r300->shader_constants[PIPE_SHADER_VERTEX].count * (sizeof(float) * 4)); draw_arrays(r300->draw, mode, start, count); @@ -92,9 +88,21 @@ static boolean r300_draw_arrays(struct pipe_context* pipe, unsigned mode, static void r300_destroy_context(struct pipe_context* context) { struct r300_context* r300 = r300_context(context); + struct r300_query* query, * temp; draw_destroy(r300->draw); + /* Free the OQ BO. */ + context->screen->buffer_destroy(r300->oqbo); + + /* If there are any queries pending or not destroyed, remove them now. */ + if (r300->query_list) { + foreach_s(query, temp, r300->query_list) { + remove_from_list(query); + FREE(query); + } + } + FREE(r300->blend_color_state); FREE(r300->rs_block); FREE(r300->scissor_state); @@ -133,7 +141,6 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, if (!r300) return NULL; - /* XXX this could be refactored now? */ r300->winsys = r300_winsys; r300->context.winsys = (struct pipe_winsys*)r300_winsys; @@ -150,14 +157,25 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300->context.is_texture_referenced = r300_is_texture_referenced; r300->context.is_buffer_referenced = r300_is_buffer_referenced; - r300->draw = draw_create(); - draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300)); - r300->blend_color_state = CALLOC_STRUCT(r300_blend_color_state); r300->rs_block = CALLOC_STRUCT(r300_rs_block); r300->scissor_state = CALLOC_STRUCT(r300_scissor_state); r300->viewport_state = CALLOC_STRUCT(r300_viewport_state); + /* Create a Draw. This is used for vert collation and SW TCL. */ + r300->draw = draw_create(); + /* Enable our renderer. */ + draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300)); + /* Disable Draw's clipping if TCL is present. */ + draw_set_driver_clipping(r300->draw, r300_screen(screen)->caps->has_tcl); + /* Force Draw to never do viewport transform, since (again) we can do + * transform in hardware, always. */ + draw_set_viewport_state(r300->draw, &r300_viewport_identity); + + /* Open up the OQ BO. */ + r300->oqbo = screen->buffer_create(screen, 4096, + PIPE_BUFFER_USAGE_VERTEX, 4096); + r300_init_flush_functions(r300); r300_init_query_functions(r300); diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h index 6f62998b35d..f78492d4aa9 100644 --- a/src/gallium/drivers/r300/r300_context.h +++ b/src/gallium/drivers/r300/r300_context.h @@ -25,15 +25,22 @@ #include "draw/draw_context.h" #include "draw/draw_vertex.h" + #include "pipe/p_context.h" + #include "tgsi/tgsi_scan.h" + #include "util/u_memory.h" +#include "util/u_simple_list.h" #include "r300_clear.h" #include "r300_query.h" #include "r300_screen.h" #include "r300_winsys.h" +struct r300_fragment_shader; +struct r300_vertex_shader; + struct r300_blend_state { uint32_t blend_control; /* R300_RB3D_CBLEND: 0x4e04 */ uint32_t alpha_blend_control; /* R300_RB3D_ABLEND: 0x4e08 */ @@ -63,6 +70,11 @@ struct r300_rs_state { /* Draw-specific rasterizer state */ struct pipe_rasterizer_state rs; + /* Whether or not to enable the VTE. This is referenced at the very + * last moment during emission of VTE state, to decide whether or not + * the VTE should be used for transformation. */ + boolean enable_vte; + uint32_t vap_control_status; /* R300_VAP_CNTL_STATUS: 0x2140 */ uint32_t point_size; /* R300_GA_POINT_SIZE: 0x421c */ uint32_t point_minmax; /* R300_GA_POINT_MINMAX: 0x4230 */ @@ -112,23 +124,24 @@ struct r300_viewport_state { uint32_t vte_control; /* R300_VAP_VTE_CNTL: 0x20b0 */ }; -#define R300_NEW_BLEND 0x0000001 -#define R300_NEW_BLEND_COLOR 0x0000002 -#define R300_NEW_CONSTANTS 0x0000004 -#define R300_NEW_DSA 0x0000008 -#define R300_NEW_FRAMEBUFFERS 0x0000010 -#define R300_NEW_FRAGMENT_SHADER 0x0000020 -#define R300_NEW_RASTERIZER 0x0000040 -#define R300_NEW_RS_BLOCK 0x0000080 -#define R300_NEW_SAMPLER 0x0000100 -#define R300_ANY_NEW_SAMPLERS 0x000ff00 -#define R300_NEW_SCISSOR 0x0010000 -#define R300_NEW_TEXTURE 0x0020000 -#define R300_ANY_NEW_TEXTURES 0x1fe0000 -#define R300_NEW_VERTEX_FORMAT 0x2000000 -#define R300_NEW_VERTEX_SHADER 0x4000000 -#define R300_NEW_VIEWPORT 0x8000000 -#define R300_NEW_KITCHEN_SINK 0xfffffff +#define R300_NEW_BLEND 0x00000001 +#define R300_NEW_BLEND_COLOR 0x00000002 +#define R300_NEW_CLIP 0x00000004 +#define R300_NEW_CONSTANTS 0x00000008 +#define R300_NEW_DSA 0x00000010 +#define R300_NEW_FRAMEBUFFERS 0x00000020 +#define R300_NEW_FRAGMENT_SHADER 0x00000040 +#define R300_NEW_RASTERIZER 0x00000080 +#define R300_NEW_RS_BLOCK 0x00000100 +#define R300_NEW_SAMPLER 0x00000200 +#define R300_ANY_NEW_SAMPLERS 0x0001fe00 +#define R300_NEW_SCISSOR 0x00020000 +#define R300_NEW_TEXTURE 0x00040000 +#define R300_ANY_NEW_TEXTURES 0x03fc0000 +#define R300_NEW_VERTEX_FORMAT 0x04000000 +#define R300_NEW_VERTEX_SHADER 0x08000000 +#define R300_NEW_VIEWPORT 0x10000000 +#define R300_NEW_KITCHEN_SINK 0x1fffffff /* The next several objects are not pure Radeon state; they inherit from * various Gallium classes. */ @@ -136,66 +149,32 @@ struct r300_viewport_state { struct r300_constant_buffer { /* Buffer of constants */ /* XXX first number should be raised */ - float constants[8][4]; - /* Number of user-defined constants */ - int user_count; + float constants[32][4]; /* Total number of constants */ - int count; -}; - -struct r3xx_fragment_shader { - /* Parent class */ - struct pipe_shader_state state; - struct tgsi_shader_info info; - - /* Has this shader been translated yet? */ - boolean translated; - - /* Pixel stack size */ - int stack_size; -}; - -struct r300_fragment_shader { - /* Parent class */ - struct r3xx_fragment_shader shader; - - /* Number of ALU instructions */ - int alu_instruction_count; - - /* Number of texture instructions */ - int tex_instruction_count; - - /* Number of texture indirections */ - int indirections; - - /* Indirection node offsets */ - int alu_offset[4]; - - /* Machine instructions */ - struct { - uint32_t alu_rgb_inst; - uint32_t alu_rgb_addr; - uint32_t alu_alpha_inst; - uint32_t alu_alpha_addr; - } instructions[64]; /* XXX magic num */ + unsigned count; }; -struct r500_fragment_shader { - /* Parent class */ - struct r3xx_fragment_shader shader; - - /* Number of used instructions */ - int instruction_count; - - /* Machine instructions */ - struct { - uint32_t inst0; - uint32_t inst1; - uint32_t inst2; - uint32_t inst3; - uint32_t inst4; - uint32_t inst5; - } instructions[256]; /*< XXX magic number */ +/* Query object. + * + * This is not a subclass of pipe_query because pipe_query is never + * actually fully defined. So, rather than have it as a member, and do + * subclass-style casting, we treat pipe_query as an opaque, and just + * trust that our state tracker does not ever mess up query objects. + */ +struct r300_query { + /* The kind of query. Currently only OQ is supported. */ + unsigned type; + /* Whether this query is currently active. Only active queries will + * get emitted into the command stream, and only active queries get + * tallied. */ + boolean active; + /* The current count of this query. Required to be at least 32 bits. */ + unsigned int count; + /* The offset of this query into the query buffer, in bytes. */ + unsigned offset; + /* Linked list members. */ + struct r300_query* prev; + struct r300_query* next; }; struct r300_texture { @@ -207,7 +186,7 @@ struct r300_texture { /* Stride (pitch?) of this texture in bytes */ unsigned stride; - + /* Total size of this texture, in bytes. */ unsigned size; @@ -232,27 +211,9 @@ struct r300_vertex_format { int fs_tab[16]; }; -struct r300_vertex_shader { - /* Parent class */ - struct pipe_shader_state state; - struct tgsi_shader_info info; - - /* Fallback shader, because Draw has issues */ - struct draw_vertex_shader* draw; - - /* Has this shader been translated yet? */ - boolean translated; - - /* Number of used instructions */ - int instruction_count; - - /* Machine instructions */ - struct { - uint32_t inst0; - uint32_t inst1; - uint32_t inst2; - uint32_t inst3; - } instructions[128]; /*< XXX magic number */ +static struct pipe_viewport_state r300_viewport_identity = { + .scale = {1.0, 1.0, 1.0, 1.0}, + .translate = {0.0, 0.0, 0.0, 0.0}, }; struct r300_context { @@ -264,17 +225,29 @@ struct r300_context { /* Draw module. Used mostly for SW TCL. */ struct draw_context* draw; + /* Vertex buffer for rendering. */ + struct pipe_buffer* vbo; + /* Offset into the VBO. */ + size_t vbo_offset; + + /* Occlusion query buffer. */ + struct pipe_buffer* oqbo; + /* Query list. */ + struct r300_query* query_list; + /* Various CSO state objects. */ /* Blend state. */ struct r300_blend_state* blend_state; /* Blend color state. */ struct r300_blend_color_state* blend_color_state; + /* User clip planes. */ + struct pipe_clip_state clip_state; /* Shader constants. */ struct r300_constant_buffer shader_constants[PIPE_SHADER_TYPES]; /* Depth, stencil, and alpha state. */ struct r300_dsa_state* dsa_state; /* Fragment shader. */ - struct r3xx_fragment_shader* fs; + struct r300_fragment_shader* fs; /* Framebuffer state. We currently don't need our own version of this. */ struct pipe_framebuffer_state framebuffer_state; /* Rasterizer state. */ @@ -289,7 +262,7 @@ struct r300_context { /* Texture states. */ struct r300_texture* textures[8]; int texture_count; - /* Vertex buffers. */ + /* Vertex buffers for Gallium. */ struct pipe_vertex_buffer vertex_buffers[PIPE_MAX_ATTRIBS]; int vertex_buffer_count; /* Vertex information. */ @@ -305,7 +278,8 @@ struct r300_context { }; /* Convenience cast wrapper. */ -static struct r300_context* r300_context(struct pipe_context* context) { +static INLINE struct r300_context* r300_context(struct pipe_context* context) +{ return (struct r300_context*)context; } diff --git a/src/gallium/drivers/r300/r300_cs.h b/src/gallium/drivers/r300/r300_cs.h index 82a3942248e..71b142c0dbf 100644 --- a/src/gallium/drivers/r300/r300_cs.h +++ b/src/gallium/drivers/r300/r300_cs.h @@ -34,6 +34,7 @@ #define MAX_CS_SIZE 64 * 1024 / 4 +#define VERY_VERBOSE_CS 0 #define VERY_VERBOSE_REGISTERS 0 /* XXX stolen from radeon_drm.h */ @@ -56,8 +57,10 @@ #define BEGIN_CS(size) do { \ CHECK_CS(size); \ - debug_printf("r300: BEGIN_CS, count %d, in %s (%s:%d)\n", \ - size, __FUNCTION__, __FILE__, __LINE__); \ + if (VERY_VERBOSE_CS) { \ + debug_printf("r300: BEGIN_CS, count %d, in %s (%s:%d)\n", \ + size, __FUNCTION__, __FILE__, __LINE__); \ + } \ cs_winsys->begin_cs(cs_winsys, (size), \ __FILE__, __FUNCTION__, __LINE__); \ cs_count = size; \ @@ -93,8 +96,9 @@ } while (0) #define OUT_CS_RELOC(bo, offset, rd, wd, flags) do { \ - debug_printf("r300: writing relocation for buffer %p, offset %d\n", \ - bo, offset); \ + debug_printf("r300: writing relocation for buffer %p, offset %d, " \ + "domains (%d, %d, %d)\n", \ + bo, offset, rd, wd, flags); \ assert(bo); \ OUT_CS(offset); \ cs_winsys->write_cs_reloc(cs_winsys, bo, rd, wd, flags); \ @@ -102,16 +106,20 @@ } while (0) #define END_CS do { \ - debug_printf("r300: END_CS in %s (%s:%d)\n", __FUNCTION__, __FILE__, \ - __LINE__); \ + if (VERY_VERBOSE_CS) { \ + debug_printf("r300: END_CS in %s (%s:%d)\n", __FUNCTION__, \ + __FILE__, __LINE__); \ + } \ if (cs_count != 0) \ debug_printf("r300: Warning: cs_count off by %d\n", cs_count); \ cs_winsys->end_cs(cs_winsys, __FILE__, __FUNCTION__, __LINE__); \ } while (0) #define FLUSH_CS do { \ - debug_printf("r300: FLUSH_CS in %s (%s:%d)\n\n", __FUNCTION__, __FILE__, \ - __LINE__); \ + if (VERY_VERBOSE_CS) { \ + debug_printf("r300: FLUSH_CS in %s (%s:%d)\n\n", __FUNCTION__, \ + __FILE__, __LINE__); \ + } \ cs_winsys->flush_cs(cs_winsys); \ } while (0) diff --git a/src/gallium/drivers/r300/r300_debug.c b/src/gallium/drivers/r300/r300_debug.c deleted file mode 100644 index dd63136c9d6..00000000000 --- a/src/gallium/drivers/r300/r300_debug.c +++ /dev/null @@ -1,238 +0,0 @@ -/* - * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#include "r300_debug.h" - -static void r300_dump_fs(struct r300_fragment_shader* fs) -{ - int i; - - for (i = 0; i < fs->alu_instruction_count; i++) { - } -} - -static char* r500_fs_swiz[] = { - " R", - " G", - " B", - " A", - " 0", - ".5", - " 1", - " U", -}; - -static char* r500_fs_op_rgb[] = { - "MAD", - "DP3", - "DP4", - "D2A", - "MIN", - "MAX", - "---", - "CND", - "CMP", - "FRC", - "SOP", - "MDH", - "MDV", -}; - -static char* r500_fs_op_alpha[] = { - "MAD", - " DP", - "MIN", - "MAX", - "---", - "CND", - "CMP", - "FRC", - "EX2", - "LN2", - "RCP", - "RSQ", - "SIN", - "COS", - "MDH", - "MDV", -}; - -static char* r500_fs_mask[] = { - "NONE", - "R ", - " G ", - "RG ", - " B ", - "R B ", - " GB ", - "RGB ", - " A", - "R A", - " G A", - "RG A", - " BA", - "R BA", - " GBA", - "RGBA", -}; - -static char* r500_fs_tex[] = { - " NOP", - " LD", - "TEXKILL", - " PROJ", - "LODBIAS", - " LOD", - " DXDY", -}; - -void r500_fs_dump(struct r500_fragment_shader* fs) -{ - int i; - uint32_t inst; - - for (i = 0; i < fs->instruction_count; i++) { - inst = fs->instructions[i].inst0; - debug_printf("%d: 0: CMN_INST 0x%08x:", i, inst); - switch (inst & 0x3) { - case R500_INST_TYPE_ALU: - debug_printf("ALU "); - break; - case R500_INST_TYPE_OUT: - debug_printf("OUT "); - break; - case R500_INST_TYPE_FC: - debug_printf("FC "); - break; - case R500_INST_TYPE_TEX: - debug_printf("TEX "); - break; - } - debug_printf("%s %s %s %s ", - inst & R500_INST_TEX_SEM_WAIT ? "TEX_WAIT" : "", - inst & R500_INST_LAST ? "LAST" : "", - inst & R500_INST_NOP ? "NOP" : "", - inst & R500_INST_ALU_WAIT ? "ALU_WAIT" : ""); - debug_printf("wmask: %s omask: %s\n", - r500_fs_mask[(inst >> 11) & 0xf], - r500_fs_mask[(inst >> 15) & 0xf]); - switch (inst & 0x3) { - case R500_INST_TYPE_ALU: - case R500_INST_TYPE_OUT: - inst = fs->instructions[i].inst1; - debug_printf(" 1: RGB_ADDR 0x%08x:", inst); - debug_printf("Addr0: %d%c, Addr1: %d%c, " - "Addr2: %d%c, srcp:%d\n", - inst & 0xff, (inst & (1 << 8)) ? 'c' : 't', - (inst >> 10) & 0xff, (inst & (1 << 18)) ? 'c' : 't', - (inst >> 20) & 0xff, (inst & (1 << 28)) ? 'c' : 't', - (inst >> 30)); - - inst = fs->instructions[i].inst2; - debug_printf(" 2: ALPHA_ADDR 0x%08x:", inst); - debug_printf("Addr0: %d%c, Addr1: %d%c, " - "Addr2: %d%c, srcp:%d\n", - inst & 0xff, (inst & (1 << 8)) ? 'c' : 't', - (inst >> 10) & 0xff, (inst & (1 << 18)) ? 'c' : 't', - (inst >> 20) & 0xff, (inst & (1 << 28)) ? 'c' : 't', - (inst >> 30)); - - inst = fs->instructions[i].inst3; - debug_printf(" 3: RGB_INST 0x%08x:", inst); - debug_printf("rgb_A_src:%d %s/%s/%s %d " - "rgb_B_src:%d %s/%s/%s %d\n", - inst & 0x3, r500_fs_swiz[(inst >> 2) & 0x7], - r500_fs_swiz[(inst >> 5) & 0x7], - r500_fs_swiz[(inst >> 8) & 0x7], - (inst >> 11) & 0x3, (inst >> 13) & 0x3, - r500_fs_swiz[(inst >> 15) & 0x7], - r500_fs_swiz[(inst >> 18) & 0x7], - r500_fs_swiz[(inst >> 21) & 0x7], - (inst >> 24) & 0x3); - - inst = fs->instructions[i].inst4; - debug_printf(" 4: ALPHA_INST 0x%08x:", inst); - debug_printf("%s dest:%d%s alp_A_src:%d %s %d " - "alp_B_src:%d %s %d w:%d\n", - r500_fs_op_alpha[inst & 0xf], (inst >> 4) & 0x7f, - inst & (1<<11) ? "(rel)":"", (inst >> 12) & 0x3, - r500_fs_swiz[(inst >> 14) & 0x7], (inst >> 17) & 0x3, - (inst >> 19) & 0x3, r500_fs_swiz[(inst >> 21) & 0x7], - (inst >> 24) & 0x3, (inst >> 31) & 0x1); - - inst = fs->instructions[i].inst5; - debug_printf(" 5: RGBA_INST 0x%08x:", inst); - debug_printf("%s dest:%d%s rgb_C_src:%d %s/%s/%s %d " - "alp_C_src:%d %s %d\n", - r500_fs_op_rgb[inst & 0xf], (inst >> 4) & 0x7f, - inst & (1 << 11) ? "(rel)":"", (inst >> 12) & 0x3, - r500_fs_swiz[(inst >> 14) & 0x7], - r500_fs_swiz[(inst >> 17) & 0x7], - r500_fs_swiz[(inst >> 20) & 0x7], - (inst >> 23) & 0x3, (inst >> 25) & 0x3, - r500_fs_swiz[(inst >> 27) & 0x7], (inst >> 30) & 0x3); - break; - case R500_INST_TYPE_FC: - /* XXX don't even bother yet */ - break; - case R500_INST_TYPE_TEX: - inst = fs->instructions[i].inst1; - debug_printf(" 1: TEX_INST 0x%08x: id: %d " - "op:%s, %s, %s %s\n", - inst, (inst >> 16) & 0xf, - r500_fs_tex[(inst >> 22) & 0x7], - (inst & (1 << 25)) ? "ACQ" : "", - (inst & (1 << 26)) ? "IGNUNC" : "", - (inst & (1 << 27)) ? "UNSCALED" : "SCALED"); - - inst = fs->instructions[i].inst2; - debug_printf(" 2: TEX_ADDR 0x%08x: " - "src: %d%s %s/%s/%s/%s dst: %d%s %s/%s/%s/%s\n", - inst, inst & 0x7f, inst & (1 << 7) ? "(rel)" : "", - r500_fs_swiz[(inst >> 8) & 0x3], - r500_fs_swiz[(inst >> 10) & 0x3], - r500_fs_swiz[(inst >> 12) & 0x3], - r500_fs_swiz[(inst >> 14) & 0x3], - (inst >> 16) & 0x7f, inst & (1 << 23) ? "(rel)" : "", - r500_fs_swiz[(inst >> 24) & 0x3], - r500_fs_swiz[(inst >> 26) & 0x3], - r500_fs_swiz[(inst >> 28) & 0x3], - r500_fs_swiz[(inst >> 30) & 0x3]); - - inst = fs->instructions[i].inst3; - debug_printf(" 3: TEX_DXDY 0x%08x\n", inst); - break; - } - } -} - -void r300_vs_dump(struct r300_vertex_shader* vs) -{ - int i; - - for (i = 0; i < vs->instruction_count; i++) { - debug_printf("inst0: 0x%x\n", vs->instructions[i].inst0); - debug_printf("inst1: 0x%x\n", vs->instructions[i].inst1); - debug_printf("inst2: 0x%x\n", vs->instructions[i].inst2); - debug_printf("inst3: 0x%x\n", vs->instructions[i].inst3); - } -} diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c index 01bac5f759c..bd4d59e6f1a 100644 --- a/src/gallium/drivers/r300/r300_emit.c +++ b/src/gallium/drivers/r300/r300_emit.c @@ -24,6 +24,9 @@ #include "r300_emit.h" +#include "r300_fs.h" +#include "r300_vs.h" + void r300_emit_blend_state(struct r300_context* r300, struct r300_blend_state* blend) { @@ -56,6 +59,36 @@ void r300_emit_blend_color_state(struct r300_context* r300, } } +void r300_emit_clip_state(struct r300_context* r300, + struct pipe_clip_state* clip) +{ + int i; + struct r300_screen* r300screen = r300_screen(r300->context.screen); + CS_LOCALS(r300); + + if (r300screen->caps->has_tcl) { + BEGIN_CS(5 + (6 * 4)); + OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, + (r300screen->caps->is_r500 ? + R500_PVS_UCP_START : R300_PVS_UCP_START)); + OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, 6 * 4); + for (i = 0; i < 6; i++) { + OUT_CS_32F(clip->ucp[i][0]); + OUT_CS_32F(clip->ucp[i][1]); + OUT_CS_32F(clip->ucp[i][2]); + OUT_CS_32F(clip->ucp[i][3]); + } + OUT_CS_REG(R300_VAP_CLIP_CNTL, ((1 << clip->nr) - 1) | + R300_PS_UCP_MODE_CLIP_AS_TRIFAN); + END_CS; + } else { + BEGIN_CS(2); + OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE); + END_CS; + } + +} + void r300_emit_dsa_state(struct r300_context* r300, struct r300_dsa_state* dsa) { @@ -79,73 +112,158 @@ void r300_emit_dsa_state(struct r300_context* r300, END_CS; } -void r300_emit_fragment_shader(struct r300_context* r300, - struct r300_fragment_shader* fs) +static const float * get_shader_constant( + struct r300_context * r300, + struct rc_constant * constant, + struct r300_constant_buffer * externals) +{ + static const float zero[4] = { 0.0, 0.0, 0.0, 0.0 }; + switch(constant->Type) { + case RC_CONSTANT_EXTERNAL: + return externals->constants[constant->u.External]; + + case RC_CONSTANT_IMMEDIATE: + return constant->u.Immediate; + + default: + debug_printf("r300: Implementation error: Unhandled constant type %i\n", + constant->Type); + return zero; + } +} + +/* Convert a normal single-precision float into the 7.16 format + * used by the R300 fragment shader. + */ +static uint32_t pack_float24(float f) +{ + union { + float fl; + uint32_t u; + } u; + float mantissa; + int exponent; + uint32_t float24 = 0; + + if (f == 0.0) + return 0; + + u.fl = f; + + mantissa = frexpf(f, &exponent); + + /* Handle -ve */ + if (mantissa < 0) { + float24 |= (1 << 23); + mantissa = mantissa * -1.0; + } + /* Handle exponent, bias of 63 */ + exponent += 62; + float24 |= (exponent << 16); + /* Kill 7 LSB of mantissa */ + float24 |= (u.u & 0x7FFFFF) >> 7; + + return float24; +} + +void r300_emit_fragment_program_code(struct r300_context* r300, + struct rX00_fragment_program_code* generic_code, + struct r300_constant_buffer* externals) { + struct r300_fragment_program_code * code = &generic_code->code.r300; + struct rc_constant_list * constants = &generic_code->constants; int i; CS_LOCALS(r300); - BEGIN_CS(22); - - OUT_CS_REG(R300_US_CONFIG, fs->indirections); - OUT_CS_REG(R300_US_PIXSIZE, fs->shader.stack_size); - /* XXX figure out exactly how big the sizes are on this reg */ - OUT_CS_REG(R300_US_CODE_OFFSET, 0x40); - /* XXX figure these ones out a bit better kthnx */ - OUT_CS_REG(R300_US_CODE_ADDR_0, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_1, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_2, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_3, 0x40 | R300_RGBA_OUT); - - for (i = 0; i < fs->alu_instruction_count; i++) { - OUT_CS_REG(R300_US_ALU_RGB_INST_0 + (4 * i), - fs->instructions[i].alu_rgb_inst); - OUT_CS_REG(R300_US_ALU_RGB_ADDR_0 + (4 * i), - fs->instructions[i].alu_rgb_addr); - OUT_CS_REG(R300_US_ALU_ALPHA_INST_0 + (4 * i), - fs->instructions[i].alu_alpha_inst); - OUT_CS_REG(R300_US_ALU_ALPHA_ADDR_0 + (4 * i), - fs->instructions[i].alu_alpha_addr); + BEGIN_CS(15 + + code->alu.length * 4 + + (code->tex.length ? (1 + code->tex.length) : 0) + + (constants->Count ? (1 + constants->Count * 4) : 0)); + + OUT_CS_REG(R300_US_CONFIG, code->config); + OUT_CS_REG(R300_US_PIXSIZE, code->pixsize); + OUT_CS_REG(R300_US_CODE_OFFSET, code->code_offset); + + OUT_CS_REG_SEQ(R300_US_CODE_ADDR_0, 4); + for(i = 0; i < 4; ++i) + OUT_CS(code->code_addr[i]); + + OUT_CS_REG_SEQ(R300_US_ALU_RGB_INST_0, code->alu.length); + for (i = 0; i < code->alu.length; i++) + OUT_CS(code->alu.inst[i].rgb_inst); + + OUT_CS_REG_SEQ(R300_US_ALU_RGB_ADDR_0, code->alu.length); + for (i = 0; i < code->alu.length; i++) + OUT_CS(code->alu.inst[i].rgb_addr); + + OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_INST_0, code->alu.length); + for (i = 0; i < code->alu.length; i++) + OUT_CS(code->alu.inst[i].alpha_inst); + + OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0, code->alu.length); + for (i = 0; i < code->alu.length; i++) + OUT_CS(code->alu.inst[i].alpha_addr); + + if (code->tex.length) { + OUT_CS_REG_SEQ(R300_US_TEX_INST_0, code->tex.length); + for(i = 0; i < code->tex.length; ++i) + OUT_CS(code->tex.inst[i]); + } + + if (constants->Count) { + OUT_CS_ONE_REG(R300_PFS_PARAM_0_X, constants->Count * 4); + for(i = 0; i < constants->Count; ++i) { + const float * data = get_shader_constant(r300, &constants->Constants[i], externals); + OUT_CS(pack_float24(data[0])); + OUT_CS(pack_float24(data[1])); + OUT_CS(pack_float24(data[2])); + OUT_CS(pack_float24(data[3])); + } } END_CS; } -void r500_emit_fragment_shader(struct r300_context* r300, - struct r500_fragment_shader* fs) +void r500_emit_fragment_program_code(struct r300_context* r300, + struct rX00_fragment_program_code* generic_code, + struct r300_constant_buffer* externals) { + struct r500_fragment_program_code * code = &generic_code->code.r500; + struct rc_constant_list * constants = &generic_code->constants; int i; - struct r300_constant_buffer* constants = - &r300->shader_constants[PIPE_SHADER_FRAGMENT]; CS_LOCALS(r300); - BEGIN_CS(9 + (fs->instruction_count * 6) + (constants->count ? 3 : 0) + - (constants->count * 4)); - OUT_CS_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO); - OUT_CS_REG(R500_US_PIXSIZE, fs->shader.stack_size); - OUT_CS_REG(R500_US_CODE_ADDR, R500_US_CODE_START_ADDR(0) | - R500_US_CODE_END_ADDR(fs->instruction_count)); + BEGIN_CS(13 + + ((code->inst_end + 1) * 6) + + (constants->Count ? (3 + (constants->Count * 4)) : 0)); + OUT_CS_REG(R500_US_CONFIG, 0); + OUT_CS_REG(R500_US_PIXSIZE, code->max_temp_idx); + OUT_CS_REG(R500_US_CODE_RANGE, + R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end)); + OUT_CS_REG(R500_US_CODE_OFFSET, 0); + OUT_CS_REG(R500_US_CODE_ADDR, + R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end)); OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_INSTR); - OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, fs->instruction_count * 6); - for (i = 0; i < fs->instruction_count; i++) { - OUT_CS(fs->instructions[i].inst0); - OUT_CS(fs->instructions[i].inst1); - OUT_CS(fs->instructions[i].inst2); - OUT_CS(fs->instructions[i].inst3); - OUT_CS(fs->instructions[i].inst4); - OUT_CS(fs->instructions[i].inst5); - } - - if (constants->count) { - OUT_CS_REG(R500_GA_US_VECTOR_INDEX, - R500_GA_US_VECTOR_INDEX_TYPE_CONST); - OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, constants->count * 4); - for (i = 0; i < constants->count; i++) { - OUT_CS_32F(constants->constants[i][0]); - OUT_CS_32F(constants->constants[i][1]); - OUT_CS_32F(constants->constants[i][2]); - OUT_CS_32F(constants->constants[i][3]); + OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, (code->inst_end + 1) * 6); + for (i = 0; i <= code->inst_end; i++) { + OUT_CS(code->inst[i].inst0); + OUT_CS(code->inst[i].inst1); + OUT_CS(code->inst[i].inst2); + OUT_CS(code->inst[i].inst3); + OUT_CS(code->inst[i].inst4); + OUT_CS(code->inst[i].inst5); + } + + if (constants->Count) { + OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST); + OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, constants->Count * 4); + for (i = 0; i < constants->Count; i++) { + const float * data = get_shader_constant(r300, &constants->Constants[i], externals); + OUT_CS_32F(data[0]); + OUT_CS_32F(data[1]); + OUT_CS_32F(data[2]); + OUT_CS_32F(data[3]); } } @@ -160,16 +278,19 @@ void r300_emit_fb_state(struct r300_context* r300, int i; CS_LOCALS(r300); - BEGIN_CS((8 * fb->nr_cbufs) + (fb->zsbuf ? 8 : 0) + 4); + BEGIN_CS((10 * fb->nr_cbufs) + (fb->zsbuf ? 10 : 0) + 4); for (i = 0; i < fb->nr_cbufs; i++) { tex = (struct r300_texture*)fb->cbufs[i]->texture; + assert(tex && tex->buffer && "cbuf is marked, but NULL!"); pixpitch = tex->stride / tex->tex.block.size; OUT_CS_REG_SEQ(R300_RB3D_COLOROFFSET0 + (4 * i), 1); OUT_CS_RELOC(tex->buffer, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); - OUT_CS_REG(R300_RB3D_COLORPITCH0 + (4 * i), pixpitch | - r300_translate_colorformat(tex->tex.format)); + OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0 + (4 * i), 1); + OUT_CS_RELOC(tex->buffer, pixpitch | + r300_translate_colorformat(tex->tex.format), 0, + RADEON_GEM_DOMAIN_VRAM, 0); OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), r300_translate_out_fmt(fb->cbufs[i]->format)); @@ -177,14 +298,16 @@ void r300_emit_fb_state(struct r300_context* r300, if (fb->zsbuf) { tex = (struct r300_texture*)fb->zsbuf->texture; - pixpitch = (tex->stride / tex->tex.block.size); + assert(tex && tex->buffer && "zsbuf is marked, but NULL!"); + pixpitch = tex->stride / tex->tex.block.size; OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET, 1); OUT_CS_RELOC(tex->buffer, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); OUT_CS_REG(R300_ZB_FORMAT, r300_translate_zsformat(tex->tex.format)); - OUT_CS_REG(R300_ZB_DEPTHPITCH, pixpitch); + OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1); + OUT_CS_RELOC(tex->buffer, pixpitch, 0, RADEON_GEM_DOMAIN_VRAM, 0); } OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, @@ -196,6 +319,79 @@ void r300_emit_fb_state(struct r300_context* r300, END_CS; } +void r300_emit_query_begin(struct r300_context* r300, + struct r300_query* query) +{ + CS_LOCALS(r300); + + /* XXX This will almost certainly not return good results + * for overlapping queries. */ + BEGIN_CS(2); + OUT_CS_REG(R300_ZB_ZPASS_DATA, 0); + END_CS; +} + +void r300_emit_query_end(struct r300_context* r300, + struct r300_query* query) +{ + struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps; + CS_LOCALS(r300); + + if (!r300->winsys->add_buffer(r300->winsys, r300->oqbo, + 0, RADEON_GEM_DOMAIN_GTT)) { + debug_printf("r300: There wasn't room for the OQ buffer!?" + " Oh noes!\n"); + } + + assert(caps->num_frag_pipes); + BEGIN_CS(6 * caps->num_frag_pipes + 2); + /* I'm not so sure I like this switch, but it's hard to be elegant + * when there's so many special cases... + * + * So here's the basic idea. For each pipe, enable writes to it only, + * then put out the relocation for ZPASS_ADDR, taking into account a + * 4-byte offset for each pipe. RV380 and older are special; they have + * only two pipes, and the second pipe's enable is on bit 3, not bit 1, + * so there's a chipset cap for that. */ + switch (caps->num_frag_pipes) { + case 4: + /* pipe 3 only */ + OUT_CS_REG(R300_SU_REG_DEST, 1 << 3); + OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); + OUT_CS_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 3), + 0, RADEON_GEM_DOMAIN_GTT, 0); + case 3: + /* pipe 2 only */ + OUT_CS_REG(R300_SU_REG_DEST, 1 << 2); + OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); + OUT_CS_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 2), + 0, RADEON_GEM_DOMAIN_GTT, 0); + case 2: + /* pipe 1 only */ + /* As mentioned above, accomodate RV380 and older. */ + OUT_CS_REG(R300_SU_REG_DEST, + 1 << (caps->high_second_pipe ? 3 : 1)); + OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); + OUT_CS_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 1), + 0, RADEON_GEM_DOMAIN_GTT, 0); + case 1: + /* pipe 0 only */ + OUT_CS_REG(R300_SU_REG_DEST, 1 << 0); + OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); + OUT_CS_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 0), + 0, RADEON_GEM_DOMAIN_GTT, 0); + default: + debug_printf("r300: Implementation error: Chipset reports %d" + " pixel pipes!\n", caps->num_frag_pipes); + assert(0); + } + + /* And, finally, reset it to normal... */ + OUT_CS_REG(R300_SU_REG_DEST, 0xF); + END_CS; + +} + void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs) { CS_LOCALS(r300); @@ -234,7 +430,7 @@ void r300_emit_rs_block_state(struct r300_context* r300, } for (i = 0; i < 8; i++) { OUT_CS(rs->ip[i]); - debug_printf("ip %d: 0x%08x\n", i, rs->ip[i]); + /* debug_printf("ip %d: 0x%08x\n", i, rs->ip[i]); */ } OUT_CS_REG_SEQ(R300_RS_COUNT, 2); @@ -248,27 +444,15 @@ void r300_emit_rs_block_state(struct r300_context* r300, } for (i = 0; i < 8; i++) { OUT_CS(rs->inst[i]); - debug_printf("inst %d: 0x%08x\n", i, rs->inst[i]); + /* debug_printf("inst %d: 0x%08x\n", i, rs->inst[i]); */ } - debug_printf("count: 0x%08x inst_count: 0x%08x\n", rs->count, - rs->inst_count); + /* debug_printf("count: 0x%08x inst_count: 0x%08x\n", rs->count, + * rs->inst_count); */ END_CS; } -void r300_emit_sampler(struct r300_context* r300, - struct r300_sampler_state* sampler, unsigned offset) -{ - CS_LOCALS(r300); - - BEGIN_CS(6); - OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), sampler->filter0); - OUT_CS_REG(R300_TX_FILTER1_0 + (offset * 4), sampler->filter1); - OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (offset * 4), sampler->border_color); - END_CS; -} - void r300_emit_scissor_state(struct r300_context* r300, struct r300_scissor_state* scissor) { @@ -282,11 +466,17 @@ void r300_emit_scissor_state(struct r300_context* r300, } void r300_emit_texture(struct r300_context* r300, - struct r300_texture* tex, unsigned offset) + struct r300_sampler_state* sampler, + struct r300_texture* tex, + unsigned offset) { CS_LOCALS(r300); - BEGIN_CS(10); + BEGIN_CS(16); + OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), sampler->filter0); + OUT_CS_REG(R300_TX_FILTER1_0 + (offset * 4), sampler->filter1); + OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (offset * 4), sampler->border_color); + OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), tex->state.format0); OUT_CS_REG(R300_TX_FORMAT1_0 + (offset * 4), tex->state.format1); OUT_CS_REG(R300_TX_FORMAT2_0 + (offset * 4), tex->state.format2); @@ -296,6 +486,30 @@ void r300_emit_texture(struct r300_context* r300, END_CS; } +void r300_emit_vertex_buffer(struct r300_context* r300) +{ + CS_LOCALS(r300); + + debug_printf("r300: Preparing vertex buffer %p for render, " + "vertex size %d\n", r300->vbo, + r300->vertex_info.vinfo.size); + /* Set the pointer to our vertex buffer. The emitted values are this: + * PACKET3 [3D_LOAD_VBPNTR] + * COUNT [1] + * FORMAT [size | stride << 8] + * OFFSET [offset into BO] + * VBPNTR [relocated BO] + */ + BEGIN_CS(7); + OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3); + OUT_CS(1); + OUT_CS(r300->vertex_info.vinfo.size | + (r300->vertex_info.vinfo.size << 8)); + OUT_CS(r300->vbo_offset); + OUT_CS_RELOC(r300->vbo, 0, RADEON_GEM_DOMAIN_GTT, 0, 0); + END_CS; +} + void r300_emit_vertex_format_state(struct r300_context* r300) { int i; @@ -310,33 +524,33 @@ void r300_emit_vertex_format_state(struct r300_context* r300) OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2); OUT_CS(r300->vertex_info.vinfo.hwfmt[2]); OUT_CS(r300->vertex_info.vinfo.hwfmt[3]); - for (i = 0; i < 4; i++) { - debug_printf("hwfmt%d: 0x%08x\n", i, - r300->vertex_info.vinfo.hwfmt[i]); - } + /* for (i = 0; i < 4; i++) { + * debug_printf("hwfmt%d: 0x%08x\n", i, + * r300->vertex_info.vinfo.hwfmt[i]); + * } */ OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, 8); for (i = 0; i < 8; i++) { OUT_CS(r300->vertex_info.vap_prog_stream_cntl[i]); - debug_printf("prog_stream_cntl%d: 0x%08x\n", i, - r300->vertex_info.vap_prog_stream_cntl[i]); + /* debug_printf("prog_stream_cntl%d: 0x%08x\n", i, + * r300->vertex_info.vap_prog_stream_cntl[i]); */ } OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, 8); for (i = 0; i < 8; i++) { OUT_CS(r300->vertex_info.vap_prog_stream_cntl_ext[i]); - debug_printf("prog_stream_cntl_ext%d: 0x%08x\n", i, - r300->vertex_info.vap_prog_stream_cntl_ext[i]); + /* debug_printf("prog_stream_cntl_ext%d: 0x%08x\n", i, + * r300->vertex_info.vap_prog_stream_cntl_ext[i]); */ } END_CS; } -void r300_emit_vertex_shader(struct r300_context* r300, - struct r300_vertex_shader* vs) +void r300_emit_vertex_program_code(struct r300_context* r300, + struct r300_vertex_program_code* code, + struct r300_constant_buffer* constants) { int i; struct r300_screen* r300screen = r300_screen(r300->context.screen); - struct r300_constant_buffer* constants = - &r300->shader_constants[PIPE_SHADER_VERTEX]; + unsigned instruction_count = code->length / 4; CS_LOCALS(r300); if (!r300screen->caps->has_tcl) { @@ -345,38 +559,40 @@ void r300_emit_vertex_shader(struct r300_context* r300, return; } - if (constants->count) { - BEGIN_CS(16 + (vs->instruction_count * 4) + (constants->count * 4)); + if (code->constants.Count) { + BEGIN_CS(14 + code->length + (code->constants.Count * 4)); } else { - BEGIN_CS(13 + (vs->instruction_count * 4) + (constants->count * 4)); + BEGIN_CS(11 + code->length); } - OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_0, R300_PVS_FIRST_INST(0) | - R300_PVS_LAST_INST(vs->instruction_count - 1)); - OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_1, vs->instruction_count - 1); - - /* XXX */ - OUT_CS_REG(R300_VAP_PVS_CONST_CNTL, 0x0); + /* R300_VAP_PVS_CODE_CNTL_0 + * R300_VAP_PVS_CONST_CNTL + * R300_VAP_PVS_CODE_CNTL_1 + * See the r5xx docs for instructions on how to use these. + * XXX these could be optimized to select better values... */ + OUT_CS_REG_SEQ(R300_VAP_PVS_CODE_CNTL_0, 3); + OUT_CS(R300_PVS_FIRST_INST(0) | + R300_PVS_XYZW_VALID_INST(instruction_count - 1) | + R300_PVS_LAST_INST(instruction_count - 1)); + OUT_CS(R300_PVS_MAX_CONST_ADDR(code->constants.Count - 1)); + OUT_CS(instruction_count - 1); OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 0); - OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, vs->instruction_count * 4); - for (i = 0; i < vs->instruction_count; i++) { - OUT_CS(vs->instructions[i].inst0); - OUT_CS(vs->instructions[i].inst1); - OUT_CS(vs->instructions[i].inst2); - OUT_CS(vs->instructions[i].inst3); - } + OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->length); + for (i = 0; i < code->length; i++) + OUT_CS(code->body.d[i]); - if (constants->count) { + if (code->constants.Count) { OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, (r300screen->caps->is_r500 ? R500_PVS_CONST_START : R300_PVS_CONST_START)); - OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, constants->count * 4); - for (i = 0; i < constants->count; i++) { - OUT_CS_32F(constants->constants[i][0]); - OUT_CS_32F(constants->constants[i][1]); - OUT_CS_32F(constants->constants[i][2]); - OUT_CS_32F(constants->constants[i][3]); + OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->constants.Count * 4); + for (i = 0; i < code->constants.Count; i++) { + const float * data = get_shader_constant(r300, &code->constants.Constants[i], constants); + OUT_CS_32F(data[0]); + OUT_CS_32F(data[1]); + OUT_CS_32F(data[2]); + OUT_CS_32F(data[3]); } } @@ -386,7 +602,12 @@ void r300_emit_vertex_shader(struct r300_context* r300, R300_PVS_VF_MAX_VTX_NUM(12)); OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0); END_CS; +} +void r300_emit_vertex_shader(struct r300_context* r300, + struct r300_vertex_shader* vs) +{ + r300_emit_vertex_program_code(r300, &vs->code, &r300->shader_constants[PIPE_SHADER_VERTEX]); } void r300_emit_viewport_state(struct r300_context* r300, @@ -403,7 +624,11 @@ void r300_emit_viewport_state(struct r300_context* r300, OUT_CS_32F(viewport->zscale); OUT_CS_32F(viewport->zoffset); - OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control); + if (r300->rs_state->enable_vte) { + OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control); + } else { + OUT_CS_REG(R300_VAP_VTE_CNTL, 0); + } END_CS; } @@ -421,23 +646,73 @@ void r300_flush_textures(struct r300_context* r300) void r300_emit_dirty_state(struct r300_context* r300) { struct r300_screen* r300screen = r300_screen(r300->context.screen); - int i; - int dirty_tex = 0; + struct r300_texture* tex; + int i, dirty_tex = 0; + boolean invalid = FALSE; - if (!(r300->dirty_hw)) { + if (!(r300->dirty_state)) { return; } r300_update_derived_state(r300); /* XXX check size */ - struct r300_texture* fb_tex = - (struct r300_texture*)r300->framebuffer_state.cbufs[0]; - r300->winsys->add_buffer(r300->winsys, fb_tex->buffer, - 0, RADEON_GEM_DOMAIN_VRAM); - if (r300->winsys->validate(r300->winsys)) { - /* XXX */ +validate: + /* Color buffers... */ + for (i = 0; i < r300->framebuffer_state.nr_cbufs; i++) { + tex = (struct r300_texture*)r300->framebuffer_state.cbufs[i]->texture; + assert(tex && tex->buffer && "cbuf is marked, but NULL!"); + if (!r300->winsys->add_buffer(r300->winsys, tex->buffer, + 0, RADEON_GEM_DOMAIN_VRAM)) { + r300->context.flush(&r300->context, 0, NULL); + goto validate; + } + } + /* ...depth buffer... */ + if (r300->framebuffer_state.zsbuf) { + tex = (struct r300_texture*)r300->framebuffer_state.zsbuf->texture; + assert(tex && tex->buffer && "zsbuf is marked, but NULL!"); + if (!r300->winsys->add_buffer(r300->winsys, tex->buffer, + 0, RADEON_GEM_DOMAIN_VRAM)) { + r300->context.flush(&r300->context, 0, NULL); + goto validate; + } + } + /* ...textures... */ + for (i = 0; i < r300->texture_count; i++) { + tex = r300->textures[i]; + assert(tex && tex->buffer && "texture is marked, but NULL!"); + if (!r300->winsys->add_buffer(r300->winsys, tex->buffer, + RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0)) { + r300->context.flush(&r300->context, 0, NULL); + goto validate; + } + } + /* ...occlusion query buffer... */ + if (!r300->winsys->add_buffer(r300->winsys, r300->oqbo, + 0, RADEON_GEM_DOMAIN_GTT)) { + r300->context.flush(&r300->context, 0, NULL); + goto validate; + } + /* ...and vertex buffer. */ + if (r300->vbo) { + if (!r300->winsys->add_buffer(r300->winsys, r300->vbo, + RADEON_GEM_DOMAIN_GTT, 0)) { + r300->context.flush(&r300->context, 0, NULL); + goto validate; + } + } else { + debug_printf("No VBO while emitting dirty state!\n"); + } + if (!r300->winsys->validate(r300->winsys)) { r300->context.flush(&r300->context, 0, NULL); + if (invalid) { + /* Well, hell. */ + debug_printf("r300: Stuck in validation loop, gonna quit now."); + exit(1); + } + invalid = TRUE; + goto validate; } if (r300->dirty_state & R300_NEW_BLEND) { @@ -450,6 +725,11 @@ void r300_emit_dirty_state(struct r300_context* r300) r300->dirty_state &= ~R300_NEW_BLEND_COLOR; } + if (r300->dirty_state & R300_NEW_CLIP) { + r300_emit_clip_state(r300, &r300->clip_state); + r300->dirty_state &= ~R300_NEW_CLIP; + } + if (r300->dirty_state & R300_NEW_DSA) { r300_emit_dsa_state(r300, r300->dsa_state); r300->dirty_state &= ~R300_NEW_DSA; @@ -457,11 +737,9 @@ void r300_emit_dirty_state(struct r300_context* r300) if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER) { if (r300screen->caps->is_r500) { - r500_emit_fragment_shader(r300, - (struct r500_fragment_shader*)r300->fs); + r500_emit_fragment_program_code(r300, &r300->fs->code, &r300->shader_constants[PIPE_SHADER_FRAGMENT]); } else { - r300_emit_fragment_shader(r300, - (struct r300_fragment_shader*)r300->fs); + r300_emit_fragment_program_code(r300, &r300->fs->code, &r300->shader_constants[PIPE_SHADER_FRAGMENT]); } r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER; } @@ -481,29 +759,27 @@ void r300_emit_dirty_state(struct r300_context* r300) r300->dirty_state &= ~R300_NEW_RS_BLOCK; } - if (r300->dirty_state & R300_ANY_NEW_SAMPLERS) { - for (i = 0; i < r300->sampler_count; i++) { - if (r300->dirty_state & (R300_NEW_SAMPLER << i)) { - r300_emit_sampler(r300, r300->sampler_states[i], i); - r300->dirty_state &= ~(R300_NEW_SAMPLER << i); - dirty_tex++; - } - } - } - if (r300->dirty_state & R300_NEW_SCISSOR) { r300_emit_scissor_state(r300, r300->scissor_state); r300->dirty_state &= ~R300_NEW_SCISSOR; } - if (r300->dirty_state & R300_ANY_NEW_TEXTURES) { - for (i = 0; i < r300->texture_count; i++) { - if (r300->dirty_state & (R300_NEW_TEXTURE << i)) { - r300_emit_texture(r300, r300->textures[i], i); - r300->dirty_state &= ~(R300_NEW_TEXTURE << i); + /* Samplers and textures are tracked separately but emitted together. */ + if (r300->dirty_state & + (R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES)) { + for (i = 0; i < MIN2(r300->sampler_count, r300->texture_count); i++) { + if (r300->dirty_state & + ((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i))) { + r300_emit_texture(r300, + r300->sampler_states[i], + r300->textures[i], + i); + r300->dirty_state &= + ~((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i)); dirty_tex++; } } + r300->dirty_state &= ~(R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES); } if (r300->dirty_state & R300_NEW_VIEWPORT) { @@ -519,4 +795,18 @@ void r300_emit_dirty_state(struct r300_context* r300) r300_emit_vertex_format_state(r300); r300->dirty_state &= ~R300_NEW_VERTEX_FORMAT; } + + if (r300->dirty_state & R300_NEW_VERTEX_SHADER) { + r300_emit_vertex_shader(r300, r300->vs); + r300->dirty_state &= ~R300_NEW_VERTEX_SHADER; + } + + /* XXX + assert(r300->dirty_state == 0); + */ + + /* Finally, emit the VBO. */ + r300_emit_vertex_buffer(r300); + + r300->dirty_hw++; } diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h index 31dbc7ab853..350691d592d 100644 --- a/src/gallium/drivers/r300/r300_emit.h +++ b/src/gallium/drivers/r300/r300_emit.h @@ -30,20 +30,28 @@ #include "r300_screen.h" #include "r300_state_inlines.h" +struct rX00_fragment_program_code; +struct r300_vertex_program_code; + void r300_emit_blend_state(struct r300_context* r300, struct r300_blend_state* blend); void r300_emit_blend_color_state(struct r300_context* r300, struct r300_blend_color_state* bc); +void r300_emit_clip_state(struct r300_context* r300, + struct pipe_clip_state* clip); + void r300_emit_dsa_state(struct r300_context* r300, struct r300_dsa_state* dsa); -void r300_emit_fragment_shader(struct r300_context* r300, - struct r300_fragment_shader* fs); +void r300_emit_fragment_program_code(struct r300_context* r300, + struct rX00_fragment_program_code* generic_code, + struct r300_constant_buffer* externals); -void r500_emit_fragment_shader(struct r300_context* r300, - struct r500_fragment_shader* fs); +void r500_emit_fragment_program_code(struct r300_context* r300, + struct rX00_fragment_program_code* generic_code, + struct r300_constant_buffer* externals); void r300_emit_fb_state(struct r300_context* r300, struct pipe_framebuffer_state* fb); @@ -53,17 +61,22 @@ void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs); void r300_emit_rs_block_state(struct r300_context* r300, struct r300_rs_block* rs); -void r300_emit_sampler(struct r300_context* r300, - struct r300_sampler_state* sampler, unsigned offset); - void r300_emit_scissor_state(struct r300_context* r300, struct r300_scissor_state* scissor); void r300_emit_texture(struct r300_context* r300, - struct r300_texture* tex, unsigned offset); + struct r300_sampler_state* sampler, + struct r300_texture* tex, + unsigned offset); + +void r300_emit_vertex_buffer(struct r300_context* r300); void r300_emit_vertex_format_state(struct r300_context* r300); +void r300_emit_vertex_program_code(struct r300_context* r300, + struct r300_vertex_program_code* code, + struct r300_constant_buffer* constants); + void r300_emit_vertex_shader(struct r300_context* r300, struct r300_vertex_shader* vs); diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c index 89a5f2b20cf..0dff1c6f4fb 100644 --- a/src/gallium/drivers/r300/r300_flush.c +++ b/src/gallium/drivers/r300/r300_flush.c @@ -29,7 +29,11 @@ static void r300_flush(struct pipe_context* pipe, struct r300_context* r300 = r300_context(pipe); CS_LOCALS(r300); - draw_flush(r300->draw); + /* We probably need to flush Draw, but we may have been called from + * within Draw. This feels kludgy, but it might be the best thing. */ + if (!r300->draw->flushing) { + draw_flush(r300->draw); + } if (r300->dirty_hw) { FLUSH_CS; diff --git a/src/gallium/drivers/r300/r300_flush.h b/src/gallium/drivers/r300/r300_flush.h index a1b224b39ce..9a83d89daab 100644 --- a/src/gallium/drivers/r300/r300_flush.h +++ b/src/gallium/drivers/r300/r300_flush.h @@ -23,6 +23,8 @@ #ifndef R300_FLUSH_H #define R300_FLUSH_H +#include "draw/draw_private.h" + #include "pipe/p_context.h" #include "r300_context.h" diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c new file mode 100644 index 00000000000..36463b9a2eb --- /dev/null +++ b/src/gallium/drivers/r300/r300_fs.c @@ -0,0 +1,137 @@ +/* + * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Joakim Sindholt <opensource@zhasha.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include "r300_fs.h" + +#include "r300_tgsi_to_rc.h" + +#include "radeon_compiler.h" + +static void find_output_registers(struct r300_fragment_program_compiler * compiler, + struct r300_fragment_shader * fs) +{ + unsigned i; + + /* Mark the outputs as not present initially */ + compiler->OutputColor = fs->info.num_outputs; + compiler->OutputDepth = fs->info.num_outputs; + + /* Now see where they really are. */ + for(i = 0; i < fs->info.num_outputs; ++i) { + switch(fs->info.output_semantic_name[i]) { + case TGSI_SEMANTIC_COLOR: + compiler->OutputColor = i; + break; + case TGSI_SEMANTIC_POSITION: + compiler->OutputDepth = i; + break; + } + } +} + +static void allocate_hardware_inputs( + struct r300_fragment_program_compiler * c, + void (*allocate)(void * data, unsigned input, unsigned hwreg), + void * mydata) +{ + struct tgsi_shader_info* info = &((struct r300_fragment_shader*)c->UserData)->info; + int total_colors = 0; + int colors = 0; + int total_generic = 0; + int generic = 0; + int i; + + for (i = 0; i < info->num_inputs; i++) { + switch (info->input_semantic_name[i]) { + case TGSI_SEMANTIC_COLOR: + total_colors++; + break; + case TGSI_SEMANTIC_FOG: + case TGSI_SEMANTIC_GENERIC: + total_generic++; + break; + } + } + + for(i = 0; i < info->num_inputs; i++) { + switch (info->input_semantic_name[i]) { + case TGSI_SEMANTIC_COLOR: + allocate(mydata, i, colors); + colors++; + break; + case TGSI_SEMANTIC_FOG: + case TGSI_SEMANTIC_GENERIC: + allocate(mydata, i, total_colors + generic); + generic++; + break; + } + } +} + +void r300_translate_fragment_shader(struct r300_context* r300, + struct r300_fragment_shader* fs) +{ + struct r300_fragment_program_compiler compiler; + struct tgsi_to_rc ttr; + + memset(&compiler, 0, sizeof(compiler)); + rc_init(&compiler.Base); + compiler.Base.Debug = 1; + + compiler.code = &fs->code; + compiler.is_r500 = r300_screen(r300->context.screen)->caps->is_r500; + compiler.AllocateHwInputs = &allocate_hardware_inputs; + compiler.UserData = fs; + + /* TODO: Program compilation depends on texture compare modes, + * which are sampler state. Therefore, programs need to be recompiled + * depending on this state as in the classic Mesa driver. + * + * This is not yet handled correctly. + */ + + find_output_registers(&compiler, fs); + + if (compiler.Base.Debug) { + debug_printf("r300: Initial fragment program\n"); + tgsi_dump(fs->state.tokens, 0); + } + + /* Translate TGSI to our internal representation */ + ttr.compiler = &compiler.Base; + ttr.info = &fs->info; + + r300_tgsi_to_rc(&ttr, fs->state.tokens); + + /* Invoke the compiler */ + r3xx_compile_fragment_program(&compiler); + if (compiler.Base.Error) { + /* Todo: Fail gracefully */ + fprintf(stderr, "r300 FP: Compiler error\n"); + abort(); + } + + /* And, finally... */ + rc_destroy(&compiler.Base); + fs->translated = TRUE; +} diff --git a/src/gallium/drivers/r300/r300_fs.h b/src/gallium/drivers/r300/r300_fs.h new file mode 100644 index 00000000000..9fab7894024 --- /dev/null +++ b/src/gallium/drivers/r300/r300_fs.h @@ -0,0 +1,51 @@ +/* + * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Joakim Sindholt <opensource@zhasha.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#ifndef R300_FS_H +#define R300_FS_H + +#include "tgsi/tgsi_dump.h" + +#include "r300_context.h" +#include "r3xx_fs.h" +#include "r5xx_fs.h" + +#include "radeon_code.h" + +struct r300_fragment_shader { + /* Parent class */ + struct pipe_shader_state state; + struct tgsi_shader_info info; + + /* Has this shader been translated yet? */ + boolean translated; + + /* Compiled code */ + struct rX00_fragment_program_code code; +}; + + +void r300_translate_fragment_shader(struct r300_context* r300, + struct r300_fragment_shader* fs); + + #endif /* R300_FS_H */ diff --git a/src/gallium/drivers/r300/r300_query.c b/src/gallium/drivers/r300/r300_query.c index 8fc61c2dec7..1d5185b417e 100644 --- a/src/gallium/drivers/r300/r300_query.c +++ b/src/gallium/drivers/r300/r300_query.c @@ -25,14 +25,30 @@ static struct pipe_query* r300_create_query(struct pipe_context* pipe, unsigned query_type) { - struct r300_query* q = CALLOC_STRUCT(r300_query); + struct r300_context* r300 = r300_context(pipe); + struct r300_screen* r300screen = r300_screen(r300->context.screen); + unsigned query_size = r300screen->caps->num_frag_pipes * 4; + struct r300_query* q, * qptr; + + q = CALLOC_STRUCT(r300_query); q->type = query_type; assert(q->type == PIPE_QUERY_OCCLUSION_COUNTER); - /* XXX this is to force winsys to give us a GTT buffer */ - q->buf = pipe->screen->buffer_create(pipe->screen, 64, - PIPE_BUFFER_USAGE_VERTEX, 64); + q->active = FALSE; + + if (!r300->query_list) { + r300->query_list = q; + } else if (!is_empty_list(r300->query_list)) { + qptr = last_elem(r300->query_list); + q->offset = qptr->offset + query_size; + insert_at_tail(r300->query_list, q); + } + + /* XXX */ + if (q->offset >= 4096) { + q->offset = 0; + } return (struct pipe_query*)q; } @@ -40,6 +56,9 @@ static struct pipe_query* r300_create_query(struct pipe_context* pipe, static void r300_destroy_query(struct pipe_context* pipe, struct pipe_query* query) { + struct r300_query* q = (struct r300_query*)query; + + remove_from_list(q); FREE(query); } @@ -49,15 +68,15 @@ static void r300_begin_query(struct pipe_context* pipe, uint32_t* map; struct r300_context* r300 = r300_context(pipe); struct r300_query* q = (struct r300_query*)query; - CS_LOCALS(r300); - map = pipe_buffer_map(pipe->screen, q->buf, PIPE_BUFFER_USAGE_CPU_WRITE); + map = pipe->screen->buffer_map(pipe->screen, r300->oqbo, + PIPE_BUFFER_USAGE_CPU_WRITE); + map += q->offset / 4; *map = ~0; - pipe_buffer_unmap(pipe->screen, q->buf); + pipe->screen->buffer_unmap(pipe->screen, r300->oqbo); - BEGIN_CS(2); - OUT_CS_REG(R300_ZB_ZPASS_DATA, 0); - END_CS; + r300_emit_dirty_state(r300); + r300_emit_query_begin(r300, q); } static void r300_end_query(struct pipe_context* pipe, @@ -65,12 +84,9 @@ static void r300_end_query(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); struct r300_query* q = (struct r300_query*)query; - CS_LOCALS(r300); - BEGIN_CS(4); - OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_RELOC(q->buf, 0, 0, RADEON_GEM_DOMAIN_GTT, 0); - END_CS; + r300_emit_dirty_state(r300); + r300_emit_query_end(r300, q); } static boolean r300_get_query_result(struct pipe_context* pipe, @@ -78,22 +94,38 @@ static boolean r300_get_query_result(struct pipe_context* pipe, boolean wait, uint64_t* result) { + struct r300_context* r300 = r300_context(pipe); + struct r300_screen* r300screen = r300_screen(r300->context.screen); struct r300_query* q = (struct r300_query*)query; + unsigned flags = PIPE_BUFFER_USAGE_CPU_READ; uint32_t* map; uint32_t temp; + unsigned i; if (wait) { - /* Well, we're expected to just sit here and spin, so let's go ahead - * and flush so we can be sure that the card's spinning... */ - /* XXX double-check these params */ pipe->flush(pipe, 0, NULL); + } else { + flags |= PIPE_BUFFER_USAGE_DONTBLOCK; } - map = pipe_buffer_map(pipe->screen, q->buf, PIPE_BUFFER_USAGE_CPU_READ); - temp = *map; - pipe_buffer_unmap(pipe->screen, q->buf); + map = pipe->screen->buffer_map(pipe->screen, r300->oqbo, flags); + map += q->offset / 4; + for (i = 0; i < r300screen->caps->num_frag_pipes; i++) { + if (*map == ~0) { + /* Looks like our results aren't ready yet. */ + if (wait) { + debug_printf("r300: Despite waiting, OQ results haven't" + " come in yet.\n"); + } + temp = ~0; + break; + } + temp += *map; + map++; + } + pipe->screen->buffer_unmap(pipe->screen, r300->oqbo); - if (temp < 0) { + if (temp == ~0) { /* Our results haven't been written yet... */ return FALSE; } diff --git a/src/gallium/drivers/r300/r300_query.h b/src/gallium/drivers/r300/r300_query.h index 4f447ea45b8..4f50e8f8440 100644 --- a/src/gallium/drivers/r300/r300_query.h +++ b/src/gallium/drivers/r300/r300_query.h @@ -27,12 +27,7 @@ #include "r300_cs.h" #include "r300_reg.h" -struct r300_query { - /* The kind of query. Currently only OQ is supported. */ - unsigned type; - /* Buffer object where we want our results to reside. */ - struct pipe_buffer* buf; -}; +struct r300_context; static INLINE struct r300_query* r300_query(struct pipe_query* q) { diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h index 660816e1da1..03cd219cde9 100644 --- a/src/gallium/drivers/r300/r300_reg.h +++ b/src/gallium/drivers/r300/r300_reg.h @@ -511,11 +511,13 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_PVS_XYZW_VALID_INST_SHIFT 10 # define R300_PVS_LAST_INST_SHIFT 20 # define R300_PVS_FIRST_INST(x) ((x) << 0) +# define R300_PVS_XYZW_VALID_INST(x) ((x) << 10) # define R300_PVS_LAST_INST(x) ((x) << 20) /* Addresses are relative the the vertex program parameters area. */ #define R300_VAP_PVS_CONST_CNTL 0x22D4 # define R300_PVS_CONST_BASE_OFFSET_SHIFT 0 # define R300_PVS_MAX_CONST_ADDR_SHIFT 16 +# define R300_PVS_MAX_CONST_ADDR(x) ((x) << 16) #define R300_VAP_PVS_CODE_CNTL_1 0x22D8 # define R300_PVS_LAST_VTX_SRC_INST_SHIFT 0 #define R300_VAP_PVS_FLOW_CNTL_OPC 0x22DC @@ -1062,8 +1064,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. R300_GA_COLOR_CONTROL_RGB2_SHADING_FLAT | \ R300_GA_COLOR_CONTROL_ALPHA2_SHADING_FLAT | \ R300_GA_COLOR_CONTROL_RGB3_SHADING_FLAT | \ - R300_GA_COLOR_CONTROL_ALPHA3_SHADING_FLAT | \ - R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST ) + R300_GA_COLOR_CONTROL_ALPHA3_SHADING_FLAT ) # define R300_SHADE_MODEL_SMOOTH ( \ R300_GA_COLOR_CONTROL_RGB0_SHADING_GOURAUD | \ @@ -1073,8 +1074,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. R300_GA_COLOR_CONTROL_RGB2_SHADING_GOURAUD | \ R300_GA_COLOR_CONTROL_ALPHA2_SHADING_GOURAUD | \ R300_GA_COLOR_CONTROL_RGB3_SHADING_GOURAUD | \ - R300_GA_COLOR_CONTROL_ALPHA3_SHADING_GOURAUD | \ - R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST ) + R300_GA_COLOR_CONTROL_ALPHA3_SHADING_GOURAUD ) /* Specifies red & green components of fill color -- S312 format -- Backwards comp. */ #define R300_GA_SOLID_RG 0x427c @@ -1478,6 +1478,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_PITCH_EN (1 << 31) # define R300_TX_WIDTH(x) ((x) << 0) # define R300_TX_HEIGHT(x) ((x) << 11) +# define R300_TX_NUM_LEVELS(x) ((x) << 26) #define R300_TX_FORMAT1_0 0x44C0 /* The interpretation of the format word by Wladimir van der Laan */ @@ -1485,9 +1486,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. They are given meanings as R, G, B and Alpha by the swizzle specification */ # define R300_TX_FORMAT_X8 0x0 -# define R500_TX_FORMAT_X1 0x0 // bit set in format 2 # define R300_TX_FORMAT_X16 0x1 -# define R500_TX_FORMAT_X1_REV 0x0 // bit set in format 2 # define R300_TX_FORMAT_Y4X4 0x2 # define R300_TX_FORMAT_Y8X8 0x3 # define R300_TX_FORMAT_Y16X16 0x4 @@ -1504,31 +1503,29 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_FORMAT_DXT1 0xF # define R300_TX_FORMAT_DXT3 0x10 # define R300_TX_FORMAT_DXT5 0x11 -# define R300_TX_FORMAT_D3DMFT_CxV8U8 0x12 /* no swizzle */ -# define R300_TX_FORMAT_A8R8G8B8 0x13 /* no swizzle */ -# define R300_TX_FORMAT_B8G8_B8G8 0x14 /* no swizzle */ -# define R300_TX_FORMAT_G8R8_G8B8 0x15 /* no swizzle */ - - /* These two values are wrong, but they're the only values that - * produce any even vaguely correct results. Can r300 only do 16-bit - * depth textures? - */ -# define R300_TX_FORMAT_X24_Y8 0x1e -# define R300_TX_FORMAT_X32 0x1e - - /* 0x16 - some 16 bit green format.. ?? */ +# define R300_TX_FORMAT_Y8 0x12 +# define R300_TX_FORMAT_AVYU444 0x13 +# define R300_TX_FORMAT_VYUY422 0x14 +# define R300_TX_FORMAT_YVYU422 0x15 +# define R300_TX_FORMAT_16_MPEG 0x16 +# define R300_TX_FORMAT_16_16_MPEG 0x17 +# define R300_TX_FORMAT_16F 0x18 +# define R300_TX_FORMAT_16F_16F 0x19 +# define R300_TX_FORMAT_16F_16F_16F_16F 0x1A +# define R300_TX_FORMAT_32F 0x1B +# define R300_TX_FORMAT_32F_32F 0x1C +# define R300_TX_FORMAT_32F_32F_32F_32F 0x1D +# define R300_TX_FORMAT_W24_FP 0x1E + +# define R300_TX_FORMAT_SIGNED_W (1 << 5) +# define R300_TX_FORMAT_SIGNED_Z (1 << 6) +# define R300_TX_FORMAT_SIGNED_Y (1 << 7) +# define R300_TX_FORMAT_SIGNED_X (1 << 8) +# define R300_TX_FORMAT_SIGNED (0xf << 5) + # define R300_TX_FORMAT_3D (1 << 25) # define R300_TX_FORMAT_CUBIC_MAP (2 << 25) - /* gap */ - /* Floating point formats */ - /* Note - hardware supports both 16 and 32 bit floating point */ -# define R300_TX_FORMAT_FL_I16 0x18 -# define R300_TX_FORMAT_FL_I16A16 0x19 -# define R300_TX_FORMAT_FL_R16G16B16A16 0x1A -# define R300_TX_FORMAT_FL_I32 0x1B -# define R300_TX_FORMAT_FL_I32A32 0x1C -# define R300_TX_FORMAT_FL_R32G32B32A32 0x1D /* alpha modes, convenience mostly */ /* if you have alpha, pick constant appropriate to the number of channels (1 for I8, 2 for I8A8, 4 for R8G8B8A8, etc */ @@ -1569,7 +1566,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_FORMAT_CONST_Z (4<<5) # define R300_TX_FORMAT_CONST_W (8<<5) -# define R300_TX_FORMAT_YUV_MODE 0x00800000 +# define R300_TX_FORMAT_GAMMA (1 << 21) +# define R300_TX_FORMAT_YUV_TO_RGB (1 << 22) #define R300_TX_FORMAT2_0 0x4500 /* obvious missing in gap */ # define R300_TX_PITCHMASK_SHIFT 0 @@ -3040,6 +3038,7 @@ enum { # define R500_INST_RGB_WMASK_R (1 << 11) # define R500_INST_RGB_WMASK_G (1 << 12) # define R500_INST_RGB_WMASK_B (1 << 13) +# define R500_INST_RGB_WMASK_RGB (7 << 11) # define R500_INST_ALPHA_WMASK (1 << 14) # define R500_INST_RGB_OMASK_R (1 << 15) # define R500_INST_RGB_OMASK_G (1 << 16) @@ -3313,6 +3312,10 @@ enum { #define R200_3D_DRAW_IMMD_2 0xC0003500 +/* XXX Oh look, stuff not brought over from docs yet */ + +#define R300_SU_REG_DEST 0x42C8 + #endif /* _R300_REG_H */ /* *INDENT-ON* */ diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c index cbd84d7c569..cd458d019ae 100644 --- a/src/gallium/drivers/r300/r300_render.c +++ b/src/gallium/drivers/r300/r300_render.c @@ -45,11 +45,7 @@ struct r300_render { /* VBO */ struct pipe_buffer* vbo; - size_t vbo_size; - size_t vbo_offset; - void* vbo_map; size_t vbo_alloc_size; - size_t vbo_max_used; }; static INLINE struct r300_render* @@ -78,24 +74,21 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render, struct pipe_screen* screen = r300->context.screen; size_t size = (size_t)vertex_size * (size_t)count; - if (r300render->vbo) { + if (r300render->vbo && (size > r300render->vbo_alloc_size)) { pipe_buffer_reference(&r300render->vbo, NULL); } + + if (!r300render->vbo) { + r300render->vbo = pipe_buffer_create(screen, + 64, + PIPE_BUFFER_USAGE_VERTEX, + size); + } - r300render->vbo_size = MAX2(size, r300render->vbo_alloc_size); - r300render->vbo_offset = 0; - r300render->vbo = pipe_buffer_create(screen, - 64, - PIPE_BUFFER_USAGE_VERTEX, - r300render->vbo_size); - + r300render->vbo_alloc_size = MAX2(size, r300render->vbo_alloc_size); r300render->vertex_size = vertex_size; - if (r300render->vbo) { - return TRUE; - } else { - return FALSE; - } + return (r300render->vbo) ? TRUE : FALSE; } static void* r300_render_map_vertices(struct vbuf_render* render) @@ -103,10 +96,8 @@ static void* r300_render_map_vertices(struct vbuf_render* render) struct r300_render* r300render = r300_render(render); struct pipe_screen* screen = r300render->r300->context.screen; - r300render->vbo_map = pipe_buffer_map(screen, r300render->vbo, - PIPE_BUFFER_USAGE_CPU_WRITE); - - return (unsigned char*)r300render->vbo_map + r300render->vbo_offset; + return (unsigned char*)pipe_buffer_map(screen, r300render->vbo, + PIPE_BUFFER_USAGE_CPU_WRITE); } static void r300_render_unmap_vertices(struct vbuf_render* render, @@ -116,9 +107,6 @@ static void r300_render_unmap_vertices(struct vbuf_render* render, struct r300_render* r300render = r300_render(render); struct pipe_screen* screen = r300render->r300->context.screen; - r300render->vbo_max_used = MAX2(r300render->vbo_max_used, - r300render->vertex_size * (max + 1)); - pipe_buffer_unmap(screen, r300render->vbo); } @@ -180,27 +168,9 @@ static void prepare_render(struct r300_render* render, unsigned count) CS_LOCALS(r300); - /* Make sure that all possible state is emitted. */ - r300_emit_dirty_state(r300); + r300->vbo = render->vbo; - debug_printf("r300: Preparing vertex buffer %p for render, " - "vertex size %d, vertex count %d\n", render->vbo, - r300->vertex_info.vinfo.size, count); - /* Set the pointer to our vertex buffer. The emitted values are this: - * PACKET3 [3D_LOAD_VBPNTR] - * COUNT [1] - * FORMAT [size | stride << 8] - * OFFSET [0] - * VBPNTR [relocated BO] - */ - BEGIN_CS(7); - OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3); - OUT_CS(1); - OUT_CS(r300->vertex_info.vinfo.size | - (r300->vertex_info.vinfo.size << 8)); - OUT_CS(render->vbo_offset); - OUT_CS_RELOC(render->vbo, 0, RADEON_GEM_DOMAIN_GTT, 0, 0); - END_CS; + r300_emit_dirty_state(r300); } static void r300_render_draw_arrays(struct vbuf_render* render, @@ -212,8 +182,6 @@ static void r300_render_draw_arrays(struct vbuf_render* render, CS_LOCALS(r300); - r300render->vbo_offset = start; - prepare_render(r300render, count); debug_printf("r300: Doing vbuf render, count %d\n", count); @@ -248,13 +216,14 @@ static void r300_render_draw(struct vbuf_render* render, return; } +/* index_map = pipe_buffer_map(screen, index_buffer, PIPE_BUFFER_USAGE_CPU_WRITE); memcpy(index_map, indices, count); pipe_buffer_unmap(screen, index_buffer); debug_printf("r300: Doing indexbuf render, count %d\n", count); -/* + BEGIN_CS(8); OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0); OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | @@ -264,13 +233,15 @@ static void r300_render_draw(struct vbuf_render* render, OUT_CS_INDEX_RELOC(index_buffer, 0, count, RADEON_GEM_DOMAIN_GTT, 0, 0); END_CS; */ - BEGIN_CS(2 + count); - OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count); + BEGIN_CS(2 + (count+1)/2); + OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2); OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | - r300render->hwprim | R300_VAP_VF_CNTL__INDEX_SIZE_32bit); - for (i = 0; i < count; i++) { - index = indices[i]; - OUT_CS(index); + r300render->hwprim); + for (i = 0; i < count-1; i += 2) { + OUT_CS(indices[i+1] << 16 | indices[i]); + } + if (count % 2) { + OUT_CS(indices[count-1]); } END_CS; } diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c index d2c5998c261..15740f61252 100644 --- a/src/gallium/drivers/r300/r300_screen.c +++ b/src/gallium/drivers/r300/r300_screen.c @@ -87,23 +87,24 @@ static int r300_get_param(struct pipe_screen* pscreen, int param) } else { return 0; } - return 0; case PIPE_CAP_GLSL: - /* IN THEORY */ - return 0; + if (r300screen->caps->is_r500) { + return 1; + } else { + return 0; + } case PIPE_CAP_S3TC: - /* IN THEORY */ - return 0; + return 1; case PIPE_CAP_ANISOTROPIC_FILTER: - /* IN THEORY */ - return 0; + return 1; case PIPE_CAP_POINT_SPRITE: /* IN THEORY */ return 0; case PIPE_CAP_MAX_RENDER_TARGETS: return 4; case PIPE_CAP_OCCLUSION_QUERY: - return 1; + /* IN THEORY */ + return 0; case PIPE_CAP_TEXTURE_SHADOW_MAP: /* IN THEORY */ return 0; @@ -143,6 +144,11 @@ static int r300_get_param(struct pipe_screen* pscreen, int param) case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS: /* XXX guessing (what a terrible guess) */ return 2; + case PIPE_CAP_TGSI_CONT_SUPPORTED: + /* XXX */ + return 0; + case PIPE_CAP_BLEND_EQUATION_SEPARATE: + return 1; default: debug_printf("r300: Implementation error: Bad param %d\n", param); @@ -152,17 +158,20 @@ static int r300_get_param(struct pipe_screen* pscreen, int param) static float r300_get_paramf(struct pipe_screen* pscreen, int param) { + struct r300_screen* r300screen = r300_screen(pscreen); + switch (param) { case PIPE_CAP_MAX_LINE_WIDTH: case PIPE_CAP_MAX_LINE_WIDTH_AA: - /* XXX this is the biggest thing that will fit in that register. - * Perhaps the actual rendering limits are less? */ - return 10922.0f; case PIPE_CAP_MAX_POINT_WIDTH: case PIPE_CAP_MAX_POINT_WIDTH_AA: - /* XXX this is the biggest thing that will fit in that register. - * Perhaps the actual rendering limits are less? */ - return 10922.0f; + /* The maximum dimensions of the colorbuffer are our practical + * rendering limits. 2048 pixels should be enough for anybody. */ + if (r300screen->caps->is_r500) { + return 4096.0f; + } else { + return 2048.0f; + } case PIPE_CAP_MAX_TEXTURE_ANISOTROPY: return 16.0f; case PIPE_CAP_MAX_TEXTURE_LOD_BIAS: @@ -174,21 +183,58 @@ static float r300_get_paramf(struct pipe_screen* pscreen, int param) } } -static boolean check_tex_2d_format(enum pipe_format format, boolean is_r500) +static boolean check_tex_2d_format(enum pipe_format format, uint32_t usage, + boolean is_r500) { switch (format) { + /* Supported formats. */ /* Colorbuffer */ case PIPE_FORMAT_A4R4G4B4_UNORM: case PIPE_FORMAT_R5G6B5_UNORM: case PIPE_FORMAT_A1R5G5B5_UNORM: - case PIPE_FORMAT_A8R8G8B8_UNORM: + return usage & + (PIPE_TEXTURE_USAGE_RENDER_TARGET | + PIPE_TEXTURE_USAGE_DISPLAY_TARGET | + PIPE_TEXTURE_USAGE_PRIMARY); + + /* Texture */ + case PIPE_FORMAT_A8R8G8B8_SRGB: + case PIPE_FORMAT_R8G8B8A8_SRGB: + case PIPE_FORMAT_DXT1_RGB: + case PIPE_FORMAT_DXT1_RGBA: + case PIPE_FORMAT_DXT3_RGBA: + case PIPE_FORMAT_DXT5_RGBA: + case PIPE_FORMAT_YCBCR: + return usage & PIPE_TEXTURE_USAGE_SAMPLER; + /* Colorbuffer or texture */ + case PIPE_FORMAT_A8R8G8B8_UNORM: + case PIPE_FORMAT_R8G8B8A8_UNORM: case PIPE_FORMAT_I8_UNORM: + return usage & + (PIPE_TEXTURE_USAGE_RENDER_TARGET | + PIPE_TEXTURE_USAGE_DISPLAY_TARGET | + PIPE_TEXTURE_USAGE_PRIMARY | + PIPE_TEXTURE_USAGE_SAMPLER); + /* Z buffer */ case PIPE_FORMAT_Z16_UNORM: - /* Z buffer with stencil */ + return usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL; + + /* Z buffer with stencil or texture */ case PIPE_FORMAT_Z24S8_UNORM: - return TRUE; + return usage & + (PIPE_TEXTURE_USAGE_DEPTH_STENCIL | + PIPE_TEXTURE_USAGE_SAMPLER); + + /* Definitely unsupported formats. */ + /* Non-usable Z buffer/stencil formats. */ + case PIPE_FORMAT_Z24X8_UNORM: + case PIPE_FORMAT_S8Z24_UNORM: + case PIPE_FORMAT_X8Z24_UNORM: + debug_printf("r300: Note: Got unsupported format: %s in %s\n", + pf_name(format), __FUNCTION__); + return FALSE; /* XXX These don't even exist case PIPE_FORMAT_A32R32G32B32: @@ -211,7 +257,8 @@ static boolean check_tex_2d_format(enum pipe_format format, boolean is_r500) return FALSE; */ default: - debug_printf("r300: Warning: Got unsupported format: %s in %s\n", + /* Unknown format... */ + debug_printf("r300: Warning: Got unknown format: %s in %s\n", pf_name(format), __FUNCTION__); break; } @@ -228,11 +275,18 @@ static boolean r300_is_format_supported(struct pipe_screen* pscreen, { switch (target) { case PIPE_TEXTURE_2D: - return check_tex_2d_format(format, + return check_tex_2d_format(format, tex_usage, r300_screen(pscreen)->caps->is_r500); + case PIPE_TEXTURE_1D: + case PIPE_TEXTURE_3D: + case PIPE_TEXTURE_CUBE: + debug_printf("r300: Implementation error: Unsupported format " + "target: %d\n", target); + break; default: - debug_printf("r300: Warning: Got unknown format target: %d\n", - format); + debug_printf("r300: Fatal: This is not a format target: %d\n", + target); + assert(0); break; } @@ -268,13 +322,14 @@ r300_get_tex_transfer(struct pipe_screen *screen, trans = CALLOC_STRUCT(r300_transfer); if (trans) { pipe_texture_reference(&trans->transfer.texture, texture); - trans->transfer.format = trans->transfer.format; + trans->transfer.format = texture->format; trans->transfer.width = w; trans->transfer.height = h; trans->transfer.block = texture->block; trans->transfer.nblocksx = texture->nblocksx[level]; trans->transfer.nblocksy = texture->nblocksy[level]; - trans->transfer.stride = tex->stride; + trans->transfer.stride = align(pf_get_stride(&trans->transfer.block, + texture->width[level]), 32); trans->transfer.usage = usage; trans->offset = offset; } diff --git a/src/gallium/drivers/r300/r300_screen.h b/src/gallium/drivers/r300/r300_screen.h index 3f52dbc3bea..2a0e41fbc3b 100644 --- a/src/gallium/drivers/r300/r300_screen.h +++ b/src/gallium/drivers/r300/r300_screen.h @@ -49,7 +49,7 @@ struct r300_transfer { }; /* Convenience cast wrapper. */ -static struct r300_screen* r300_screen(struct pipe_screen* screen) { +static INLINE struct r300_screen* r300_screen(struct pipe_screen* screen) { return (struct r300_screen*)screen; } diff --git a/src/gallium/drivers/r300/r300_shader_inlines.h b/src/gallium/drivers/r300/r300_shader_inlines.h new file mode 100644 index 00000000000..a04f45b03e2 --- /dev/null +++ b/src/gallium/drivers/r300/r300_shader_inlines.h @@ -0,0 +1,47 @@ +/* + * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com> + * Joakim Sindholt <opensource@zhasha.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#ifndef R300_SHADER_INLINES_H +#define R300_SHADER_INLINES_H + +/* TGSI constants. TGSI is like XML: If it can't solve your problems, you're + * not using enough of it. */ +static const struct tgsi_full_src_register r300_constant_zero = { + .SrcRegister.Extended = TRUE, + .SrcRegister.File = TGSI_FILE_NULL, + .SrcRegisterExtSwz.ExtSwizzleX = TGSI_EXTSWIZZLE_ZERO, + .SrcRegisterExtSwz.ExtSwizzleY = TGSI_EXTSWIZZLE_ZERO, + .SrcRegisterExtSwz.ExtSwizzleZ = TGSI_EXTSWIZZLE_ZERO, + .SrcRegisterExtSwz.ExtSwizzleW = TGSI_EXTSWIZZLE_ZERO, +}; + +static const struct tgsi_full_src_register r300_constant_one = { + .SrcRegister.Extended = TRUE, + .SrcRegister.File = TGSI_FILE_NULL, + .SrcRegisterExtSwz.ExtSwizzleX = TGSI_EXTSWIZZLE_ONE, + .SrcRegisterExtSwz.ExtSwizzleY = TGSI_EXTSWIZZLE_ONE, + .SrcRegisterExtSwz.ExtSwizzleZ = TGSI_EXTSWIZZLE_ONE, + .SrcRegisterExtSwz.ExtSwizzleW = TGSI_EXTSWIZZLE_ONE, +}; + +#endif /* R300_SHADER_INLINES_H */ diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index 184a23c9e68..c16cadd0407 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -24,12 +24,15 @@ #include "util/u_pack_color.h" #include "util/u_debug.h" + +#include "pipe/p_config.h" #include "pipe/internal/p_winsys_screen.h" #include "r300_context.h" #include "r300_reg.h" #include "r300_state_inlines.h" -#include "r300_state_shader.h" +#include "r300_fs.h" +#include "r300_vs.h" /* r300_state: Functions used to intialize state context by translating * Gallium state objects into semi-native r300 state objects. */ @@ -62,8 +65,6 @@ static void* r300_create_blend_state(struct pipe_context* pipe, } /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ - /* XXX are logicops still allowed if blending's disabled? - * Does Gallium take care of it for us? */ if (state->logicop_enable) { blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE | (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; @@ -121,9 +122,14 @@ static void r300_set_clip_state(struct pipe_context* pipe, const struct pipe_clip_state* state) { struct r300_context* r300 = r300_context(pipe); - /* XXX Draw */ - draw_flush(r300->draw); - draw_set_clip_state(r300->draw, state); + + if (r300_screen(pipe->screen)->caps->has_tcl) { + r300->clip_state = *state; + r300->dirty_state |= R300_NEW_CLIP; + } else { + draw_flush(r300->draw); + draw_set_clip_state(r300->draw, state); + } } static void @@ -132,7 +138,6 @@ static void const struct pipe_constant_buffer* buffer) { struct r300_context* r300 = r300_context(pipe); - int i = r300->shader_constants[shader].user_count; /* This entire chunk of code seems ever-so-slightly baked. * It's as if I've got pipe_buffer* matryoshkas... */ @@ -143,24 +148,13 @@ static void map, buffer->buffer->size); pipe->winsys->buffer_unmap(pipe->winsys, buffer->buffer); - r300->shader_constants[shader].user_count = + r300->shader_constants[shader].count = buffer->buffer->size / (sizeof(float) * 4); } else { - r300->shader_constants[shader].user_count = 0; + r300->shader_constants[shader].count = 0; } r300->dirty_state |= R300_NEW_CONSTANTS; - - /* If the number of constants have changed, invalidate the shader. */ - if (r300->shader_constants[shader].user_count != i) { - if (shader == PIPE_SHADER_FRAGMENT && r300->fs) { - r300->fs->translated = FALSE; - r300_translate_fragment_shader(r300, r300->fs); - } else if (shader == PIPE_SHADER_VERTEX && r300->vs) { - r300->vs->translated = FALSE; - r300_translate_vertex_shader(r300, r300->vs); - } - } } /* Create a new depth, stencil, and alpha state based on the CSO dsa state. @@ -230,7 +224,8 @@ static void* dsa->alpha_reference = CLAMP(state->alpha.ref_value * 1023.0f, 0, 1023); } else { - dsa->z_buffer_top = R300_ZTOP_ENABLE; + /* XXX need to fix this to be dynamically set + dsa->z_buffer_top = R300_ZTOP_ENABLE; */ } return (void*)dsa; @@ -257,6 +252,7 @@ static void r300_set_edgeflags(struct pipe_context* pipe, const unsigned* bitfield) { /* XXX you know it's bad when i915 has this blank too */ + /* XXX and even worse, I have no idea WTF the bitfield is */ } static void @@ -276,19 +272,13 @@ static void static void* r300_create_fs_state(struct pipe_context* pipe, const struct pipe_shader_state* shader) { - struct r300_context* r300 = r300_context(pipe); - struct r3xx_fragment_shader* fs = NULL; + struct r300_fragment_shader* fs = NULL; - if (r300_screen(r300->context.screen)->caps->is_r500) { - fs = - (struct r3xx_fragment_shader*)CALLOC_STRUCT(r500_fragment_shader); - } else { - fs = - (struct r3xx_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); - } + fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); /* Copy state directly into shader. */ fs->state = *shader; + fs->state.tokens = tgsi_dup_tokens(shader->tokens); tgsi_scan_shader(shader->tokens, &fs->info); @@ -299,7 +289,7 @@ static void* r300_create_fs_state(struct pipe_context* pipe, static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) { struct r300_context* r300 = r300_context(pipe); - struct r3xx_fragment_shader* fs = (struct r3xx_fragment_shader*)shader; + struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; if (fs == NULL) { r300->fs = NULL; @@ -308,7 +298,6 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) r300_translate_fragment_shader(r300, fs); } - fs->translated = TRUE; r300->fs = fs; r300->dirty_state |= R300_NEW_FRAGMENT_SHADER; @@ -317,13 +306,16 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) /* Delete fragment shader state. */ static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) { + struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; + rc_constants_destroy(&fs->code.constants); + FREE(fs->state.tokens); FREE(shader); } static void r300_set_polygon_stipple(struct pipe_context* pipe, const struct pipe_poly_stipple* state) { - /* XXX */ + /* XXX no idea how to set this up, but not terribly important */ } /* Create a new rasterizer state based on the CSO rasterizer state. @@ -341,14 +333,21 @@ static void* r300_create_rs_state(struct pipe_context* pipe, /* Copy rasterizer state for Draw. */ rs->rs = *state; + rs->enable_vte = !state->bypass_vs_clip_and_viewport; + +#ifdef PIPE_ARCH_LITTLE_ENDIAN + rs->vap_control_status = R300_VC_NO_SWAP; +#else + rs->vap_control_status = R300_VC_32BIT_SWAP; +#endif + /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL. * Else, enable HW TCL and force Draw's TCL off. */ if (state->bypass_vs_clip_and_viewport || !r300_screen(pipe->screen)->caps->has_tcl) { - rs->vap_control_status = R300_VAP_TCL_BYPASS; + rs->vap_control_status |= R300_VAP_TCL_BYPASS; } else { rs->rs.bypass_vs_clip_and_viewport = TRUE; - rs->vap_control_status = 0; } rs->point_size = pack_float_16_6x(state->point_size) | @@ -412,6 +411,10 @@ static void* r300_create_rs_state(struct pipe_context* pipe, rs->color_control = R300_SHADE_MODEL_SMOOTH; } + if (!state->flatshade_first) { + rs->color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST; + } + return (void*)rs; } @@ -535,16 +538,16 @@ static void r300_set_scissor_state(struct pipe_context* pipe, (state->minx << R300_SCISSORS_X_SHIFT) | (state->miny << R300_SCISSORS_Y_SHIFT); r300->scissor_state->scissor_bottom_right = - (state->maxx << R300_SCISSORS_X_SHIFT) | - (state->maxy << R300_SCISSORS_Y_SHIFT); + ((state->maxx - 1) << R300_SCISSORS_X_SHIFT) | + ((state->maxy - 1) << R300_SCISSORS_Y_SHIFT); } else { /* Offset of 1440 in non-R500 chipsets. */ r300->scissor_state->scissor_top_left = ((state->minx + 1440) << R300_SCISSORS_X_SHIFT) | ((state->miny + 1440) << R300_SCISSORS_Y_SHIFT); r300->scissor_state->scissor_bottom_right = - ((state->maxx + 1440) << R300_SCISSORS_X_SHIFT) | - ((state->maxy + 1440) << R300_SCISSORS_Y_SHIFT); + (((state->maxx - 1) + 1440) << R300_SCISSORS_X_SHIFT) | + (((state->maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT); } r300->dirty_state |= R300_NEW_SCISSOR; @@ -555,40 +558,35 @@ static void r300_set_viewport_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); - draw_flush(r300->draw); - - if (r300_screen(r300->context.screen)->caps->has_tcl) { - /* Do the transform in HW. */ - r300->viewport_state->vte_control = R300_VTX_W0_FMT; + /* Do the transform in HW. */ + r300->viewport_state->vte_control = R300_VTX_W0_FMT; - if (state->scale[0] != 1.0f) { - r300->viewport_state->xscale = state->scale[0]; - r300->viewport_state->vte_control |= R300_VPORT_X_SCALE_ENA; - } - if (state->scale[1] != 1.0f) { - r300->viewport_state->yscale = state->scale[1]; - r300->viewport_state->vte_control |= R300_VPORT_Y_SCALE_ENA; - } - if (state->scale[2] != 1.0f) { - r300->viewport_state->zscale = state->scale[2]; - r300->viewport_state->vte_control |= R300_VPORT_Z_SCALE_ENA; - } - if (state->translate[0] != 0.0f) { - r300->viewport_state->xoffset = state->translate[0]; - r300->viewport_state->vte_control |= R300_VPORT_X_OFFSET_ENA; - } - if (state->translate[1] != 0.0f) { - r300->viewport_state->yoffset = state->translate[1]; - r300->viewport_state->vte_control |= R300_VPORT_Y_OFFSET_ENA; - } - if (state->translate[2] != 0.0f) { - r300->viewport_state->zoffset = state->translate[2]; - r300->viewport_state->vte_control |= R300_VPORT_Z_OFFSET_ENA; - } - } else { - r300->viewport_state->vte_control = 0; - /* Have Draw do the actual transform. */ - draw_set_viewport_state(r300->draw, state); + if (state->scale[0] != 1.0f) { + assert(state->scale[0] != 0.0f); + r300->viewport_state->xscale = state->scale[0]; + r300->viewport_state->vte_control |= R300_VPORT_X_SCALE_ENA; + } + if (state->scale[1] != 1.0f) { + assert(state->scale[1] != 0.0f); + r300->viewport_state->yscale = state->scale[1]; + r300->viewport_state->vte_control |= R300_VPORT_Y_SCALE_ENA; + } + if (state->scale[2] != 1.0f) { + assert(state->scale[2] != 0.0f); + r300->viewport_state->zscale = state->scale[2]; + r300->viewport_state->vte_control |= R300_VPORT_Z_SCALE_ENA; + } + if (state->translate[0] != 0.0f) { + r300->viewport_state->xoffset = state->translate[0]; + r300->viewport_state->vte_control |= R300_VPORT_X_OFFSET_ENA; + } + if (state->translate[1] != 0.0f) { + r300->viewport_state->yoffset = state->translate[1]; + r300->viewport_state->vte_control |= R300_VPORT_Y_OFFSET_ENA; + } + if (state->translate[2] != 0.0f) { + r300->viewport_state->zoffset = state->translate[2]; + r300->viewport_state->vte_control |= R300_VPORT_Z_OFFSET_ENA; } r300->dirty_state |= R300_NEW_VIEWPORT; @@ -628,6 +626,7 @@ static void* r300_create_vs_state(struct pipe_context* pipe, struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); /* Copy state directly into shader. */ vs->state = *shader; + vs->state.tokens = tgsi_dup_tokens(shader->tokens); tgsi_scan_shader(shader->tokens, &vs->info); @@ -672,7 +671,9 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) if (r300_screen(pipe->screen)->caps->has_tcl) { struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; + rc_constants_destroy(&vs->code.constants); draw_delete_vertex_shader(r300->draw, vs->draw); + FREE(vs->state.tokens); FREE(shader); } else { draw_delete_vertex_shader(r300->draw, diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index c4c9784a00c..c01e61a9b19 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -22,6 +22,9 @@ #include "r300_state_derived.h" +#include "r300_fs.h" +#include "r300_vs.h" + /* r300_state_derived: Various bits of state which are dependent upon * currently bound CSO data. */ @@ -46,63 +49,71 @@ static void r300_vs_tab_routes(struct r300_context* r300, assert(info->num_inputs <= 16); - if (r300screen->caps->has_tcl) { - /* Just copy vert attribs over as-is. */ + if (!r300screen->caps->has_tcl || !r300->rs_state->enable_vte) + { for (i = 0; i < info->num_inputs; i++) { - tab[i] = i; - } - for (i = 0; i < info->num_outputs; i++) { - switch (info->output_semantic_name[i]) { + switch (info->input_semantic_name[i]) { case TGSI_SEMANTIC_POSITION: pos = TRUE; + tab[i] = 0; break; case TGSI_SEMANTIC_COLOR: + tab[i] = 2 + cols; cols++; break; case TGSI_SEMANTIC_PSIZE: psize = TRUE; + tab[i] = 15; break; case TGSI_SEMANTIC_FOG: fog = TRUE; + /* Fall through */ case TGSI_SEMANTIC_GENERIC: + tab[i] = 6 + texs; texs++; break; default: - debug_printf("r300: Unknown vertex output %d\n", - info->output_semantic_name[i]); + debug_printf("r300: Unknown vertex input %d\n", + info->input_semantic_name[i]); break; } } - } else { + } + else + { + /* Just copy vert attribs over as-is. */ for (i = 0; i < info->num_inputs; i++) { - switch (info->input_semantic_name[i]) { + tab[i] = i; + } + + for (i = 0; i < info->num_outputs; i++) { + switch (info->output_semantic_name[i]) { case TGSI_SEMANTIC_POSITION: pos = TRUE; - tab[i] = 0; break; case TGSI_SEMANTIC_COLOR: - tab[i] = 2 + cols; cols++; break; case TGSI_SEMANTIC_PSIZE: psize = TRUE; - tab[i] = 15; break; case TGSI_SEMANTIC_FOG: fog = TRUE; /* Fall through */ case TGSI_SEMANTIC_GENERIC: - tab[i] = 6 + texs; texs++; break; default: - debug_printf("r300: Unknown vertex input %d\n", - info->input_semantic_name[i]); + debug_printf("r300: Unknown vertex output %d\n", + info->output_semantic_name[i]); break; } } } + /* XXX magic */ + assert(texs <= 8); + /* Do the actual vertex_info setup. * * vertex_info has four uints of hardware-specific data in it. @@ -140,21 +151,32 @@ static void r300_vs_tab_routes(struct r300_context* r300, vinfo->hwfmt[2] |= (R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i); } - for (i = 0; i < texs; i++) { + /* Init i right here, increment it if fog is enabled. + * This gets around a double-increment problem. */ + i = 0; + + if (fog) { + i++; draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, - draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i)); + draw_find_vs_output(r300->draw, TGSI_SEMANTIC_FOG, 0)); vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i); vinfo->hwfmt[3] |= (4 << (3 * i)); } - if (fog) { - i++; + for (i; i < texs; i++) { draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, - draw_find_vs_output(r300->draw, TGSI_SEMANTIC_FOG, 0)); + draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i)); vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i); vinfo->hwfmt[3] |= (4 << (3 * i)); } + /* Handle the case where the vertex shader will be generating some of + * the attribs based on its inputs. */ + if (r300screen->caps->has_tcl && + info->num_inputs < info->num_outputs) { + vinfo->num_attribs = info->num_inputs; + } + draw_compute_vertex_size(vinfo); } @@ -162,26 +184,40 @@ static void r300_vs_tab_routes(struct r300_context* r300, static void r300_vertex_psc(struct r300_context* r300, struct r300_vertex_format* vformat) { + struct r300_screen* r300screen = r300_screen(r300->context.screen); struct vertex_info* vinfo = &vformat->vinfo; int* tab = vformat->vs_tab; uint32_t temp; - int i; + int i, attrib_count; - debug_printf("r300: attrib count: %d\n", vinfo->num_attribs); - for (i = 0; i < vinfo->num_attribs; i++) { - debug_printf("r300: attrib: offset %d, interp %d, size %d," - " tab %d\n", vinfo->attrib[i].src_index, - vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, - tab[i]); + /* Vertex shaders have no semantics on their inputs, + * so PSC should just route stuff based on their info, + * and not on attrib information. */ + if (r300screen->caps->has_tcl) { + attrib_count = r300->vs->info.num_inputs; + debug_printf("r300: routing %d attribs in psc for vs\n", + attrib_count); + } else { + attrib_count = vinfo->num_attribs; + debug_printf("r300: attrib count: %d\n", attrib_count); + for (i = 0; i < attrib_count; i++) { + debug_printf("r300: attrib: offset %d, interp %d, size %d," + " tab %d\n", vinfo->attrib[i].src_index, + vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, + tab[i]); + } } - for (i = 0; i < vinfo->num_attribs; i++) { + for (i = 0; i < attrib_count; i++) { /* Make sure we have a proper destination for our attribute */ assert(tab[i] != -1); /* Add the attribute to the PSC table. */ - temp = translate_vertex_data_type(vinfo->attrib[i].emit) | - (tab[i] << R300_DST_VEC_LOC_SHIFT); + temp = r300screen->caps->has_tcl ? + R300_DATA_TYPE_FLOAT_4 : + translate_vertex_data_type(vinfo->attrib[i].emit); + temp |= tab[i] << R300_DST_VEC_LOC_SHIFT; + if (i & 1) { vformat->vap_prog_stream_cntl[i >> 1] &= 0x0000ffff; vformat->vap_prog_stream_cntl[i >> 1] |= temp << 16; @@ -206,7 +242,6 @@ static void r300_vertex_psc(struct r300_context* r300, /* Update the vertex format. */ static void r300_update_vertex_format(struct r300_context* r300) { - struct r300_screen* r300screen = r300_screen(r300->context.screen); struct r300_vertex_format vformat; int i; diff --git a/src/gallium/drivers/r300/r300_state_invariant.c b/src/gallium/drivers/r300/r300_state_invariant.c index 8bd9b41bd74..7d822fec483 100644 --- a/src/gallium/drivers/r300/r300_state_invariant.c +++ b/src/gallium/drivers/r300/r300_state_invariant.c @@ -34,42 +34,31 @@ void r300_emit_invariant_state(struct r300_context* r300) struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps; CS_LOCALS(r300); - BEGIN_CS(30 + (caps->has_tcl ? 2: 0)); + BEGIN_CS(24 + (caps->has_tcl ? 2: 0)); /*** Graphics Backend (GB) ***/ /* Various GB enables */ - OUT_CS_REG(R300_GB_ENABLE, 0x0); - /* Subpixel multisampling for AA */ - OUT_CS_REG(R300_GB_MSPOS0, 0x66666666); - OUT_CS_REG(R300_GB_MSPOS1, 0x66666666); - /* GB tile config and pipe setup */ - OUT_CS_REG(R300_GB_TILE_CONFIG, R300_GB_TILE_DISABLE | - r300_translate_gb_pipes(caps->num_frag_pipes)); + OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE | + R300_GB_LINE_STUFF_ENABLE | + R300_GB_TRIANGLE_STUFF_ENABLE); + /* Subpixel multisampling for AA + * These are commented out because glisse's CS checker doesn't like them. + * I presume these will be re-enabled later. + * OUT_CS_REG(R300_GB_MSPOS0, 0x66666666); + * OUT_CS_REG(R300_GB_MSPOS1, 0x6666666); + */ /* Source of fog depth */ OUT_CS_REG(R300_GB_SELECT, R300_GB_FOG_SELECT_1_1_W); /* AA enable */ OUT_CS_REG(R300_GB_AA_CONFIG, 0x0); - /*** Geometry Assembly (GA) ***/ - /* GA errata fixes. */ - if (caps->is_r500) { - OUT_CS_REG(R300_GA_ENHANCE, - R300_GA_ENHANCE_DEADLOCK_CNTL_PREVENT_TCL | - R300_GA_ENHANCE_FASTSYNC_CNTL_ENABLE | - R500_GA_ENHANCE_REG_READWRITE_ENABLE | - R500_GA_ENHANCE_REG_NOSTALL_ENABLE); - } else { - OUT_CS_REG(R300_GA_ENHANCE, - R300_GA_ENHANCE_DEADLOCK_CNTL_PREVENT_TCL | - R300_GA_ENHANCE_FASTSYNC_CNTL_ENABLE); - } - /*** Fog (FG) ***/ OUT_CS_REG(R300_FG_FOG_BLEND, 0x0); OUT_CS_REG(R300_FG_FOG_COLOR_R, 0x0); OUT_CS_REG(R300_FG_FOG_COLOR_G, 0x0); OUT_CS_REG(R300_FG_FOG_COLOR_B, 0x0); OUT_CS_REG(R300_FG_DEPTH_SRC, 0x0); + OUT_CS_REG(R300_US_W_FMT, 0x0); /*** VAP ***/ /* Max and min vertex index clamp. */ @@ -86,7 +75,7 @@ void r300_emit_invariant_state(struct r300_context* r300) END_CS; /* XXX unsorted stuff from surface_fill */ - BEGIN_CS(79 + (caps->has_tcl ? 7 : 0)); + BEGIN_CS(64 + (caps->has_tcl ? 5 : 0) + (caps->is_r500 ? 4 : 0)); /* Flush PVS. */ OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0); @@ -94,30 +83,26 @@ void r300_emit_invariant_state(struct r300_context* r300) R300_VPORT_X_OFFSET_ENA | R300_VPORT_Y_SCALE_ENA | R300_VPORT_Y_OFFSET_ENA | R300_VPORT_Z_SCALE_ENA | R300_VPORT_Z_OFFSET_ENA | R300_VTX_W0_FMT); - /* XXX endian */ if (caps->has_tcl) { - OUT_CS_REG(R300_VAP_CNTL_STATUS, R300_VC_NO_SWAP); - OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE | - R300_PS_UCP_MODE_CLIP_AS_TRIFAN); OUT_CS_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4); OUT_CS_32F(1.0); OUT_CS_32F(1.0); OUT_CS_32F(1.0); OUT_CS_32F(1.0); - } else { - OUT_CS_REG(R300_VAP_CNTL_STATUS, R300_VC_NO_SWAP | - R300_VAP_TCL_BYPASS); } /* XXX point tex stuffing */ OUT_CS_REG_SEQ(R300_GA_POINT_S0, 1); OUT_CS_32F(0.0); OUT_CS_REG_SEQ(R300_GA_POINT_S1, 1); OUT_CS_32F(1.0); + /* XXX line tex stuffing */ + OUT_CS_REG_SEQ(R300_GA_LINE_S0, 1); + OUT_CS_32F(0.0); + OUT_CS_REG_SEQ(R300_GA_LINE_S1, 1); + OUT_CS_32F(1.0); OUT_CS_REG(R300_GA_TRIANGLE_STIPPLE, 0x5 | (0x5 << R300_GA_TRIANGLE_STIPPLE_Y_SHIFT_SHIFT)); /* XXX this big chunk should be refactored into rs_state */ - OUT_CS_REG(R300_GA_LINE_S0, 0x00000000); - OUT_CS_REG(R300_GA_LINE_S1, 0x3F800000); OUT_CS_REG(R300_GA_SOLID_RG, 0x00000000); OUT_CS_REG(R300_GA_SOLID_BA, 0x00000000); OUT_CS_REG(R300_GA_POLY_MODE, 0x00000000); @@ -133,8 +118,10 @@ void r300_emit_invariant_state(struct r300_context* r300) OUT_CS_REG(R300_RB3D_CCTL, 0x00000000); OUT_CS_REG(RB3D_COLOR_CHANNEL_MASK, 0x0000000F); OUT_CS_REG(R300_RB3D_AARESOLVE_CTL, 0x00000000); - OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x00000000); - OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFFFFFFFF); + if (caps->is_r500) { + OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x00000000); + OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFFFFFFFF); + } OUT_CS_REG(R300_ZB_FORMAT, 0x00000002); OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT, 0x00000003); OUT_CS_REG(R300_ZB_BW_CNTL, 0x00000000); @@ -144,17 +131,9 @@ void r300_emit_invariant_state(struct r300_context* r300) OUT_CS_REG(R300_VAP_VTX_STATE_CNTL, 0x1); OUT_CS_REG(R300_VAP_VSM_VTX_ASSM, 0x405); OUT_CS_REG(R300_SE_VTE_CNTL, 0x0000043F); - /* Vertex size. */ - OUT_CS_REG(R300_VAP_VTX_SIZE, 0x8); /* XXX */ OUT_CS_REG(R300_SC_CLIP_RULE, 0xaaaa); - OUT_CS_REG_SEQ(R300_US_OUT_FMT_0, 4); - OUT_CS(R300_C0_SEL_B | R300_C1_SEL_G | R300_C2_SEL_R | R300_C3_SEL_A); - OUT_CS(R300_US_OUT_FMT_UNUSED); - OUT_CS(R300_US_OUT_FMT_UNUSED); - OUT_CS(R300_US_OUT_FMT_UNUSED); - OUT_CS_REG(R300_US_W_FMT, R300_W_FMT_W0); END_CS; } diff --git a/src/gallium/drivers/r300/r300_state_shader.c b/src/gallium/drivers/r300/r300_state_shader.c deleted file mode 100644 index 1b02239ee76..00000000000 --- a/src/gallium/drivers/r300/r300_state_shader.c +++ /dev/null @@ -1,649 +0,0 @@ -/* - * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#include "r300_state_shader.h" - -static void r300_copy_passthrough_shader(struct r300_fragment_shader* fs) -{ - struct r300_fragment_shader* pt = &r300_passthrough_fragment_shader; - fs->shader.stack_size = pt->shader.stack_size; - fs->alu_instruction_count = pt->alu_instruction_count; - fs->tex_instruction_count = pt->tex_instruction_count; - fs->indirections = pt->indirections; - fs->instructions[0] = pt->instructions[0]; -} - -static void r500_copy_passthrough_shader(struct r500_fragment_shader* fs) -{ - struct r500_fragment_shader* pt = &r500_passthrough_fragment_shader; - fs->shader.stack_size = pt->shader.stack_size; - fs->instruction_count = pt->instruction_count; - fs->instructions[0] = pt->instructions[0]; -} - -static void r300_fs_declare(struct r300_fs_asm* assembler, - struct tgsi_full_declaration* decl) -{ - switch (decl->Declaration.File) { - case TGSI_FILE_INPUT: - switch (decl->Semantic.SemanticName) { - case TGSI_SEMANTIC_COLOR: - assembler->color_count++; - break; - case TGSI_SEMANTIC_GENERIC: - assembler->tex_count++; - break; - default: - debug_printf("r300: fs: Bad semantic declaration %d\n", - decl->Semantic.SemanticName); - break; - } - break; - case TGSI_FILE_OUTPUT: - case TGSI_FILE_CONSTANT: - break; - case TGSI_FILE_TEMPORARY: - assembler->temp_count++; - break; - default: - debug_printf("r300: fs: Bad file %d\n", decl->Declaration.File); - break; - } - - assembler->temp_offset = assembler->color_count + assembler->tex_count; -} - -static INLINE unsigned r300_fs_src(struct r300_fs_asm* assembler, - struct tgsi_src_register* src) -{ - switch (src->File) { - case TGSI_FILE_NULL: - return 0; - case TGSI_FILE_INPUT: - /* XXX may be wrong */ - return src->Index; - break; - case TGSI_FILE_TEMPORARY: - return src->Index + assembler->temp_offset; - break; - case TGSI_FILE_IMMEDIATE: - return (src->Index + assembler->imm_offset) | (1 << 8); - break; - case TGSI_FILE_CONSTANT: - /* XXX magic */ - return src->Index | (1 << 8); - break; - default: - debug_printf("r300: fs: Unimplemented src %d\n", src->File); - break; - } - return 0; -} - -static INLINE unsigned r300_fs_dst(struct r300_fs_asm* assembler, - struct tgsi_dst_register* dst) -{ - switch (dst->File) { - case TGSI_FILE_NULL: - /* This happens during KIL instructions. */ - return 0; - break; - case TGSI_FILE_OUTPUT: - return 0; - break; - case TGSI_FILE_TEMPORARY: - return dst->Index + assembler->temp_offset; - break; - default: - debug_printf("r300: fs: Unimplemented dst %d\n", dst->File); - break; - } - return 0; -} - -static INLINE unsigned r500_fix_swiz(unsigned s) -{ - /* For historical reasons, the swizzle values x, y, z, w, and 0 are - * equivalent to the actual machine code, but 1 is not. Thus, we just - * adjust it a bit... */ - if (s == TGSI_EXTSWIZZLE_ONE) { - return R500_SWIZZLE_ONE; - } else { - return s; - } -} - -static uint32_t r500_rgba_swiz(struct tgsi_full_src_register* reg) -{ - if (reg->SrcRegister.Extended) { - return r500_fix_swiz(reg->SrcRegisterExtSwz.ExtSwizzleX) | - (r500_fix_swiz(reg->SrcRegisterExtSwz.ExtSwizzleY) << 3) | - (r500_fix_swiz(reg->SrcRegisterExtSwz.ExtSwizzleZ) << 6) | - (r500_fix_swiz(reg->SrcRegisterExtSwz.ExtSwizzleW) << 9); - } else { - return reg->SrcRegister.SwizzleX | - (reg->SrcRegister.SwizzleY << 3) | - (reg->SrcRegister.SwizzleZ << 6) | - (reg->SrcRegister.SwizzleW << 9); - } -} - -static uint32_t r500_strq_swiz(struct tgsi_full_src_register* reg) -{ - return reg->SrcRegister.SwizzleX | - (reg->SrcRegister.SwizzleY << 2) | - (reg->SrcRegister.SwizzleZ << 4) | - (reg->SrcRegister.SwizzleW << 6); -} - -static INLINE uint32_t r500_rgb_swiz(struct tgsi_full_src_register* reg) -{ - /* Only the first 9 bits... */ - return (r500_rgba_swiz(reg) & 0x1ff) | - (reg->SrcRegister.Negate ? (1 << 9) : 0) | - (reg->SrcRegisterExtMod.Absolute ? (1 << 10) : 0); -} - -static INLINE uint32_t r500_alpha_swiz(struct tgsi_full_src_register* reg) -{ - /* Only the last 3 bits... */ - return (r500_rgba_swiz(reg) >> 9) | - (reg->SrcRegister.Negate ? (1 << 9) : 0) | - (reg->SrcRegisterExtMod.Absolute ? (1 << 10) : 0); -} - -static INLINE uint32_t r300_rgb_op(unsigned op) -{ - switch (op) { - case TGSI_OPCODE_MOV: - return R300_ALU_OUTC_CMP; - default: - return 0; - } -} - -static INLINE uint32_t r300_alpha_op(unsigned op) -{ - switch (op) { - case TGSI_OPCODE_MOV: - return R300_ALU_OUTA_CMP; - default: - return 0; - } -} - -static INLINE uint32_t r500_rgba_op(unsigned op) -{ - switch (op) { - case TGSI_OPCODE_EX2: - case TGSI_OPCODE_LG2: - case TGSI_OPCODE_RCP: - case TGSI_OPCODE_RSQ: - return R500_ALU_RGBA_OP_SOP; - case TGSI_OPCODE_FRC: - return R500_ALU_RGBA_OP_FRC; - case TGSI_OPCODE_DP3: - return R500_ALU_RGBA_OP_DP3; - case TGSI_OPCODE_DP4: - case TGSI_OPCODE_DPH: - return R500_ALU_RGBA_OP_DP4; - case TGSI_OPCODE_ABS: - case TGSI_OPCODE_CMP: - case TGSI_OPCODE_MOV: - case TGSI_OPCODE_SWZ: - return R500_ALU_RGBA_OP_CMP; - case TGSI_OPCODE_ADD: - case TGSI_OPCODE_MAD: - case TGSI_OPCODE_MUL: - case TGSI_OPCODE_SUB: - return R500_ALU_RGBA_OP_MAD; - default: - return 0; - } -} - -static INLINE uint32_t r500_alpha_op(unsigned op) -{ - switch (op) { - case TGSI_OPCODE_EX2: - return R500_ALPHA_OP_EX2; - case TGSI_OPCODE_LG2: - return R500_ALPHA_OP_LN2; - case TGSI_OPCODE_RCP: - return R500_ALPHA_OP_RCP; - case TGSI_OPCODE_RSQ: - return R500_ALPHA_OP_RSQ; - case TGSI_OPCODE_FRC: - return R500_ALPHA_OP_FRC; - case TGSI_OPCODE_DP3: - case TGSI_OPCODE_DP4: - case TGSI_OPCODE_DPH: - return R500_ALPHA_OP_DP; - case TGSI_OPCODE_ABS: - case TGSI_OPCODE_CMP: - case TGSI_OPCODE_MOV: - case TGSI_OPCODE_SWZ: - return R500_ALPHA_OP_CMP; - case TGSI_OPCODE_ADD: - case TGSI_OPCODE_MAD: - case TGSI_OPCODE_MUL: - case TGSI_OPCODE_SUB: - return R500_ALPHA_OP_MAD; - default: - return 0; - } -} - -static INLINE uint32_t r500_tex_op(unsigned op) -{ - switch (op) { - case TGSI_OPCODE_KIL: - return R500_TEX_INST_TEXKILL; - case TGSI_OPCODE_TEX: - return R500_TEX_INST_LD; - case TGSI_OPCODE_TXB: - return R500_TEX_INST_LODBIAS; - case TGSI_OPCODE_TXP: - return R500_TEX_INST_PROJ; - default: - return 0; - } -} - -static INLINE void r300_emit_maths(struct r300_fragment_shader* fs, - struct r300_fs_asm* assembler, - struct tgsi_full_src_register* src, - struct tgsi_full_dst_register* dst, - unsigned op, - unsigned count) -{ - int i = fs->alu_instruction_count; - - fs->instructions[i].alu_rgb_inst = R300_RGB_SWIZA(R300_ALU_ARGC_SRC0C_XYZ) | - R300_RGB_SWIZB(R300_ALU_ARGC_SRC0C_XYZ) | - R300_RGB_SWIZC(R300_ALU_ARGC_ZERO) | - r300_rgb_op(op); - fs->instructions[i].alu_rgb_addr = R300_RGB_ADDR0(0) | R300_RGB_ADDR1(0) | - R300_RGB_ADDR2(0) | R300_ALU_DSTC_OUTPUT_XYZ; - fs->instructions[i].alu_alpha_inst = R300_ALPHA_SWIZA(R300_ALU_ARGA_SRC0A) | - R300_ALPHA_SWIZB(R300_ALU_ARGA_SRC0A) | - R300_ALPHA_SWIZC(R300_ALU_ARGA_ZERO) | - r300_alpha_op(op); - fs->instructions[i].alu_alpha_addr = R300_ALPHA_ADDR0(0) | - R300_ALPHA_ADDR1(0) | R300_ALPHA_ADDR2(0) | R300_ALU_DSTA_OUTPUT; - - fs->alu_instruction_count++; -} - -/* Setup an ALU operation. */ -static INLINE void r500_emit_alu(struct r500_fragment_shader* fs, - struct r300_fs_asm* assembler, - struct tgsi_full_dst_register* dst) -{ - int i = fs->instruction_count; - - if (dst->DstRegister.File == TGSI_FILE_OUTPUT) { - fs->instructions[i].inst0 = R500_INST_TYPE_OUT | - R500_ALU_OMASK(dst->DstRegister.WriteMask); - } else { - fs->instructions[i].inst0 = R500_INST_TYPE_ALU | - R500_ALU_WMASK(dst->DstRegister.WriteMask); - } - - fs->instructions[i].inst0 |= R500_INST_TEX_SEM_WAIT; - - fs->instructions[i].inst4 = - R500_ALPHA_ADDRD(r300_fs_dst(assembler, &dst->DstRegister)); - fs->instructions[i].inst5 = - R500_ALU_RGBA_ADDRD(r300_fs_dst(assembler, &dst->DstRegister)); -} - -static INLINE void r500_emit_maths(struct r500_fragment_shader* fs, - struct r300_fs_asm* assembler, - struct tgsi_full_src_register* src, - struct tgsi_full_dst_register* dst, - unsigned op, - unsigned count) -{ - int i = fs->instruction_count; - - r500_emit_alu(fs, assembler, dst); - - switch (count) { - case 3: - fs->instructions[i].inst1 = - R500_RGB_ADDR2(r300_fs_src(assembler, &src[2].SrcRegister)); - fs->instructions[i].inst2 = - R500_ALPHA_ADDR2(r300_fs_src(assembler, &src[2].SrcRegister)); - fs->instructions[i].inst5 |= - R500_ALU_RGBA_SEL_C_SRC2 | - R500_SWIZ_RGBA_C(r500_rgb_swiz(&src[2])) | - R500_ALU_RGBA_ALPHA_SEL_C_SRC2 | - R500_SWIZ_ALPHA_C(r500_alpha_swiz(&src[2])); - case 2: - fs->instructions[i].inst1 |= - R500_RGB_ADDR1(r300_fs_src(assembler, &src[1].SrcRegister)); - fs->instructions[i].inst2 |= - R500_ALPHA_ADDR1(r300_fs_src(assembler, &src[1].SrcRegister)); - fs->instructions[i].inst3 = - R500_ALU_RGB_SEL_B_SRC1 | - R500_SWIZ_RGB_B(r500_rgb_swiz(&src[1])); - fs->instructions[i].inst4 |= - R500_SWIZ_ALPHA_B(r500_alpha_swiz(&src[1])) | - R500_ALPHA_SEL_B_SRC1; - case 1: - case 0: - default: - fs->instructions[i].inst1 |= - R500_RGB_ADDR0(r300_fs_src(assembler, &src[0].SrcRegister)); - fs->instructions[i].inst2 |= - R500_ALPHA_ADDR0(r300_fs_src(assembler, &src[0].SrcRegister)); - fs->instructions[i].inst3 |= - R500_ALU_RGB_SEL_A_SRC0 | - R500_SWIZ_RGB_A(r500_rgb_swiz(&src[0])); - fs->instructions[i].inst4 |= - R500_SWIZ_ALPHA_A(r500_alpha_swiz(&src[0])) | - R500_ALPHA_SEL_A_SRC0; - break; - } - - fs->instructions[i].inst4 |= r500_alpha_op(op); - fs->instructions[i].inst5 |= r500_rgba_op(op); - - fs->instruction_count++; -} - -static INLINE void r500_emit_tex(struct r500_fragment_shader* fs, - struct r300_fs_asm* assembler, - struct tgsi_full_src_register* src, - struct tgsi_full_dst_register* dst, - uint32_t op) -{ - int i = fs->instruction_count; - - fs->instructions[i].inst0 = R500_INST_TYPE_TEX | - R500_TEX_WMASK(dst->DstRegister.WriteMask) | - R500_INST_TEX_SEM_WAIT; - fs->instructions[i].inst1 = R500_TEX_ID(0) | - R500_TEX_SEM_ACQUIRE | //R500_TEX_IGNORE_UNCOVERED | - r500_tex_op(op); - fs->instructions[i].inst2 = - R500_TEX_SRC_ADDR(r300_fs_src(assembler, &src->SrcRegister)) | - R500_SWIZ_TEX_STRQ(r500_strq_swiz(src)) | - R500_TEX_DST_ADDR(r300_fs_dst(assembler, &dst->DstRegister)) | - R500_TEX_DST_R_SWIZ_R | R500_TEX_DST_G_SWIZ_G | - R500_TEX_DST_B_SWIZ_B | R500_TEX_DST_A_SWIZ_A; - - if (dst->DstRegister.File == TGSI_FILE_OUTPUT) { - fs->instructions[i].inst2 |= - R500_TEX_DST_ADDR(assembler->temp_count + - assembler->temp_offset); - - fs->instruction_count++; - - /* Setup and emit a MOV. */ - src[0].SrcRegister.Index = assembler->temp_count; - src[0].SrcRegister.File = TGSI_FILE_TEMPORARY; - - src[1] = src[0]; - src[2] = r500_constant_zero; - r500_emit_maths(fs, assembler, src, dst, TGSI_OPCODE_MOV, 3); - } else { - fs->instruction_count++; - } -} - -static void r300_fs_instruction(struct r300_fragment_shader* fs, - struct r300_fs_asm* assembler, - struct tgsi_full_instruction* inst) -{ - switch (inst->Instruction.Opcode) { - case TGSI_OPCODE_MOV: - /* src0 -> src1 and src2 forced to zero */ - inst->FullSrcRegisters[1] = inst->FullSrcRegisters[0]; - inst->FullSrcRegisters[2] = r500_constant_zero; - r300_emit_maths(fs, assembler, inst->FullSrcRegisters, - &inst->FullDstRegisters[0], inst->Instruction.Opcode, 3); - break; - case TGSI_OPCODE_END: - break; - default: - debug_printf("r300: fs: Bad opcode %d\n", - inst->Instruction.Opcode); - break; - } -} - -static void r500_fs_instruction(struct r500_fragment_shader* fs, - struct r300_fs_asm* assembler, - struct tgsi_full_instruction* inst) -{ - /* Switch between opcodes. When possible, prefer using the official - * AMD/ATI names for opcodes, please, as it facilitates using the - * documentation. */ - switch (inst->Instruction.Opcode) { - /* The simple scalar ops. */ - case TGSI_OPCODE_EX2: - case TGSI_OPCODE_LG2: - case TGSI_OPCODE_RCP: - case TGSI_OPCODE_RSQ: - /* Copy red swizzle to alpha for src0 */ - inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtSwizzleW = - inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtSwizzleX; - inst->FullSrcRegisters[0].SrcRegister.SwizzleW = - inst->FullSrcRegisters[0].SrcRegister.SwizzleX; - /* Fall through */ - case TGSI_OPCODE_FRC: - r500_emit_maths(fs, assembler, inst->FullSrcRegisters, - &inst->FullDstRegisters[0], inst->Instruction.Opcode, 1); - break; - - /* The dot products. */ - case TGSI_OPCODE_DPH: - /* Set alpha swizzle to one for src0 */ - if (!inst->FullSrcRegisters[0].SrcRegister.Extended) { - inst->FullSrcRegisters[0].SrcRegister.Extended = TRUE; - inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtSwizzleX = - inst->FullSrcRegisters[0].SrcRegister.SwizzleX; - inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtSwizzleY = - inst->FullSrcRegisters[0].SrcRegister.SwizzleY; - inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtSwizzleZ = - inst->FullSrcRegisters[0].SrcRegister.SwizzleZ; - } - inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtSwizzleW = - TGSI_EXTSWIZZLE_ONE; - /* Fall through */ - case TGSI_OPCODE_DP3: - case TGSI_OPCODE_DP4: - r500_emit_maths(fs, assembler, inst->FullSrcRegisters, - &inst->FullDstRegisters[0], inst->Instruction.Opcode, 2); - break; - - /* Simple three-source operations. */ - case TGSI_OPCODE_CMP: - /* Swap src0 and src2 */ - inst->FullSrcRegisters[3] = inst->FullSrcRegisters[2]; - inst->FullSrcRegisters[2] = inst->FullSrcRegisters[0]; - inst->FullSrcRegisters[0] = inst->FullSrcRegisters[3]; - r500_emit_maths(fs, assembler, inst->FullSrcRegisters, - &inst->FullDstRegisters[0], inst->Instruction.Opcode, 3); - break; - - /* The MAD variants. */ - case TGSI_OPCODE_SUB: - /* Just like ADD, but flip the negation on src1 first */ - inst->FullSrcRegisters[1].SrcRegister.Negate = - !inst->FullSrcRegisters[1].SrcRegister.Negate; - /* Fall through */ - case TGSI_OPCODE_ADD: - /* Force src0 to one, move all registers over */ - inst->FullSrcRegisters[2] = inst->FullSrcRegisters[1]; - inst->FullSrcRegisters[1] = inst->FullSrcRegisters[0]; - inst->FullSrcRegisters[0] = r500_constant_one; - r500_emit_maths(fs, assembler, inst->FullSrcRegisters, - &inst->FullDstRegisters[0], inst->Instruction.Opcode, 3); - break; - case TGSI_OPCODE_MUL: - /* Force our src2 to zero */ - inst->FullSrcRegisters[2] = r500_constant_zero; - r500_emit_maths(fs, assembler, inst->FullSrcRegisters, - &inst->FullDstRegisters[0], inst->Instruction.Opcode, 3); - break; - case TGSI_OPCODE_MAD: - r500_emit_maths(fs, assembler, inst->FullSrcRegisters, - &inst->FullDstRegisters[0], inst->Instruction.Opcode, 3); - break; - - /* The MOV variants. */ - case TGSI_OPCODE_ABS: - /* Set absolute value modifiers. */ - inst->FullSrcRegisters[0].SrcRegisterExtMod.Absolute = TRUE; - /* Fall through */ - case TGSI_OPCODE_MOV: - case TGSI_OPCODE_SWZ: - /* src0 -> src1 and src2 forced to zero */ - inst->FullSrcRegisters[1] = inst->FullSrcRegisters[0]; - inst->FullSrcRegisters[2] = r500_constant_zero; - r500_emit_maths(fs, assembler, inst->FullSrcRegisters, - &inst->FullDstRegisters[0], inst->Instruction.Opcode, 3); - break; - - /* The texture instruction set. */ - case TGSI_OPCODE_KIL: - case TGSI_OPCODE_TEX: - case TGSI_OPCODE_TXB: - case TGSI_OPCODE_TXP: - r500_emit_tex(fs, assembler, &inst->FullSrcRegisters[0], - &inst->FullDstRegisters[0], inst->Instruction.Opcode); - break; - - /* This is the end. My only friend, the end. */ - case TGSI_OPCODE_END: - break; - default: - debug_printf("r300: fs: Bad opcode %d\n", - inst->Instruction.Opcode); - break; - } - - /* Clamp, if saturation flags are set. */ - if (inst->Instruction.Saturate == TGSI_SAT_ZERO_ONE) { - fs->instructions[fs->instruction_count - 1].inst0 |= - R500_INST_RGB_CLAMP | R500_INST_ALPHA_CLAMP; - } -} - -static void r300_fs_finalize(struct r3xx_fragment_shader* fs, - struct r300_fs_asm* assembler) -{ - fs->stack_size = assembler->temp_count + assembler->temp_offset; -} - -static void r500_fs_finalize(struct r500_fragment_shader* fs, - struct r300_fs_asm* assembler) -{ - /* XXX should this just go with OPCODE_END? */ - fs->instructions[fs->instruction_count - 1].inst0 |= - R500_INST_LAST; -} - -void r300_translate_fragment_shader(struct r300_context* r300, - struct r3xx_fragment_shader* fs) -{ - struct tgsi_parse_context parser; - int i; - boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500; - struct r300_constant_buffer* consts = - &r300->shader_constants[PIPE_SHADER_FRAGMENT]; - - struct r300_fs_asm* assembler = CALLOC_STRUCT(r300_fs_asm); - if (assembler == NULL) { - return; - } - /* Setup starting offset for immediates. */ - assembler->imm_offset = consts->user_count; - - /* Make sure we start at the beginning of the shader. */ - if (is_r500) { - ((struct r500_fragment_shader*)fs)->instruction_count = 0; - } - - tgsi_parse_init(&parser, fs->state.tokens); - - while (!tgsi_parse_end_of_tokens(&parser)) { - tgsi_parse_token(&parser); - - /* This is seriously the lamest way to create fragment programs ever. - * I blame TGSI. */ - switch (parser.FullToken.Token.Type) { - case TGSI_TOKEN_TYPE_DECLARATION: - /* Allocated registers sitting at the beginning - * of the program. */ - r300_fs_declare(assembler, &parser.FullToken.FullDeclaration); - break; - case TGSI_TOKEN_TYPE_IMMEDIATE: - debug_printf("r300: Emitting immediate to constant buffer, " - "position %d\n", - assembler->imm_offset + assembler->imm_count); - /* I am not amused by the length of these. */ - for (i = 0; i < 4; i++) { - consts->constants[assembler->imm_offset + - assembler->imm_count][i] = - parser.FullToken.FullImmediate.u.ImmediateFloat32[i] - .Float; - } - assembler->imm_count++; - break; - case TGSI_TOKEN_TYPE_INSTRUCTION: - if (is_r500) { - r500_fs_instruction((struct r500_fragment_shader*)fs, - assembler, &parser.FullToken.FullInstruction); - } else { - r300_fs_instruction((struct r300_fragment_shader*)fs, - assembler, &parser.FullToken.FullInstruction); - } - break; - } - } - - debug_printf("r300: fs: %d texs and %d colors, first free reg is %d\n", - assembler->tex_count, assembler->color_count, - assembler->tex_count + assembler->color_count); - - consts->count = consts->user_count + assembler->imm_count; - debug_printf("r300: fs: %d total constants, " - "%d from user and %d from immediates\n", consts->count, - consts->user_count, assembler->imm_count); - r300_fs_finalize(fs, assembler); - if (is_r500) { - r500_fs_finalize((struct r500_fragment_shader*)fs, assembler); - } - - tgsi_dump(fs->state.tokens); - /* XXX finish r300 dumper too */ - if (is_r500) { - r500_fs_dump((struct r500_fragment_shader*)fs); - } - - tgsi_parse_free(&parser); - FREE(assembler); -} diff --git a/src/gallium/drivers/r300/r300_state_shader.h b/src/gallium/drivers/r300/r300_state_shader.h deleted file mode 100644 index 185fdd90f0c..00000000000 --- a/src/gallium/drivers/r300/r300_state_shader.h +++ /dev/null @@ -1,230 +0,0 @@ -/* - * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#ifndef R300_STATE_SHADER_H -#define R300_STATE_SHADER_H - -#include "tgsi/tgsi_parse.h" - -#include "r300_context.h" -#include "r300_debug.h" -#include "r300_reg.h" -#include "r300_screen.h" - -/* XXX this all should find its way back to r300_reg */ -/* Swizzle tools */ -#define R500_SWIZZLE_ZERO 4 -#define R500_SWIZZLE_HALF 5 -#define R500_SWIZZLE_ONE 6 -#define R500_SWIZ_RGB_ZERO ((4 << 0) | (4 << 3) | (4 << 6)) -#define R500_SWIZ_RGB_ONE ((6 << 0) | (6 << 3) | (6 << 6)) -#define R500_SWIZ_RGB_RGB ((0 << 0) | (1 << 3) | (2 << 6)) -#define R500_SWIZ_MOD_NEG 1 -#define R500_SWIZ_MOD_ABS 2 -#define R500_SWIZ_MOD_NEG_ABS 3 -/* Swizzles for inst2 */ -#define R500_SWIZ_TEX_STRQ(x) ((x) << 8) -#define R500_SWIZ_TEX_RGBA(x) ((x) << 24) -/* Swizzles for inst3 */ -#define R500_SWIZ_RGB_A(x) ((x) << 2) -#define R500_SWIZ_RGB_B(x) ((x) << 15) -/* Swizzles for inst4 */ -#define R500_SWIZ_ALPHA_A(x) ((x) << 14) -#define R500_SWIZ_ALPHA_B(x) ((x) << 21) -/* Swizzle for inst5 */ -#define R500_SWIZ_RGBA_C(x) ((x) << 14) -#define R500_SWIZ_ALPHA_C(x) ((x) << 27) -/* Writemasks */ -#define R500_TEX_WMASK(x) ((x) << 11) -#define R500_ALU_WMASK(x) ((x) << 11) -#define R500_ALU_OMASK(x) ((x) << 15) - -/* TGSI constants. TGSI is like XML: If it can't solve your problems, you're - * not using enough of it. */ -static const struct tgsi_full_src_register r500_constant_zero = { - .SrcRegister.Extended = TRUE, - .SrcRegister.File = TGSI_FILE_NULL, - .SrcRegisterExtSwz.ExtSwizzleX = TGSI_EXTSWIZZLE_ZERO, - .SrcRegisterExtSwz.ExtSwizzleY = TGSI_EXTSWIZZLE_ZERO, - .SrcRegisterExtSwz.ExtSwizzleZ = TGSI_EXTSWIZZLE_ZERO, - .SrcRegisterExtSwz.ExtSwizzleW = TGSI_EXTSWIZZLE_ZERO, -}; - -static const struct tgsi_full_src_register r500_constant_one = { - .SrcRegister.Extended = TRUE, - .SrcRegister.File = TGSI_FILE_NULL, - .SrcRegisterExtSwz.ExtSwizzleX = TGSI_EXTSWIZZLE_ONE, - .SrcRegisterExtSwz.ExtSwizzleY = TGSI_EXTSWIZZLE_ONE, - .SrcRegisterExtSwz.ExtSwizzleZ = TGSI_EXTSWIZZLE_ONE, - .SrcRegisterExtSwz.ExtSwizzleW = TGSI_EXTSWIZZLE_ONE, -}; - -/* Temporary struct used to hold assembly state while putting together - * fragment programs. */ -struct r300_fs_asm { - /* Pipe context. */ - struct r300_context* r300; - /* Number of colors. */ - unsigned color_count; - /* Number of texcoords. */ - unsigned tex_count; - /* Offset for temporary registers. Inputs and temporaries have no - * distinguishing markings, so inputs start at 0 and the first usable - * temporary register is after all inputs. */ - unsigned temp_offset; - /* Number of requested temporary registers. */ - unsigned temp_count; - /* Offset for immediate constants. Neither R300 nor R500 can do four - * inline constants per source, so instead we copy immediates into the - * constant buffer. */ - unsigned imm_offset; - /* Number of immediate constants. */ - unsigned imm_count; -}; - -void r300_translate_fragment_shader(struct r300_context* r300, - struct r3xx_fragment_shader* fs); - -static struct r300_fragment_shader r300_passthrough_fragment_shader = { - /* XXX This is the emission code. TODO: decode - OUT_CS_REG(R300_US_CONFIG, 0); - OUT_CS_REG(R300_US_CODE_OFFSET, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_0, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_1, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_2, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_3, 0x400000); -*/ - .alu_instruction_count = 1, - .tex_instruction_count = 0, - .indirections = 0, - .shader.stack_size = 1, - - .instructions[0].alu_rgb_inst = R300_RGB_SWIZA(R300_ALU_ARGC_SRC0C_XYZ) | - R300_RGB_SWIZB(R300_ALU_ARGC_SRC0C_XYZ) | - R300_RGB_SWIZC(R300_ALU_ARGC_ZERO) | - R300_ALU_OUTC_CMP, - .instructions[0].alu_rgb_addr = R300_RGB_ADDR0(0) | R300_RGB_ADDR1(0) | - R300_RGB_ADDR2(0) | R300_ALU_DSTC_OUTPUT_XYZ, - .instructions[0].alu_alpha_inst = R300_ALPHA_SWIZA(R300_ALU_ARGA_SRC0A) | - R300_ALPHA_SWIZB(R300_ALU_ARGA_SRC0A) | - R300_ALPHA_SWIZC(R300_ALU_ARGA_ZERO) | - R300_ALU_OUTA_CMP, - .instructions[0].alu_alpha_addr = R300_ALPHA_ADDR0(0) | - R300_ALPHA_ADDR1(0) | R300_ALPHA_ADDR2(0) | R300_ALU_DSTA_OUTPUT, -}; - -static struct r500_fragment_shader r500_passthrough_fragment_shader = { - .shader.stack_size = 0, - .instruction_count = 1, - .instructions[0].inst0 = R500_INST_TYPE_OUT | - R500_INST_TEX_SEM_WAIT | R500_INST_LAST | - R500_INST_RGB_OMASK_RGB | R500_INST_ALPHA_OMASK | - R500_INST_RGB_CLAMP | R500_INST_ALPHA_CLAMP, - .instructions[0].inst1 = - R500_RGB_ADDR0(0) | R500_RGB_ADDR1(0) | R500_RGB_ADDR1_CONST | - R500_RGB_ADDR2(0) | R500_RGB_ADDR2_CONST, - .instructions[0].inst2 = - R500_ALPHA_ADDR0(0) | R500_ALPHA_ADDR1(0) | R500_ALPHA_ADDR1_CONST | - R500_ALPHA_ADDR2(0) | R500_ALPHA_ADDR2_CONST, - .instructions[0].inst3 = - R500_ALU_RGB_SEL_A_SRC0 | R500_ALU_RGB_R_SWIZ_A_R | - R500_ALU_RGB_G_SWIZ_A_G | R500_ALU_RGB_B_SWIZ_A_B | - R500_ALU_RGB_SEL_B_SRC0 | R500_ALU_RGB_R_SWIZ_B_R | - R500_ALU_RGB_B_SWIZ_B_G | R500_ALU_RGB_G_SWIZ_B_B, - .instructions[0].inst4 = - R500_ALPHA_OP_CMP | R500_ALPHA_SWIZ_A_A | R500_ALPHA_SWIZ_B_A, - .instructions[0].inst5 = - R500_ALU_RGBA_OP_CMP | R500_ALU_RGBA_R_SWIZ_0 | - R500_ALU_RGBA_G_SWIZ_0 | R500_ALU_RGBA_B_SWIZ_0 | - R500_ALU_RGBA_A_SWIZ_0, -}; - -static struct r300_fragment_shader r300_texture_fragment_shader = { - /* XXX This is the emission code. TODO: decode - OUT_CS_REG(R300_US_CONFIG, 0); - OUT_CS_REG(R300_US_CODE_OFFSET, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_0, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_1, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_2, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_3, 0x400000); -*/ - .alu_instruction_count = 1, - .tex_instruction_count = 0, - .indirections = 0, - .shader.stack_size = 1, - - .instructions[0].alu_rgb_inst = R300_RGB_SWIZA(R300_ALU_ARGC_SRC0C_XYZ) | - R300_RGB_SWIZB(R300_ALU_ARGC_SRC0C_XYZ) | - R300_RGB_SWIZC(R300_ALU_ARGC_ZERO) | - R300_ALU_OUTC_CMP, - .instructions[0].alu_rgb_addr = R300_RGB_ADDR0(0) | R300_RGB_ADDR1(0) | - R300_RGB_ADDR2(0) | R300_ALU_DSTC_OUTPUT_XYZ, - .instructions[0].alu_alpha_inst = R300_ALPHA_SWIZA(R300_ALU_ARGA_SRC0A) | - R300_ALPHA_SWIZB(R300_ALU_ARGA_SRC0A) | - R300_ALPHA_SWIZC(R300_ALU_ARGA_ZERO) | - R300_ALU_OUTA_CMP, - .instructions[0].alu_alpha_addr = R300_ALPHA_ADDR0(0) | - R300_ALPHA_ADDR1(0) | R300_ALPHA_ADDR2(0) | R300_ALU_DSTA_OUTPUT, -}; - -static struct r500_fragment_shader r500_texture_fragment_shader = { - .shader.stack_size = 1, - .instruction_count = 2, - .instructions[0].inst0 = R500_INST_TYPE_TEX | - R500_INST_TEX_SEM_WAIT | - R500_INST_RGB_OMASK_RGB | R500_INST_ALPHA_OMASK | - R500_INST_RGB_CLAMP | R500_INST_ALPHA_CLAMP, - .instructions[0].inst1 = R500_TEX_ID(0) | R500_TEX_INST_LD | - R500_TEX_SEM_ACQUIRE | R500_TEX_IGNORE_UNCOVERED, - .instructions[0].inst2 = R500_TEX_SRC_ADDR(0) | - R500_TEX_SRC_S_SWIZ_R | R500_TEX_SRC_T_SWIZ_G | - R500_TEX_SRC_R_SWIZ_B | R500_TEX_SRC_Q_SWIZ_A | - R500_TEX_DST_ADDR(0) | - R500_TEX_DST_R_SWIZ_R | R500_TEX_DST_G_SWIZ_G | - R500_TEX_DST_B_SWIZ_B | R500_TEX_DST_A_SWIZ_A, - .instructions[0].inst3 = 0x0, - .instructions[0].inst4 = 0x0, - .instructions[0].inst5 = 0x0, - .instructions[1].inst0 = R500_INST_TYPE_OUT | - R500_INST_TEX_SEM_WAIT | R500_INST_LAST | - R500_INST_RGB_OMASK_RGB | R500_INST_ALPHA_OMASK | - R500_INST_RGB_CLAMP | R500_INST_ALPHA_CLAMP, - .instructions[1].inst1 = - R500_RGB_ADDR0(0) | R500_RGB_ADDR1(0) | R500_RGB_ADDR1_CONST | - R500_RGB_ADDR2(0) | R500_RGB_ADDR2_CONST, - .instructions[1].inst2 = - R500_ALPHA_ADDR0(0) | R500_ALPHA_ADDR1(0) | R500_ALPHA_ADDR1_CONST | - R500_ALPHA_ADDR2(0) | R500_ALPHA_ADDR2_CONST, - .instructions[1].inst3 = - R500_ALU_RGB_SEL_A_SRC0 | R500_ALU_RGB_R_SWIZ_A_R | - R500_ALU_RGB_G_SWIZ_A_G | R500_ALU_RGB_B_SWIZ_A_B | - R500_ALU_RGB_SEL_B_SRC0 | R500_ALU_RGB_R_SWIZ_B_R | - R500_ALU_RGB_B_SWIZ_B_G | R500_ALU_RGB_G_SWIZ_B_B, - .instructions[1].inst4 = - R500_ALPHA_OP_CMP | R500_ALPHA_SWIZ_A_A | R500_ALPHA_SWIZ_B_A, - .instructions[1].inst5 = - R500_ALU_RGBA_OP_CMP | R500_ALU_RGBA_R_SWIZ_0 | - R500_ALU_RGBA_G_SWIZ_0 | R500_ALU_RGBA_B_SWIZ_0 | - R500_ALU_RGBA_A_SWIZ_0, -}; - -#endif /* R300_STATE_SHADER_H */ diff --git a/src/gallium/drivers/r300/r300_state_tcl.c b/src/gallium/drivers/r300/r300_state_tcl.c deleted file mode 100644 index d84912de48f..00000000000 --- a/src/gallium/drivers/r300/r300_state_tcl.c +++ /dev/null @@ -1,353 +0,0 @@ -/* - * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#include "r300_state_tcl.h" - -static void r300_vs_declare(struct r300_vs_asm* assembler, - struct tgsi_full_declaration* decl) -{ - switch (decl->Declaration.File) { - case TGSI_FILE_INPUT: - break; - case TGSI_FILE_OUTPUT: - switch (decl->Semantic.SemanticName) { - case TGSI_SEMANTIC_POSITION: - assembler->tab[decl->DeclarationRange.First] = 0; - break; - case TGSI_SEMANTIC_COLOR: - assembler->tab[decl->DeclarationRange.First] = - (assembler->point_size ? 1 : 0) + - assembler->out_colors; - break; - case TGSI_SEMANTIC_FOG: - case TGSI_SEMANTIC_GENERIC: - /* XXX multiple? */ - assembler->tab[decl->DeclarationRange.First] = - (assembler->point_size ? 1 : 0) + - assembler->out_colors + - assembler->out_texcoords; - break; - case TGSI_SEMANTIC_PSIZE: - assembler->tab[decl->DeclarationRange.First] = 1; - break; - default: - debug_printf("r300: vs: Bad semantic declaration %d\n", - decl->Semantic.SemanticName); - break; - } - break; - case TGSI_FILE_CONSTANT: - break; - case TGSI_FILE_TEMPORARY: - assembler->temp_count++; - break; - default: - debug_printf("r300: vs: Bad file %d\n", decl->Declaration.File); - break; - } -} - -static INLINE unsigned r300_vs_src_type(struct r300_vs_asm* assembler, - struct tgsi_src_register* src) -{ - switch (src->File) { - case TGSI_FILE_NULL: - /* Probably a zero or one swizzle */ - return R300_PVS_SRC_REG_INPUT; - break; - case TGSI_FILE_INPUT: - return R300_PVS_SRC_REG_INPUT; - break; - case TGSI_FILE_TEMPORARY: - return R300_PVS_SRC_REG_TEMPORARY; - break; - case TGSI_FILE_CONSTANT: - return R300_PVS_SRC_REG_CONSTANT; - default: - debug_printf("r300: vs: Unimplemented src type %d\n", src->File); - break; - } - return 0; -} - -static INLINE unsigned r300_vs_dst_type(struct r300_vs_asm* assembler, - struct tgsi_dst_register* dst) -{ - switch (dst->File) { - case TGSI_FILE_TEMPORARY: - return R300_PVS_DST_REG_TEMPORARY; - break; - case TGSI_FILE_OUTPUT: - return R300_PVS_DST_REG_OUT; - break; - default: - debug_printf("r300: vs: Unimplemented dst type %d\n", dst->File); - break; - } - return 0; -} - -static INLINE unsigned r300_vs_dst(struct r300_vs_asm* assembler, - struct tgsi_dst_register* dst) -{ - switch (dst->File) { - case TGSI_FILE_TEMPORARY: - return dst->Index; - break; - case TGSI_FILE_OUTPUT: - return assembler->tab[dst->Index]; - break; - default: - debug_printf("r300: vs: Unimplemented dst %d\n", dst->File); - break; - } - return 0; -} - -static uint32_t r300_vs_op(unsigned op) -{ - switch (op) { - case TGSI_OPCODE_DP3: - case TGSI_OPCODE_DP4: - return R300_VE_DOT_PRODUCT; - case TGSI_OPCODE_MUL: - return R300_VE_MULTIPLY; - case TGSI_OPCODE_ADD: - case TGSI_OPCODE_MOV: - case TGSI_OPCODE_SWZ: - return R300_VE_ADD; - case TGSI_OPCODE_MAD: - return R300_PVS_DST_MACRO_INST | R300_PVS_MACRO_OP_2CLK_MADD; - default: - break; - } - return 0; -} - -static uint32_t r300_vs_swiz(struct tgsi_full_src_register* reg) -{ - if (reg->SrcRegister.Extended) { - return reg->SrcRegisterExtSwz.ExtSwizzleX | - (reg->SrcRegisterExtSwz.ExtSwizzleY << 3) | - (reg->SrcRegisterExtSwz.ExtSwizzleZ << 6) | - (reg->SrcRegisterExtSwz.ExtSwizzleW << 9); - } else { - return reg->SrcRegister.SwizzleX | - (reg->SrcRegister.SwizzleY << 3) | - (reg->SrcRegister.SwizzleZ << 6) | - (reg->SrcRegister.SwizzleW << 9); - } -} - -static void r300_vs_emit_inst(struct r300_vertex_shader* vs, - struct r300_vs_asm* assembler, - struct tgsi_full_src_register* src, - struct tgsi_full_dst_register* dst, - unsigned op, - unsigned count) -{ - int i = vs->instruction_count; - vs->instructions[i].inst0 = R300_PVS_DST_OPCODE(r300_vs_op(op)) | - R300_PVS_DST_REG_TYPE(r300_vs_dst_type(assembler, &dst->DstRegister)) | - R300_PVS_DST_OFFSET(r300_vs_dst(assembler, &dst->DstRegister)) | - R300_PVS_DST_WE_XYZW; - switch (count) { - case 3: - vs->instructions[i].inst3 = - R300_PVS_SRC_REG_TYPE(r300_vs_src_type(assembler, - &src[2].SrcRegister)) | - R300_PVS_SRC_OFFSET(src[2].SrcRegister.Index) | - R300_PVS_SRC_SWIZZLE(r300_vs_swiz(&src[2])); - /* Fall through */ - case 2: - vs->instructions[i].inst2 = - R300_PVS_SRC_REG_TYPE(r300_vs_src_type(assembler, - &src[1].SrcRegister)) | - R300_PVS_SRC_OFFSET(src[1].SrcRegister.Index) | - R300_PVS_SRC_SWIZZLE(r300_vs_swiz(&src[1])); - /* Fall through */ - case 1: - vs->instructions[i].inst1 = - R300_PVS_SRC_REG_TYPE(r300_vs_src_type(assembler, - &src[0].SrcRegister)) | - R300_PVS_SRC_OFFSET(src[0].SrcRegister.Index) | - R300_PVS_SRC_SWIZZLE(r300_vs_swiz(&src[0])); - break; - } - vs->instruction_count++; -} - -static void r300_vs_instruction(struct r300_vertex_shader* vs, - struct r300_vs_asm* assembler, - struct tgsi_full_instruction* inst) -{ - switch (inst->Instruction.Opcode) { - case TGSI_OPCODE_ADD: - case TGSI_OPCODE_MUL: - r300_vs_emit_inst(vs, assembler, inst->FullSrcRegisters, - &inst->FullDstRegisters[0], inst->Instruction.Opcode, - 2); - break; - case TGSI_OPCODE_DP3: - /* Set alpha swizzle to zero for src0 and src1 */ - if (!inst->FullSrcRegisters[0].SrcRegister.Extended) { - inst->FullSrcRegisters[0].SrcRegister.Extended = TRUE; - inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtSwizzleX = - inst->FullSrcRegisters[0].SrcRegister.SwizzleX; - inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtSwizzleY = - inst->FullSrcRegisters[0].SrcRegister.SwizzleY; - inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtSwizzleZ = - inst->FullSrcRegisters[0].SrcRegister.SwizzleZ; - } - inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtSwizzleW = - TGSI_EXTSWIZZLE_ZERO; - if (!inst->FullSrcRegisters[1].SrcRegister.Extended) { - inst->FullSrcRegisters[1].SrcRegister.Extended = TRUE; - inst->FullSrcRegisters[1].SrcRegisterExtSwz.ExtSwizzleX = - inst->FullSrcRegisters[1].SrcRegister.SwizzleX; - inst->FullSrcRegisters[1].SrcRegisterExtSwz.ExtSwizzleY = - inst->FullSrcRegisters[1].SrcRegister.SwizzleY; - inst->FullSrcRegisters[1].SrcRegisterExtSwz.ExtSwizzleZ = - inst->FullSrcRegisters[1].SrcRegister.SwizzleZ; - } - inst->FullSrcRegisters[1].SrcRegisterExtSwz.ExtSwizzleW = - TGSI_EXTSWIZZLE_ZERO; - /* Fall through */ - case TGSI_OPCODE_DP4: - r300_vs_emit_inst(vs, assembler, inst->FullSrcRegisters, - &inst->FullDstRegisters[0], inst->Instruction.Opcode, - 2); - break; - case TGSI_OPCODE_MOV: - case TGSI_OPCODE_SWZ: - inst->FullSrcRegisters[1] = r300_constant_zero; - r300_vs_emit_inst(vs, assembler, inst->FullSrcRegisters, - &inst->FullDstRegisters[0], inst->Instruction.Opcode, - 2); - break; - case TGSI_OPCODE_MAD: - r300_vs_emit_inst(vs, assembler, inst->FullSrcRegisters, - &inst->FullDstRegisters[0], inst->Instruction.Opcode, - 3); - break; - case TGSI_OPCODE_END: - break; - default: - debug_printf("r300: vs: Bad opcode %d\n", - inst->Instruction.Opcode); - break; - } -} - -static void r300_vs_init(struct r300_vertex_shader* vs, - struct r300_vs_asm* assembler) -{ - struct tgsi_shader_info* info = &vs->info; - int i; - - for (i = 0; i < info->num_outputs; i++) { - switch (info->output_semantic_name[i]) { - case TGSI_SEMANTIC_PSIZE: - assembler->point_size = TRUE; - break; - case TGSI_SEMANTIC_COLOR: - assembler->out_colors++; - break; - case TGSI_SEMANTIC_FOG: - case TGSI_SEMANTIC_GENERIC: - assembler->out_texcoords++; - break; - } - } -} - -void r300_translate_vertex_shader(struct r300_context* r300, - struct r300_vertex_shader* vs) -{ - struct tgsi_parse_context parser; - int i; - struct r300_constant_buffer* consts = - &r300->shader_constants[PIPE_SHADER_VERTEX]; - - struct r300_vs_asm* assembler = CALLOC_STRUCT(r300_vs_asm); - if (assembler == NULL) { - return; - } - - /* Init assembler. */ - r300_vs_init(vs, assembler); - - /* Setup starting offset for immediates. */ - assembler->imm_offset = consts->user_count; - - tgsi_parse_init(&parser, vs->state.tokens); - - while (!tgsi_parse_end_of_tokens(&parser)) { - tgsi_parse_token(&parser); - - /* This is seriously the lamest way to create fragment programs ever. - * I blame TGSI. */ - switch (parser.FullToken.Token.Type) { - case TGSI_TOKEN_TYPE_DECLARATION: - /* Allocated registers sitting at the beginning - * of the program. */ - r300_vs_declare(assembler, &parser.FullToken.FullDeclaration); - break; - case TGSI_TOKEN_TYPE_IMMEDIATE: - debug_printf("r300: Emitting immediate to constant buffer, " - "position %d\n", - assembler->imm_offset + assembler->imm_count); - /* I am not amused by the length of these. */ - for (i = 0; i < 4; i++) { - consts->constants[assembler->imm_offset + - assembler->imm_count][i] = - parser.FullToken.FullImmediate.u.ImmediateFloat32[i] - .Float; - } - assembler->imm_count++; - break; - case TGSI_TOKEN_TYPE_INSTRUCTION: - r300_vs_instruction(vs, assembler, - &parser.FullToken.FullInstruction); - break; - } - } - - debug_printf("r300: vs: %d texs and %d colors, first free reg is %d\n", - assembler->tex_count, assembler->color_count, - assembler->tex_count + assembler->color_count); - - consts->count = consts->user_count + assembler->imm_count; - debug_printf("r300: vs: %d total constants, " - "%d from user and %d from immediates\n", consts->count, - consts->user_count, assembler->imm_count); - - debug_printf("r300: vs: tab: %d %d %d %d\n", assembler->tab[0], - assembler->tab[1], assembler->tab[2], assembler->tab[3]); - - tgsi_dump(vs->state.tokens); - /* XXX finish r300 vertex shader dumper */ - r300_vs_dump(vs); - - tgsi_parse_free(&parser); - FREE(assembler); -} diff --git a/src/gallium/drivers/r300/r300_state_tcl.h b/src/gallium/drivers/r300/r300_state_tcl.h deleted file mode 100644 index e2e1357d436..00000000000 --- a/src/gallium/drivers/r300/r300_state_tcl.h +++ /dev/null @@ -1,153 +0,0 @@ -/* - * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#ifndef R300_STATE_TCL_H -#define R300_STATE_TCL_H - -#include "tgsi/tgsi_parse.h" - -#include "r300_context.h" -#include "r300_debug.h" -#include "r300_reg.h" -#include "r300_screen.h" - -/* XXX get these to r300_reg */ -#define R300_PVS_DST_OPCODE(x) ((x) << 0) -# define R300_VE_DOT_PRODUCT 1 -# define R300_VE_MULTIPLY 2 -# define R300_VE_ADD 3 -#define R300_PVS_DST_MACRO_INST (1 << 7) -# define R300_PVS_MACRO_OP_2CLK_MADD 0 -#define R300_PVS_DST_REG_TYPE(x) ((x) << 8) -# define R300_PVS_DST_REG_TEMPORARY 0 -# define R300_PVS_DST_REG_A0 1 -# define R300_PVS_DST_REG_OUT 2 -# define R300_PVS_DST_REG_OUT_REPL_X 3 -# define R300_PVS_DST_REG_ALT_TEMPORARY 4 -# define R300_PVS_DST_REG_INPUT 5 -#define R300_PVS_DST_OFFSET(x) ((x) << 13) -#define R300_PVS_DST_WE(x) ((x) << 20) -#define R300_PVS_DST_WE_XYZW (0xf << 20) - -#define R300_PVS_SRC_REG_TYPE(x) ((x) << 0) -# define R300_PVS_SRC_REG_TEMPORARY 0 -# define R300_PVS_SRC_REG_INPUT 1 -# define R300_PVS_SRC_REG_CONSTANT 2 -# define R300_PVS_SRC_REG_ALT_TEMPORARY 3 -#define R300_PVS_SRC_OFFSET(x) ((x) << 5) -#define R300_PVS_SRC_SWIZZLE(x) ((x) << 13) -# define R300_PVS_SRC_SELECT_X 0 -# define R300_PVS_SRC_SELECT_Y 1 -# define R300_PVS_SRC_SELECT_Z 2 -# define R300_PVS_SRC_SELECT_W 3 -# define R300_PVS_SRC_SELECT_FORCE_0 4 -# define R300_PVS_SRC_SELECT_FORCE_1 5 -# define R300_PVS_SRC_SWIZZLE_XYZW \ - ((R300_PVS_SRC_SELECT_X | (R300_PVS_SRC_SELECT_Y << 3) | \ - (R300_PVS_SRC_SELECT_Z << 6) | (R300_PVS_SRC_SELECT_W << 9)) << 13) -# define R300_PVS_SRC_SWIZZLE_ZERO \ - ((R300_PVS_SRC_SELECT_FORCE_0 | (R300_PVS_SRC_SELECT_FORCE_0 << 3) | \ - (R300_PVS_SRC_SELECT_FORCE_0 << 6) | \ - (R300_PVS_SRC_SELECT_FORCE_0 << 9)) << 13) -# define R300_PVS_SRC_SWIZZLE_ONE \ - ((R300_PVS_SRC_SELECT_FORCE_1 | (R300_PVS_SRC_SELECT_FORCE_1 << 3) | \ - (R300_PVS_SRC_SELECT_FORCE_1 << 6) | \ - (R300_PVS_SRC_SELECT_FORCE_1 << 9)) << 13) - -static const struct tgsi_full_src_register r300_constant_zero = { - .SrcRegister.Extended = TRUE, - .SrcRegister.File = TGSI_FILE_NULL, - .SrcRegisterExtSwz.ExtSwizzleX = TGSI_EXTSWIZZLE_ZERO, - .SrcRegisterExtSwz.ExtSwizzleY = TGSI_EXTSWIZZLE_ZERO, - .SrcRegisterExtSwz.ExtSwizzleZ = TGSI_EXTSWIZZLE_ZERO, - .SrcRegisterExtSwz.ExtSwizzleW = TGSI_EXTSWIZZLE_ZERO, -}; - -/* Temporary struct used to hold assembly state while putting together - * fragment programs. */ -struct r300_vs_asm { - /* Pipe context. */ - struct r300_context* r300; - /* Number of colors. */ - unsigned color_count; - /* Number of texcoords. */ - unsigned tex_count; - /* Number of requested temporary registers. */ - unsigned temp_count; - /* Offset for immediate constants. Neither R300 nor R500 can do four - * inline constants per source, so instead we copy immediates into the - * constant buffer. */ - unsigned imm_offset; - /* Number of immediate constants. */ - unsigned imm_count; - /* Number of colors to write. */ - unsigned out_colors; - /* Number of texcoords to write. */ - unsigned out_texcoords; - /* Whether to emit point size. */ - boolean point_size; - /* Tab of declared outputs to OVM outputs. */ - unsigned tab[16]; -}; - -static struct r300_vertex_shader r300_passthrough_vertex_shader = { - /* XXX translate these back into normal instructions */ - .instruction_count = 2, - .instructions[0].inst0 = R300_PVS_DST_OPCODE(R300_VE_ADD) | - R300_PVS_DST_REG_TYPE(R300_PVS_DST_REG_OUT) | - R300_PVS_DST_OFFSET(0) | R300_PVS_DST_WE_XYZW, - .instructions[0].inst1 = R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) | - R300_PVS_SRC_OFFSET(0) | R300_PVS_SRC_SWIZZLE_XYZW, - .instructions[0].inst2 = R300_PVS_SRC_SWIZZLE_ZERO, - .instructions[0].inst3 = 0x0, - .instructions[1].inst0 = R300_PVS_DST_OPCODE(R300_VE_ADD) | - R300_PVS_DST_REG_TYPE(R300_PVS_DST_REG_OUT) | - R300_PVS_DST_OFFSET(1) | R300_PVS_DST_WE_XYZW, - .instructions[1].inst1 = R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) | - R300_PVS_SRC_OFFSET(1) | R300_PVS_SRC_SWIZZLE_XYZW, - .instructions[1].inst2 = R300_PVS_SRC_SWIZZLE_ZERO, - .instructions[1].inst3 = 0x0, -}; - -static struct r300_vertex_shader r300_texture_vertex_shader = { - /* XXX translate these back into normal instructions */ - .instruction_count = 2, - .instructions[0].inst0 = R300_PVS_DST_OPCODE(R300_VE_ADD) | - R300_PVS_DST_REG_TYPE(R300_PVS_DST_REG_OUT) | - R300_PVS_DST_OFFSET(0) | R300_PVS_DST_WE_XYZW, - .instructions[0].inst1 = R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) | - R300_PVS_SRC_OFFSET(0) | R300_PVS_SRC_SWIZZLE_XYZW, - .instructions[0].inst2 = R300_PVS_SRC_SWIZZLE_ZERO, - .instructions[0].inst3 = 0x0, - .instructions[1].inst0 = R300_PVS_DST_OPCODE(R300_VE_ADD) | - R300_PVS_DST_REG_TYPE(R300_PVS_DST_REG_OUT) | - R300_PVS_DST_OFFSET(1) | R300_PVS_DST_WE_XYZW, - .instructions[1].inst1 = R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) | - R300_PVS_SRC_OFFSET(1) | R300_PVS_SRC_SWIZZLE_XYZW, - .instructions[1].inst2 = R300_PVS_SRC_SWIZZLE_ZERO, - .instructions[1].inst3 = 0x0, -}; - -void r300_translate_vertex_shader(struct r300_context* r300, - struct r300_vertex_shader* vs); - -#endif /* R300_STATE_TCL_H */ diff --git a/src/gallium/drivers/r300/r300_surface.c b/src/gallium/drivers/r300/r300_surface.c index 4dd5b8af99a..96e6e4a77d4 100644 --- a/src/gallium/drivers/r300/r300_surface.c +++ b/src/gallium/drivers/r300/r300_surface.c @@ -23,42 +23,48 @@ #include "r300_surface.h" -static void r300_surface_setup(struct pipe_context* pipe, - struct pipe_surface* dest, +static void r300_surface_setup(struct r300_context* r300, + struct r300_texture* dest, unsigned x, unsigned y, unsigned w, unsigned h) { - struct r300_context* r300 = r300_context(pipe); - struct r300_capabilities* caps = r300_screen(pipe->screen)->caps; - struct r300_texture* tex = (struct r300_texture*)dest->texture; - unsigned pixpitch = tex->stride / tex->tex.block.size; + struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps; + unsigned pixpitch = dest->stride / dest->tex.block.size; CS_LOCALS(r300); - /* Make sure our target BO is okay. */ - r300->winsys->add_buffer(r300->winsys, tex->buffer, - 0, RADEON_GEM_DOMAIN_VRAM); - if (r300->winsys->validate(r300->winsys)) { - r300->context.flush(&r300->context, 0, NULL); - } - r300_emit_blend_state(r300, &blend_clear_state); r300_emit_blend_color_state(r300, &blend_color_clear_state); r300_emit_dsa_state(r300, &dsa_clear_state); r300_emit_rs_state(r300, &rs_clear_state); - BEGIN_CS(15); + BEGIN_CS(26); + + /* Viewport setup */ + OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6); + OUT_CS_32F((float)w); + OUT_CS_32F((float)x); + OUT_CS_32F((float)h); + OUT_CS_32F((float)y); + OUT_CS_32F(1.0); + OUT_CS_32F(0.0); + + OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VPORT_X_SCALE_ENA | + R300_VPORT_X_OFFSET_ENA | + R300_VPORT_Y_SCALE_ENA | + R300_VPORT_Y_OFFSET_ENA | + R300_VTX_XY_FMT | R300_VTX_Z_FMT); /* Pixel scissors. */ OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2); if (caps->is_r500) { OUT_CS((x << R300_SCISSORS_X_SHIFT) | (y << R300_SCISSORS_Y_SHIFT)); - OUT_CS((w << R300_SCISSORS_X_SHIFT) | (h << R300_SCISSORS_Y_SHIFT)); + OUT_CS(((w - 1) << R300_SCISSORS_X_SHIFT) | ((h - 1) << R300_SCISSORS_Y_SHIFT)); } else { /* Non-R500 chipsets have an offset of 1440 in their scissors. */ OUT_CS(((x + 1440) << R300_SCISSORS_X_SHIFT) | ((y + 1440) << R300_SCISSORS_Y_SHIFT)); - OUT_CS(((w + 1440) << R300_SCISSORS_X_SHIFT) | - ((h + 1440) << R300_SCISSORS_Y_SHIFT)); + OUT_CS((((w - 1) + 1440) << R300_SCISSORS_X_SHIFT) | + (((h - 1) + 1440) << R300_SCISSORS_Y_SHIFT)); } /* Flush colorbuffer and blend caches. */ @@ -71,9 +77,11 @@ static void r300_surface_setup(struct pipe_context* pipe, /* Setup colorbuffer. */ OUT_CS_REG_SEQ(R300_RB3D_COLOROFFSET0, 1); - OUT_CS_RELOC(tex->buffer, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); - OUT_CS_REG(R300_RB3D_COLORPITCH0, pixpitch | - r300_translate_colorformat(tex->tex.format)); + OUT_CS_RELOC(dest->buffer, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); + OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0, 1); + OUT_CS_RELOC(dest->buffer, pixpitch | + r300_translate_colorformat(dest->tex.format), 0, + RADEON_GEM_DOMAIN_VRAM, 0); OUT_CS_REG(RB3D_COLOR_CHANNEL_MASK, 0xf); END_CS; @@ -93,6 +101,7 @@ static void r300_surface_fill(struct pipe_context* pipe, struct r300_capabilities* caps = r300_screen(pipe->screen)->caps; struct r300_texture* tex = (struct r300_texture*)dest->texture; unsigned pixpitch = tex->stride / tex->tex.block.size; + boolean invalid = FALSE; CS_LOCALS(r300); a = (float)((color >> 24) & 0xff) / 255.0f; @@ -105,19 +114,41 @@ static void r300_surface_fill(struct pipe_context* pipe, /* Fallback? */ if (FALSE) { +fallback: debug_printf("r300: Falling back on surface clear..."); util_surface_fill(pipe, dest, x, y, w, h, color); return; } - r300_surface_setup(r300, dest, x, y, w, h); + /* Make sure our target BO is okay. */ +validate: + if (!r300->winsys->add_buffer(r300->winsys, tex->buffer, + 0, RADEON_GEM_DOMAIN_VRAM)) { + r300->context.flush(&r300->context, 0, NULL); + goto validate; + } + if (!r300->winsys->validate(r300->winsys)) { + r300->context.flush(&r300->context, 0, NULL); + if (invalid) { + debug_printf("r300: Stuck in validation loop, gonna fallback."); + goto fallback; + } + invalid = TRUE; + goto validate; + } + + r300_surface_setup(r300, tex, x, y, w, h); /* Vertex shader setup */ if (caps->has_tcl) { - r300_emit_vertex_shader(r300, &r300_passthrough_vertex_shader); + r300_emit_vertex_program_code(r300, &r300_passthrough_vertex_shader, 0); } else { BEGIN_CS(4); - OUT_CS_REG(R300_VAP_CNTL_STATUS, R300_VAP_TCL_BYPASS); + OUT_CS_REG(R300_VAP_CNTL_STATUS, +#ifdef PIPE_ARCH_BIG_ENDIAN + R300_VC_32BIT_SWAP | +#endif + R300_VAP_TCL_BYPASS); OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(5) | R300_PVS_NUM_CNTLRS(5) | R300_PVS_NUM_FPUS(caps->num_vert_fpus) | @@ -127,14 +158,14 @@ static void r300_surface_fill(struct pipe_context* pipe, /* Fragment shader setup */ if (caps->is_r500) { - r500_emit_fragment_shader(r300, &r500_passthrough_fragment_shader); - r300_emit_rs_block_state(r300, &r500_rs_block_clear_state); + r500_emit_fragment_program_code(r300, &r5xx_passthrough_fragment_shader, 0); + r300_emit_rs_block_state(r300, &r5xx_rs_block_clear_state); } else { - r300_emit_fragment_shader(r300, &r300_passthrough_fragment_shader); - r300_emit_rs_block_state(r300, &r300_rs_block_clear_state); + r300_emit_fragment_program_code(r300, &r3xx_passthrough_fragment_shader, 0); + r300_emit_rs_block_state(r300, &r3xx_rs_block_clear_state); } - BEGIN_CS(31); + BEGIN_CS(26); /* VAP stream control, mapping from input memory to PVS/RS memory */ if (caps->has_tcl) { @@ -161,27 +192,21 @@ static void r300_surface_fill(struct pipe_context* pipe, /* Disable textures */ OUT_CS_REG(R300_TX_ENABLE, 0x0); - /* Viewport setup */ - OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6); - OUT_CS_32F(1.0); - OUT_CS_32F((float)x); - OUT_CS_32F(1.0); - OUT_CS_32F((float)y); - OUT_CS_32F(1.0); - OUT_CS_32F(0.0); - /* The size of the point we're about to draw, in sixths of pixels */ OUT_CS_REG(R300_GA_POINT_SIZE, - ((h * 6) & R300_POINTSIZE_Y_MASK) | + ((h * 6) & R300_POINTSIZE_Y_MASK) | ((w * 6) << R300_POINTSIZE_X_SHIFT)); + /* Vertex size. */ + OUT_CS_REG(R300_VAP_VTX_SIZE, 0x8); + /* Packet3 with our point vertex */ OUT_CS_PKT3(R200_3D_DRAW_IMMD_2, 8); OUT_CS(R300_PRIM_TYPE_POINT | R300_PRIM_WALK_RING | (1 << R300_PRIM_NUM_VERTICES_SHIFT)); /* Position */ - OUT_CS_32F(w / 2.0); - OUT_CS_32F(h / 2.0); + OUT_CS_32F(0.5); + OUT_CS_32F(0.5); OUT_CS_32F(1.0); OUT_CS_32F(1.0); /* Color */ @@ -190,11 +215,7 @@ static void r300_surface_fill(struct pipe_context* pipe, OUT_CS_32F(b); OUT_CS_32F(a); - /* XXX figure out why this is 0xA and not 0x2 */ OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, 0xA); - /* XXX OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT, - R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE | - R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE); */ END_CS; @@ -213,30 +234,63 @@ static void r300_surface_copy(struct pipe_context* pipe, struct r300_texture* srctex = (struct r300_texture*)src->texture; struct r300_texture* desttex = (struct r300_texture*)dest->texture; unsigned pixpitch = srctex->stride / srctex->tex.block.size; + boolean invalid = FALSE; + float fsrcx = srcx, fsrcy = srcy, fdestx = destx, fdesty = desty; CS_LOCALS(r300); debug_printf("r300: Copying surface %p at (%d,%d) to %p at (%d, %d)," " dimensions %dx%d (pixel pitch %d)\n", src, srcx, srcy, dest, destx, desty, w, h, pixpitch); - if ((srctex == desttex) && + if ((srctex->buffer == desttex->buffer) && ((destx < srcx + w) || (srcx < destx + w)) && ((desty < srcy + h) || (srcy < desty + h))) { +fallback: debug_printf("r300: Falling back on surface_copy\n"); util_surface_copy(pipe, FALSE, dest, destx, desty, src, srcx, srcy, w, h); } - r300_emit_sampler(r300, &r300_sampler_copy_state, 0); - r300_emit_texture(r300, srctex, 0); + /* Add our target BOs to the list. */ +validate: + if (!r300->winsys->add_buffer(r300->winsys, srctex->buffer, + RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0)) { + r300->context.flush(&r300->context, 0, NULL); + goto validate; + } + if (!r300->winsys->add_buffer(r300->winsys, desttex->buffer, + 0, RADEON_GEM_DOMAIN_VRAM)) { + r300->context.flush(&r300->context, 0, NULL); + goto validate; + } + if (!r300->winsys->validate(r300->winsys)) { + r300->context.flush(&r300->context, 0, NULL); + if (invalid) { + debug_printf("r300: Stuck in validation loop, gonna fallback."); + goto fallback; + } + invalid = TRUE; + goto validate; + } + + r300_surface_setup(r300, desttex, destx, desty, w, h); + + /* Setup the texture. */ + r300_emit_texture(r300, &r300_sampler_copy_state, srctex, 0); + + /* Flush and enable. */ r300_flush_textures(r300); /* Vertex shader setup */ if (caps->has_tcl) { - r300_emit_vertex_shader(r300, &r300_texture_vertex_shader); + r300_emit_vertex_program_code(r300, &r300_passthrough_vertex_shader, 0); } else { BEGIN_CS(4); - OUT_CS_REG(R300_VAP_CNTL_STATUS, R300_VAP_TCL_BYPASS); + OUT_CS_REG(R300_VAP_CNTL_STATUS, +#ifdef PIPE_ARCH_BIG_ENDIAN + R300_VC_32BIT_SWAP | +#endif + R300_VAP_TCL_BYPASS); OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(5) | R300_PVS_NUM_CNTLRS(5) | R300_PVS_NUM_FPUS(caps->num_vert_fpus) | @@ -246,13 +300,14 @@ static void r300_surface_copy(struct pipe_context* pipe, /* Fragment shader setup */ if (caps->is_r500) { - r500_emit_fragment_shader(r300, &r500_texture_fragment_shader); - r300_emit_rs_block_state(r300, &r500_rs_block_copy_state); + r500_emit_fragment_program_code(r300, &r5xx_texture_fragment_shader, 0); + r300_emit_rs_block_state(r300, &r5xx_rs_block_copy_state); } else { - r300_emit_fragment_shader(r300, &r300_texture_fragment_shader); - r300_emit_rs_block_state(r300, &r300_rs_block_copy_state); + r300_emit_fragment_program_code(r300, &r3xx_texture_fragment_shader, 0); + r300_emit_rs_block_state(r300, &r3xx_rs_block_copy_state); } + BEGIN_CS(30); /* VAP stream control, mapping from input memory to PVS/RS memory */ if (caps->has_tcl) { OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0, @@ -275,33 +330,38 @@ static void r300_surface_copy(struct pipe_context* pipe, /* Two components of texture 0 */ OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_1, 0x2); + /* Vertex size. */ + OUT_CS_REG(R300_VAP_VTX_SIZE, 0x4); + /* Packet3 with our texcoords */ - OUT_CS_PKT3(R200_3D_DRAW_IMMD_2, 8); + OUT_CS_PKT3(R200_3D_DRAW_IMMD_2, 16); OUT_CS(R300_PRIM_TYPE_QUADS | R300_PRIM_WALK_RING | (4 << R300_PRIM_NUM_VERTICES_SHIFT)); /* (x , y ) */ - OUT_CS_32F((float)destx); - OUT_CS_32F((float)desty); - OUT_CS_32F((float)srcx); - OUT_CS_32F((float)srcy); + OUT_CS_32F(fdestx / dest->width); + OUT_CS_32F(fdesty / dest->height); + OUT_CS_32F(fsrcx / src->width); + OUT_CS_32F(fsrcy / src->height); /* (x , y + h) */ - OUT_CS_32F((float)destx); - OUT_CS_32F((float)(desty + h)); - OUT_CS_32F((float)srcx); - OUT_CS_32F((float)(srcy + h)); + OUT_CS_32F(fdestx / dest->width); + OUT_CS_32F((fdesty + h) / dest->height); + OUT_CS_32F(fsrcx / src->width); + OUT_CS_32F((fsrcy + h) / src->height); /* (x + w, y + h) */ - OUT_CS_32F((float)(destx + w)); - OUT_CS_32F((float)(desty + h)); - OUT_CS_32F((float)(srcx + w)); - OUT_CS_32F((float)(srcy + h)); + OUT_CS_32F((fdestx + w) / dest->width); + OUT_CS_32F((fdesty + h) / dest->height); + OUT_CS_32F((fsrcx + w) / src->width); + OUT_CS_32F((fsrcy + h) / src->height); /* (x + w, y ) */ - OUT_CS_32F((float)(destx + w)); - OUT_CS_32F((float)desty); - OUT_CS_32F((float)(srcx + w)); - OUT_CS_32F((float)srcy); + OUT_CS_32F((fdestx + w) / dest->width); + OUT_CS_32F(fdesty / dest->height); + OUT_CS_32F((fsrcx + w) / src->width); + OUT_CS_32F(fsrcy / src->height); OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, 0xA); + END_CS; + r300->dirty_hw++; } diff --git a/src/gallium/drivers/r300/r300_surface.h b/src/gallium/drivers/r300/r300_surface.h index 894def07aa8..f9e98b2ec9c 100644 --- a/src/gallium/drivers/r300/r300_surface.h +++ b/src/gallium/drivers/r300/r300_surface.h @@ -31,8 +31,8 @@ #include "r300_context.h" #include "r300_cs.h" #include "r300_emit.h" -#include "r300_state_shader.h" -#include "r300_state_tcl.h" +#include "r300_fs.h" +#include "r300_vs.h" #include "r300_state_inlines.h" static struct r300_blend_state blend_clear_state = { @@ -72,17 +72,17 @@ static struct r300_rs_state rs_clear_state = { .color_control = R300_SHADE_MODEL_FLAT, }; -static struct r300_rs_block r300_rs_block_clear_state = { - .ip[0] = R500_RS_SEL_S(R300_RS_SEL_K0) | - R500_RS_SEL_T(R300_RS_SEL_K0) | - R500_RS_SEL_R(R300_RS_SEL_K0) | +static struct r300_rs_block r3xx_rs_block_clear_state = { + .ip[0] = R500_RS_SEL_S(R300_RS_SEL_C0) | + R500_RS_SEL_T(R300_RS_SEL_C0) | + R500_RS_SEL_R(R300_RS_SEL_C0) | R500_RS_SEL_Q(R300_RS_SEL_K1), .inst[0] = R300_RS_INST_COL_CN_WRITE, .count = R300_IT_COUNT(0) | R300_IC_COUNT(1) | R300_HIRES_EN, .inst_count = 0, }; -static struct r300_rs_block r500_rs_block_clear_state = { +static struct r300_rs_block r5xx_rs_block_clear_state = { .ip[0] = R500_RS_SEL_S(R500_RS_IP_PTR_K0) | R500_RS_SEL_T(R500_RS_IP_PTR_K0) | R500_RS_SEL_R(R500_RS_IP_PTR_K0) | @@ -94,24 +94,24 @@ static struct r300_rs_block r500_rs_block_clear_state = { /* The following state is used for surface_copy only. */ -static struct r300_rs_block r300_rs_block_copy_state = { +static struct r300_rs_block r3xx_rs_block_copy_state = { .ip[0] = R500_RS_SEL_S(R300_RS_SEL_K0) | R500_RS_SEL_T(R300_RS_SEL_K0) | R500_RS_SEL_R(R300_RS_SEL_K0) | R500_RS_SEL_Q(R300_RS_SEL_K1), .inst[0] = R300_RS_INST_COL_CN_WRITE, .count = R300_IT_COUNT(2) | R300_IC_COUNT(0) | R300_HIRES_EN, - .inst_count = R300_RS_TX_OFFSET(6), + .inst_count = R300_RS_TX_OFFSET(0), }; -static struct r300_rs_block r500_rs_block_copy_state = { +static struct r300_rs_block r5xx_rs_block_copy_state = { .ip[0] = R500_RS_SEL_S(0) | R500_RS_SEL_T(1) | R500_RS_SEL_R(R500_RS_IP_PTR_K0) | R500_RS_SEL_Q(R500_RS_IP_PTR_K1), .inst[0] = R500_RS_INST_TEX_CN_WRITE, .count = R300_IT_COUNT(2) | R300_IC_COUNT(0) | R300_HIRES_EN, - .inst_count = R300_RS_TX_OFFSET(6), + .inst_count = R300_RS_TX_OFFSET(0), }; static struct r300_sampler_state r300_sampler_copy_state = { diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c index fe91f4e1844..590052509cc 100644 --- a/src/gallium/drivers/r300/r300_texture.c +++ b/src/gallium/drivers/r300/r300_texture.c @@ -22,33 +22,35 @@ #include "r300_texture.h" -static int minify(int i) -{ - return MAX2(1, i >> 1); -} - static void r300_setup_texture_state(struct r300_texture* tex, unsigned width, unsigned height, - unsigned pitch) + unsigned pitch, + unsigned levels) { struct r300_texture_state* state = &tex->state; state->format0 = R300_TX_WIDTH((width - 1) & 0x7ff) | - R300_TX_HEIGHT((height - 1) & 0x7ff) | R300_TX_PITCH_EN; + R300_TX_HEIGHT((height - 1) & 0x7ff) | + R300_TX_NUM_LEVELS(levels) | + R300_TX_PITCH_EN; /* XXX */ - state->format1 = R300_TX_FORMAT_A8R8G8B8; + state->format1 = r300_translate_texformat(tex->tex.format); state->format2 = pitch - 1; - /* XXX + /* Assume (somewhat foolishly) that oversized textures will + * not be permitted by the state tracker. */ if (width > 2048) { - state->pitch |= R300_TXWIDTH_11; + state->format2 |= R500_TXWIDTH_BIT11; } if (height > 2048) { - state->pitch |= R300_TXHEIGHT_11; - } */ + state->format2 |= R500_TXHEIGHT_BIT11; + } + + debug_printf("r300: Set texture state (%dx%d, pitch %d, %d levels)\n", + width, height, pitch, levels); } static void r300_setup_miptree(struct r300_texture* tex) @@ -65,16 +67,24 @@ static void r300_setup_miptree(struct r300_texture* tex) } base->nblocksx[i] = pf_get_nblocksx(&base->block, base->width[i]); - base->nblocksy[i] = pf_get_nblocksy(&base->block, base->width[i]); - - /* Radeons enjoy things in multiples of 32. */ - /* XXX this can be 32 when POT */ - stride = (base->nblocksx[i] * base->block.size + 63) & ~63; + base->nblocksy[i] = pf_get_nblocksy(&base->block, base->height[i]); + + /* Radeons enjoy things in multiples of 64. + * + * XXX + * POT, uncompressed, unmippmapped textures can be aligned to 32, + * instead of 64. */ + stride = align(pf_get_stride(&base->block, base->width[i]), 32); size = stride * base->nblocksy[i] * base->depth[i]; - tex->offset[i] = (tex->size + 63) & ~63; + tex->offset[i] = align(tex->size, 32); tex->size = tex->offset[i] + size; + debug_printf("r300: Texture miptree: Level %d " + "(%dx%dx%d px, pitch %d bytes)\n", + i, base->width[i], base->height[i], base->depth[i], + stride); + /* Save stride of first level to the texture. */ if (i == 0) { tex->stride = stride; } @@ -86,8 +96,6 @@ static struct pipe_texture* r300_texture_create(struct pipe_screen* screen, const struct pipe_texture* template) { - /* XXX struct r300_screen* r300screen = r300_screen(screen); */ - struct r300_texture* tex = CALLOC_STRUCT(r300_texture); if (!tex) { @@ -100,11 +108,10 @@ static struct pipe_texture* r300_setup_miptree(tex); - /* XXX */ - r300_setup_texture_state(tex, tex->tex.width[0], tex->tex.height[0], - tex->tex.width[0]); + r300_setup_texture_state(tex, template->width[0], template->height[0], + template->width[0], template->last_level); - tex->buffer = screen->buffer_create(screen, 64, + tex->buffer = screen->buffer_create(screen, 1024, PIPE_BUFFER_USAGE_PIXEL, tex->size); @@ -166,6 +173,7 @@ static struct pipe_texture* { struct r300_texture* tex; + /* XXX we should start doing mips now... */ if (base->target != PIPE_TEXTURE_2D || base->last_level != 0 || base->depth[0] != 1) { @@ -183,8 +191,9 @@ static struct pipe_texture* tex->stride = *stride; + /* XXX */ r300_setup_texture_state(tex, tex->tex.width[0], tex->tex.height[0], - tex->stride); + tex->stride, 0); pipe_buffer_reference(&tex->buffer, buffer); diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h index 98fb5c9a08f..3b56f0307c0 100644 --- a/src/gallium/drivers/r300/r300_texture.h +++ b/src/gallium/drivers/r300/r300_texture.h @@ -32,6 +32,52 @@ void r300_init_screen_texture_functions(struct pipe_screen* screen); +/* Note the signature of R300_EASY_TX_FORMAT(A, R, G, B, FORMAT)... */ +static INLINE uint32_t r300_translate_texformat(enum pipe_format format) +{ + switch (format) { + /* X8 */ + case PIPE_FORMAT_I8_UNORM: + return R300_EASY_TX_FORMAT(X, X, X, X, X8); + /* W8Z8Y8X8 */ + case PIPE_FORMAT_A8R8G8B8_UNORM: + return R300_EASY_TX_FORMAT(X, Y, Z, W, W8Z8Y8X8); + case PIPE_FORMAT_R8G8B8A8_UNORM: + return R300_EASY_TX_FORMAT(Y, Z, W, X, W8Z8Y8X8); + case PIPE_FORMAT_A8R8G8B8_SRGB: + return R300_EASY_TX_FORMAT(X, Y, Z, W, W8Z8Y8X8) | + R300_TX_FORMAT_GAMMA; + case PIPE_FORMAT_R8G8B8A8_SRGB: + return R300_EASY_TX_FORMAT(Y, Z, W, X, W8Z8Y8X8) | + R300_TX_FORMAT_GAMMA; + /* DXT1 */ + case PIPE_FORMAT_DXT1_RGB: + return R300_EASY_TX_FORMAT(X, Y, Z, ONE, DXT1); + case PIPE_FORMAT_DXT1_RGBA: + return R300_EASY_TX_FORMAT(X, Y, Z, W, DXT1); + /* DXT3 */ + case PIPE_FORMAT_DXT3_RGBA: + return R300_EASY_TX_FORMAT(X, Y, Z, W, DXT3); + /* DXT5 */ + case PIPE_FORMAT_DXT5_RGBA: + return R300_EASY_TX_FORMAT(Y, Z, W, X, DXT5); + /* YVYU422 */ + case PIPE_FORMAT_YCBCR: + return R300_EASY_TX_FORMAT(X, Y, Z, ONE, YVYU422) | + R300_TX_FORMAT_YUV_TO_RGB; + /* W24_FP */ + case PIPE_FORMAT_Z24S8_UNORM: + return R300_EASY_TX_FORMAT(X, X, X, X, W24_FP); + default: + debug_printf("r300: Implementation error: " + "Got unsupported texture format %s in %s\n", + pf_name(format), __FUNCTION__); + assert(0); + break; + } + return 0; +} + #ifndef R300_WINSYS_H boolean r300_get_texture_buffer(struct pipe_texture* texture, diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c new file mode 100644 index 00000000000..d68a1041063 --- /dev/null +++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c @@ -0,0 +1,336 @@ +/* + * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include "r300_tgsi_to_rc.h" + +#include "radeon_compiler.h" +#include "radeon_program.h" + +#include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_scan.h" +#include "tgsi/tgsi_util.h" + + +static unsigned translate_opcode(unsigned opcode) +{ + switch(opcode) { + case TGSI_OPCODE_ARL: return OPCODE_ARL; + case TGSI_OPCODE_MOV: return OPCODE_MOV; + case TGSI_OPCODE_LIT: return OPCODE_LIT; + case TGSI_OPCODE_RCP: return OPCODE_RCP; + case TGSI_OPCODE_RSQ: return OPCODE_RSQ; + case TGSI_OPCODE_EXP: return OPCODE_EXP; + case TGSI_OPCODE_LOG: return OPCODE_LOG; + case TGSI_OPCODE_MUL: return OPCODE_MUL; + case TGSI_OPCODE_ADD: return OPCODE_ADD; + case TGSI_OPCODE_DP3: return OPCODE_DP3; + case TGSI_OPCODE_DP4: return OPCODE_DP4; + case TGSI_OPCODE_DST: return OPCODE_DST; + case TGSI_OPCODE_MIN: return OPCODE_MIN; + case TGSI_OPCODE_MAX: return OPCODE_MAX; + case TGSI_OPCODE_SLT: return OPCODE_SLT; + case TGSI_OPCODE_SGE: return OPCODE_SGE; + case TGSI_OPCODE_MAD: return OPCODE_MAD; + case TGSI_OPCODE_SUB: return OPCODE_SUB; + case TGSI_OPCODE_LRP: return OPCODE_LRP; + /* case TGSI_OPCODE_CND: return OPCODE_CND; */ + case TGSI_OPCODE_DP2A: return OPCODE_DP2A; + /* gap */ + case TGSI_OPCODE_FRC: return OPCODE_FRC; + /* case TGSI_OPCODE_CLAMP: return OPCODE_CLAMP; */ + case TGSI_OPCODE_FLR: return OPCODE_FLR; + /* case TGSI_OPCODE_ROUND: return OPCODE_ROUND; */ + case TGSI_OPCODE_EX2: return OPCODE_EX2; + case TGSI_OPCODE_LG2: return OPCODE_LG2; + case TGSI_OPCODE_POW: return OPCODE_POW; + case TGSI_OPCODE_XPD: return OPCODE_XPD; + /* gap */ + case TGSI_OPCODE_ABS: return OPCODE_ABS; + case TGSI_OPCODE_RCC: return OPCODE_RCC; + case TGSI_OPCODE_DPH: return OPCODE_DPH; + case TGSI_OPCODE_COS: return OPCODE_COS; + case TGSI_OPCODE_DDX: return OPCODE_DDX; + case TGSI_OPCODE_DDY: return OPCODE_DDY; + /* case TGSI_OPCODE_KILP: return OPCODE_KILP; */ + case TGSI_OPCODE_PK2H: return OPCODE_PK2H; + case TGSI_OPCODE_PK2US: return OPCODE_PK2US; + case TGSI_OPCODE_PK4B: return OPCODE_PK4B; + case TGSI_OPCODE_PK4UB: return OPCODE_PK4UB; + case TGSI_OPCODE_RFL: return OPCODE_RFL; + case TGSI_OPCODE_SEQ: return OPCODE_SEQ; + case TGSI_OPCODE_SFL: return OPCODE_SFL; + case TGSI_OPCODE_SGT: return OPCODE_SGT; + case TGSI_OPCODE_SIN: return OPCODE_SIN; + case TGSI_OPCODE_SLE: return OPCODE_SLE; + case TGSI_OPCODE_SNE: return OPCODE_SNE; + case TGSI_OPCODE_STR: return OPCODE_STR; + case TGSI_OPCODE_TEX: return OPCODE_TEX; + case TGSI_OPCODE_TXD: return OPCODE_TXD; + case TGSI_OPCODE_TXP: return OPCODE_TXP; + case TGSI_OPCODE_UP2H: return OPCODE_UP2H; + case TGSI_OPCODE_UP2US: return OPCODE_UP2US; + case TGSI_OPCODE_UP4B: return OPCODE_UP4B; + case TGSI_OPCODE_UP4UB: return OPCODE_UP4UB; + case TGSI_OPCODE_X2D: return OPCODE_X2D; + case TGSI_OPCODE_ARA: return OPCODE_ARA; + case TGSI_OPCODE_ARR: return OPCODE_ARR; + case TGSI_OPCODE_BRA: return OPCODE_BRA; + case TGSI_OPCODE_CAL: return OPCODE_CAL; + case TGSI_OPCODE_RET: return OPCODE_RET; + case TGSI_OPCODE_SSG: return OPCODE_SSG; + case TGSI_OPCODE_CMP: return OPCODE_CMP; + case TGSI_OPCODE_SCS: return OPCODE_SCS; + case TGSI_OPCODE_TXB: return OPCODE_TXB; + /* case TGSI_OPCODE_NRM: return OPCODE_NRM; */ + /* case TGSI_OPCODE_DIV: return OPCODE_DIV; */ + case TGSI_OPCODE_DP2: return OPCODE_DP2; + case TGSI_OPCODE_TXL: return OPCODE_TXL; + case TGSI_OPCODE_BRK: return OPCODE_BRK; + case TGSI_OPCODE_IF: return OPCODE_IF; + /* case TGSI_OPCODE_LOOP: return OPCODE_LOOP; */ + /* case TGSI_OPCODE_REP: return OPCODE_REP; */ + case TGSI_OPCODE_ELSE: return OPCODE_ELSE; + case TGSI_OPCODE_ENDIF: return OPCODE_ENDIF; + case TGSI_OPCODE_ENDLOOP: return OPCODE_ENDLOOP; + /* case TGSI_OPCODE_ENDREP: return OPCODE_ENDREP; */ + case TGSI_OPCODE_PUSHA: return OPCODE_PUSHA; + case TGSI_OPCODE_POPA: return OPCODE_POPA; + /* case TGSI_OPCODE_CEIL: return OPCODE_CEIL; */ + /* case TGSI_OPCODE_I2F: return OPCODE_I2F; */ + case TGSI_OPCODE_NOT: return OPCODE_NOT; + case TGSI_OPCODE_TRUNC: return OPCODE_TRUNC; + /* case TGSI_OPCODE_SHL: return OPCODE_SHL; */ + /* case TGSI_OPCODE_SHR: return OPCODE_SHR; */ + case TGSI_OPCODE_AND: return OPCODE_AND; + case TGSI_OPCODE_OR: return OPCODE_OR; + /* case TGSI_OPCODE_MOD: return OPCODE_MOD; */ + case TGSI_OPCODE_XOR: return OPCODE_XOR; + /* case TGSI_OPCODE_SAD: return OPCODE_SAD; */ + /* case TGSI_OPCODE_TXF: return OPCODE_TXF; */ + /* case TGSI_OPCODE_TXQ: return OPCODE_TXQ; */ + case TGSI_OPCODE_CONT: return OPCODE_CONT; + /* case TGSI_OPCODE_EMIT: return OPCODE_EMIT; */ + /* case TGSI_OPCODE_ENDPRIM: return OPCODE_ENDPRIM; */ + /* case TGSI_OPCODE_BGNLOOP2: return OPCODE_BGNLOOP2; */ + case TGSI_OPCODE_BGNSUB: return OPCODE_BGNSUB; + /* case TGSI_OPCODE_ENDLOOP2: return OPCODE_ENDLOOP2; */ + case TGSI_OPCODE_ENDSUB: return OPCODE_ENDSUB; + case TGSI_OPCODE_NOISE1: return OPCODE_NOISE1; + case TGSI_OPCODE_NOISE2: return OPCODE_NOISE2; + case TGSI_OPCODE_NOISE3: return OPCODE_NOISE3; + case TGSI_OPCODE_NOISE4: return OPCODE_NOISE4; + case TGSI_OPCODE_NOP: return OPCODE_NOP; + /* gap */ + case TGSI_OPCODE_NRM4: return OPCODE_NRM4; + /* case TGSI_OPCODE_CALLNZ: return OPCODE_CALLNZ; */ + /* case TGSI_OPCODE_IFC: return OPCODE_IFC; */ + /* case TGSI_OPCODE_BREAKC: return OPCODE_BREAKC; */ + case TGSI_OPCODE_KIL: return OPCODE_KIL; + case TGSI_OPCODE_END: return OPCODE_END; + case TGSI_OPCODE_SWZ: return OPCODE_SWZ; + } + + fprintf(stderr, "Unknown opcode: %i\n", opcode); + abort(); +} + +static unsigned translate_saturate(unsigned saturate) +{ + switch(saturate) { + case TGSI_SAT_NONE: return SATURATE_OFF; + case TGSI_SAT_ZERO_ONE: return SATURATE_ZERO_ONE; + case TGSI_SAT_MINUS_PLUS_ONE: return SATURATE_PLUS_MINUS_ONE; + } + + fprintf(stderr, "Unknown saturate mode: %i\n", saturate); + abort(); +} + +static unsigned translate_register_file(unsigned file) +{ + switch(file) { + case TGSI_FILE_CONSTANT: return PROGRAM_CONSTANT; + case TGSI_FILE_IMMEDIATE: return PROGRAM_CONSTANT; + case TGSI_FILE_INPUT: return PROGRAM_INPUT; + case TGSI_FILE_OUTPUT: return PROGRAM_OUTPUT; + case TGSI_FILE_TEMPORARY: return PROGRAM_TEMPORARY; + case TGSI_FILE_ADDRESS: return PROGRAM_ADDRESS; + } + + fprintf(stderr, "Unhandled register file: %i\n", file); + abort(); +} + +static int translate_register_index( + struct tgsi_to_rc * ttr, + unsigned file, + int index) +{ + if (file == TGSI_FILE_IMMEDIATE) + return ttr->immediate_offset + index; + + return index; +} + +static void transform_dstreg( + struct tgsi_to_rc * ttr, + struct prog_dst_register * dst, + struct tgsi_full_dst_register * src) +{ + dst->File = translate_register_file(src->DstRegister.File); + dst->Index = translate_register_index(ttr, src->DstRegister.File, src->DstRegister.Index); + dst->WriteMask = src->DstRegister.WriteMask; + dst->RelAddr = src->DstRegister.Indirect; +} + +static void transform_srcreg( + struct tgsi_to_rc * ttr, + struct prog_src_register * dst, + struct tgsi_full_src_register * src) +{ + dst->File = translate_register_file(src->SrcRegister.File); + dst->Index = translate_register_index(ttr, src->SrcRegister.File, src->SrcRegister.Index); + dst->RelAddr = src->SrcRegister.Indirect; + dst->Swizzle = tgsi_util_get_full_src_register_extswizzle(src, 0); + dst->Swizzle |= tgsi_util_get_full_src_register_extswizzle(src, 1) << 3; + dst->Swizzle |= tgsi_util_get_full_src_register_extswizzle(src, 2) << 6; + dst->Swizzle |= tgsi_util_get_full_src_register_extswizzle(src, 3) << 9; + dst->Abs = src->SrcRegisterExtMod.Absolute; + dst->Negate = + src->SrcRegisterExtSwz.NegateX | + (src->SrcRegisterExtSwz.NegateY << 1) | + (src->SrcRegisterExtSwz.NegateZ << 2) | + (src->SrcRegisterExtSwz.NegateW << 3); + dst->Negate ^= src->SrcRegister.Negate ? NEGATE_XYZW : 0; +} + +static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_ext_texture src) +{ + switch(src.Texture) { + case TGSI_TEXTURE_1D: + dst->I.TexSrcTarget = TEXTURE_1D_INDEX; + break; + case TGSI_TEXTURE_2D: + dst->I.TexSrcTarget = TEXTURE_2D_INDEX; + break; + case TGSI_TEXTURE_3D: + dst->I.TexSrcTarget = TEXTURE_3D_INDEX; + break; + case TGSI_TEXTURE_CUBE: + dst->I.TexSrcTarget = TEXTURE_CUBE_INDEX; + break; + case TGSI_TEXTURE_RECT: + dst->I.TexSrcTarget = TEXTURE_RECT_INDEX; + break; + case TGSI_TEXTURE_SHADOW1D: + dst->I.TexSrcTarget = TEXTURE_1D_INDEX; + dst->I.TexShadow = 1; + break; + case TGSI_TEXTURE_SHADOW2D: + dst->I.TexSrcTarget = TEXTURE_2D_INDEX; + dst->I.TexShadow = 1; + break; + case TGSI_TEXTURE_SHADOWRECT: + dst->I.TexSrcTarget = TEXTURE_RECT_INDEX; + dst->I.TexShadow = 1; + break; + } +} + +static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_instruction * src) +{ + if (src->Instruction.Opcode == TGSI_OPCODE_END) + return; + + struct rc_instruction * dst = rc_insert_new_instruction(ttr->compiler, ttr->compiler->Program.Instructions.Prev); + int i; + + dst->I.Opcode = translate_opcode(src->Instruction.Opcode); + dst->I.SaturateMode = translate_saturate(src->Instruction.Saturate); + + if (src->Instruction.NumDstRegs) + transform_dstreg(ttr, &dst->I.DstReg, &src->FullDstRegisters[0]); + + for(i = 0; i < src->Instruction.NumSrcRegs; ++i) { + if (src->FullSrcRegisters[i].SrcRegister.File == TGSI_FILE_SAMPLER) + dst->I.TexSrcUnit = src->FullSrcRegisters[i].SrcRegister.Index; + else + transform_srcreg(ttr, &dst->I.SrcReg[i], &src->FullSrcRegisters[i]); + } + + /* Texturing. */ + transform_texture(dst, src->InstructionExtTexture); +} + +static void handle_immediate(struct tgsi_to_rc * ttr, struct tgsi_full_immediate * imm) +{ + struct rc_constant constant; + int i; + + constant.Type = RC_CONSTANT_IMMEDIATE; + constant.Size = 4; + for(i = 0; i < 4; ++i) + constant.u.Immediate[i] = imm->u[i].Float; + rc_constants_add(&ttr->compiler->Program.Constants, &constant); +} + +void r300_tgsi_to_rc(struct tgsi_to_rc * ttr, const struct tgsi_token * tokens) +{ + struct tgsi_parse_context parser; + int i; + + /* Allocate constants placeholders. + * + * Note: What if declared constants are not contiguous? */ + for(i = 0; i <= ttr->info->file_max[TGSI_FILE_CONSTANT]; ++i) { + struct rc_constant constant; + memset(&constant, 0, sizeof(constant)); + constant.Type = RC_CONSTANT_EXTERNAL; + constant.Size = 4; + constant.u.External = i; + rc_constants_add(&ttr->compiler->Program.Constants, &constant); + } + + ttr->immediate_offset = ttr->compiler->Program.Constants.Count; + + tgsi_parse_init(&parser, tokens); + + while (!tgsi_parse_end_of_tokens(&parser)) { + tgsi_parse_token(&parser); + + switch (parser.FullToken.Token.Type) { + case TGSI_TOKEN_TYPE_DECLARATION: + break; + case TGSI_TOKEN_TYPE_IMMEDIATE: + handle_immediate(ttr, &parser.FullToken.FullImmediate); + break; + case TGSI_TOKEN_TYPE_INSTRUCTION: + transform_instruction(ttr, &parser.FullToken.FullInstruction); + break; + } + } + + tgsi_parse_free(&parser); + + rc_calculate_inputs_outputs(ttr->compiler); +} + diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.h b/src/gallium/drivers/r300/r300_tgsi_to_rc.h new file mode 100644 index 00000000000..93e90ec6d2c --- /dev/null +++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.h @@ -0,0 +1,41 @@ +/* + * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#ifndef R300_TGSI_TO_RC_H +#define R300_TGSI_TO_RC_H + +struct radeon_compiler; + +struct tgsi_full_declaration; +struct tgsi_shader_info; +struct tgsi_token; + +struct tgsi_to_rc { + struct radeon_compiler * compiler; + const struct tgsi_shader_info * info; + + int immediate_offset; +}; + +void r300_tgsi_to_rc(struct tgsi_to_rc * ttr, const struct tgsi_token * tokens); + +#endif /* R300_TGSI_TO_RC_H */ diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c new file mode 100644 index 00000000000..2cb903bba2f --- /dev/null +++ b/src/gallium/drivers/r300/r300_vs.c @@ -0,0 +1,234 @@ +/* + * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include "r300_vs.h" + +#include "r300_context.h" +#include "r300_tgsi_to_rc.h" + +#include "tgsi/tgsi_dump.h" +#include "tgsi/tgsi_parse.h" + +#include "radeon_compiler.h" + + +static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c) +{ + struct r300_vertex_shader * vs = c->UserData; + struct tgsi_shader_info* info = &vs->info; + boolean pointsize = false; + int out_colors = 0; + int colors = 0; + int out_generic = 0; + int generic = 0; + int i; + + /* Fill in the input mapping */ + for (i = 0; i < info->num_inputs; i++) + c->code->inputs[i] = i; + + /* Fill in the output mapping */ + for (i = 0; i < info->num_outputs; i++) { + switch (info->output_semantic_name[i]) { + case TGSI_SEMANTIC_PSIZE: + pointsize = true; + break; + case TGSI_SEMANTIC_COLOR: + out_colors++; + break; + case TGSI_SEMANTIC_FOG: + case TGSI_SEMANTIC_GENERIC: + out_generic++; + break; + } + } + + struct tgsi_parse_context parser; + + tgsi_parse_init(&parser, vs->state.tokens); + + while (!tgsi_parse_end_of_tokens(&parser)) { + tgsi_parse_token(&parser); + + if (parser.FullToken.Token.Type != TGSI_TOKEN_TYPE_DECLARATION) + continue; + + struct tgsi_full_declaration * decl = &parser.FullToken.FullDeclaration; + + if (decl->Declaration.File != TGSI_FILE_OUTPUT) + continue; + + switch (decl->Semantic.SemanticName) { + case TGSI_SEMANTIC_POSITION: + c->code->outputs[decl->DeclarationRange.First] = 0; + break; + case TGSI_SEMANTIC_PSIZE: + c->code->outputs[decl->DeclarationRange.First] = 1; + break; + case TGSI_SEMANTIC_COLOR: + c->code->outputs[decl->DeclarationRange.First] = 1 + + (pointsize ? 1 : 0) + + colors++; + break; + case TGSI_SEMANTIC_FOG: + case TGSI_SEMANTIC_GENERIC: + c->code->outputs[decl->DeclarationRange.First] = 1 + + (pointsize ? 1 : 0) + + out_colors + + generic++; + break; + default: + debug_printf("r300: vs: Bad semantic declaration %d\n", + decl->Semantic.SemanticName); + break; + } + } + + tgsi_parse_free(&parser); +} + + +void r300_translate_vertex_shader(struct r300_context* r300, + struct r300_vertex_shader* vs) +{ + struct r300_vertex_program_compiler compiler; + struct tgsi_to_rc ttr; + + /* Setup the compiler */ + rc_init(&compiler.Base); + + compiler.Base.Debug = 1; + compiler.code = &vs->code; + compiler.UserData = vs; + + if (compiler.Base.Debug) { + debug_printf("r300: Initial vertex program\n"); + tgsi_dump(vs->state.tokens, 0); + } + + /* Translate TGSI to our internal representation */ + ttr.compiler = &compiler.Base; + ttr.info = &vs->info; + + r300_tgsi_to_rc(&ttr, vs->state.tokens); + + compiler.RequiredOutputs = ~(~0 << vs->info.num_outputs); + compiler.SetHwInputOutput = &set_vertex_inputs_outputs; + + /* Invoke the compiler */ + r3xx_compile_vertex_program(&compiler); + if (compiler.Base.Error) { + /* Todo: Fail gracefully */ + fprintf(stderr, "r300 VP: Compiler error\n"); + abort(); + } + + /* And, finally... */ + rc_destroy(&compiler.Base); + vs->translated = TRUE; +} + + +/* XXX get these to r300_reg */ +#define R300_PVS_DST_OPCODE(x) ((x) << 0) +# define R300_VE_DOT_PRODUCT 1 +# define R300_VE_MULTIPLY 2 +# define R300_VE_ADD 3 +# define R300_VE_MAXIMUM 7 +# define R300_VE_SET_LESS_THAN 10 +#define R300_PVS_DST_MATH_INST (1 << 6) +# define R300_ME_RECIP_DX 6 +#define R300_PVS_DST_MACRO_INST (1 << 7) +# define R300_PVS_MACRO_OP_2CLK_MADD 0 +#define R300_PVS_DST_REG_TYPE(x) ((x) << 8) +# define R300_PVS_DST_REG_TEMPORARY 0 +# define R300_PVS_DST_REG_A0 1 +# define R300_PVS_DST_REG_OUT 2 +# define R300_PVS_DST_REG_OUT_REPL_X 3 +# define R300_PVS_DST_REG_ALT_TEMPORARY 4 +# define R300_PVS_DST_REG_INPUT 5 +#define R300_PVS_DST_OFFSET(x) ((x) << 13) +#define R300_PVS_DST_WE(x) ((x) << 20) +#define R300_PVS_DST_WE_XYZW (0xf << 20) + +#define R300_PVS_SRC_REG_TYPE(x) ((x) << 0) +# define R300_PVS_SRC_REG_TEMPORARY 0 +# define R300_PVS_SRC_REG_INPUT 1 +# define R300_PVS_SRC_REG_CONSTANT 2 +# define R300_PVS_SRC_REG_ALT_TEMPORARY 3 +#define R300_PVS_SRC_OFFSET(x) ((x) << 5) +#define R300_PVS_SRC_SWIZZLE(x) ((x) << 13) +# define R300_PVS_SRC_SELECT_X 0 +# define R300_PVS_SRC_SELECT_Y 1 +# define R300_PVS_SRC_SELECT_Z 2 +# define R300_PVS_SRC_SELECT_W 3 +# define R300_PVS_SRC_SELECT_FORCE_0 4 +# define R300_PVS_SRC_SELECT_FORCE_1 5 +# define R300_PVS_SRC_SWIZZLE_XYZW \ + ((R300_PVS_SRC_SELECT_X | (R300_PVS_SRC_SELECT_Y << 3) | \ + (R300_PVS_SRC_SELECT_Z << 6) | (R300_PVS_SRC_SELECT_W << 9)) << 13) +# define R300_PVS_SRC_SWIZZLE_ZERO \ + ((R300_PVS_SRC_SELECT_FORCE_0 | (R300_PVS_SRC_SELECT_FORCE_0 << 3) | \ + (R300_PVS_SRC_SELECT_FORCE_0 << 6) | \ + (R300_PVS_SRC_SELECT_FORCE_0 << 9)) << 13) +# define R300_PVS_SRC_SWIZZLE_ONE \ + ((R300_PVS_SRC_SELECT_FORCE_1 | (R300_PVS_SRC_SELECT_FORCE_1 << 3) | \ + (R300_PVS_SRC_SELECT_FORCE_1 << 6) | \ + (R300_PVS_SRC_SELECT_FORCE_1 << 9)) << 13) +#define R300_PVS_MODIFIER_X (1 << 25) +#define R300_PVS_MODIFIER_Y (1 << 26) +#define R300_PVS_MODIFIER_Z (1 << 27) +#define R300_PVS_MODIFIER_W (1 << 28) +#define R300_PVS_NEGATE_XYZW \ + (R300_PVS_MODIFIER_X | R300_PVS_MODIFIER_Y | \ + R300_PVS_MODIFIER_Z | R300_PVS_MODIFIER_W) + +struct r300_vertex_program_code r300_passthrough_vertex_shader = { + .length = 8, /* two instructions */ + + /* MOV out[0], in[0] */ + .body.d[0] = R300_PVS_DST_OPCODE(R300_VE_ADD) | + R300_PVS_DST_REG_TYPE(R300_PVS_DST_REG_OUT) | + R300_PVS_DST_OFFSET(0) | R300_PVS_DST_WE_XYZW, + .body.d[1] = R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) | + R300_PVS_SRC_OFFSET(0) | R300_PVS_SRC_SWIZZLE_XYZW, + .body.d[2] = R300_PVS_SRC_SWIZZLE_ZERO, + .body.d[3] = 0x0, + + /* MOV out[1], in[1] */ + .body.d[4] = R300_PVS_DST_OPCODE(R300_VE_ADD) | + R300_PVS_DST_REG_TYPE(R300_PVS_DST_REG_OUT) | + R300_PVS_DST_OFFSET(1) | R300_PVS_DST_WE_XYZW, + .body.d[5] = R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) | + R300_PVS_SRC_OFFSET(1) | R300_PVS_SRC_SWIZZLE_XYZW, + .body.d[6] = R300_PVS_SRC_SWIZZLE_ZERO, + .body.d[7] = 0x0, + + .inputs[0] = 0, + .inputs[1] = 1, + .outputs[0] = 0, + .outputs[1] = 1, + + .InputsRead = 3, + .OutputsWritten = 3 +}; + diff --git a/src/gallium/drivers/r300/r300_vs.h b/src/gallium/drivers/r300/r300_vs.h new file mode 100644 index 00000000000..2a4ce315e32 --- /dev/null +++ b/src/gallium/drivers/r300/r300_vs.h @@ -0,0 +1,54 @@ +/* + * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#ifndef R300_VS_H +#define R300_VS_H + +#include "pipe/p_state.h" +#include "tgsi/tgsi_scan.h" + +#include "radeon_code.h" + +struct r300_context; + +struct r300_vertex_shader { + /* Parent class */ + struct pipe_shader_state state; + struct tgsi_shader_info info; + + /* Fallback shader, because Draw has issues */ + struct draw_vertex_shader* draw; + + /* Has this shader been translated yet? */ + boolean translated; + + /* Machine code (if translated) */ + struct r300_vertex_program_code code; +}; + + +extern struct r300_vertex_program_code r300_passthrough_vertex_shader; + +void r300_translate_vertex_shader(struct r300_context* r300, + struct r300_vertex_shader* vs); + +#endif /* R300_VS_H */ diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h index a833bb0399a..f18ad75a47d 100644 --- a/src/gallium/drivers/r300/r300_winsys.h +++ b/src/gallium/drivers/r300/r300_winsys.h @@ -55,10 +55,10 @@ struct r300_winsys { uint32_t vram_size; /* Add a pipe_buffer to the list of buffer objects to validate. */ - void (*add_buffer)(struct r300_winsys* winsys, - struct pipe_buffer* pbuffer, - uint32_t rd, - uint32_t wd); + boolean (*add_buffer)(struct r300_winsys* winsys, + struct pipe_buffer* pbuffer, + uint32_t rd, + uint32_t wd); /* Revalidate all currently setup pipe_buffers. * Returns TRUE if a flush is required. */ diff --git a/src/gallium/drivers/r300/r3xx_fs.c b/src/gallium/drivers/r300/r3xx_fs.c new file mode 100644 index 00000000000..c1c1194d58e --- /dev/null +++ b/src/gallium/drivers/r300/r3xx_fs.c @@ -0,0 +1,74 @@ +/* + * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Joakim Sindholt <opensource@zhasha.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include "r3xx_fs.h" + +#include "r300_reg.h" + +struct rX00_fragment_program_code r3xx_passthrough_fragment_shader = { + .code.r300.alu.length = 1, + .code.r300.tex.length = 0, + + .code.r300.config = 0, + .code.r300.pixsize = 0, + .code.r300.code_offset = 0, + .code.r300.code_addr[3] = R300_RGBA_OUT, + + .code.r300.alu.inst[0].rgb_inst = R300_RGB_SWIZA(R300_ALU_ARGC_SRC0C_XYZ) | + R300_RGB_SWIZB(R300_ALU_ARGC_SRC0C_XYZ) | + R300_RGB_SWIZC(R300_ALU_ARGC_ZERO) | + R300_ALU_OUTC_CMP, + .code.r300.alu.inst[0].rgb_addr = R300_RGB_ADDR0(0) | R300_RGB_ADDR1(0) | + R300_RGB_ADDR2(0) | R300_ALU_DSTC_OUTPUT_XYZ, + .code.r300.alu.inst[0].alpha_inst = R300_ALPHA_SWIZA(R300_ALU_ARGA_SRC0A) | + R300_ALPHA_SWIZB(R300_ALU_ARGA_SRC0A) | + R300_ALPHA_SWIZC(R300_ALU_ARGA_ZERO) | + R300_ALU_OUTA_CMP, + .code.r300.alu.inst[0].alpha_addr = R300_ALPHA_ADDR0(0) | + R300_ALPHA_ADDR1(0) | R300_ALPHA_ADDR2(0) | R300_ALU_DSTA_OUTPUT, +}; + +struct rX00_fragment_program_code r3xx_texture_fragment_shader = { + .code.r300.alu.length = 1, + .code.r300.tex.length = 1, + + .code.r300.config = R300_PFS_CNTL_FIRST_NODE_HAS_TEX, + .code.r300.pixsize = 0, + .code.r300.code_offset = 0, + .code.r300.code_addr[3] = R300_RGBA_OUT, + + .code.r300.tex.inst[0] = R300_TEX_OP_LD << R300_TEX_INST_SHIFT, + + .code.r300.alu.inst[0].rgb_inst = R300_RGB_SWIZA(R300_ALU_ARGC_SRC0C_XYZ) | + R300_RGB_SWIZB(R300_ALU_ARGC_SRC0C_XYZ) | + R300_RGB_SWIZC(R300_ALU_ARGC_ZERO) | + R300_ALU_OUTC_CMP, + .code.r300.alu.inst[0].rgb_addr = R300_RGB_ADDR0(0) | R300_RGB_ADDR1(0) | + R300_RGB_ADDR2(0) | R300_ALU_DSTC_OUTPUT_XYZ, + .code.r300.alu.inst[0].alpha_inst = R300_ALPHA_SWIZA(R300_ALU_ARGA_SRC0A) | + R300_ALPHA_SWIZB(R300_ALU_ARGA_SRC0A) | + R300_ALPHA_SWIZC(R300_ALU_ARGA_ZERO) | + R300_ALU_OUTA_CMP, + .code.r300.alu.inst[0].alpha_addr = R300_ALPHA_ADDR0(0) | + R300_ALPHA_ADDR1(0) | R300_ALPHA_ADDR2(0) | R300_ALU_DSTA_OUTPUT, +}; diff --git a/src/gallium/drivers/r300/r300_debug.h b/src/gallium/drivers/r300/r3xx_fs.h index a1f873656dc..51cd245724d 100644 --- a/src/gallium/drivers/r300/r300_debug.h +++ b/src/gallium/drivers/r300/r3xx_fs.h @@ -1,5 +1,6 @@ /* - * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com> + * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Joakim Sindholt <opensource@zhasha.com> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -20,15 +21,12 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ -#ifndef R300_DEBUG_H -#define R300_DEBUG_H +#ifndef R3XX_FS_H +#define R3XX_FS_H -#include "r300_reg.h" -#include "r300_state_shader.h" -#include "r300_state_tcl.h" +#include "radeon_code.h" -void r500_fs_dump(struct r500_fragment_shader* fs); +struct rX00_fragment_program_code r3xx_passthrough_fragment_shader; +struct rX00_fragment_program_code r3xx_texture_fragment_shader; -void r300_vs_dump(struct r300_vertex_shader* vs); - -#endif /* R300_DEBUG_H */ +#endif /* R3XX_FS_H */ diff --git a/src/gallium/drivers/r300/r5xx_fs.c b/src/gallium/drivers/r300/r5xx_fs.c new file mode 100644 index 00000000000..f072deab0d9 --- /dev/null +++ b/src/gallium/drivers/r300/r5xx_fs.c @@ -0,0 +1,125 @@ +/* + * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Joakim Sindholt <opensource@zhasha.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include "r5xx_fs.h" + +#include "r300_reg.h" + +/* XXX this all should find its way back to r300_reg */ +/* Swizzle tools */ +#define R500_SWIZZLE_ZERO 4 +#define R500_SWIZZLE_HALF 5 +#define R500_SWIZZLE_ONE 6 +#define R500_SWIZ_RGB_ZERO ((4 << 0) | (4 << 3) | (4 << 6)) +#define R500_SWIZ_RGB_ONE ((6 << 0) | (6 << 3) | (6 << 6)) +#define R500_SWIZ_RGB_RGB ((0 << 0) | (1 << 3) | (2 << 6)) +#define R500_SWIZ_MOD_NEG 1 +#define R500_SWIZ_MOD_ABS 2 +#define R500_SWIZ_MOD_NEG_ABS 3 +/* Swizzles for inst2 */ +#define R500_SWIZ_TEX_STRQ(x) ((x) << 8) +#define R500_SWIZ_TEX_RGBA(x) ((x) << 24) +/* Swizzles for inst3 */ +#define R500_SWIZ_RGB_A(x) ((x) << 2) +#define R500_SWIZ_RGB_B(x) ((x) << 15) +/* Swizzles for inst4 */ +#define R500_SWIZ_ALPHA_A(x) ((x) << 14) +#define R500_SWIZ_ALPHA_B(x) ((x) << 21) +/* Swizzle for inst5 */ +#define R500_SWIZ_RGBA_C(x) ((x) << 14) +#define R500_SWIZ_ALPHA_C(x) ((x) << 27) +/* Writemasks */ +#define R500_TEX_WMASK(x) ((x) << 11) +#define R500_ALU_WMASK(x) ((x) << 11) +#define R500_ALU_OMASK(x) ((x) << 15) +#define R500_W_OMASK (1 << 31) + +struct rX00_fragment_program_code r5xx_passthrough_fragment_shader = { + .code.r500.max_temp_idx = 0, + .code.r500.inst_end = 0, + + .code.r500.inst[0].inst0 = R500_INST_TYPE_OUT | + R500_INST_TEX_SEM_WAIT | R500_INST_LAST | + R500_INST_RGB_OMASK_RGB | R500_INST_ALPHA_OMASK | + R500_INST_RGB_CLAMP | R500_INST_ALPHA_CLAMP, + .code.r500.inst[0].inst1 = + R500_RGB_ADDR0(0) | R500_RGB_ADDR1(0) | R500_RGB_ADDR1_CONST | + R500_RGB_ADDR2(0) | R500_RGB_ADDR2_CONST, + .code.r500.inst[0].inst2 = + R500_ALPHA_ADDR0(0) | R500_ALPHA_ADDR1(0) | R500_ALPHA_ADDR1_CONST | + R500_ALPHA_ADDR2(0) | R500_ALPHA_ADDR2_CONST, + .code.r500.inst[0].inst3 = + R500_ALU_RGB_SEL_A_SRC0 | R500_ALU_RGB_R_SWIZ_A_R | + R500_ALU_RGB_G_SWIZ_A_G | R500_ALU_RGB_B_SWIZ_A_B | + R500_ALU_RGB_SEL_B_SRC0 | R500_ALU_RGB_R_SWIZ_B_R | + R500_ALU_RGB_B_SWIZ_B_G | R500_ALU_RGB_G_SWIZ_B_B, + .code.r500.inst[0].inst4 = + R500_ALPHA_OP_CMP | R500_ALPHA_SWIZ_A_A | R500_ALPHA_SWIZ_B_A, + .code.r500.inst[0].inst5 = + R500_ALU_RGBA_OP_CMP | R500_ALU_RGBA_R_SWIZ_0 | + R500_ALU_RGBA_G_SWIZ_0 | R500_ALU_RGBA_B_SWIZ_0 | + R500_ALU_RGBA_A_SWIZ_0, +}; + +struct rX00_fragment_program_code r5xx_texture_fragment_shader = { + .code.r500.max_temp_idx = 0, + .code.r500.inst_end = 1, + + .code.r500.inst[0].inst0 = R500_INST_TYPE_TEX | + R500_INST_TEX_SEM_WAIT | + R500_INST_RGB_WMASK_RGB | R500_INST_ALPHA_WMASK | + R500_INST_RGB_CLAMP | R500_INST_ALPHA_CLAMP, + .code.r500.inst[0].inst1 = R500_TEX_ID(0) | R500_TEX_INST_LD | + R500_TEX_SEM_ACQUIRE | R500_TEX_IGNORE_UNCOVERED, + .code.r500.inst[0].inst2 = R500_TEX_SRC_ADDR(0) | + R500_TEX_SRC_S_SWIZ_R | R500_TEX_SRC_T_SWIZ_G | + R500_TEX_SRC_R_SWIZ_B | R500_TEX_SRC_Q_SWIZ_A | + R500_TEX_DST_ADDR(0) | + R500_TEX_DST_R_SWIZ_R | R500_TEX_DST_G_SWIZ_G | + R500_TEX_DST_B_SWIZ_B | R500_TEX_DST_A_SWIZ_A, + .code.r500.inst[0].inst3 = 0x0, + .code.r500.inst[0].inst4 = 0x0, + .code.r500.inst[0].inst5 = 0x0, + + .code.r500.inst[1].inst0 = R500_INST_TYPE_OUT | + R500_INST_TEX_SEM_WAIT | R500_INST_LAST | + R500_INST_RGB_OMASK_RGB | R500_INST_ALPHA_OMASK | + R500_INST_RGB_CLAMP | R500_INST_ALPHA_CLAMP, + .code.r500.inst[1].inst1 = + R500_RGB_ADDR0(0) | R500_RGB_ADDR1(0) | R500_RGB_ADDR1_CONST | + R500_RGB_ADDR2(0) | R500_RGB_ADDR2_CONST, + .code.r500.inst[1].inst2 = + R500_ALPHA_ADDR0(0) | R500_ALPHA_ADDR1(0) | R500_ALPHA_ADDR1_CONST | + R500_ALPHA_ADDR2(0) | R500_ALPHA_ADDR2_CONST, + .code.r500.inst[1].inst3 = + R500_ALU_RGB_SEL_A_SRC0 | R500_ALU_RGB_R_SWIZ_A_R | + R500_ALU_RGB_G_SWIZ_A_G | R500_ALU_RGB_B_SWIZ_A_B | + R500_ALU_RGB_SEL_B_SRC0 | R500_ALU_RGB_R_SWIZ_B_R | + R500_ALU_RGB_B_SWIZ_B_G | R500_ALU_RGB_G_SWIZ_B_B, + .code.r500.inst[1].inst4 = + R500_ALPHA_OP_CMP | R500_ALPHA_SWIZ_A_A | R500_ALPHA_SWIZ_B_A, + .code.r500.inst[1].inst5 = + R500_ALU_RGBA_OP_CMP | R500_ALU_RGBA_R_SWIZ_0 | + R500_ALU_RGBA_G_SWIZ_0 | R500_ALU_RGBA_B_SWIZ_0 | + R500_ALU_RGBA_A_SWIZ_0, +}; diff --git a/src/gallium/drivers/r300/r5xx_fs.h b/src/gallium/drivers/r300/r5xx_fs.h new file mode 100644 index 00000000000..a4addde32b2 --- /dev/null +++ b/src/gallium/drivers/r300/r5xx_fs.h @@ -0,0 +1,32 @@ +/* + * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Joakim Sindholt <opensource@zhasha.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#ifndef R5XX_FS_H +#define R5XX_FS_H + +#include "radeon_code.h" + +struct rX00_fragment_program_code r5xx_passthrough_fragment_shader; +struct rX00_fragment_program_code r5xx_texture_fragment_shader; + +#endif /* R5XX_FS_H */ diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h index dffc15a4f19..7888c2f644b 100644 --- a/src/gallium/drivers/softpipe/sp_context.h +++ b/src/gallium/drivers/softpipe/sp_context.h @@ -92,6 +92,7 @@ struct softpipe_context { * queries. */ uint64_t occlusion_count; + unsigned active_query_count; /** Mapped vertex buffers */ ubyte *mapped_vbuffer[PIPE_MAX_ATTRIBS]; diff --git a/src/gallium/drivers/softpipe/sp_draw_arrays.c b/src/gallium/drivers/softpipe/sp_draw_arrays.c index ba2766ff139..d4045816d03 100644 --- a/src/gallium/drivers/softpipe/sp_draw_arrays.c +++ b/src/gallium/drivers/softpipe/sp_draw_arrays.c @@ -35,6 +35,7 @@ #include "pipe/p_context.h" #include "pipe/internal/p_winsys_screen.h" #include "pipe/p_inlines.h" +#include "util/u_prim.h" #include "sp_context.h" #include "sp_state.h" @@ -65,6 +66,7 @@ softpipe_map_constant_buffers(struct softpipe_context *sp) size); } + static void softpipe_unmap_constant_buffers(struct softpipe_context *sp) { @@ -86,20 +88,6 @@ softpipe_unmap_constant_buffers(struct softpipe_context *sp) } -static unsigned reduced_prim[PIPE_PRIM_POLYGON + 1] = { - PIPE_PRIM_POINTS, - PIPE_PRIM_LINES, - PIPE_PRIM_LINES, - PIPE_PRIM_LINES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES -}; - - boolean softpipe_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) @@ -108,15 +96,11 @@ softpipe_draw_arrays(struct pipe_context *pipe, unsigned mode, } - /** * Draw vertex arrays, with optional indexing. * Basically, map the vertex buffers (and drawing surfaces), then hand off * the drawing to the 'draw' module. - * - * XXX should the element buffer be specified/bound with a separate function? */ - boolean softpipe_draw_range_elements(struct pipe_context *pipe, struct pipe_buffer *indexBuffer, @@ -129,7 +113,7 @@ softpipe_draw_range_elements(struct pipe_context *pipe, struct draw_context *draw = sp->draw; unsigned i; - sp->reduced_api_prim = reduced_prim[mode]; + sp->reduced_api_prim = u_reduced_prim(mode); if (sp->dirty) softpipe_update_derived( sp ); @@ -147,6 +131,7 @@ softpipe_draw_range_elements(struct pipe_context *pipe, PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_vertex_buffer(draw, i, buf); } + /* Map index buffer, if present */ if (indexBuffer) { void *mapped_indexes @@ -159,10 +144,10 @@ softpipe_draw_range_elements(struct pipe_context *pipe, } else { /* no index/element buffer */ - draw_set_mapped_element_buffer_range(draw, 0, start, start + count - 1, NULL); + draw_set_mapped_element_buffer_range(draw, 0, start, + start + count - 1, NULL); } - /* draw! */ draw_arrays(draw, mode, start, count); @@ -187,6 +172,7 @@ softpipe_draw_range_elements(struct pipe_context *pipe, return TRUE; } + boolean softpipe_draw_elements(struct pipe_context *pipe, struct pipe_buffer *indexBuffer, @@ -200,11 +186,9 @@ softpipe_draw_elements(struct pipe_context *pipe, } - void softpipe_set_edgeflags(struct pipe_context *pipe, const unsigned *edgeflags) { struct softpipe_context *sp = softpipe_context(pipe); draw_set_edgeflags(sp->draw, edgeflags); } - diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c index 31c3ca21c51..f4fa0905d74 100644 --- a/src/gallium/drivers/softpipe/sp_fs_sse.c +++ b/src/gallium/drivers/softpipe/sp_fs_sse.c @@ -45,17 +45,6 @@ #include "rtasm/rtasm_x86sse.h" -/* Surely this should be defined somewhere in a tgsi header: - */ -typedef void (PIPE_CDECL *codegen_function)( - const struct tgsi_exec_vector *input, - struct tgsi_exec_vector *output, - const float (*constant)[4], - struct tgsi_exec_vector *temporary, - const struct tgsi_interp_coef *coef, - float (*immediates)[4] - //, const struct tgsi_exec_vector *quadPos - ); /** @@ -65,7 +54,7 @@ struct sp_sse_fragment_shader { struct sp_fragment_shader base; struct x86_function sse2_program; - codegen_function func; + tgsi_sse2_fs_function func; float immediates[TGSI_EXEC_NUM_IMMEDIATES][4]; }; @@ -83,6 +72,7 @@ fs_sse_prepare( const struct sp_fragment_shader *base, struct tgsi_exec_machine *machine, struct tgsi_sampler **samplers ) { + machine->Samplers = samplers; } @@ -107,12 +97,10 @@ fs_sse_run( const struct sp_fragment_shader *base, tgsi_set_kill_mask(machine, 0x0); tgsi_set_exec_mask(machine, 1, 1, 1, 1); - shader->func( machine->Inputs, - machine->Outputs, + shader->func( machine, machine->Consts, - machine->Temps, - machine->InterpCoefs, - shader->immediates + (const float (*)[4])shader->immediates, + machine->InterpCoefs // , &machine->QuadPos ); @@ -151,7 +139,7 @@ softpipe_create_fs_sse(struct softpipe_context *softpipe, return NULL; } - shader->func = (codegen_function) x86_get_func( &shader->sse2_program ); + shader->func = (tgsi_sse2_fs_function) x86_get_func( &shader->sse2_program ); if (!shader->func) { x86_release_func( &shader->sse2_program ); FREE(shader); diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c index 06725fd09b4..42021789ea8 100644 --- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c +++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c @@ -238,65 +238,117 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr) case PIPE_PRIM_TRIANGLES: for (i = 2; i < nr; i += 3) { - setup_tri( setup_ctx, - get_vert(vertex_buffer, indices[i-2], stride), - get_vert(vertex_buffer, indices[i-1], stride), - get_vert(vertex_buffer, indices[i-0], stride)); + if (softpipe->rasterizer->flatshade_first) { + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride), + get_vert(vertex_buffer, indices[i-2], stride) ); + } + else { + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-2], stride), + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } } break; case PIPE_PRIM_TRIANGLE_STRIP: for (i = 2; i < nr; i += 1) { - setup_tri( setup_ctx, - get_vert(vertex_buffer, indices[i+(i&1)-2], stride), - get_vert(vertex_buffer, indices[i-(i&1)-1], stride), - get_vert(vertex_buffer, indices[i-0], stride)); + if (softpipe->rasterizer->flatshade_first) { + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i+(i&1)-1], stride), + get_vert(vertex_buffer, indices[i-(i&1)], stride), + get_vert(vertex_buffer, indices[i-2], stride) ); + } + else { + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i+(i&1)-2], stride), + get_vert(vertex_buffer, indices[i-(i&1)-1], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } } break; case PIPE_PRIM_TRIANGLE_FAN: for (i = 2; i < nr; i += 1) { - setup_tri( setup_ctx, - get_vert(vertex_buffer, indices[0], stride), - get_vert(vertex_buffer, indices[i-1], stride), - get_vert(vertex_buffer, indices[i-0], stride)); + if (softpipe->rasterizer->flatshade_first) { + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-0], stride), + get_vert(vertex_buffer, indices[0], stride), + get_vert(vertex_buffer, indices[i-1], stride) ); + } + else { + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[0], stride), + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } } break; case PIPE_PRIM_QUADS: for (i = 3; i < nr; i += 4) { - setup_tri( setup_ctx, - get_vert(vertex_buffer, indices[i-3], stride), - get_vert(vertex_buffer, indices[i-2], stride), - get_vert(vertex_buffer, indices[i-0], stride)); - - setup_tri( setup_ctx, - get_vert(vertex_buffer, indices[i-2], stride), - get_vert(vertex_buffer, indices[i-1], stride), - get_vert(vertex_buffer, indices[i-0], stride)); + if (softpipe->rasterizer->flatshade_first) { + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-2], stride), + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-3], stride) ); + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride), + get_vert(vertex_buffer, indices[i-3], stride) ); + } + else { + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-3], stride), + get_vert(vertex_buffer, indices[i-2], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-2], stride), + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } } break; case PIPE_PRIM_QUAD_STRIP: for (i = 3; i < nr; i += 2) { - setup_tri( setup_ctx, - get_vert(vertex_buffer, indices[i-3], stride), - get_vert(vertex_buffer, indices[i-2], stride), - get_vert(vertex_buffer, indices[i-0], stride)); - - setup_tri( setup_ctx, - get_vert(vertex_buffer, indices[i-1], stride), - get_vert(vertex_buffer, indices[i-3], stride), - get_vert(vertex_buffer, indices[i-0], stride)); + if (softpipe->rasterizer->flatshade_first) { + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-0], stride), + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-3], stride)); + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-2], stride), + get_vert(vertex_buffer, indices[i-0], stride), + get_vert(vertex_buffer, indices[i-3], stride) ); + } + else { + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-3], stride), + get_vert(vertex_buffer, indices[i-2], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + setup_tri( setup_ctx, + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[i-3], stride), + get_vert(vertex_buffer, indices[i-0], stride) ); + } } break; case PIPE_PRIM_POLYGON: + /* Almost same as tri fan but the _first_ vertex specifies the flat + * shading color. Note that the first polygon vertex is passed as + * the last triangle vertex here. + * flatshade_first state makes no difference. + */ for (i = 2; i < nr; i += 1) { setup_tri( setup_ctx, - get_vert(vertex_buffer, indices[0-1], stride), get_vert(vertex_buffer, indices[i-0], stride), - get_vert(vertex_buffer, indices[0], stride)); + get_vert(vertex_buffer, indices[i-1], stride), + get_vert(vertex_buffer, indices[0], stride) ); } break; @@ -368,58 +420,104 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr) } break; - case PIPE_PRIM_TRIANGLES: for (i = 2; i < nr; i += 3) { - setup_tri( setup_ctx, - get_vert(vertex_buffer, i-2, stride), - get_vert(vertex_buffer, i-1, stride), - get_vert(vertex_buffer, i-0, stride)); + if (softpipe->rasterizer->flatshade_first) { + setup_tri( setup_ctx, + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride), + get_vert(vertex_buffer, i-2, stride) ); + } + else { + setup_tri( setup_ctx, + get_vert(vertex_buffer, i-2, stride), + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride) ); + } } break; case PIPE_PRIM_TRIANGLE_STRIP: - for (i = 2; i < nr; i += 1) { - setup_tri( setup_ctx, - get_vert(vertex_buffer, i+(i&1)-2, stride), - get_vert(vertex_buffer, i-(i&1)-1, stride), - get_vert(vertex_buffer, i-0, stride)); + for (i = 2; i < nr; i++) { + if (softpipe->rasterizer->flatshade_first) { + setup_tri( setup_ctx, + get_vert(vertex_buffer, i+(i&1)-1, stride), + get_vert(vertex_buffer, i-(i&1), stride), + get_vert(vertex_buffer, i-2, stride) ); + } + else { + setup_tri( setup_ctx, + get_vert(vertex_buffer, i+(i&1)-2, stride), + get_vert(vertex_buffer, i-(i&1)-1, stride), + get_vert(vertex_buffer, i-0, stride) ); + } } break; case PIPE_PRIM_TRIANGLE_FAN: for (i = 2; i < nr; i += 1) { - setup_tri( setup_ctx, - get_vert(vertex_buffer, 0, stride), - get_vert(vertex_buffer, i-1, stride), - get_vert(vertex_buffer, i-0, stride)); + if (softpipe->rasterizer->flatshade_first) { + setup_tri( setup_ctx, + get_vert(vertex_buffer, i-0, stride), + get_vert(vertex_buffer, 0, stride), + get_vert(vertex_buffer, i-1, stride) ); + } + else { + setup_tri( setup_ctx, + get_vert(vertex_buffer, 0, stride), + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride) ); + } } break; case PIPE_PRIM_QUADS: for (i = 3; i < nr; i += 4) { - setup_tri( setup_ctx, - get_vert(vertex_buffer, i-3, stride), - get_vert(vertex_buffer, i-2, stride), - get_vert(vertex_buffer, i-0, stride)); - - setup_tri( setup_ctx, - get_vert(vertex_buffer, i-2, stride), - get_vert(vertex_buffer, i-1, stride), - get_vert(vertex_buffer, i-0, stride)); + if (softpipe->rasterizer->flatshade_first) { + setup_tri( setup_ctx, + get_vert(vertex_buffer, i-2, stride), + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-3, stride) ); + setup_tri( setup_ctx, + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride), + get_vert(vertex_buffer, i-3, stride) ); + } + else { + setup_tri( setup_ctx, + get_vert(vertex_buffer, i-3, stride), + get_vert(vertex_buffer, i-2, stride), + get_vert(vertex_buffer, i-0, stride) ); + setup_tri( setup_ctx, + get_vert(vertex_buffer, i-2, stride), + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-0, stride) ); + } } break; + case PIPE_PRIM_QUAD_STRIP: for (i = 3; i < nr; i += 2) { - setup_tri( setup_ctx, - get_vert(vertex_buffer, i-3, stride), - get_vert(vertex_buffer, i-2, stride), - get_vert(vertex_buffer, i-0, stride)); - - setup_tri( setup_ctx, - get_vert(vertex_buffer, i-1, stride), - get_vert(vertex_buffer, i-3, stride), - get_vert(vertex_buffer, i-0, stride)); + if (softpipe->rasterizer->flatshade_first) { + setup_tri( setup_ctx, + get_vert(vertex_buffer, i-0, stride), + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-3, stride) ); + setup_tri( setup_ctx, + get_vert(vertex_buffer, i-2, stride), + get_vert(vertex_buffer, i-0, stride), + get_vert(vertex_buffer, i-3, stride) ); + } + else { + setup_tri( setup_ctx, + get_vert(vertex_buffer, i-3, stride), + get_vert(vertex_buffer, i-2, stride), + get_vert(vertex_buffer, i-0, stride) ); + setup_tri( setup_ctx, + get_vert(vertex_buffer, i-1, stride), + get_vert(vertex_buffer, i-3, stride), + get_vert(vertex_buffer, i-0, stride) ); + } } break; @@ -427,12 +525,13 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr) /* Almost same as tri fan but the _first_ vertex specifies the flat * shading color. Note that the first polygon vertex is passed as * the last triangle vertex here. + * flatshade_first state makes no difference. */ for (i = 2; i < nr; i += 1) { setup_tri( setup_ctx, get_vert(vertex_buffer, i-1, stride), get_vert(vertex_buffer, i-0, stride), - get_vert(vertex_buffer, 0, stride)); + get_vert(vertex_buffer, 0, stride) ); } break; diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c index ca637a1d6a4..28f8d1a60ea 100644 --- a/src/gallium/drivers/softpipe/sp_quad_fs.c +++ b/src/gallium/drivers/softpipe/sp_quad_fs.c @@ -52,7 +52,7 @@ struct quad_shade_stage { struct quad_stage stage; /**< base class */ - struct tgsi_exec_machine machine; + struct tgsi_exec_machine *machine; struct tgsi_exec_vector *inputs, *outputs; }; @@ -73,7 +73,7 @@ shade_quad(struct quad_stage *qs, struct quad_header *quad) { struct quad_shade_stage *qss = quad_shade_stage( qs ); struct softpipe_context *softpipe = qs->softpipe; - struct tgsi_exec_machine *machine = &qss->machine; + struct tgsi_exec_machine *machine = qss->machine; boolean z_written; /* Consts do not require 16 byte alignment. */ @@ -146,7 +146,7 @@ shade_begin(struct quad_stage *qs) struct softpipe_context *softpipe = qs->softpipe; softpipe->fs->prepare( softpipe->fs, - &qss->machine, + qss->machine, (struct tgsi_sampler **) softpipe->tgsi.frag_samplers_list ); @@ -159,9 +159,8 @@ shade_destroy(struct quad_stage *qs) { struct quad_shade_stage *qss = (struct quad_shade_stage *) qs; - tgsi_exec_machine_free_data(&qss->machine); - FREE( qss->inputs ); - FREE( qss->outputs ); + tgsi_exec_machine_destroy(qss->machine); + FREE( qs ); } @@ -170,19 +169,24 @@ struct quad_stage * sp_quad_shade_stage( struct softpipe_context *softpipe ) { struct quad_shade_stage *qss = CALLOC_STRUCT(quad_shade_stage); - - /* allocate storage for program inputs/outputs, aligned to 16 bytes */ - qss->inputs = MALLOC(PIPE_MAX_ATTRIBS * sizeof(*qss->inputs) + 16); - qss->outputs = MALLOC(PIPE_MAX_ATTRIBS * sizeof(*qss->outputs) + 16); - qss->machine.Inputs = align16(qss->inputs); - qss->machine.Outputs = align16(qss->outputs); + if (!qss) + goto fail; qss->stage.softpipe = softpipe; qss->stage.begin = shade_begin; qss->stage.run = shade_quad; qss->stage.destroy = shade_destroy; - tgsi_exec_machine_init( &qss->machine ); + qss->machine = tgsi_exec_machine_create(); + if (!qss->machine) + goto fail; return &qss->stage; + +fail: + if (qss && qss->machine) + tgsi_exec_machine_destroy(qss->machine); + + FREE(qss); + return NULL; } diff --git a/src/gallium/drivers/softpipe/sp_quad_pipe.c b/src/gallium/drivers/softpipe/sp_quad_pipe.c index 892ef87ee9f..b5f69b74264 100644 --- a/src/gallium/drivers/softpipe/sp_quad_pipe.c +++ b/src/gallium/drivers/softpipe/sp_quad_pipe.c @@ -80,7 +80,7 @@ sp_build_quad_pipeline(struct softpipe_context *sp) sp_push_quad_first( sp, sp->quad[i].blend, i ); } - if (sp->depth_stencil->depth.occlusion_count) { + if (sp->active_query_count) { sp_push_quad_first( sp, sp->quad[i].occlusion, i ); } diff --git a/src/gallium/drivers/softpipe/sp_query.c b/src/gallium/drivers/softpipe/sp_query.c index 93dab236d66..379cf4ad064 100644 --- a/src/gallium/drivers/softpipe/sp_query.c +++ b/src/gallium/drivers/softpipe/sp_query.c @@ -34,6 +34,7 @@ #include "util/u_memory.h" #include "sp_context.h" #include "sp_query.h" +#include "sp_state.h" struct softpipe_query { uint64_t start; @@ -69,6 +70,8 @@ softpipe_begin_query(struct pipe_context *pipe, struct pipe_query *q) struct softpipe_query *sq = softpipe_query(q); sq->start = softpipe->occlusion_count; + softpipe->active_query_count++; + softpipe->dirty |= SP_NEW_QUERY; } @@ -78,7 +81,9 @@ softpipe_end_query(struct pipe_context *pipe, struct pipe_query *q) struct softpipe_context *softpipe = softpipe_context( pipe ); struct softpipe_query *sq = softpipe_query(q); + softpipe->active_query_count--; sq->end = softpipe->occlusion_count; + softpipe->dirty |= SP_NEW_QUERY; } diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c index cc39d33ede6..6178c4ac7e3 100644 --- a/src/gallium/drivers/softpipe/sp_screen.c +++ b/src/gallium/drivers/softpipe/sp_screen.c @@ -82,13 +82,15 @@ softpipe_get_param(struct pipe_screen *screen, int param) case PIPE_CAP_TEXTURE_SHADOW_MAP: return 1; case PIPE_CAP_MAX_TEXTURE_2D_LEVELS: - return 12; /* max 2Kx2K */ + return 13; /* max 4Kx4K */ case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: return 8; /* max 128x128x128 */ case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS: - return 12; /* max 2Kx2K */ + return 13; /* max 4Kx4K */ case PIPE_CAP_TGSI_CONT_SUPPORTED: return 1; + case PIPE_CAP_BLEND_EQUATION_SEPARATE: + return 1; default: return 0; } @@ -108,7 +110,7 @@ softpipe_get_paramf(struct pipe_screen *screen, int param) case PIPE_CAP_MAX_POINT_WIDTH_AA: return 255.0; /* arbitrary */ case PIPE_CAP_MAX_TEXTURE_ANISOTROPY: - return 0.0; + return 16.0; /* not actually signficant at this time */ case PIPE_CAP_MAX_TEXTURE_LOD_BIAS: return 16.0; /* arbitrary */ default: diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c index 1eb23cd6b6f..de3ae3c3696 100644 --- a/src/gallium/drivers/softpipe/sp_setup.c +++ b/src/gallium/drivers/softpipe/sp_setup.c @@ -507,6 +507,8 @@ static void print_vertex(const struct setup_context *setup, #endif /** + * Sort the vertices from top to bottom order, setting up the triangle + * edge fields (ebot, emaj, etop). * \return FALSE if coords are inf/nan (cull the tri), TRUE otherwise */ static boolean setup_sort_vertices( struct setup_context *setup, @@ -1051,7 +1053,10 @@ setup_line_coefficients(struct setup_context *setup, float area; /* use setup->vmin, vmax to point to vertices */ - setup->vprovoke = v1; + if (softpipe->rasterizer->flatshade_first) + setup->vprovoke = v0; + else + setup->vprovoke = v1; setup->vmin = v0; setup->vmax = v1; diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 5de358dae93..a1d3bade27a 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -705,15 +705,18 @@ get_texel(const struct tgsi_sampler *tgsi_sampler, * Compare texcoord 'p' (aka R) against texture value 'rgba[0]' * When we sampled the depth texture, the depth value was put into all * RGBA channels. We look at the red channel here. + * \param rgba quad of (depth) texel values + * \param p texture 'P' components for four pixels in quad + * \param j which pixel in the quad to test [0..3] */ static INLINE void -shadow_compare(uint compare_func, +shadow_compare(const struct pipe_sampler_state *sampler, float rgba[NUM_CHANNELS][QUAD_SIZE], const float p[QUAD_SIZE], uint j) { int k; - switch (compare_func) { + switch (sampler->compare_func) { case PIPE_FUNC_LESS: k = p[j] < rgba[0][j]; break; @@ -751,6 +754,78 @@ shadow_compare(uint compare_func, /** + * As above, but do four z/texture comparisons. + */ +static INLINE void +shadow_compare4(const struct pipe_sampler_state *sampler, + float rgba[NUM_CHANNELS][QUAD_SIZE], + const float p[QUAD_SIZE]) +{ + int j, k0, k1, k2, k3; + float val; + + /* compare four texcoords vs. four texture samples */ + switch (sampler->compare_func) { + case PIPE_FUNC_LESS: + k0 = p[0] < rgba[0][0]; + k1 = p[1] < rgba[0][1]; + k2 = p[2] < rgba[0][2]; + k3 = p[3] < rgba[0][3]; + break; + case PIPE_FUNC_LEQUAL: + k0 = p[0] <= rgba[0][0]; + k1 = p[1] <= rgba[0][1]; + k2 = p[2] <= rgba[0][2]; + k3 = p[3] <= rgba[0][3]; + break; + case PIPE_FUNC_GREATER: + k0 = p[0] > rgba[0][0]; + k1 = p[1] > rgba[0][1]; + k2 = p[2] > rgba[0][2]; + k3 = p[3] > rgba[0][3]; + break; + case PIPE_FUNC_GEQUAL: + k0 = p[0] >= rgba[0][0]; + k1 = p[1] >= rgba[0][1]; + k2 = p[2] >= rgba[0][2]; + k3 = p[3] >= rgba[0][3]; + break; + case PIPE_FUNC_EQUAL: + k0 = p[0] == rgba[0][0]; + k1 = p[1] == rgba[0][1]; + k2 = p[2] == rgba[0][2]; + k3 = p[3] == rgba[0][3]; + break; + case PIPE_FUNC_NOTEQUAL: + k0 = p[0] != rgba[0][0]; + k1 = p[1] != rgba[0][1]; + k2 = p[2] != rgba[0][2]; + k3 = p[3] != rgba[0][3]; + break; + case PIPE_FUNC_ALWAYS: + k0 = k1 = k2 = k3 = 1; + break; + case PIPE_FUNC_NEVER: + k0 = k1 = k2 = k3 = 0; + break; + default: + k0 = k1 = k2 = k3 = 0; + assert(0); + break; + } + + /* convert four pass/fail values to an intensity in [0,1] */ + val = 0.25F * (k0 + k1 + k2 + k3); + + /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */ + for (j = 0; j < 4; j++) { + rgba[0][j] = rgba[1][j] = rgba[2][j] = val; + rgba[3][j] = 1.0F; + } +} + + +/** * Common code for sampling 1D/2D/cube textures. * Could probably extend for 3D... */ @@ -769,7 +844,6 @@ sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler, const uint unit = samp->unit; const struct pipe_texture *texture = sp->texture[unit]; const struct pipe_sampler_state *sampler = sp->sampler[unit]; - const uint compare_func = sampler->compare_func; unsigned level0, level1, j, imgFilter; int width, height; float levelBlend; @@ -794,7 +868,7 @@ sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler, for (j = 0; j < QUAD_SIZE; j++) { get_texel(tgsi_sampler, faces[j], level0, x[j], y[j], 0, rgba, j); if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { - shadow_compare(compare_func, rgba, p, j); + shadow_compare(sampler, rgba, p, j); } if (level0 != level1) { @@ -806,7 +880,7 @@ sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler, get_texel(tgsi_sampler, faces[j], level1, x[j], y[j], 0, rgba2, j); if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){ - shadow_compare(compare_func, rgba2, p, j); + shadow_compare(sampler, rgba2, p, j); } for (c = 0; c < NUM_CHANNELS; c++) { @@ -833,10 +907,7 @@ sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler, get_texel(tgsi_sampler, faces[j], level0, x0[j], y1[j], 0, tx, 2); get_texel(tgsi_sampler, faces[j], level0, x1[j], y1[j], 0, tx, 3); if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { - shadow_compare(compare_func, tx, p, 0); - shadow_compare(compare_func, tx, p, 1); - shadow_compare(compare_func, tx, p, 2); - shadow_compare(compare_func, tx, p, 3); + shadow_compare4(sampler, tx, p); } /* interpolate R, G, B, A */ @@ -858,10 +929,7 @@ sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler, get_texel(tgsi_sampler, faces[j], level1, x0[j], y1[j], 0, tx, 2); get_texel(tgsi_sampler, faces[j], level1, x1[j], y1[j], 0, tx, 3); if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){ - shadow_compare(compare_func, tx, p, 0); - shadow_compare(compare_func, tx, p, 1); - shadow_compare(compare_func, tx, p, 2); - shadow_compare(compare_func, tx, p, 3); + shadow_compare4(sampler, tx, p); } /* interpolate R, G, B, A */ @@ -1076,7 +1144,6 @@ sp_get_samples_rect(const struct tgsi_sampler *tgsi_sampler, const struct pipe_texture *texture = sp->texture[unit]; const struct pipe_sampler_state *sampler = sp->sampler[unit]; const uint face = 0; - const uint compare_func = sampler->compare_func; unsigned level0, level1, j, imgFilter; int width, height; float levelBlend; @@ -1101,7 +1168,7 @@ sp_get_samples_rect(const struct tgsi_sampler *tgsi_sampler, for (j = 0; j < QUAD_SIZE; j++) { get_texel(tgsi_sampler, face, level0, x[j], y[j], 0, rgba, j); if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { - shadow_compare(compare_func, rgba, p, j); + shadow_compare(sampler, rgba, p, j); } } } @@ -1121,10 +1188,7 @@ sp_get_samples_rect(const struct tgsi_sampler *tgsi_sampler, get_texel(tgsi_sampler, face, level0, x0[j], y1[j], 0, tx, 2); get_texel(tgsi_sampler, face, level0, x1[j], y1[j], 0, tx, 3); if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { - shadow_compare(compare_func, tx, p, 0); - shadow_compare(compare_func, tx, p, 1); - shadow_compare(compare_func, tx, p, 2); - shadow_compare(compare_func, tx, p, 3); + shadow_compare4(sampler, tx, p); } for (c = 0; c < 4; c++) { rgba[c][j] = lerp_2d(xw[j], yw[j], diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c index 7a533dad9f0..70f09324311 100644 --- a/src/gallium/drivers/softpipe/sp_texture.c +++ b/src/gallium/drivers/softpipe/sp_texture.c @@ -48,11 +48,6 @@ /* Simple, maximally packed layout. */ -static unsigned minify( unsigned d ) -{ - return MAX2(1, d>>1); -} - /* Conventional allocation path for non-display textures: */ @@ -100,6 +95,7 @@ softpipe_displaytarget_layout(struct pipe_screen *screen, { unsigned usage = (PIPE_BUFFER_USAGE_CPU_READ_WRITE | PIPE_BUFFER_USAGE_GPU_READ_WRITE); + unsigned tex_usage = spt->base.tex_usage; spt->base.nblocksx[0] = pf_get_nblocksx(&spt->base.block, spt->base.width[0]); spt->base.nblocksy[0] = pf_get_nblocksy(&spt->base.block, spt->base.height[0]); @@ -109,6 +105,7 @@ softpipe_displaytarget_layout(struct pipe_screen *screen, spt->base.height[0], spt->base.format, usage, + tex_usage, &spt->stride[0]); return spt->buffer != NULL; @@ -130,7 +127,8 @@ softpipe_texture_create(struct pipe_screen *screen, pipe_reference_init(&spt->base.reference, 1); spt->base.screen = screen; - if (spt->base.tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET) { + if (spt->base.tex_usage & (PIPE_TEXTURE_USAGE_DISPLAY_TARGET | + PIPE_TEXTURE_USAGE_PRIMARY)) { if (!softpipe_displaytarget_layout(screen, spt)) goto fail; } @@ -224,12 +222,6 @@ softpipe_get_tex_surface(struct pipe_screen *screen, if (ps->usage & PIPE_BUFFER_USAGE_GPU_READ) ps->usage |= PIPE_BUFFER_USAGE_CPU_READ; - if (ps->usage & (PIPE_BUFFER_USAGE_CPU_WRITE | - PIPE_BUFFER_USAGE_GPU_WRITE)) { - /* Mark the surface as dirty. The tile cache will look for this. */ - spt->modified = TRUE; - } - ps->face = face; ps->level = level; ps->zslice = zslice; @@ -376,6 +368,11 @@ softpipe_transfer_unmap(struct pipe_screen *screen, spt = softpipe_texture(transfer->texture); pipe_buffer_unmap( screen, spt->buffer ); + + if (transfer->usage != PIPE_TRANSFER_READ) { + /* Mark the texture as dirty to expire the tile caches. */ + spt->modified = TRUE; + } } diff --git a/src/gallium/drivers/trace/Makefile b/src/gallium/drivers/trace/Makefile index e087db169aa..dd6831c70ab 100644 --- a/src/gallium/drivers/trace/Makefile +++ b/src/gallium/drivers/trace/Makefile @@ -7,8 +7,11 @@ C_SOURCES = \ tr_buffer.c \ tr_context.c \ tr_dump.c \ + tr_dump_state.c \ tr_screen.c \ tr_state.c \ + tr_rbug.c \ + tr_drm.c \ tr_texture.c include ../../Makefile.template diff --git a/src/gallium/drivers/trace/README b/src/gallium/drivers/trace/README index 73dce20372e..1000c31e49a 100644 --- a/src/gallium/drivers/trace/README +++ b/src/gallium/drivers/trace/README @@ -3,7 +3,8 @@ = About = -This directory contains a Gallium3D pipe driver which traces all incoming calls. +This directory contains a Gallium3D debugger pipe driver. +It can traces all incoming calls and/or provide remote debugging functionality. = Build Instructions = @@ -23,7 +24,9 @@ ensure the right libGL.so is being picked by doing ldd progs/trivial/tri -and then try running +== Traceing == + +For traceing then do export XMESA_TRACE=y GALLIUM_TRACE=tri.trace progs/trivial/tri @@ -32,6 +35,16 @@ which should create a tri.trace file, which is an XML file. You can view copying trace.xsl to the same directory, and opening with a XSLT capable browser such as Firefox or Internet Explorer. +== Remote debugging == + +For remote debugging + + export XMESA_TRACE=y + GALLIUM_RBUG=true progs/trivial/tri + +which should open gallium remote debugging session. While the program is running +you can launch the small remote debugging application from progs/rbug. More +information is in that directory. = Integrating = @@ -62,3 +75,4 @@ trace_screen with real_screen when creating them. -- Jose Fonseca <jrfonseca@tungstengraphics.com> +Jakob Bornecrantz <jakob@vmware.com> diff --git a/src/gallium/drivers/trace/SConscript b/src/gallium/drivers/trace/SConscript index 4215215d1a1..c1675d1c165 100644 --- a/src/gallium/drivers/trace/SConscript +++ b/src/gallium/drivers/trace/SConscript @@ -7,9 +7,12 @@ trace = env.ConvenienceLibrary( source = [ 'tr_buffer.c', 'tr_context.c', + 'tr_drm.c', 'tr_dump.c', + 'tr_dump_state.c', 'tr_screen.c', 'tr_state.c', + 'tr_rbug.c', 'tr_texture.c', ]) diff --git a/src/gallium/drivers/trace/tr_context.c b/src/gallium/drivers/trace/tr_context.c index 47280459a75..ae0af4d0557 100644 --- a/src/gallium/drivers/trace/tr_context.c +++ b/src/gallium/drivers/trace/tr_context.c @@ -31,11 +31,11 @@ #include "pipe/p_screen.h" #include "tr_dump.h" +#include "tr_dump_state.h" #include "tr_state.h" #include "tr_buffer.h" #include "tr_screen.h" #include "tr_texture.h" -#include "tr_context.h" static INLINE struct pipe_buffer * @@ -113,6 +113,65 @@ trace_context_set_edgeflags(struct pipe_context *_pipe, } +static INLINE void +trace_context_draw_block(struct trace_context *tr_ctx, int flag) +{ + int k; + + pipe_mutex_lock(tr_ctx->draw_mutex); + + if (tr_ctx->draw_blocker & flag) { + tr_ctx->draw_blocked |= flag; + } else if ((tr_ctx->draw_rule.blocker & flag) && + (tr_ctx->draw_blocker & 4)) { + boolean block = FALSE; + debug_printf("%s (%lu %lu) (%lu %lu) (%lu %u) (%lu %u)\n", __FUNCTION__, + tr_ctx->draw_rule.fs, tr_ctx->curr.fs, + tr_ctx->draw_rule.vs, tr_ctx->curr.vs, + tr_ctx->draw_rule.surf, 0, + tr_ctx->draw_rule.tex, 0); + if (tr_ctx->draw_rule.fs && + tr_ctx->draw_rule.fs == tr_ctx->curr.fs) + block = TRUE; + if (tr_ctx->draw_rule.vs && + tr_ctx->draw_rule.vs == tr_ctx->curr.vs) + block = TRUE; + if (tr_ctx->draw_rule.surf && + tr_ctx->draw_rule.surf == tr_ctx->curr.zsbuf) + block = TRUE; + if (tr_ctx->draw_rule.surf) + for (k = 0; k < tr_ctx->curr.nr_cbufs; k++) + if (tr_ctx->draw_rule.surf == tr_ctx->curr.cbufs[k]) + block = TRUE; + if (tr_ctx->draw_rule.tex) + for (k = 0; k < tr_ctx->curr.num_texs; k++) + if (tr_ctx->draw_rule.tex == tr_ctx->curr.tex[k]) + block = TRUE; + + if (block) + tr_ctx->draw_blocked |= (flag | 4); + } + + if (tr_ctx->draw_blocked) + trace_rbug_notify_draw_blocked(tr_ctx); + + /* wait for rbug to clear the blocked flag */ + while (tr_ctx->draw_blocked & flag) { + tr_ctx->draw_blocked |= flag; +#ifdef PIPE_THREAD_HAVE_CONDVAR + pipe_condvar_wait(tr_ctx->draw_cond, tr_ctx->draw_mutex); +#else + pipe_mutex_unlock(tr_ctx->draw_mutex); +#ifdef PIPE_SUBSYSTEM_WINDOWS_USER + Sleep(1); +#endif + pipe_mutex_lock(tr_ctx->draw_mutex); +#endif + } + + pipe_mutex_unlock(tr_ctx->draw_mutex); +} + static INLINE boolean trace_context_draw_arrays(struct pipe_context *_pipe, unsigned mode, unsigned start, unsigned count) @@ -121,6 +180,11 @@ trace_context_draw_arrays(struct pipe_context *_pipe, struct pipe_context *pipe = tr_ctx->pipe; boolean result; + if (tr_ctx->curr.fs->disabled || tr_ctx->curr.vs->disabled) + return 0; + + trace_context_draw_block(tr_ctx, 1); + trace_dump_call_begin("pipe_context", "draw_arrays"); trace_dump_arg(ptr, pipe); @@ -134,6 +198,8 @@ trace_context_draw_arrays(struct pipe_context *_pipe, trace_dump_call_end(); + trace_context_draw_block(tr_ctx, 2); + return result; } @@ -150,6 +216,11 @@ trace_context_draw_elements(struct pipe_context *_pipe, struct pipe_buffer *indexBuffer = tr_buf->buffer; boolean result; + if (tr_ctx->curr.fs->disabled || tr_ctx->curr.vs->disabled) + return 0; + + trace_context_draw_block(tr_ctx, 1); + trace_screen_user_buffer_update(_pipe->screen, indexBuffer); trace_dump_call_begin("pipe_context", "draw_elements"); @@ -167,6 +238,8 @@ trace_context_draw_elements(struct pipe_context *_pipe, trace_dump_call_end(); + trace_context_draw_block(tr_ctx, 2); + return result; } @@ -187,6 +260,11 @@ trace_context_draw_range_elements(struct pipe_context *_pipe, struct pipe_buffer *indexBuffer = tr_buf->buffer; boolean result; + if (tr_ctx->curr.fs->disabled || tr_ctx->curr.vs->disabled) + return 0; + + trace_context_draw_block(tr_ctx, 1); + trace_screen_user_buffer_update(_pipe->screen, indexBuffer); trace_dump_call_begin("pipe_context", "draw_range_elements"); @@ -209,6 +287,8 @@ trace_context_draw_range_elements(struct pipe_context *_pipe, trace_dump_call_end(); + trace_context_draw_block(tr_ctx, 2); + return result; } @@ -573,23 +653,32 @@ trace_context_create_fs_state(struct pipe_context *_pipe, trace_dump_call_end(); + result = trace_shader_create(tr_ctx, state, result, TRACE_SHADER_FRAGMENT); + return result; } static INLINE void trace_context_bind_fs_state(struct pipe_context *_pipe, - void *state) + void *_state) { struct trace_context *tr_ctx = trace_context(_pipe); + struct trace_shader *tr_shdr = trace_shader(_state); struct pipe_context *pipe = tr_ctx->pipe; + void *state = tr_shdr ? tr_shdr->state : NULL; trace_dump_call_begin("pipe_context", "bind_fs_state"); trace_dump_arg(ptr, pipe); trace_dump_arg(ptr, state); - pipe->bind_fs_state(pipe, state);; + tr_ctx->curr.fs = tr_shdr; + + if (tr_shdr && tr_shdr->replaced) + state = tr_shdr->replaced; + + pipe->bind_fs_state(pipe, state); trace_dump_call_end(); } @@ -597,19 +686,23 @@ trace_context_bind_fs_state(struct pipe_context *_pipe, static INLINE void trace_context_delete_fs_state(struct pipe_context *_pipe, - void *state) + void *_state) { struct trace_context *tr_ctx = trace_context(_pipe); + struct trace_shader *tr_shdr = trace_shader(_state); struct pipe_context *pipe = tr_ctx->pipe; + void *state = tr_shdr->state; trace_dump_call_begin("pipe_context", "delete_fs_state"); trace_dump_arg(ptr, pipe); trace_dump_arg(ptr, state); - pipe->delete_fs_state(pipe, state);; + pipe->delete_fs_state(pipe, state); trace_dump_call_end(); + + trace_shader_destroy(tr_ctx, tr_shdr); } @@ -626,28 +719,37 @@ trace_context_create_vs_state(struct pipe_context *_pipe, trace_dump_arg(ptr, pipe); trace_dump_arg(shader_state, state); - result = pipe->create_vs_state(pipe, state);; + result = pipe->create_vs_state(pipe, state); trace_dump_ret(ptr, result); trace_dump_call_end(); + result = trace_shader_create(tr_ctx, state, result, TRACE_SHADER_VERTEX); + return result; } static INLINE void trace_context_bind_vs_state(struct pipe_context *_pipe, - void *state) + void *_state) { struct trace_context *tr_ctx = trace_context(_pipe); + struct trace_shader *tr_shdr = trace_shader(_state); struct pipe_context *pipe = tr_ctx->pipe; + void *state = tr_shdr ? tr_shdr->state : NULL; trace_dump_call_begin("pipe_context", "bind_vs_state"); trace_dump_arg(ptr, pipe); trace_dump_arg(ptr, state); + tr_ctx->curr.vs = tr_shdr; + + if (tr_shdr && tr_shdr->replaced) + state = tr_shdr->replaced; + pipe->bind_vs_state(pipe, state);; trace_dump_call_end(); @@ -656,10 +758,12 @@ trace_context_bind_vs_state(struct pipe_context *_pipe, static INLINE void trace_context_delete_vs_state(struct pipe_context *_pipe, - void *state) + void *_state) { struct trace_context *tr_ctx = trace_context(_pipe); + struct trace_shader *tr_shdr = trace_shader(_state); struct pipe_context *pipe = tr_ctx->pipe; + void *state = tr_shdr->state; trace_dump_call_begin("pipe_context", "delete_vs_state"); @@ -669,6 +773,8 @@ trace_context_delete_vs_state(struct pipe_context *_pipe, pipe->delete_vs_state(pipe, state);; trace_dump_call_end(); + + trace_shader_destroy(tr_ctx, tr_shdr); } @@ -747,6 +853,19 @@ trace_context_set_framebuffer_state(struct pipe_context *_pipe, struct pipe_framebuffer_state unwrapped_state; unsigned i; + { + tr_ctx->curr.nr_cbufs = state->nr_cbufs; + for (i = 0; i < state->nr_cbufs; i++) + if (state->cbufs[i]) + tr_ctx->curr.cbufs[i] = trace_texture(state->cbufs[i]->texture); + else + tr_ctx->curr.cbufs[i] = NULL; + if (state->zsbuf) + tr_ctx->curr.zsbuf = trace_texture(state->zsbuf->texture); + else + tr_ctx->curr.zsbuf = NULL; + } + /* Unwrap the input state */ memcpy(&unwrapped_state, state, sizeof(unwrapped_state)); for(i = 0; i < state->nr_cbufs; ++i) @@ -827,12 +946,17 @@ trace_context_set_sampler_textures(struct pipe_context *_pipe, struct pipe_texture **textures) { struct trace_context *tr_ctx = trace_context(_pipe); + struct trace_texture *tr_tex; struct pipe_context *pipe = tr_ctx->pipe; struct pipe_texture *unwrapped_textures[PIPE_MAX_SAMPLERS]; unsigned i; - for(i = 0; i < num_textures; ++i) - unwrapped_textures[i] = trace_texture_unwrap(tr_ctx, textures[i]); + tr_ctx->curr.num_texs = num_textures; + for(i = 0; i < num_textures; ++i) { + tr_tex = trace_texture(textures[i]); + tr_ctx->curr.tex[i] = tr_tex; + unwrapped_textures[i] = tr_tex ? tr_tex->texture : NULL; + } textures = unwrapped_textures; trace_dump_call_begin("pipe_context", "set_sampler_textures"); @@ -1078,6 +1202,12 @@ trace_is_buffer_referenced( struct pipe_context *_pipe, return referenced; } +static const struct debug_named_value rbug_blocker_flags[] = { + {"before", 1}, + {"after", 2}, + {NULL, 0}, +}; + struct pipe_context * trace_context_create(struct pipe_screen *_screen, struct pipe_context *pipe) @@ -1089,7 +1219,7 @@ trace_context_create(struct pipe_screen *_screen, if(!pipe) goto error1; - if(!trace_dump_trace_enabled()) + if(!trace_enabled()) goto error1; tr_scr = trace_screen(_screen); @@ -1099,6 +1229,14 @@ trace_context_create(struct pipe_screen *_screen, if(!tr_ctx) goto error1; + tr_ctx->draw_blocker = debug_get_flags_option("RBUG_BLOCK", + rbug_blocker_flags, + 0); + pipe_mutex_init(tr_ctx->draw_mutex); + pipe_condvar_init(tr_ctx->draw_cond); + pipe_mutex_init(tr_ctx->list_mutex); + make_empty_list(&tr_ctx->shaders); + tr_ctx->base.winsys = _screen->winsys; tr_ctx->base.screen = _screen; tr_ctx->base.destroy = trace_context_destroy; @@ -1139,8 +1277,10 @@ trace_context_create(struct pipe_screen *_screen, tr_ctx->base.set_sampler_textures = trace_context_set_sampler_textures; tr_ctx->base.set_vertex_buffers = trace_context_set_vertex_buffers; tr_ctx->base.set_vertex_elements = trace_context_set_vertex_elements; - tr_ctx->base.surface_copy = trace_context_surface_copy; - tr_ctx->base.surface_fill = trace_context_surface_fill; + if (pipe->surface_copy) + tr_ctx->base.surface_copy = trace_context_surface_copy; + if (pipe->surface_fill) + tr_ctx->base.surface_fill = trace_context_surface_fill; tr_ctx->base.clear = trace_context_clear; tr_ctx->base.flush = trace_context_flush; tr_ctx->base.is_texture_referenced = trace_is_texture_referenced; diff --git a/src/gallium/drivers/trace/tr_context.h b/src/gallium/drivers/trace/tr_context.h index 2512a0a2328..6febe4b4114 100644 --- a/src/gallium/drivers/trace/tr_context.h +++ b/src/gallium/drivers/trace/tr_context.h @@ -33,6 +33,7 @@ #include "util/u_debug.h" #include "pipe/p_context.h" +#include "tr_screen.h" #ifdef __cplusplus extern "C" { @@ -45,7 +46,41 @@ struct trace_context struct pipe_context *pipe; + /* current state */ + struct { + struct trace_shader *fs; + struct trace_shader *vs; + + struct trace_texture *tex[PIPE_MAX_SAMPLERS]; + unsigned num_texs; + + unsigned nr_cbufs; + struct trace_texture *cbufs[PIPE_MAX_COLOR_BUFS]; + struct trace_texture *zsbuf; + } curr; + + struct { + struct trace_shader *fs; + struct trace_shader *vs; + + struct trace_texture *tex; + struct trace_texture *surf; + + int blocker; + } draw_rule; + unsigned draw_num_rules; + pipe_condvar draw_cond; + pipe_mutex draw_mutex; + int draw_blocker; + int draw_blocked; + + /* for list on screen */ struct tr_list list; + + /* list of state objects */ + pipe_mutex list_mutex; + unsigned num_shaders; + struct tr_list shaders; }; @@ -62,6 +97,9 @@ struct pipe_context * trace_context_create(struct pipe_screen *screen, struct pipe_context *pipe); +void +trace_rbug_notify_draw_blocked(struct trace_context *tr_ctx); + #ifdef __cplusplus } diff --git a/src/gallium/drivers/trace/tr_drm.c b/src/gallium/drivers/trace/tr_drm.c new file mode 100644 index 00000000000..781ca5d3bc0 --- /dev/null +++ b/src/gallium/drivers/trace/tr_drm.c @@ -0,0 +1,186 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "state_tracker/drm_api.h" + +#include "util/u_memory.h" +#include "trace/tr_drm.h" +#include "trace/tr_screen.h" +#include "trace/tr_context.h" +#include "trace/tr_buffer.h" +#include "trace/tr_texture.h" + +struct trace_drm_api +{ + struct drm_api base; + + struct drm_api *api; +}; + +static INLINE struct trace_drm_api * +trace_drm_api(struct drm_api *_api) +{ + return (struct trace_drm_api *)_api; +} + +static struct pipe_screen * +trace_drm_create_screen(struct drm_api *_api, int fd, + struct drm_create_screen_arg *arg) +{ + struct trace_drm_api *tr_api = trace_drm_api(_api); + struct drm_api *api = tr_api->api; + struct pipe_screen *screen; + + /* TODO trace call */ + + if (arg && arg->mode != DRM_CREATE_NORMAL) + return NULL; + + screen = api->create_screen(api, fd, arg); + + return trace_screen_create(screen); +} + +static struct pipe_context * +trace_drm_create_context(struct drm_api *_api, + struct pipe_screen *_screen) +{ + struct trace_screen *tr_screen = trace_screen(_screen); + struct trace_drm_api *tr_api = trace_drm_api(_api); + struct pipe_screen *screen = tr_screen->screen; + struct drm_api *api = tr_api->api; + struct pipe_context *pipe; + + /* TODO trace call */ + + pipe = api->create_context(api, screen); + + pipe = trace_context_create(_screen, pipe); + + return pipe; +} + +static struct pipe_texture * +trace_drm_texture_from_shared_handle(struct drm_api *_api, + struct pipe_screen *_screen, + struct pipe_texture *templ, + const char *name, + unsigned stride, + unsigned handle) +{ + struct trace_screen *tr_screen = trace_screen(_screen); + struct trace_drm_api *tr_api = trace_drm_api(_api); + struct pipe_screen *screen = tr_screen->screen; + struct drm_api *api = tr_api->api; + struct pipe_texture *result; + + /* TODO trace call */ + + result = api->texture_from_shared_handle(api, screen, templ, name, stride, handle); + + result = trace_texture_create(trace_screen(_screen), result); + + return result; +} + +static boolean +trace_drm_shared_handle_from_texture(struct drm_api *_api, + struct pipe_screen *_screen, + struct pipe_texture *_texture, + unsigned *stride, + unsigned *handle) +{ + struct trace_screen *tr_screen = trace_screen(_screen); + struct trace_texture *tr_texture = trace_texture(_texture); + struct trace_drm_api *tr_api = trace_drm_api(_api); + struct pipe_screen *screen = tr_screen->screen; + struct pipe_texture *texture = tr_texture->texture; + struct drm_api *api = tr_api->api; + + /* TODO trace call */ + + return api->shared_handle_from_texture(api, screen, texture, stride, handle); +} + +static boolean +trace_drm_local_handle_from_texture(struct drm_api *_api, + struct pipe_screen *_screen, + struct pipe_texture *_texture, + unsigned *stride, + unsigned *handle) +{ + struct trace_screen *tr_screen = trace_screen(_screen); + struct trace_texture *tr_texture = trace_texture(_texture); + struct trace_drm_api *tr_api = trace_drm_api(_api); + struct pipe_screen *screen = tr_screen->screen; + struct pipe_texture *texture = tr_texture->texture; + struct drm_api *api = tr_api->api; + + /* TODO trace call */ + + return api->local_handle_from_texture(api, screen, texture, stride, handle); +} + +static void +trace_drm_destroy(struct drm_api *_api) +{ + struct trace_drm_api *tr_api = trace_drm_api(_api); + struct drm_api *api = tr_api->api; + api->destroy(api); + + free(tr_api); +} + +struct drm_api * +trace_drm_create(struct drm_api *api) +{ + struct trace_drm_api *tr_api; + + if (!api) + goto error; + + if (!trace_enabled()) + goto error; + + tr_api = CALLOC_STRUCT(trace_drm_api); + + if (!tr_api) + goto error; + + tr_api->base.create_screen = trace_drm_create_screen; + tr_api->base.create_context = trace_drm_create_context; + tr_api->base.texture_from_shared_handle = trace_drm_texture_from_shared_handle; + tr_api->base.shared_handle_from_texture = trace_drm_shared_handle_from_texture; + tr_api->base.local_handle_from_texture = trace_drm_local_handle_from_texture; + tr_api->base.destroy = trace_drm_destroy; + tr_api->api = api; + + return &tr_api->base; + +error: + return api; +} diff --git a/src/gallium/drivers/trace/tr_drm.h b/src/gallium/drivers/trace/tr_drm.h new file mode 100644 index 00000000000..845c66a32aa --- /dev/null +++ b/src/gallium/drivers/trace/tr_drm.h @@ -0,0 +1,35 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef TR_DRM_H +#define TR_DRM_H + +struct drm_api; + +struct drm_api* trace_drm_create(struct drm_api *api); + +#endif /* ID_DRM_H */ diff --git a/src/gallium/drivers/trace/tr_dump_state.c b/src/gallium/drivers/trace/tr_dump_state.c new file mode 100644 index 00000000000..bcf6751af4f --- /dev/null +++ b/src/gallium/drivers/trace/tr_dump_state.c @@ -0,0 +1,549 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "pipe/p_compiler.h" +#include "util/u_memory.h" +#include "tgsi/tgsi_dump.h" + +#include "tr_dump.h" +#include "tr_dump_state.h" + + +void trace_dump_format(enum pipe_format format) +{ + if (!trace_dumping_enabled_locked()) + return; + + trace_dump_enum(pf_name(format) ); +} + + +void trace_dump_block(const struct pipe_format_block *block) +{ + if (!trace_dumping_enabled_locked()) + return; + + trace_dump_struct_begin("pipe_format_block"); + trace_dump_member(uint, block, size); + trace_dump_member(uint, block, width); + trace_dump_member(uint, block, height); + trace_dump_struct_end(); +} + + +static void trace_dump_reference(const struct pipe_reference *reference) +{ + if (!trace_dumping_enabled_locked()) + return; + + trace_dump_struct_begin("pipe_reference"); + trace_dump_member(int, &reference->count, count); + trace_dump_struct_end(); +} + + +void trace_dump_template(const struct pipe_texture *templat) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!templat) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_texture"); + + trace_dump_member(int, templat, target); + trace_dump_member(format, templat, format); + + trace_dump_member_begin("width"); + trace_dump_array(uint, templat->width, 1); + trace_dump_member_end(); + + trace_dump_member_begin("height"); + trace_dump_array(uint, templat->height, 1); + trace_dump_member_end(); + + trace_dump_member_begin("depth"); + trace_dump_array(uint, templat->depth, 1); + trace_dump_member_end(); + + trace_dump_member_begin("block"); + trace_dump_block(&templat->block); + trace_dump_member_end(); + + trace_dump_member(uint, templat, last_level); + trace_dump_member(uint, templat, tex_usage); + + trace_dump_struct_end(); +} + + +void trace_dump_rasterizer_state(const struct pipe_rasterizer_state *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_rasterizer_state"); + + trace_dump_member(bool, state, flatshade); + trace_dump_member(bool, state, light_twoside); + trace_dump_member(uint, state, front_winding); + trace_dump_member(uint, state, cull_mode); + trace_dump_member(uint, state, fill_cw); + trace_dump_member(uint, state, fill_ccw); + trace_dump_member(bool, state, offset_cw); + trace_dump_member(bool, state, offset_ccw); + trace_dump_member(bool, state, scissor); + trace_dump_member(bool, state, poly_smooth); + trace_dump_member(bool, state, poly_stipple_enable); + trace_dump_member(bool, state, point_smooth); + trace_dump_member(bool, state, point_sprite); + trace_dump_member(bool, state, point_size_per_vertex); + trace_dump_member(bool, state, multisample); + trace_dump_member(bool, state, line_smooth); + trace_dump_member(bool, state, line_stipple_enable); + trace_dump_member(uint, state, line_stipple_factor); + trace_dump_member(uint, state, line_stipple_pattern); + trace_dump_member(bool, state, line_last_pixel); + trace_dump_member(bool, state, bypass_vs_clip_and_viewport); + trace_dump_member(bool, state, flatshade_first); + trace_dump_member(bool, state, gl_rasterization_rules); + + trace_dump_member(float, state, line_width); + trace_dump_member(float, state, point_size); + trace_dump_member(float, state, point_size_min); + trace_dump_member(float, state, point_size_max); + trace_dump_member(float, state, offset_units); + trace_dump_member(float, state, offset_scale); + + trace_dump_member_array(uint, state, sprite_coord_mode); + + trace_dump_struct_end(); +} + + +void trace_dump_poly_stipple(const struct pipe_poly_stipple *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_poly_stipple"); + + trace_dump_member_begin("stipple"); + trace_dump_array(uint, + state->stipple, + Elements(state->stipple)); + trace_dump_member_end(); + + trace_dump_struct_end(); +} + + +void trace_dump_viewport_state(const struct pipe_viewport_state *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_viewport_state"); + + trace_dump_member_array(float, state, scale); + trace_dump_member_array(float, state, translate); + + trace_dump_struct_end(); +} + + +void trace_dump_scissor_state(const struct pipe_scissor_state *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_scissor_state"); + + trace_dump_member(uint, state, minx); + trace_dump_member(uint, state, miny); + trace_dump_member(uint, state, maxx); + trace_dump_member(uint, state, maxy); + + trace_dump_struct_end(); +} + + +void trace_dump_clip_state(const struct pipe_clip_state *state) +{ + unsigned i; + + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_clip_state"); + + trace_dump_member_begin("ucp"); + trace_dump_array_begin(); + for(i = 0; i < PIPE_MAX_CLIP_PLANES; ++i) { + trace_dump_elem_begin(); + trace_dump_array(float, state->ucp[i], 4); + trace_dump_elem_end(); + } + trace_dump_array_end(); + trace_dump_member_end(); + + trace_dump_member(uint, state, nr); + + trace_dump_struct_end(); +} + + +void trace_dump_constant_buffer(const struct pipe_constant_buffer *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_constant_buffer"); + + trace_dump_member(buffer_ptr, state, buffer); + + trace_dump_struct_end(); +} + + +void trace_dump_shader_state(const struct pipe_shader_state *state) +{ + static char str[8192]; + + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + tgsi_dump_str(state->tokens, 0, str, sizeof(str)); + + trace_dump_struct_begin("pipe_shader_state"); + + trace_dump_member_begin("tokens"); + trace_dump_string(str); + trace_dump_member_end(); + + trace_dump_struct_end(); +} + + +void trace_dump_depth_stencil_alpha_state(const struct pipe_depth_stencil_alpha_state *state) +{ + unsigned i; + + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_depth_stencil_alpha_state"); + + trace_dump_member_begin("depth"); + trace_dump_struct_begin("pipe_depth_state"); + trace_dump_member(bool, &state->depth, enabled); + trace_dump_member(bool, &state->depth, writemask); + trace_dump_member(uint, &state->depth, func); + trace_dump_struct_end(); + trace_dump_member_end(); + + trace_dump_member_begin("stencil"); + trace_dump_array_begin(); + for(i = 0; i < Elements(state->stencil); ++i) { + trace_dump_elem_begin(); + trace_dump_struct_begin("pipe_stencil_state"); + trace_dump_member(bool, &state->stencil[i], enabled); + trace_dump_member(uint, &state->stencil[i], func); + trace_dump_member(uint, &state->stencil[i], fail_op); + trace_dump_member(uint, &state->stencil[i], zpass_op); + trace_dump_member(uint, &state->stencil[i], zfail_op); + trace_dump_member(uint, &state->stencil[i], ref_value); + trace_dump_member(uint, &state->stencil[i], valuemask); + trace_dump_member(uint, &state->stencil[i], writemask); + trace_dump_struct_end(); + trace_dump_elem_end(); + } + trace_dump_array_end(); + trace_dump_member_end(); + + trace_dump_member_begin("alpha"); + trace_dump_struct_begin("pipe_alpha_state"); + trace_dump_member(bool, &state->alpha, enabled); + trace_dump_member(uint, &state->alpha, func); + trace_dump_member(float, &state->alpha, ref_value); + trace_dump_struct_end(); + trace_dump_member_end(); + + trace_dump_struct_end(); +} + + +void trace_dump_blend_state(const struct pipe_blend_state *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_blend_state"); + + trace_dump_member(bool, state, blend_enable); + + trace_dump_member(uint, state, rgb_func); + trace_dump_member(uint, state, rgb_src_factor); + trace_dump_member(uint, state, rgb_dst_factor); + + trace_dump_member(uint, state, alpha_func); + trace_dump_member(uint, state, alpha_src_factor); + trace_dump_member(uint, state, alpha_dst_factor); + + trace_dump_member(bool, state, logicop_enable); + trace_dump_member(uint, state, logicop_func); + + trace_dump_member(uint, state, colormask); + trace_dump_member(bool, state, dither); + + trace_dump_struct_end(); +} + + +void trace_dump_blend_color(const struct pipe_blend_color *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_blend_color"); + + trace_dump_member_array(float, state, color); + + trace_dump_struct_end(); +} + + +void trace_dump_framebuffer_state(const struct pipe_framebuffer_state *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + trace_dump_struct_begin("pipe_framebuffer_state"); + + trace_dump_member(uint, state, width); + trace_dump_member(uint, state, height); + trace_dump_member(uint, state, nr_cbufs); + trace_dump_member_array(ptr, state, cbufs); + trace_dump_member(ptr, state, zsbuf); + + trace_dump_struct_end(); +} + + +void trace_dump_sampler_state(const struct pipe_sampler_state *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_sampler_state"); + + trace_dump_member(uint, state, wrap_s); + trace_dump_member(uint, state, wrap_t); + trace_dump_member(uint, state, wrap_r); + trace_dump_member(uint, state, min_img_filter); + trace_dump_member(uint, state, min_mip_filter); + trace_dump_member(uint, state, mag_img_filter); + trace_dump_member(bool, state, compare_mode); + trace_dump_member(uint, state, compare_func); + trace_dump_member(bool, state, normalized_coords); + trace_dump_member(uint, state, prefilter); + trace_dump_member(float, state, lod_bias); + trace_dump_member(float, state, min_lod); + trace_dump_member(float, state, max_lod); + trace_dump_member_array(float, state, border_color); + trace_dump_member(float, state, max_anisotropy); + + trace_dump_struct_end(); +} + + +void trace_dump_surface(const struct pipe_surface *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_surface"); + + trace_dump_reference(&state->reference); + + trace_dump_member(format, state, format); + trace_dump_member(uint, state, width); + trace_dump_member(uint, state, height); + + trace_dump_member(uint, state, layout); + trace_dump_member(uint, state, offset); + trace_dump_member(uint, state, usage); + + trace_dump_member(ptr, state, texture); + trace_dump_member(uint, state, face); + trace_dump_member(uint, state, level); + trace_dump_member(uint, state, zslice); + + trace_dump_struct_end(); +} + + +void trace_dump_transfer(const struct pipe_transfer *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_transfer"); + + trace_dump_member(format, state, format); + trace_dump_member(uint, state, width); + trace_dump_member(uint, state, height); + + trace_dump_member_begin("block"); + trace_dump_block(&state->block); + trace_dump_member_end(); + + trace_dump_member(uint, state, nblocksx); + trace_dump_member(uint, state, nblocksy); + trace_dump_member(uint, state, stride); + trace_dump_member(uint, state, usage); + + trace_dump_member(ptr, state, texture); + trace_dump_member(uint, state, face); + trace_dump_member(uint, state, level); + trace_dump_member(uint, state, zslice); + + trace_dump_struct_end(); +} + + +void trace_dump_vertex_buffer(const struct pipe_vertex_buffer *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_vertex_buffer"); + + trace_dump_member(uint, state, stride); + trace_dump_member(uint, state, max_index); + trace_dump_member(uint, state, buffer_offset); + trace_dump_member(buffer_ptr, state, buffer); + + trace_dump_struct_end(); +} + + +void trace_dump_vertex_element(const struct pipe_vertex_element *state) +{ + if (!trace_dumping_enabled_locked()) + return; + + if(!state) { + trace_dump_null(); + return; + } + + trace_dump_struct_begin("pipe_vertex_element"); + + trace_dump_member(uint, state, src_offset); + + trace_dump_member(uint, state, vertex_buffer_index); + trace_dump_member(uint, state, nr_components); + + trace_dump_member(format, state, src_format); + + trace_dump_struct_end(); +} diff --git a/src/gallium/drivers/trace/tr_dump_state.h b/src/gallium/drivers/trace/tr_dump_state.h new file mode 100644 index 00000000000..05b821adb64 --- /dev/null +++ b/src/gallium/drivers/trace/tr_dump_state.h @@ -0,0 +1,78 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef TR_DUMP_STATE_H_ +#define TR_DUMP_STATE_H_ + +#include "pipe/p_format.h" +#include "pipe/p_state.h" +#include "pipe/p_shader_tokens.h" + + +void trace_dump_format(enum pipe_format format); + +void trace_dump_block(const struct pipe_format_block *block); + +void trace_dump_template(const struct pipe_texture *templat); + + +void trace_dump_rasterizer_state(const struct pipe_rasterizer_state *state); + +void trace_dump_poly_stipple(const struct pipe_poly_stipple *state); + +void trace_dump_viewport_state(const struct pipe_viewport_state *state); + +void trace_dump_scissor_state(const struct pipe_scissor_state *state); + +void trace_dump_clip_state(const struct pipe_clip_state *state); + +void trace_dump_constant_buffer(const struct pipe_constant_buffer *state); + +void trace_dump_token(const struct tgsi_token *token); + +void trace_dump_shader_state(const struct pipe_shader_state *state); + +void trace_dump_depth_stencil_alpha_state(const struct pipe_depth_stencil_alpha_state *state); + +void trace_dump_blend_state(const struct pipe_blend_state *state); + +void trace_dump_blend_color(const struct pipe_blend_color *state); + +void trace_dump_framebuffer_state(const struct pipe_framebuffer_state *state); + +void trace_dump_sampler_state(const struct pipe_sampler_state *state); + +void trace_dump_surface(const struct pipe_surface *state); + +void trace_dump_transfer(const struct pipe_transfer *state); + +void trace_dump_vertex_buffer(const struct pipe_vertex_buffer *state); + +void trace_dump_vertex_element(const struct pipe_vertex_element *state); + + +#endif /* TR_STATE_H */ diff --git a/src/gallium/drivers/trace/tr_rbug.c b/src/gallium/drivers/trace/tr_rbug.c new file mode 100644 index 00000000000..e85ac15edca --- /dev/null +++ b/src/gallium/drivers/trace/tr_rbug.c @@ -0,0 +1,864 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "util/u_string.h" +#include "util/u_memory.h" +#include "util/u_simple_list.h" +#include "util/u_network.h" + +#include "tgsi/tgsi_parse.h" + +#include "tr_dump.h" +#include "tr_state.h" +#include "tr_buffer.h" +#include "tr_texture.h" + +#include "rbug/rbug.h" + +#include <errno.h> + +#if defined(PIPE_SUBSYSTEM_WINDOWS_USER) +# define sleep Sleep +#elif defined(PIPE_OS_LINUX) +void usleep(int); +# define sleep usleep +#else +# warning "No socket implementation" +#endif + +#define U642VOID(x) ((void *)(unsigned long)(x)) +#define VOID2U64(x) ((uint64_t)(unsigned long)(x)) + +struct trace_rbug +{ + struct trace_screen *tr_scr; + struct rbug_connection *con; + pipe_thread thread; + boolean running; +}; + +PIPE_THREAD_ROUTINE(trace_rbug_thread, void_tr_rbug); + + +/********************************************************** + * Helper functions + */ + + +static struct trace_context * +trace_rbug_get_context_locked(struct trace_screen *tr_scr, rbug_context_t ctx) +{ + struct trace_context *tr_ctx = NULL; + struct tr_list *ptr; + + foreach(ptr, &tr_scr->contexts) { + tr_ctx = (struct trace_context *)((char*)ptr - offsetof(struct trace_context, list)); + if (ctx == VOID2U64(tr_ctx)) + break; + tr_ctx = NULL; + } + + return tr_ctx; +} + +static struct trace_shader * +trace_rbug_get_shader_locked(struct trace_context *tr_ctx, rbug_shader_t shdr) +{ + struct trace_shader *tr_shdr = NULL; + struct tr_list *ptr; + + foreach(ptr, &tr_ctx->shaders) { + tr_shdr = (struct trace_shader *)((char*)ptr - offsetof(struct trace_shader, list)); + if (shdr == VOID2U64(tr_shdr)) + break; + tr_shdr = NULL; + } + + return tr_shdr; +} + +static void * +trace_shader_create_locked(struct pipe_context *pipe, + struct trace_shader *tr_shdr, + struct tgsi_token *tokens) +{ + void *state = NULL; + struct pipe_shader_state pss = { 0 }; + pss.tokens = tokens; + + if (tr_shdr->type == TRACE_SHADER_FRAGMENT) { + state = pipe->create_fs_state(pipe, &pss); + } else if (tr_shdr->type == TRACE_SHADER_VERTEX) { + state = pipe->create_vs_state(pipe, &pss); + } else + assert(0); + + return state; +} + +static void +trace_shader_bind_locked(struct pipe_context *pipe, + struct trace_shader *tr_shdr, + void *state) +{ + if (tr_shdr->type == TRACE_SHADER_FRAGMENT) { + pipe->bind_fs_state(pipe, state); + } else if (tr_shdr->type == TRACE_SHADER_VERTEX) { + pipe->bind_vs_state(pipe, state); + } else + assert(0); +} + +static void +trace_shader_delete_locked(struct pipe_context *pipe, + struct trace_shader *tr_shdr, + void *state) +{ + if (tr_shdr->type == TRACE_SHADER_FRAGMENT) { + pipe->delete_fs_state(pipe, state); + } else if (tr_shdr->type == TRACE_SHADER_VERTEX) { + pipe->delete_vs_state(pipe, state); + } else + assert(0); +} + +/************************************************ + * Request handler functions + */ + + +static int +trace_rbug_texture_list(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_texture *tr_tex = NULL; + struct tr_list *ptr; + rbug_texture_t *texs; + int i = 0; + + pipe_mutex_lock(tr_scr->list_mutex); + texs = MALLOC(tr_scr->num_textures * sizeof(rbug_texture_t)); + foreach(ptr, &tr_scr->textures) { + tr_tex = (struct trace_texture *)((char*)ptr - offsetof(struct trace_texture, list)); + texs[i++] = VOID2U64(tr_tex); + } + pipe_mutex_unlock(tr_scr->list_mutex); + + rbug_send_texture_list_reply(tr_rbug->con, serial, texs, i, NULL); + FREE(texs); + + return 0; +} + +static int +trace_rbug_texture_info(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_texture *tr_tex; + struct rbug_proto_texture_info *gpti = (struct rbug_proto_texture_info *)header; + struct tr_list *ptr; + struct pipe_texture *t; + + pipe_mutex_lock(tr_scr->list_mutex); + foreach(ptr, &tr_scr->textures) { + tr_tex = (struct trace_texture *)((char*)ptr - offsetof(struct trace_texture, list)); + if (gpti->texture == VOID2U64(tr_tex)) + break; + tr_tex = NULL; + } + + if (!tr_tex) { + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + t = tr_tex->texture; + rbug_send_texture_info_reply(tr_rbug->con, serial, + t->target, t->format, + t->width, t->last_level + 1, + t->height, t->last_level + 1, + t->depth, t->last_level + 1, + t->block.width, t->block.height, t->block.size, + t->last_level, + t->nr_samples, + t->tex_usage, + NULL); + + pipe_mutex_unlock(tr_scr->list_mutex); + + return 0; +} + +static int +trace_rbug_texture_read(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + struct rbug_proto_texture_read *gptr = (struct rbug_proto_texture_read *)header; + + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_texture *tr_tex; + struct tr_list *ptr; + + struct pipe_screen *screen = tr_scr->screen; + struct pipe_texture *tex; + struct pipe_transfer *t; + + void *map; + + pipe_mutex_lock(tr_scr->list_mutex); + foreach(ptr, &tr_scr->textures) { + tr_tex = (struct trace_texture *)((char*)ptr - offsetof(struct trace_texture, list)); + if (gptr->texture == VOID2U64(tr_tex)) + break; + tr_tex = NULL; + } + + if (!tr_tex) { + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + tex = tr_tex->texture; + t = screen->get_tex_transfer(tr_scr->screen, tex, + gptr->face, gptr->level, gptr->zslice, + PIPE_TRANSFER_READ, + gptr->x, gptr->y, gptr->w, gptr->h); + + map = screen->transfer_map(screen, t); + + rbug_send_texture_read_reply(tr_rbug->con, serial, + t->format, + t->block.width, t->block.height, t->block.size, + (uint8_t*)map, t->stride * t->nblocksy, + t->stride, + NULL); + + screen->transfer_unmap(screen, t); + screen->tex_transfer_destroy(t); + + pipe_mutex_unlock(tr_scr->list_mutex); + + return 0; +} + +static int +trace_rbug_context_list(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct tr_list *ptr; + struct trace_context *tr_ctx = NULL; + rbug_context_t *ctxs; + int i = 0; + + pipe_mutex_lock(tr_scr->list_mutex); + ctxs = MALLOC(tr_scr->num_contexts * sizeof(rbug_context_t)); + foreach(ptr, &tr_scr->contexts) { + tr_ctx = (struct trace_context *)((char*)ptr - offsetof(struct trace_context, list)); + ctxs[i++] = VOID2U64(tr_ctx); + } + pipe_mutex_unlock(tr_scr->list_mutex); + + rbug_send_context_list_reply(tr_rbug->con, serial, ctxs, i, NULL); + FREE(ctxs); + + return 0; +} + +static int +trace_rbug_context_info(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + struct rbug_proto_context_info *info = (struct rbug_proto_context_info *)header; + + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_context *tr_ctx = NULL; + rbug_texture_t cbufs[PIPE_MAX_COLOR_BUFS]; + rbug_texture_t texs[PIPE_MAX_SAMPLERS]; + int i; + + pipe_mutex_lock(tr_scr->list_mutex); + tr_ctx = trace_rbug_get_context_locked(tr_scr, info->context); + + if (!tr_ctx) { + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + /* protect the pipe context */ + pipe_mutex_lock(tr_ctx->draw_mutex); + trace_dump_call_lock(); + + for (i = 0; i < tr_ctx->curr.nr_cbufs; i++) + cbufs[i] = VOID2U64(tr_ctx->curr.cbufs[i]); + + for (i = 0; i < tr_ctx->curr.num_texs; i++) + texs[i] = VOID2U64(tr_ctx->curr.tex[i]); + + rbug_send_context_info_reply(tr_rbug->con, serial, + VOID2U64(tr_ctx->curr.vs), VOID2U64(tr_ctx->curr.fs), + texs, tr_ctx->curr.num_texs, + cbufs, tr_ctx->curr.nr_cbufs, + VOID2U64(tr_ctx->curr.zsbuf), + tr_ctx->draw_blocker, tr_ctx->draw_blocked, NULL); + + trace_dump_call_unlock(); + pipe_mutex_unlock(tr_ctx->draw_mutex); + + pipe_mutex_unlock(tr_scr->list_mutex); + + return 0; +} + +static int +trace_rbug_context_draw_block(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + struct rbug_proto_context_draw_block *block = (struct rbug_proto_context_draw_block *)header; + + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_context *tr_ctx = NULL; + + pipe_mutex_lock(tr_scr->list_mutex); + tr_ctx = trace_rbug_get_context_locked(tr_scr, block->context); + + if (!tr_ctx) { + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + pipe_mutex_lock(tr_ctx->draw_mutex); + tr_ctx->draw_blocker |= block->block; + pipe_mutex_unlock(tr_ctx->draw_mutex); + + pipe_mutex_unlock(tr_scr->list_mutex); + + return 0; +} + +static int +trace_rbug_context_draw_step(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + struct rbug_proto_context_draw_step *step = (struct rbug_proto_context_draw_step *)header; + + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_context *tr_ctx = NULL; + + pipe_mutex_lock(tr_scr->list_mutex); + tr_ctx = trace_rbug_get_context_locked(tr_scr, step->context); + + if (!tr_ctx) { + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + pipe_mutex_lock(tr_ctx->draw_mutex); + if (tr_ctx->draw_blocked & RBUG_BLOCK_RULE) { + if (step->step & RBUG_BLOCK_RULE) + tr_ctx->draw_blocked &= ~RBUG_BLOCK_MASK; + } else { + tr_ctx->draw_blocked &= ~step->step; + } + pipe_mutex_unlock(tr_ctx->draw_mutex); + +#ifdef PIPE_THREAD_HAVE_CONDVAR + pipe_condvar_broadcast(tr_ctx->draw_cond); +#endif + + pipe_mutex_unlock(tr_scr->list_mutex); + + return 0; +} + +static int +trace_rbug_context_draw_unblock(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + struct rbug_proto_context_draw_unblock *unblock = (struct rbug_proto_context_draw_unblock *)header; + + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_context *tr_ctx = NULL; + + pipe_mutex_lock(tr_scr->list_mutex); + tr_ctx = trace_rbug_get_context_locked(tr_scr, unblock->context); + + if (!tr_ctx) { + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + pipe_mutex_lock(tr_ctx->draw_mutex); + if (tr_ctx->draw_blocked & RBUG_BLOCK_RULE) { + if (unblock->unblock & RBUG_BLOCK_RULE) + tr_ctx->draw_blocked &= ~RBUG_BLOCK_MASK; + } else { + tr_ctx->draw_blocked &= ~unblock->unblock; + } + tr_ctx->draw_blocker &= ~unblock->unblock; + pipe_mutex_unlock(tr_ctx->draw_mutex); + +#ifdef PIPE_THREAD_HAVE_CONDVAR + pipe_condvar_broadcast(tr_ctx->draw_cond); +#endif + + pipe_mutex_unlock(tr_scr->list_mutex); + + return 0; +} + +static int +trace_rbug_context_draw_rule(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + struct rbug_proto_context_draw_rule *rule = (struct rbug_proto_context_draw_rule *)header; + + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_context *tr_ctx = NULL; + + pipe_mutex_lock(tr_scr->list_mutex); + tr_ctx = trace_rbug_get_context_locked(tr_scr, rule->context); + + if (!tr_ctx) { + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + pipe_mutex_lock(tr_ctx->draw_mutex); + tr_ctx->draw_rule.vs = U642VOID(rule->vertex); + tr_ctx->draw_rule.fs = U642VOID(rule->fragment); + tr_ctx->draw_rule.tex = U642VOID(rule->texture); + tr_ctx->draw_rule.surf = U642VOID(rule->surface); + tr_ctx->draw_rule.blocker = rule->block; + tr_ctx->draw_blocker |= RBUG_BLOCK_RULE; + pipe_mutex_unlock(tr_ctx->draw_mutex); + +#ifdef PIPE_THREAD_HAVE_CONDVAR + pipe_condvar_broadcast(tr_ctx->draw_cond); +#endif + + pipe_mutex_unlock(tr_scr->list_mutex); + + return 0; +} + +static int +trace_rbug_context_flush(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + struct rbug_proto_context_flush *flush = (struct rbug_proto_context_flush *)header; + + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_context *tr_ctx = NULL; + + pipe_mutex_lock(tr_scr->list_mutex); + tr_ctx = trace_rbug_get_context_locked(tr_scr, flush->context); + + if (!tr_ctx) { + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + /* protect the pipe context */ + trace_dump_call_lock(); + + tr_ctx->pipe->flush(tr_ctx->pipe, flush->flags, NULL); + + trace_dump_call_unlock(); + pipe_mutex_unlock(tr_scr->list_mutex); + + return 0; +} + +static int +trace_rbug_shader_list(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + struct rbug_proto_shader_list *list = (struct rbug_proto_shader_list *)header; + + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_context *tr_ctx = NULL; + struct trace_shader *tr_shdr = NULL; + struct tr_list *ptr; + rbug_shader_t *shdrs; + int i = 0; + + pipe_mutex_lock(tr_scr->list_mutex); + tr_ctx = trace_rbug_get_context_locked(tr_scr, list->context); + + if (!tr_ctx) { + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + pipe_mutex_lock(tr_ctx->list_mutex); + shdrs = MALLOC(tr_ctx->num_shaders * sizeof(rbug_shader_t)); + foreach(ptr, &tr_ctx->shaders) { + tr_shdr = (struct trace_shader *)((char*)ptr - offsetof(struct trace_shader, list)); + shdrs[i++] = VOID2U64(tr_shdr); + } + + pipe_mutex_unlock(tr_ctx->list_mutex); + pipe_mutex_unlock(tr_scr->list_mutex); + + rbug_send_shader_list_reply(tr_rbug->con, serial, shdrs, i, NULL); + FREE(shdrs); + + return 0; +} + +static int +trace_rbug_shader_info(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + struct rbug_proto_shader_info *info = (struct rbug_proto_shader_info *)header; + + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_context *tr_ctx = NULL; + struct trace_shader *tr_shdr = NULL; + unsigned original_len; + unsigned replaced_len; + + pipe_mutex_lock(tr_scr->list_mutex); + tr_ctx = trace_rbug_get_context_locked(tr_scr, info->context); + + if (!tr_ctx) { + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + pipe_mutex_lock(tr_ctx->list_mutex); + + tr_shdr = trace_rbug_get_shader_locked(tr_ctx, info->shader); + + if (!tr_shdr) { + pipe_mutex_unlock(tr_ctx->list_mutex); + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + /* just in case */ + assert(sizeof(struct tgsi_token) == 4); + + original_len = tgsi_num_tokens(tr_shdr->tokens); + if (tr_shdr->replaced_tokens) + replaced_len = tgsi_num_tokens(tr_shdr->replaced_tokens); + else + replaced_len = 0; + + rbug_send_shader_info_reply(tr_rbug->con, serial, + (uint32_t*)tr_shdr->tokens, original_len, + (uint32_t*)tr_shdr->replaced_tokens, replaced_len, + tr_shdr->disabled, + NULL); + + pipe_mutex_unlock(tr_ctx->list_mutex); + pipe_mutex_unlock(tr_scr->list_mutex); + + return 0; +} + +static int +trace_rbug_shader_disable(struct trace_rbug *tr_rbug, struct rbug_header *header) +{ + struct rbug_proto_shader_disable *dis = (struct rbug_proto_shader_disable *)header; + + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_context *tr_ctx = NULL; + struct trace_shader *tr_shdr = NULL; + + pipe_mutex_lock(tr_scr->list_mutex); + tr_ctx = trace_rbug_get_context_locked(tr_scr, dis->context); + + if (!tr_ctx) { + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + pipe_mutex_lock(tr_ctx->list_mutex); + + tr_shdr = trace_rbug_get_shader_locked(tr_ctx, dis->shader); + + if (!tr_shdr) { + pipe_mutex_unlock(tr_ctx->list_mutex); + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + tr_shdr->disabled = dis->disable; + + pipe_mutex_unlock(tr_ctx->list_mutex); + pipe_mutex_unlock(tr_scr->list_mutex); + + return 0; +} + +static int +trace_rbug_shader_replace(struct trace_rbug *tr_rbug, struct rbug_header *header) +{ + struct rbug_proto_shader_replace *rep = (struct rbug_proto_shader_replace *)header; + + struct trace_screen *tr_scr = tr_rbug->tr_scr; + struct trace_context *tr_ctx = NULL; + struct trace_shader *tr_shdr = NULL; + struct pipe_context *pipe = NULL; + void *state; + + pipe_mutex_lock(tr_scr->list_mutex); + tr_ctx = trace_rbug_get_context_locked(tr_scr, rep->context); + + if (!tr_ctx) { + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + pipe_mutex_lock(tr_ctx->list_mutex); + + tr_shdr = trace_rbug_get_shader_locked(tr_ctx, rep->shader); + + if (!tr_shdr) { + pipe_mutex_unlock(tr_ctx->list_mutex); + pipe_mutex_unlock(tr_scr->list_mutex); + return -ESRCH; + } + + /* protect the pipe context */ + trace_dump_call_lock(); + + pipe = tr_ctx->pipe; + + /* remove old replaced shader */ + if (tr_shdr->replaced) { + if (tr_ctx->curr.fs == tr_shdr || tr_ctx->curr.vs == tr_shdr) + trace_shader_bind_locked(pipe, tr_shdr, tr_shdr->state); + + FREE(tr_shdr->replaced_tokens); + trace_shader_delete_locked(pipe, tr_shdr, tr_shdr->replaced); + tr_shdr->replaced = NULL; + tr_shdr->replaced_tokens = NULL; + } + + /* empty inputs means restore old which we did above */ + if (rep->tokens_len == 0) + goto out; + + tr_shdr->replaced_tokens = tgsi_dup_tokens((struct tgsi_token *)rep->tokens); + if (!tr_shdr->replaced_tokens) + goto err; + + state = trace_shader_create_locked(pipe, tr_shdr, tr_shdr->replaced_tokens); + if (!state) + goto err; + + /* bind new shader if the shader is currently a bound */ + if (tr_ctx->curr.fs == tr_shdr || tr_ctx->curr.vs == tr_shdr) + trace_shader_bind_locked(pipe, tr_shdr, state); + + /* save state */ + tr_shdr->replaced = state; + +out: + trace_dump_call_unlock(); + pipe_mutex_unlock(tr_ctx->list_mutex); + pipe_mutex_unlock(tr_scr->list_mutex); + + return 0; + +err: + FREE(tr_shdr->replaced_tokens); + tr_shdr->replaced = NULL; + tr_shdr->replaced_tokens = NULL; + + trace_dump_call_unlock(); + pipe_mutex_unlock(tr_ctx->list_mutex); + pipe_mutex_unlock(tr_scr->list_mutex); + return -EINVAL; +} + +static boolean +trace_rbug_header(struct trace_rbug *tr_rbug, struct rbug_header *header, uint32_t serial) +{ + int ret = 0; + + switch(header->opcode) { + case RBUG_OP_PING: + rbug_send_ping_reply(tr_rbug->con, serial, NULL); + break; + case RBUG_OP_TEXTURE_LIST: + ret = trace_rbug_texture_list(tr_rbug, header, serial); + break; + case RBUG_OP_TEXTURE_INFO: + ret = trace_rbug_texture_info(tr_rbug, header, serial); + break; + case RBUG_OP_TEXTURE_READ: + ret = trace_rbug_texture_read(tr_rbug, header, serial); + break; + case RBUG_OP_CONTEXT_LIST: + ret = trace_rbug_context_list(tr_rbug, header, serial); + break; + case RBUG_OP_CONTEXT_INFO: + ret = trace_rbug_context_info(tr_rbug, header, serial); + break; + case RBUG_OP_CONTEXT_DRAW_BLOCK: + ret = trace_rbug_context_draw_block(tr_rbug, header, serial); + break; + case RBUG_OP_CONTEXT_DRAW_STEP: + ret = trace_rbug_context_draw_step(tr_rbug, header, serial); + break; + case RBUG_OP_CONTEXT_DRAW_UNBLOCK: + ret = trace_rbug_context_draw_unblock(tr_rbug, header, serial); + break; + case RBUG_OP_CONTEXT_DRAW_RULE: + ret = trace_rbug_context_draw_rule(tr_rbug, header, serial); + break; + case RBUG_OP_CONTEXT_FLUSH: + ret = trace_rbug_context_flush(tr_rbug, header, serial); + break; + case RBUG_OP_SHADER_LIST: + ret = trace_rbug_shader_list(tr_rbug, header, serial); + break; + case RBUG_OP_SHADER_INFO: + ret = trace_rbug_shader_info(tr_rbug, header, serial); + break; + case RBUG_OP_SHADER_DISABLE: + ret = trace_rbug_shader_disable(tr_rbug, header); + break; + case RBUG_OP_SHADER_REPLACE: + ret = trace_rbug_shader_replace(tr_rbug, header); + break; + default: + debug_printf("%s - unsupported opcode %u\n", __FUNCTION__, header->opcode); + ret = -ENOSYS; + break; + } + rbug_free_header(header); + + if (ret) + rbug_send_error_reply(tr_rbug->con, serial, ret, NULL); + + return TRUE; +} + +static void +trace_rbug_con(struct trace_rbug *tr_rbug) +{ + struct rbug_header *header; + uint32_t serial; + + debug_printf("%s - connection received\n", __FUNCTION__); + + while(tr_rbug->running) { + header = rbug_get_message(tr_rbug->con, &serial); + if (!header) + break; + + if (!trace_rbug_header(tr_rbug, header, serial)) + break; + } + + debug_printf("%s - connection closed\n", __FUNCTION__); + + rbug_disconnect(tr_rbug->con); + tr_rbug->con = NULL; +} + +PIPE_THREAD_ROUTINE(trace_rbug_thread, void_tr_rbug) +{ + struct trace_rbug *tr_rbug = void_tr_rbug; + uint16_t port = 13370; + int s = -1; + int c; + + u_socket_init(); + + for (;port <= 13379 && s < 0; port++) + s = u_socket_listen_on_port(port); + + if (s < 0) { + debug_printf("trace_rbug - failed to listen\n"); + return NULL; + } + + u_socket_block(s, false); + + debug_printf("trace_rbug - remote debugging listening on port %u\n", --port); + + while(tr_rbug->running) { + sleep(1); + + c = u_socket_accept(s); + if (c < 0) + continue; + + u_socket_block(c, true); + tr_rbug->con = rbug_from_socket(c); + + trace_rbug_con(tr_rbug); + + u_socket_close(c); + } + + u_socket_close(s); + + u_socket_stop(); + + return NULL; +} + +/********************************************************** + * + */ + +struct trace_rbug * +trace_rbug_start(struct trace_screen *tr_scr) +{ + struct trace_rbug *tr_rbug = CALLOC_STRUCT(trace_rbug); + if (!tr_rbug) + return NULL; + + tr_rbug->tr_scr = tr_scr; + tr_rbug->running = TRUE; + tr_rbug->thread = pipe_thread_create(trace_rbug_thread, tr_rbug); + + return tr_rbug; +} + +void +trace_rbug_stop(struct trace_rbug *tr_rbug) +{ + if (!tr_rbug) + return; + + tr_rbug->running = false; + pipe_thread_wait(tr_rbug->thread); + + FREE(tr_rbug); + + return; +} + +void +trace_rbug_notify_draw_blocked(struct trace_context *tr_ctx) +{ + struct trace_screen *tr_scr = trace_screen(tr_ctx->base.screen); + struct trace_rbug *tr_rbug = tr_scr->rbug; + + if (tr_rbug && tr_rbug->con) + rbug_send_context_draw_blocked(tr_rbug->con, + VOID2U64(tr_ctx), tr_ctx->draw_blocked, NULL); +} diff --git a/src/gallium/drivers/trace/tr_screen.c b/src/gallium/drivers/trace/tr_screen.c index 12a85353428..26f1c04594f 100644 --- a/src/gallium/drivers/trace/tr_screen.c +++ b/src/gallium/drivers/trace/tr_screen.c @@ -30,13 +30,16 @@ #include "tr_buffer.h" #include "tr_dump.h" -#include "tr_state.h" +#include "tr_dump_state.h" #include "tr_texture.h" #include "tr_screen.h" #include "pipe/p_inlines.h" +static boolean trace = FALSE; +static boolean rbug = FALSE; + static const char * trace_screen_get_name(struct pipe_screen *_screen) { @@ -212,10 +215,12 @@ static struct pipe_texture * trace_screen_texture_blanket(struct pipe_screen *_screen, const struct pipe_texture *templat, const unsigned *ppitch, - struct pipe_buffer *buffer) + struct pipe_buffer *_buffer) { struct trace_screen *tr_scr = trace_screen(_screen); + struct trace_buffer *tr_buf = trace_buffer(_buffer); struct pipe_screen *screen = tr_scr->screen; + struct pipe_buffer *buffer = tr_buf->buffer; unsigned pitch = *ppitch; struct pipe_texture *result; @@ -457,6 +462,7 @@ trace_screen_surface_buffer_create(struct pipe_screen *_screen, unsigned width, unsigned height, enum pipe_format format, unsigned usage, + unsigned tex_usage, unsigned *pstride) { struct trace_screen *tr_scr = trace_screen(_screen); @@ -471,11 +477,13 @@ trace_screen_surface_buffer_create(struct pipe_screen *_screen, trace_dump_arg(uint, height); trace_dump_arg(format, format); trace_dump_arg(uint, usage); + trace_dump_arg(uint, tex_usage); result = screen->surface_buffer_create(screen, width, height, format, usage, + tex_usage, pstride); stride = *pstride; @@ -818,18 +826,41 @@ trace_screen_destroy(struct pipe_screen *_screen) struct pipe_screen *screen = tr_scr->screen; trace_dump_call_begin("pipe_screen", "destroy"); - trace_dump_arg(ptr, screen); - - screen->destroy(screen); - trace_dump_call_end(); - trace_dump_trace_end(); + if (tr_scr->rbug) + trace_rbug_stop(tr_scr->rbug); + + screen->destroy(screen); + FREE(tr_scr); } +boolean +trace_enabled(void) +{ + static boolean firstrun = TRUE; + + if (!firstrun) + return trace; + firstrun = FALSE; + + trace_dump_init(); + + if(trace_dump_trace_begin()) { + trace_dumping_start(); + trace = TRUE; + } + + if (debug_get_bool_option("GALLIUM_RBUG", FALSE)) { + trace = TRUE; + rbug = TRUE; + } + + return trace; +} struct pipe_screen * trace_screen_create(struct pipe_screen *screen) @@ -840,13 +871,9 @@ trace_screen_create(struct pipe_screen *screen) if(!screen) goto error1; - trace_dump_init(); - - if(!trace_dump_trace_begin()) + if (!trace_enabled()) goto error1; - trace_dumping_start(); - trace_dump_call_begin("", "pipe_screen_create"); tr_scr = CALLOC_STRUCT(trace_screen); @@ -904,6 +931,9 @@ trace_screen_create(struct pipe_screen *screen) trace_dump_ret(ptr, screen); trace_dump_call_end(); + if (rbug) + tr_scr->rbug = trace_rbug_start(tr_scr); + return &tr_scr->base; #if 0 diff --git a/src/gallium/drivers/trace/tr_screen.h b/src/gallium/drivers/trace/tr_screen.h index 59f254166d4..dba8cd7c653 100644 --- a/src/gallium/drivers/trace/tr_screen.h +++ b/src/gallium/drivers/trace/tr_screen.h @@ -57,6 +57,9 @@ struct trace_screen struct pipe_screen *screen; + /* remote debugger */ + struct trace_rbug *rbug; + pipe_mutex list_mutex; int num_buffers; int num_contexts; @@ -71,14 +74,34 @@ struct trace_screen }; +/* + * tr_rbug.c + */ + + +struct trace_rbug; + +struct trace_rbug * +trace_rbug_start(struct trace_screen *tr_scr); + +void +trace_rbug_stop(struct trace_rbug *tr_rbug); + + +/* + * tr_screen.c + */ + + +boolean +trace_enabled(void); + struct trace_screen * trace_screen(struct pipe_screen *screen); - struct pipe_screen * trace_screen_create(struct pipe_screen *screen); - void trace_screen_user_buffer_update(struct pipe_screen *screen, struct pipe_buffer *buffer); @@ -99,6 +122,7 @@ trace_screen_user_buffer_update(struct pipe_screen *screen, pipe_mutex_unlock(tr_scr->list_mutex); \ } while (0) + #ifdef __cplusplus } #endif diff --git a/src/gallium/drivers/trace/tr_state.c b/src/gallium/drivers/trace/tr_state.c index a9570c1aebd..d8c11640bf3 100644 --- a/src/gallium/drivers/trace/tr_state.c +++ b/src/gallium/drivers/trace/tr_state.c @@ -1,491 +1,66 @@ -/************************************************************************** - * - * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. +/* + * Copyright 2009 VMware, Inc. * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. * - **************************************************************************/ - + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ -#include "pipe/p_compiler.h" -#include "util/u_memory.h" -#include "tgsi/tgsi_dump.h" - -#include "tr_dump.h" #include "tr_state.h" +#include "util/u_memory.h" +#include "util/u_simple_list.h" -void trace_dump_format(enum pipe_format format) -{ - trace_dump_enum(pf_name(format) ); -} - - -void trace_dump_block(const struct pipe_format_block *block) -{ - trace_dump_struct_begin("pipe_format_block"); - trace_dump_member(uint, block, size); - trace_dump_member(uint, block, width); - trace_dump_member(uint, block, height); - trace_dump_struct_end(); -} - - -static void trace_dump_reference(const struct pipe_reference *reference) -{ - trace_dump_struct_begin("pipe_reference"); - trace_dump_member(int, &reference->count, count); - trace_dump_struct_end(); -} - - -void trace_dump_template(const struct pipe_texture *templat) -{ - if(!templat) { - trace_dump_null(); - return; - } - - trace_dump_struct_begin("pipe_texture"); - - trace_dump_member(int, templat, target); - trace_dump_member(format, templat, format); - - trace_dump_member_begin("width"); - trace_dump_array(uint, templat->width, 1); - trace_dump_member_end(); - - trace_dump_member_begin("height"); - trace_dump_array(uint, templat->height, 1); - trace_dump_member_end(); - - trace_dump_member_begin("depth"); - trace_dump_array(uint, templat->depth, 1); - trace_dump_member_end(); - - trace_dump_member_begin("block"); - trace_dump_block(&templat->block); - trace_dump_member_end(); - - trace_dump_member(uint, templat, last_level); - trace_dump_member(uint, templat, tex_usage); - - trace_dump_struct_end(); -} - - -void trace_dump_rasterizer_state(const struct pipe_rasterizer_state *state) -{ - if(!state) { - trace_dump_null(); - return; - } - - trace_dump_struct_begin("pipe_rasterizer_state"); - - trace_dump_member(bool, state, flatshade); - trace_dump_member(bool, state, light_twoside); - trace_dump_member(uint, state, front_winding); - trace_dump_member(uint, state, cull_mode); - trace_dump_member(uint, state, fill_cw); - trace_dump_member(uint, state, fill_ccw); - trace_dump_member(bool, state, offset_cw); - trace_dump_member(bool, state, offset_ccw); - trace_dump_member(bool, state, scissor); - trace_dump_member(bool, state, poly_smooth); - trace_dump_member(bool, state, poly_stipple_enable); - trace_dump_member(bool, state, point_smooth); - trace_dump_member(bool, state, point_sprite); - trace_dump_member(bool, state, point_size_per_vertex); - trace_dump_member(bool, state, multisample); - trace_dump_member(bool, state, line_smooth); - trace_dump_member(bool, state, line_stipple_enable); - trace_dump_member(uint, state, line_stipple_factor); - trace_dump_member(uint, state, line_stipple_pattern); - trace_dump_member(bool, state, line_last_pixel); - trace_dump_member(bool, state, bypass_vs_clip_and_viewport); - trace_dump_member(bool, state, flatshade_first); - trace_dump_member(bool, state, gl_rasterization_rules); - - trace_dump_member(float, state, line_width); - trace_dump_member(float, state, point_size); - trace_dump_member(float, state, point_size_min); - trace_dump_member(float, state, point_size_max); - trace_dump_member(float, state, offset_units); - trace_dump_member(float, state, offset_scale); - - trace_dump_member_array(uint, state, sprite_coord_mode); - - trace_dump_struct_end(); -} - - -void trace_dump_poly_stipple(const struct pipe_poly_stipple *state) -{ - if(!state) { - trace_dump_null(); - return; - } - - trace_dump_struct_begin("pipe_poly_stipple"); - - trace_dump_member_begin("stipple"); - trace_dump_array(uint, - state->stipple, - Elements(state->stipple)); - trace_dump_member_end(); - - trace_dump_struct_end(); -} - - -void trace_dump_viewport_state(const struct pipe_viewport_state *state) -{ - if(!state) { - trace_dump_null(); - return; - } - - trace_dump_struct_begin("pipe_viewport_state"); - - trace_dump_member_array(float, state, scale); - trace_dump_member_array(float, state, translate); - - trace_dump_struct_end(); -} - - -void trace_dump_scissor_state(const struct pipe_scissor_state *state) -{ - if(!state) { - trace_dump_null(); - return; - } - - trace_dump_struct_begin("pipe_scissor_state"); - - trace_dump_member(uint, state, minx); - trace_dump_member(uint, state, miny); - trace_dump_member(uint, state, maxx); - trace_dump_member(uint, state, maxy); - - trace_dump_struct_end(); -} - - -void trace_dump_clip_state(const struct pipe_clip_state *state) -{ - unsigned i; - - if(!state) { - trace_dump_null(); - return; - } - - trace_dump_struct_begin("pipe_clip_state"); - - trace_dump_member_begin("ucp"); - trace_dump_array_begin(); - for(i = 0; i < PIPE_MAX_CLIP_PLANES; ++i) { - trace_dump_elem_begin(); - trace_dump_array(float, state->ucp[i], 4); - trace_dump_elem_end(); - } - trace_dump_array_end(); - trace_dump_member_end(); - - trace_dump_member(uint, state, nr); - - trace_dump_struct_end(); -} - - -void trace_dump_constant_buffer(const struct pipe_constant_buffer *state) -{ - if(!state) { - trace_dump_null(); - return; - } - - trace_dump_struct_begin("pipe_constant_buffer"); - - trace_dump_member(buffer_ptr, state, buffer); - - trace_dump_struct_end(); -} - - -void trace_dump_shader_state(const struct pipe_shader_state *state) -{ - static char str[8192]; - - if(!state) { - trace_dump_null(); - return; - } - - tgsi_dump_str(state->tokens, 0, str, sizeof(str)); - - trace_dump_struct_begin("pipe_shader_state"); - - trace_dump_member_begin("tokens"); - trace_dump_string(str); - trace_dump_member_end(); - - trace_dump_struct_end(); -} - - -void trace_dump_depth_stencil_alpha_state(const struct pipe_depth_stencil_alpha_state *state) -{ - unsigned i; - - if(!state) { - trace_dump_null(); - return; - } - - trace_dump_struct_begin("pipe_depth_stencil_alpha_state"); - - trace_dump_member_begin("depth"); - trace_dump_struct_begin("pipe_depth_state"); - trace_dump_member(bool, &state->depth, enabled); - trace_dump_member(bool, &state->depth, writemask); - trace_dump_member(uint, &state->depth, func); - trace_dump_member(bool, &state->depth, occlusion_count); - trace_dump_struct_end(); - trace_dump_member_end(); - - trace_dump_member_begin("stencil"); - trace_dump_array_begin(); - for(i = 0; i < Elements(state->stencil); ++i) { - trace_dump_elem_begin(); - trace_dump_struct_begin("pipe_stencil_state"); - trace_dump_member(bool, &state->stencil[i], enabled); - trace_dump_member(uint, &state->stencil[i], func); - trace_dump_member(uint, &state->stencil[i], fail_op); - trace_dump_member(uint, &state->stencil[i], zpass_op); - trace_dump_member(uint, &state->stencil[i], zfail_op); - trace_dump_member(uint, &state->stencil[i], ref_value); - trace_dump_member(uint, &state->stencil[i], valuemask); - trace_dump_member(uint, &state->stencil[i], writemask); - trace_dump_struct_end(); - trace_dump_elem_end(); - } - trace_dump_array_end(); - trace_dump_member_end(); - - trace_dump_member_begin("alpha"); - trace_dump_struct_begin("pipe_alpha_state"); - trace_dump_member(bool, &state->alpha, enabled); - trace_dump_member(uint, &state->alpha, func); - trace_dump_member(float, &state->alpha, ref_value); - trace_dump_struct_end(); - trace_dump_member_end(); - - trace_dump_struct_end(); -} - - -void trace_dump_blend_state(const struct pipe_blend_state *state) -{ - if(!state) { - trace_dump_null(); - return; - } - - trace_dump_struct_begin("pipe_blend_state"); - - trace_dump_member(bool, state, blend_enable); - - trace_dump_member(uint, state, rgb_func); - trace_dump_member(uint, state, rgb_src_factor); - trace_dump_member(uint, state, rgb_dst_factor); - - trace_dump_member(uint, state, alpha_func); - trace_dump_member(uint, state, alpha_src_factor); - trace_dump_member(uint, state, alpha_dst_factor); - - trace_dump_member(bool, state, logicop_enable); - trace_dump_member(uint, state, logicop_func); - - trace_dump_member(uint, state, colormask); - trace_dump_member(bool, state, dither); - - trace_dump_struct_end(); -} - - -void trace_dump_blend_color(const struct pipe_blend_color *state) -{ - if(!state) { - trace_dump_null(); - return; - } - - trace_dump_struct_begin("pipe_blend_color"); - - trace_dump_member_array(float, state, color); - - trace_dump_struct_end(); -} - - -void trace_dump_framebuffer_state(const struct pipe_framebuffer_state *state) -{ - trace_dump_struct_begin("pipe_framebuffer_state"); - - trace_dump_member(uint, state, width); - trace_dump_member(uint, state, height); - trace_dump_member(uint, state, nr_cbufs); - trace_dump_member_array(ptr, state, cbufs); - trace_dump_member(ptr, state, zsbuf); - - trace_dump_struct_end(); -} - - -void trace_dump_sampler_state(const struct pipe_sampler_state *state) -{ - if(!state) { - trace_dump_null(); - return; - } - - trace_dump_struct_begin("pipe_sampler_state"); - - trace_dump_member(uint, state, wrap_s); - trace_dump_member(uint, state, wrap_t); - trace_dump_member(uint, state, wrap_r); - trace_dump_member(uint, state, min_img_filter); - trace_dump_member(uint, state, min_mip_filter); - trace_dump_member(uint, state, mag_img_filter); - trace_dump_member(bool, state, compare_mode); - trace_dump_member(uint, state, compare_func); - trace_dump_member(bool, state, normalized_coords); - trace_dump_member(uint, state, prefilter); - trace_dump_member(float, state, shadow_ambient); - trace_dump_member(float, state, lod_bias); - trace_dump_member(float, state, min_lod); - trace_dump_member(float, state, max_lod); - trace_dump_member_array(float, state, border_color); - trace_dump_member(float, state, max_anisotropy); - - trace_dump_struct_end(); -} - - -void trace_dump_surface(const struct pipe_surface *state) -{ - if(!state) { - trace_dump_null(); - return; - } - - trace_dump_struct_begin("pipe_surface"); - - trace_dump_reference(&state->reference); - - trace_dump_member(format, state, format); - trace_dump_member(uint, state, width); - trace_dump_member(uint, state, height); - - trace_dump_member(uint, state, layout); - trace_dump_member(uint, state, offset); - trace_dump_member(uint, state, usage); - - trace_dump_member(ptr, state, texture); - trace_dump_member(uint, state, face); - trace_dump_member(uint, state, level); - trace_dump_member(uint, state, zslice); - - trace_dump_struct_end(); -} - +#include "tgsi/tgsi_parse.h" -void trace_dump_transfer(const struct pipe_transfer *state) +struct trace_shader * trace_shader_create(struct trace_context *tr_ctx, + const struct pipe_shader_state *state, + void *result, + enum trace_shader_type type) { - if(!state) { - trace_dump_null(); - return; - } + struct trace_shader *tr_shdr = CALLOC_STRUCT(trace_shader); - trace_dump_struct_begin("pipe_transfer"); + tr_shdr->state = result; + tr_shdr->type = type; + tr_shdr->tokens = tgsi_dup_tokens(state->tokens); - trace_dump_member(format, state, format); - trace_dump_member(uint, state, width); - trace_dump_member(uint, state, height); + /* works on context as well */ + trace_screen_add_to_list(tr_ctx, shaders, tr_shdr); - trace_dump_member_begin("block"); - trace_dump_block(&state->block); - trace_dump_member_end(); - - trace_dump_member(uint, state, nblocksx); - trace_dump_member(uint, state, nblocksy); - trace_dump_member(uint, state, stride); - trace_dump_member(uint, state, usage); - - trace_dump_member(ptr, state, texture); - trace_dump_member(uint, state, face); - trace_dump_member(uint, state, level); - trace_dump_member(uint, state, zslice); - - trace_dump_struct_end(); + return tr_shdr; } - -void trace_dump_vertex_buffer(const struct pipe_vertex_buffer *state) +void trace_shader_destroy(struct trace_context *tr_ctx, + struct trace_shader *tr_shdr) { - if(!state) { - trace_dump_null(); - return; - } + trace_screen_remove_from_list(tr_ctx, shaders, tr_shdr); - trace_dump_struct_begin("pipe_vertex_buffer"); - - trace_dump_member(uint, state, stride); - trace_dump_member(uint, state, max_index); - trace_dump_member(uint, state, buffer_offset); - trace_dump_member(buffer_ptr, state, buffer); - - trace_dump_struct_end(); -} - - -void trace_dump_vertex_element(const struct pipe_vertex_element *state) -{ - if(!state) { - trace_dump_null(); - return; + if (tr_shdr->replaced) { + if (tr_shdr->type == TRACE_SHADER_FRAGMENT) + tr_ctx->pipe->delete_fs_state(tr_ctx->pipe, tr_shdr->replaced); + else if (tr_shdr->type == TRACE_SHADER_VERTEX) + tr_ctx->pipe->delete_vs_state(tr_ctx->pipe, tr_shdr->replaced); + else + assert(0); } - trace_dump_struct_begin("pipe_vertex_element"); - - trace_dump_member(uint, state, src_offset); - - trace_dump_member(uint, state, vertex_buffer_index); - trace_dump_member(uint, state, nr_components); - - trace_dump_member(format, state, src_format); - - trace_dump_struct_end(); + FREE(tr_shdr->replaced_tokens); + FREE(tr_shdr->tokens); + FREE(tr_shdr); } diff --git a/src/gallium/drivers/trace/tr_state.h b/src/gallium/drivers/trace/tr_state.h index 513ed0ac98d..1c16042ee5a 100644 --- a/src/gallium/drivers/trace/tr_state.h +++ b/src/gallium/drivers/trace/tr_state.h @@ -1,78 +1,68 @@ -/************************************************************************** - * - * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. +/* + * Copyright 2009 VMware, Inc. * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. * - **************************************************************************/ - -#ifndef TR_STATE_H -#define TR_STATE_H - -#include "pipe/p_format.h" -#include "pipe/p_state.h" -#include "pipe/p_shader_tokens.h" - - -void trace_dump_format(enum pipe_format format); - -void trace_dump_block(const struct pipe_format_block *block); - -void trace_dump_template(const struct pipe_texture *templat); - - -void trace_dump_rasterizer_state(const struct pipe_rasterizer_state *state); - -void trace_dump_poly_stipple(const struct pipe_poly_stipple *state); - -void trace_dump_viewport_state(const struct pipe_viewport_state *state); - -void trace_dump_scissor_state(const struct pipe_scissor_state *state); - -void trace_dump_clip_state(const struct pipe_clip_state *state); + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ -void trace_dump_constant_buffer(const struct pipe_constant_buffer *state); +#ifndef TR_STATE_H_ +#define TR_STATE_H_ -void trace_dump_token(const struct tgsi_token *token); +#include "tr_context.h" -void trace_dump_shader_state(const struct pipe_shader_state *state); +struct tgsi_token; -void trace_dump_depth_stencil_alpha_state(const struct pipe_depth_stencil_alpha_state *state); +enum trace_shader_type { + TRACE_SHADER_FRAGMENT = 0, + TRACE_SHADER_VERTEX = 1, + TRACE_SHADER_GEOMETRY = 2, +}; -void trace_dump_blend_state(const struct pipe_blend_state *state); +struct trace_shader +{ + struct tr_list list; -void trace_dump_blend_color(const struct pipe_blend_color *state); + enum trace_shader_type type; -void trace_dump_framebuffer_state(const struct pipe_framebuffer_state *state); + void *state; + void *replaced; -void trace_dump_sampler_state(const struct pipe_sampler_state *state); + struct tgsi_token *tokens; + struct tgsi_token *replaced_tokens; -void trace_dump_surface(const struct pipe_surface *state); + boolean disabled; +}; -void trace_dump_transfer(const struct pipe_transfer *state); -void trace_dump_vertex_buffer(const struct pipe_vertex_buffer *state); +static INLINE struct trace_shader * +trace_shader(void *state) +{ + return (struct trace_shader *)state; +} -void trace_dump_vertex_element(const struct pipe_vertex_element *state); +struct trace_shader * trace_shader_create(struct trace_context *tr_ctx, + const struct pipe_shader_state *state, + void *result, + enum trace_shader_type type); +void trace_shader_destroy(struct trace_context *tr_ctx, + struct trace_shader *tr_shdr); -#endif /* TR_STATE_H */ +#endif diff --git a/src/gallium/include/pipe/internal/p_winsys_screen.h b/src/gallium/include/pipe/internal/p_winsys_screen.h index f4a29e63c7e..a1542dada70 100644 --- a/src/gallium/include/pipe/internal/p_winsys_screen.h +++ b/src/gallium/include/pipe/internal/p_winsys_screen.h @@ -140,6 +140,7 @@ struct pipe_winsys unsigned width, unsigned height, enum pipe_format format, unsigned usage, + unsigned tex_usage, unsigned *stride); diff --git a/src/gallium/include/pipe/p_compiler.h b/src/gallium/include/pipe/p_compiler.h index e6a67f8c2fd..c13cffceb0a 100644 --- a/src/gallium/include/pipe/p_compiler.h +++ b/src/gallium/include/pipe/p_compiler.h @@ -93,9 +93,11 @@ typedef int _Bool; #endif +#ifndef __HAIKU__ typedef unsigned int uint; -typedef unsigned char ubyte; typedef unsigned short ushort; +#endif +typedef unsigned char ubyte; #if 0 #define boolean bool diff --git a/src/gallium/include/pipe/p_config.h b/src/gallium/include/pipe/p_config.h index 63238ea46e9..de99957d9d0 100644 --- a/src/gallium/include/pipe/p_config.h +++ b/src/gallium/include/pipe/p_config.h @@ -102,6 +102,19 @@ /* + * Endian detection. + */ + +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) +#define PIPE_ARCH_LITTLE_ENDIAN +#elif defined(PIPE_ARCH_PPC) || defined(PIPE_ARCH_PPC_64) +#define PIPE_ARCH_BIG_ENDIAN +#else +#define PIPE_ARCH_UNKNOWN_ENDIAN +#endif + + +/* * Operating system family. * * See subsystem below for a more fine-grained distinction. @@ -119,10 +132,17 @@ #define PIPE_OS_SOLARIS #endif +#if defined(__APPLE__) +#define PIPE_OS_APPLE +#endif + #if defined(_WIN32) || defined(WIN32) #define PIPE_OS_WINDOWS #endif +#if defined(__HAIKU__) +#define PIPE_OS_HAIKU +#endif /* * Subsystem. diff --git a/src/gallium/include/pipe/p_context.h b/src/gallium/include/pipe/p_context.h index 57e966ac3b0..39620a71980 100644 --- a/src/gallium/include/pipe/p_context.h +++ b/src/gallium/include/pipe/p_context.h @@ -98,7 +98,7 @@ struct pipe_context { */ /*@{*/ struct pipe_query *(*create_query)( struct pipe_context *pipe, - unsigned query_type ); + unsigned query_type ); void (*destroy_query)(struct pipe_context *pipe, struct pipe_query *q); @@ -215,10 +215,12 @@ struct pipe_context { /** * Clear the specified set of currently bound buffers to specified values. + * The entire buffers are cleared (no scissor, no colormask, etc). * - * buffers is a bitfield of PIPE_CLEAR_* values. - * - * rgba is a pointer to an array of one float for each of r, g, b, a. + * \param buffers bitfield of PIPE_CLEAR_* values. + * \param rgba pointer to an array of one float for each of r, g, b, a. + * \param depth depth clear value in [0,1]. + * \param stencil stencil clear value */ void (*clear)(struct pipe_context *pipe, unsigned buffers, @@ -226,7 +228,9 @@ struct pipe_context { double depth, unsigned stencil); - /** Flush rendering (flags = bitmask of PIPE_FLUSH_x tokens) */ + /** Flush rendering + * \param flags bitmask of PIPE_FLUSH_x tokens) + */ void (*flush)( struct pipe_context *pipe, unsigned flags, struct pipe_fence_handle **fence ); @@ -242,10 +246,10 @@ struct pipe_context { * \param texture texture to check. * \param face cubemap face. Use 0 for non-cubemap texture. */ - unsigned int (*is_texture_referenced) (struct pipe_context *pipe, struct pipe_texture *texture, unsigned face, unsigned level); + /** * Check whether a buffer is referenced by an unflushed hw command. * The state-tracker uses this function to optimize away unnecessary @@ -255,7 +259,6 @@ struct pipe_context { * checked. * \param buf Buffer to check. */ - unsigned int (*is_buffer_referenced) (struct pipe_context *pipe, struct pipe_buffer *buf); }; diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h index 5c16884c16f..b01ab6d137c 100644 --- a/src/gallium/include/pipe/p_defines.h +++ b/src/gallium/include/pipe/p_defines.h @@ -132,6 +132,7 @@ enum pipe_texture_target { #define PIPE_TEX_FACE_NEG_Y 3 #define PIPE_TEX_FACE_POS_Z 4 #define PIPE_TEX_FACE_NEG_Z 5 +#define PIPE_TEX_FACE_MAX 6 #define PIPE_TEX_WRAP_REPEAT 0 #define PIPE_TEX_WRAP_CLAMP 1 @@ -158,14 +159,6 @@ enum pipe_texture_target { #define PIPE_TEX_COMPARE_NONE 0 #define PIPE_TEX_COMPARE_R_TO_TEXTURE 1 -#define PIPE_TEX_FACE_POS_X 0 -#define PIPE_TEX_FACE_NEG_X 1 -#define PIPE_TEX_FACE_POS_Y 2 -#define PIPE_TEX_FACE_NEG_Y 3 -#define PIPE_TEX_FACE_POS_Z 4 -#define PIPE_TEX_FACE_NEG_Z 5 -#define PIPE_TEX_FACE_MAX 6 - #define PIPE_TEXTURE_USAGE_RENDER_TARGET 0x1 #define PIPE_TEXTURE_USAGE_DISPLAY_TARGET 0x2 /* ie a backbuffer */ #define PIPE_TEXTURE_USAGE_PRIMARY 0x4 /* ie a frontbuffer */ @@ -198,9 +191,9 @@ enum pipe_texture_target { * Transfer object usage flags */ enum pipe_transfer_usage { - PIPE_TRANSFER_READ, - PIPE_TRANSFER_WRITE, - PIPE_TRANSFER_READ_WRITE /**< Read/modify/write */ + PIPE_TRANSFER_READ = (1 << 0), + PIPE_TRANSFER_WRITE = (1 << 1), + PIPE_TRANSFER_READ_WRITE = PIPE_TRANSFER_READ | PIPE_TRANSFER_WRITE /**< Read/modify/write */ }; @@ -281,9 +274,8 @@ enum pipe_transfer_usage { /** - * Implementation capabilities/limits - * Passed to pipe->get_param() - * XXX this will need some fine tuning... + * Implementation capabilities/limits which are queried through + * pipe_screen::get_param() and pipe_screen::get_paramf(). */ #define PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS 1 #define PIPE_CAP_NPOT_TEXTURES 2 @@ -312,6 +304,7 @@ enum pipe_transfer_usage { #define PIPE_CAP_TEXTURE_MIRROR_REPEAT 25 #define PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS 26 #define PIPE_CAP_TGSI_CONT_SUPPORTED 27 +#define PIPE_CAP_BLEND_EQUATION_SEPARATE 28 /** diff --git a/src/gallium/include/pipe/p_screen.h b/src/gallium/include/pipe/p_screen.h index 6cbdd759434..3f30c52a169 100644 --- a/src/gallium/include/pipe/p_screen.h +++ b/src/gallium/include/pipe/p_screen.h @@ -194,6 +194,7 @@ struct pipe_screen { unsigned width, unsigned height, enum pipe_format format, unsigned usage, + unsigned tex_usage, unsigned *stride); diff --git a/src/gallium/include/pipe/p_shader_tokens.h b/src/gallium/include/pipe/p_shader_tokens.h index b00cfe3423c..5fa6c9af30b 100644 --- a/src/gallium/include/pipe/p_shader_tokens.h +++ b/src/gallium/include/pipe/p_shader_tokens.h @@ -78,6 +78,7 @@ enum tgsi_file_type { TGSI_FILE_SAMPLER =5, TGSI_FILE_ADDRESS =6, TGSI_FILE_IMMEDIATE =7, + TGSI_FILE_LOOP =8, TGSI_FILE_COUNT /**< how many TGSI_FILE_ types */ }; @@ -152,13 +153,16 @@ struct tgsi_immediate unsigned Extended : 1; /**< BOOL */ }; -struct tgsi_immediate_float32 +union tgsi_immediate_data { float Float; }; -/* - * GL_NV_vertex_program +/* TGSI opcodes. + * + * For more information on semantics of opcodes and + * which APIs are known to use which opcodes, see + * auxiliary/tgsi/tgsi-instruction-set.txt */ #define TGSI_OPCODE_ARL 0 #define TGSI_OPCODE_MOV 1 @@ -177,62 +181,32 @@ struct tgsi_immediate_float32 #define TGSI_OPCODE_SLT 14 #define TGSI_OPCODE_SGE 15 #define TGSI_OPCODE_MAD 16 - -/* - * GL_ATI_fragment_shader - */ #define TGSI_OPCODE_SUB 17 -#define TGSI_OPCODE_DOT3 TGSI_OPCODE_DP3 -#define TGSI_OPCODE_DOT4 TGSI_OPCODE_DP4 -#define TGSI_OPCODE_LERP 18 +#define TGSI_OPCODE_LRP 18 #define TGSI_OPCODE_CND 19 -#define TGSI_OPCODE_CND0 20 -#define TGSI_OPCODE_DOT2ADD 21 - -/* - * GL_EXT_vertex_shader - */ -#define TGSI_OPCODE_INDEX 22 /* considered for removal */ -#define TGSI_OPCODE_NEGATE 23 /* considered for removal */ -#define TGSI_OPCODE_MADD TGSI_OPCODE_MAD -#define TGSI_OPCODE_FRAC 24 -#define TGSI_OPCODE_SETGE TGSI_OPCODE_SGE -#define TGSI_OPCODE_SETLT TGSI_OPCODE_SLT + /* gap */ +#define TGSI_OPCODE_DP2A 21 + /* gap */ +#define TGSI_OPCODE_FRC 24 #define TGSI_OPCODE_CLAMP 25 -#define TGSI_OPCODE_FLOOR 26 +#define TGSI_OPCODE_FLR 26 #define TGSI_OPCODE_ROUND 27 -#define TGSI_OPCODE_EXPBASE2 28 -#define TGSI_OPCODE_LOGBASE2 29 -#define TGSI_OPCODE_POWER 30 -#define TGSI_OPCODE_RECIP TGSI_OPCODE_RCP -#define TGSI_OPCODE_RECIPSQRT TGSI_OPCODE_RSQ -#define TGSI_OPCODE_CROSSPRODUCT 31 -#define TGSI_OPCODE_MULTIPLYMATRIX 32 /* considered for removal */ - -/* - * GL_NV_vertex_program1_1 - */ +#define TGSI_OPCODE_EX2 28 +#define TGSI_OPCODE_LG2 29 +#define TGSI_OPCODE_POW 30 +#define TGSI_OPCODE_XPD 31 + /* gap */ #define TGSI_OPCODE_ABS 33 #define TGSI_OPCODE_RCC 34 #define TGSI_OPCODE_DPH 35 - -/* - * GL_NV_fragment_program - */ #define TGSI_OPCODE_COS 36 #define TGSI_OPCODE_DDX 37 #define TGSI_OPCODE_DDY 38 -#define TGSI_OPCODE_EX2 TGSI_OPCODE_EXPBASE2 -#define TGSI_OPCODE_FLR TGSI_OPCODE_FLOOR -#define TGSI_OPCODE_FRC TGSI_OPCODE_FRAC #define TGSI_OPCODE_KILP 39 /* predicated kill */ -#define TGSI_OPCODE_LG2 TGSI_OPCODE_LOGBASE2 -#define TGSI_OPCODE_LRP TGSI_OPCODE_LERP #define TGSI_OPCODE_PK2H 40 #define TGSI_OPCODE_PK2US 41 #define TGSI_OPCODE_PK4B 42 #define TGSI_OPCODE_PK4UB 43 -#define TGSI_OPCODE_POW TGSI_OPCODE_POWER #define TGSI_OPCODE_RFL 44 #define TGSI_OPCODE_SEQ 45 #define TGSI_OPCODE_SFL 46 @@ -249,66 +223,29 @@ struct tgsi_immediate_float32 #define TGSI_OPCODE_UP4B 57 #define TGSI_OPCODE_UP4UB 58 #define TGSI_OPCODE_X2D 59 - -/* - * GL_NV_vertex_program2 - */ #define TGSI_OPCODE_ARA 60 #define TGSI_OPCODE_ARR 61 #define TGSI_OPCODE_BRA 62 #define TGSI_OPCODE_CAL 63 #define TGSI_OPCODE_RET 64 -#define TGSI_OPCODE_SSG 65 - -/* - * GL_ARB_vertex_program - */ -#define TGSI_OPCODE_SWZ 118 -#define TGSI_OPCODE_XPD TGSI_OPCODE_CROSSPRODUCT - -/* - * GL_ARB_fragment_program - */ +#define TGSI_OPCODE_SSG 65 /* SGN */ #define TGSI_OPCODE_CMP 66 -#define TGSI_OPCODE_KIL 116 /* conditional kill */ #define TGSI_OPCODE_SCS 67 #define TGSI_OPCODE_TXB 68 - -/* - * GL_NV_fragment_program_option - */ -/* No new opcode */ - -/* - * GL_NV_fragment_program2 - */ #define TGSI_OPCODE_NRM 69 #define TGSI_OPCODE_DIV 70 #define TGSI_OPCODE_DP2 71 -#define TGSI_OPCODE_DP2A TGSI_OPCODE_DOT2ADD #define TGSI_OPCODE_TXL 72 #define TGSI_OPCODE_BRK 73 #define TGSI_OPCODE_IF 74 -#define TGSI_OPCODE_LOOP 75 +#define TGSI_OPCODE_BGNFOR 75 #define TGSI_OPCODE_REP 76 #define TGSI_OPCODE_ELSE 77 #define TGSI_OPCODE_ENDIF 78 -#define TGSI_OPCODE_ENDLOOP 79 +#define TGSI_OPCODE_ENDFOR 79 #define TGSI_OPCODE_ENDREP 80 - -/* - * GL_NV_vertex_program2_option - */ - -/* - * GL_NV_vertex_program3 - */ #define TGSI_OPCODE_PUSHA 81 #define TGSI_OPCODE_POPA 82 - -/* - * GL_NV_gpu_program4 - */ #define TGSI_OPCODE_CEIL 83 #define TGSI_OPCODE_I2F 84 #define TGSI_OPCODE_NOT 85 @@ -323,103 +260,25 @@ struct tgsi_immediate_float32 #define TGSI_OPCODE_TXF 94 #define TGSI_OPCODE_TXQ 95 #define TGSI_OPCODE_CONT 96 - -/* - * GL_NV_vertex_program4 - */ -/* Same as GL_NV_gpu_program4 */ - -/* - * GL_NV_fragment_program4 - */ -/* Same as GL_NV_gpu_program4 */ - -/* - * GL_NV_geometry_program4 - */ -/* Same as GL_NV_gpu_program4 */ #define TGSI_OPCODE_EMIT 97 #define TGSI_OPCODE_ENDPRIM 98 - -/* - * GLSL - */ -#define TGSI_OPCODE_BGNLOOP2 99 +#define TGSI_OPCODE_BGNLOOP 99 #define TGSI_OPCODE_BGNSUB 100 -#define TGSI_OPCODE_ENDLOOP2 101 +#define TGSI_OPCODE_ENDLOOP 101 #define TGSI_OPCODE_ENDSUB 102 -#define TGSI_OPCODE_INT TGSI_OPCODE_TRUNC #define TGSI_OPCODE_NOISE1 103 #define TGSI_OPCODE_NOISE2 104 #define TGSI_OPCODE_NOISE3 105 #define TGSI_OPCODE_NOISE4 106 #define TGSI_OPCODE_NOP 107 - -/* - * ps_1_1 - */ -#define TGSI_OPCODE_TEXKILL TGSI_OPCODE_KIL - -/* - * ps_1_2 - */ -/* CMP - use TGSI_OPCODE_CND0 */ - -/* - * ps_1_3 - */ -/* CMP - use TGSI_OPCODE_CND0 */ - -/* - * ps_1_4 - */ -#define TGSI_OPCODE_TEXLD TGSI_OPCODE_TEX - -/* - * ps_2_0 - */ -#define TGSI_OPCODE_M4X4 TGSI_OPCODE_MULTIPLYMATRIX -#define TGSI_OPCODE_M4X3 108 -#define TGSI_OPCODE_M3X4 109 -#define TGSI_OPCODE_M3X3 110 -#define TGSI_OPCODE_M3X2 111 -#define TGSI_OPCODE_CRS TGSI_OPCODE_XPD + /* gap */ #define TGSI_OPCODE_NRM4 112 -#define TGSI_OPCODE_SINCOS TGSI_OPCODE_SCS -#define TGSI_OPCODE_TEXLDB TGSI_OPCODE_TXB -#define TGSI_OPCODE_DP2ADD TGSI_OPCODE_DP2A - -/* - * ps_2_x - */ -#define TGSI_OPCODE_CALL TGSI_OPCODE_CAL #define TGSI_OPCODE_CALLNZ 113 #define TGSI_OPCODE_IFC 114 -#define TGSI_OPCODE_BREAK TGSI_OPCODE_BRK #define TGSI_OPCODE_BREAKC 115 -#define TGSI_OPCODE_DSX TGSI_OPCODE_DDX -#define TGSI_OPCODE_DSY TGSI_OPCODE_DDY -#define TGSI_OPCODE_TEXLDD TGSI_OPCODE_TXD - -/* - * vs_1_1 - */ -#define TGSI_OPCODE_EXPP TGSI_OPCODE_EXP -#define TGSI_OPCODE_LOGP TGSI_OPCODE_LG2 - -/* - * vs_2_0 - */ -#define TGSI_OPCODE_SGN TGSI_OPCODE_SSG -#define TGSI_OPCODE_MOVA TGSI_OPCODE_ARR -/* EXPP - use TGSI_OPCODE_EX2 */ - -/* - * vs_2_x - */ - +#define TGSI_OPCODE_KIL 116 /* conditional kill */ #define TGSI_OPCODE_END 117 /* aka HALT */ - +#define TGSI_OPCODE_SWZ 118 #define TGSI_OPCODE_LAST 119 #define TGSI_SAT_NONE 0 /* do not saturate */ diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h index 4b590bdc906..626bedb35a8 100644 --- a/src/gallium/include/pipe/p_state.h +++ b/src/gallium/include/pipe/p_state.h @@ -75,11 +75,11 @@ struct pipe_surface; */ struct pipe_buffer { - struct pipe_reference reference; - struct pipe_screen *screen; - unsigned alignment; - unsigned usage; - unsigned size; + struct pipe_reference reference; + unsigned size; + struct pipe_screen *screen; + unsigned alignment; + unsigned usage; }; @@ -265,7 +265,6 @@ struct pipe_sampler_state unsigned compare_func:3; /**< PIPE_FUNC_x */ unsigned normalized_coords:1; /**< Are coords normalized to [0,1]? */ unsigned prefilter:4; /**< Wierd sampling state exposed by some api's */ - float shadow_ambient; /**< shadow test fail color/intensity */ float lod_bias; /**< LOD/lambda bias */ float min_lod, max_lod; /**< LOD clamp range, after bias */ float border_color[4]; @@ -287,10 +286,10 @@ struct pipe_surface unsigned offset; /**< offset from start of buffer, in bytes */ unsigned usage; /**< PIPE_BUFFER_USAGE_* */ + unsigned zslice; struct pipe_texture *texture; /**< texture into which this is a view */ unsigned face; unsigned level; - unsigned zslice; }; diff --git a/src/gallium/include/pipe/p_thread.h b/src/gallium/include/pipe/p_thread.h index ce8d79549ad..25e41482325 100644 --- a/src/gallium/include/pipe/p_thread.h +++ b/src/gallium/include/pipe/p_thread.h @@ -39,11 +39,13 @@ #include "util/u_debug.h" /* for assert */ -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) #include <pthread.h> /* POSIX threads headers */ #include <stdio.h> /* for perror() */ +#define PIPE_THREAD_HAVE_CONDVAR + typedef pthread_t pipe_thread; #define PIPE_THREAD_ROUTINE( name, param ) \ @@ -211,7 +213,7 @@ typedef unsigned pipe_condvar; */ typedef struct { -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) pthread_key_t key; #elif defined(PIPE_SUBSYSTEM_WINDOWS_USER) DWORD key; @@ -226,7 +228,7 @@ typedef struct { static INLINE void pipe_tsd_init(pipe_tsd *tsd) { -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) if (pthread_key_create(&tsd->key, NULL/*free*/) != 0) { perror("pthread_key_create(): failed to allocate key for thread specific data"); exit(-1); @@ -243,7 +245,7 @@ pipe_tsd_get(pipe_tsd *tsd) if (tsd->initMagic != (int) PIPE_TSD_INIT_MAGIC) { pipe_tsd_init(tsd); } -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) return pthread_getspecific(tsd->key); #elif defined(PIPE_SUBSYSTEM_WINDOWS_USER) assert(0); @@ -260,7 +262,7 @@ pipe_tsd_set(pipe_tsd *tsd, void *value) if (tsd->initMagic != (int) PIPE_TSD_INIT_MAGIC) { pipe_tsd_init(tsd); } -#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) +#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU) if (pthread_setspecific(tsd->key, value) != 0) { perror("pthread_set_specific() failed"); exit(-1); diff --git a/src/gallium/include/state_tracker/drm_api.h b/src/gallium/include/state_tracker/drm_api.h index 5790b2f6c7e..4d1259e1ee7 100644 --- a/src/gallium/include/state_tracker/drm_api.h +++ b/src/gallium/include/state_tracker/drm_api.h @@ -32,33 +32,38 @@ struct drm_api * Special buffer functions */ /*@{*/ - struct pipe_screen* (*create_screen)(int drm_fd, - struct drm_create_screen_arg *arg); - struct pipe_context* (*create_context)(struct pipe_screen *screen); + struct pipe_screen* (*create_screen)(struct drm_api *api, int drm_fd, + struct drm_create_screen_arg *arg); + struct pipe_context* (*create_context)(struct drm_api *api, + struct pipe_screen *screen); /*@}*/ /** * Special buffer functions */ /*@{*/ - boolean (*buffer_from_texture)(struct pipe_texture *texture, - struct pipe_buffer **buffer, - unsigned *stride); - struct pipe_buffer* (*buffer_from_handle)(struct pipe_screen *screen, - const char *name, - unsigned handle); - boolean (*handle_from_buffer)(struct pipe_screen *screen, - struct pipe_buffer *buffer, - unsigned *handle); - boolean (*global_handle_from_buffer)(struct pipe_screen *screen, - struct pipe_buffer *buffer, - unsigned *handle); + struct pipe_texture* + (*texture_from_shared_handle)(struct drm_api *api, + struct pipe_screen *screen, + struct pipe_texture *templ, + const char *name, + unsigned stride, + unsigned handle); + boolean (*shared_handle_from_texture)(struct drm_api *api, + struct pipe_screen *screen, + struct pipe_texture *texture, + unsigned *stride, + unsigned *handle); + boolean (*local_handle_from_texture)(struct drm_api *api, + struct pipe_screen *screen, + struct pipe_texture *texture, + unsigned *stride, + unsigned *handle); /*@}*/ + + void (*destroy)(struct drm_api *api); }; -/** - * A driver needs to export this symbol - */ -extern struct drm_api drm_api_hooks; +extern struct drm_api * drm_api_create(void); #endif diff --git a/src/gallium/state_trackers/dri/dri_context.c b/src/gallium/state_trackers/dri/dri_context.c index 45eaec4ed39..8819936fcaf 100644 --- a/src/gallium/state_trackers/dri/dri_context.c +++ b/src/gallium/state_trackers/dri/dri_context.c @@ -69,7 +69,7 @@ dri_create_context(const __GLcontextModes * visual, driParseConfigFiles(&ctx->optionCache, &screen->optionCache, sPriv->myNum, "dri"); - ctx->pipe = drm_api_hooks.create_context(screen->pipe_screen); + ctx->pipe = screen->api->create_context(screen->api, screen->pipe_screen); if (ctx->pipe == NULL) goto fail; @@ -100,7 +100,6 @@ void dri_destroy_context(__DRIcontextPrivate * cPriv) { struct dri_context *ctx = dri_context(cPriv); - struct dri_screen *screen = dri_screen(cPriv->driScreenPriv); /* No particular reason to wait for command completion before * destroying a context, but it is probably worthwhile flushing it @@ -109,9 +108,6 @@ dri_destroy_context(__DRIcontextPrivate * cPriv) */ st_flush(ctx->st, 0, NULL); - if (screen->dummyContext == ctx) - screen->dummyContext = NULL; - /* Also frees ctx->pipe? */ st_destroy_context(ctx->st); @@ -143,7 +139,6 @@ dri_make_current(__DRIcontextPrivate * cPriv, { if (cPriv) { struct dri_context *ctx = dri_context(cPriv); - struct dri_screen *screen = dri_screen(cPriv->driScreenPriv); struct dri_drawable *draw = dri_drawable(driDrawPriv); struct dri_drawable *read = dri_drawable(driReadPriv); struct st_context *old_st = st_get_current(); @@ -153,11 +148,6 @@ dri_make_current(__DRIcontextPrivate * cPriv, ++ctx->bind_count; - /* This is for situations in which we need a rendering context but - * there may not be any currently bound. - */ - screen->dummyContext = ctx; - if (ctx->dPriv != driDrawPriv) { ctx->dPriv = driDrawPriv; ctx->d_stamp = driDrawPriv->lastStamp - 1; diff --git a/src/gallium/state_trackers/dri/dri_drawable.c b/src/gallium/state_trackers/dri/dri_drawable.c index 5846390d02c..5cec9e329d9 100644 --- a/src/gallium/state_trackers/dri/dri_drawable.c +++ b/src/gallium/state_trackers/dri/dri_drawable.c @@ -36,6 +36,8 @@ #include "pipe/p_context.h" #include "pipe/p_screen.h" #include "pipe/p_inlines.h" +#include "main/mtypes.h" +#include "main/renderbuffer.h" #include "state_tracker/drm_api.h" #include "state_tracker/dri1_api.h" #include "state_tracker/st_public.h" @@ -44,16 +46,9 @@ #include "util/u_memory.h" -static void -dri_copy_to_front(__DRIdrawablePrivate * dPriv, - struct pipe_surface *from, - int x, int y, unsigned w, unsigned h) -{ - /* TODO send a message to the Xserver to copy to the real front buffer */ -} - static struct pipe_surface * -dri_surface_from_handle(struct pipe_screen *screen, +dri_surface_from_handle(struct drm_api *api, + struct pipe_screen *screen, unsigned handle, enum pipe_format format, unsigned width, unsigned height, unsigned pitch) @@ -61,11 +56,6 @@ dri_surface_from_handle(struct pipe_screen *screen, struct pipe_surface *surface = NULL; struct pipe_texture *texture = NULL; struct pipe_texture templat; - struct pipe_buffer *buf = NULL; - - buf = drm_api_hooks.buffer_from_handle(screen, "dri2 buffer", handle); - if (!buf) - return NULL; memset(&templat, 0, sizeof(templat)); templat.tex_usage |= PIPE_TEXTURE_USAGE_RENDER_TARGET; @@ -77,13 +67,13 @@ dri_surface_from_handle(struct pipe_screen *screen, templat.height[0] = height; pf_get_block(templat.format, &templat.block); - texture = screen->texture_blanket(screen, &templat, &pitch, buf); - - /* we don't need the buffer from this point on */ - pipe_buffer_reference(&buf, NULL); + texture = api->texture_from_shared_handle(api, screen, &templat, + "dri2 buffer", pitch, handle); - if (!texture) + if (!texture) { + debug_printf("%s: Failed to blanket the buffer with a texture\n", __func__); return NULL; + } surface = screen->get_tex_surface(screen, texture, 0, 0, 0, PIPE_BUFFER_USAGE_GPU_READ | @@ -95,17 +85,48 @@ dri_surface_from_handle(struct pipe_screen *screen, } /** + * Pixmaps have will have the same name of fake front and front. + */ +static boolean +dri2_check_if_pixmap(__DRIbuffer *buffers, int count) +{ + boolean found = FALSE; + boolean is_pixmap = FALSE; + unsigned name; + int i; + + for (i = 0; i < count; i++) { + switch (buffers[i].attachment) { + case __DRI_BUFFER_FRONT_LEFT: + case __DRI_BUFFER_FAKE_FRONT_LEFT: + if (found) { + is_pixmap = buffers[i].name == name; + } else { + name = buffers[i].name; + found = TRUE; + } + default: + continue; + } + } + + return is_pixmap; +} + +/** * This will be called a drawable is known to have been resized. */ void dri_get_buffers(__DRIdrawablePrivate * dPriv) { + struct dri_drawable *drawable = dri_drawable(dPriv); struct pipe_surface *surface = NULL; struct pipe_screen *screen = dri_screen(drawable->sPriv)->pipe_screen; __DRIbuffer *buffers = NULL; __DRIscreen *dri_screen = drawable->sPriv; __DRIdrawable *dri_drawable = drawable->dPriv; + struct drm_api *api = ((struct dri_screen*)(dri_screen->private))->api; boolean have_depth = FALSE; int i, count; @@ -138,36 +159,45 @@ dri_get_buffers(__DRIdrawablePrivate * dPriv) dri_drawable->pBackClipRects[0].x2 = dri_drawable->w; dri_drawable->pBackClipRects[0].y2 = dri_drawable->h; + if (drawable->old_num == count && + drawable->old_w == dri_drawable->w && + drawable->old_h == dri_drawable->h && + memcmp(drawable->old, buffers, sizeof(__DRIbuffer) * count) == 0) { + return; + } else { + drawable->old_num = count; + drawable->old_w = dri_drawable->w; + drawable->old_h = dri_drawable->h; + memcpy(drawable->old, buffers, sizeof(__DRIbuffer) * count); + } + + drawable->is_pixmap = dri2_check_if_pixmap(buffers, count); + for (i = 0; i < count; i++) { enum pipe_format format = 0; int index = 0; switch (buffers[i].attachment) { case __DRI_BUFFER_FRONT_LEFT: - index = ST_SURFACE_FRONT_LEFT; - format = PIPE_FORMAT_A8R8G8B8_UNORM; - break; + continue; case __DRI_BUFFER_FAKE_FRONT_LEFT: index = ST_SURFACE_FRONT_LEFT; - format = PIPE_FORMAT_A8R8G8B8_UNORM; + format = drawable->color_format; break; case __DRI_BUFFER_BACK_LEFT: index = ST_SURFACE_BACK_LEFT; - format = PIPE_FORMAT_A8R8G8B8_UNORM; + format = drawable->color_format; break; case __DRI_BUFFER_DEPTH: - index = ST_SURFACE_DEPTH; - format = PIPE_FORMAT_Z24S8_UNORM; - break; + case __DRI_BUFFER_DEPTH_STENCIL: case __DRI_BUFFER_STENCIL: index = ST_SURFACE_DEPTH; - format = PIPE_FORMAT_Z24S8_UNORM; + format = drawable->depth_stencil_format; break; case __DRI_BUFFER_ACCUM: default: assert(0); } - assert(buffers[i].cpp == 4); if (index == ST_SURFACE_DEPTH) { if (have_depth) @@ -176,12 +206,25 @@ dri_get_buffers(__DRIdrawablePrivate * dPriv) have_depth = TRUE; } - surface = dri_surface_from_handle(screen, + surface = dri_surface_from_handle(api, + screen, buffers[i].name, format, dri_drawable->w, dri_drawable->h, buffers[i].pitch); + switch (buffers[i].attachment) { + case __DRI_BUFFER_FAKE_FRONT_LEFT: + case __DRI_BUFFER_BACK_LEFT: + drawable->color_format = surface->format; + break; + case __DRI_BUFFER_DEPTH: + case __DRI_BUFFER_DEPTH_STENCIL: + case __DRI_BUFFER_STENCIL: + drawable->depth_stencil_format = surface->format; + break; + } + st_set_framebuffer_surface(drawable->stfb, index, surface); pipe_surface_reference(&surface, NULL); } @@ -189,13 +232,60 @@ dri_get_buffers(__DRIdrawablePrivate * dPriv) st_resize_framebuffer(drawable->stfb, dri_drawable->w, dri_drawable->h); } +/** + * These are used for GLX_EXT_texture_from_pixmap + */ +void dri2_set_tex_buffer2(__DRIcontext *pDRICtx, GLint target, + GLint format, __DRIdrawable *dPriv) +{ + struct dri_drawable *drawable = dri_drawable(dPriv); + struct pipe_surface *ps; + + if (!drawable->stfb->Base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer) { + struct gl_renderbuffer *rb = + st_new_renderbuffer_fb(drawable->color_format, 0 /*XXX*/, FALSE); + _mesa_add_renderbuffer(&drawable->stfb->Base, BUFFER_FRONT_LEFT, rb); + } + + dri_get_buffers(drawable->dPriv); + st_get_framebuffer_surface(drawable->stfb, ST_SURFACE_FRONT_LEFT, &ps); + + st_bind_texture_surface(ps, target == GL_TEXTURE_2D ? ST_TEXTURE_2D : + ST_TEXTURE_RECT, 0, drawable->color_format); +} + +void dri2_set_tex_buffer(__DRIcontext *pDRICtx, GLint target, + __DRIdrawable *dPriv) +{ + dri2_set_tex_buffer2(pDRICtx, target, GLX_TEXTURE_FORMAT_RGBA_EXT, dPriv); +} + void dri_flush_frontbuffer(struct pipe_screen *screen, struct pipe_surface *surf, void *context_private) { struct dri_context *ctx = (struct dri_context *)context_private; + struct dri_drawable *drawable = dri_drawable(ctx->dPriv); + __DRIdrawable *dri_drawable = ctx->dPriv; + __DRIscreen *dri_screen = ctx->sPriv; + + /* XXX Does this function get called with DRI1? */ - dri_copy_to_front(ctx->dPriv, surf, 0, 0, surf->width, surf->height); + if (ctx->dPriv == NULL) { + debug_printf("%s: no drawable bound to context\n", __func__); + return; + } + +#if 0 + /* TODO if rendering to pixmaps is slow enable this code. */ + if (drawable->is_pixmap) + return; +#else + (void)drawable; +#endif + + (*dri_screen->dri2.loader->flushFrontBuffer)(dri_drawable, + dri_drawable->loaderPrivate); } /** @@ -206,10 +296,8 @@ dri_create_buffer(__DRIscreenPrivate * sPriv, __DRIdrawablePrivate * dPriv, const __GLcontextModes * visual, boolean isPixmap) { - enum pipe_format colorFormat, depthFormat, stencilFormat; struct dri_screen *screen = sPriv->private; struct dri_drawable *drawable = NULL; - struct pipe_screen *pscreen = screen->pipe_screen; int i; if (isPixmap) @@ -219,39 +307,43 @@ dri_create_buffer(__DRIscreenPrivate * sPriv, if (drawable == NULL) goto fail; - /* XXX: todo: use the pipe_screen queries to figure out which - * render targets are supportable. - */ - assert(visual->redBits == 8); - assert(visual->depthBits == 24 || visual->depthBits == 0); - assert(visual->stencilBits == 8 || visual->stencilBits == 0); - - colorFormat = PIPE_FORMAT_A8R8G8B8_UNORM; - - if (visual->depthBits) { - if (pscreen->is_format_supported(pscreen, PIPE_FORMAT_Z24S8_UNORM, - PIPE_TEXTURE_2D, - PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0)) - depthFormat = PIPE_FORMAT_Z24S8_UNORM; - else - depthFormat = PIPE_FORMAT_S8Z24_UNORM; - } else - depthFormat = PIPE_FORMAT_NONE; - - if (visual->stencilBits) { - if (pscreen->is_format_supported(pscreen, PIPE_FORMAT_Z24S8_UNORM, - PIPE_TEXTURE_2D, - PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0)) - stencilFormat = PIPE_FORMAT_Z24S8_UNORM; + if (visual->redBits == 8) { + if (visual->alphaBits == 8) + drawable->color_format = PIPE_FORMAT_A8R8G8B8_UNORM; else - stencilFormat = PIPE_FORMAT_S8Z24_UNORM; - } else - stencilFormat = PIPE_FORMAT_NONE; + drawable->color_format = PIPE_FORMAT_X8R8G8B8_UNORM; + } else { + drawable->color_format = PIPE_FORMAT_R5G6B5_UNORM; + } + + switch(visual->depthBits) { + default: + case 0: + drawable->depth_stencil_format = PIPE_FORMAT_NONE; + break; + case 16: + drawable->depth_stencil_format = PIPE_FORMAT_Z16_UNORM; + break; + case 24: + if (visual->stencilBits == 0) { + drawable->depth_stencil_format = (screen->d_depth_bits_last) ? + PIPE_FORMAT_X8Z24_UNORM: + PIPE_FORMAT_Z24X8_UNORM; + } else { + drawable->depth_stencil_format = (screen->sd_depth_bits_last) ? + PIPE_FORMAT_S8Z24_UNORM: + PIPE_FORMAT_Z24S8_UNORM; + } + break; + case 32: + drawable->depth_stencil_format = PIPE_FORMAT_Z32_UNORM; + break; + } drawable->stfb = st_create_framebuffer(visual, - colorFormat, - depthFormat, - stencilFormat, + drawable->color_format, + drawable->depth_stencil_format, + drawable->depth_stencil_format, dPriv->w, dPriv->h, (void *)drawable); if (drawable->stfb == NULL) @@ -262,17 +354,19 @@ dri_create_buffer(__DRIscreenPrivate * sPriv, dPriv->driverPrivate = (void *)drawable; /* setup dri2 buffers information */ + /* TODO incase of double buffer visual, delay fake creation */ i = 0; drawable->attachments[i++] = __DRI_BUFFER_FRONT_LEFT; -#if 0 - /* TODO incase of double buffer visual, delay fake creation */ - drawable->attachments[i++] = __DRI_BUFFER_FAKE_FRONT_LEFT; -#endif + if (visual->doubleBufferMode) drawable->attachments[i++] = __DRI_BUFFER_BACK_LEFT; - if (visual->depthBits) + else + drawable->attachments[i++] = __DRI_BUFFER_FAKE_FRONT_LEFT; + if (visual->depthBits && visual->stencilBits) + drawable->attachments[i++] = __DRI_BUFFER_DEPTH_STENCIL; + else if (visual->depthBits) drawable->attachments[i++] = __DRI_BUFFER_DEPTH; - if (visual->stencilBits) + else if (visual->stencilBits) drawable->attachments[i++] = __DRI_BUFFER_STENCIL; drawable->num_attachments = i; diff --git a/src/gallium/state_trackers/dri/dri_drawable.h b/src/gallium/state_trackers/dri/dri_drawable.h index 78a66624aaf..b910930db42 100644 --- a/src/gallium/state_trackers/dri/dri_drawable.h +++ b/src/gallium/state_trackers/dri/dri_drawable.h @@ -33,6 +33,7 @@ struct pipe_surface; struct pipe_fence_handle; struct st_framebuffer; +struct dri_context; #define DRI_SWAP_FENCES_MAX 8 #define DRI_SWAP_FENCES_MASK 7 @@ -46,6 +47,13 @@ struct dri_drawable unsigned attachments[8]; unsigned num_attachments; + boolean is_pixmap; + + __DRIbuffer old[8]; + unsigned old_num; + unsigned old_w; + unsigned old_h; + /* gallium */ struct st_framebuffer *stfb; struct pipe_fence_handle *swap_fences[DRI_SWAP_FENCES_MAX]; @@ -53,6 +61,9 @@ struct dri_drawable unsigned int tail; unsigned int desired_fences; unsigned int cur_fences; + + enum pipe_format color_format; + enum pipe_format depth_stencil_format; }; static INLINE struct dri_drawable * @@ -82,6 +93,12 @@ void dri_get_buffers(__DRIdrawablePrivate * dPriv); void dri_destroy_buffer(__DRIdrawablePrivate * dPriv); +void dri2_set_tex_buffer2(__DRIcontext *pDRICtx, GLint target, + GLint glx_texture_format, __DRIdrawable *dPriv); + +void dri2_set_tex_buffer(__DRIcontext *pDRICtx, GLint target, + __DRIdrawable *dPriv); + void dri1_update_drawables(struct dri_context *ctx, struct dri_drawable *draw, struct dri_drawable *read); diff --git a/src/gallium/state_trackers/dri/dri_extensions.c b/src/gallium/state_trackers/dri/dri_extensions.c index 0c59d42d5c6..4349a4d1d2f 100644 --- a/src/gallium/state_trackers/dri/dri_extensions.c +++ b/src/gallium/state_trackers/dri/dri_extensions.c @@ -33,12 +33,15 @@ #include "dri_context.h" #include "state_tracker/st_context.h" +#define need_GL_ARB_map_buffer_range #define need_GL_ARB_multisample #define need_GL_ARB_occlusion_query #define need_GL_ARB_point_parameters +#define need_GL_ARB_shader_objects #define need_GL_ARB_texture_compression #define need_GL_ARB_vertex_buffer_object #define need_GL_ARB_vertex_program +#define need_GL_ARB_vertex_shader #define need_GL_ARB_window_pos #define need_GL_EXT_blend_color #define need_GL_EXT_blend_equation_separate @@ -50,16 +53,24 @@ #define need_GL_EXT_multi_draw_arrays #define need_GL_EXT_secondary_color #define need_GL_NV_vertex_program +#define need_GL_VERSION_2_0 +#define need_GL_VERSION_2_1 #include "extension_helper.h" /** * Extension strings exported by the driver. */ const struct dri_extension card_extensions[] = { + {"GL_ARB_fragment_shader", NULL}, + {"GL_ARB_map_buffer_range", GL_ARB_map_buffer_range_functions}, {"GL_ARB_multisample", GL_ARB_multisample_functions}, {"GL_ARB_multitexture", NULL}, {"GL_ARB_occlusion_query", GL_ARB_occlusion_query_functions}, + {"GL_ARB_pixel_buffer_object", NULL}, {"GL_ARB_point_parameters", GL_ARB_point_parameters_functions}, + {"GL_ARB_shading_language_100", GL_VERSION_2_0_functions }, + {"GL_ARB_shading_language_120", GL_VERSION_2_1_functions }, + {"GL_ARB_shader_objects", GL_ARB_shader_objects_functions}, {"GL_ARB_texture_border_clamp", NULL}, {"GL_ARB_texture_compression", GL_ARB_texture_compression_functions}, {"GL_ARB_texture_cube_map", NULL}, @@ -69,12 +80,11 @@ const struct dri_extension card_extensions[] = { {"GL_ARB_texture_mirrored_repeat", NULL}, {"GL_ARB_texture_rectangle", NULL}, {"GL_ARB_vertex_buffer_object", GL_ARB_vertex_buffer_object_functions}, - {"GL_ARB_pixel_buffer_object", NULL}, + {"GL_ARB_vertex_shader", GL_ARB_vertex_shader_functions}, {"GL_ARB_vertex_program", GL_ARB_vertex_program_functions}, {"GL_ARB_window_pos", GL_ARB_window_pos_functions}, {"GL_EXT_blend_color", GL_EXT_blend_color_functions}, - {"GL_EXT_blend_equation_separate", - GL_EXT_blend_equation_separate_functions}, + {"GL_EXT_blend_equation_separate", GL_EXT_blend_equation_separate_functions}, {"GL_EXT_blend_func_separate", GL_EXT_blend_func_separate_functions}, {"GL_EXT_blend_minmax", GL_EXT_blend_minmax_functions}, {"GL_EXT_blend_subtract", NULL}, diff --git a/src/gallium/state_trackers/dri/dri_screen.c b/src/gallium/state_trackers/dri/dri_screen.c index d3392ee690b..884b6d50111 100644 --- a/src/gallium/state_trackers/dri/dri_screen.c +++ b/src/gallium/state_trackers/dri/dri_screen.c @@ -57,82 +57,144 @@ PUBLIC const char __driConfigOptions[] = const uint __driNConfigOptions = 3; +static const __DRItexBufferExtension dri2TexBufferExtension = { + { __DRI_TEX_BUFFER, __DRI_TEX_BUFFER_VERSION }, + dri2_set_tex_buffer, + dri2_set_tex_buffer2, +}; + static const __DRIextension *dri_screen_extensions[] = { &driReadDrawableExtension, &driCopySubBufferExtension.base, &driSwapControlExtension.base, &driFrameTrackingExtension.base, &driMediaStreamCounterExtension.base, + &dri2TexBufferExtension.base, NULL }; struct dri1_api *__dri1_api_hooks = NULL; static const __DRIconfig ** -dri_fill_in_modes(__DRIscreenPrivate * psp, - unsigned pixel_bits, unsigned depth_bits, - unsigned stencil_bits, GLboolean have_back_buffer) +dri_fill_in_modes(struct dri_screen *screen, + unsigned pixel_bits) { - __DRIconfig **configs; - __GLcontextModes *m; + __DRIconfig **configs = NULL; unsigned num_modes; - uint8_t depth_bits_array[3]; - uint8_t stencil_bits_array[3]; + uint8_t depth_bits_array[5]; + uint8_t stencil_bits_array[5]; uint8_t msaa_samples_array[1]; unsigned depth_buffer_factor; unsigned back_buffer_factor; unsigned msaa_samples_factor; - GLenum fb_format; - GLenum fb_type; - int i; + struct pipe_screen *p_screen = screen->pipe_screen; + boolean pf_r5g6b5, pf_a8r8g8b8, pf_x8r8g8b8; + boolean pf_z16, pf_x8z24, pf_z24x8, pf_s8z24, pf_z24s8, pf_z32; static const GLenum back_buffer_modes[] = { GLX_NONE, GLX_SWAP_UNDEFINED_OML, GLX_SWAP_COPY_OML }; - /* TODO probe the hardware of what is supports */ depth_bits_array[0] = 0; - depth_bits_array[1] = 24; - depth_bits_array[2] = 24; - - stencil_bits_array[0] = 0; /* no depth or stencil */ - stencil_bits_array[1] = 0; /* z24x8 */ - stencil_bits_array[2] = 8; /* z24s8 */ + stencil_bits_array[0] = 0; + depth_buffer_factor = 1; + + pf_z16 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_Z16_UNORM, + PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0); + pf_z32 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_Z32_UNORM, + PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0); + pf_x8z24 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_X8Z24_UNORM, + PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0); + pf_z24x8 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_Z24X8_UNORM, + PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0); + pf_s8z24 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_S8Z24_UNORM, + PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0); + pf_z24s8 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_Z24S8_UNORM, + PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0); + pf_r5g6b5 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_R5G6B5_UNORM, + PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_RENDER_TARGET, 0); + pf_a8r8g8b8 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_A8R8G8B8_UNORM, + PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_RENDER_TARGET, 0); + pf_x8r8g8b8 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_X8R8G8B8_UNORM, + PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_RENDER_TARGET, 0); + + if (pf_z16) { + depth_bits_array[depth_buffer_factor] = 16; + stencil_bits_array[depth_buffer_factor++] = 0; + } + if (pf_x8z24 || pf_z24x8) { + depth_bits_array[depth_buffer_factor] = 24; + stencil_bits_array[depth_buffer_factor++] = 0; + screen->d_depth_bits_last = pf_x8z24; + } + if (pf_s8z24 || pf_z24s8) { + depth_bits_array[depth_buffer_factor] = 24; + stencil_bits_array[depth_buffer_factor++] = 8; + screen->sd_depth_bits_last = pf_s8z24; + } + if (pf_z32) { + depth_bits_array[depth_buffer_factor] = 32; + stencil_bits_array[depth_buffer_factor++] = 0; + } msaa_samples_array[0] = 0; - - depth_buffer_factor = 3; back_buffer_factor = 3; msaa_samples_factor = 1; num_modes = depth_buffer_factor * back_buffer_factor * msaa_samples_factor * 4; - if (pixel_bits == 16) { - fb_format = GL_RGB; - fb_type = GL_UNSIGNED_SHORT_5_6_5; + if (pixel_bits == 16 && pf_r5g6b5) { + configs = driCreateConfigs(GL_RGB, GL_UNSIGNED_SHORT_5_6_5, + depth_bits_array, stencil_bits_array, + depth_buffer_factor, back_buffer_modes, + back_buffer_factor, + msaa_samples_array, 1); } else { - fb_format = GL_BGRA; - fb_type = GL_UNSIGNED_INT_8_8_8_8_REV; + __DRIconfig **configs_a8r8g8b8 = NULL; + __DRIconfig **configs_x8r8g8b8 = NULL; + + if (pf_a8r8g8b8) + configs_a8r8g8b8 = driCreateConfigs(GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, + depth_bits_array, + stencil_bits_array, + depth_buffer_factor, + back_buffer_modes, + back_buffer_factor, + msaa_samples_array, 1); + if (pf_x8r8g8b8) + configs_x8r8g8b8 = driCreateConfigs(GL_BGR, GL_UNSIGNED_INT_8_8_8_8_REV, + depth_bits_array, + stencil_bits_array, + depth_buffer_factor, + back_buffer_modes, + back_buffer_factor, + msaa_samples_array, 1); + + if (configs_a8r8g8b8 && configs_x8r8g8b8) + configs = driConcatConfigs(configs_x8r8g8b8, configs_a8r8g8b8); + else if (configs_a8r8g8b8) + configs = configs_a8r8g8b8; + else if (configs_x8r8g8b8) + configs = configs_x8r8g8b8; + else + configs = NULL; } - configs = driCreateConfigs(fb_format, fb_type, - depth_bits_array, - stencil_bits_array, depth_buffer_factor, - back_buffer_modes, back_buffer_factor, - msaa_samples_array, msaa_samples_factor); if (configs == NULL) { debug_printf("%s: driCreateConfigs failed\n", __FUNCTION__); return NULL; } - for (i = 0; configs[i]; i++) { - m = &configs[i]->modes; - if ((m->stencilBits != 0) && (m->stencilBits != stencil_bits)) { - m->visualRating = GLX_SLOW_CONFIG; - } - } - return (const const __DRIconfig **)configs; } @@ -170,6 +232,7 @@ dri_init_screen(__DRIscreenPrivate * sPriv) if (!screen) return NULL; + screen->api = drm_api_create(); screen->sPriv = sPriv; screen->fd = sPriv->fd; screen->drmLock = (drmLock *) & sPriv->pSAREA->lock; @@ -187,7 +250,7 @@ dri_init_screen(__DRIscreenPrivate * sPriv) dri_copy_version(&arg.drm_version, &sPriv->drm_version); arg.api = NULL; - screen->pipe_screen = drm_api_hooks.create_screen(screen->fd, &arg.base); + screen->pipe_screen = screen->api->create_screen(screen->api, screen->fd, &arg.base); if (!screen->pipe_screen || !arg.api) { debug_printf("%s: failed to create dri1 screen\n", __FUNCTION__); @@ -200,7 +263,13 @@ dri_init_screen(__DRIscreenPrivate * sPriv) driParseOptionInfo(&screen->optionCache, __driConfigOptions, __driNConfigOptions); - configs = dri_fill_in_modes(sPriv, sPriv->fbBPP, 24, 8, 1); + /** + * FIXME: If the driver supports format conversion swapbuffer blits, we might + * want to support other color bit depths than the server is currently + * using. + */ + + configs = dri_fill_in_modes(screen, sPriv->fbBPP); if (!configs) goto out_no_configs; @@ -230,13 +299,14 @@ dri_init_screen2(__DRIscreenPrivate * sPriv) if (!screen) goto fail; + screen->api = drm_api_create(); screen->sPriv = sPriv; screen->fd = sPriv->fd; sPriv->private = (void *)screen; sPriv->extensions = dri_screen_extensions; arg.mode = DRM_CREATE_NORMAL; - screen->pipe_screen = drm_api_hooks.create_screen(screen->fd, &arg); + screen->pipe_screen = screen->api->create_screen(screen->api, screen->fd, &arg); if (!screen->pipe_screen) { debug_printf("%s: failed to create pipe_screen\n", __FUNCTION__); goto fail; @@ -248,7 +318,7 @@ dri_init_screen2(__DRIscreenPrivate * sPriv) driParseOptionInfo(&screen->optionCache, __driConfigOptions, __driNConfigOptions); - return dri_fill_in_modes(sPriv, 4 * 8, 24, 8, 1); + return dri_fill_in_modes(screen, 32); fail: return NULL; } diff --git a/src/gallium/state_trackers/dri/dri_screen.h b/src/gallium/state_trackers/dri/dri_screen.h index 100d9e50e07..f6c56d0f0c5 100644 --- a/src/gallium/state_trackers/dri/dri_screen.h +++ b/src/gallium/state_trackers/dri/dri_screen.h @@ -49,19 +49,16 @@ struct dri_screen */ driOptionCache optionCache; - /** - * Temporary(?) context to use for SwapBuffers or other situations in - * which we need a rendering context, but none is currently bound. - */ - struct dri_context *dummyContext; - /* drm */ int fd; drmLock *drmLock; /* gallium */ + struct drm_api *api; struct pipe_winsys *pipe_winsys; struct pipe_screen *pipe_screen; + boolean d_depth_bits_last; + boolean sd_depth_bits_last; }; /** cast wrapper */ diff --git a/src/gallium/state_trackers/egl/Makefile b/src/gallium/state_trackers/egl/Makefile index 692a3c8b76f..e825aa718b6 100644 --- a/src/gallium/state_trackers/egl/Makefile +++ b/src/gallium/state_trackers/egl/Makefile @@ -1,29 +1,19 @@ -TARGET = libegldrm.a -CFILES = $(wildcard ./*.c) -OBJECTS = $(patsubst ./%.c,./%.o,$(CFILES)) -GALLIUMDIR = ../.. -TOP = ../../../.. +TOP = ../../../.. +include $(TOP)/configs/current -include ${TOP}/configs/current +LIBNAME = egldrm -CFLAGS := \ - -I${GALLIUMDIR}/include \ - -I${GALLIUMDIR}/auxiliary \ - -I${TOP}/src/mesa/drivers/dri/common \ - -I${TOP}/src/mesa \ - -I$(TOP)/include \ - -I$(TOP)/src/egl/main \ - ${LIBDRM_CFLAGS} \ - ${CFLAGS} +LIBRARY_INCLUDES = \ + -I$(TOP)/src/gallium/include \ + -I$(TOP)/src/gallium/auxiliary \ + -I$(TOP)/src/mesa/drivers/dri/common \ + -I$(TOP)/src/mesa \ + -I$(TOP)/include \ + -I$(TOP)/src/egl/main \ + $(shell pkg-config --cflags-only-I libdrm) -############################################# -.PHONY = all clean +C_SOURCES = $(wildcard ./*.c) -all: ${TARGET} -${TARGET}: ${OBJECTS} - ar rcs $@ $^ - -clean: - rm -rf ${OBJECTS} ${TARGET} +include ../../Makefile.template diff --git a/src/gallium/state_trackers/egl/egl_context.c b/src/gallium/state_trackers/egl/egl_context.c index 36548fae263..c4f7361ca0b 100644 --- a/src/gallium/state_trackers/egl/egl_context.c +++ b/src/gallium/state_trackers/egl/egl_context.c @@ -83,23 +83,16 @@ const struct dri_extension card_extensions[] = { {NULL, NULL} }; -EGLContext -drm_create_context(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, EGLContext share_list, const EGLint *attrib_list) +_EGLContext * +drm_create_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, _EGLContext *share_list, const EGLint *attrib_list) { - struct drm_device *dev = (struct drm_device *)drv; + struct drm_device *dev = lookup_drm_device(dpy); struct drm_context *ctx; struct drm_context *share = NULL; struct st_context *st_share = NULL; - _EGLConfig *conf; int i; __GLcontextModes *visual; - conf = _eglLookupConfig(drv, dpy, config); - if (!conf) { - _eglError(EGL_BAD_CONFIG, "eglCreateContext"); - return EGL_NO_CONTEXT; - } - for (i = 0; attrib_list && attrib_list[i] != EGL_NONE; i++) { switch (attrib_list[i]) { /* no attribs defined for now */ @@ -113,9 +106,9 @@ drm_create_context(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, EGLContext if (!ctx) goto err_c; - _eglInitContext(drv, dpy, &ctx->base, config, attrib_list); + _eglInitContext(drv, &ctx->base, conf, attrib_list); - ctx->pipe = drm_api_hooks.create_context(dev->screen); + ctx->pipe = dev->api->create_context(dev->api, dev->screen); if (!ctx->pipe) goto err_pipe; @@ -129,28 +122,21 @@ drm_create_context(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, EGLContext if (!ctx->st) goto err_gl; - /* generate handle and insert into hash table */ - _eglSaveContext(&ctx->base); - assert(_eglGetContextHandle(&ctx->base)); - - return _eglGetContextHandle(&ctx->base); + return &ctx->base; err_gl: ctx->pipe->destroy(ctx->pipe); err_pipe: free(ctx); err_c: - return EGL_NO_CONTEXT; + return NULL; } EGLBoolean -drm_destroy_context(_EGLDriver *drv, EGLDisplay dpy, EGLContext context) +drm_destroy_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *context) { struct drm_context *c = lookup_drm_context(context); - _eglRemoveContext(&c->base); - if (c->base.IsBound) { - c->base.DeletePending = EGL_TRUE; - } else { + if (!_eglIsContextBound(&c->base)) { st_destroy_context(c->st); c->pipe->destroy(c->pipe); free(c); @@ -159,7 +145,7 @@ drm_destroy_context(_EGLDriver *drv, EGLDisplay dpy, EGLContext context) } EGLBoolean -drm_make_current(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext context) +drm_make_current(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw, _EGLSurface *read, _EGLContext *context) { struct drm_surface *readSurf = lookup_drm_surface(read); struct drm_surface *drawSurf = lookup_drm_surface(draw); diff --git a/src/gallium/state_trackers/egl/egl_surface.c b/src/gallium/state_trackers/egl/egl_surface.c index 489aa8d9af5..69e2d6b7081 100644 --- a/src/gallium/state_trackers/egl/egl_surface.c +++ b/src/gallium/state_trackers/egl/egl_surface.c @@ -67,39 +67,21 @@ drm_create_framebuffer(const __GLcontextModes *visual, } static void -drm_create_texture(_EGLDriver *drv, +drm_create_texture(_EGLDisplay *dpy, struct drm_screen *scrn, unsigned w, unsigned h) { - struct drm_device *dev = (struct drm_device *)drv; + struct drm_device *dev = lookup_drm_device(dpy); struct pipe_screen *screen = dev->screen; struct pipe_surface *surface; struct pipe_texture *texture; struct pipe_texture templat; - struct pipe_buffer *buf; - unsigned stride = 1024; + struct pipe_buffer *buf = NULL; unsigned pitch = 0; - unsigned size = 0; - - /* ugly */ - if (stride < w) - stride = 2048; - - pitch = stride * 4; - size = h * 2 * pitch; - - buf = pipe_buffer_create(screen, - 0, /* alignment */ - PIPE_BUFFER_USAGE_GPU_READ_WRITE | - PIPE_BUFFER_USAGE_CPU_READ_WRITE, - size); - - if (!buf) - goto err_buf; memset(&templat, 0, sizeof(templat)); - templat.tex_usage |= PIPE_TEXTURE_USAGE_DISPLAY_TARGET; - templat.tex_usage |= PIPE_TEXTURE_USAGE_RENDER_TARGET; + templat.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET; + templat.tex_usage |= PIPE_TEXTURE_USAGE_PRIMARY; templat.target = PIPE_TEXTURE_2D; templat.last_level = 0; templat.depth[0] = 1; @@ -108,10 +90,9 @@ drm_create_texture(_EGLDriver *drv, templat.height[0] = h; pf_get_block(templat.format, &templat.block); - texture = screen->texture_blanket(dev->screen, - &templat, - &pitch, - buf); + texture = screen->texture_create(dev->screen, + &templat); + if (!texture) goto err_tex; @@ -125,14 +106,13 @@ drm_create_texture(_EGLDriver *drv, if (!surface) goto err_surf; - scrn->tex = texture; scrn->surface = surface; - scrn->buffer = buf; scrn->front.width = w; scrn->front.height = h; scrn->front.pitch = pitch; - drm_api_hooks.handle_from_buffer(screen, scrn->buffer, &scrn->front.handle); + dev->api->local_handle_from_texture(dev->api, screen, texture, + &scrn->front.pitch, &scrn->front.handle); if (0) goto err_handle; @@ -144,7 +124,6 @@ err_surf: pipe_texture_reference(&texture, NULL); err_tex: pipe_buffer_reference(&buf, NULL); -err_buf: return; } @@ -153,9 +132,9 @@ err_buf: */ void -drm_takedown_shown_screen(_EGLDriver *drv, struct drm_screen *screen) +drm_takedown_shown_screen(_EGLDisplay *dpy, struct drm_screen *screen) { - struct drm_device *dev = (struct drm_device *)drv; + struct drm_device *dev = lookup_drm_device(dpy); screen->surf = NULL; @@ -178,22 +157,22 @@ drm_takedown_shown_screen(_EGLDriver *drv, struct drm_screen *screen) screen->shown = 0; } -EGLSurface -drm_create_window_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, NativeWindowType window, const EGLint *attrib_list) +_EGLSurface * +drm_create_window_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, NativeWindowType window, const EGLint *attrib_list) { - return EGL_NO_SURFACE; + return NULL; } -EGLSurface -drm_create_pixmap_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, NativePixmapType pixmap, const EGLint *attrib_list) +_EGLSurface * +drm_create_pixmap_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, NativePixmapType pixmap, const EGLint *attrib_list) { - return EGL_NO_SURFACE; + return NULL; } -EGLSurface -drm_create_pbuffer_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, +_EGLSurface * +drm_create_pbuffer_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, const EGLint *attrib_list) { int i; @@ -201,13 +180,6 @@ drm_create_pbuffer_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, int height = -1; struct drm_surface *surf = NULL; __GLcontextModes *visual; - _EGLConfig *conf; - - conf = _eglLookupConfig(drv, dpy, config); - if (!conf) { - _eglError(EGL_BAD_CONFIG, "eglCreatePbufferSurface"); - return EGL_NO_CONTEXT; - } for (i = 0; attrib_list && attrib_list[i] != EGL_NONE; i++) { switch (attrib_list[i]) { @@ -225,14 +197,14 @@ drm_create_pbuffer_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, if (width < 1 || height < 1) { _eglError(EGL_BAD_ATTRIBUTE, "eglCreatePbufferSurface"); - return EGL_NO_SURFACE; + return NULL; } surf = (struct drm_surface *) calloc(1, sizeof(struct drm_surface)); if (!surf) goto err; - if (!_eglInitSurface(drv, dpy, &surf->base, EGL_PBUFFER_BIT, config, attrib_list)) + if (!_eglInitSurface(drv, &surf->base, EGL_PBUFFER_BIT, conf, attrib_list)) goto err_surf; surf->w = width; @@ -245,17 +217,16 @@ drm_create_pbuffer_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, (void*)surf); drm_visual_modes_destroy(visual); - _eglSaveSurface(&surf->base); - return surf->base.Handle; + return &surf->base; err_surf: free(surf); err: - return EGL_NO_SURFACE; + return NULL; } -EGLSurface -drm_create_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy, EGLConfig cfg, +_EGLSurface * +drm_create_screen_surface_mesa(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *cfg, const EGLint *attrib_list) { EGLSurface surf = drm_create_pbuffer_surface(drv, dpy, cfg, attrib_list); @@ -264,22 +235,21 @@ drm_create_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy, EGLConfig cfg, } EGLBoolean -drm_show_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy, - EGLScreenMESA screen, - EGLSurface surface, EGLModeMESA m) +drm_show_screen_surface_mesa(_EGLDriver *drv, _EGLDisplay *dpy, + _EGLScreen *screen, + _EGLSurface *surface, _EGLMode *mode) { - struct drm_device *dev = (struct drm_device *)drv; + struct drm_device *dev = lookup_drm_device(dpy); struct drm_surface *surf = lookup_drm_surface(surface); - struct drm_screen *scrn = lookup_drm_screen(dpy, screen); - _EGLMode *mode = _eglLookupMode(dpy, m); + struct drm_screen *scrn = lookup_drm_screen(screen); int ret; unsigned int i, k; if (scrn->shown) - drm_takedown_shown_screen(drv, scrn); + drm_takedown_shown_screen(dpy, scrn); - drm_create_texture(drv, scrn, mode->Width, mode->Height); + drm_create_texture(dpy, scrn, mode->Width, mode->Height); if (!scrn->buffer) return EGL_FALSE; @@ -361,15 +331,12 @@ err_bo: } EGLBoolean -drm_destroy_surface(_EGLDriver *drv, EGLDisplay dpy, EGLSurface surface) +drm_destroy_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface) { struct drm_surface *surf = lookup_drm_surface(surface); - _eglRemoveSurface(&surf->base); - if (surf->base.IsBound) { - surf->base.DeletePending = EGL_TRUE; - } else { + if (!_eglIsSurfaceBound(&surf->base)) { if (surf->screen) - drm_takedown_shown_screen(drv, surf->screen); + drm_takedown_shown_screen(dpy, surf->screen); st_unreference_framebuffer(surf->stfb); free(surf); } @@ -377,8 +344,9 @@ drm_destroy_surface(_EGLDriver *drv, EGLDisplay dpy, EGLSurface surface) } EGLBoolean -drm_swap_buffers(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw) +drm_swap_buffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw) { + struct drm_device *dev = lookup_drm_device(dpy); struct drm_surface *surf = lookup_drm_surface(draw); struct pipe_surface *back_surf; @@ -396,7 +364,6 @@ drm_swap_buffers(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw) st_notify_swapbuffers(surf->stfb); if (surf->screen) { - surf->user->pipe->flush(surf->user->pipe, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_TEXTURE_CACHE, NULL); surf->user->pipe->surface_copy(surf->user->pipe, surf->screen->surface, 0, 0, @@ -404,7 +371,15 @@ drm_swap_buffers(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw) 0, 0, surf->w, surf->h); surf->user->pipe->flush(surf->user->pipe, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_TEXTURE_CACHE, NULL); - /* TODO stuff here */ + +#ifdef DRM_MODE_FEATURE_DIRTYFB + /* TODO query connector property to see if this is needed */ + drmModeDirtyFB(dev->drmFD, surf->screen->fbID, NULL, 0); +#else + (void)dev; +#endif + + /* TODO more stuff here */ } } diff --git a/src/gallium/state_trackers/egl/egl_tracker.c b/src/gallium/state_trackers/egl/egl_tracker.c index 8e620084617..5140755001c 100644 --- a/src/gallium/state_trackers/egl/egl_tracker.c +++ b/src/gallium/state_trackers/egl/egl_tracker.c @@ -6,6 +6,8 @@ #include <string.h> #include "egl_tracker.h" +#include <fcntl.h> + #include "egllog.h" #include "state_tracker/drm_api.h" @@ -21,44 +23,46 @@ extern const struct dri_extension card_extensions[]; * Exported functions */ +static void +drm_unload(_EGLDriver *drv) +{ + free(drv); +} + /** * The bootstrap function. Return a new drm_driver object and * plug in API functions. */ _EGLDriver * -_eglMain(_EGLDisplay *dpy, const char *args) +_eglMain(const char *args) { - struct drm_device *drm; + _EGLDriver *drv; - drm = (struct drm_device *) calloc(1, sizeof(struct drm_device)); - if (!drm) { + drv = (_EGLDriver *) calloc(1, sizeof(_EGLDriver)); + if (!drv) { return NULL; } /* First fill in the dispatch table with defaults */ - _eglInitDriverFallbacks(&drm->base); + _eglInitDriverFallbacks(drv); /* then plug in our Drm-specific functions */ - drm->base.API.Initialize = drm_initialize; - drm->base.API.Terminate = drm_terminate; - drm->base.API.CreateContext = drm_create_context; - drm->base.API.MakeCurrent = drm_make_current; - drm->base.API.CreateWindowSurface = drm_create_window_surface; - drm->base.API.CreatePixmapSurface = drm_create_pixmap_surface; - drm->base.API.CreatePbufferSurface = drm_create_pbuffer_surface; - drm->base.API.DestroySurface = drm_destroy_surface; - drm->base.API.DestroyContext = drm_destroy_context; - drm->base.API.CreateScreenSurfaceMESA = drm_create_screen_surface_mesa; - drm->base.API.ShowScreenSurfaceMESA = drm_show_screen_surface_mesa; - drm->base.API.SwapBuffers = drm_swap_buffers; - - drm->base.ClientAPIsMask = EGL_OPENGL_BIT /*| EGL_OPENGL_ES_BIT*/; - drm->base.Name = "DRM/Gallium/Win"; - - /* enable supported extensions */ - drm->base.Extensions.MESA_screen_surface = EGL_TRUE; - drm->base.Extensions.MESA_copy_context = EGL_TRUE; - - return &drm->base; + drv->API.Initialize = drm_initialize; + drv->API.Terminate = drm_terminate; + drv->API.CreateContext = drm_create_context; + drv->API.MakeCurrent = drm_make_current; + drv->API.CreateWindowSurface = drm_create_window_surface; + drv->API.CreatePixmapSurface = drm_create_pixmap_surface; + drv->API.CreatePbufferSurface = drm_create_pbuffer_surface; + drv->API.DestroySurface = drm_destroy_surface; + drv->API.DestroyContext = drm_destroy_context; + drv->API.CreateScreenSurfaceMESA = drm_create_screen_surface_mesa; + drv->API.ShowScreenSurfaceMESA = drm_show_screen_surface_mesa; + drv->API.SwapBuffers = drm_swap_buffers; + + drv->Name = "DRM/Gallium/Win"; + drv->Unload = drm_unload; + + return drv; } static void @@ -126,11 +130,18 @@ drm_find_dpms(struct drm_device *dev, struct drm_screen *screen) screen->dpms = p; } +static int drm_open_minor(int minor) +{ + char buf[64]; + + sprintf(buf, DRM_DEV_NAME, DRM_DIR_NAME, minor); + return open(buf, O_RDWR, 0); +} + EGLBoolean -drm_initialize(_EGLDriver *drv, EGLDisplay dpy, EGLint *major, EGLint *minor) +drm_initialize(_EGLDriver *drv, _EGLDisplay *disp, EGLint *major, EGLint *minor) { - _EGLDisplay *disp = _eglLookupDisplay(dpy); - struct drm_device *dev = (struct drm_device *)drv; + struct drm_device *dev; struct drm_screen *screen = NULL; drmModeConnectorPtr connector = NULL; drmModeResPtr res = NULL; @@ -139,14 +150,20 @@ drm_initialize(_EGLDriver *drv, EGLDisplay dpy, EGLint *major, EGLint *minor) EGLint i; int fd; - fd = drmOpen("i915", NULL); + dev = (struct drm_device *) calloc(1, sizeof(struct drm_device)); + if (!dev) + return EGL_FALSE; + dev->api = drm_api_create(); + + /* try the first node */ + fd = drm_open_minor(0); if (fd < 0) goto err_fd; dev->drmFD = fd; drm_get_device_id(dev); - dev->screen = drm_api_hooks.create_screen(dev->drmFD, NULL); + dev->screen = dev->api->create_screen(dev->api, dev->drmFD, NULL); if (!dev->screen) goto err_screen; dev->winsys = dev->screen->winsys; @@ -184,6 +201,8 @@ drm_initialize(_EGLDriver *drv, EGLDisplay dpy, EGLint *major, EGLint *minor) } dev->count_screens = num_screens; + disp->DriverData = dev; + /* for now we only have one config */ _EGLConfig *config = calloc(1, sizeof(*config)); memset(config, 1, sizeof(*config)); @@ -198,7 +217,10 @@ drm_initialize(_EGLDriver *drv, EGLDisplay dpy, EGLint *major, EGLint *minor) _eglSetConfigAttrib(config, EGL_SURFACE_TYPE, EGL_PBUFFER_BIT); _eglAddConfig(disp, config); - drv->Initialized = EGL_TRUE; + disp->ClientAPIsMask = EGL_OPENGL_BIT /*| EGL_OPENGL_ES_BIT*/; + /* enable supported extensions */ + disp->Extensions.MESA_screen_surface = EGL_TRUE; + disp->Extensions.MESA_copy_context = EGL_TRUE; *major = 1; *minor = 4; @@ -208,23 +230,27 @@ drm_initialize(_EGLDriver *drv, EGLDisplay dpy, EGLint *major, EGLint *minor) err_screen: drmClose(fd); err_fd: + free(dev); return EGL_FALSE; } EGLBoolean -drm_terminate(_EGLDriver *drv, EGLDisplay dpy) +drm_terminate(_EGLDriver *drv, _EGLDisplay *dpy) { - struct drm_device *dev = (struct drm_device *)drv; + struct drm_device *dev = lookup_drm_device(dpy); struct drm_screen *screen; int i = 0; + _eglReleaseDisplayResources(drv, dpy); + _eglCleanupDisplay(dpy); + drmFreeVersion(dev->version); for (i = 0; i < dev->count_screens; i++) { screen = dev->screens[i]; if (screen->shown) - drm_takedown_shown_screen(drv, screen); + drm_takedown_shown_screen(dpy, screen); drmModeFreeProperty(screen->dpms); drmModeFreeConnector(screen->connector); @@ -237,8 +263,9 @@ drm_terminate(_EGLDriver *drv, EGLDisplay dpy) drmClose(dev->drmFD); - _eglCleanupDisplay(_eglLookupDisplay(dpy)); + dev->api->destroy(dev->api); free(dev); + dpy->DriverData = NULL; return EGL_TRUE; } diff --git a/src/gallium/state_trackers/egl/egl_tracker.h b/src/gallium/state_trackers/egl/egl_tracker.h index ce2717de639..dd4730f9579 100644 --- a/src/gallium/state_trackers/egl/egl_tracker.h +++ b/src/gallium/state_trackers/egl/egl_tracker.h @@ -32,12 +32,11 @@ struct drm_context; struct drm_device { - _EGLDriver base; /* base class/object */ - /* * pipe */ + struct drm_api *api; struct pipe_winsys *winsys; struct pipe_screen *screen; @@ -135,25 +134,29 @@ struct drm_screen }; +static INLINE struct drm_device * +lookup_drm_device(_EGLDisplay *d) +{ + return (struct drm_device *) d->DriverData; +} + + static INLINE struct drm_context * -lookup_drm_context(EGLContext context) +lookup_drm_context(_EGLContext *c) { - _EGLContext *c = _eglLookupContext(context); return (struct drm_context *) c; } static INLINE struct drm_surface * -lookup_drm_surface(EGLSurface surface) +lookup_drm_surface(_EGLSurface *s) { - _EGLSurface *s = _eglLookupSurface(surface); return (struct drm_surface *) s; } static INLINE struct drm_screen * -lookup_drm_screen(EGLDisplay dpy, EGLScreenMESA screen) +lookup_drm_screen(_EGLScreen *s) { - _EGLScreen *s = _eglLookupScreen(dpy, screen); return (struct drm_screen *) s; } @@ -170,25 +173,25 @@ __GLcontextModes* drm_visual_from_config(_EGLConfig *conf); * egl_surface.h */ /*@{*/ -void drm_takedown_shown_screen(_EGLDriver *drv, struct drm_screen *screen); +void drm_takedown_shown_screen(_EGLDisplay *dpy, struct drm_screen *screen); /*@}*/ /** * All function exported to the egl side. */ /*@{*/ -EGLBoolean drm_initialize(_EGLDriver *drv, EGLDisplay dpy, EGLint *major, EGLint *minor); -EGLBoolean drm_terminate(_EGLDriver *drv, EGLDisplay dpy); -EGLContext drm_create_context(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, EGLContext share_list, const EGLint *attrib_list); -EGLBoolean drm_destroy_context(_EGLDriver *drv, EGLDisplay dpy, EGLContext context); -EGLSurface drm_create_window_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, NativeWindowType window, const EGLint *attrib_list); -EGLSurface drm_create_pixmap_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, NativePixmapType pixmap, const EGLint *attrib_list); -EGLSurface drm_create_pbuffer_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, const EGLint *attrib_list); -EGLSurface drm_create_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy, EGLConfig cfg, const EGLint *attrib_list); -EGLBoolean drm_show_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy, EGLScreenMESA screen, EGLSurface surface, EGLModeMESA m); -EGLBoolean drm_destroy_surface(_EGLDriver *drv, EGLDisplay dpy, EGLSurface surface); -EGLBoolean drm_make_current(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext context); -EGLBoolean drm_swap_buffers(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw); +EGLBoolean drm_initialize(_EGLDriver *drv, _EGLDisplay *dpy, EGLint *major, EGLint *minor); +EGLBoolean drm_terminate(_EGLDriver *drv, _EGLDisplay *dpy); +_EGLContext *drm_create_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, _EGLContext *share_list, const EGLint *attrib_list); +EGLBoolean drm_destroy_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *context); +_EGLSurface *drm_create_window_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, NativeWindowType window, const EGLint *attrib_list); +_EGLSurface *drm_create_pixmap_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, NativePixmapType pixmap, const EGLint *attrib_list); +_EGLSurface *drm_create_pbuffer_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, const EGLint *attrib_list); +_EGLSurface *drm_create_screen_surface_mesa(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, const EGLint *attrib_list); +EGLBoolean drm_show_screen_surface_mesa(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *screen, _EGLSurface *surface, _EGLMode *mode); +EGLBoolean drm_destroy_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface); +EGLBoolean drm_make_current(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw, _EGLSurface *read, _EGLContext *context); +EGLBoolean drm_swap_buffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw); /*@}*/ #endif diff --git a/src/gallium/state_trackers/g3dvl/vl_basic_csc.c b/src/gallium/state_trackers/g3dvl/vl_basic_csc.c index 16d4f1e32c9..20d682de3fb 100644 --- a/src/gallium/state_trackers/g3dvl/vl_basic_csc.c +++ b/src/gallium/state_trackers/g3dvl/vl_basic_csc.c @@ -82,7 +82,6 @@ static int vlResizeFrameBuffer template.width[0] = width; template.height[0] = height; template.depth[0] = 1; - template.compressed = 0; pf_get_block(template.format, &template.block); template.tex_usage = PIPE_TEXTURE_USAGE_DISPLAY_TARGET; @@ -669,7 +668,6 @@ static int vlInit sampler.compare_func = PIPE_FUNC_ALWAYS; sampler.normalized_coords = 1; /*sampler.prefilter = ;*/ - /*sampler.shadow_ambient = ;*/ /*sampler.lod_bias = ;*/ /*sampler.min_lod = ;*/ /*sampler.max_lod = ;*/ diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c index eb8270ecad3..23631adb693 100644 --- a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c +++ b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c @@ -1074,7 +1074,6 @@ static int vlInit sampler.compare_func = PIPE_FUNC_ALWAYS; sampler.normalized_coords = 1; /*sampler.prefilter = ;*/ - /*sampler.shadow_ambient = ;*/ /*sampler.lod_bias = ;*/ sampler.min_lod = 0; /*sampler.max_lod = ;*/ @@ -1090,7 +1089,6 @@ static int vlInit template.width[0] = vlRoundUpPOT(mc->picture_width); template.height[0] = vlRoundUpPOT(mc->picture_height); template.depth[0] = 1; - template.compressed = 0; pf_get_block(template.format, &template.block); template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER | PIPE_TEXTURE_USAGE_DYNAMIC; diff --git a/src/gallium/state_trackers/g3dvl/vl_surface.c b/src/gallium/state_trackers/g3dvl/vl_surface.c index 92388f7978d..7f60852cae8 100644 --- a/src/gallium/state_trackers/g3dvl/vl_surface.c +++ b/src/gallium/state_trackers/g3dvl/vl_surface.c @@ -45,7 +45,6 @@ int vlCreateSurface template.width[0] = vlRoundUpPOT(sfc->width); template.height[0] = vlRoundUpPOT(sfc->height); template.depth[0] = 1; - template.compressed = 0; pf_get_block(template.format, &template.block); template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER | PIPE_TEXTURE_USAGE_RENDER_TARGET; diff --git a/src/gallium/state_trackers/glx/xlib/Makefile b/src/gallium/state_trackers/glx/xlib/Makefile index 6d10b090aa0..7b2adc62c34 100644 --- a/src/gallium/state_trackers/glx/xlib/Makefile +++ b/src/gallium/state_trackers/glx/xlib/Makefile @@ -5,13 +5,12 @@ LIBNAME = xlib LIBRARY_INCLUDES = \ -I$(TOP)/include \ - -I$(TOP)/src/mesa \ - -I$(TOP)/src/mesa/main + -I$(TOP)/src/mesa C_SOURCES = \ - glxapi.c \ - fakeglx.c \ - fakeglx_fonts.c \ + glx_api.c \ + glx_getproc.c \ + glx_usefont.c \ xm_api.c include ../../../Makefile.template diff --git a/src/gallium/state_trackers/glx/xlib/SConscript b/src/gallium/state_trackers/glx/xlib/SConscript index 0dbe3413972..fa96df357d5 100644 --- a/src/gallium/state_trackers/glx/xlib/SConscript +++ b/src/gallium/state_trackers/glx/xlib/SConscript @@ -4,8 +4,7 @@ Import('*') if env['platform'] == 'linux' \ - and 'mesa' in env['statetrackers'] \ - and ('softpipe' or 'i915simple' or 'trace') in env['drivers']: + and 'mesa' in env['statetrackers']: env = env.Clone() @@ -18,9 +17,10 @@ if env['platform'] == 'linux' \ st_xlib = env.ConvenienceLibrary( target = 'st_xlib', - source = [ 'glxapi.c', - 'fakeglx.c', - 'fakeglx_fonts.c', + source = [ + 'glx_api.c', + 'glx_getproc.c', + 'glx_usefont.c', 'xm_api.c', ] ) diff --git a/src/gallium/state_trackers/glx/xlib/fakeglx.h b/src/gallium/state_trackers/glx/xlib/fakeglx.h deleted file mode 100644 index e5fd960072e..00000000000 --- a/src/gallium/state_trackers/glx/xlib/fakeglx.h +++ /dev/null @@ -1,41 +0,0 @@ - -/* - * Mesa 3-D graphics library - * Version: 3.5 - * - * Copyright (C) 1999-2000 Brian Paul All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN - * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN - * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - - -#ifndef FAKEGLX_H -#define FAKEGLX_H - - -#include <X11/Xlib.h> - -struct _glxapi_table; - -extern struct _glxapi_table *_mesa_GetGLXDispatchTable(void); - -extern void Fake_glXUseXFont( Font font, int first, int count, int listbase ); - - -#endif - diff --git a/src/gallium/state_trackers/glx/xlib/fakeglx.c b/src/gallium/state_trackers/glx/xlib/glx_api.c index 6df4c7d6936..d1a98f89917 100644 --- a/src/gallium/state_trackers/glx/xlib/fakeglx.c +++ b/src/gallium/state_trackers/glx/xlib/glx_api.c @@ -1,8 +1,9 @@ /* * Mesa 3-D graphics library - * Version: 7.1 + * Version: 7.6 * * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. + * Copyright (C) 2009 VMware, Inc. All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -23,31 +24,21 @@ */ -/* - * This is an emulation of the GLX API which allows Mesa/GLX-based programs - * to run on X servers which do not have the real GLX extension. - * - * Thanks to the contributors: - * - * Initial version: Philip Brown (phil@bolthole.com) - * Better glXGetConfig() support: Armin Liebchen (liebchen@asylum.cs.utah.edu) - * Further visual-handling refinements: Wolfram Gloger - * (wmglo@Dent.MED.Uni-Muenchen.DE). - * - * Notes: - * Don't be fooled, stereo isn't supported yet. +/** + * "Fake" GLX API implemented in terms of the XMesa*() functions. */ -#include "glxapi.h" +#define GLX_GLXEXT_PROTOTYPES +#include "GL/glx.h" + #include "xm_api.h" -#include "context.h" -#include "config.h" -#include "macros.h" -#include "imports.h" -#include "version.h" -#include "fakeglx.h" +#include "main/context.h" +#include "main/config.h" +#include "main/macros.h" +#include "main/imports.h" +#include "main/version.h" #include "state_tracker/st_context.h" #include "state_tracker/st_public.h" @@ -80,23 +71,44 @@ "GLX_SGIX_fbconfig " \ "GLX_SGIX_pbuffer " -/* - * Our fake GLX context will contain a "real" GLX context and an XMesa context. - * - * Note that a pointer to a __GLXcontext is a pointer to a fake_glx_context, - * and vice versa. - * - * We really just need this structure in order to make the libGL functions - * glXGetCurrentContext(), glXGetCurrentDrawable() and glXGetCurrentDisplay() - * work correctly. +#define DEFAULT_DIRECT GL_TRUE + + + +/** + * The GLXContext typedef is defined as a pointer to this structure. */ -struct fake_glx_context { - __GLXcontext glxContext; /* this MUST be first! */ +struct __GLXcontextRec +{ + Display *currentDpy; + GLboolean isDirect; + GLXDrawable currentDrawable; + GLXDrawable currentReadable; + XID xid; + XMesaContext xmesaContext; }; +static pipe_tsd ContextTSD; + +/** Set current context for calling thread */ +static void +SetCurrentContext(GLXContext c) +{ + pipe_tsd_set(&ContextTSD, c); +} + +/** Get current context for calling thread */ +static GLXContext +GetCurrentContext(void) +{ + return pipe_tsd_get(&ContextTSD); +} + + + /**********************************************************************/ /*** GLX Visual Code ***/ /**********************************************************************/ @@ -363,10 +375,6 @@ find_glx_visual( Display *dpy, XVisualInfo *vinfo ) } - - - - /** * Try to get an X visual which matches the given arguments. */ @@ -415,7 +423,6 @@ get_visual( Display *dpy, int scr, unsigned int depth, int xclass ) } - /* * Retrieve the value of the given environment variable and find * the X visual which matches it. @@ -999,8 +1006,8 @@ choose_visual( Display *dpy, int screen, const int *list, GLboolean fbConfig ) } -static XVisualInfo * -Fake_glXChooseVisual( Display *dpy, int screen, int *list ) +XVisualInfo * +glXChooseVisual( Display *dpy, int screen, int *list ) { XMesaVisual xmvis; @@ -1021,18 +1028,18 @@ Fake_glXChooseVisual( Display *dpy, int screen, int *list ) } -static GLXContext -Fake_glXCreateContext( Display *dpy, XVisualInfo *visinfo, - GLXContext share_list, Bool direct ) +GLXContext +glXCreateContext( Display *dpy, XVisualInfo *visinfo, + GLXContext share_list, Bool direct ) { XMesaVisual xmvis; - struct fake_glx_context *glxCtx; - struct fake_glx_context *shareCtx = (struct fake_glx_context *) share_list; + GLXContext glxCtx; + GLXContext shareCtx = share_list; if (!dpy || !visinfo) return 0; - glxCtx = CALLOC_STRUCT(fake_glx_context); + glxCtx = CALLOC_STRUCT(__GLXcontextRec); if (!glxCtx) return 0; @@ -1059,13 +1066,11 @@ Fake_glXCreateContext( Display *dpy, XVisualInfo *visinfo, return NULL; } - glxCtx->glxContext.isDirect = GL_FALSE; - glxCtx->glxContext.currentDpy = dpy; - glxCtx->glxContext.xid = (XID) glxCtx; /* self pointer */ - - assert((void *) glxCtx == (void *) &(glxCtx->glxContext)); + glxCtx->isDirect = DEFAULT_DIRECT; + glxCtx->currentDpy = dpy; + glxCtx->xid = (XID) glxCtx; /* self pointer */ - return (GLXContext) glxCtx; + return glxCtx; } @@ -1078,11 +1083,11 @@ static XMesaBuffer MakeCurrent_PrevReadBuffer = 0; /* GLX 1.3 and later */ -static Bool -Fake_glXMakeContextCurrent( Display *dpy, GLXDrawable draw, - GLXDrawable read, GLXContext ctx ) +Bool +glXMakeContextCurrent( Display *dpy, GLXDrawable draw, + GLXDrawable read, GLXContext ctx ) { - struct fake_glx_context *glxCtx = (struct fake_glx_context *) ctx; + GLXContext glxCtx = ctx; static boolean firsttime = 1, no_rast = 0; if (firsttime) { @@ -1090,7 +1095,6 @@ Fake_glXMakeContextCurrent( Display *dpy, GLXDrawable draw, firsttime = 0; } - if (ctx && draw && read) { XMesaBuffer drawBuffer, readBuffer; XMesaContext xmctx = glxCtx->xmesaContext; @@ -1145,9 +1149,10 @@ Fake_glXMakeContextCurrent( Display *dpy, GLXDrawable draw, /* Now make current! */ if (XMesaMakeCurrent2(xmctx, drawBuffer, readBuffer)) { - ((__GLXcontext *) ctx)->currentDpy = dpy; - ((__GLXcontext *) ctx)->currentDrawable = draw; - ((__GLXcontext *) ctx)->currentReadable = read; + ctx->currentDpy = dpy; + ctx->currentDrawable = draw; + ctx->currentReadable = read; + SetCurrentContext(ctx); return True; } else { @@ -1162,6 +1167,7 @@ Fake_glXMakeContextCurrent( Display *dpy, GLXDrawable draw, MakeCurrent_PrevReadable = 0; MakeCurrent_PrevDrawBuffer = 0; MakeCurrent_PrevReadBuffer = 0; + SetCurrentContext(NULL); return True; } else { @@ -1173,15 +1179,61 @@ Fake_glXMakeContextCurrent( Display *dpy, GLXDrawable draw, } -static Bool -Fake_glXMakeCurrent( Display *dpy, GLXDrawable drawable, GLXContext ctx ) +Bool +glXMakeCurrent( Display *dpy, GLXDrawable drawable, GLXContext ctx ) +{ + return glXMakeContextCurrent( dpy, drawable, drawable, ctx ); +} + + +GLXContext +glXGetCurrentContext(void) +{ + return GetCurrentContext(); +} + + +Display * +glXGetCurrentDisplay(void) +{ + GLXContext glxCtx = glXGetCurrentContext(); + + return glxCtx ? glxCtx->currentDpy : NULL; +} + + +Display * +glXGetCurrentDisplayEXT(void) +{ + return glXGetCurrentDisplay(); +} + + +GLXDrawable +glXGetCurrentDrawable(void) +{ + GLXContext gc = glXGetCurrentContext(); + return gc ? gc->currentDrawable : 0; +} + + +GLXDrawable +glXGetCurrentReadDrawable(void) +{ + GLXContext gc = glXGetCurrentContext(); + return gc ? gc->currentReadable : 0; +} + + +GLXDrawable +glXGetCurrentReadDrawableSGI(void) { - return Fake_glXMakeContextCurrent( dpy, drawable, drawable, ctx ); + return glXGetCurrentReadDrawable(); } -static GLXPixmap -Fake_glXCreateGLXPixmap( Display *dpy, XVisualInfo *visinfo, Pixmap pixmap ) +GLXPixmap +glXCreateGLXPixmap( Display *dpy, XVisualInfo *visinfo, Pixmap pixmap ) { XMesaVisual v; XMesaBuffer b; @@ -1205,9 +1257,9 @@ Fake_glXCreateGLXPixmap( Display *dpy, XVisualInfo *visinfo, Pixmap pixmap ) /*** GLX_MESA_pixmap_colormap ***/ -static GLXPixmap -Fake_glXCreateGLXPixmapMESA( Display *dpy, XVisualInfo *visinfo, - Pixmap pixmap, Colormap cmap ) +GLXPixmap +glXCreateGLXPixmapMESA( Display *dpy, XVisualInfo *visinfo, + Pixmap pixmap, Colormap cmap ) { XMesaVisual v; XMesaBuffer b; @@ -1229,8 +1281,8 @@ Fake_glXCreateGLXPixmapMESA( Display *dpy, XVisualInfo *visinfo, } -static void -Fake_glXDestroyGLXPixmap( Display *dpy, GLXPixmap pixmap ) +void +glXDestroyGLXPixmap( Display *dpy, GLXPixmap pixmap ) { XMesaBuffer b = XMesaFindBuffer(dpy, pixmap); if (b) { @@ -1242,14 +1294,12 @@ Fake_glXDestroyGLXPixmap( Display *dpy, GLXPixmap pixmap ) } -static void -Fake_glXCopyContext( Display *dpy, GLXContext src, GLXContext dst, - unsigned long mask ) +void +glXCopyContext( Display *dpy, GLXContext src, GLXContext dst, + unsigned long mask ) { - struct fake_glx_context *fakeSrc = (struct fake_glx_context *) src; - struct fake_glx_context *fakeDst = (struct fake_glx_context *) dst; - XMesaContext xm_src = fakeSrc->xmesaContext; - XMesaContext xm_dst = fakeDst->xmesaContext; + XMesaContext xm_src = src->xmesaContext; + XMesaContext xm_dst = dst->xmesaContext; (void) dpy; if (MakeCurrent_PrevContext == src) { _mesa_Flush(); @@ -1258,8 +1308,8 @@ Fake_glXCopyContext( Display *dpy, GLXContext src, GLXContext dst, } -static Bool -Fake_glXQueryExtension( Display *dpy, int *errorb, int *event ) +Bool +glXQueryExtension( Display *dpy, int *errorb, int *event ) { /* Mesa's GLX isn't really an X extension but we try to act like one. */ (void) dpy; @@ -1269,18 +1319,10 @@ Fake_glXQueryExtension( Display *dpy, int *errorb, int *event ) } -extern void _kw_ungrab_all( Display *dpy ); -void _kw_ungrab_all( Display *dpy ) -{ - XUngrabPointer( dpy, CurrentTime ); - XUngrabKeyboard( dpy, CurrentTime ); -} - - -static void -Fake_glXDestroyContext( Display *dpy, GLXContext ctx ) +void +glXDestroyContext( Display *dpy, GLXContext ctx ) { - struct fake_glx_context *glxCtx = (struct fake_glx_context *) ctx; + GLXContext glxCtx = ctx; (void) dpy; MakeCurrent_PrevContext = 0; MakeCurrent_PrevDrawable = 0; @@ -1293,18 +1335,18 @@ Fake_glXDestroyContext( Display *dpy, GLXContext ctx ) } -static Bool -Fake_glXIsDirect( Display *dpy, GLXContext ctx ) +Bool +glXIsDirect( Display *dpy, GLXContext ctx ) { - (void) dpy; + GLXContext glxCtx = ctx; (void) ctx; - return False; + return glxCtx->isDirect; } -static void -Fake_glXSwapBuffers( Display *dpy, GLXDrawable drawable ) +void +glXSwapBuffers( Display *dpy, GLXDrawable drawable ) { XMesaBuffer buffer = XMesaFindBuffer( dpy, drawable ); static boolean firsttime = 1, no_rast = 0; @@ -1330,8 +1372,8 @@ Fake_glXSwapBuffers( Display *dpy, GLXDrawable drawable ) /*** GLX_MESA_copy_sub_buffer ***/ -static void -Fake_glXCopySubBufferMESA( Display *dpy, GLXDrawable drawable, +void +glXCopySubBufferMESA( Display *dpy, GLXDrawable drawable, int x, int y, int width, int height ) { XMesaBuffer buffer = XMesaFindBuffer( dpy, drawable ); @@ -1344,8 +1386,8 @@ Fake_glXCopySubBufferMESA( Display *dpy, GLXDrawable drawable, } -static Bool -Fake_glXQueryVersion( Display *dpy, int *maj, int *min ) +Bool +glXQueryVersion( Display *dpy, int *maj, int *min ) { (void) dpy; /* Return GLX version, not Mesa version */ @@ -1561,8 +1603,8 @@ get_config( XMesaVisual xmvis, int attrib, int *value, GLboolean fbconfig ) } -static int -Fake_glXGetConfig( Display *dpy, XVisualInfo *visinfo, +int +glXGetConfig( Display *dpy, XVisualInfo *visinfo, int attrib, int *value ) { XMesaVisual xmvis; @@ -1591,8 +1633,8 @@ Fake_glXGetConfig( Display *dpy, XVisualInfo *visinfo, } -static void -Fake_glXWaitGL( void ) +void +glXWaitGL( void ) { XMesaContext xmesa = XMesaGetCurrentContext(); XMesaFlush( xmesa ); @@ -1600,8 +1642,8 @@ Fake_glXWaitGL( void ) -static void -Fake_glXWaitX( void ) +void +glXWaitX( void ) { XMesaContext xmesa = XMesaGetCurrentContext(); XMesaFlush( xmesa ); @@ -1617,8 +1659,8 @@ get_extensions( void ) /* GLX 1.1 and later */ -static const char * -Fake_glXQueryExtensionsString( Display *dpy, int screen ) +const char * +glXQueryExtensionsString( Display *dpy, int screen ) { (void) dpy; (void) screen; @@ -1628,8 +1670,8 @@ Fake_glXQueryExtensionsString( Display *dpy, int screen ) /* GLX 1.1 and later */ -static const char * -Fake_glXQueryServerString( Display *dpy, int screen, int name ) +const char * +glXQueryServerString( Display *dpy, int screen, int name ) { static char version[1000]; _mesa_sprintf(version, "%d.%d %s", @@ -1653,8 +1695,8 @@ Fake_glXQueryServerString( Display *dpy, int screen, int name ) /* GLX 1.1 and later */ -static const char * -Fake_glXGetClientString( Display *dpy, int name ) +const char * +glXGetClientString( Display *dpy, int name ) { static char version[1000]; _mesa_sprintf(version, "%d.%d %s", CLIENT_MAJOR_VERSION, @@ -1681,8 +1723,8 @@ Fake_glXGetClientString( Display *dpy, int name ) */ -static int -Fake_glXGetFBConfigAttrib( Display *dpy, GLXFBConfig config, +int +glXGetFBConfigAttrib( Display *dpy, GLXFBConfig config, int attribute, int *value ) { XMesaVisual v = (XMesaVisual) config; @@ -1696,8 +1738,8 @@ Fake_glXGetFBConfigAttrib( Display *dpy, GLXFBConfig config, } -static GLXFBConfig * -Fake_glXGetFBConfigs( Display *dpy, int screen, int *nelements ) +GLXFBConfig * +glXGetFBConfigs( Display *dpy, int screen, int *nelements ) { XVisualInfo *visuals, visTemplate; const long visMask = VisualScreenMask; @@ -1722,15 +1764,15 @@ Fake_glXGetFBConfigs( Display *dpy, int screen, int *nelements ) } -static GLXFBConfig * -Fake_glXChooseFBConfig( Display *dpy, int screen, +GLXFBConfig * +glXChooseFBConfig( Display *dpy, int screen, const int *attribList, int *nitems ) { XMesaVisual xmvis; if (!attribList || !attribList[0]) { /* return list of all configs (per GLX_SGIX_fbconfig spec) */ - return Fake_glXGetFBConfigs(dpy, screen, nitems); + return glXGetFBConfigs(dpy, screen, nitems); } xmvis = choose_visual(dpy, screen, attribList, GL_TRUE); @@ -1751,8 +1793,8 @@ Fake_glXChooseFBConfig( Display *dpy, int screen, } -static XVisualInfo * -Fake_glXGetVisualFromFBConfig( Display *dpy, GLXFBConfig config ) +XVisualInfo * +glXGetVisualFromFBConfig( Display *dpy, GLXFBConfig config ) { if (dpy && config) { XMesaVisual xmvis = (XMesaVisual) config; @@ -1773,8 +1815,8 @@ Fake_glXGetVisualFromFBConfig( Display *dpy, GLXFBConfig config ) } -static GLXWindow -Fake_glXCreateWindow( Display *dpy, GLXFBConfig config, Window win, +GLXWindow +glXCreateWindow( Display *dpy, GLXFBConfig config, Window win, const int *attribList ) { XMesaVisual xmvis = (XMesaVisual) config; @@ -1793,8 +1835,8 @@ Fake_glXCreateWindow( Display *dpy, GLXFBConfig config, Window win, } -static void -Fake_glXDestroyWindow( Display *dpy, GLXWindow window ) +void +glXDestroyWindow( Display *dpy, GLXWindow window ) { XMesaBuffer b = XMesaFindBuffer(dpy, (Drawable) window); if (b) @@ -1804,8 +1846,8 @@ Fake_glXDestroyWindow( Display *dpy, GLXWindow window ) /* XXX untested */ -static GLXPixmap -Fake_glXCreatePixmap( Display *dpy, GLXFBConfig config, Pixmap pixmap, +GLXPixmap +glXCreatePixmap( Display *dpy, GLXFBConfig config, Pixmap pixmap, const int *attribList ) { XMesaVisual v = (XMesaVisual) config; @@ -1914,8 +1956,8 @@ Fake_glXCreatePixmap( Display *dpy, GLXFBConfig config, Pixmap pixmap, } -static void -Fake_glXDestroyPixmap( Display *dpy, GLXPixmap pixmap ) +void +glXDestroyPixmap( Display *dpy, GLXPixmap pixmap ) { XMesaBuffer b = XMesaFindBuffer(dpy, (Drawable)pixmap); if (b) @@ -1924,8 +1966,8 @@ Fake_glXDestroyPixmap( Display *dpy, GLXPixmap pixmap ) } -static GLXPbuffer -Fake_glXCreatePbuffer( Display *dpy, GLXFBConfig config, +GLXPbuffer +glXCreatePbuffer( Display *dpy, GLXFBConfig config, const int *attribList ) { XMesaVisual xmvis = (XMesaVisual) config; @@ -1977,8 +2019,8 @@ Fake_glXCreatePbuffer( Display *dpy, GLXFBConfig config, } -static void -Fake_glXDestroyPbuffer( Display *dpy, GLXPbuffer pbuf ) +void +glXDestroyPbuffer( Display *dpy, GLXPbuffer pbuf ) { XMesaBuffer b = XMesaFindBuffer(dpy, pbuf); if (b) { @@ -1987,8 +2029,8 @@ Fake_glXDestroyPbuffer( Display *dpy, GLXPbuffer pbuf ) } -static void -Fake_glXQueryDrawable( Display *dpy, GLXDrawable draw, int attribute, +void +glXQueryDrawable( Display *dpy, GLXDrawable draw, int attribute, unsigned int *value ) { XMesaBuffer xmbuf = XMesaFindBuffer(dpy, draw); @@ -2029,19 +2071,19 @@ Fake_glXQueryDrawable( Display *dpy, GLXDrawable draw, int attribute, } -static GLXContext -Fake_glXCreateNewContext( Display *dpy, GLXFBConfig config, +GLXContext +glXCreateNewContext( Display *dpy, GLXFBConfig config, int renderType, GLXContext shareList, Bool direct ) { - struct fake_glx_context *glxCtx; - struct fake_glx_context *shareCtx = (struct fake_glx_context *) shareList; + GLXContext glxCtx; + GLXContext shareCtx = shareList; XMesaVisual xmvis = (XMesaVisual) config; if (!dpy || !config || (renderType != GLX_RGBA_TYPE && renderType != GLX_COLOR_INDEX_TYPE)) return 0; - glxCtx = CALLOC_STRUCT(fake_glx_context); + glxCtx = CALLOC_STRUCT(__GLXcontextRec); if (!glxCtx) return 0; @@ -2055,20 +2097,18 @@ Fake_glXCreateNewContext( Display *dpy, GLXFBConfig config, return NULL; } - glxCtx->glxContext.isDirect = GL_FALSE; - glxCtx->glxContext.currentDpy = dpy; - glxCtx->glxContext.xid = (XID) glxCtx; /* self pointer */ - - assert((void *) glxCtx == (void *) &(glxCtx->glxContext)); + glxCtx->isDirect = DEFAULT_DIRECT; + glxCtx->currentDpy = dpy; + glxCtx->xid = (XID) glxCtx; /* self pointer */ - return (GLXContext) glxCtx; + return glxCtx; } -static int -Fake_glXQueryContext( Display *dpy, GLXContext ctx, int attribute, int *value ) +int +glXQueryContext( Display *dpy, GLXContext ctx, int attribute, int *value ) { - struct fake_glx_context *glxCtx = (struct fake_glx_context *) ctx; + GLXContext glxCtx = ctx; XMesaContext xmctx = glxCtx->xmesaContext; (void) dpy; @@ -2094,8 +2134,8 @@ Fake_glXQueryContext( Display *dpy, GLXContext ctx, int attribute, int *value ) } -static void -Fake_glXSelectEvent( Display *dpy, GLXDrawable drawable, unsigned long mask ) +void +glXSelectEvent( Display *dpy, GLXDrawable drawable, unsigned long mask ) { XMesaBuffer xmbuf = XMesaFindBuffer(dpy, drawable); if (xmbuf) @@ -2103,8 +2143,8 @@ Fake_glXSelectEvent( Display *dpy, GLXDrawable drawable, unsigned long mask ) } -static void -Fake_glXGetSelectedEvent( Display *dpy, GLXDrawable drawable, +void +glXGetSelectedEvent( Display *dpy, GLXDrawable drawable, unsigned long *mask ) { XMesaBuffer xmbuf = XMesaFindBuffer(dpy, drawable); @@ -2118,8 +2158,8 @@ Fake_glXGetSelectedEvent( Display *dpy, GLXDrawable drawable, /*** GLX_SGI_swap_control ***/ -static int -Fake_glXSwapIntervalSGI(int interval) +int +glXSwapIntervalSGI(int interval) { (void) interval; return 0; @@ -2131,16 +2171,16 @@ Fake_glXSwapIntervalSGI(int interval) static unsigned int FrameCounter = 0; -static int -Fake_glXGetVideoSyncSGI(unsigned int *count) +int +glXGetVideoSyncSGI(unsigned int *count) { /* this is a bogus implementation */ *count = FrameCounter++; return 0; } -static int -Fake_glXWaitVideoSyncSGI(int divisor, int remainder, unsigned int *count) +int +glXWaitVideoSyncSGI(int divisor, int remainder, unsigned int *count) { if (divisor <= 0 || remainder < 0) return GLX_BAD_VALUE; @@ -2156,15 +2196,15 @@ Fake_glXWaitVideoSyncSGI(int divisor, int remainder, unsigned int *count) /*** GLX_SGI_make_current_read ***/ -static Bool -Fake_glXMakeCurrentReadSGI(Display *dpy, GLXDrawable draw, GLXDrawable read, GLXContext ctx) +Bool +glXMakeCurrentReadSGI(Display *dpy, GLXDrawable draw, GLXDrawable read, GLXContext ctx) { - return Fake_glXMakeContextCurrent( dpy, draw, read, ctx ); + return glXMakeContextCurrent( dpy, draw, read, ctx ); } /* not used static GLXDrawable -Fake_glXGetCurrentReadDrawableSGI(void) +glXGetCurrentReadDrawableSGI(void) { return 0; } @@ -2174,8 +2214,8 @@ Fake_glXGetCurrentReadDrawableSGI(void) /*** GLX_SGIX_video_source ***/ #if defined(_VL_H) -static GLXVideoSourceSGIX -Fake_glXCreateGLXVideoSourceSGIX(Display *dpy, int screen, VLServer server, VLPath path, int nodeClass, VLNode drainNode) +GLXVideoSourceSGIX +glXCreateGLXVideoSourceSGIX(Display *dpy, int screen, VLServer server, VLPath path, int nodeClass, VLNode drainNode) { (void) dpy; (void) screen; @@ -2186,8 +2226,8 @@ Fake_glXCreateGLXVideoSourceSGIX(Display *dpy, int screen, VLServer server, VLPa return 0; } -static void -Fake_glXDestroyGLXVideoSourceSGIX(Display *dpy, GLXVideoSourceSGIX src) +void +glXDestroyGLXVideoSourceSGIX(Display *dpy, GLXVideoSourceSGIX src) { (void) dpy; (void) src; @@ -2198,30 +2238,30 @@ Fake_glXDestroyGLXVideoSourceSGIX(Display *dpy, GLXVideoSourceSGIX src) /*** GLX_EXT_import_context ***/ -static void -Fake_glXFreeContextEXT(Display *dpy, GLXContext context) +void +glXFreeContextEXT(Display *dpy, GLXContext context) { (void) dpy; (void) context; } -static GLXContextID -Fake_glXGetContextIDEXT(const GLXContext context) +GLXContextID +glXGetContextIDEXT(const GLXContext context) { (void) context; return 0; } -static GLXContext -Fake_glXImportContextEXT(Display *dpy, GLXContextID contextID) +GLXContext +glXImportContextEXT(Display *dpy, GLXContextID contextID) { (void) dpy; (void) contextID; return 0; } -static int -Fake_glXQueryContextInfoEXT(Display *dpy, GLXContext context, int attribute, int *value) +int +glXQueryContextInfoEXT(Display *dpy, GLXContext context, int attribute, int *value) { (void) dpy; (void) context; @@ -2234,21 +2274,21 @@ Fake_glXQueryContextInfoEXT(Display *dpy, GLXContext context, int attribute, int /*** GLX_SGIX_fbconfig ***/ -static int -Fake_glXGetFBConfigAttribSGIX(Display *dpy, GLXFBConfigSGIX config, int attribute, int *value) +int +glXGetFBConfigAttribSGIX(Display *dpy, GLXFBConfigSGIX config, int attribute, int *value) { - return Fake_glXGetFBConfigAttrib(dpy, config, attribute, value); + return glXGetFBConfigAttrib(dpy, config, attribute, value); } -static GLXFBConfigSGIX * -Fake_glXChooseFBConfigSGIX(Display *dpy, int screen, int *attrib_list, int *nelements) +GLXFBConfigSGIX * +glXChooseFBConfigSGIX(Display *dpy, int screen, int *attrib_list, int *nelements) { - return (GLXFBConfig *) Fake_glXChooseFBConfig(dpy, screen, attrib_list, nelements); + return (GLXFBConfig *) glXChooseFBConfig(dpy, screen, attrib_list, nelements); } -static GLXPixmap -Fake_glXCreateGLXPixmapWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, Pixmap pixmap) +GLXPixmap +glXCreateGLXPixmapWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, Pixmap pixmap) { XMesaVisual xmvis = (XMesaVisual) config; XMesaBuffer xmbuf = XMesaCreatePixmapBuffer(xmvis, pixmap, 0); @@ -2256,14 +2296,14 @@ Fake_glXCreateGLXPixmapWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, Pixm } -static GLXContext -Fake_glXCreateContextWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, int render_type, GLXContext share_list, Bool direct) +GLXContext +glXCreateContextWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, int render_type, GLXContext share_list, Bool direct) { XMesaVisual xmvis = (XMesaVisual) config; - struct fake_glx_context *glxCtx; - struct fake_glx_context *shareCtx = (struct fake_glx_context *) share_list; + GLXContext glxCtx; + GLXContext shareCtx = share_list; - glxCtx = CALLOC_STRUCT(fake_glx_context); + glxCtx = CALLOC_STRUCT(__GLXcontextRec); if (!glxCtx) return 0; @@ -2277,25 +2317,23 @@ Fake_glXCreateContextWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, int re return NULL; } - glxCtx->glxContext.isDirect = GL_FALSE; - glxCtx->glxContext.currentDpy = dpy; - glxCtx->glxContext.xid = (XID) glxCtx; /* self pointer */ - - assert((void *) glxCtx == (void *) &(glxCtx->glxContext)); + glxCtx->isDirect = DEFAULT_DIRECT; + glxCtx->currentDpy = dpy; + glxCtx->xid = (XID) glxCtx; /* self pointer */ - return (GLXContext) glxCtx; + return glxCtx; } -static XVisualInfo * -Fake_glXGetVisualFromFBConfigSGIX(Display *dpy, GLXFBConfigSGIX config) +XVisualInfo * +glXGetVisualFromFBConfigSGIX(Display *dpy, GLXFBConfigSGIX config) { - return Fake_glXGetVisualFromFBConfig(dpy, config); + return glXGetVisualFromFBConfig(dpy, config); } -static GLXFBConfigSGIX -Fake_glXGetFBConfigFromVisualSGIX(Display *dpy, XVisualInfo *vis) +GLXFBConfigSGIX +glXGetFBConfigFromVisualSGIX(Display *dpy, XVisualInfo *vis) { XMesaVisual xmvis = find_glx_visual(dpy, vis); if (!xmvis) { @@ -2310,8 +2348,8 @@ Fake_glXGetFBConfigFromVisualSGIX(Display *dpy, XVisualInfo *vis) /*** GLX_SGIX_pbuffer ***/ -static GLXPbufferSGIX -Fake_glXCreateGLXPbufferSGIX(Display *dpy, GLXFBConfigSGIX config, +GLXPbufferSGIX +glXCreateGLXPbufferSGIX(Display *dpy, GLXFBConfigSGIX config, unsigned int width, unsigned int height, int *attribList) { @@ -2349,8 +2387,8 @@ Fake_glXCreateGLXPbufferSGIX(Display *dpy, GLXFBConfigSGIX config, } -static void -Fake_glXDestroyGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf) +void +glXDestroyGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf) { XMesaBuffer xmbuf = XMesaFindBuffer(dpy, pbuf); if (xmbuf) { @@ -2359,8 +2397,8 @@ Fake_glXDestroyGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf) } -static int -Fake_glXQueryGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf, int attribute, unsigned int *value) +int +glXQueryGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf, int attribute, unsigned int *value) { const XMesaBuffer xmbuf = XMesaFindBuffer(dpy, pbuf); @@ -2392,8 +2430,8 @@ Fake_glXQueryGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf, int attribute, un } -static void -Fake_glXSelectEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long mask) +void +glXSelectEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long mask) { XMesaBuffer xmbuf = XMesaFindBuffer(dpy, drawable); if (xmbuf) { @@ -2403,8 +2441,8 @@ Fake_glXSelectEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long mask) } -static void -Fake_glXGetSelectedEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long *mask) +void +glXGetSelectedEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long *mask) { XMesaBuffer xmbuf = XMesaFindBuffer(dpy, drawable); if (xmbuf) { @@ -2419,8 +2457,8 @@ Fake_glXGetSelectedEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long * /*** GLX_SGI_cushion ***/ -static void -Fake_glXCushionSGI(Display *dpy, Window win, float cushion) +void +glXCushionSGI(Display *dpy, Window win, float cushion) { (void) dpy; (void) win; @@ -2431,8 +2469,8 @@ Fake_glXCushionSGI(Display *dpy, Window win, float cushion) /*** GLX_SGIX_video_resize ***/ -static int -Fake_glXBindChannelToWindowSGIX(Display *dpy, int screen, int channel , Window window) +int +glXBindChannelToWindowSGIX(Display *dpy, int screen, int channel , Window window) { (void) dpy; (void) screen; @@ -2441,8 +2479,8 @@ Fake_glXBindChannelToWindowSGIX(Display *dpy, int screen, int channel , Window w return 0; } -static int -Fake_glXChannelRectSGIX(Display *dpy, int screen, int channel, int x, int y, int w, int h) +int +glXChannelRectSGIX(Display *dpy, int screen, int channel, int x, int y, int w, int h) { (void) dpy; (void) screen; @@ -2454,8 +2492,8 @@ Fake_glXChannelRectSGIX(Display *dpy, int screen, int channel, int x, int y, int return 0; } -static int -Fake_glXQueryChannelRectSGIX(Display *dpy, int screen, int channel, int *x, int *y, int *w, int *h) +int +glXQueryChannelRectSGIX(Display *dpy, int screen, int channel, int *x, int *y, int *w, int *h) { (void) dpy; (void) screen; @@ -2467,8 +2505,8 @@ Fake_glXQueryChannelRectSGIX(Display *dpy, int screen, int channel, int *x, int return 0; } -static int -Fake_glXQueryChannelDeltasSGIX(Display *dpy, int screen, int channel, int *dx, int *dy, int *dw, int *dh) +int +glXQueryChannelDeltasSGIX(Display *dpy, int screen, int channel, int *dx, int *dy, int *dw, int *dh) { (void) dpy; (void) screen; @@ -2480,8 +2518,8 @@ Fake_glXQueryChannelDeltasSGIX(Display *dpy, int screen, int channel, int *dx, i return 0; } -static int -Fake_glXChannelRectSyncSGIX(Display *dpy, int screen, int channel, GLenum synctype) +int +glXChannelRectSyncSGIX(Display *dpy, int screen, int channel, GLenum synctype) { (void) dpy; (void) screen; @@ -2495,8 +2533,8 @@ Fake_glXChannelRectSyncSGIX(Display *dpy, int screen, int channel, GLenum syncty /*** GLX_SGIX_dmbuffer **/ #if defined(_DM_BUFFER_H_) -static Bool -Fake_glXAssociateDMPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuffer, DMparams *params, DMbuffer dmbuffer) +Bool +glXAssociateDMPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuffer, DMparams *params, DMbuffer dmbuffer) { (void) dpy; (void) pbuffer; @@ -2509,8 +2547,8 @@ Fake_glXAssociateDMPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuffer, DMparams *p /*** GLX_SGIX_swap_group ***/ -static void -Fake_glXJoinSwapGroupSGIX(Display *dpy, GLXDrawable drawable, GLXDrawable member) +void +glXJoinSwapGroupSGIX(Display *dpy, GLXDrawable drawable, GLXDrawable member) { (void) dpy; (void) drawable; @@ -2521,16 +2559,16 @@ Fake_glXJoinSwapGroupSGIX(Display *dpy, GLXDrawable drawable, GLXDrawable member /*** GLX_SGIX_swap_barrier ***/ -static void -Fake_glXBindSwapBarrierSGIX(Display *dpy, GLXDrawable drawable, int barrier) +void +glXBindSwapBarrierSGIX(Display *dpy, GLXDrawable drawable, int barrier) { (void) dpy; (void) drawable; (void) barrier; } -static Bool -Fake_glXQueryMaxSwapBarriersSGIX(Display *dpy, int screen, int *max) +Bool +glXQueryMaxSwapBarriersSGIX(Display *dpy, int screen, int *max) { (void) dpy; (void) screen; @@ -2542,8 +2580,8 @@ Fake_glXQueryMaxSwapBarriersSGIX(Display *dpy, int screen, int *max) /*** GLX_SUN_get_transparent_index ***/ -static Status -Fake_glXGetTransparentIndexSUN(Display *dpy, Window overlay, Window underlay, long *pTransparent) +Status +glXGetTransparentIndexSUN(Display *dpy, Window overlay, Window underlay, long *pTransparent) { (void) dpy; (void) overlay; @@ -2560,8 +2598,8 @@ Fake_glXGetTransparentIndexSUN(Display *dpy, Window overlay, Window underlay, lo * Release the depth, stencil, accum buffers attached to a GLXDrawable * (a window or pixmap) prior to destroying the GLXDrawable. */ -static Bool -Fake_glXReleaseBuffersMESA( Display *dpy, GLXDrawable d ) +Bool +glXReleaseBuffersMESA( Display *dpy, GLXDrawable d ) { XMesaBuffer b = XMesaFindBuffer(dpy, d); if (b) { @@ -2573,8 +2611,8 @@ Fake_glXReleaseBuffersMESA( Display *dpy, GLXDrawable d ) /*** GLX_EXT_texture_from_pixmap ***/ -static void -Fake_glXBindTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer, +void +glXBindTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer, const int *attrib_list) { XMesaBuffer b = XMesaFindBuffer(dpy, drawable); @@ -2582,162 +2620,10 @@ Fake_glXBindTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer, XMesaBindTexImage(dpy, b, buffer, attrib_list); } -static void -Fake_glXReleaseTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer) +void +glXReleaseTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer) { XMesaBuffer b = XMesaFindBuffer(dpy, drawable); if (b) XMesaReleaseTexImage(dpy, b, buffer); } - - - -/** - * Create a new GLX API dispatch table with its function pointers - * initialized to point to Mesa's "fake" GLX API functions. - * - * Note: there used to be a similar function - * (_real_GetGLXDispatchTable) that returns a new dispatch table with - * all pointers initalized to point to "real" GLX functions (which - * understand GLX wire protocol, etc). - */ -struct _glxapi_table * -_mesa_GetGLXDispatchTable(void) -{ - static struct _glxapi_table glx; - - /* be sure our dispatch table size <= libGL's table */ - { - GLuint size = sizeof(struct _glxapi_table) / sizeof(void *); - (void) size; - assert(_glxapi_get_dispatch_table_size() >= size); - } - - /* initialize the whole table to no-ops */ - _glxapi_set_no_op_table(&glx); - - /* now initialize the table with the functions I implement */ - glx.ChooseVisual = Fake_glXChooseVisual; - glx.CopyContext = Fake_glXCopyContext; - glx.CreateContext = Fake_glXCreateContext; - glx.CreateGLXPixmap = Fake_glXCreateGLXPixmap; - glx.DestroyContext = Fake_glXDestroyContext; - glx.DestroyGLXPixmap = Fake_glXDestroyGLXPixmap; - glx.GetConfig = Fake_glXGetConfig; - /*glx.GetCurrentContext = Fake_glXGetCurrentContext;*/ - /*glx.GetCurrentDrawable = Fake_glXGetCurrentDrawable;*/ - glx.IsDirect = Fake_glXIsDirect; - glx.MakeCurrent = Fake_glXMakeCurrent; - glx.QueryExtension = Fake_glXQueryExtension; - glx.QueryVersion = Fake_glXQueryVersion; - glx.SwapBuffers = Fake_glXSwapBuffers; - glx.UseXFont = Fake_glXUseXFont; - glx.WaitGL = Fake_glXWaitGL; - glx.WaitX = Fake_glXWaitX; - - /*** GLX_VERSION_1_1 ***/ - glx.GetClientString = Fake_glXGetClientString; - glx.QueryExtensionsString = Fake_glXQueryExtensionsString; - glx.QueryServerString = Fake_glXQueryServerString; - - /*** GLX_VERSION_1_2 ***/ - /*glx.GetCurrentDisplay = Fake_glXGetCurrentDisplay;*/ - - /*** GLX_VERSION_1_3 ***/ - glx.ChooseFBConfig = Fake_glXChooseFBConfig; - glx.CreateNewContext = Fake_glXCreateNewContext; - glx.CreatePbuffer = Fake_glXCreatePbuffer; - glx.CreatePixmap = Fake_glXCreatePixmap; - glx.CreateWindow = Fake_glXCreateWindow; - glx.DestroyPbuffer = Fake_glXDestroyPbuffer; - glx.DestroyPixmap = Fake_glXDestroyPixmap; - glx.DestroyWindow = Fake_glXDestroyWindow; - /*glx.GetCurrentReadDrawable = Fake_glXGetCurrentReadDrawable;*/ - glx.GetFBConfigAttrib = Fake_glXGetFBConfigAttrib; - glx.GetFBConfigs = Fake_glXGetFBConfigs; - glx.GetSelectedEvent = Fake_glXGetSelectedEvent; - glx.GetVisualFromFBConfig = Fake_glXGetVisualFromFBConfig; - glx.MakeContextCurrent = Fake_glXMakeContextCurrent; - glx.QueryContext = Fake_glXQueryContext; - glx.QueryDrawable = Fake_glXQueryDrawable; - glx.SelectEvent = Fake_glXSelectEvent; - - /*** GLX_SGI_swap_control ***/ - glx.SwapIntervalSGI = Fake_glXSwapIntervalSGI; - - /*** GLX_SGI_video_sync ***/ - glx.GetVideoSyncSGI = Fake_glXGetVideoSyncSGI; - glx.WaitVideoSyncSGI = Fake_glXWaitVideoSyncSGI; - - /*** GLX_SGI_make_current_read ***/ - glx.MakeCurrentReadSGI = Fake_glXMakeCurrentReadSGI; - /*glx.GetCurrentReadDrawableSGI = Fake_glXGetCurrentReadDrawableSGI;*/ - -/*** GLX_SGIX_video_source ***/ -#if defined(_VL_H) - glx.CreateGLXVideoSourceSGIX = Fake_glXCreateGLXVideoSourceSGIX; - glx.DestroyGLXVideoSourceSGIX = Fake_glXDestroyGLXVideoSourceSGIX; -#endif - - /*** GLX_EXT_import_context ***/ - glx.FreeContextEXT = Fake_glXFreeContextEXT; - glx.GetContextIDEXT = Fake_glXGetContextIDEXT; - /*glx.GetCurrentDisplayEXT = Fake_glXGetCurrentDisplayEXT;*/ - glx.ImportContextEXT = Fake_glXImportContextEXT; - glx.QueryContextInfoEXT = Fake_glXQueryContextInfoEXT; - - /*** GLX_SGIX_fbconfig ***/ - glx.GetFBConfigAttribSGIX = Fake_glXGetFBConfigAttribSGIX; - glx.ChooseFBConfigSGIX = Fake_glXChooseFBConfigSGIX; - glx.CreateGLXPixmapWithConfigSGIX = Fake_glXCreateGLXPixmapWithConfigSGIX; - glx.CreateContextWithConfigSGIX = Fake_glXCreateContextWithConfigSGIX; - glx.GetVisualFromFBConfigSGIX = Fake_glXGetVisualFromFBConfigSGIX; - glx.GetFBConfigFromVisualSGIX = Fake_glXGetFBConfigFromVisualSGIX; - - /*** GLX_SGIX_pbuffer ***/ - glx.CreateGLXPbufferSGIX = Fake_glXCreateGLXPbufferSGIX; - glx.DestroyGLXPbufferSGIX = Fake_glXDestroyGLXPbufferSGIX; - glx.QueryGLXPbufferSGIX = Fake_glXQueryGLXPbufferSGIX; - glx.SelectEventSGIX = Fake_glXSelectEventSGIX; - glx.GetSelectedEventSGIX = Fake_glXGetSelectedEventSGIX; - - /*** GLX_SGI_cushion ***/ - glx.CushionSGI = Fake_glXCushionSGI; - - /*** GLX_SGIX_video_resize ***/ - glx.BindChannelToWindowSGIX = Fake_glXBindChannelToWindowSGIX; - glx.ChannelRectSGIX = Fake_glXChannelRectSGIX; - glx.QueryChannelRectSGIX = Fake_glXQueryChannelRectSGIX; - glx.QueryChannelDeltasSGIX = Fake_glXQueryChannelDeltasSGIX; - glx.ChannelRectSyncSGIX = Fake_glXChannelRectSyncSGIX; - - /*** GLX_SGIX_dmbuffer **/ -#if defined(_DM_BUFFER_H_) - glx.AssociateDMPbufferSGIX = NULL; -#endif - - /*** GLX_SGIX_swap_group ***/ - glx.JoinSwapGroupSGIX = Fake_glXJoinSwapGroupSGIX; - - /*** GLX_SGIX_swap_barrier ***/ - glx.BindSwapBarrierSGIX = Fake_glXBindSwapBarrierSGIX; - glx.QueryMaxSwapBarriersSGIX = Fake_glXQueryMaxSwapBarriersSGIX; - - /*** GLX_SUN_get_transparent_index ***/ - glx.GetTransparentIndexSUN = Fake_glXGetTransparentIndexSUN; - - /*** GLX_MESA_copy_sub_buffer ***/ - glx.CopySubBufferMESA = Fake_glXCopySubBufferMESA; - - /*** GLX_MESA_release_buffers ***/ - glx.ReleaseBuffersMESA = Fake_glXReleaseBuffersMESA; - - /*** GLX_MESA_pixmap_colormap ***/ - glx.CreateGLXPixmapMESA = Fake_glXCreateGLXPixmapMESA; - - /*** GLX_EXT_texture_from_pixmap ***/ - glx.BindTexImageEXT = Fake_glXBindTexImageEXT; - glx.ReleaseTexImageEXT = Fake_glXReleaseTexImageEXT; - - return &glx; -} diff --git a/src/gallium/state_trackers/glx/xlib/glx_getproc.c b/src/gallium/state_trackers/glx/xlib/glx_getproc.c new file mode 100644 index 00000000000..ca7d88c9227 --- /dev/null +++ b/src/gallium/state_trackers/glx/xlib/glx_getproc.c @@ -0,0 +1,214 @@ +/* + * Mesa 3-D graphics library + * Version: 7.6 + * + * Copyright (C) 1999-2008 Brian Paul All Rights Reserved. + * Copyright (C) 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + + +/** + * glXGetProcAddress() + */ + + +#define GLX_GLXEXT_PROTOTYPES + +#include <string.h> +#include "GL/glx.h" +#include "glapi/glapi.h" + + +struct name_address_pair { + const char *Name; + __GLXextFuncPtr Address; +}; + + +static struct name_address_pair GLX_functions[] = { + /*** GLX_VERSION_1_0 ***/ + { "glXChooseVisual", (__GLXextFuncPtr) glXChooseVisual }, + { "glXCopyContext", (__GLXextFuncPtr) glXCopyContext }, + { "glXCreateContext", (__GLXextFuncPtr) glXCreateContext }, + { "glXCreateGLXPixmap", (__GLXextFuncPtr) glXCreateGLXPixmap }, + { "glXDestroyContext", (__GLXextFuncPtr) glXDestroyContext }, + { "glXDestroyGLXPixmap", (__GLXextFuncPtr) glXDestroyGLXPixmap }, + { "glXGetConfig", (__GLXextFuncPtr) glXGetConfig }, + { "glXGetCurrentContext", (__GLXextFuncPtr) glXGetCurrentContext }, + { "glXGetCurrentDrawable", (__GLXextFuncPtr) glXGetCurrentDrawable }, + { "glXIsDirect", (__GLXextFuncPtr) glXIsDirect }, + { "glXMakeCurrent", (__GLXextFuncPtr) glXMakeCurrent }, + { "glXQueryExtension", (__GLXextFuncPtr) glXQueryExtension }, + { "glXQueryVersion", (__GLXextFuncPtr) glXQueryVersion }, + { "glXSwapBuffers", (__GLXextFuncPtr) glXSwapBuffers }, + { "glXUseXFont", (__GLXextFuncPtr) glXUseXFont }, + { "glXWaitGL", (__GLXextFuncPtr) glXWaitGL }, + { "glXWaitX", (__GLXextFuncPtr) glXWaitX }, + + /*** GLX_VERSION_1_1 ***/ + { "glXGetClientString", (__GLXextFuncPtr) glXGetClientString }, + { "glXQueryExtensionsString", (__GLXextFuncPtr) glXQueryExtensionsString }, + { "glXQueryServerString", (__GLXextFuncPtr) glXQueryServerString }, + + /*** GLX_VERSION_1_2 ***/ + { "glXGetCurrentDisplay", (__GLXextFuncPtr) glXGetCurrentDisplay }, + + /*** GLX_VERSION_1_3 ***/ + { "glXChooseFBConfig", (__GLXextFuncPtr) glXChooseFBConfig }, + { "glXCreateNewContext", (__GLXextFuncPtr) glXCreateNewContext }, + { "glXCreatePbuffer", (__GLXextFuncPtr) glXCreatePbuffer }, + { "glXCreatePixmap", (__GLXextFuncPtr) glXCreatePixmap }, + { "glXCreateWindow", (__GLXextFuncPtr) glXCreateWindow }, + { "glXDestroyPbuffer", (__GLXextFuncPtr) glXDestroyPbuffer }, + { "glXDestroyPixmap", (__GLXextFuncPtr) glXDestroyPixmap }, + { "glXDestroyWindow", (__GLXextFuncPtr) glXDestroyWindow }, + { "glXGetCurrentReadDrawable", (__GLXextFuncPtr) glXGetCurrentReadDrawable }, + { "glXGetFBConfigAttrib", (__GLXextFuncPtr) glXGetFBConfigAttrib }, + { "glXGetFBConfigs", (__GLXextFuncPtr) glXGetFBConfigs }, + { "glXGetSelectedEvent", (__GLXextFuncPtr) glXGetSelectedEvent }, + { "glXGetVisualFromFBConfig", (__GLXextFuncPtr) glXGetVisualFromFBConfig }, + { "glXMakeContextCurrent", (__GLXextFuncPtr) glXMakeContextCurrent }, + { "glXQueryContext", (__GLXextFuncPtr) glXQueryContext }, + { "glXQueryDrawable", (__GLXextFuncPtr) glXQueryDrawable }, + { "glXSelectEvent", (__GLXextFuncPtr) glXSelectEvent }, + + /*** GLX_VERSION_1_4 ***/ + { "glXGetProcAddress", (__GLXextFuncPtr) glXGetProcAddress }, + + /*** GLX_SGI_swap_control ***/ + { "glXSwapIntervalSGI", (__GLXextFuncPtr) glXSwapIntervalSGI }, + + /*** GLX_SGI_video_sync ***/ + { "glXGetVideoSyncSGI", (__GLXextFuncPtr) glXGetVideoSyncSGI }, + { "glXWaitVideoSyncSGI", (__GLXextFuncPtr) glXWaitVideoSyncSGI }, + + /*** GLX_SGI_make_current_read ***/ + { "glXMakeCurrentReadSGI", (__GLXextFuncPtr) glXMakeCurrentReadSGI }, + { "glXGetCurrentReadDrawableSGI", (__GLXextFuncPtr) glXGetCurrentReadDrawableSGI }, + + /*** GLX_SGIX_video_source ***/ +#if defined(_VL_H) + { "glXCreateGLXVideoSourceSGIX", (__GLXextFuncPtr) glXCreateGLXVideoSourceSGIX }, + { "glXDestroyGLXVideoSourceSGIX", (__GLXextFuncPtr) glXDestroyGLXVideoSourceSGIX }, +#endif + + /*** GLX_EXT_import_context ***/ + { "glXFreeContextEXT", (__GLXextFuncPtr) glXFreeContextEXT }, + { "glXGetContextIDEXT", (__GLXextFuncPtr) glXGetContextIDEXT }, + { "glXGetCurrentDisplayEXT", (__GLXextFuncPtr) glXGetCurrentDisplayEXT }, + { "glXImportContextEXT", (__GLXextFuncPtr) glXImportContextEXT }, + { "glXQueryContextInfoEXT", (__GLXextFuncPtr) glXQueryContextInfoEXT }, + + /*** GLX_SGIX_fbconfig ***/ + { "glXGetFBConfigAttribSGIX", (__GLXextFuncPtr) glXGetFBConfigAttribSGIX }, + { "glXChooseFBConfigSGIX", (__GLXextFuncPtr) glXChooseFBConfigSGIX }, + { "glXCreateGLXPixmapWithConfigSGIX", (__GLXextFuncPtr) glXCreateGLXPixmapWithConfigSGIX }, + { "glXCreateContextWithConfigSGIX", (__GLXextFuncPtr) glXCreateContextWithConfigSGIX }, + { "glXGetVisualFromFBConfigSGIX", (__GLXextFuncPtr) glXGetVisualFromFBConfigSGIX }, + { "glXGetFBConfigFromVisualSGIX", (__GLXextFuncPtr) glXGetFBConfigFromVisualSGIX }, + + /*** GLX_SGIX_pbuffer ***/ + { "glXCreateGLXPbufferSGIX", (__GLXextFuncPtr) glXCreateGLXPbufferSGIX }, + { "glXDestroyGLXPbufferSGIX", (__GLXextFuncPtr) glXDestroyGLXPbufferSGIX }, + { "glXQueryGLXPbufferSGIX", (__GLXextFuncPtr) glXQueryGLXPbufferSGIX }, + { "glXSelectEventSGIX", (__GLXextFuncPtr) glXSelectEventSGIX }, + { "glXGetSelectedEventSGIX", (__GLXextFuncPtr) glXGetSelectedEventSGIX }, + + /*** GLX_SGI_cushion ***/ + { "glXCushionSGI", (__GLXextFuncPtr) glXCushionSGI }, + + /*** GLX_SGIX_video_resize ***/ + { "glXBindChannelToWindowSGIX", (__GLXextFuncPtr) glXBindChannelToWindowSGIX }, + { "glXChannelRectSGIX", (__GLXextFuncPtr) glXChannelRectSGIX }, + { "glXQueryChannelRectSGIX", (__GLXextFuncPtr) glXQueryChannelRectSGIX }, + { "glXQueryChannelDeltasSGIX", (__GLXextFuncPtr) glXQueryChannelDeltasSGIX }, + { "glXChannelRectSyncSGIX", (__GLXextFuncPtr) glXChannelRectSyncSGIX }, + + /*** GLX_SGIX_dmbuffer **/ +#if defined(_DM_BUFFER_H_) + { "glXAssociateDMPbufferSGIX", (__GLXextFuncPtr) glXAssociateDMPbufferSGIX }, +#endif + + /*** GLX_SGIX_swap_group ***/ + { "glXJoinSwapGroupSGIX", (__GLXextFuncPtr) glXJoinSwapGroupSGIX }, + + /*** GLX_SGIX_swap_barrier ***/ + { "glXBindSwapBarrierSGIX", (__GLXextFuncPtr) glXBindSwapBarrierSGIX }, + { "glXQueryMaxSwapBarriersSGIX", (__GLXextFuncPtr) glXQueryMaxSwapBarriersSGIX }, + + /*** GLX_SUN_get_transparent_index ***/ + { "glXGetTransparentIndexSUN", (__GLXextFuncPtr) glXGetTransparentIndexSUN }, + + /*** GLX_MESA_copy_sub_buffer ***/ + { "glXCopySubBufferMESA", (__GLXextFuncPtr) glXCopySubBufferMESA }, + + /*** GLX_MESA_pixmap_colormap ***/ + { "glXCreateGLXPixmapMESA", (__GLXextFuncPtr) glXCreateGLXPixmapMESA }, + + /*** GLX_MESA_release_buffers ***/ + { "glXReleaseBuffersMESA", (__GLXextFuncPtr) glXReleaseBuffersMESA }, + + /*** GLX_ARB_get_proc_address ***/ + { "glXGetProcAddressARB", (__GLXextFuncPtr) glXGetProcAddressARB }, + + /*** GLX_EXT_texture_from_pixmap ***/ + { "glXBindTexImageEXT", (__GLXextFuncPtr) glXBindTexImageEXT }, + { "glXReleaseTexImageEXT", (__GLXextFuncPtr) glXReleaseTexImageEXT }, + + { NULL, NULL } /* end of list */ +}; + + + +/** + * Return address of named glX function, or NULL if not found. + */ +static __GLXextFuncPtr +_glxapi_get_proc_address(const char *funcName) +{ + GLuint i; + for (i = 0; GLX_functions[i].Name; i++) { + if (strcmp(GLX_functions[i].Name, funcName) == 0) + return GLX_functions[i].Address; + } + return NULL; +} + + +__GLXextFuncPtr +glXGetProcAddressARB(const GLubyte *procName) +{ + __GLXextFuncPtr f; + + f = _glxapi_get_proc_address((const char *) procName); + if (f) { + return f; + } + + f = (__GLXextFuncPtr) _glapi_get_proc_address((const char *) procName); + return f; +} + + +/* GLX 1.4 */ +void (*glXGetProcAddress(const GLubyte *procName))() +{ + return glXGetProcAddressARB(procName); +} diff --git a/src/gallium/state_trackers/glx/xlib/fakeglx_fonts.c b/src/gallium/state_trackers/glx/xlib/glx_usefont.c index e3590467563..acc64df62b6 100644 --- a/src/gallium/state_trackers/glx/xlib/fakeglx_fonts.c +++ b/src/gallium/state_trackers/glx/xlib/glx_usefont.c @@ -1,9 +1,10 @@ - /* * Mesa 3-D graphics library - * Version: 3.5 + * Version: 7.6 * - * Copyright (C) 1999-2000 Brian Paul All Rights Reserved. + * Copyright (C) 1995 Thorsten.Ohl @ Physik.TH-Darmstadt.de + * Copyright (C) 1999-2008 Brian Paul All Rights Reserved. + * Copyright (C) 2009 VMware, Inc. All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -24,13 +25,13 @@ */ -/* xfonts.c -- glXUseXFont() for Mesa written by - * Copyright (C) 1995 Thorsten.Ohl @ Physik.TH-Darmstadt.de +/** + * Fake implementation of glXUseXFont(). */ -#include "context.h" -#include "imports.h" -#include "fakeglx.h" + +#include "main/context.h" +#include "main/imports.h" #include <GL/glx.h> @@ -210,7 +211,7 @@ isvalid(XFontStruct * fs, unsigned int which) void -Fake_glXUseXFont(Font font, int first, int count, int listbase) +glXUseXFont(Font font, int first, int count, int listbase) { Display *dpy; Window win; @@ -228,7 +229,8 @@ Fake_glXUseXFont(Font font, int first, int count, int listbase) dpy = glXGetCurrentDisplay(); if (!dpy) return; /* I guess glXMakeCurrent wasn't called */ - win = RootWindow(dpy, DefaultScreen(dpy)); + i = DefaultScreen(dpy); + win = RootWindow(dpy, i); fs = XQueryFont(dpy, font); if (!fs) { diff --git a/src/gallium/state_trackers/glx/xlib/glxapi.c b/src/gallium/state_trackers/glx/xlib/glxapi.c deleted file mode 100644 index c2cb34d7cf7..00000000000 --- a/src/gallium/state_trackers/glx/xlib/glxapi.c +++ /dev/null @@ -1,1254 +0,0 @@ -/* - * Mesa 3-D graphics library - * Version: 7.1 - * - * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN - * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN - * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - - -/* - * This is the GLX API dispatcher. Calls to the glX* functions are - * either routed to the real GLX encoders or to Mesa's pseudo-GLX functions. - * See the glxapi.h file for more details. - */ - - -#include <assert.h> -#include <stdlib.h> -#include <stdio.h> -#include <string.h> -#include "glapi/glapi.h" -#include "glxapi.h" -#include "fakeglx.h" -#include "pipe/p_thread.h" - - -#if defined(__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__) >= 303 -# define PUBLIC __attribute__((visibility("default"))) -# define USED __attribute__((used)) -#else -# define PUBLIC -# define USED -#endif - - -struct display_dispatch { - Display *Dpy; - struct _glxapi_table *Table; - struct display_dispatch *Next; -}; - -static struct display_dispatch *DispatchList = NULL; - - -/* Display -> Dispatch caching */ -static Display *prevDisplay = NULL; -static struct _glxapi_table *prevTable = NULL; - - -static struct _glxapi_table * -get_dispatch(Display *dpy) -{ - if (!dpy) - return NULL; - - /* search list of display/dispatch pairs for this display */ - { - const struct display_dispatch *d = DispatchList; - while (d) { - if (d->Dpy == dpy) { - prevDisplay = dpy; - prevTable = d->Table; - return d->Table; /* done! */ - } - d = d->Next; - } - } - - /* A new display, determine if we should use real GLX - * or Mesa's pseudo-GLX. - */ - { - struct _glxapi_table *t = _mesa_GetGLXDispatchTable(); - - if (t) { - struct display_dispatch *d; - d = (struct display_dispatch *) malloc(sizeof(struct display_dispatch)); - if (d) { - d->Dpy = dpy; - d->Table = t; - /* insert at head of list */ - d->Next = DispatchList; - DispatchList = d; - /* update cache */ - prevDisplay = dpy; - prevTable = t; - return t; - } - } - } - - /* If we get here that means we can't use real GLX on this display - * and the Mesa pseudo-GLX software renderer wasn't compiled in. - * Or, we ran out of memory! - */ - return NULL; -} - - -/* Don't use the GET_DISPATCH defined in glthread.h */ -#undef GET_DISPATCH - -#define GET_DISPATCH(DPY, TABLE) \ - if (DPY == prevDisplay) { \ - TABLE = prevTable; \ - } \ - else if (!DPY) { \ - TABLE = NULL; \ - } \ - else { \ - TABLE = get_dispatch(DPY); \ - } - - - - -/** - * GLX API current context. - */ -pipe_tsd ContextTSD; - - -static void -SetCurrentContext(GLXContext c) -{ - pipe_tsd_set(&ContextTSD, c); -} - - -/* - * GLX API entrypoints - */ - -/*** GLX_VERSION_1_0 ***/ - -XVisualInfo PUBLIC * -glXChooseVisual(Display *dpy, int screen, int *list) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return NULL; - return (t->ChooseVisual)(dpy, screen, list); -} - - -void PUBLIC -glXCopyContext(Display *dpy, GLXContext src, GLXContext dst, unsigned long mask) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->CopyContext)(dpy, src, dst, mask); -} - - -GLXContext PUBLIC -glXCreateContext(Display *dpy, XVisualInfo *visinfo, GLXContext shareList, Bool direct) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->CreateContext)(dpy, visinfo, shareList, direct); -} - - -GLXPixmap PUBLIC -glXCreateGLXPixmap(Display *dpy, XVisualInfo *visinfo, Pixmap pixmap) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->CreateGLXPixmap)(dpy, visinfo, pixmap); -} - - -void PUBLIC -glXDestroyContext(Display *dpy, GLXContext ctx) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - if (glXGetCurrentContext() == ctx) - SetCurrentContext(NULL); - (t->DestroyContext)(dpy, ctx); -} - - -void PUBLIC -glXDestroyGLXPixmap(Display *dpy, GLXPixmap pixmap) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->DestroyGLXPixmap)(dpy, pixmap); -} - - -int PUBLIC -glXGetConfig(Display *dpy, XVisualInfo *visinfo, int attrib, int *value) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return GLX_NO_EXTENSION; - return (t->GetConfig)(dpy, visinfo, attrib, value); -} - - -GLXContext PUBLIC -glXGetCurrentContext(void) -{ - return (GLXContext) pipe_tsd_get(&ContextTSD); -} - - -GLXDrawable PUBLIC -glXGetCurrentDrawable(void) -{ - __GLXcontext *gc = (__GLXcontext *) glXGetCurrentContext(); - return gc ? gc->currentDrawable : 0; -} - - -Bool PUBLIC -glXIsDirect(Display *dpy, GLXContext ctx) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return False; - return (t->IsDirect)(dpy, ctx); -} - - -Bool PUBLIC -glXMakeCurrent(Display *dpy, GLXDrawable drawable, GLXContext ctx) -{ - Bool b; - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) { - return False; - } - b = (*t->MakeCurrent)(dpy, drawable, ctx); - if (b) { - SetCurrentContext(ctx); - } - return b; -} - - -Bool PUBLIC -glXQueryExtension(Display *dpy, int *errorb, int *event) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return False; - return (t->QueryExtension)(dpy, errorb, event); -} - - -Bool PUBLIC -glXQueryVersion(Display *dpy, int *maj, int *min) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return False; - return (t->QueryVersion)(dpy, maj, min); -} - - -void PUBLIC -glXSwapBuffers(Display *dpy, GLXDrawable drawable) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->SwapBuffers)(dpy, drawable); -} - - -void PUBLIC -glXUseXFont(Font font, int first, int count, int listBase) -{ - struct _glxapi_table *t; - Display *dpy = glXGetCurrentDisplay(); - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->UseXFont)(font, first, count, listBase); -} - - -void PUBLIC -glXWaitGL(void) -{ - struct _glxapi_table *t; - Display *dpy = glXGetCurrentDisplay(); - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->WaitGL)(); -} - - -void PUBLIC -glXWaitX(void) -{ - struct _glxapi_table *t; - Display *dpy = glXGetCurrentDisplay(); - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->WaitX)(); -} - - - -/*** GLX_VERSION_1_1 ***/ - -const char PUBLIC * -glXGetClientString(Display *dpy, int name) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return NULL; - return (t->GetClientString)(dpy, name); -} - - -const char PUBLIC * -glXQueryExtensionsString(Display *dpy, int screen) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return NULL; - return (t->QueryExtensionsString)(dpy, screen); -} - - -const char PUBLIC * -glXQueryServerString(Display *dpy, int screen, int name) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return NULL; - return (t->QueryServerString)(dpy, screen, name); -} - - -/*** GLX_VERSION_1_2 ***/ - -Display PUBLIC * -glXGetCurrentDisplay(void) -{ - /* Same code as in libGL's glxext.c */ - __GLXcontext *gc = (__GLXcontext *) glXGetCurrentContext(); - if (NULL == gc) return NULL; - return gc->currentDpy; -} - - - -/*** GLX_VERSION_1_3 ***/ - -GLXFBConfig PUBLIC * -glXChooseFBConfig(Display *dpy, int screen, const int *attribList, int *nitems) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->ChooseFBConfig)(dpy, screen, attribList, nitems); -} - - -GLXContext PUBLIC -glXCreateNewContext(Display *dpy, GLXFBConfig config, int renderType, GLXContext shareList, Bool direct) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->CreateNewContext)(dpy, config, renderType, shareList, direct); -} - - -GLXPbuffer PUBLIC -glXCreatePbuffer(Display *dpy, GLXFBConfig config, const int *attribList) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->CreatePbuffer)(dpy, config, attribList); -} - - -GLXPixmap PUBLIC -glXCreatePixmap(Display *dpy, GLXFBConfig config, Pixmap pixmap, const int *attribList) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->CreatePixmap)(dpy, config, pixmap, attribList); -} - - -GLXWindow PUBLIC -glXCreateWindow(Display *dpy, GLXFBConfig config, Window win, const int *attribList) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->CreateWindow)(dpy, config, win, attribList); -} - - -void PUBLIC -glXDestroyPbuffer(Display *dpy, GLXPbuffer pbuf) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->DestroyPbuffer)(dpy, pbuf); -} - - -void PUBLIC -glXDestroyPixmap(Display *dpy, GLXPixmap pixmap) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->DestroyPixmap)(dpy, pixmap); -} - - -void PUBLIC -glXDestroyWindow(Display *dpy, GLXWindow window) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->DestroyWindow)(dpy, window); -} - - -GLXDrawable PUBLIC -glXGetCurrentReadDrawable(void) -{ - __GLXcontext *gc = (__GLXcontext *) glXGetCurrentContext(); - return gc ? gc->currentReadable : 0; -} - - -int PUBLIC -glXGetFBConfigAttrib(Display *dpy, GLXFBConfig config, int attribute, int *value) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return GLX_NO_EXTENSION; - return (t->GetFBConfigAttrib)(dpy, config, attribute, value); -} - - -GLXFBConfig PUBLIC * -glXGetFBConfigs(Display *dpy, int screen, int *nelements) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->GetFBConfigs)(dpy, screen, nelements); -} - -void PUBLIC -glXGetSelectedEvent(Display *dpy, GLXDrawable drawable, unsigned long *mask) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->GetSelectedEvent)(dpy, drawable, mask); -} - - -XVisualInfo PUBLIC * -glXGetVisualFromFBConfig(Display *dpy, GLXFBConfig config) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return NULL; - return (t->GetVisualFromFBConfig)(dpy, config); -} - - -Bool PUBLIC -glXMakeContextCurrent(Display *dpy, GLXDrawable draw, GLXDrawable read, GLXContext ctx) -{ - Bool b; - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return False; - b = (t->MakeContextCurrent)(dpy, draw, read, ctx); - if (b) { - SetCurrentContext(ctx); - } - return b; -} - - -int PUBLIC -glXQueryContext(Display *dpy, GLXContext ctx, int attribute, int *value) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - assert(t); - if (!t) - return 0; /* XXX correct? */ - return (t->QueryContext)(dpy, ctx, attribute, value); -} - - -void PUBLIC -glXQueryDrawable(Display *dpy, GLXDrawable draw, int attribute, unsigned int *value) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->QueryDrawable)(dpy, draw, attribute, value); -} - - -void PUBLIC -glXSelectEvent(Display *dpy, GLXDrawable drawable, unsigned long mask) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->SelectEvent)(dpy, drawable, mask); -} - - - -/*** GLX_SGI_swap_control ***/ - -int PUBLIC -glXSwapIntervalSGI(int interval) -{ - struct _glxapi_table *t; - Display *dpy = glXGetCurrentDisplay(); - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->SwapIntervalSGI)(interval); -} - - - -/*** GLX_SGI_video_sync ***/ - -int PUBLIC -glXGetVideoSyncSGI(unsigned int *count) -{ - struct _glxapi_table *t; - Display *dpy = glXGetCurrentDisplay(); - GET_DISPATCH(dpy, t); - if (!t || !glXGetCurrentContext()) - return GLX_BAD_CONTEXT; - return (t->GetVideoSyncSGI)(count); -} - -int PUBLIC -glXWaitVideoSyncSGI(int divisor, int remainder, unsigned int *count) -{ - struct _glxapi_table *t; - Display *dpy = glXGetCurrentDisplay(); - GET_DISPATCH(dpy, t); - if (!t || !glXGetCurrentContext()) - return GLX_BAD_CONTEXT; - return (t->WaitVideoSyncSGI)(divisor, remainder, count); -} - - - -/*** GLX_SGI_make_current_read ***/ - -Bool PUBLIC -glXMakeCurrentReadSGI(Display *dpy, GLXDrawable draw, GLXDrawable read, GLXContext ctx) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return False; - return (t->MakeCurrentReadSGI)(dpy, draw, read, ctx); -} - -GLXDrawable PUBLIC -glXGetCurrentReadDrawableSGI(void) -{ - return glXGetCurrentReadDrawable(); -} - - -#if defined(_VL_H) - -GLXVideoSourceSGIX PUBLIC -glXCreateGLXVideoSourceSGIX(Display *dpy, int screen, VLServer server, VLPath path, int nodeClass, VLNode drainNode) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->CreateGLXVideoSourceSGIX)(dpy, screen, server, path, nodeClass, drainNode); -} - -void PUBLIC -glXDestroyGLXVideoSourceSGIX(Display *dpy, GLXVideoSourceSGIX src) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->DestroyGLXVideoSourceSGIX)(dpy, src); -} - -#endif - - -/*** GLX_EXT_import_context ***/ - -void PUBLIC -glXFreeContextEXT(Display *dpy, GLXContext context) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->FreeContextEXT)(dpy, context); -} - -GLXContextID PUBLIC -glXGetContextIDEXT(const GLXContext context) -{ - return ((__GLXcontext *) context)->xid; -} - -Display PUBLIC * -glXGetCurrentDisplayEXT(void) -{ - return glXGetCurrentDisplay(); -} - -GLXContext PUBLIC -glXImportContextEXT(Display *dpy, GLXContextID contextID) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->ImportContextEXT)(dpy, contextID); -} - -int PUBLIC -glXQueryContextInfoEXT(Display *dpy, GLXContext context, int attribute,int *value) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; /* XXX ok? */ - return (t->QueryContextInfoEXT)(dpy, context, attribute, value); -} - - - -/*** GLX_SGIX_fbconfig ***/ - -int PUBLIC -glXGetFBConfigAttribSGIX(Display *dpy, GLXFBConfigSGIX config, int attribute, int *value) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->GetFBConfigAttribSGIX)(dpy, config, attribute, value); -} - -GLXFBConfigSGIX PUBLIC * -glXChooseFBConfigSGIX(Display *dpy, int screen, int *attrib_list, int *nelements) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->ChooseFBConfigSGIX)(dpy, screen, attrib_list, nelements); -} - -GLXPixmap PUBLIC -glXCreateGLXPixmapWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, Pixmap pixmap) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->CreateGLXPixmapWithConfigSGIX)(dpy, config, pixmap); -} - -GLXContext PUBLIC -glXCreateContextWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, int render_type, GLXContext share_list, Bool direct) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->CreateContextWithConfigSGIX)(dpy, config, render_type, share_list, direct); -} - -XVisualInfo PUBLIC * -glXGetVisualFromFBConfigSGIX(Display *dpy, GLXFBConfigSGIX config) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->GetVisualFromFBConfigSGIX)(dpy, config); -} - -GLXFBConfigSGIX PUBLIC -glXGetFBConfigFromVisualSGIX(Display *dpy, XVisualInfo *vis) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->GetFBConfigFromVisualSGIX)(dpy, vis); -} - - - -/*** GLX_SGIX_pbuffer ***/ - -GLXPbufferSGIX PUBLIC -glXCreateGLXPbufferSGIX(Display *dpy, GLXFBConfigSGIX config, unsigned int width, unsigned int height, int *attrib_list) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->CreateGLXPbufferSGIX)(dpy, config, width, height, attrib_list); -} - -void PUBLIC -glXDestroyGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->DestroyGLXPbufferSGIX)(dpy, pbuf); -} - -int PUBLIC -glXQueryGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf, int attribute, unsigned int *value) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->QueryGLXPbufferSGIX)(dpy, pbuf, attribute, value); -} - -void PUBLIC -glXSelectEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long mask) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->SelectEventSGIX)(dpy, drawable, mask); -} - -void PUBLIC -glXGetSelectedEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long *mask) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->GetSelectedEventSGIX)(dpy, drawable, mask); -} - - - -/*** GLX_SGI_cushion ***/ - -void PUBLIC -glXCushionSGI(Display *dpy, Window win, float cushion) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->CushionSGI)(dpy, win, cushion); -} - - - -/*** GLX_SGIX_video_resize ***/ - -int PUBLIC -glXBindChannelToWindowSGIX(Display *dpy, int screen, int channel , Window window) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->BindChannelToWindowSGIX)(dpy, screen, channel, window); -} - -int PUBLIC -glXChannelRectSGIX(Display *dpy, int screen, int channel, int x, int y, int w, int h) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->ChannelRectSGIX)(dpy, screen, channel, x, y, w, h); -} - -int PUBLIC -glXQueryChannelRectSGIX(Display *dpy, int screen, int channel, int *x, int *y, int *w, int *h) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->QueryChannelRectSGIX)(dpy, screen, channel, x, y, w, h); -} - -int PUBLIC -glXQueryChannelDeltasSGIX(Display *dpy, int screen, int channel, int *dx, int *dy, int *dw, int *dh) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->QueryChannelDeltasSGIX)(dpy, screen, channel, dx, dy, dw, dh); -} - -int PUBLIC -glXChannelRectSyncSGIX(Display *dpy, int screen, int channel, GLenum synctype) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->ChannelRectSyncSGIX)(dpy, screen, channel, synctype); -} - - - -#if defined(_DM_BUFFER_H_) - -Bool PUBLIC -glXAssociateDMPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuffer, DMparams *params, DMbuffer dmbuffer) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return False; - return (t->AssociateDMPbufferSGIX)(dpy, pbuffer, params, dmbuffer); -} - -#endif - - -/*** GLX_SGIX_swap_group ***/ - -void PUBLIC -glXJoinSwapGroupSGIX(Display *dpy, GLXDrawable drawable, GLXDrawable member) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (*t->JoinSwapGroupSGIX)(dpy, drawable, member); -} - - -/*** GLX_SGIX_swap_barrier ***/ - -void PUBLIC -glXBindSwapBarrierSGIX(Display *dpy, GLXDrawable drawable, int barrier) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (*t->BindSwapBarrierSGIX)(dpy, drawable, barrier); -} - -Bool PUBLIC -glXQueryMaxSwapBarriersSGIX(Display *dpy, int screen, int *max) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return False; - return (*t->QueryMaxSwapBarriersSGIX)(dpy, screen, max); -} - - - -/*** GLX_SUN_get_transparent_index ***/ - -Status PUBLIC -glXGetTransparentIndexSUN(Display *dpy, Window overlay, Window underlay, long *pTransparent) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return False; - return (*t->GetTransparentIndexSUN)(dpy, overlay, underlay, pTransparent); -} - - - -/*** GLX_MESA_copy_sub_buffer ***/ - -void PUBLIC -glXCopySubBufferMESA(Display *dpy, GLXDrawable drawable, int x, int y, int width, int height) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return; - (t->CopySubBufferMESA)(dpy, drawable, x, y, width, height); -} - - - -/*** GLX_MESA_release_buffers ***/ - -Bool PUBLIC -glXReleaseBuffersMESA(Display *dpy, Window w) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return False; - return (t->ReleaseBuffersMESA)(dpy, w); -} - - - -/*** GLX_MESA_pixmap_colormap ***/ - -GLXPixmap PUBLIC -glXCreateGLXPixmapMESA(Display *dpy, XVisualInfo *visinfo, Pixmap pixmap, Colormap cmap) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (!t) - return 0; - return (t->CreateGLXPixmapMESA)(dpy, visinfo, pixmap, cmap); -} - -/*** GLX_EXT_texture_from_pixmap */ - -void -glXBindTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer, - const int *attrib_list) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (t) - t->BindTexImageEXT(dpy, drawable, buffer, attrib_list); -} - -void -glXReleaseTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer) -{ - struct _glxapi_table *t; - GET_DISPATCH(dpy, t); - if (t) - t->ReleaseTexImageEXT(dpy, drawable, buffer); -} - - -/**********************************************************************/ -/* GLX API management functions */ -/**********************************************************************/ - - -const char * -_glxapi_get_version(void) -{ - return "1.3"; -} - - - -/* - * Return size of the GLX dispatch table, in entries, not bytes. - */ -GLuint -_glxapi_get_dispatch_table_size(void) -{ - return sizeof(struct _glxapi_table) / sizeof(void *); -} - - -static int -generic_no_op_func(void) -{ - return 0; -} - - -/* - * Initialize all functions in given dispatch table to be no-ops - */ -void -_glxapi_set_no_op_table(struct _glxapi_table *t) -{ - typedef int (*nop_func)(void); - nop_func *dispatch = (nop_func *) t; - GLuint n = _glxapi_get_dispatch_table_size(); - GLuint i; - for (i = 0; i < n; i++) { - dispatch[i] = generic_no_op_func; - } -} - - -struct name_address_pair { - const char *Name; - __GLXextFuncPtr Address; -}; - -static struct name_address_pair GLX_functions[] = { - /*** GLX_VERSION_1_0 ***/ - { "glXChooseVisual", (__GLXextFuncPtr) glXChooseVisual }, - { "glXCopyContext", (__GLXextFuncPtr) glXCopyContext }, - { "glXCreateContext", (__GLXextFuncPtr) glXCreateContext }, - { "glXCreateGLXPixmap", (__GLXextFuncPtr) glXCreateGLXPixmap }, - { "glXDestroyContext", (__GLXextFuncPtr) glXDestroyContext }, - { "glXDestroyGLXPixmap", (__GLXextFuncPtr) glXDestroyGLXPixmap }, - { "glXGetConfig", (__GLXextFuncPtr) glXGetConfig }, - { "glXGetCurrentContext", (__GLXextFuncPtr) glXGetCurrentContext }, - { "glXGetCurrentDrawable", (__GLXextFuncPtr) glXGetCurrentDrawable }, - { "glXIsDirect", (__GLXextFuncPtr) glXIsDirect }, - { "glXMakeCurrent", (__GLXextFuncPtr) glXMakeCurrent }, - { "glXQueryExtension", (__GLXextFuncPtr) glXQueryExtension }, - { "glXQueryVersion", (__GLXextFuncPtr) glXQueryVersion }, - { "glXSwapBuffers", (__GLXextFuncPtr) glXSwapBuffers }, - { "glXUseXFont", (__GLXextFuncPtr) glXUseXFont }, - { "glXWaitGL", (__GLXextFuncPtr) glXWaitGL }, - { "glXWaitX", (__GLXextFuncPtr) glXWaitX }, - - /*** GLX_VERSION_1_1 ***/ - { "glXGetClientString", (__GLXextFuncPtr) glXGetClientString }, - { "glXQueryExtensionsString", (__GLXextFuncPtr) glXQueryExtensionsString }, - { "glXQueryServerString", (__GLXextFuncPtr) glXQueryServerString }, - - /*** GLX_VERSION_1_2 ***/ - { "glXGetCurrentDisplay", (__GLXextFuncPtr) glXGetCurrentDisplay }, - - /*** GLX_VERSION_1_3 ***/ - { "glXChooseFBConfig", (__GLXextFuncPtr) glXChooseFBConfig }, - { "glXCreateNewContext", (__GLXextFuncPtr) glXCreateNewContext }, - { "glXCreatePbuffer", (__GLXextFuncPtr) glXCreatePbuffer }, - { "glXCreatePixmap", (__GLXextFuncPtr) glXCreatePixmap }, - { "glXCreateWindow", (__GLXextFuncPtr) glXCreateWindow }, - { "glXDestroyPbuffer", (__GLXextFuncPtr) glXDestroyPbuffer }, - { "glXDestroyPixmap", (__GLXextFuncPtr) glXDestroyPixmap }, - { "glXDestroyWindow", (__GLXextFuncPtr) glXDestroyWindow }, - { "glXGetCurrentReadDrawable", (__GLXextFuncPtr) glXGetCurrentReadDrawable }, - { "glXGetFBConfigAttrib", (__GLXextFuncPtr) glXGetFBConfigAttrib }, - { "glXGetFBConfigs", (__GLXextFuncPtr) glXGetFBConfigs }, - { "glXGetSelectedEvent", (__GLXextFuncPtr) glXGetSelectedEvent }, - { "glXGetVisualFromFBConfig", (__GLXextFuncPtr) glXGetVisualFromFBConfig }, - { "glXMakeContextCurrent", (__GLXextFuncPtr) glXMakeContextCurrent }, - { "glXQueryContext", (__GLXextFuncPtr) glXQueryContext }, - { "glXQueryDrawable", (__GLXextFuncPtr) glXQueryDrawable }, - { "glXSelectEvent", (__GLXextFuncPtr) glXSelectEvent }, - - /*** GLX_VERSION_1_4 ***/ - { "glXGetProcAddress", (__GLXextFuncPtr) glXGetProcAddress }, - - /*** GLX_SGI_swap_control ***/ - { "glXSwapIntervalSGI", (__GLXextFuncPtr) glXSwapIntervalSGI }, - - /*** GLX_SGI_video_sync ***/ - { "glXGetVideoSyncSGI", (__GLXextFuncPtr) glXGetVideoSyncSGI }, - { "glXWaitVideoSyncSGI", (__GLXextFuncPtr) glXWaitVideoSyncSGI }, - - /*** GLX_SGI_make_current_read ***/ - { "glXMakeCurrentReadSGI", (__GLXextFuncPtr) glXMakeCurrentReadSGI }, - { "glXGetCurrentReadDrawableSGI", (__GLXextFuncPtr) glXGetCurrentReadDrawableSGI }, - - /*** GLX_SGIX_video_source ***/ -#if defined(_VL_H) - { "glXCreateGLXVideoSourceSGIX", (__GLXextFuncPtr) glXCreateGLXVideoSourceSGIX }, - { "glXDestroyGLXVideoSourceSGIX", (__GLXextFuncPtr) glXDestroyGLXVideoSourceSGIX }, -#endif - - /*** GLX_EXT_import_context ***/ - { "glXFreeContextEXT", (__GLXextFuncPtr) glXFreeContextEXT }, - { "glXGetContextIDEXT", (__GLXextFuncPtr) glXGetContextIDEXT }, - { "glXGetCurrentDisplayEXT", (__GLXextFuncPtr) glXGetCurrentDisplayEXT }, - { "glXImportContextEXT", (__GLXextFuncPtr) glXImportContextEXT }, - { "glXQueryContextInfoEXT", (__GLXextFuncPtr) glXQueryContextInfoEXT }, - - /*** GLX_SGIX_fbconfig ***/ - { "glXGetFBConfigAttribSGIX", (__GLXextFuncPtr) glXGetFBConfigAttribSGIX }, - { "glXChooseFBConfigSGIX", (__GLXextFuncPtr) glXChooseFBConfigSGIX }, - { "glXCreateGLXPixmapWithConfigSGIX", (__GLXextFuncPtr) glXCreateGLXPixmapWithConfigSGIX }, - { "glXCreateContextWithConfigSGIX", (__GLXextFuncPtr) glXCreateContextWithConfigSGIX }, - { "glXGetVisualFromFBConfigSGIX", (__GLXextFuncPtr) glXGetVisualFromFBConfigSGIX }, - { "glXGetFBConfigFromVisualSGIX", (__GLXextFuncPtr) glXGetFBConfigFromVisualSGIX }, - - /*** GLX_SGIX_pbuffer ***/ - { "glXCreateGLXPbufferSGIX", (__GLXextFuncPtr) glXCreateGLXPbufferSGIX }, - { "glXDestroyGLXPbufferSGIX", (__GLXextFuncPtr) glXDestroyGLXPbufferSGIX }, - { "glXQueryGLXPbufferSGIX", (__GLXextFuncPtr) glXQueryGLXPbufferSGIX }, - { "glXSelectEventSGIX", (__GLXextFuncPtr) glXSelectEventSGIX }, - { "glXGetSelectedEventSGIX", (__GLXextFuncPtr) glXGetSelectedEventSGIX }, - - /*** GLX_SGI_cushion ***/ - { "glXCushionSGI", (__GLXextFuncPtr) glXCushionSGI }, - - /*** GLX_SGIX_video_resize ***/ - { "glXBindChannelToWindowSGIX", (__GLXextFuncPtr) glXBindChannelToWindowSGIX }, - { "glXChannelRectSGIX", (__GLXextFuncPtr) glXChannelRectSGIX }, - { "glXQueryChannelRectSGIX", (__GLXextFuncPtr) glXQueryChannelRectSGIX }, - { "glXQueryChannelDeltasSGIX", (__GLXextFuncPtr) glXQueryChannelDeltasSGIX }, - { "glXChannelRectSyncSGIX", (__GLXextFuncPtr) glXChannelRectSyncSGIX }, - - /*** GLX_SGIX_dmbuffer **/ -#if defined(_DM_BUFFER_H_) - { "glXAssociateDMPbufferSGIX", (__GLXextFuncPtr) glXAssociateDMPbufferSGIX }, -#endif - - /*** GLX_SGIX_swap_group ***/ - { "glXJoinSwapGroupSGIX", (__GLXextFuncPtr) glXJoinSwapGroupSGIX }, - - /*** GLX_SGIX_swap_barrier ***/ - { "glXBindSwapBarrierSGIX", (__GLXextFuncPtr) glXBindSwapBarrierSGIX }, - { "glXQueryMaxSwapBarriersSGIX", (__GLXextFuncPtr) glXQueryMaxSwapBarriersSGIX }, - - /*** GLX_SUN_get_transparent_index ***/ - { "glXGetTransparentIndexSUN", (__GLXextFuncPtr) glXGetTransparentIndexSUN }, - - /*** GLX_MESA_copy_sub_buffer ***/ - { "glXCopySubBufferMESA", (__GLXextFuncPtr) glXCopySubBufferMESA }, - - /*** GLX_MESA_pixmap_colormap ***/ - { "glXCreateGLXPixmapMESA", (__GLXextFuncPtr) glXCreateGLXPixmapMESA }, - - /*** GLX_MESA_release_buffers ***/ - { "glXReleaseBuffersMESA", (__GLXextFuncPtr) glXReleaseBuffersMESA }, - - /*** GLX_ARB_get_proc_address ***/ - { "glXGetProcAddressARB", (__GLXextFuncPtr) glXGetProcAddressARB }, - - /*** GLX_EXT_texture_from_pixmap ***/ - { "glXBindTexImageEXT", (__GLXextFuncPtr) glXBindTexImageEXT }, - { "glXReleaseTexImageEXT", (__GLXextFuncPtr) glXReleaseTexImageEXT }, - - { NULL, NULL } /* end of list */ -}; - - - -/* - * Return address of named glX function, or NULL if not found. - */ -__GLXextFuncPtr -_glxapi_get_proc_address(const char *funcName) -{ - GLuint i; - for (i = 0; GLX_functions[i].Name; i++) { - if (strcmp(GLX_functions[i].Name, funcName) == 0) - return GLX_functions[i].Address; - } - return NULL; -} - - - -/* - * This function does not get dispatched through the dispatch table - * since it's really a "meta" function. - */ -__GLXextFuncPtr -glXGetProcAddressARB(const GLubyte *procName) -{ - __GLXextFuncPtr f; - - f = _glxapi_get_proc_address((const char *) procName); - if (f) { - return f; - } - - f = (__GLXextFuncPtr) _glapi_get_proc_address((const char *) procName); - return f; -} - - -/* GLX 1.4 */ -void (*glXGetProcAddress(const GLubyte *procName))() -{ - return glXGetProcAddressARB(procName); -} diff --git a/src/gallium/state_trackers/glx/xlib/glxapi.h b/src/gallium/state_trackers/glx/xlib/glxapi.h deleted file mode 100644 index b4e12b4162b..00000000000 --- a/src/gallium/state_trackers/glx/xlib/glxapi.h +++ /dev/null @@ -1,213 +0,0 @@ -/* - * Mesa 3-D graphics library - * Version: 6.3 - * - * Copyright (C) 1999-2004 Brian Paul All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN - * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN - * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - - -#ifndef _glxapi_h_ -#define _glxapi_h_ - - -#define GLX_GLXEXT_PROTOTYPES -#include "GL/glx.h" - - -/* The GLX API dispatcher (i.e. this code) is being built into stand-alone - * Mesa. We don't know anything about XFree86 or real GLX so we define a - * minimal __GLXContextRec here so some of the functions in this file can - * work properly. - */ -typedef struct __GLXcontextRec { - Display *currentDpy; - GLboolean isDirect; - GLXDrawable currentDrawable; - GLXDrawable currentReadable; - XID xid; -} __GLXcontext; - - -/* - * Almost all the GLX API functions get routed through this dispatch table. - * The exceptions are the glXGetCurrentXXX() functions. - * - * This dispatch table allows multiple GLX client-side modules to coexist. - * Specifically, a real GLX library (like SGI's or the Utah GLX) and Mesa's - * pseudo-GLX can be present at the same time. The former being used on - * GLX-enabled X servers and the later on non-GLX X servers. - * - * Red Hat has been using this since Red Hat Linux 7.0 (I think). - * This'll be a standard feature in XFree86 4.3. It basically allows one - * libGL to do both DRI-rendering and "fake GLX" rendering to X displays - * that lack the GLX extension. - */ -struct _glxapi_table { - /*** GLX_VERSION_1_0 ***/ - XVisualInfo *(*ChooseVisual)(Display *dpy, int screen, int *list); - void (*CopyContext)(Display *dpy, GLXContext src, GLXContext dst, unsigned long mask); - GLXContext (*CreateContext)(Display *dpy, XVisualInfo *visinfo, GLXContext shareList, Bool direct); - GLXPixmap (*CreateGLXPixmap)(Display *dpy, XVisualInfo *visinfo, Pixmap pixmap); - void (*DestroyContext)(Display *dpy, GLXContext ctx); - void (*DestroyGLXPixmap)(Display *dpy, GLXPixmap pixmap); - int (*GetConfig)(Display *dpy, XVisualInfo *visinfo, int attrib, int *value); - /*GLXContext (*GetCurrentContext)(void);*/ - /*GLXDrawable (*GetCurrentDrawable)(void);*/ - Bool (*IsDirect)(Display *dpy, GLXContext ctx); - Bool (*MakeCurrent)(Display *dpy, GLXDrawable drawable, GLXContext ctx); - Bool (*QueryExtension)(Display *dpy, int *errorb, int *event); - Bool (*QueryVersion)(Display *dpy, int *maj, int *min); - void (*SwapBuffers)(Display *dpy, GLXDrawable drawable); - void (*UseXFont)(Font font, int first, int count, int listBase); - void (*WaitGL)(void); - void (*WaitX)(void); - - /*** GLX_VERSION_1_1 ***/ - const char *(*GetClientString)(Display *dpy, int name); - const char *(*QueryExtensionsString)(Display *dpy, int screen); - const char *(*QueryServerString)(Display *dpy, int screen, int name); - - /*** GLX_VERSION_1_2 ***/ - /*Display *(*GetCurrentDisplay)(void);*/ - - /*** GLX_VERSION_1_3 ***/ - GLXFBConfig *(*ChooseFBConfig)(Display *dpy, int screen, const int *attribList, int *nitems); - GLXContext (*CreateNewContext)(Display *dpy, GLXFBConfig config, int renderType, GLXContext shareList, Bool direct); - GLXPbuffer (*CreatePbuffer)(Display *dpy, GLXFBConfig config, const int *attribList); - GLXPixmap (*CreatePixmap)(Display *dpy, GLXFBConfig config, Pixmap pixmap, const int *attribList); - GLXWindow (*CreateWindow)(Display *dpy, GLXFBConfig config, Window win, const int *attribList); - void (*DestroyPbuffer)(Display *dpy, GLXPbuffer pbuf); - void (*DestroyPixmap)(Display *dpy, GLXPixmap pixmap); - void (*DestroyWindow)(Display *dpy, GLXWindow window); - /*GLXDrawable (*GetCurrentReadDrawable)(void);*/ - int (*GetFBConfigAttrib)(Display *dpy, GLXFBConfig config, int attribute, int *value); - GLXFBConfig *(*GetFBConfigs)(Display *dpy, int screen, int *nelements); - void (*GetSelectedEvent)(Display *dpy, GLXDrawable drawable, unsigned long *mask); - XVisualInfo *(*GetVisualFromFBConfig)(Display *dpy, GLXFBConfig config); - Bool (*MakeContextCurrent)(Display *dpy, GLXDrawable draw, GLXDrawable read, GLXContext ctx); - int (*QueryContext)(Display *dpy, GLXContext ctx, int attribute, int *value); - void (*QueryDrawable)(Display *dpy, GLXDrawable draw, int attribute, unsigned int *value); - void (*SelectEvent)(Display *dpy, GLXDrawable drawable, unsigned long mask); - - /*** GLX_SGI_swap_control ***/ - int (*SwapIntervalSGI)(int); - - /*** GLX_SGI_video_sync ***/ - int (*GetVideoSyncSGI)(unsigned int *count); - int (*WaitVideoSyncSGI)(int divisor, int remainder, unsigned int *count); - - /*** GLX_SGI_make_current_read ***/ - Bool (*MakeCurrentReadSGI)(Display *, GLXDrawable, GLXDrawable, GLXContext); - /*GLXDrawable (*GetCurrentReadDrawableSGI)(void);*/ - - /*** GLX_SGIX_video_source (needs video library) ***/ -#if defined(_VL_H_) - GLXVideoSourceSGIX (*CreateGLXVideoSourceSGIX)(Display *, int, VLServer, VLPath, int, VLNode); - void (*DestroyGLXVideoSourceSGIX)(Display *, GLXVideoSourceSGIX); -#else - void *CreateGLXVideoSourceSGIX; - void *DestroyGLXVideoSourceSGIX; -#endif - - /*** GLX_EXT_import_context ***/ - void (*FreeContextEXT)(Display *dpy, GLXContext context); - GLXContextID (*GetContextIDEXT)(const GLXContext context); - /*Display *(*GetCurrentDisplayEXT)(void);*/ - GLXContext (*ImportContextEXT)(Display *dpy, GLXContextID contextID); - int (*QueryContextInfoEXT)(Display *dpy, GLXContext context, int attribute,int *value); - - /*** GLX_SGIX_fbconfig ***/ - int (*GetFBConfigAttribSGIX)(Display *, GLXFBConfigSGIX, int, int *); - GLXFBConfigSGIX * (*ChooseFBConfigSGIX)(Display *, int, int *, int *); - GLXPixmap (*CreateGLXPixmapWithConfigSGIX)(Display *, GLXFBConfigSGIX, Pixmap); - GLXContext (*CreateContextWithConfigSGIX)(Display *, GLXFBConfigSGIX, int, GLXContext, Bool); - XVisualInfo * (*GetVisualFromFBConfigSGIX)(Display *, GLXFBConfigSGIX); - GLXFBConfigSGIX (*GetFBConfigFromVisualSGIX)(Display *, XVisualInfo *); - - /*** GLX_SGIX_pbuffer ***/ - GLXPbufferSGIX (*CreateGLXPbufferSGIX)(Display *, GLXFBConfigSGIX, unsigned int, unsigned int, int *); - void (*DestroyGLXPbufferSGIX)(Display *, GLXPbufferSGIX); - int (*QueryGLXPbufferSGIX)(Display *, GLXPbufferSGIX, int, unsigned int *); - void (*SelectEventSGIX)(Display *, GLXDrawable, unsigned long); - void (*GetSelectedEventSGIX)(Display *, GLXDrawable, unsigned long *); - - /*** GLX_SGI_cushion ***/ - void (*CushionSGI)(Display *, Window, float); - - /*** GLX_SGIX_video_resize ***/ - int (*BindChannelToWindowSGIX)(Display *, int, int, Window); - int (*ChannelRectSGIX)(Display *, int, int, int, int, int, int); - int (*QueryChannelRectSGIX)(Display *, int, int, int *, int *, int *, int *); - int (*QueryChannelDeltasSGIX)(Display *, int, int, int *, int *, int *, int *); - int (*ChannelRectSyncSGIX)(Display *, int, int, GLenum); - - /*** GLX_SGIX_dmbuffer (needs dmedia library) ***/ -#if defined (_DM_BUFFER_H_) - Bool (*AssociateDMPbufferSGIX)(Display *, GLXPbufferSGIX, DMparams *, DMbuffer); -#else - void *AssociciateDMPbufferSGIX; -#endif - - /*** GLX_SGIX_swap_group ***/ - void (*JoinSwapGroupSGIX)(Display *, GLXDrawable, GLXDrawable); - - /*** GLX_SGIX_swap_barrier ***/ - void (*BindSwapBarrierSGIX)(Display *, GLXDrawable, int); - Bool (*QueryMaxSwapBarriersSGIX)(Display *, int, int *); - - /*** GLX_SUN_get_transparent_index ***/ - Status (*GetTransparentIndexSUN)(Display *, Window, Window, long *); - - /*** GLX_MESA_copy_sub_buffer ***/ - void (*CopySubBufferMESA)(Display *dpy, GLXDrawable drawable, int x, int y, int width, int height); - - /*** GLX_MESA_release_buffers ***/ - Bool (*ReleaseBuffersMESA)(Display *dpy, Window w); - - /*** GLX_MESA_pixmap_colormap ***/ - GLXPixmap (*CreateGLXPixmapMESA)(Display *dpy, XVisualInfo *visinfo, Pixmap pixmap, Colormap cmap); - - /*** GLX_EXT_texture_from_pixmap ***/ - void (*BindTexImageEXT)(Display *dpy, GLXDrawable drawable, int buffer, - const int *attrib_list); - void (*ReleaseTexImageEXT)(Display *dpy, GLXDrawable drawable, int buffer); -}; - - - -extern const char * -_glxapi_get_version(void); - - - - -extern GLuint -_glxapi_get_dispatch_table_size(void); - - -extern void -_glxapi_set_no_op_table(struct _glxapi_table *t); - - -extern __GLXextFuncPtr -_glxapi_get_proc_address(const char *funcName); - - -#endif diff --git a/src/gallium/state_trackers/glx/xlib/xm_api.c b/src/gallium/state_trackers/glx/xlib/xm_api.c index 45ad694e234..957002ddd55 100644 --- a/src/gallium/state_trackers/glx/xlib/xm_api.c +++ b/src/gallium/state_trackers/glx/xlib/xm_api.c @@ -286,6 +286,18 @@ choose_pixel_format(XMesaVisual v) return PIPE_FORMAT_B8G8R8A8_UNORM; } } + else if ( GET_REDMASK(v) == 0x0000ff00 + && GET_GREENMASK(v) == 0x00ff0000 + && GET_BLUEMASK(v) == 0xff000000 + && v->BitsPerPixel == 32) { + if (native_byte_order) { + /* no byteswapping needed */ + return PIPE_FORMAT_B8G8R8A8_UNORM; + } + else { + return PIPE_FORMAT_A8R8G8B8_UNORM; + } + } else if ( GET_REDMASK(v) == 0xf800 && GET_GREENMASK(v) == 0x07e0 && GET_BLUEMASK(v) == 0x001f @@ -656,7 +668,7 @@ XMesaVisual XMesaCreateVisual( Display *display, * at a later time. */ v->visinfo = (XVisualInfo *) MALLOC(sizeof(*visinfo)); - if(!v->visinfo) { + if (!v->visinfo) { _mesa_free(v); return NULL; } @@ -742,7 +754,7 @@ PUBLIC XMesaContext XMesaCreateContext( XMesaVisual v, XMesaContext share_list ) { static GLboolean firstTime = GL_TRUE; - struct pipe_screen *screen; + static struct pipe_screen *screen = NULL; struct pipe_context *pipe = NULL; XMesaContext c; GLcontext *mesaCtx; @@ -750,6 +762,7 @@ XMesaContext XMesaCreateContext( XMesaVisual v, XMesaContext share_list ) if (firstTime) { pipe_mutex_init(_xmesa_lock); + screen = driver.create_pipe_screen(); firstTime = GL_FALSE; } @@ -765,14 +778,10 @@ XMesaContext XMesaCreateContext( XMesaVisual v, XMesaContext share_list ) c->xm_buffer = NULL; /* set later by XMesaMakeCurrent */ c->xm_read_buffer = NULL; - /* XXX: create once per Xlib Display. - */ - screen = driver.create_pipe_screen(); if (screen == NULL) goto fail; - pipe = driver.create_pipe_context( screen, - (void *)c ); + pipe = driver.create_pipe_context(screen, (void *) c); if (pipe == NULL) goto fail; @@ -785,26 +794,15 @@ XMesaContext XMesaCreateContext( XMesaVisual v, XMesaContext share_list ) mesaCtx = c->st->ctx; c->st->ctx->DriverCtx = c; -#if 00 - _mesa_enable_sw_extensions(mesaCtx); - _mesa_enable_1_3_extensions(mesaCtx); - _mesa_enable_1_4_extensions(mesaCtx); - _mesa_enable_1_5_extensions(mesaCtx); - _mesa_enable_2_0_extensions(mesaCtx); -#endif - return c; - fail: +fail: if (c->st) st_destroy_context(c->st); else if (pipe) pipe->destroy(pipe); - if (screen) - screen->destroy( screen ); - - FREE(c); + _mesa_free(c); return NULL; } @@ -1100,26 +1098,19 @@ XMesaContext XMesaGetCurrentContext( void ) - - - -/* - * Copy the back buffer to the front buffer. If there's no back buffer - * this is a no-op. +/** + * Swap front and back color buffers and have winsys display front buffer. + * If there's no front color buffer no swap actually occurs. */ PUBLIC void XMesaSwapBuffers( XMesaBuffer b ) { - struct pipe_surface *surf; + struct pipe_surface *frontLeftSurf; - /* If we're swapping the buffer associated with the current context - * we have to flush any pending rendering commands first. - */ - st_notify_swapbuffers(b->stfb); + st_swapbuffers(b->stfb, &frontLeftSurf, NULL); - st_get_framebuffer_surface(b->stfb, ST_SURFACE_BACK_LEFT, &surf); - if (surf) { - driver.display_surface(b, surf); + if (frontLeftSurf) { + driver.display_surface(b, frontLeftSurf); } xmesa_check_and_update_buffer_size(NULL, b); @@ -1165,7 +1156,7 @@ void XMesaFlush( XMesaContext c ) XMesaBuffer XMesaFindBuffer( Display *dpy, Drawable d ) { XMesaBuffer b; - for (b=XMesaBufferList; b; b=b->Next) { + for (b = XMesaBufferList; b; b = b->Next) { if (b->drawable == d && b->xm_visual->display == dpy) { return b; } diff --git a/src/gallium/state_trackers/glx/xlib/xm_api.h b/src/gallium/state_trackers/glx/xlib/xm_api.h index bdd434cd364..ce97a3ec768 100644 --- a/src/gallium/state_trackers/glx/xlib/xm_api.h +++ b/src/gallium/state_trackers/glx/xlib/xm_api.h @@ -57,7 +57,7 @@ and create a window, you must do the following to use the X/Mesa interface: #define XMESA_H -#include "mtypes.h" +#include "main/mtypes.h" #include "state_tracker/st_context.h" #include "state_tracker/st_public.h" #include "pipe/p_thread.h" diff --git a/src/gallium/state_trackers/vega/Makefile b/src/gallium/state_trackers/vega/Makefile new file mode 100644 index 00000000000..b8c805b06c4 --- /dev/null +++ b/src/gallium/state_trackers/vega/Makefile @@ -0,0 +1,128 @@ +# src/mesa/Makefile + +TOP = ../../../.. +include $(TOP)/configs/current +GALLIUM = $(TOP) + +### Lists of source files, included by Makefiles + +VG_SOURCES = \ + api_context.c \ + api_filters.c \ + api_images.c \ + api_masks.c \ + api_misc.c \ + api_paint.c \ + api_params.c \ + api_path.c \ + api_text.c \ + api_transform.c \ + vgu.c \ + vg_context.c \ + vg_state.c \ + vg_tracker.c \ + vg_translate.c \ + polygon.c \ + bezier.c \ + path.c \ + paint.c \ + arc.c \ + image.c \ + renderer.c \ + stroker.c \ + mask.c \ + shader.c \ + shaders_cache.c + + +### All the core C sources + +ALL_SOURCES = \ + $(VG_SOURCES) + + +### Object files +VG_OBJECTS = \ + $(VG_SOURCES:.c=.o) + +### Include directories + +INCLUDE_DIRS = \ + -I$(TOP)/include \ + -I$(GALLIUM)/include \ + -I$(GALLIUM)/src/gallium/include \ + -I$(GALLIUM)/src/gallium/auxiliary + +VG_LIB = OpenVG +VG_LIB_NAME = lib$(VG_LIB).so + +VG_MAJOR = 1 +VG_MINOR = 0 +VG_TINY = 0 + +GALLIUM_LIBS = \ + $(GALLIUM)/src/gallium/auxiliary/pipebuffer/libpipebuffer.a \ + $(GALLIUM)/src/gallium/auxiliary/sct/libsct.a \ + $(GALLIUM)/src/gallium/auxiliary/draw/libdraw.a \ + $(GALLIUM)/src/gallium/auxiliary/rtasm/librtasm.a \ + $(GALLIUM)/src/gallium/auxiliary/translate/libtranslate.a \ + $(GALLIUM)/src/gallium/auxiliary/cso_cache/libcso_cache.a \ + $(GALLIUM)/src/gallium/auxiliary/util/libutil.a \ + $(GALLIUM)/src/gallium/auxiliary/tgsi/libtgsi.a + +.SUFFIXES : .cpp + +.c.o: + $(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $< -o $@ + +.cpp.o: + $(CXX) -c $(INCLUDE_DIRS) $(CXXFLAGS) $< -o $@ + +.S.o: + $(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $< -o $@ + + +default: depend subdirs $(TOP)/$(LIB_DIR)/$(VG_LIB_NAME) + +# Make the OpenVG library +$(TOP)/$(LIB_DIR)/$(VG_LIB_NAME): $(VG_OBJECTS) $(GALLIUM_LIBS) + $(TOP)/bin/mklib -o $(VG_LIB) \ + -major $(VG_MAJOR) \ + -minor $(VG_MINOR) \ + -patch $(VG_TINY) \ + -install $(TOP)/$(LIB_DIR) \ + $(VG_OBJECTS) $(GALLIUM_LIBS) \ + -Wl,--whole-archive $(LIBS) -Wl,--no-whole-archive $(SYS_LIBS) + +###################################################################### +# Generic stuff + +depend: $(ALL_SOURCES) + @ echo "running $(MKDEP)" + @ rm -f depend # workaround oops on gutsy?!? + @ touch depend + @ $(MKDEP) $(MKDEP_OPTIONS) $(DEFINES) $(INCLUDE_DIRS) $(ALL_SOURCES) \ + > /dev/null 2>/dev/null + + +subdirs: + +install: default + $(INSTALL) -d $(INSTALL_DIR)/include/VG + $(INSTALL) -d $(INSTALL_DIR)/$(LIB_DIR) + $(INSTALL) -m 644 $(TOP)/include/VG/*.h $(INSTALL_DIR)/include/VG + @if [ -e $(TOP)/$(LIB_DIR)/$(VG_LIB_NAME) ]; then \ + $(INSTALL) $(TOP)/$(LIB_DIR)/libOpenVG* $(INSTALL_DIR)/$(LIB_DIR); \ + fi + +# Emacs tags +tags: + etags `find . -name \*.[ch]` $(TOP)/include/VG/*.h + +clean: + -rm -f *.o + -rm -f */*.o + -rm -f */*/*.o + -rm -f depend depend.bak + +include depend diff --git a/src/gallium/state_trackers/vega/api_consts.h b/src/gallium/state_trackers/vega/api_consts.h new file mode 100644 index 00000000000..e1b48d4a46c --- /dev/null +++ b/src/gallium/state_trackers/vega/api_consts.h @@ -0,0 +1,56 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef API_CONSTS_H +#define API_CONSTS_H + +/*must be at least 32*/ +#define VEGA_MAX_SCISSOR_RECTS 32 + +/*must be at least 16*/ +#define VEGA_MAX_DASH_COUNT 32 + +/*must be at least 7*/ +#define VEGA_MAX_KERNEL_SIZE 7 + +/*must be at least 15*/ +#define VEGA_MAX_SEPARABLE_KERNEL_SIZE 15 + +/*must be at least 32*/ +#define VEGA_MAX_COLOR_RAMP_STOPS 256 + +#define VEGA_MAX_IMAGE_WIDTH 2048 + +#define VEGA_MAX_IMAGE_HEIGHT 2048 + +#define VEGA_MAX_IMAGE_PIXELS (2048*2048) + +#define VEGA_MAX_IMAGE_BYTES (2048*2048 * 4) + +/*must be at least 128*/ +#define VEGA_MAX_GAUSSIAN_STD_DEVIATION 128 + +#endif diff --git a/src/gallium/state_trackers/vega/api_context.c b/src/gallium/state_trackers/vega/api_context.c new file mode 100644 index 00000000000..47db102dd2d --- /dev/null +++ b/src/gallium/state_trackers/vega/api_context.c @@ -0,0 +1,75 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "VG/openvg.h" + +#include "vg_context.h" + +#include "pipe/p_context.h" +#include "pipe/p_screen.h" + +VGErrorCode vgGetError(void) +{ + struct vg_context *ctx = vg_current_context(); + VGErrorCode error = VG_NO_CONTEXT_ERROR; + + if (!ctx) + return error; + + error = ctx->_error; + ctx->_error = VG_NO_ERROR; + + return error; +} + +void vgFlush(void) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_context *pipe; + + if (!ctx) + return; + + pipe = ctx->pipe; + pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); +} + +void vgFinish(void) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_fence_handle *fence = NULL; + struct pipe_context *pipe; + + if (!ctx) + return; + + pipe = ctx->pipe; + + pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME, &fence); + + pipe->screen->fence_finish(pipe->screen, fence, 0); + pipe->screen->fence_reference(pipe->screen, &fence, NULL); +} diff --git a/src/gallium/state_trackers/vega/api_filters.c b/src/gallium/state_trackers/vega/api_filters.c new file mode 100644 index 00000000000..862cbb03c43 --- /dev/null +++ b/src/gallium/state_trackers/vega/api_filters.c @@ -0,0 +1,805 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "VG/openvg.h" + +#include "vg_context.h" +#include "image.h" +#include "renderer.h" +#include "shaders_cache.h" +#include "st_inlines.h" + +#include "pipe/p_context.h" +#include "pipe/p_state.h" +#include "pipe/p_inlines.h" +#include "pipe/p_screen.h" +#include "pipe/p_shader_tokens.h" + +#include "util/u_memory.h" + + +#include "asm_filters.h" + + +struct filter_info { + struct vg_image *dst; + struct vg_image *src; + struct vg_shader * (*setup_shader)(struct vg_context *, void *); + void *user_data; + const void *const_buffer; + VGint const_buffer_len; + VGTilingMode tiling_mode; + struct pipe_texture *extra_texture; +}; + +static INLINE struct pipe_texture *create_texture_1d(struct vg_context *ctx, + const VGuint *color_data, + const VGint color_data_len) +{ + struct pipe_context *pipe = ctx->pipe; + struct pipe_screen *screen = pipe->screen; + struct pipe_texture *tex = 0; + struct pipe_texture templ; + + memset(&templ, 0, sizeof(templ)); + templ.target = PIPE_TEXTURE_1D; + templ.format = PIPE_FORMAT_A8R8G8B8_UNORM; + templ.last_level = 0; + templ.width[0] = color_data_len; + templ.height[0] = 1; + templ.depth[0] = 1; + pf_get_block(PIPE_FORMAT_A8R8G8B8_UNORM, &templ.block); + templ.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER; + + tex = screen->texture_create(screen, &templ); + + { /* upload color_data */ + struct pipe_transfer *transfer = + screen->get_tex_transfer(screen, tex, + 0, 0, 0, + PIPE_TRANSFER_READ_WRITE , + 0, 0, tex->width[0], tex->height[0]); + void *map = screen->transfer_map(screen, transfer); + memcpy(map, color_data, sizeof(VGint)*color_data_len); + screen->transfer_unmap(screen, transfer); + screen->tex_transfer_destroy(transfer); + } + + return tex; +} + +static INLINE struct pipe_surface * setup_framebuffer(struct vg_image *dst) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_context *pipe = ctx->pipe; + struct pipe_framebuffer_state fb; + struct pipe_surface *dst_surf = pipe->screen->get_tex_surface( + pipe->screen, dst->texture, 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_WRITE); + + /* drawing dest */ + memset(&fb, 0, sizeof(fb)); + fb.width = dst->x + dst_surf->width; + fb.height = dst->y + dst_surf->height; + fb.nr_cbufs = 1; + fb.cbufs[0] = dst_surf; + { + VGint i; + for (i = 1; i < PIPE_MAX_COLOR_BUFS; ++i) + fb.cbufs[i] = 0; + } + cso_set_framebuffer(ctx->cso_context, &fb); + + return dst_surf; +} + +static void setup_viewport(struct vg_image *dst) +{ + struct vg_context *ctx = vg_current_context(); + vg_set_viewport(ctx, VEGA_Y0_TOP); +} + +static void setup_blend() +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_blend_state blend; + memset(&blend, 0, sizeof(blend)); + blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE; + blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE; + blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO; + blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO; + if (ctx->state.vg.filter_channel_mask & VG_RED) + blend.colormask |= PIPE_MASK_R; + if (ctx->state.vg.filter_channel_mask & VG_GREEN) + blend.colormask |= PIPE_MASK_G; + if (ctx->state.vg.filter_channel_mask & VG_BLUE) + blend.colormask |= PIPE_MASK_B; + if (ctx->state.vg.filter_channel_mask & VG_ALPHA) + blend.colormask |= PIPE_MASK_A; + blend.blend_enable = 1; + cso_set_blend(ctx->cso_context, &blend); +} + +static void setup_constant_buffer(struct vg_context *ctx, const void *buffer, + VGint param_bytes) +{ + struct pipe_context *pipe = ctx->pipe; + struct pipe_constant_buffer *cbuf = &ctx->filter.buffer; + + /* We always need to get a new buffer, to keep the drivers simple and + * avoid gratuitous rendering synchronization. */ + pipe_buffer_reference(&cbuf->buffer, NULL); + + cbuf->buffer = pipe_buffer_create(pipe->screen, 16, + PIPE_BUFFER_USAGE_CONSTANT, + param_bytes); + + if (cbuf->buffer) { + st_no_flush_pipe_buffer_write(ctx, cbuf->buffer, + 0, param_bytes, buffer); + } + + ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, cbuf); +} + +static void setup_samplers(struct vg_context *ctx, struct filter_info *info) +{ + struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS]; + struct pipe_texture *textures[PIPE_MAX_SAMPLERS]; + struct pipe_sampler_state sampler[3]; + int num_samplers = 0; + int num_textures = 0; + + samplers[0] = NULL; + samplers[1] = NULL; + samplers[2] = NULL; + samplers[3] = NULL; + textures[0] = NULL; + textures[1] = NULL; + textures[2] = NULL; + textures[3] = NULL; + + memset(&sampler[0], 0, sizeof(struct pipe_sampler_state)); + sampler[0].wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + sampler[0].wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + sampler[0].wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + sampler[0].min_img_filter = PIPE_TEX_MIPFILTER_LINEAR; + sampler[0].mag_img_filter = PIPE_TEX_MIPFILTER_LINEAR; + sampler[0].normalized_coords = 1; + + switch(info->tiling_mode) { + case VG_TILE_FILL: + sampler[0].wrap_s = PIPE_TEX_WRAP_CLAMP_TO_BORDER; + sampler[0].wrap_t = PIPE_TEX_WRAP_CLAMP_TO_BORDER; + memcpy(sampler[0].border_color, + ctx->state.vg.tile_fill_color, + sizeof(VGfloat) * 4); + break; + case VG_TILE_PAD: + sampler[0].wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + sampler[0].wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + break; + case VG_TILE_REPEAT: + sampler[0].wrap_s = PIPE_TEX_WRAP_REPEAT; + sampler[0].wrap_t = PIPE_TEX_WRAP_REPEAT; + break; + case VG_TILE_REFLECT: + sampler[0].wrap_s = PIPE_TEX_WRAP_MIRROR_REPEAT; + sampler[0].wrap_t = PIPE_TEX_WRAP_MIRROR_REPEAT; + break; + default: + debug_assert(!"Unknown tiling mode"); + } + + samplers[0] = &sampler[0]; + textures[0] = info->src->texture; + ++num_samplers; + ++num_textures; + + if (info->extra_texture) { + memcpy(&sampler[1], &sampler[0], sizeof(struct pipe_sampler_state)); + samplers[1] = &sampler[1]; + textures[1] = info->extra_texture; + ++num_samplers; + ++num_textures; + } + + + cso_set_samplers(ctx->cso_context, num_samplers, (const struct pipe_sampler_state **)samplers); + cso_set_sampler_textures(ctx->cso_context, num_textures, textures); +} + +static struct vg_shader * setup_color_matrix(struct vg_context *ctx, void *user_data) +{ + struct vg_shader *shader = + shader_create_from_text(ctx->pipe, color_matrix_asm, 200, + PIPE_SHADER_FRAGMENT); + cso_set_fragment_shader_handle(ctx->cso_context, shader->driver); + return shader; +} + +static struct vg_shader * setup_convolution(struct vg_context *ctx, void *user_data) +{ + char buffer[1024]; + VGint num_consts = (VGint)(long)(user_data); + struct vg_shader *shader; + + snprintf(buffer, 1023, convolution_asm, num_consts, num_consts / 2 + 1); + + shader = shader_create_from_text(ctx->pipe, buffer, 200, + PIPE_SHADER_FRAGMENT); + + cso_set_fragment_shader_handle(ctx->cso_context, shader->driver); + return shader; +} + +static struct vg_shader * setup_lookup(struct vg_context *ctx, void *user_data) +{ + struct vg_shader *shader = + shader_create_from_text(ctx->pipe, lookup_asm, + 200, PIPE_SHADER_FRAGMENT); + + cso_set_fragment_shader_handle(ctx->cso_context, shader->driver); + return shader; +} + + +static struct vg_shader * setup_lookup_single(struct vg_context *ctx, void *user_data) +{ + char buffer[1024]; + VGImageChannel channel = (VGImageChannel)(user_data); + struct vg_shader *shader; + + switch(channel) { + case VG_RED: + snprintf(buffer, 1023, lookup_single_asm, "xxxx"); + break; + case VG_GREEN: + snprintf(buffer, 1023, lookup_single_asm, "yyyy"); + break; + case VG_BLUE: + snprintf(buffer, 1023, lookup_single_asm, "zzzz"); + break; + case VG_ALPHA: + snprintf(buffer, 1023, lookup_single_asm, "wwww"); + break; + default: + debug_assert(!"Unknown color channel"); + } + + shader = shader_create_from_text(ctx->pipe, buffer, 200, + PIPE_SHADER_FRAGMENT); + + cso_set_fragment_shader_handle(ctx->cso_context, shader->driver); + return shader; +} + +static void execute_filter(struct vg_context *ctx, + struct filter_info *info) +{ + struct pipe_surface *dst_surf; + struct vg_shader *shader; + + cso_save_framebuffer(ctx->cso_context); + cso_save_fragment_shader(ctx->cso_context); + cso_save_viewport(ctx->cso_context); + cso_save_blend(ctx->cso_context); + cso_save_samplers(ctx->cso_context); + cso_save_sampler_textures(ctx->cso_context); + + dst_surf = setup_framebuffer(info->dst); + setup_viewport(info->dst); + setup_blend(); + setup_constant_buffer(ctx, info->const_buffer, info->const_buffer_len); + shader = info->setup_shader(ctx, info->user_data); + setup_samplers(ctx, info); + + renderer_draw_texture(ctx->renderer, + info->src->texture, + info->dst->x, info->dst->y, + info->dst->x + info->dst->width, + info->dst->y + info->dst->height, + info->dst->x, info->dst->y, + info->dst->x + info->dst->width, + info->dst->y + info->dst->height); + + cso_restore_framebuffer(ctx->cso_context); + cso_restore_fragment_shader(ctx->cso_context); + cso_restore_viewport(ctx->cso_context); + cso_restore_blend(ctx->cso_context); + cso_restore_samplers(ctx->cso_context); + cso_restore_sampler_textures(ctx->cso_context); + + vg_shader_destroy(ctx, shader); + + pipe_surface_reference(&dst_surf, NULL); +} + +void vgColorMatrix(VGImage dst, VGImage src, + const VGfloat * matrix) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *d, *s; + struct filter_info info; + + if (dst == VG_INVALID_HANDLE || src == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (!matrix || !is_aligned(matrix)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + d = (struct vg_image*)dst; + s = (struct vg_image*)src; + + if (vg_image_overlaps(d, s)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + info.dst = d; + info.src = s; + info.setup_shader = &setup_color_matrix; + info.user_data = NULL; + info.const_buffer = matrix; + info.const_buffer_len = 20 * sizeof(VGfloat); + info.tiling_mode = VG_TILE_PAD; + info.extra_texture = 0; + execute_filter(ctx, &info); +} + +static VGfloat texture_offset(VGfloat width, VGint kernelSize, VGint current, VGint shift) +{ + VGfloat diff = current - shift; + + return diff / width; +} + +void vgConvolve(VGImage dst, VGImage src, + VGint kernelWidth, VGint kernelHeight, + VGint shiftX, VGint shiftY, + const VGshort * kernel, + VGfloat scale, + VGfloat bias, + VGTilingMode tilingMode) +{ + struct vg_context *ctx = vg_current_context(); + VGfloat *buffer; + VGint buffer_len; + VGint i, j; + VGint idx = 0; + struct vg_image *d, *s; + VGint kernel_size = kernelWidth * kernelHeight; + struct filter_info info; + const VGint max_kernel_size = vgGeti(VG_MAX_KERNEL_SIZE); + + if (dst == VG_INVALID_HANDLE || src == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (kernelWidth <= 0 || kernelHeight <= 0 || + kernelWidth > max_kernel_size || kernelHeight > max_kernel_size) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (!kernel || !is_aligned_to(kernel, 2)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (tilingMode < VG_TILE_FILL || + tilingMode > VG_TILE_REFLECT) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + d = (struct vg_image*)dst; + s = (struct vg_image*)src; + + if (vg_image_overlaps(d, s)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + vg_validate_state(ctx); + + buffer_len = 8 + 2 * 4 * kernel_size; + buffer = (VGfloat*)malloc(buffer_len * sizeof(VGfloat)); + + buffer[0] = 0.f; + buffer[1] = 1.f; + buffer[2] = 2.f; /*unused*/ + buffer[3] = 4.f; /*unused*/ + + buffer[4] = kernelWidth * kernelHeight; + buffer[5] = scale; + buffer[6] = bias; + buffer[7] = 0.f; + + idx = 8; + for (j = 0; j < kernelHeight; ++j) { + for (i = 0; i < kernelWidth; ++i) { + VGint index = j * kernelWidth + i; + VGfloat x, y; + + x = texture_offset(s->width, kernelWidth, i, shiftX); + y = texture_offset(s->height, kernelHeight, j, shiftY); + + buffer[idx + index*4 + 0] = x; + buffer[idx + index*4 + 1] = y; + buffer[idx + index*4 + 2] = 0.f; + buffer[idx + index*4 + 3] = 0.f; + } + } + idx += kernel_size * 4; + + for (j = 0; j < kernelHeight; ++j) { + for (i = 0; i < kernelWidth; ++i) { + /* transpose the kernel */ + VGint index = j * kernelWidth + i; + VGint kindex = (kernelWidth - i - 1) * kernelHeight + (kernelHeight - j - 1); + buffer[idx + index*4 + 0] = kernel[kindex]; + buffer[idx + index*4 + 1] = kernel[kindex]; + buffer[idx + index*4 + 2] = kernel[kindex]; + buffer[idx + index*4 + 3] = kernel[kindex]; + } + } + + info.dst = d; + info.src = s; + info.setup_shader = &setup_convolution; + info.user_data = (void*)(long)(buffer_len/4); + info.const_buffer = buffer; + info.const_buffer_len = buffer_len * sizeof(VGfloat); + info.tiling_mode = tilingMode; + info.extra_texture = 0; + execute_filter(ctx, &info); + + free(buffer); +} + +void vgSeparableConvolve(VGImage dst, VGImage src, + VGint kernelWidth, + VGint kernelHeight, + VGint shiftX, VGint shiftY, + const VGshort * kernelX, + const VGshort * kernelY, + VGfloat scale, + VGfloat bias, + VGTilingMode tilingMode) +{ + struct vg_context *ctx = vg_current_context(); + VGshort *kernel; + VGint i, j, idx = 0; + const VGint max_kernel_size = vgGeti(VG_MAX_SEPARABLE_KERNEL_SIZE); + + if (dst == VG_INVALID_HANDLE || src == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (kernelWidth <= 0 || kernelHeight <= 0 || + kernelWidth > max_kernel_size || kernelHeight > max_kernel_size) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (!kernelX || !kernelY || + !is_aligned_to(kernelX, 2) || !is_aligned_to(kernelY, 2)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (tilingMode < VG_TILE_FILL || + tilingMode > VG_TILE_REFLECT) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + kernel = malloc(sizeof(VGshort)*kernelWidth*kernelHeight); + for (i = 0; i < kernelWidth; ++i) { + for (j = 0; j < kernelHeight; ++j) { + kernel[idx] = kernelX[i] * kernelY[j]; + ++idx; + } + } + vgConvolve(dst, src, kernelWidth, kernelHeight, shiftX, shiftY, + kernel, scale, bias, tilingMode); + free(kernel); +} + +static INLINE VGfloat compute_gaussian_componenet(VGfloat x, VGfloat y, + VGfloat stdDeviationX, + VGfloat stdDeviationY) +{ + VGfloat mult = 1 / ( 2 * M_PI * stdDeviationX * stdDeviationY); + VGfloat e = exp( - ( pow(x, 2)/(2*pow(stdDeviationX, 2)) + + pow(y, 2)/(2*pow(stdDeviationY, 2)) ) ); + return mult * e; +} + +static INLINE VGint compute_kernel_size(VGfloat deviation) +{ + VGint size = ceil(2.146 * deviation); + if (size > 11) + return 11; + return size; +} + +static void compute_gaussian_kernel(VGfloat *kernel, + VGint width, VGint height, + VGfloat stdDeviationX, + VGfloat stdDeviationY) +{ + VGint i, j; + VGfloat scale = 0.0f; + + for (j = 0; j < height; ++j) { + for (i = 0; i < width; ++i) { + VGint idx = (height - j -1) * width + (width - i -1); + kernel[idx] = compute_gaussian_componenet(i-(ceil(width/2))-1, + j-ceil(height/2)-1, + stdDeviationX, stdDeviationY); + scale += kernel[idx]; + } + } + + for (j = 0; j < height; ++j) { + for (i = 0; i < width; ++i) { + VGint idx = j * width + i; + kernel[idx] /= scale; + } + } +} + +void vgGaussianBlur(VGImage dst, VGImage src, + VGfloat stdDeviationX, + VGfloat stdDeviationY, + VGTilingMode tilingMode) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *d, *s; + VGfloat *buffer, *kernel; + VGint kernel_width, kernel_height, kernel_size; + VGint buffer_len; + VGint idx, i, j; + struct filter_info info; + + if (dst == VG_INVALID_HANDLE || src == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (stdDeviationX <= 0 || stdDeviationY <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (tilingMode < VG_TILE_FILL || + tilingMode > VG_TILE_REFLECT) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + d = (struct vg_image*)dst; + s = (struct vg_image*)src; + + if (vg_image_overlaps(d, s)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + kernel_width = compute_kernel_size(stdDeviationX); + kernel_height = compute_kernel_size(stdDeviationY); + kernel_size = kernel_width * kernel_height; + kernel = malloc(sizeof(VGfloat)*kernel_size); + compute_gaussian_kernel(kernel, kernel_width, kernel_height, + stdDeviationX, stdDeviationY); + + buffer_len = 8 + 2 * 4 * kernel_size; + buffer = (VGfloat*)malloc(buffer_len * sizeof(VGfloat)); + + buffer[0] = 0.f; + buffer[1] = 1.f; + buffer[2] = 2.f; /*unused*/ + buffer[3] = 4.f; /*unused*/ + + buffer[4] = kernel_width * kernel_height; + buffer[5] = 1.f;/*scale*/ + buffer[6] = 0.f;/*bias*/ + buffer[7] = 0.f; + + idx = 8; + for (j = 0; j < kernel_height; ++j) { + for (i = 0; i < kernel_width; ++i) { + VGint index = j * kernel_width + i; + VGfloat x, y; + + x = texture_offset(s->width, kernel_width, i, kernel_width/2); + y = texture_offset(s->height, kernel_height, j, kernel_height/2); + + buffer[idx + index*4 + 0] = x; + buffer[idx + index*4 + 1] = y; + buffer[idx + index*4 + 2] = 0.f; + buffer[idx + index*4 + 3] = 0.f; + } + } + idx += kernel_size * 4; + + for (j = 0; j < kernel_height; ++j) { + for (i = 0; i < kernel_width; ++i) { + /* transpose the kernel */ + VGint index = j * kernel_width + i; + VGint kindex = (kernel_width - i - 1) * kernel_height + (kernel_height - j - 1); + buffer[idx + index*4 + 0] = kernel[kindex]; + buffer[idx + index*4 + 1] = kernel[kindex]; + buffer[idx + index*4 + 2] = kernel[kindex]; + buffer[idx + index*4 + 3] = kernel[kindex]; + } + } + + info.dst = d; + info.src = s; + info.setup_shader = &setup_convolution; + info.user_data = (void*)(long)(buffer_len/4); + info.const_buffer = buffer; + info.const_buffer_len = buffer_len * sizeof(VGfloat); + info.tiling_mode = tilingMode; + info.extra_texture = 0; + execute_filter(ctx, &info); + + free(buffer); + free(kernel); +} + +void vgLookup(VGImage dst, VGImage src, + const VGubyte * redLUT, + const VGubyte * greenLUT, + const VGubyte * blueLUT, + const VGubyte * alphaLUT, + VGboolean outputLinear, + VGboolean outputPremultiplied) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *d, *s; + VGuint color_data[256]; + VGint i; + struct pipe_texture *lut_texture; + VGfloat buffer[4]; + struct filter_info info; + + if (dst == VG_INVALID_HANDLE || src == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (!redLUT || !greenLUT || !blueLUT || !alphaLUT) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + d = (struct vg_image*)dst; + s = (struct vg_image*)src; + + if (vg_image_overlaps(d, s)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + for (i = 0; i < 256; ++i) { + color_data[i] = blueLUT[i] << 24 | greenLUT[i] << 16 | + redLUT[i] << 8 | alphaLUT[i]; + } + lut_texture = create_texture_1d(ctx, color_data, 255); + + buffer[0] = 0.f; + buffer[1] = 0.f; + buffer[2] = 1.f; + buffer[3] = 1.f; + + info.dst = d; + info.src = s; + info.setup_shader = &setup_lookup; + info.user_data = NULL; + info.const_buffer = buffer; + info.const_buffer_len = 4 * sizeof(VGfloat); + info.tiling_mode = VG_TILE_PAD; + info.extra_texture = lut_texture; + + execute_filter(ctx, &info); + + pipe_texture_reference(&lut_texture, NULL); +} + +void vgLookupSingle(VGImage dst, VGImage src, + const VGuint * lookupTable, + VGImageChannel sourceChannel, + VGboolean outputLinear, + VGboolean outputPremultiplied) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *d, *s; + struct pipe_texture *lut_texture; + VGfloat buffer[4]; + struct filter_info info; + VGuint color_data[256]; + VGint i; + + if (dst == VG_INVALID_HANDLE || src == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (!lookupTable || !is_aligned(lookupTable)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (sourceChannel != VG_RED && sourceChannel != VG_GREEN && + sourceChannel != VG_BLUE && sourceChannel != VG_ALPHA) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + d = (struct vg_image*)dst; + s = (struct vg_image*)src; + + if (vg_image_overlaps(d, s)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + for (i = 0; i < 256; ++i) { + VGuint rgba = lookupTable[i]; + VGubyte blue, green, red, alpha; + red = (rgba & 0xff000000)>>24; + green = (rgba & 0x00ff0000)>>16; + blue = (rgba & 0x0000ff00)>> 8; + alpha = (rgba & 0x000000ff)>> 0; + color_data[i] = blue << 24 | green << 16 | + red << 8 | alpha; + } + lut_texture = create_texture_1d(ctx, color_data, 256); + + buffer[0] = 0.f; + buffer[1] = 0.f; + buffer[2] = 1.f; + buffer[3] = 1.f; + + info.dst = d; + info.src = s; + info.setup_shader = &setup_lookup_single; + info.user_data = (void*)sourceChannel; + info.const_buffer = buffer; + info.const_buffer_len = 4 * sizeof(VGfloat); + info.tiling_mode = VG_TILE_PAD; + info.extra_texture = lut_texture; + + execute_filter(ctx, &info); + + pipe_texture_reference(&lut_texture, NULL); +} diff --git a/src/gallium/state_trackers/vega/api_images.c b/src/gallium/state_trackers/vega/api_images.c new file mode 100644 index 00000000000..c437553bc23 --- /dev/null +++ b/src/gallium/state_trackers/vega/api_images.c @@ -0,0 +1,489 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "image.h" + +#include "VG/openvg.h" + +#include "vg_context.h" +#include "vg_translate.h" +#include "api_consts.h" +#include "image.h" + +#include "pipe/p_context.h" +#include "pipe/p_screen.h" +#include "pipe/p_inlines.h" +#include "util/u_blit.h" +#include "util/u_tile.h" +#include "util/u_memory.h" + +static INLINE VGboolean supported_image_format(VGImageFormat format) +{ + switch(format) { + case VG_sRGBX_8888: + case VG_sRGBA_8888: + case VG_sRGBA_8888_PRE: + case VG_sRGB_565: + case VG_sRGBA_5551: + case VG_sRGBA_4444: + case VG_sL_8: + case VG_lRGBX_8888: + case VG_lRGBA_8888: + case VG_lRGBA_8888_PRE: + case VG_lL_8: + case VG_A_8: + case VG_BW_1: +#ifdef OPENVG_VERSION_1_1 + case VG_A_1: + case VG_A_4: +#endif + case VG_sXRGB_8888: + case VG_sARGB_8888: + case VG_sARGB_8888_PRE: + case VG_sARGB_1555: + case VG_sARGB_4444: + case VG_lXRGB_8888: + case VG_lARGB_8888: + case VG_lARGB_8888_PRE: + case VG_sBGRX_8888: + case VG_sBGRA_8888: + case VG_sBGRA_8888_PRE: + case VG_sBGR_565: + case VG_sBGRA_5551: + case VG_sBGRA_4444: + case VG_lBGRX_8888: + case VG_lBGRA_8888: + case VG_lBGRA_8888_PRE: + case VG_sXBGR_8888: + case VG_sABGR_8888: + case VG_sABGR_8888_PRE: + case VG_sABGR_1555: + case VG_sABGR_4444: + case VG_lXBGR_8888: + case VG_lABGR_8888: + case VG_lABGR_8888_PRE: + return VG_TRUE; + default: + return VG_FALSE; + } + return VG_FALSE; +} + +VGImage vgCreateImage(VGImageFormat format, + VGint width, VGint height, + VGbitfield allowedQuality) +{ + struct vg_context *ctx = vg_current_context(); + + if (!supported_image_format(format)) { + vg_set_error(ctx, VG_UNSUPPORTED_IMAGE_FORMAT_ERROR); + return VG_INVALID_HANDLE; + } + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_INVALID_HANDLE; + } + if (width > vgGeti(VG_MAX_IMAGE_WIDTH) || + height > vgGeti(VG_MAX_IMAGE_HEIGHT)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_INVALID_HANDLE; + } + if (width * height > vgGeti(VG_MAX_IMAGE_PIXELS)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_INVALID_HANDLE; + } + + if (!(allowedQuality & ((VG_IMAGE_QUALITY_NONANTIALIASED | + VG_IMAGE_QUALITY_FASTER | + VG_IMAGE_QUALITY_BETTER)))) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_INVALID_HANDLE; + } + + return (VGImage)image_create(format, width, height); +} + +void vgDestroyImage(VGImage image) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *img = (struct vg_image *)image; + + if (image == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (!vg_object_is_valid((void*)image, VG_OBJECT_IMAGE)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + image_destroy(img); +} + +void vgClearImage(VGImage image, + VGint x, VGint y, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *img; + + if (image == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + img = (struct vg_image*)image; + + if (x + width < 0 || y + height < 0) + return; + + image_clear(img, x, y, width, height); + +} + +void vgImageSubData(VGImage image, + const void * data, + VGint dataStride, + VGImageFormat dataFormat, + VGint x, VGint y, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *img; + + if (image == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (!supported_image_format(dataFormat)) { + vg_set_error(ctx, VG_UNSUPPORTED_IMAGE_FORMAT_ERROR); + return; + } + if (width <= 0 || height <= 0 || !data || !is_aligned(data)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + img = (struct vg_image*)(image); + image_sub_data(img, data, dataStride, dataFormat, + x, y, width, height); +} + +void vgGetImageSubData(VGImage image, + void * data, + VGint dataStride, + VGImageFormat dataFormat, + VGint x, VGint y, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *img; + + if (image == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (!supported_image_format(dataFormat)) { + vg_set_error(ctx, VG_UNSUPPORTED_IMAGE_FORMAT_ERROR); + return; + } + if (width <= 0 || height <= 0 || !data || !is_aligned(data)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + img = (struct vg_image*)image; + image_get_sub_data(img, data, dataStride, dataFormat, + x, y, width, height); +} + +VGImage vgChildImage(VGImage parent, + VGint x, VGint y, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *p; + + if (parent == VG_INVALID_HANDLE || + !vg_context_is_object_valid(ctx, VG_OBJECT_IMAGE, (void*)parent) || + !vg_object_is_valid((void*)parent, VG_OBJECT_IMAGE)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return VG_INVALID_HANDLE; + } + if (width <= 0 || height <= 0 || x < 0 || y < 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_INVALID_HANDLE; + } + p = (struct vg_image *)parent; + if (x > p->width || y > p->height) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_INVALID_HANDLE; + } + if (x + width > p->width || y + height > p->height) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_INVALID_HANDLE; + } + + return (VGImage)image_child_image(p, x, y, width, height); +} + +VGImage vgGetParent(VGImage image) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *img; + + if (image == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return VG_INVALID_HANDLE; + } + + img = (struct vg_image*)image; + if (img->parent) + return (VGImage)img->parent; + else + return image; +} + +void vgCopyImage(VGImage dst, VGint dx, VGint dy, + VGImage src, VGint sx, VGint sy, + VGint width, VGint height, + VGboolean dither) +{ + struct vg_context *ctx = vg_current_context(); + + if (src == VG_INVALID_HANDLE || dst == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + vg_validate_state(ctx); + image_copy((struct vg_image*)dst, dx, dy, + (struct vg_image*)src, sx, sy, + width, height, dither); +} + +void vgDrawImage(VGImage image) +{ + struct vg_context *ctx = vg_current_context(); + + if (!ctx) + return; + + if (image == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + vg_validate_state(ctx); + image_draw((struct vg_image*)image); +} + +void vgSetPixels(VGint dx, VGint dy, + VGImage src, VGint sx, VGint sy, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + + vg_validate_state(ctx); + + if (src == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + image_set_pixels(dx, dy, (struct vg_image*)src, sx, sy, width, + height); +} + +void vgGetPixels(VGImage dst, VGint dx, VGint dy, + VGint sx, VGint sy, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *img; + + if (dst == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + img = (struct vg_image*)dst; + + image_get_pixels(img, dx, dy, + sx, sy, width, height); +} + +void vgWritePixels(const void * data, VGint dataStride, + VGImageFormat dataFormat, + VGint dx, VGint dy, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_context *pipe = ctx->pipe; + + if (!supported_image_format(dataFormat)) { + vg_set_error(ctx, VG_UNSUPPORTED_IMAGE_FORMAT_ERROR); + return; + } + if (!data || !is_aligned(data)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + vg_validate_state(ctx); + { + struct vg_image *img = image_create(dataFormat, width, height); + image_sub_data(img, data, dataStride, dataFormat, 0, 0, + width, height); +#if 0 + struct matrix *matrix = &ctx->state.vg.image_user_to_surface_matrix; + matrix_translate(matrix, dx, dy); + image_draw(img); + matrix_translate(matrix, -dx, -dy); +#else + /* this looks like a better approach */ + image_set_pixels(dx, dy, img, 0, 0, width, height); +#endif + image_destroy(img); + } + /* make sure rendering has completed */ + pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); +} + +void vgReadPixels(void * data, VGint dataStride, + VGImageFormat dataFormat, + VGint sx, VGint sy, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_context *pipe = ctx->pipe; + struct pipe_screen *screen = pipe->screen; + + struct st_framebuffer *stfb = ctx->draw_buffer; + struct st_renderbuffer *strb = stfb->strb; + struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb; + + VGfloat temp[VEGA_MAX_IMAGE_WIDTH][4]; + VGfloat *df = (VGfloat*)temp; + VGint y = (fb->height - sy) - 1, yStep = -1; + VGint i; + VGubyte *dst = (VGubyte *)data; + VGint xoffset = 0, yoffset = 0; + + if (!supported_image_format(dataFormat)) { + vg_set_error(ctx, VG_UNSUPPORTED_IMAGE_FORMAT_ERROR); + return; + } + if (!data || !is_aligned(data)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + /* make sure rendering has completed */ + pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); + if (sx < 0) { + xoffset = -sx; + xoffset *= _vega_size_for_format(dataFormat); + width += sx; + sx = 0; + } + if (sy < 0) { + yoffset = -sy; + height += sy; + sy = 0; + y = (fb->height - sy) - 1; + yoffset *= dataStride; + } + + { + struct pipe_transfer *transfer; + + transfer = screen->get_tex_transfer(screen, strb->texture, 0, 0, 0, + PIPE_TRANSFER_READ, + 0, 0, width, height); + + /* Do a row at a time to flip image data vertically */ + for (i = 0; i < height; i++) { +#if 0 + debug_printf("%d-%d == %d\n", sy, height, y); +#endif + pipe_get_tile_rgba(transfer, sx, y, width, 1, df); + y += yStep; + _vega_pack_rgba_span_float(ctx, width, temp, dataFormat, + dst + yoffset + xoffset); + dst += dataStride; + } + + screen->tex_transfer_destroy(transfer); + } +} + +void vgCopyPixels(VGint dx, VGint dy, + VGint sx, VGint sy, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb; + struct st_renderbuffer *strb = ctx->draw_buffer->strb; + + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + /* do nothing if we copy from outside the fb */ + if (dx >= (VGint)fb->width || dy >= (VGint)fb->height || + sx >= (VGint)fb->width || sy >= (VGint)fb->height) + return; + + vg_validate_state(ctx); + /* make sure rendering has completed */ + vgFinish(); + + vg_copy_surface(ctx, strb->surface, dx, dy, + strb->surface, sx, sy, width, height); +} diff --git a/src/gallium/state_trackers/vega/api_masks.c b/src/gallium/state_trackers/vega/api_masks.c new file mode 100644 index 00000000000..4f9f3dae173 --- /dev/null +++ b/src/gallium/state_trackers/vega/api_masks.c @@ -0,0 +1,373 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "VG/openvg.h" + +#include "mask.h" +#include "renderer.h" + +#include "vg_context.h" +#include "pipe/p_context.h" +#include "pipe/p_inlines.h" +#include "pipe/internal/p_winsys_screen.h" /* for winsys->update_buffer */ + +#include "util/u_pack_color.h" +#include "util/u_draw_quad.h" +#include "util/u_memory.h" + + +#define DISABLE_1_1_MASKING 1 + +/** + * Draw a screen-aligned quadrilateral. + * Coords are window coords with y=0=bottom. These coords will be transformed + * by the vertex shader and viewport transform. + */ +static void +draw_clear_quad(struct vg_context *st, + float x0, float y0, float x1, float y1, float z, + const VGfloat color[4]) +{ + struct pipe_context *pipe = st->pipe; + struct pipe_buffer *buf; + VGuint i; + + /* positions */ + st->clear.vertices[0][0][0] = x0; + st->clear.vertices[0][0][1] = y0; + + st->clear.vertices[1][0][0] = x1; + st->clear.vertices[1][0][1] = y0; + + st->clear.vertices[2][0][0] = x1; + st->clear.vertices[2][0][1] = y1; + + st->clear.vertices[3][0][0] = x0; + st->clear.vertices[3][0][1] = y1; + + /* same for all verts: */ + for (i = 0; i < 4; i++) { + st->clear.vertices[i][0][2] = z; + st->clear.vertices[i][0][3] = 1.0; + st->clear.vertices[i][1][0] = color[0]; + st->clear.vertices[i][1][1] = color[1]; + st->clear.vertices[i][1][2] = color[2]; + st->clear.vertices[i][1][3] = color[3]; + } + + + /* put vertex data into vbuf */ + buf = pipe_user_buffer_create(pipe->screen, + st->clear.vertices, + sizeof(st->clear.vertices)); + + + /* draw */ + if (buf) { + util_draw_vertex_buffer(pipe, buf, 0, + PIPE_PRIM_TRIANGLE_FAN, + 4, /* verts */ + 2); /* attribs/vert */ + + pipe_buffer_reference(&buf, NULL); + } +} + +/** + * Do vgClear by drawing a quadrilateral. + */ +static void +clear_with_quad(struct vg_context *st, float x0, float y0, + float width, float height, const VGfloat clear_color[4]) +{ + VGfloat x1, y1; + + vg_validate_state(st); + + x1 = x0 + width; + y1 = y0 + height; + + /* + printf("%s %f,%f %f,%f\n", __FUNCTION__, + x0, y0, + x1, y1); + */ + + if (st->pipe->winsys && st->pipe->winsys->update_buffer) + st->pipe->winsys->update_buffer( st->pipe->winsys, + st->pipe->priv ); + + cso_save_blend(st->cso_context); + cso_save_rasterizer(st->cso_context); + cso_save_fragment_shader(st->cso_context); + cso_save_vertex_shader(st->cso_context); + + /* blend state: RGBA masking */ + { + struct pipe_blend_state blend; + memset(&blend, 0, sizeof(blend)); + blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE; + blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE; + blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO; + blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO; + blend.colormask |= PIPE_MASK_R; + blend.colormask |= PIPE_MASK_G; + blend.colormask |= PIPE_MASK_B; + blend.colormask |= PIPE_MASK_A; + cso_set_blend(st->cso_context, &blend); + } + + cso_set_rasterizer(st->cso_context, &st->clear.raster); + + cso_set_fragment_shader_handle(st->cso_context, st->clear.fs); + cso_set_vertex_shader_handle(st->cso_context, vg_clear_vs(st)); + + /* draw quad matching scissor rect (XXX verify coord round-off) */ + draw_clear_quad(st, x0, y0, x1, y1, 0, clear_color); + + /* Restore pipe state */ + cso_restore_blend(st->cso_context); + cso_restore_rasterizer(st->cso_context); + cso_restore_fragment_shader(st->cso_context); + cso_restore_vertex_shader(st->cso_context); +} + + +void vgMask(VGHandle mask, VGMaskOperation operation, + VGint x, VGint y, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + + if (width <=0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (operation < VG_CLEAR_MASK || operation > VG_SUBTRACT_MASK) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + + vg_validate_state(ctx); + + if (operation == VG_CLEAR_MASK) { + mask_fill(x, y, width, height, 0.f); + } else if (operation == VG_FILL_MASK) { + mask_fill(x, y, width, height, 1.f); + } else if (vg_object_is_valid((void*)mask, VG_OBJECT_IMAGE)) { + struct vg_image *image = (struct vg_image *)mask; + mask_using_image(image, operation, x, y, width, height); + } else if (vg_object_is_valid((void*)mask, VG_OBJECT_MASK)) { +#if DISABLE_1_1_MASKING + return; +#else + struct vg_mask_layer *layer = (struct vg_mask_layer *)mask; + mask_using_layer(layer, operation, x, y, width, height); +#endif + } else { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + } +} + +void vgClear(VGint x, VGint y, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_framebuffer_state *fb; + + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + vg_validate_state(ctx); +#if 0 + debug_printf("Clear [%d, %d, %d, %d] with [%f, %f, %f, %f]\n", + x, y, width, height, + ctx->state.vg.clear_color[0], + ctx->state.vg.clear_color[1], + ctx->state.vg.clear_color[2], + ctx->state.vg.clear_color[3]); +#endif + + fb = &ctx->state.g3d.fb; + /* check for a whole surface clear */ + if (!ctx->state.vg.scissoring && + (x == 0 && y == 0 && width == fb->width && height == fb->height)) { + ctx->pipe->clear(ctx->pipe, PIPE_CLEAR_COLOR | PIPE_CLEAR_DEPTHSTENCIL, + ctx->state.vg.clear_color, 1., 0); + } else { + clear_with_quad(ctx, x, y, width, height, ctx->state.vg.clear_color); + } +} + + +#ifdef OPENVG_VERSION_1_1 + + +void vgRenderToMask(VGPath path, + VGbitfield paintModes, + VGMaskOperation operation) +{ + struct vg_context *ctx = vg_current_context(); + + if (path == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (!paintModes || (paintModes&(~(VG_STROKE_PATH|VG_FILL_PATH)))) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (operation < VG_CLEAR_MASK || + operation > VG_SUBTRACT_MASK) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (!vg_object_is_valid((void*)path, VG_OBJECT_PATH)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + +#if DISABLE_1_1_MASKING + return; +#endif + + vg_validate_state(ctx); + + mask_render_to((struct path *)path, paintModes, operation); +} + +VGMaskLayer vgCreateMaskLayer(VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + + if (width <= 0 || height <= 0 || + width > vgGeti(VG_MAX_IMAGE_WIDTH) || + height > vgGeti(VG_MAX_IMAGE_HEIGHT)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_INVALID_HANDLE; + } + + return (VGMaskLayer)mask_layer_create(width, height); +} + +void vgDestroyMaskLayer(VGMaskLayer maskLayer) +{ + struct vg_mask_layer *mask = 0; + struct vg_context *ctx = vg_current_context(); + + if (maskLayer == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (!vg_object_is_valid((void*)maskLayer, VG_OBJECT_MASK)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + mask = (struct vg_mask_layer *)maskLayer; + mask_layer_destroy(mask); +} + +void vgFillMaskLayer(VGMaskLayer maskLayer, + VGint x, VGint y, + VGint width, VGint height, + VGfloat value) +{ + struct vg_mask_layer *mask = 0; + struct vg_context *ctx = vg_current_context(); + + if (maskLayer == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (value < 0 || value > 1) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (x < 0 || y < 0 || (x + width) < 0 || (y + height) < 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (!vg_object_is_valid((void*)maskLayer, VG_OBJECT_MASK)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + mask = (struct vg_mask_layer*)maskLayer; + + if (x + width > mask_layer_width(mask) || + y + height > mask_layer_height(mask)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + +#if DISABLE_1_1_MASKING + return; +#endif + mask_layer_fill(mask, x, y, width, height, value); +} + +void vgCopyMask(VGMaskLayer maskLayer, + VGint sx, VGint sy, + VGint dx, VGint dy, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_mask_layer *mask = 0; + + if (maskLayer == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (!vg_object_is_valid((void*)maskLayer, VG_OBJECT_MASK)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + +#if DISABLE_1_1_MASKING + return; +#endif + + mask = (struct vg_mask_layer*)maskLayer; + mask_copy(mask, sx, sy, dx, dy, width, height); +} + +#endif diff --git a/src/gallium/state_trackers/vega/api_misc.c b/src/gallium/state_trackers/vega/api_misc.c new file mode 100644 index 00000000000..78ba0bc1103 --- /dev/null +++ b/src/gallium/state_trackers/vega/api_misc.c @@ -0,0 +1,83 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "VG/openvg.h" + +#include "vg_context.h" + +/* Hardware Queries */ +VGHardwareQueryResult vgHardwareQuery(VGHardwareQueryType key, + VGint setting) +{ + struct vg_context *ctx = vg_current_context(); + + if (key < VG_IMAGE_FORMAT_QUERY || + key > VG_PATH_DATATYPE_QUERY) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_HARDWARE_UNACCELERATED; + } + + if (key == VG_IMAGE_FORMAT_QUERY) { + if (setting < VG_sRGBX_8888 || + setting > VG_lABGR_8888_PRE) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_HARDWARE_UNACCELERATED; + } + } else if (key == VG_PATH_DATATYPE_QUERY) { + if (setting < VG_PATH_DATATYPE_S_8 || + setting > VG_PATH_DATATYPE_F) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_HARDWARE_UNACCELERATED; + } + } + /* we're supposed to accelerate everything */ + return VG_HARDWARE_ACCELERATED; +} + +/* Renderer and Extension Information */ +const VGubyte *vgGetString(VGStringID name) +{ + struct vg_context *ctx = vg_current_context(); + static const VGubyte *vendor = (VGubyte *)"Tungsten Graphics, Inc"; + static const VGubyte *renderer = (VGubyte *)"Vega OpenVG 1.0"; + static const VGubyte *version = (VGubyte *)"1.0"; + + if (!ctx) + return NULL; + + switch(name) { + case VG_VENDOR: + return vendor; + case VG_RENDERER: + return renderer; + case VG_VERSION: + return version; + case VG_EXTENSIONS: + return NULL; + default: + return NULL; + } +} diff --git a/src/gallium/state_trackers/vega/api_paint.c b/src/gallium/state_trackers/vega/api_paint.c new file mode 100644 index 00000000000..dd3ac5bdb09 --- /dev/null +++ b/src/gallium/state_trackers/vega/api_paint.c @@ -0,0 +1,166 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "VG/openvg.h" + +#include "vg_context.h" +#include "paint.h" +#include "image.h" + +VGPaint vgCreatePaint(void) +{ + return (VGPaint) paint_create(vg_current_context()); +} + +void vgDestroyPaint(VGPaint p) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_paint *paint; + + if (p == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + paint = (struct vg_paint *)p; + paint_destroy(paint); +} + +void vgSetPaint(VGPaint paint, VGbitfield paintModes) +{ + struct vg_context *ctx = vg_current_context(); + + if (paint == VG_INVALID_HANDLE) { + /* restore the default */ + paint = (VGPaint)ctx->default_paint; + } else if (!vg_object_is_valid((void*)paint, VG_OBJECT_PAINT)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (!(paintModes & ((VG_FILL_PATH|VG_STROKE_PATH)))) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (paintModes & VG_FILL_PATH) { + ctx->state.vg.fill_paint = (struct vg_paint *)paint; + } + if (paintModes & VG_STROKE_PATH) { + ctx->state.vg.stroke_paint = (struct vg_paint *)paint; + } +} + +VGPaint vgGetPaint(VGPaintMode paintMode) +{ + struct vg_context *ctx = vg_current_context(); + VGPaint paint = VG_INVALID_HANDLE; + + if (paintMode < VG_STROKE_PATH || paintMode > VG_FILL_PATH) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_INVALID_HANDLE; + } + + if (paintMode == VG_FILL_PATH) + paint = (VGPaint)ctx->state.vg.fill_paint; + else if (paintMode == VG_STROKE_PATH) + paint = (VGPaint)ctx->state.vg.stroke_paint; + + if (paint == (VGPaint)ctx->default_paint) + paint = VG_INVALID_HANDLE; + + return paint; +} + +void vgSetColor(VGPaint paint, VGuint rgba) +{ + struct vg_context *ctx = vg_current_context(); + + if (paint == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (!vg_object_is_valid((void*)paint, VG_OBJECT_PAINT)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + { + struct vg_paint *p = (struct vg_paint *)paint; + paint_set_colori(p, rgba); + } +} + +VGuint vgGetColor(VGPaint paint) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_paint *p; + VGuint rgba = 0; + + if (paint == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return rgba; + } + + if (!vg_object_is_valid((void*)paint, VG_OBJECT_PAINT)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return rgba; + } + p = (struct vg_paint *)paint; + + return paint_colori(p); +} + +void vgPaintPattern(VGPaint paint, VGImage pattern) +{ + struct vg_context *ctx = vg_current_context(); + + if (paint == VG_INVALID_HANDLE || + !vg_context_is_object_valid(ctx, VG_OBJECT_PAINT, (void *)paint)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (pattern == VG_INVALID_HANDLE) { + paint_set_type((struct vg_paint*)paint, VG_PAINT_TYPE_COLOR); + return; + } + + if (!vg_context_is_object_valid(ctx, VG_OBJECT_IMAGE, (void *)pattern)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + + if (!vg_object_is_valid((void*)paint, VG_OBJECT_PAINT) || + !vg_object_is_valid((void*)pattern, VG_OBJECT_IMAGE)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + paint_set_pattern((struct vg_paint*)paint, + (struct vg_image*)pattern); +} + diff --git a/src/gallium/state_trackers/vega/api_params.c b/src/gallium/state_trackers/vega/api_params.c new file mode 100644 index 00000000000..db77fd9cb05 --- /dev/null +++ b/src/gallium/state_trackers/vega/api_params.c @@ -0,0 +1,1673 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "VG/openvg.h" + +#include "vg_context.h" +#include "paint.h" +#include "path.h" +#include "image.h" +#include "matrix.h" +#include "api_consts.h" + +#include "pipe/p_compiler.h" +#include "util/u_pointer.h" +#include "util/u_math.h" + +#include <math.h> + +static INLINE struct vg_state *current_state() +{ + struct vg_context *ctx = vg_current_context(); + if (!ctx) + return 0; + else + return &ctx->state.vg; +} + +static INLINE VGboolean count_in_bounds(VGParamType type, VGint count) +{ + if (count < 0) + return VG_FALSE; + + if (type == VG_SCISSOR_RECTS) + return (!(count % 4) && (count >= 0 || count <= VEGA_MAX_SCISSOR_RECTS * 4)); + else if (type == VG_STROKE_DASH_PATTERN) { + return count <= VEGA_MAX_DASH_COUNT; + } else { + VGint real_count = vgGetVectorSize(type); + return count == real_count; + } +} + +void vgSetf (VGParamType type, VGfloat value) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_state *state = current_state(); + VGErrorCode error = VG_NO_ERROR; + + switch(type) { + case VG_MATRIX_MODE: + case VG_FILL_RULE: + case VG_IMAGE_QUALITY: + case VG_RENDERING_QUALITY: + case VG_BLEND_MODE: + case VG_IMAGE_MODE: +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM: +#endif + case VG_STROKE_CAP_STYLE: + case VG_STROKE_JOIN_STYLE: + case VG_STROKE_DASH_PHASE_RESET: + case VG_MASKING: + case VG_SCISSORING: + case VG_PIXEL_LAYOUT: + case VG_SCREEN_LAYOUT: + case VG_FILTER_FORMAT_LINEAR: + case VG_FILTER_FORMAT_PREMULTIPLIED: + case VG_FILTER_CHANNEL_MASK: + + case VG_MAX_SCISSOR_RECTS: + case VG_MAX_DASH_COUNT: + case VG_MAX_KERNEL_SIZE: + case VG_MAX_SEPARABLE_KERNEL_SIZE: + case VG_MAX_COLOR_RAMP_STOPS: + case VG_MAX_IMAGE_WIDTH: + case VG_MAX_IMAGE_HEIGHT: + case VG_MAX_IMAGE_PIXELS: + case VG_MAX_IMAGE_BYTES: + case VG_MAX_GAUSSIAN_STD_DEVIATION: + case VG_MAX_FLOAT: + vgSeti(type, floor(value)); + return; + break; + case VG_STROKE_LINE_WIDTH: + state->stroke.line_width.f = value; + state->stroke.line_width.i = float_to_int_floor(*((VGuint*)(&value))); + break; + case VG_STROKE_MITER_LIMIT: + state->stroke.miter_limit.f = value; + state->stroke.miter_limit.i = float_to_int_floor(*((VGuint*)(&value))); + break; + case VG_STROKE_DASH_PHASE: + state->stroke.dash_phase.f = value; + state->stroke.dash_phase.i = float_to_int_floor(*((VGuint*)(&value))); + break; + default: + error = VG_ILLEGAL_ARGUMENT_ERROR; + break; + } + vg_set_error(ctx, error); +} + +void vgSeti (VGParamType type, VGint value) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_state *state = current_state(); + VGErrorCode error = VG_NO_ERROR; + + switch(type) { + case VG_MATRIX_MODE: + if (value < VG_MATRIX_PATH_USER_TO_SURFACE || +#ifdef OPENVG_VERSION_1_1 + value > VG_MATRIX_GLYPH_USER_TO_SURFACE) +#else + value > VG_MATRIX_STROKE_PAINT_TO_USER) +#endif + error = VG_ILLEGAL_ARGUMENT_ERROR; + else + state->matrix_mode = value; + break; + case VG_FILL_RULE: + if (value < VG_EVEN_ODD || + value > VG_NON_ZERO) + error = VG_ILLEGAL_ARGUMENT_ERROR; + else + state->fill_rule = value; + break; + case VG_IMAGE_QUALITY: + state->image_quality = value; + break; + case VG_RENDERING_QUALITY: + if (value < VG_RENDERING_QUALITY_NONANTIALIASED || + value > VG_RENDERING_QUALITY_BETTER) + error = VG_ILLEGAL_ARGUMENT_ERROR; + else + state->rendering_quality = value; + break; + case VG_BLEND_MODE: + if (value < VG_BLEND_SRC || + value > VG_BLEND_ADDITIVE) + error = VG_ILLEGAL_ARGUMENT_ERROR; + else { + ctx->state.dirty |= BLEND_DIRTY; + state->blend_mode = value; + } + break; + case VG_IMAGE_MODE: + if (value < VG_DRAW_IMAGE_NORMAL || + value > VG_DRAW_IMAGE_STENCIL) + error = VG_ILLEGAL_ARGUMENT_ERROR; + else + state->image_mode = value; +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM: + state->color_transform = value; +#endif + break; + case VG_STROKE_LINE_WIDTH: + state->stroke.line_width.f = value; + state->stroke.line_width.i = value; + break; + case VG_STROKE_CAP_STYLE: + if (value < VG_CAP_BUTT || + value > VG_CAP_SQUARE) + error = VG_ILLEGAL_ARGUMENT_ERROR; + else + state->stroke.cap_style = value; + break; + case VG_STROKE_JOIN_STYLE: + if (value < VG_JOIN_MITER || + value > VG_JOIN_BEVEL) + error = VG_ILLEGAL_ARGUMENT_ERROR; + else + state->stroke.join_style = value; + break; + case VG_STROKE_MITER_LIMIT: + state->stroke.miter_limit.f = value; + state->stroke.miter_limit.i = value; + break; + case VG_STROKE_DASH_PHASE: + state->stroke.dash_phase.f = value; + state->stroke.dash_phase.i = value; + break; + case VG_STROKE_DASH_PHASE_RESET: + state->stroke.dash_phase_reset = value; + break; + case VG_MASKING: + state->masking = value; + break; + case VG_SCISSORING: + state->scissoring = value; + ctx->state.dirty |= DEPTH_STENCIL_DIRTY; + break; + case VG_PIXEL_LAYOUT: + if (value < VG_PIXEL_LAYOUT_UNKNOWN || + value > VG_PIXEL_LAYOUT_BGR_HORIZONTAL) + error = VG_ILLEGAL_ARGUMENT_ERROR; + else + state->pixel_layout = value; + break; + case VG_SCREEN_LAYOUT: + /* read only ignore */ + break; + case VG_FILTER_FORMAT_LINEAR: + state->filter_format_linear = value; + break; + case VG_FILTER_FORMAT_PREMULTIPLIED: + state->filter_format_premultiplied = value; + break; + case VG_FILTER_CHANNEL_MASK: + state->filter_channel_mask = value; + break; + + case VG_MAX_SCISSOR_RECTS: + case VG_MAX_DASH_COUNT: + case VG_MAX_KERNEL_SIZE: + case VG_MAX_SEPARABLE_KERNEL_SIZE: + case VG_MAX_COLOR_RAMP_STOPS: + case VG_MAX_IMAGE_WIDTH: + case VG_MAX_IMAGE_HEIGHT: + case VG_MAX_IMAGE_PIXELS: + case VG_MAX_IMAGE_BYTES: + case VG_MAX_GAUSSIAN_STD_DEVIATION: + case VG_MAX_FLOAT: + /* read only ignore */ + break; + default: + error = VG_ILLEGAL_ARGUMENT_ERROR; + break; + } + vg_set_error(ctx, error); +} + +void vgSetfv(VGParamType type, VGint count, + const VGfloat * values) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_state *state = current_state(); + VGErrorCode error = VG_NO_ERROR; + + if ((count && !values) || !count_in_bounds(type, count) || !is_aligned(values)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + switch(type) { + case VG_MATRIX_MODE: + case VG_FILL_RULE: + case VG_IMAGE_QUALITY: + case VG_RENDERING_QUALITY: + case VG_BLEND_MODE: + case VG_IMAGE_MODE: +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM: +#endif + case VG_STROKE_CAP_STYLE: + case VG_STROKE_JOIN_STYLE: + case VG_STROKE_DASH_PHASE_RESET: + case VG_MASKING: + case VG_SCISSORING: + case VG_PIXEL_LAYOUT: + case VG_SCREEN_LAYOUT: + case VG_FILTER_FORMAT_LINEAR: + case VG_FILTER_FORMAT_PREMULTIPLIED: + case VG_FILTER_CHANNEL_MASK: + vgSeti(type, floor(values[0])); + return; + break; + case VG_SCISSOR_RECTS: { + VGint i; + VGuint *x = (VGuint*)values; + for (i = 0; i < count; ++i) { + state->scissor_rects[i].f = values[i]; + state->scissor_rects[i].i = float_to_int_floor(x[i]); + } + state->scissor_rects_num = count / 4; + ctx->state.dirty |= DEPTH_STENCIL_DIRTY; + } + break; +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM_VALUES: { + VGint i; + for (i = 0; i < count; ++i) { + state->color_transform_values[i] = values[i]; + } + } + break; +#endif + case VG_STROKE_LINE_WIDTH: + state->stroke.line_width.f = values[0]; + state->stroke.line_width.i = float_to_int_floor(*((VGuint*)(values))); + break; + case VG_STROKE_MITER_LIMIT: + state->stroke.miter_limit.f = values[0]; + state->stroke.miter_limit.i = float_to_int_floor(*((VGuint*)(values))); + break; + case VG_STROKE_DASH_PATTERN: { + int i; + for (i = 0; i < count; ++i) { + state->stroke.dash_pattern[i].f = values[i]; + state->stroke.dash_pattern[i].i = + float_to_int_floor(*((VGuint*)(values + i))); + } + state->stroke.dash_pattern_num = count; + } + break; + case VG_STROKE_DASH_PHASE: + state->stroke.dash_phase.f = values[0]; + state->stroke.dash_phase.i = float_to_int_floor(*((VGuint*)(values))); + break; + case VG_TILE_FILL_COLOR: + state->tile_fill_color[0] = values[0]; + state->tile_fill_color[1] = values[1]; + state->tile_fill_color[2] = values[2]; + state->tile_fill_color[3] = values[3]; + + state->tile_fill_colori[0] = float_to_int_floor(*((VGuint*)(values + 0))); + state->tile_fill_colori[1] = float_to_int_floor(*((VGuint*)(values + 1))); + state->tile_fill_colori[2] = float_to_int_floor(*((VGuint*)(values + 2))); + state->tile_fill_colori[3] = float_to_int_floor(*((VGuint*)(values + 3))); + break; + case VG_CLEAR_COLOR: + state->clear_color[0] = values[0]; + state->clear_color[1] = values[1]; + state->clear_color[2] = values[2]; + state->clear_color[3] = values[3]; + + state->clear_colori[0] = float_to_int_floor(*((VGuint*)(values + 0))); + state->clear_colori[1] = float_to_int_floor(*((VGuint*)(values + 1))); + state->clear_colori[2] = float_to_int_floor(*((VGuint*)(values + 2))); + state->clear_colori[3] = float_to_int_floor(*((VGuint*)(values + 3))); + break; +#ifdef OPENVG_VERSION_1_1 + case VG_GLYPH_ORIGIN: + state->glyph_origin[0].f = values[0]; + state->glyph_origin[1].f = values[1]; + + state->glyph_origin[0].i = float_to_int_floor(*((VGuint*)(values + 0))); + state->glyph_origin[1].i = float_to_int_floor(*((VGuint*)(values + 1))); + break; +#endif + + case VG_MAX_SCISSOR_RECTS: + case VG_MAX_DASH_COUNT: + case VG_MAX_KERNEL_SIZE: + case VG_MAX_SEPARABLE_KERNEL_SIZE: + case VG_MAX_COLOR_RAMP_STOPS: + case VG_MAX_IMAGE_WIDTH: + case VG_MAX_IMAGE_HEIGHT: + case VG_MAX_IMAGE_PIXELS: + case VG_MAX_IMAGE_BYTES: + case VG_MAX_GAUSSIAN_STD_DEVIATION: + case VG_MAX_FLOAT: + break; + default: + error = VG_ILLEGAL_ARGUMENT_ERROR; + break; + } + vg_set_error(ctx, error); +} + +void vgSetiv(VGParamType type, VGint count, + const VGint * values) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_state *state = current_state(); + + if ((count && !values) || !count_in_bounds(type, count) || !is_aligned(values)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + switch(type) { + case VG_MATRIX_MODE: + case VG_FILL_RULE: + case VG_IMAGE_QUALITY: + case VG_RENDERING_QUALITY: + case VG_BLEND_MODE: + case VG_IMAGE_MODE: +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM: +#endif + case VG_STROKE_CAP_STYLE: + case VG_STROKE_JOIN_STYLE: + case VG_STROKE_DASH_PHASE_RESET: + case VG_MASKING: + case VG_SCISSORING: + case VG_PIXEL_LAYOUT: + case VG_SCREEN_LAYOUT: + case VG_FILTER_FORMAT_LINEAR: + case VG_FILTER_FORMAT_PREMULTIPLIED: + case VG_FILTER_CHANNEL_MASK: + vgSeti(type, values[0]); + return; + break; + case VG_SCISSOR_RECTS: { + VGint i; + for (i = 0; i < count; ++i) { + state->scissor_rects[i].i = values[i]; + state->scissor_rects[i].f = values[i]; + } + state->scissor_rects_num = count / 4; + ctx->state.dirty |= DEPTH_STENCIL_DIRTY; + } + break; +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM_VALUES: { + VGint i; + for (i = 0; i < count; ++i) { + state->color_transform_values[i] = values[i]; + } + } + break; +#endif + case VG_STROKE_LINE_WIDTH: + state->stroke.line_width.f = values[0]; + state->stroke.line_width.i = values[0]; + break; + case VG_STROKE_MITER_LIMIT: + state->stroke.miter_limit.f = values[0]; + state->stroke.miter_limit.i = values[0]; + break; + case VG_STROKE_DASH_PATTERN: { + int i; + for (i = 0; i < count; ++i) { + state->stroke.dash_pattern[i].f = values[i]; + state->stroke.dash_pattern[i].i = values[i]; + } + state->stroke.dash_pattern_num = count; + } + break; + case VG_STROKE_DASH_PHASE: + state->stroke.dash_phase.f = values[0]; + state->stroke.dash_phase.i = values[0]; + break; + case VG_TILE_FILL_COLOR: + state->tile_fill_color[0] = values[0]; + state->tile_fill_color[1] = values[1]; + state->tile_fill_color[2] = values[2]; + state->tile_fill_color[3] = values[3]; + + state->tile_fill_colori[0] = values[0]; + state->tile_fill_colori[1] = values[1]; + state->tile_fill_colori[2] = values[2]; + state->tile_fill_colori[3] = values[3]; + break; + case VG_CLEAR_COLOR: + state->clear_color[0] = values[0]; + state->clear_color[1] = values[1]; + state->clear_color[2] = values[2]; + state->clear_color[3] = values[3]; + + state->clear_colori[0] = values[0]; + state->clear_colori[1] = values[1]; + state->clear_colori[2] = values[2]; + state->clear_colori[3] = values[3]; + break; +#ifdef OPENVG_VERSION_1_1 + case VG_GLYPH_ORIGIN: + state->glyph_origin[0].f = values[0]; + state->glyph_origin[1].f = values[1]; + state->glyph_origin[0].i = values[0]; + state->glyph_origin[1].i = values[1]; + break; +#endif + + case VG_MAX_SCISSOR_RECTS: + case VG_MAX_DASH_COUNT: + case VG_MAX_KERNEL_SIZE: + case VG_MAX_SEPARABLE_KERNEL_SIZE: + case VG_MAX_COLOR_RAMP_STOPS: + case VG_MAX_IMAGE_WIDTH: + case VG_MAX_IMAGE_HEIGHT: + case VG_MAX_IMAGE_PIXELS: + case VG_MAX_IMAGE_BYTES: + case VG_MAX_GAUSSIAN_STD_DEVIATION: + case VG_MAX_FLOAT: + break; + + default: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + } +} + +VGfloat vgGetf(VGParamType type) +{ + struct vg_context *ctx = vg_current_context(); + const struct vg_state *state = current_state(); + VGErrorCode error = VG_NO_ERROR; + VGfloat value = 0.0f; + + switch(type) { + case VG_MATRIX_MODE: + case VG_FILL_RULE: + case VG_IMAGE_QUALITY: + case VG_RENDERING_QUALITY: + case VG_BLEND_MODE: + case VG_IMAGE_MODE: +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM: +#endif + case VG_STROKE_CAP_STYLE: + case VG_STROKE_JOIN_STYLE: + case VG_STROKE_DASH_PHASE_RESET: + case VG_MASKING: + case VG_SCISSORING: + case VG_PIXEL_LAYOUT: + case VG_SCREEN_LAYOUT: + case VG_FILTER_FORMAT_LINEAR: + case VG_FILTER_FORMAT_PREMULTIPLIED: + case VG_FILTER_CHANNEL_MASK: + return vgGeti(type); + break; + case VG_STROKE_LINE_WIDTH: + value = state->stroke.line_width.f; + break; + case VG_STROKE_MITER_LIMIT: + value = state->stroke.miter_limit.f; + break; + case VG_STROKE_DASH_PHASE: + value = state->stroke.dash_phase.f; + break; + + case VG_MAX_SCISSOR_RECTS: + case VG_MAX_DASH_COUNT: + case VG_MAX_KERNEL_SIZE: + case VG_MAX_SEPARABLE_KERNEL_SIZE: + case VG_MAX_COLOR_RAMP_STOPS: + case VG_MAX_IMAGE_WIDTH: + case VG_MAX_IMAGE_HEIGHT: + case VG_MAX_IMAGE_PIXELS: + case VG_MAX_IMAGE_BYTES: + case VG_MAX_GAUSSIAN_STD_DEVIATION: + return vgGeti(type); + break; + case VG_MAX_FLOAT: + value = 1e+10;/*must be at least 1e+10*/ + break; + default: + error = VG_ILLEGAL_ARGUMENT_ERROR; + break; + } + vg_set_error(ctx, error); + return value; +} + +VGint vgGeti(VGParamType type) +{ + const struct vg_state *state = current_state(); + struct vg_context *ctx = vg_current_context(); + VGErrorCode error = VG_NO_ERROR; + VGint value = 0; + + switch(type) { + case VG_MATRIX_MODE: + value = state->matrix_mode; + break; + case VG_FILL_RULE: + value = state->fill_rule; + break; + case VG_IMAGE_QUALITY: + value = state->image_quality; + break; + case VG_RENDERING_QUALITY: + value = state->rendering_quality; + break; + case VG_BLEND_MODE: + value = state->blend_mode; + break; + case VG_IMAGE_MODE: + value = state->image_mode; + break; +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM: + value = state->color_transform; + break; +#endif + case VG_STROKE_LINE_WIDTH: + value = state->stroke.line_width.i; + break; + case VG_STROKE_CAP_STYLE: + value = state->stroke.cap_style; + break; + case VG_STROKE_JOIN_STYLE: + value = state->stroke.join_style; + break; + case VG_STROKE_MITER_LIMIT: + value = state->stroke.miter_limit.i; + break; + case VG_STROKE_DASH_PHASE: + value = state->stroke.dash_phase.i; + break; + case VG_STROKE_DASH_PHASE_RESET: + value = state->stroke.dash_phase_reset; + break; + case VG_MASKING: + value = state->masking; + break; + case VG_SCISSORING: + value = state->scissoring; + break; + case VG_PIXEL_LAYOUT: + value = state->pixel_layout; + break; + case VG_SCREEN_LAYOUT: + value = state->screen_layout; + break; + case VG_FILTER_FORMAT_LINEAR: + value = state->filter_format_linear; + break; + case VG_FILTER_FORMAT_PREMULTIPLIED: + value = state->filter_format_premultiplied; + break; + case VG_FILTER_CHANNEL_MASK: + value = state->filter_channel_mask; + break; + + case VG_MAX_SCISSOR_RECTS: + value = 32; /*must be at least 32*/ + break; + case VG_MAX_DASH_COUNT: + value = 16; /*must be at least 16*/ + break; + case VG_MAX_KERNEL_SIZE: + value = 7; /*must be at least 7*/ + break; + case VG_MAX_SEPARABLE_KERNEL_SIZE: + value = 15; /*must be at least 15*/ + break; + case VG_MAX_COLOR_RAMP_STOPS: + value = 256; /*must be at least 32*/ + break; + case VG_MAX_IMAGE_WIDTH: + value = 2048; + break; + case VG_MAX_IMAGE_HEIGHT: + value = 2048; + break; + case VG_MAX_IMAGE_PIXELS: + value = 2048*2048; + break; + case VG_MAX_IMAGE_BYTES: + value = 2048*2048 * 4; + break; + case VG_MAX_GAUSSIAN_STD_DEVIATION: + value = 128; /*must be at least 128*/ + break; + + case VG_MAX_FLOAT: { + VGfloat val = vgGetf(type); + value = float_to_int_floor(*((VGuint*)&val)); + } + break; + default: + error = VG_ILLEGAL_ARGUMENT_ERROR; + break; + } + vg_set_error(ctx, error); + return value; +} + +VGint vgGetVectorSize(VGParamType type) +{ + struct vg_context *ctx = vg_current_context(); + const struct vg_state *state = current_state(); + switch(type) { + case VG_MATRIX_MODE: + case VG_FILL_RULE: + case VG_IMAGE_QUALITY: + case VG_RENDERING_QUALITY: + case VG_BLEND_MODE: + case VG_IMAGE_MODE: + return 1; + case VG_SCISSOR_RECTS: + return state->scissor_rects_num * 4; +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM: + return 1; + case VG_COLOR_TRANSFORM_VALUES: + return 8; +#endif + case VG_STROKE_LINE_WIDTH: + case VG_STROKE_CAP_STYLE: + case VG_STROKE_JOIN_STYLE: + case VG_STROKE_MITER_LIMIT: + return 1; + case VG_STROKE_DASH_PATTERN: + return state->stroke.dash_pattern_num; + case VG_STROKE_DASH_PHASE: + return 1; + case VG_STROKE_DASH_PHASE_RESET: + return 1; + case VG_TILE_FILL_COLOR: + return 4; + case VG_CLEAR_COLOR: + return 4; +#ifdef OPENVG_VERSION_1_1 + case VG_GLYPH_ORIGIN: + return 2; +#endif + case VG_MASKING: + return 1; + case VG_SCISSORING: + return 1; + case VG_PIXEL_LAYOUT: + return 1; + case VG_SCREEN_LAYOUT: + return 1; + case VG_FILTER_FORMAT_LINEAR: + return 1; + case VG_FILTER_FORMAT_PREMULTIPLIED: + return 1; + case VG_FILTER_CHANNEL_MASK: + return 1; + + case VG_MAX_COLOR_RAMP_STOPS: + return 1; + case VG_MAX_SCISSOR_RECTS: + case VG_MAX_DASH_COUNT: + case VG_MAX_KERNEL_SIZE: + case VG_MAX_SEPARABLE_KERNEL_SIZE: + case VG_MAX_IMAGE_WIDTH: + case VG_MAX_IMAGE_HEIGHT: + case VG_MAX_IMAGE_PIXELS: + case VG_MAX_IMAGE_BYTES: + case VG_MAX_FLOAT: + case VG_MAX_GAUSSIAN_STD_DEVIATION: + return 1; + default: + if (ctx) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return 0; + } +} + +void vgGetfv(VGParamType type, VGint count, + VGfloat * values) +{ + const struct vg_state *state = current_state(); + struct vg_context *ctx = vg_current_context(); + VGint real_count = vgGetVectorSize(type); + + if (!values || count <= 0 || count > real_count || !is_aligned(values)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + switch(type) { + case VG_MATRIX_MODE: + case VG_FILL_RULE: + case VG_IMAGE_QUALITY: + case VG_RENDERING_QUALITY: + case VG_BLEND_MODE: + case VG_IMAGE_MODE: +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM: +#endif + case VG_STROKE_CAP_STYLE: + case VG_STROKE_JOIN_STYLE: + case VG_STROKE_DASH_PHASE_RESET: + case VG_MASKING: + case VG_SCISSORING: + case VG_PIXEL_LAYOUT: + case VG_SCREEN_LAYOUT: + case VG_FILTER_FORMAT_LINEAR: + case VG_FILTER_FORMAT_PREMULTIPLIED: + case VG_FILTER_CHANNEL_MASK: + case VG_MAX_SCISSOR_RECTS: + case VG_MAX_DASH_COUNT: + case VG_MAX_KERNEL_SIZE: + case VG_MAX_SEPARABLE_KERNEL_SIZE: + case VG_MAX_COLOR_RAMP_STOPS: + case VG_MAX_IMAGE_WIDTH: + case VG_MAX_IMAGE_HEIGHT: + case VG_MAX_IMAGE_PIXELS: + case VG_MAX_IMAGE_BYTES: + case VG_MAX_GAUSSIAN_STD_DEVIATION: + values[0] = vgGeti(type); + break; + case VG_MAX_FLOAT: + values[0] = vgGetf(type); + break; + case VG_SCISSOR_RECTS: { + VGint i; + for (i = 0; i < count; ++i) { + values[i] = state->scissor_rects[i].f; + } + } + break; +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM_VALUES: { + memcpy(values, state->color_transform_values, + sizeof(VGfloat) * count); + } + break; +#endif + case VG_STROKE_LINE_WIDTH: + values[0] = state->stroke.line_width.f; + break; + case VG_STROKE_MITER_LIMIT: + values[0] = state->stroke.miter_limit.f; + break; + case VG_STROKE_DASH_PATTERN: { + VGint i; + for (i = 0; i < count; ++i) { + values[i] = state->stroke.dash_pattern[i].f; + } + } + break; + case VG_STROKE_DASH_PHASE: + values[0] = state->stroke.dash_phase.f; + break; + case VG_TILE_FILL_COLOR: + values[0] = state->tile_fill_color[0]; + values[1] = state->tile_fill_color[1]; + values[2] = state->tile_fill_color[2]; + values[3] = state->tile_fill_color[3]; + break; + case VG_CLEAR_COLOR: + values[0] = state->clear_color[0]; + values[1] = state->clear_color[1]; + values[2] = state->clear_color[2]; + values[3] = state->clear_color[3]; + break; +#ifdef OPENVG_VERSION_1_1 + case VG_GLYPH_ORIGIN: + values[0] = state->glyph_origin[0].f; + values[1] = state->glyph_origin[1].f; + break; +#endif + default: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + } +} + +void vgGetiv(VGParamType type, VGint count, + VGint * values) +{ + const struct vg_state *state = current_state(); + struct vg_context *ctx = vg_current_context(); + VGint real_count = vgGetVectorSize(type); + + if (!values || count <= 0 || count > real_count || !is_aligned(values)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + switch(type) { + case VG_MATRIX_MODE: + case VG_FILL_RULE: + case VG_IMAGE_QUALITY: + case VG_RENDERING_QUALITY: + case VG_BLEND_MODE: + case VG_IMAGE_MODE: +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM: +#endif + case VG_STROKE_CAP_STYLE: + case VG_STROKE_JOIN_STYLE: + case VG_STROKE_DASH_PHASE_RESET: + case VG_MASKING: + case VG_SCISSORING: + case VG_PIXEL_LAYOUT: + case VG_SCREEN_LAYOUT: + case VG_FILTER_FORMAT_LINEAR: + case VG_FILTER_FORMAT_PREMULTIPLIED: + case VG_FILTER_CHANNEL_MASK: + case VG_MAX_SCISSOR_RECTS: + case VG_MAX_DASH_COUNT: + case VG_MAX_KERNEL_SIZE: + case VG_MAX_SEPARABLE_KERNEL_SIZE: + case VG_MAX_COLOR_RAMP_STOPS: + case VG_MAX_IMAGE_WIDTH: + case VG_MAX_IMAGE_HEIGHT: + case VG_MAX_IMAGE_PIXELS: + case VG_MAX_IMAGE_BYTES: + case VG_MAX_GAUSSIAN_STD_DEVIATION: + values[0] = vgGeti(type); + break; + case VG_MAX_FLOAT: { + VGfloat val = vgGetf(type); + values[0] = float_to_int_floor(*((VGuint*)&val)); + } + break; + case VG_SCISSOR_RECTS: { + VGint i; + for (i = 0; i < count; ++i) { + values[i] = state->scissor_rects[i].i; + } + } + break; +#ifdef OPENVG_VERSION_1_1 + case VG_COLOR_TRANSFORM_VALUES: { + VGint i; + VGuint *x = (VGuint*)state->color_transform_values; + for (i = 0; i < count; ++i) { + values[i] = float_to_int_floor(x[i]); + } + } + break; +#endif + case VG_STROKE_LINE_WIDTH: + values[0] = state->stroke.line_width.i; + break; + case VG_STROKE_MITER_LIMIT: + values[0] = state->stroke.miter_limit.i; + break; + case VG_STROKE_DASH_PATTERN: { + VGint i; + for (i = 0; i < count; ++i) { + values[i] = state->stroke.dash_pattern[i].i; + } + } + break; + case VG_STROKE_DASH_PHASE: + values[0] = state->stroke.dash_phase.i; + break; + case VG_TILE_FILL_COLOR: + values[0] = state->tile_fill_colori[0]; + values[1] = state->tile_fill_colori[1]; + values[2] = state->tile_fill_colori[2]; + values[3] = state->tile_fill_colori[3]; + break; + case VG_CLEAR_COLOR: + values[0] = state->clear_colori[0]; + values[1] = state->clear_colori[1]; + values[2] = state->clear_colori[2]; + values[3] = state->clear_colori[3]; + break; +#ifdef OPENVG_VERSION_1_1 + case VG_GLYPH_ORIGIN: + values[0] = state->glyph_origin[0].i; + values[1] = state->glyph_origin[1].i; + break; +#endif + default: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + } +} + +void vgSetParameterf(VGHandle object, + VGint paramType, + VGfloat value) +{ + struct vg_context *ctx = vg_current_context(); + void *ptr = (void*)object; + + if (!object || object == VG_INVALID_HANDLE || !is_aligned(ptr)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + switch(paramType) { + case VG_PAINT_TYPE: + case VG_PAINT_COLOR_RAMP_SPREAD_MODE: + case VG_PAINT_PATTERN_TILING_MODE: + vgSetParameteri(object, paramType, floor(value)); + return; + break; + case VG_PAINT_COLOR: + case VG_PAINT_COLOR_RAMP_STOPS: + case VG_PAINT_LINEAR_GRADIENT: + case VG_PAINT_RADIAL_GRADIENT: + /* it's an error if paramType refers to a vector parameter */ + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + case VG_PAINT_COLOR_RAMP_PREMULTIPLIED: { + struct vg_paint *p = (struct vg_paint *)object; + paint_set_color_ramp_premultiplied(p, value); + } + break; + + case VG_PATH_DATATYPE: + case VG_PATH_FORMAT: + case VG_PATH_SCALE: + case VG_PATH_BIAS: + case VG_PATH_NUM_SEGMENTS: + case VG_PATH_NUM_COORDS: + + case VG_IMAGE_FORMAT: + case VG_IMAGE_WIDTH: + case VG_IMAGE_HEIGHT: + +#ifdef OPENVG_VERSION_1_1 + case VG_FONT_NUM_GLYPHS: + /* read only don't produce an error */ + break; +#endif + default: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + } +} + +void vgSetParameteri(VGHandle object, + VGint paramType, + VGint value) +{ + struct vg_context *ctx = vg_current_context(); + void *ptr = (void*)object; + + if (!object || object == VG_INVALID_HANDLE || !is_aligned(ptr)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + switch(paramType) { + case VG_PAINT_TYPE: + if (value < VG_PAINT_TYPE_COLOR || + value > VG_PAINT_TYPE_PATTERN) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else { + struct vg_paint *paint = (struct vg_paint *)ptr; + paint_set_type(paint, value); + } + break; + case VG_PAINT_COLOR: + case VG_PAINT_COLOR_RAMP_STOPS: + case VG_PAINT_LINEAR_GRADIENT: + case VG_PAINT_RADIAL_GRADIENT: + /* it's an error if paramType refers to a vector parameter */ + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + case VG_PAINT_COLOR_RAMP_SPREAD_MODE: + if (value < VG_COLOR_RAMP_SPREAD_PAD || + value > VG_COLOR_RAMP_SPREAD_REFLECT) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else { + struct vg_paint *paint = (struct vg_paint *)ptr; + paint_set_spread_mode(paint, value); + } + break; + case VG_PAINT_COLOR_RAMP_PREMULTIPLIED: { + struct vg_paint *p = (struct vg_paint *)object; + paint_set_color_ramp_premultiplied(p, value); + } + break; + case VG_PAINT_PATTERN_TILING_MODE: + if (value < VG_TILE_FILL || + value > VG_TILE_REFLECT) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else { + struct vg_paint *paint = (struct vg_paint *)ptr; + paint_set_pattern_tiling(paint, value); + } + break; + + case VG_PATH_DATATYPE: + case VG_PATH_FORMAT: + case VG_PATH_SCALE: + case VG_PATH_BIAS: + case VG_PATH_NUM_SEGMENTS: + case VG_PATH_NUM_COORDS: + + case VG_IMAGE_FORMAT: + case VG_IMAGE_WIDTH: + case VG_IMAGE_HEIGHT: + +#ifdef OPENVG_VERSION_1_1 + case VG_FONT_NUM_GLYPHS: + /* read only don't produce an error */ + break; +#endif + default: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } +} + +void vgSetParameterfv(VGHandle object, + VGint paramType, + VGint count, + const VGfloat * values) +{ + struct vg_context *ctx = vg_current_context(); + void *ptr = (void*)object; + VGint real_count = vgGetParameterVectorSize(object, paramType); + + if (!object || object == VG_INVALID_HANDLE || !is_aligned(ptr)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (count < 0 || count < real_count || + (values == NULL && count != 0) || + !is_aligned(values)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + switch(paramType) { + case VG_PAINT_TYPE: + case VG_PAINT_COLOR_RAMP_SPREAD_MODE: + case VG_PAINT_COLOR_RAMP_PREMULTIPLIED: + case VG_PAINT_PATTERN_TILING_MODE: + if (count != 1) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else + vgSetParameterf(object, paramType, values[0]); + return; + break; + case VG_PAINT_COLOR: { + if (count != 4) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else { + struct vg_paint *paint = (struct vg_paint *)object; + paint_set_color(paint, values); + } + } + break; + case VG_PAINT_COLOR_RAMP_STOPS: { + if (count && count < 4) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else { + struct vg_paint *paint = (struct vg_paint *)object; + count = MIN2(count, VEGA_MAX_COLOR_RAMP_STOPS); + paint_set_ramp_stops(paint, values, count); + { + VGint stopsi[VEGA_MAX_COLOR_RAMP_STOPS]; + int i = 0; + for (i = 0; i < count; ++i) { + stopsi[i] = float_to_int_floor(*((VGuint*)(values + i))); + } + paint_set_ramp_stopsi(paint, stopsi, count); + } + } + } + break; + case VG_PAINT_LINEAR_GRADIENT: { + if (count != 4) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else { + struct vg_paint *paint = (struct vg_paint *)object; + paint_set_linear_gradient(paint, values); + { + VGint vals[4]; + vals[0] = FLT_TO_INT(values[0]); + vals[1] = FLT_TO_INT(values[1]); + vals[2] = FLT_TO_INT(values[2]); + vals[3] = FLT_TO_INT(values[3]); + paint_set_linear_gradienti(paint, vals); + } + } + } + break; + case VG_PAINT_RADIAL_GRADIENT: { + if (count != 5) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else { + struct vg_paint *paint = (struct vg_paint *)object; + paint_set_radial_gradient(paint, values); + { + VGint vals[5]; + vals[0] = FLT_TO_INT(values[0]); + vals[1] = FLT_TO_INT(values[1]); + vals[2] = FLT_TO_INT(values[2]); + vals[3] = FLT_TO_INT(values[3]); + vals[4] = FLT_TO_INT(values[4]); + paint_set_radial_gradienti(paint, vals); + } + } + } + break; + + case VG_PATH_DATATYPE: + case VG_PATH_FORMAT: + case VG_PATH_SCALE: + case VG_PATH_BIAS: + case VG_PATH_NUM_SEGMENTS: + case VG_PATH_NUM_COORDS: + +#ifdef OPENVG_VERSION_1_1 + case VG_FONT_NUM_GLYPHS: + /* read only don't produce an error */ + break; +#endif + default: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } +} + +void vgSetParameteriv(VGHandle object, + VGint paramType, + VGint count, + const VGint * values) +{ + struct vg_context *ctx = vg_current_context(); + void *ptr = (void*)object; + VGint real_count = vgGetParameterVectorSize(object, paramType); + + if (!object || object == VG_INVALID_HANDLE || !is_aligned(ptr)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (count < 0 || count < real_count || + (values == NULL && count != 0) || + !is_aligned(values)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + switch(paramType) { + case VG_PAINT_TYPE: + case VG_PAINT_COLOR_RAMP_SPREAD_MODE: + case VG_PAINT_COLOR_RAMP_PREMULTIPLIED: + case VG_PAINT_PATTERN_TILING_MODE: + if (count != 1) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else + vgSetParameteri(object, paramType, values[0]); + return; + break; + case VG_PAINT_COLOR: { + if (count != 4) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else { + struct vg_paint *paint = (struct vg_paint *)object; + paint_set_coloriv(paint, values); + } + } + break; + case VG_PAINT_COLOR_RAMP_STOPS: { + if ((count % 5)) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else { + VGfloat *vals = 0; + int i; + struct vg_paint *paint = (struct vg_paint *)object; + if (count) { + vals = malloc(sizeof(VGfloat)*count); + for (i = 0; i < count; ++i) + vals[i] = values[i]; + } + + paint_set_ramp_stopsi(paint, values, count); + paint_set_ramp_stops(paint, vals, count); + free(vals); + } + } + break; + case VG_PAINT_LINEAR_GRADIENT: { + if (count != 4) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else { + VGfloat vals[4]; + struct vg_paint *paint = (struct vg_paint *)object; + vals[0] = values[0]; + vals[1] = values[1]; + vals[2] = values[2]; + vals[3] = values[3]; + paint_set_linear_gradient(paint, vals); + paint_set_linear_gradienti(paint, values); + } + } + break; + case VG_PAINT_RADIAL_GRADIENT: { + if (count != 5) + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + else { + VGfloat vals[5]; + struct vg_paint *paint = (struct vg_paint *)object; + vals[0] = values[0]; + vals[1] = values[1]; + vals[2] = values[2]; + vals[3] = values[3]; + vals[4] = values[4]; + paint_set_radial_gradient(paint, vals); + paint_set_radial_gradienti(paint, values); + } + } + break; + case VG_PATH_DATATYPE: + case VG_PATH_FORMAT: + case VG_PATH_SCALE: + case VG_PATH_BIAS: + case VG_PATH_NUM_SEGMENTS: + case VG_PATH_NUM_COORDS: + /* read only don't produce an error */ + break; + default: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } +} + +VGint vgGetParameterVectorSize(VGHandle object, + VGint paramType) +{ + struct vg_context *ctx = vg_current_context(); + void *ptr = (void*)object; + + if (!ptr || object == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return 0; + } + + switch(paramType) { + case VG_PAINT_TYPE: + case VG_PAINT_COLOR_RAMP_SPREAD_MODE: + case VG_PAINT_COLOR_RAMP_PREMULTIPLIED: + case VG_PAINT_PATTERN_TILING_MODE: + return 1; + case VG_PAINT_COLOR: + return 4; + case VG_PAINT_COLOR_RAMP_STOPS: { + struct vg_paint *p = (struct vg_paint *)object; + return paint_num_ramp_stops(p); + } + break; + case VG_PAINT_LINEAR_GRADIENT: + return 4; + case VG_PAINT_RADIAL_GRADIENT: + return 5; + + + case VG_PATH_FORMAT: + case VG_PATH_DATATYPE: + case VG_PATH_SCALE: + case VG_PATH_BIAS: + case VG_PATH_NUM_SEGMENTS: + case VG_PATH_NUM_COORDS: + return 1; + + case VG_IMAGE_FORMAT: + case VG_IMAGE_WIDTH: + case VG_IMAGE_HEIGHT: + return 1; + +#ifdef OPENVG_VERSION_1_1 + case VG_FONT_NUM_GLYPHS: + return 1; +#endif + + default: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + } + return 0; +} + + +VGfloat vgGetParameterf(VGHandle object, + VGint paramType) +{ + struct vg_context *ctx = vg_current_context(); + void *ptr = (void*)object; + + if (!ptr || object == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return 0; + } + + switch(paramType) { + case VG_PAINT_TYPE: + case VG_PAINT_COLOR_RAMP_SPREAD_MODE: + case VG_PAINT_COLOR_RAMP_PREMULTIPLIED: + case VG_PAINT_PATTERN_TILING_MODE: + return vgGetParameteri(object, paramType); + break; + case VG_PAINT_COLOR: + case VG_PAINT_COLOR_RAMP_STOPS: + case VG_PAINT_LINEAR_GRADIENT: + case VG_PAINT_RADIAL_GRADIENT: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + + case VG_PATH_FORMAT: + return VG_PATH_FORMAT_STANDARD; + case VG_PATH_SCALE: { + struct path *p = (struct path*)object; + return path_scale(p); + } + case VG_PATH_BIAS: { + struct path *p = (struct path*)object; + return path_bias(p); + } + case VG_PATH_DATATYPE: + case VG_PATH_NUM_SEGMENTS: + case VG_PATH_NUM_COORDS: + return vgGetParameteri(object, paramType); + break; + + case VG_IMAGE_FORMAT: + case VG_IMAGE_WIDTH: + case VG_IMAGE_HEIGHT: +#ifdef OPENVG_VERSION_1_1 + case VG_FONT_NUM_GLYPHS: + return vgGetParameteri(object, paramType); + break; +#endif + + default: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + } + return 0; +} + +VGint vgGetParameteri(VGHandle object, + VGint paramType) +{ + struct vg_context *ctx = vg_current_context(); + void *ptr = (void*)object; + + if (!ptr || object == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return 0; + } + + switch(paramType) { + case VG_PAINT_TYPE: { + struct vg_paint *paint = (struct vg_paint *)ptr; + return paint_type(paint); + } + break; + case VG_PAINT_COLOR_RAMP_SPREAD_MODE: { + struct vg_paint *p = (struct vg_paint *)object; + return paint_spread_mode(p); + } + case VG_PAINT_COLOR_RAMP_PREMULTIPLIED: { + struct vg_paint *p = (struct vg_paint *)object; + return paint_color_ramp_premultiplied(p); + } + break; + case VG_PAINT_PATTERN_TILING_MODE: { + struct vg_paint *p = (struct vg_paint *)object; + return paint_pattern_tiling(p); + } + break; + case VG_PAINT_COLOR: + case VG_PAINT_COLOR_RAMP_STOPS: + case VG_PAINT_LINEAR_GRADIENT: + case VG_PAINT_RADIAL_GRADIENT: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + + case VG_PATH_FORMAT: + return VG_PATH_FORMAT_STANDARD; + case VG_PATH_SCALE: + case VG_PATH_BIAS: + return vgGetParameterf(object, paramType); + case VG_PATH_DATATYPE: { + struct path *p = (struct path*)object; + return path_datatype(p); + } + case VG_PATH_NUM_SEGMENTS: { + struct path *p = (struct path*)object; + return path_num_segments(p); + } + case VG_PATH_NUM_COORDS: { + struct path *p = (struct path*)object; + return path_num_coords(p); + } + break; + + case VG_IMAGE_FORMAT: { + struct vg_image *img = (struct vg_image*)object; + return img->format; + } + break; + case VG_IMAGE_WIDTH: { + struct vg_image *img = (struct vg_image*)object; + return img->width; + } + break; + case VG_IMAGE_HEIGHT: { + struct vg_image *img = (struct vg_image*)object; + return img->height; + } + break; + +#ifdef OPENVG_VERSION_1_1 + case VG_FONT_NUM_GLYPHS: { + return 1; + } + break; +#endif + + default: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + } + return 0; +} + +void vgGetParameterfv(VGHandle object, + VGint paramType, + VGint count, + VGfloat * values) +{ + struct vg_context *ctx = vg_current_context(); + void *ptr = (void*)object; + VGint real_count = vgGetParameterVectorSize(object, paramType); + + if (!ptr || object == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (!values || count <= 0 || count > real_count || + !is_aligned(values)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + switch(paramType) { + case VG_PAINT_TYPE: { + struct vg_paint *p = (struct vg_paint *)object; + values[0] = paint_type(p); + } + break; + case VG_PAINT_COLOR_RAMP_SPREAD_MODE: { + struct vg_paint *p = (struct vg_paint *)object; + values[0] = paint_spread_mode(p); + } + break; + case VG_PAINT_COLOR_RAMP_PREMULTIPLIED: { + struct vg_paint *p = (struct vg_paint *)object; + values[0] = paint_color_ramp_premultiplied(p); + } + break; + case VG_PAINT_PATTERN_TILING_MODE: { + values[0] = vgGetParameterf(object, paramType); + } + break; + case VG_PAINT_COLOR: { + struct vg_paint *paint = (struct vg_paint *)object; + paint_get_color(paint, values); + } + break; + case VG_PAINT_COLOR_RAMP_STOPS: { + struct vg_paint *paint = (struct vg_paint *)object; + paint_ramp_stops(paint, values, count); + } + break; + case VG_PAINT_LINEAR_GRADIENT: { + struct vg_paint *paint = (struct vg_paint *)object; + paint_linear_gradient(paint, values); + } + break; + case VG_PAINT_RADIAL_GRADIENT: { + struct vg_paint *paint = (struct vg_paint *)object; + paint_radial_gradient(paint, values); + } + break; + + case VG_PATH_FORMAT: + case VG_PATH_DATATYPE: + case VG_PATH_NUM_SEGMENTS: + case VG_PATH_NUM_COORDS: + values[0] = vgGetParameteri(object, paramType); + break; + case VG_PATH_SCALE: + case VG_PATH_BIAS: + values[0] = vgGetParameterf(object, paramType); + break; + + case VG_IMAGE_FORMAT: + case VG_IMAGE_WIDTH: + case VG_IMAGE_HEIGHT: +#ifdef OPENVG_VERSION_1_1 + case VG_FONT_NUM_GLYPHS: + values[0] = vgGetParameteri(object, paramType); + break; +#endif + + default: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + } +} + +void vgGetParameteriv(VGHandle object, + VGint paramType, + VGint count, + VGint * values) +{ + struct vg_context *ctx = vg_current_context(); + void *ptr = (void*)object; + VGint real_count = vgGetParameterVectorSize(object, paramType); + + if (!ptr || object == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (!values || count <= 0 || count > real_count || + !is_aligned(values)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + switch(paramType) { + case VG_PAINT_TYPE: + case VG_PAINT_COLOR_RAMP_SPREAD_MODE: + case VG_PAINT_COLOR_RAMP_PREMULTIPLIED: + case VG_PAINT_PATTERN_TILING_MODE: +#ifdef OPENVG_VERSION_1_1 + case VG_FONT_NUM_GLYPHS: + values[0] = vgGetParameteri(object, paramType); + break; +#endif + case VG_PAINT_COLOR: { + struct vg_paint *paint = (struct vg_paint *)object; + paint_get_coloriv(paint, values); + } + break; + case VG_PAINT_COLOR_RAMP_STOPS: { + struct vg_paint *paint = (struct vg_paint *)object; + paint_ramp_stopsi(paint, values, count); + } + break; + case VG_PAINT_LINEAR_GRADIENT: { + struct vg_paint *paint = (struct vg_paint *)object; + paint_linear_gradienti(paint, values); + } + break; + case VG_PAINT_RADIAL_GRADIENT: { + struct vg_paint *paint = (struct vg_paint *)object; + paint_radial_gradienti(paint, values); + } + break; + + case VG_PATH_SCALE: + case VG_PATH_BIAS: + values[0] = vgGetParameterf(object, paramType); + break; + case VG_PATH_FORMAT: + case VG_PATH_DATATYPE: + case VG_PATH_NUM_SEGMENTS: + case VG_PATH_NUM_COORDS: + values[0] = vgGetParameteri(object, paramType); + break; + + case VG_IMAGE_FORMAT: + case VG_IMAGE_WIDTH: + case VG_IMAGE_HEIGHT: + values[0] = vgGetParameteri(object, paramType); + break; + + default: + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + break; + } +} diff --git a/src/gallium/state_trackers/vega/api_path.c b/src/gallium/state_trackers/vega/api_path.c new file mode 100644 index 00000000000..a6b7a2bb93a --- /dev/null +++ b/src/gallium/state_trackers/vega/api_path.c @@ -0,0 +1,488 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "VG/openvg.h" + +#include "vg_context.h" +#include "path.h" +#include "polygon.h" +#include "paint.h" + +#include "pipe/p_context.h" +#include "pipe/p_inlines.h" +#include "util/u_draw_quad.h" + +VGPath vgCreatePath(VGint pathFormat, + VGPathDatatype datatype, + VGfloat scale, VGfloat bias, + VGint segmentCapacityHint, + VGint coordCapacityHint, + VGbitfield capabilities) +{ + struct vg_context *ctx = vg_current_context(); + + if (pathFormat != VG_PATH_FORMAT_STANDARD) { + vg_set_error(ctx, VG_UNSUPPORTED_PATH_FORMAT_ERROR); + return VG_INVALID_HANDLE; + } + if (datatype < VG_PATH_DATATYPE_S_8 || + datatype > VG_PATH_DATATYPE_F) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_INVALID_HANDLE; + } + if (!scale) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_INVALID_HANDLE; + } + + return (VGPath)path_create(datatype, scale, bias, + segmentCapacityHint, coordCapacityHint, + capabilities); +} + +void vgClearPath(VGPath path, VGbitfield capabilities) +{ + struct vg_context *ctx = vg_current_context(); + struct path *p = 0; + + if (path == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + p = (struct path *)path; + path_clear(p, capabilities); +} + +void vgDestroyPath(VGPath p) +{ + struct path *path = 0; + struct vg_context *ctx = vg_current_context(); + + if (p == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + path = (struct path *)p; + path_destroy(path); +} + +void vgRemovePathCapabilities(VGPath path, + VGbitfield capabilities) +{ + struct vg_context *ctx = vg_current_context(); + VGbitfield current; + struct path *p; + + if (path == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + p = (struct path*)path; + current = path_capabilities(p); + path_set_capabilities(p, (current & + (~(capabilities & VG_PATH_CAPABILITY_ALL)))); +} + +VGbitfield vgGetPathCapabilities(VGPath path) +{ + struct vg_context *ctx = vg_current_context(); + struct path *p = 0; + + if (path == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return 0; + } + p = (struct path*)path; + return path_capabilities(p); +} + +void vgAppendPath(VGPath dstPath, VGPath srcPath) +{ + struct vg_context *ctx = vg_current_context(); + struct path *src, *dst; + + if (dstPath == VG_INVALID_HANDLE || srcPath == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + src = (struct path *)srcPath; + dst = (struct path *)dstPath; + + if (!(path_capabilities(src) & VG_PATH_CAPABILITY_APPEND_FROM) || + !(path_capabilities(dst) & VG_PATH_CAPABILITY_APPEND_TO)) { + vg_set_error(ctx, VG_PATH_CAPABILITY_ERROR); + return; + } + path_append_path(dst, src); +} + +void vgAppendPathData(VGPath dstPath, + VGint numSegments, + const VGubyte * pathSegments, + const void * pathData) +{ + struct vg_context *ctx = vg_current_context(); + struct path *p = 0; + VGint i; + + if (dstPath == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (!pathSegments) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (numSegments <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + for (i = 0; i < numSegments; ++i) { + if (pathSegments[i] < VG_CLOSE_PATH || + pathSegments[i] > VG_LCWARC_TO_REL) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + } + + p = (struct path*)dstPath; + + if (!pathData || !is_aligned_to(pathData, path_datatype_size(p))) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (!(path_capabilities(p)&VG_PATH_CAPABILITY_APPEND_TO)) { + vg_set_error(ctx, VG_PATH_CAPABILITY_ERROR); + return; + } + + path_append_data(p, numSegments, pathSegments, pathData); +} + +void vgModifyPathCoords(VGPath dstPath, + VGint startIndex, + VGint numSegments, + const void * pathData) +{ + struct vg_context *ctx = vg_current_context(); + struct path *p = 0; + + if (dstPath == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (startIndex < 0 || numSegments <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + p = (struct path *)dstPath; + + if (!pathData || !is_aligned_to(pathData, path_datatype_size(p))) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (startIndex + numSegments > path_num_segments(p)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (!(path_capabilities(p)&VG_PATH_CAPABILITY_MODIFY)) { + vg_set_error(ctx, VG_PATH_CAPABILITY_ERROR); + return; + } + path_modify_coords(p, startIndex, numSegments, pathData); +} + +void vgTransformPath(VGPath dstPath, VGPath srcPath) +{ + struct vg_context *ctx = vg_current_context(); + struct path *src = 0, *dst = 0; + + if (dstPath == VG_INVALID_HANDLE || srcPath == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + src = (struct path *)srcPath; + dst = (struct path *)dstPath; + + if (!(path_capabilities(src) & VG_PATH_CAPABILITY_TRANSFORM_FROM) || + !(path_capabilities(dst) & VG_PATH_CAPABILITY_TRANSFORM_TO)) { + vg_set_error(ctx, VG_PATH_CAPABILITY_ERROR); + return; + } + path_transform(dst, src); +} + +VGboolean vgInterpolatePath(VGPath dstPath, + VGPath startPath, + VGPath endPath, + VGfloat amount) +{ + struct vg_context *ctx = vg_current_context(); + struct path *start = 0, *dst = 0, *end = 0; + + if (dstPath == VG_INVALID_HANDLE || + startPath == VG_INVALID_HANDLE || + endPath == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return VG_FALSE; + } + dst = (struct path *)dstPath; + start = (struct path *)startPath; + end = (struct path *)endPath; + + if (!(path_capabilities(dst) & VG_PATH_CAPABILITY_INTERPOLATE_TO) || + !(path_capabilities(start) & VG_PATH_CAPABILITY_INTERPOLATE_FROM) || + !(path_capabilities(end) & VG_PATH_CAPABILITY_INTERPOLATE_FROM)) { + vg_set_error(ctx, VG_PATH_CAPABILITY_ERROR); + return VG_FALSE; + } + + return path_interpolate(dst, + start, end, amount); +} + +VGfloat vgPathLength(VGPath path, + VGint startSegment, + VGint numSegments) +{ + struct vg_context *ctx = vg_current_context(); + struct path *p = 0; + + if (path == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return -1; + } + if (startSegment < 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return -1; + } + if (numSegments <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return -1; + } + p = (struct path*)path; + + if (!(path_capabilities(p) & VG_PATH_CAPABILITY_PATH_LENGTH)) { + vg_set_error(ctx, VG_PATH_CAPABILITY_ERROR); + return -1; + } + if (startSegment + numSegments > path_num_segments(p)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return -1; + } + + return path_length(p, startSegment, numSegments); +} + +void vgPointAlongPath(VGPath path, + VGint startSegment, + VGint numSegments, + VGfloat distance, + VGfloat * x, VGfloat * y, + VGfloat * tangentX, + VGfloat * tangentY) +{ + struct vg_context *ctx = vg_current_context(); + struct path *p = 0; + VGbitfield caps; + + if (path == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (startSegment < 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (numSegments <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (!is_aligned(x) || !is_aligned(y) || + !is_aligned(tangentX) || !is_aligned(tangentY)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + p = (struct path*)path; + + caps = path_capabilities(p); + if (!(caps & VG_PATH_CAPABILITY_POINT_ALONG_PATH) || + !(caps & VG_PATH_CAPABILITY_TANGENT_ALONG_PATH)) { + vg_set_error(ctx, VG_PATH_CAPABILITY_ERROR); + return; + } + + if (startSegment + numSegments > path_num_segments(p)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + { + VGfloat point[2], normal[2]; + path_point(p, startSegment, numSegments, distance, + point, normal); + if (x) + *x = point[0]; + if (y) + *y = point[1]; + if (tangentX) + *tangentX = -normal[1]; + if (tangentY) + *tangentY = normal[0]; + } +} + +void vgPathBounds(VGPath path, + VGfloat * minX, + VGfloat * minY, + VGfloat * width, + VGfloat * height) +{ + struct vg_context *ctx = vg_current_context(); + struct path *p = 0; + VGbitfield caps; + + if (path == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (!minX || !minY || !width || !height) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (!is_aligned(minX) || !is_aligned(minY) || + !is_aligned(width) || !is_aligned(height)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + p = (struct path*)path; + + caps = path_capabilities(p); + if (!(caps & VG_PATH_CAPABILITY_PATH_BOUNDS)) { + vg_set_error(ctx, VG_PATH_CAPABILITY_ERROR); + return; + } + + path_bounding_rect(p, minX, minY, width, height); +} + +void vgPathTransformedBounds(VGPath path, + VGfloat * minX, + VGfloat * minY, + VGfloat * width, + VGfloat * height) +{ + struct vg_context *ctx = vg_current_context(); + struct path *p = 0; + VGbitfield caps; + + if (path == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (!minX || !minY || !width || !height) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (!is_aligned(minX) || !is_aligned(minY) || + !is_aligned(width) || !is_aligned(height)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + p = (struct path*)path; + + caps = path_capabilities(p); + if (!(caps & VG_PATH_CAPABILITY_PATH_TRANSFORMED_BOUNDS)) { + vg_set_error(ctx, VG_PATH_CAPABILITY_ERROR); + return; + } + +#if 0 + /* faster, but seems to have precision problems... */ + path_bounding_rect(p, minX, minY, width, height); + if (*width > 0 && *height > 0) { + VGfloat pts[] = {*minX, *minY, + *minX + *width, *minY, + *minX + *width, *minY + *height, + *minX, *minY + *height}; + struct matrix *matrix = &ctx->state.vg.path_user_to_surface_matrix; + VGfloat maxX, maxY; + matrix_map_point(matrix, pts[0], pts[1], pts + 0, pts + 1); + matrix_map_point(matrix, pts[2], pts[3], pts + 2, pts + 3); + matrix_map_point(matrix, pts[4], pts[5], pts + 4, pts + 5); + matrix_map_point(matrix, pts[6], pts[7], pts + 6, pts + 7); + *minX = MIN2(pts[0], MIN2(pts[2], MIN2(pts[4], pts[6]))); + *minY = MIN2(pts[1], MIN2(pts[3], MIN2(pts[5], pts[7]))); + maxX = MAX2(pts[0], MAX2(pts[2], MAX2(pts[4], pts[6]))); + maxY = MAX2(pts[1], MAX2(pts[3], MAX2(pts[5], pts[7]))); + *width = maxX - *minX; + *height = maxY - *minY; + } +#else + { + struct path *dst = path_create(VG_PATH_DATATYPE_F, 1.0, 0, + 0, 0, VG_PATH_CAPABILITY_ALL); + path_transform(dst, p); + path_bounding_rect(dst, minX, minY, width, height); + path_destroy(dst); + } +#endif +} + + +void vgDrawPath(VGPath path, VGbitfield paintModes) +{ + struct vg_context *ctx = vg_current_context(); + + if (path == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + if (!(paintModes & (VG_STROKE_PATH | VG_FILL_PATH))) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (path_is_empty((struct path*)path)) + return; + path_render((struct path*)path, paintModes); +} + diff --git a/src/gallium/state_trackers/vega/api_text.c b/src/gallium/state_trackers/vega/api_text.c new file mode 100644 index 00000000000..d8411cf3e87 --- /dev/null +++ b/src/gallium/state_trackers/vega/api_text.c @@ -0,0 +1,258 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "VG/openvg.h" + +#include "vg_context.h" + +#include "util/u_memory.h" + +#ifdef OPENVG_VERSION_1_1 + +struct vg_font { + struct vg_object base; + + VGint glyph_indices[200]; + VGint num_glyphs; +}; + +VGFont vgCreateFont(VGint glyphCapacityHint) +{ + struct vg_font *font = 0; + struct vg_context *ctx = vg_current_context(); + + if (glyphCapacityHint < 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return VG_INVALID_HANDLE; + } + + font = CALLOC_STRUCT(vg_font); + vg_init_object(&font->base, ctx, VG_OBJECT_FONT); + vg_context_add_object(ctx, VG_OBJECT_FONT, font); + return (VGFont)font; +} + +void vgDestroyFont(VGFont f) +{ + struct vg_font *font = (struct vg_font *)f; + struct vg_context *ctx = vg_current_context(); + + if (f == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + vg_context_remove_object(ctx, VG_OBJECT_FONT, font); + /*free(font);*/ +} + +void vgSetGlyphToPath(VGFont font, + VGuint glyphIndex, + VGPath path, + VGboolean isHinted, + VGfloat glyphOrigin [2], + VGfloat escapement[2]) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_object *pathObj; + struct vg_font *f; + + if (font == VG_INVALID_HANDLE || + !vg_context_is_object_valid(ctx, VG_OBJECT_FONT, (void *)font)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (!glyphOrigin || !escapement || + !is_aligned(glyphOrigin) || !is_aligned(escapement)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (path != VG_INVALID_HANDLE && + !vg_context_is_object_valid(ctx, VG_OBJECT_PATH, (void *)path)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + pathObj = (struct vg_object*)path; + if (pathObj && pathObj->type != VG_OBJECT_PATH) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + + f = (struct vg_font*)font; + f->glyph_indices[f->num_glyphs] = glyphIndex; + ++f->num_glyphs; +} + +void vgSetGlyphToImage(VGFont font, + VGuint glyphIndex, + VGImage image, + VGfloat glyphOrigin [2], + VGfloat escapement[2]) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_object *img_obj; + struct vg_font *f; + + if (font == VG_INVALID_HANDLE || + !vg_context_is_object_valid(ctx, VG_OBJECT_FONT, (void *)font)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (!glyphOrigin || !escapement || + !is_aligned(glyphOrigin) || !is_aligned(escapement)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (image != VG_INVALID_HANDLE && + !vg_context_is_object_valid(ctx, VG_OBJECT_IMAGE, (void *)image)) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + img_obj = (struct vg_object*)image; + if (img_obj && img_obj->type != VG_OBJECT_IMAGE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + f = (struct vg_font*)font; + f->glyph_indices[f->num_glyphs] = glyphIndex; + ++f->num_glyphs; +} + +static INLINE VGboolean font_contains_glyph(struct vg_font *font, + VGuint glyph_index) +{ + VGint i; + for (i = 0; i < font->num_glyphs; ++i) { + if (font->glyph_indices[i] == glyph_index) { + return VG_TRUE; + } + } + return VG_FALSE; +} + +void vgClearGlyph(VGFont font, + VGuint glyphIndex) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_font *f; + VGint i; + + if (font == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (glyphIndex <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + f = (struct vg_font*)font; + if (!font_contains_glyph(f, glyphIndex)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + for (i = 0; i < f->num_glyphs; ++i) { + if (f->glyph_indices[i] == glyphIndex) { + /*FIXME*/ + f->glyph_indices[f->num_glyphs] = 0; + --f->num_glyphs; + return; + } + } +} + +void vgDrawGlyph(VGFont font, + VGuint glyphIndex, + VGbitfield paintModes, + VGboolean allowAutoHinting) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_font *f; + + if (font == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (glyphIndex <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (paintModes & (~(VG_STROKE_PATH|VG_FILL_PATH))) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + f = (struct vg_font*)font; + if (!font_contains_glyph(f, glyphIndex)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } +} + +void vgDrawGlyphs(VGFont font, + VGint glyphCount, + VGuint *glyphIndices, + VGfloat *adjustments_x, + VGfloat *adjustments_y, + VGbitfield paintModes, + VGboolean allowAutoHinting) +{ + struct vg_context *ctx = vg_current_context(); + VGint i; + struct vg_font *f; + + if (font == VG_INVALID_HANDLE) { + vg_set_error(ctx, VG_BAD_HANDLE_ERROR); + return; + } + if (glyphCount <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (!glyphIndices || !is_aligned(glyphIndices)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (!adjustments_x || !is_aligned(adjustments_x) || + !adjustments_y || !is_aligned(adjustments_y)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + if (paintModes & (~(VG_STROKE_PATH|VG_FILL_PATH))) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + f = (struct vg_font*)font; + for (i = 0; i < glyphCount; ++i) { + VGuint glyph_index = glyphIndices[i]; + if (!font_contains_glyph(f, glyph_index)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + } +} + +#endif diff --git a/src/gallium/state_trackers/vega/api_transform.c b/src/gallium/state_trackers/vega/api_transform.c new file mode 100644 index 00000000000..763a5ec415c --- /dev/null +++ b/src/gallium/state_trackers/vega/api_transform.c @@ -0,0 +1,128 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "VG/openvg.h" + +#include "vg_context.h" + +#include "matrix.h" + +void vgLoadIdentity(void) +{ + struct vg_context *ctx = vg_current_context(); + struct matrix *mat = vg_state_matrix(&ctx->state.vg); + matrix_load_identity(mat); +} + +void vgLoadMatrix(const VGfloat * m) +{ + struct vg_context *ctx = vg_current_context(); + struct matrix *mat; + + if (!ctx) + return; + + if (!m || !is_aligned(m)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + mat = vg_state_matrix(&ctx->state.vg); + matrix_init(mat, m); + if (!matrix_is_affine(mat)) { + if (ctx->state.vg.matrix_mode != VG_MATRIX_IMAGE_USER_TO_SURFACE) { + matrix_make_affine(mat); + } + } +} + +void vgGetMatrix(VGfloat * m) +{ + struct vg_context *ctx = vg_current_context(); + struct matrix *mat; + + if (!ctx) + return; + + if (!m || !is_aligned(m)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + mat = vg_state_matrix(&ctx->state.vg); + memcpy(m, mat->m, sizeof(VGfloat)*9); +} + +void vgMultMatrix(const VGfloat * m) +{ + struct vg_context *ctx = vg_current_context(); + struct matrix *dst, src; + + if (!ctx) + return; + + if (!m || !is_aligned(m)) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + matrix_init(&src, m); + dst = vg_state_matrix(&ctx->state.vg); + if (!matrix_is_affine(&src)) { + if (ctx->state.vg.matrix_mode != VG_MATRIX_IMAGE_USER_TO_SURFACE) { + matrix_make_affine(&src); + } + } + matrix_mult(dst, &src); + +} + +void vgTranslate(VGfloat tx, VGfloat ty) +{ + struct vg_context *ctx = vg_current_context(); + struct matrix *dst = vg_state_matrix(&ctx->state.vg); + matrix_translate(dst, tx, ty); +} + +void vgScale(VGfloat sx, VGfloat sy) +{ + struct vg_context *ctx = vg_current_context(); + struct matrix *dst = vg_state_matrix(&ctx->state.vg); + matrix_scale(dst, sx, sy); +} + +void vgShear(VGfloat shx, VGfloat shy) +{ + struct vg_context *ctx = vg_current_context(); + struct matrix *dst = vg_state_matrix(&ctx->state.vg); + matrix_shear(dst, shx, shy); +} + +void vgRotate(VGfloat angle) +{ + struct vg_context *ctx = vg_current_context(); + struct matrix *dst = vg_state_matrix(&ctx->state.vg); + matrix_rotate(dst, angle); +} diff --git a/src/gallium/state_trackers/vega/arc.c b/src/gallium/state_trackers/vega/arc.c new file mode 100644 index 00000000000..e74c7f03345 --- /dev/null +++ b/src/gallium/state_trackers/vega/arc.c @@ -0,0 +1,708 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "arc.h" + +#include "matrix.h" +#include "bezier.h" +#include "polygon.h" +#include "stroker.h" +#include "path.h" + +#include "util/u_debug.h" + +#include <math.h> + +#ifndef M_PI +#define M_PI 3.14159265358979323846 +#endif + +#define DEBUG_ARCS 0 + +static const VGfloat two_pi = M_PI * 2; + + +static const double coeffs3Low[2][4][4] = { + { + { 3.85268, -21.229, -0.330434, 0.0127842 }, + { -1.61486, 0.706564, 0.225945, 0.263682 }, + { -0.910164, 0.388383, 0.00551445, 0.00671814 }, + { -0.630184, 0.192402, 0.0098871, 0.0102527 } + }, + { + { -0.162211, 9.94329, 0.13723, 0.0124084 }, + { -0.253135, 0.00187735, 0.0230286, 0.01264 }, + { -0.0695069, -0.0437594, 0.0120636, 0.0163087 }, + { -0.0328856, -0.00926032, -0.00173573, 0.00527385 } + } +}; + +/* coefficients for error estimation + while using cubic Bézier curves for approximation + 1/4 <= b/a <= 1 */ +static const double coeffs3High[2][4][4] = { + { + { 0.0899116, -19.2349, -4.11711, 0.183362 }, + { 0.138148, -1.45804, 1.32044, 1.38474 }, + { 0.230903, -0.450262, 0.219963, 0.414038 }, + { 0.0590565, -0.101062, 0.0430592, 0.0204699 } + }, + { + { 0.0164649, 9.89394, 0.0919496, 0.00760802 }, + { 0.0191603, -0.0322058, 0.0134667, -0.0825018 }, + { 0.0156192, -0.017535, 0.00326508, -0.228157 }, + { -0.0236752, 0.0405821, -0.0173086, 0.176187 } + } +}; + +/* safety factor to convert the "best" error approximation + into a "max bound" error */ +static const double safety3[] = { + 0.001, 4.98, 0.207, 0.0067 +}; + +/* The code below is from the OpenVG 1.1 Spec + * Section 18.4 */ + +/* Given: Points (x0, y0) and (x1, y1) + * Return: TRUE if a solution exists, FALSE otherwise + * Circle centers are written to (cx0, cy0) and (cx1, cy1) + */ +static VGboolean +find_unit_circles(double x0, double y0, double x1, double y1, + double *cx0, double *cy0, + double *cx1, double *cy1) +{ + /* Compute differences and averages */ + double dx = x0 - x1; + double dy = y0 - y1; + double xm = (x0 + x1)/2; + double ym = (y0 + y1)/2; + double dsq, disc, s, sdx, sdy; + + /* Solve for intersecting unit circles */ + dsq = dx*dx + dy*dy; + if (dsq == 0.0) return VG_FALSE; /* Points are coincident */ + disc = 1.0/dsq - 1.0/4.0; + + /* the precision we care about here is around float so if we're + * around the float defined zero then make it official to avoid + * precision problems later on */ + if (floatIsZero(disc)) + disc = 0.0; + + if (disc < 0.0) return VG_FALSE; /* Points are too far apart */ + s = sqrt(disc); + sdx = s*dx; + sdy = s*dy; + *cx0 = xm + sdy; + *cy0 = ym - sdx; + *cx1 = xm - sdy; + *cy1 = ym + sdx; + return VG_TRUE; +} + + +/* Given: Ellipse parameters rh, rv, rot (in degrees), + * endpoints (x0, y0) and (x1, y1) + * Return: TRUE if a solution exists, FALSE otherwise + * Ellipse centers are written to (cx0, cy0) and (cx1, cy1) + */ +static VGboolean +find_ellipses(double rh, double rv, double rot, + double x0, double y0, double x1, double y1, + double *cx0, double *cy0, double *cx1, double *cy1) +{ + double COS, SIN, x0p, y0p, x1p, y1p, pcx0, pcy0, pcx1, pcy1; + /* Convert rotation angle from degrees to radians */ + rot *= M_PI/180.0; + /* Pre-compute rotation matrix entries */ + COS = cos(rot); SIN = sin(rot); + /* Transform (x0, y0) and (x1, y1) into unit space */ + /* using (inverse) rotate, followed by (inverse) scale */ + x0p = (x0*COS + y0*SIN)/rh; + y0p = (-x0*SIN + y0*COS)/rv; + x1p = (x1*COS + y1*SIN)/rh; + y1p = (-x1*SIN + y1*COS)/rv; + if (!find_unit_circles(x0p, y0p, x1p, y1p, + &pcx0, &pcy0, &pcx1, &pcy1)) { + return VG_FALSE; + } + /* Transform back to original coordinate space */ + /* using (forward) scale followed by (forward) rotate */ + pcx0 *= rh; pcy0 *= rv; + pcx1 *= rh; pcy1 *= rv; + *cx0 = pcx0*COS - pcy0*SIN; + *cy0 = pcx0*SIN + pcy0*COS; + *cx1 = pcx1*COS - pcy1*SIN; + *cy1 = pcx1*SIN + pcy1*COS; + return VG_TRUE; +} + +static INLINE VGboolean +try_to_fix_radii(struct arc *arc) +{ + double COS, SIN, rot, x0p, y0p, x1p, y1p; + double dx, dy, dsq, scale; + + /* Convert rotation angle from degrees to radians */ + rot = DEGREES_TO_RADIANS(arc->theta); + + /* Pre-compute rotation matrix entries */ + COS = cos(rot); SIN = sin(rot); + + /* Transform (x0, y0) and (x1, y1) into unit space */ + /* using (inverse) rotate, followed by (inverse) scale */ + x0p = (arc->x1*COS + arc->y1*SIN)/arc->a; + y0p = (-arc->x1*SIN + arc->y1*COS)/arc->b; + x1p = (arc->x2*COS + arc->y2*SIN)/arc->a; + y1p = (-arc->x2*SIN + arc->y2*COS)/arc->b; + /* Compute differences and averages */ + dx = x0p - x1p; + dy = y0p - y1p; + + dsq = dx*dx + dy*dy; +#if 0 + if (dsq <= 0.001) { + debug_printf("AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAaaaaa\n"); + } +#endif + scale = 1/(2/sqrt(dsq)); + arc->a *= scale; + arc->b *= scale; + return VG_TRUE; +} + +static INLINE double vector_normalize(double *v) +{ + double sq = v[0] * v[0] + v[1] * v[1]; + return sqrt(sq); +} +static INLINE double vector_orientation(double *v) +{ + double norm = vector_normalize(v); + double cosa = v[0] / norm; + double sina = v[1] / norm; + return (sina>=0 ? acos(cosa) : 2*M_PI - acos(cosa)); +} +static INLINE double vector_dot(double *v0, + double *v1) +{ + return v0[0] * v1[0] + v0[1] * v1[1]; +} + +static INLINE double vector_angles(double *v0, + double *v1) +{ + double dot = vector_dot(v0, v1); + double norm0 = vector_normalize(v0); + double norm1 = vector_normalize(v1); + + return acos(dot / (norm0 * norm1)); +} + +static VGboolean find_angles(struct arc *arc) +{ + double vec0[2], vec1[2]; + double lambda1, lambda2; + double angle; + struct matrix matrix; + + if (floatIsZero(arc->a) || floatIsZero(arc->b)) { + return VG_FALSE; + } + /* map the points to an identity circle */ + matrix_load_identity(&matrix); + matrix_scale(&matrix, 1.f, arc->a/arc->b); + matrix_rotate(&matrix, -arc->theta); + matrix_map_point(&matrix, + arc->x1, arc->y1, + &arc->x1, &arc->y1); + matrix_map_point(&matrix, + arc->x2, arc->y2, + &arc->x2, &arc->y2); + matrix_map_point(&matrix, + arc->cx, arc->cy, + &arc->cx, &arc->cy); + +#if DEBUG_ARCS + debug_printf("Matrix 3 [%f, %f, %f| %f, %f, %f| %f, %f, %f]\n", + matrix.m[0], matrix.m[1], matrix.m[2], + matrix.m[3], matrix.m[4], matrix.m[5], + matrix.m[6], matrix.m[7], matrix.m[8]); + debug_printf("Endpoints [%f, %f], [%f, %f]\n", + arc->x1, arc->y1, arc->x2, arc->y2); +#endif + + vec0[0] = arc->x1 - arc->cx; + vec0[1] = arc->y1 - arc->cy; + vec1[0] = arc->x2 - arc->cx; + vec1[1] = arc->y2 - arc->cy; + +#if DEBUG_ARCS + debug_printf("Vec is [%f, %f], [%f, %f], [%f, %f]\n", + vec0[0], vec0[1], vec1[0], vec1[1], arc->cx, arc->cy); +#endif + + lambda1 = vector_orientation(vec0); + + if (isnan(lambda1)) + lambda1 = 0.f; + + if (arc->type == VG_SCWARC_TO || + arc->type == VG_SCCWARC_TO) + angle = vector_angles(vec0, vec1); + else if (arc->type == VG_LCWARC_TO || + arc->type == VG_LCCWARC_TO) { + angle = 2*M_PI - vector_angles(vec0, vec1); + } else + abort(); + + if (isnan(angle)) + angle = M_PI; + + + if (arc->type == VG_SCWARC_TO || + arc->type == VG_LCWARC_TO) + lambda2 = lambda1 - angle; + else + lambda2 = lambda1 + angle; + +#if DEBUG_ARCS + debug_printf("Angle is %f and (%f, %f)\n", angle, lambda1, lambda2); +#endif + +#if 0 + arc->eta1 = atan2(sin(lambda1) / arc->b, + cos(lambda1) / arc->a); + arc->eta2 = atan2(sin(lambda2) / arc->b, + cos(lambda2) / arc->a); + + /* make sure we have eta1 <= eta2 <= eta1 + 2 PI */ + arc->eta2 -= two_pi * floor((arc->eta2 - arc->eta1) / two_pi); + + /* the preceding correction fails if we have exactly et2 - eta1 = 2 PI + it reduces the interval to zero length */ + if ((lambda2 - lambda1 > M_PI) && (arc->eta2 - arc->eta1 < M_PI)) { + arc->eta2 += 2 * M_PI; + } +#else + arc->eta1 = lambda1; + arc->eta2 = lambda2; +#endif + + return VG_TRUE; +} + +#if DEBUG_ARCS +static void check_endpoints(struct arc *arc) +{ + double x1, y1, x2, y2; + + double a_cos_eta1 = arc->a * cos(arc->eta1); + double b_sin_eta1 = arc->b * sin(arc->eta1); + x1 = arc->cx + a_cos_eta1 * arc->cos_theta - + b_sin_eta1 * arc->sin_theta; + y1 = arc->cy + a_cos_eta1 * arc->sin_theta + + b_sin_eta1 * arc->cos_theta; + + double a_cos_eta2 = arc->a * cos(arc->eta2); + double b_sin_eta2 = arc->b * sin(arc->eta2); + x2 = arc->cx + a_cos_eta2 * arc->cos_theta - + b_sin_eta2 * arc->sin_theta; + y2 = arc->cy + a_cos_eta2 * arc->sin_theta + + b_sin_eta2 * arc->cos_theta; + + debug_printf("Computed (%f, %f), (%f, %f)\n", + x1, y1, x2, y2); + debug_printf("Real (%f, %f), (%f, %f)\n", + arc->x1, arc->y1, + arc->x2, arc->y2); +} +#endif + +void arc_init(struct arc *arc, + VGPathSegment type, + VGfloat x1, VGfloat y1, + VGfloat x2, VGfloat y2, + VGfloat rh, VGfloat rv, + VGfloat rot) +{ + assert(type == VG_SCCWARC_TO || + type == VG_SCWARC_TO || + type == VG_LCCWARC_TO || + type == VG_LCWARC_TO); + arc->type = type; + arc->x1 = x1; + arc->y1 = y1; + arc->x2 = x2; + arc->y2 = y2; + arc->a = rh; + arc->b = rv; + arc->theta = rot; + arc->cos_theta = cos(arc->theta); + arc->sin_theta = sin(arc->theta); + { + double cx0, cy0, cx1, cy1; + double cx, cy; + arc->is_valid = find_ellipses(rh, rv, rot, x1, y1, x2, y2, + &cx0, &cy0, &cx1, &cy1); + + if (!arc->is_valid && try_to_fix_radii(arc)) { + rh = arc->a; + rv = arc->b; + arc->is_valid = + find_ellipses(rh, rv, rot, x1, y1, x2, y2, + &cx0, &cy0, &cx1, &cy1); + } + + if (type == VG_SCWARC_TO || + type == VG_LCCWARC_TO) { + cx = cx1; + cy = cy1; + } else { + cx = cx0; + cy = cy0; + } +#if DEBUG_ARCS + debug_printf("Centers are : (%f, %f) , (%f, %f). Real (%f, %f)\n", + cx0, cy0, cx1, cy1, cx, cy); +#endif + arc->cx = cx; + arc->cy = cy; + if (arc->is_valid) { + arc->is_valid = find_angles(arc); +#if DEBUG_ARCS + check_endpoints(arc); +#endif + /* remap a few points. find_angles requires + * rot in angles, the rest of the code + * will need them in radians. and find_angles + * modifies the center to match an identity + * circle so lets reset it */ + arc->theta = DEGREES_TO_RADIANS(rot); + arc->cos_theta = cos(arc->theta); + arc->sin_theta = sin(arc->theta); + arc->cx = cx; + arc->cy = cy; + } + } +} + +static INLINE double rational_function(double x, const double *c) +{ + return (x * (x * c[0] + c[1]) + c[2]) / (x + c[3]); +} + +static double estimate_error(struct arc *arc, + double etaA, double etaB) +{ + double eta = 0.5 * (etaA + etaB); + + double x = arc->b / arc->a; + double dEta = etaB - etaA; + double cos2 = cos(2 * eta); + double cos4 = cos(4 * eta); + double cos6 = cos(6 * eta); + double c0, c1; + + /* select the right coeficients set according to degree and b/a */ + const double (*coeffs)[4][4]; + const double *safety; + coeffs = (x < 0.25) ? coeffs3Low : coeffs3High; + safety = safety3; + + c0 = rational_function(x, coeffs[0][0]) + + cos2 * rational_function(x, coeffs[0][1]) + + cos4 * rational_function(x, coeffs[0][2]) + + cos6 * rational_function(x, coeffs[0][3]); + + c1 = rational_function(x, coeffs[1][0]) + + cos2 * rational_function(x, coeffs[1][1]) + + cos4 * rational_function(x, coeffs[1][2]) + + cos6 * rational_function(x, coeffs[1][3]); + + return rational_function(x, safety) * arc->a * exp(c0 + c1 * dEta); +} + +struct arc_cb { + void (*move)(struct arc_cb *cb, VGfloat x, VGfloat y); + void (*point)(struct arc_cb *cb, VGfloat x, VGfloat y); + void (*bezier)(struct arc_cb *cb, struct bezier *bezier); + + void *user_data; +}; + +static void cb_null_move(struct arc_cb *cb, VGfloat x, VGfloat y) +{ +} + +static void polygon_point(struct arc_cb *cb, VGfloat x, VGfloat y) +{ + struct polygon *poly = (struct polygon*)cb->user_data; + polygon_vertex_append(poly, x, y); +} + +static void polygon_bezier(struct arc_cb *cb, struct bezier *bezier) +{ + struct polygon *poly = (struct polygon*)cb->user_data; + bezier_add_to_polygon(bezier, poly); +} + +static void stroke_point(struct arc_cb *cb, VGfloat x, VGfloat y) +{ + struct stroker *stroker = (struct stroker*)cb->user_data; + stroker_line_to(stroker, x, y); +} + +static void stroke_curve(struct arc_cb *cb, struct bezier *bezier) +{ + struct stroker *stroker = (struct stroker*)cb->user_data; + stroker_curve_to(stroker, + bezier->x2, bezier->y2, + bezier->x3, bezier->y3, + bezier->x4, bezier->y4); +} + +static void stroke_emit_point(struct arc_cb *cb, VGfloat x, VGfloat y) +{ + struct stroker *stroker = (struct stroker*)cb->user_data; + stroker_emit_line_to(stroker, x, y); +} + +static void stroke_emit_curve(struct arc_cb *cb, struct bezier *bezier) +{ + struct stroker *stroker = (struct stroker*)cb->user_data; + stroker_emit_curve_to(stroker, + bezier->x2, bezier->y2, + bezier->x3, bezier->y3, + bezier->x4, bezier->y4); +} + +static void arc_path_move(struct arc_cb *cb, VGfloat x, VGfloat y) +{ + struct path *path = (struct path*)cb->user_data; + path_move_to(path, x, y); +} + +static void arc_path_point(struct arc_cb *cb, VGfloat x, VGfloat y) +{ + struct path *path = (struct path*)cb->user_data; + path_line_to(path, x, y); +} + +static void arc_path_bezier(struct arc_cb *cb, struct bezier *bezier) +{ + struct path *path = (struct path*)cb->user_data; + path_cubic_to(path, + bezier->x2, bezier->y2, + bezier->x3, bezier->y3, + bezier->x4, bezier->y4); +} + +static INLINE int num_beziers_needed(struct arc *arc) +{ + double threshold = 0.05; + VGboolean found = VG_FALSE; + int n = 1; + double min_eta, max_eta; + + min_eta = MIN2(arc->eta1, arc->eta2); + max_eta = MAX2(arc->eta1, arc->eta2); + + while ((! found) && (n < 1024)) { + double d_eta = (max_eta - min_eta) / n; + if (d_eta <= 0.5 * M_PI) { + double eta_b = min_eta; + found = VG_TRUE; + for (int i = 0; found && (i < n); ++i) { + double etaA = eta_b; + eta_b += d_eta; + found = (estimate_error(arc, etaA, eta_b) <= threshold); + } + } + n = n << 1; + } + + return n; +} + +static void arc_to_beziers(struct arc *arc, + struct arc_cb cb, + struct matrix *matrix) +{ + int n = 1; + double d_eta, eta_b, cos_eta_b, + sin_eta_b, a_cos_eta_b, b_sin_eta_b, a_sin_eta_b, + b_cos_eta_b, x_b, y_b, x_b_dot, y_b_dot, lx, ly; + double t, alpha; + + { /* always move to the start of the arc */ + VGfloat x = arc->x1; + VGfloat y = arc->y1; + matrix_map_point(matrix, x, y, &x, &y); + cb.move(&cb, x, y); + } + + if (!arc->is_valid) { + VGfloat x = arc->x2; + VGfloat y = arc->y2; + matrix_map_point(matrix, x, y, &x, &y); + cb.point(&cb, x, y); + return; + } + + /* find the number of Bézier curves needed */ + n = num_beziers_needed(arc); + + d_eta = (arc->eta2 - arc->eta1) / n; + eta_b = arc->eta1; + + cos_eta_b = cos(eta_b); + sin_eta_b = sin(eta_b); + a_cos_eta_b = arc->a * cos_eta_b; + b_sin_eta_b = arc->b * sin_eta_b; + a_sin_eta_b = arc->a * sin_eta_b; + b_cos_eta_b = arc->b * cos_eta_b; + x_b = arc->cx + a_cos_eta_b * arc->cos_theta - + b_sin_eta_b * arc->sin_theta; + y_b = arc->cy + a_cos_eta_b * arc->sin_theta + + b_sin_eta_b * arc->cos_theta; + x_b_dot = -a_sin_eta_b * arc->cos_theta - + b_cos_eta_b * arc->sin_theta; + y_b_dot = -a_sin_eta_b * arc->sin_theta + + b_cos_eta_b * arc->cos_theta; + + { + VGfloat x = x_b, y = y_b; + matrix_map_point(matrix, x, y, &x, &y); + cb.point(&cb, x, y); + } + lx = x_b; + ly = y_b; + + t = tan(0.5 * d_eta); + alpha = sin(d_eta) * (sqrt(4 + 3 * t * t) - 1) / 3; + + for (int i = 0; i < n; ++i) { + struct bezier bezier; + double xA = x_b; + double yA = y_b; + double xADot = x_b_dot; + double yADot = y_b_dot; + + eta_b += d_eta; + cos_eta_b = cos(eta_b); + sin_eta_b = sin(eta_b); + a_cos_eta_b = arc->a * cos_eta_b; + b_sin_eta_b = arc->b * sin_eta_b; + a_sin_eta_b = arc->a * sin_eta_b; + b_cos_eta_b = arc->b * cos_eta_b; + x_b = arc->cx + a_cos_eta_b * arc->cos_theta - + b_sin_eta_b * arc->sin_theta; + y_b = arc->cy + a_cos_eta_b * arc->sin_theta + + b_sin_eta_b * arc->cos_theta; + x_b_dot = -a_sin_eta_b * arc->cos_theta - + b_cos_eta_b * arc->sin_theta; + y_b_dot = -a_sin_eta_b * arc->sin_theta + + b_cos_eta_b * arc->cos_theta; + + bezier_init(&bezier, + lx, ly, + (float) (xA + alpha * xADot), (float) (yA + alpha * yADot), + (float) (x_b - alpha * x_b_dot), (float) (y_b - alpha * y_b_dot), + (float) x_b, (float) y_b); +#if 0 + debug_printf("%d) Bezier (%f, %f), (%f, %f), (%f, %f), (%f, %f)\n", + i, + bezier.x1, bezier.y1, + bezier.x2, bezier.y2, + bezier.x3, bezier.y3, + bezier.x4, bezier.y4); +#endif + bezier_transform(&bezier, matrix); + cb.bezier(&cb, &bezier); + lx = x_b; + ly = y_b; + } +} + + +void arc_add_to_polygon(struct arc *arc, + struct polygon *poly, + struct matrix *matrix) +{ + struct arc_cb cb; + + cb.move = cb_null_move; + cb.point = polygon_point; + cb.bezier = polygon_bezier; + cb.user_data = poly; + + arc_to_beziers(arc, cb, matrix); +} + +void arc_stroke_cb(struct arc *arc, + struct stroker *stroke, + struct matrix *matrix) +{ + struct arc_cb cb; + + cb.move = cb_null_move; + cb.point = stroke_point; + cb.bezier = stroke_curve; + cb.user_data = stroke; + + arc_to_beziers(arc, cb, matrix); +} + +void arc_stroker_emit(struct arc *arc, + struct stroker *stroker, + struct matrix *matrix) +{ + struct arc_cb cb; + + cb.move = cb_null_move; + cb.point = stroke_emit_point; + cb.bezier = stroke_emit_curve; + cb.user_data = stroker; + + arc_to_beziers(arc, cb, matrix); +} + +void arc_to_path(struct arc *arc, + struct path *path, + struct matrix *matrix) +{ + struct arc_cb cb; + + cb.move = arc_path_move; + cb.point = arc_path_point; + cb.bezier = arc_path_bezier; + cb.user_data = path; + + arc_to_beziers(arc, cb, matrix); +} diff --git a/src/gallium/state_trackers/vega/arc.h b/src/gallium/state_trackers/vega/arc.h new file mode 100644 index 00000000000..3205cd5021d --- /dev/null +++ b/src/gallium/state_trackers/vega/arc.h @@ -0,0 +1,80 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef ARC_H +#define ARC_H + +#include "VG/openvg.h" + +struct polygon; +struct matrix; +struct stroker; +struct path; + +struct arc { + VGPathSegment type; + + VGfloat cx, cy; + + VGfloat a, b; + + VGfloat theta; + VGfloat cos_theta, sin_theta; + + VGfloat eta1; + VGfloat eta2; + + VGfloat x1, y1, x2, y2; + + VGboolean is_valid; +}; + +void arc_init(struct arc *arc, + VGPathSegment type, + VGfloat x1, VGfloat y1, + VGfloat x2, VGfloat y2, + VGfloat rh, VGfloat rv, + VGfloat rot); + +void arc_add_to_polygon(struct arc *arc, + struct polygon *poly, + struct matrix *matrix); + + +void arc_to_path(struct arc *arc, + struct path *p, + struct matrix *matrix); + +void arc_stroke_cb(struct arc *arc, + struct stroker *stroke, + struct matrix *matrix); + +void arc_stroker_emit(struct arc *arc, + struct stroker *stroke, + struct matrix *matrix); + + +#endif diff --git a/src/gallium/state_trackers/vega/asm_fill.h b/src/gallium/state_trackers/vega/asm_fill.h new file mode 100644 index 00000000000..2f394ad6c55 --- /dev/null +++ b/src/gallium/state_trackers/vega/asm_fill.h @@ -0,0 +1,246 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef ASM_FILL_H +#define ASM_FILL_H + +static const char solid_fill_asm[] = + "MOV %s, CONST[0]\n"; + + +static const char linear_grad_asm[] = + "MOV TEMP[0].xy, IN[0]\n" + "MOV TEMP[0].z, CONST[1].yyyy\n" + "DP3 TEMP[1], CONST[2], TEMP[0]\n" + "DP3 TEMP[2], CONST[3], TEMP[0]\n" + "DP3 TEMP[3], CONST[4], TEMP[0]\n" + "RCP TEMP[3], TEMP[3]\n" + "MUL TEMP[1], TEMP[1], TEMP[3]\n" + "MUL TEMP[2], TEMP[2], TEMP[3]\n" + "MOV TEMP[4].x, TEMP[1]\n" + "MOV TEMP[4].y, TEMP[2]\n" + "MUL TEMP[0], CONST[0].yyyy, TEMP[4].yyyy\n" + "MAD TEMP[1], CONST[0].xxxx, TEMP[4].xxxx, TEMP[0]\n" + "MUL TEMP[2], TEMP[1], CONST[0].zzzz\n" + "TEX %s, TEMP[2], SAMP[0], 1D\n"; + +static const char radial_grad_asm[] = + "MOV TEMP[0].xy, IN[0]\n" + "MOV TEMP[0].z, CONST[1].yyyy\n" + "DP3 TEMP[1], CONST[2], TEMP[0]\n" + "DP3 TEMP[2], CONST[3], TEMP[0]\n" + "DP3 TEMP[3], CONST[4], TEMP[0]\n" + "RCP TEMP[3], TEMP[3]\n" + "MUL TEMP[1], TEMP[1], TEMP[3]\n" + "MUL TEMP[2], TEMP[2], TEMP[3]\n" + "MOV TEMP[5].x, TEMP[1]\n" + "MOV TEMP[5].y, TEMP[2]\n" + "MUL TEMP[0], CONST[0].yyyy, TEMP[5].yyyy\n" + "MAD TEMP[1], CONST[0].xxxx, TEMP[5].xxxx, TEMP[0]\n" + "ADD TEMP[1], TEMP[1], TEMP[1]\n" + "MUL TEMP[3], TEMP[5].yyyy, TEMP[5].yyyy\n" + "MAD TEMP[4], TEMP[5].xxxx, TEMP[5].xxxx, TEMP[3]\n" + "MOV TEMP[4], -TEMP[4]\n" + "MUL TEMP[2], CONST[0].zzzz, TEMP[4]\n" + "MUL TEMP[0], CONST[1].wwww, TEMP[2]\n" + "MUL TEMP[3], TEMP[1], TEMP[1]\n" + "SUB TEMP[2], TEMP[3], TEMP[0]\n" + "RSQ TEMP[2], |TEMP[2]|\n" + "RCP TEMP[2], TEMP[2]\n" + "SUB TEMP[1], TEMP[2], TEMP[1]\n" + "ADD TEMP[0], CONST[0].zzzz, CONST[0].zzzz\n" + "RCP TEMP[0], TEMP[0]\n" + "MUL TEMP[2], TEMP[1], TEMP[0]\n" + "TEX %s, TEMP[2], SAMP[0], 1D\n"; + +static const char pattern_asm[] = + "MOV TEMP[0].xy, IN[0]\n" + "MOV TEMP[0].z, CONST[1].yyyy\n" + "DP3 TEMP[1], CONST[2], TEMP[0]\n" + "DP3 TEMP[2], CONST[3], TEMP[0]\n" + "DP3 TEMP[3], CONST[4], TEMP[0]\n" + "RCP TEMP[3], TEMP[3]\n" + "MUL TEMP[1], TEMP[1], TEMP[3]\n" + "MUL TEMP[2], TEMP[2], TEMP[3]\n" + "MOV TEMP[4].x, TEMP[1]\n" + "MOV TEMP[4].y, TEMP[2]\n" + "RCP TEMP[0], CONST[1].zwzw\n" + "MOV TEMP[1], TEMP[4]\n" + "MUL TEMP[1].x, TEMP[1], TEMP[0]\n" + "MUL TEMP[1].y, TEMP[1], TEMP[0]\n" + "TEX %s, TEMP[1], SAMP[0], 2D\n"; + + +static const char mask_asm[] = + "TEX TEMP[1], IN[0], SAMP[1], 2D\n" + "MUL TEMP[0].w, TEMP[0].wwww, TEMP[1].wwww\n" + "MOV %s, TEMP[0]\n"; + + +static const char image_normal_asm[] = + "TEX %s, IN[1], SAMP[3], 2D\n"; + +static const char image_multiply_asm[] = + "TEX TEMP[1], IN[1], SAMP[3], 2D\n" + "MUL %s, TEMP[0], TEMP[1]\n"; + +static const char image_stencil_asm[] = + "TEX TEMP[1], IN[1], SAMP[3], 2D\n" + "MUL %s, TEMP[0], TEMP[1]\n"; + + +#define EXTENDED_BLEND_OVER \ + "SUB TEMP[3], CONST[1].yyyy, TEMP[1].wwww\n" \ + "SUB TEMP[4], CONST[1].yyyy, TEMP[0].wwww\n" \ + "MUL TEMP[3], TEMP[0], TEMP[3]\n" \ + "MUL TEMP[4], TEMP[1], TEMP[4]\n" \ + "ADD TEMP[3], TEMP[3], TEMP[4]\n" + +static const char blend_multiply_asm[] = + "TEX TEMP[1], IN[0], SAMP[2], 2D\n" + EXTENDED_BLEND_OVER + "MUL TEMP[4], TEMP[0], TEMP[1]\n" + "ADD TEMP[1], TEMP[4], TEMP[3]\n"/*result.rgb*/ + "MUL TEMP[2], TEMP[0].wwww, TEMP[1].wwww\n" + "ADD TEMP[3], TEMP[0].wwww, TEMP[1].wwww\n" + "SUB TEMP[1].w, TEMP[3], TEMP[2]\n" + "MOV %s, TEMP[1]\n"; +#if 1 +static const char blend_screen_asm[] = + "TEX TEMP[1], IN[0], SAMP[2], 2D\n" + "ADD TEMP[3], TEMP[0], TEMP[1]\n" + "MUL TEMP[2], TEMP[0], TEMP[1]\n" + "SUB %s, TEMP[3], TEMP[2]\n"; +#else +static const char blend_screen_asm[] = + "TEX TEMP[1], IN[0], SAMP[2], 2D\n" + "MOV %s, TEMP[1]\n"; +#endif + +static const char blend_darken_asm[] = + "TEX TEMP[1], IN[0], SAMP[2], 2D\n" + EXTENDED_BLEND_OVER + "MUL TEMP[4], TEMP[0], TEMP[1].wwww\n" + "MUL TEMP[5], TEMP[1], TEMP[0].wwww\n" + "MIN TEMP[4], TEMP[4], TEMP[5]\n" + "ADD TEMP[1], TEMP[3], TEMP[4]\n" + "MUL TEMP[2], TEMP[0].wwww, TEMP[1].wwww\n" + "ADD TEMP[3], TEMP[0].wwww, TEMP[1].wwww\n" + "SUB TEMP[1].w, TEMP[3], TEMP[2]\n" + "MOV %s, TEMP[1]\n"; + +static const char blend_lighten_asm[] = + "TEX TEMP[1], IN[0], SAMP[2], 2D\n" + EXTENDED_BLEND_OVER + "MUL TEMP[4], TEMP[0], TEMP[1].wwww\n" + "MUL TEMP[5], TEMP[1], TEMP[0].wwww\n" + "MAX TEMP[4], TEMP[4], TEMP[5]\n" + "ADD TEMP[1], TEMP[3], TEMP[4]\n" + "MUL TEMP[2], TEMP[0].wwww, TEMP[1].wwww\n" + "ADD TEMP[3], TEMP[0].wwww, TEMP[1].wwww\n" + "SUB TEMP[1].w, TEMP[3], TEMP[2]\n" + "MOV %s, TEMP[1]\n"; + + +static const char premultiply_asm[] = + "MUL TEMP[0].xyz, TEMP[0], TEMP[0].wwww\n"; + +static const char unpremultiply_asm[] = + "TEX TEMP[0], IN[0], SAMP[1], 2D\n"; + + +static const char color_bw_asm[] = + "ADD TEMP[1], CONST[1].yyyy, CONST[1].yyyy\n" + "RCP TEMP[2], TEMP[1]\n" + "ADD TEMP[1], CONST[1].yyyy, TEMP[2]\n" + "ADD TEMP[2].x, TEMP[0].xxxx, TEMP[0].yyyy\n" + "ADD TEMP[2].x, TEMP[0].zzzz, TEMP[0].xxxx\n" + "SGE TEMP[0].xyz, TEMP[2].xxxx, TEMP[1]\n" + "SGE TEMP[0].w, TEMP[0].wwww, TEMP[2].yyyy\n" + "MOV %s, TEMP[0]\n"; + + +struct shader_asm_info { + VGint id; + VGint num_tokens; + const char * txt; + + VGboolean needs_position; + + VGint start_const; + VGint num_consts; + + VGint start_sampler; + VGint num_samplers; + + VGint start_temp; + VGint num_temps; +}; + + +static const struct shader_asm_info shaders_asm[] = { + /* fills */ + {VEGA_SOLID_FILL_SHADER, 40, solid_fill_asm, + VG_FALSE, 0, 1, 0, 0, 0, 0}, + {VEGA_LINEAR_GRADIENT_SHADER, 200, linear_grad_asm, + VG_TRUE, 0, 5, 0, 1, 0, 5}, + {VEGA_RADIAL_GRADIENT_SHADER, 200, radial_grad_asm, + VG_TRUE, 0, 5, 0, 1, 0, 6}, + {VEGA_PATTERN_SHADER, 100, pattern_asm, + VG_TRUE, 1, 4, 0, 1, 0, 5}, + + /* image draw modes */ + {VEGA_IMAGE_NORMAL_SHADER, 200, image_normal_asm, + VG_TRUE, 0, 0, 3, 1, 0, 0}, + {VEGA_IMAGE_MULTIPLY_SHADER, 200, image_multiply_asm, + VG_TRUE, 0, 0, 3, 1, 0, 2}, + {VEGA_IMAGE_STENCIL_SHADER, 200, image_stencil_asm, + VG_TRUE, 0, 0, 3, 1, 0, 2}, + + {VEGA_MASK_SHADER, 100, mask_asm, + VG_TRUE, 0, 0, 1, 1, 0, 2}, + + /* extra blend modes */ + {VEGA_BLEND_MULTIPLY_SHADER, 200, blend_multiply_asm, + VG_TRUE, 1, 1, 2, 1, 0, 5}, + {VEGA_BLEND_SCREEN_SHADER, 200, blend_screen_asm, + VG_TRUE, 0, 0, 2, 1, 0, 4}, + {VEGA_BLEND_DARKEN_SHADER, 200, blend_darken_asm, + VG_TRUE, 1, 1, 2, 1, 0, 6}, + {VEGA_BLEND_LIGHTEN_SHADER, 200, blend_lighten_asm, + VG_TRUE, 1, 1, 2, 1, 0, 6}, + + /* premultiply */ + {VEGA_PREMULTIPLY_SHADER, 100, premultiply_asm, + VG_FALSE, 0, 0, 0, 0, 0, 1}, + {VEGA_UNPREMULTIPLY_SHADER, 100, unpremultiply_asm, + VG_FALSE, 0, 0, 0, 0, 0, 1}, + + /* color transform to black and white */ + {VEGA_BW_SHADER, 150, color_bw_asm, + VG_FALSE, 1, 1, 0, 0, 0, 3}, +}; +#endif diff --git a/src/gallium/state_trackers/vega/asm_filters.h b/src/gallium/state_trackers/vega/asm_filters.h new file mode 100644 index 00000000000..9a49f2e12d0 --- /dev/null +++ b/src/gallium/state_trackers/vega/asm_filters.h @@ -0,0 +1,117 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef ASM_FILTERS_H +#define ASM_FILTERS_H + +static const char color_matrix_asm[] = + "FRAG1.1\n" + "DCL IN[0], GENERIC[0], PERSPECTIVE\n" + "DCL OUT[0], COLOR, CONSTANT\n" + "DCL CONST[0..4], CONSTANT\n" + "DCL TEMP[0..4], CONSTANT\n" + "DCL SAMP[0], CONSTANT\n" + "TEX TEMP[0], IN[0], SAMP[0], 2D\n" + "MOV TEMP[1], TEMP[0].xxxx\n" + "MOV TEMP[2], TEMP[0].yyyy\n" + "MOV TEMP[3], TEMP[0].zzzz\n" + "MOV TEMP[4], TEMP[0].wwww\n" + "MUL TEMP[1], TEMP[1], CONST[0]\n" + "MUL TEMP[2], TEMP[2], CONST[1]\n" + "MUL TEMP[3], TEMP[3], CONST[2]\n" + "MUL TEMP[4], TEMP[4], CONST[3]\n" + "ADD TEMP[0], TEMP[1], CONST[4]\n" + "ADD TEMP[0], TEMP[0], TEMP[2]\n" + "ADD TEMP[0], TEMP[0], TEMP[3]\n" + "ADD TEMP[0], TEMP[0], TEMP[4]\n" + "MOV OUT[0], TEMP[0]\n" + "END\n"; + +static const char convolution_asm[] = + "FRAG1.1\n" + "DCL IN[0], GENERIC[0], PERSPECTIVE\n" + "DCL OUT[0], COLOR, CONSTANT\n" + "DCL TEMP[0..4], CONSTANT\n" + "DCL ADDR[0], CONSTANT\n" + "DCL CONST[0..%d], CONSTANT\n" + "DCL SAMP[0], CONSTANT\n" + "0: MOV TEMP[0], CONST[0].xxxx\n" + "1: MOV TEMP[1], CONST[0].xxxx\n" + "2: BGNLOOP :14\n" + "3: SGE TEMP[0].z, TEMP[0].yyyy, CONST[1].xxxx\n" + "4: IF TEMP[0].zzzz :7\n" + "5: BRK\n" + "6: ENDIF\n" + "7: ARL ADDR[0].x, TEMP[0].yyyy\n" + "8: MOV TEMP[3], CONST[ADDR[0]+2]\n" + "9: ADD TEMP[4].xy, IN[0], TEMP[3]\n" + "10: TEX TEMP[2], TEMP[4], SAMP[0], 2D\n" + "11: MOV TEMP[3], CONST[ADDR[0]+%d]\n" + "12: MAD TEMP[1], TEMP[2], TEMP[3], TEMP[1]\n" + "13: ADD TEMP[0].y, TEMP[0].yyyy, CONST[0].yyyy\n" + "14: ENDLOOP :2\n" + "15: MAD OUT[0], TEMP[1], CONST[1].yyyy, CONST[1].zzzz\n" + "16: END\n"; + + +static const char lookup_asm[] = + "FRAG1.1\n" + "DCL IN[0], GENERIC[0], PERSPECTIVE\n" + "DCL OUT[0], COLOR, CONSTANT\n" + "DCL TEMP[0..2], CONSTANT\n" + "DCL CONST[0], CONSTANT\n" + "DCL SAMP[0..1], CONSTANT\n" + "TEX TEMP[0], IN[0], SAMP[0], 2D\n" + "MOV TEMP[1], TEMP[0]\n" + /* do red */ + "TEX TEMP[2], TEMP[1].xxxx, SAMP[1], 1D\n" + "MOV TEMP[0].x, TEMP[2].xxxx\n" + /* do blue */ + "TEX TEMP[2], TEMP[1].yyyy, SAMP[1], 1D\n" + "MOV TEMP[0].y, TEMP[2].yyyy\n" + /* do green */ + "TEX TEMP[2], TEMP[1].zzzz, SAMP[1], 1D\n" + "MOV TEMP[0].z, TEMP[2].zzzz\n" + /* do alpha */ + "TEX TEMP[2], TEMP[1].wwww, SAMP[1], 1D\n" + "MOV TEMP[0].w, TEMP[2].wwww\n" + "MOV OUT[0], TEMP[0]\n" + "END\n"; + + +static const char lookup_single_asm[] = + "FRAG1.1\n" + "DCL IN[0], GENERIC[0], PERSPECTIVE\n" + "DCL OUT[0], COLOR, CONSTANT\n" + "DCL TEMP[0..2], CONSTANT\n" + "DCL CONST[0], CONSTANT\n" + "DCL SAMP[0..1], CONSTANT\n" + "TEX TEMP[0], IN[0], SAMP[0], 2D\n" + "TEX TEMP[1], TEMP[0].%s, SAMP[1], 1D\n" + "MOV OUT[0], TEMP[1]\n" + "END\n"; + +#endif diff --git a/src/gallium/state_trackers/vega/asm_util.h b/src/gallium/state_trackers/vega/asm_util.h new file mode 100644 index 00000000000..218e1d166db --- /dev/null +++ b/src/gallium/state_trackers/vega/asm_util.h @@ -0,0 +1,136 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef ASM_UTIL_H +#define ASM_UTIL_H + + +static const char pass_through_depth_asm[] = + "FRAG1.1\n" + "DCL IN[0], POSITION, LINEAR\n" + "DCL OUT[0].z, POSITION, CONSTANT\n" + "0: MOV OUT[0].z, IN[0].zzzz\n" + "1: END\n"; + + + +/* μnew = μmask */ +static const char set_mask_asm[] = + "FRAG1.1\n" + "DCL IN[0], GENERIC[0], PERSPECTIVE\n" + "DCL SAMP[0], CONSTANT\n" + "DCL OUT[0], COLOR, CONSTANT\n" + "0: TEX OUT[0], IN[0], SAMP[0], 2D\n"/*umask*/ + "1: END\n"; + +/* μnew = 1 – (1 – μmask)*(1 – μprev) */ +static const char union_mask_asm[] = + "FRAG1.1\n" + "DCL IN[0], GENERIC[0], PERSPECTIVE\n" + "DCL IN[1], POSITION, LINEAR\n" + "DCL CONST[0], CONSTANT\n" + "DCL SAMP[0..1], CONSTANT\n" + "DCL TEMP[0..3], CONSTANT\n" + "DCL OUT[0], COLOR, CONSTANT\n" + "0: TEX TEMP[1], IN[0], SAMP[0], 2D\n"/*umask*/ + "1: TEX TEMP[0], IN[1], SAMP[1], 2D\n"/*uprev*/ + "2: SUB TEMP[2], CONST[0], TEMP[0]\n" + "3: SUB TEMP[3], CONST[0], TEMP[1]\n" + "4: MUL TEMP[0].w, TEMP[2].wwww, TEMP[3].wwww\n" + "5: SUB OUT[0], CONST[0], TEMP[0]\n" + "6: END\n"; + +/* μnew = μmask *μprev */ +static const char intersect_mask_asm[] = + "FRAG1.1\n" + "DCL IN[0], GENERIC[0], PERSPECTIVE\n" + "DCL IN[1], POSITION, LINEAR\n" + "DCL CONST[0], CONSTANT\n" + "DCL SAMP[0..1], CONSTANT\n" + "DCL TEMP[0..1], CONSTANT\n" + "DCL OUT[0], COLOR, CONSTANT\n" + "0: TEX TEMP[0], IN[1], SAMP[1], 2D\n"/*uprev*/ + "1: TEX TEMP[1], IN[0], SAMP[0], 2D\n"/*umask*/ + "2: MUL OUT[0], TEMP[0].wwww, TEMP[1].wwww\n" + "3: END\n"; + +/* μnew = μprev*(1 – μmask) */ +static const char subtract_mask_asm[] = + "FRAG1.1\n" + "DCL IN[0], GENERIC[0], PERSPECTIVE\n" + "DCL IN[1], POSITION, LINEAR\n" + "DCL CONST[0], CONSTANT\n" + "DCL SAMP[0..1], CONSTANT\n" + "DCL TEMP[0..2], CONSTANT\n" + "DCL OUT[0], COLOR, CONSTANT\n" + "0: TEX TEMP[1], IN[0], SAMP[0], 2D\n"/*umask*/ + "1: TEX TEMP[0], IN[1], SAMP[1], 2D\n"/*uprev*/ + "2: SUB TEMP[2], CONST[0], TEMP[1]\n" + "3: MUL OUT[0], TEMP[2].wwww, TEMP[0].wwww\n" + "4: END\n"; + + +static const char vs_plain_asm[] = + "VERT1.1\n" + "DCL IN[0]\n" + "DCL OUT[0], POSITION\n" + "DCL TEMP[0]\n" + "DCL CONST[0..1]\n" + "0: MUL TEMP[0], IN[0], CONST[0]\n" + "1: ADD TEMP[0], TEMP[0], CONST[1]\n" + "2: MOV OUT[0], TEMP[0]\n" + "3: END\n"; + +static const char vs_clear_asm[] = + "VERT1.1\n" + "DCL IN[0]\n" + "DCL IN[1]\n" + "DCL OUT[0], POSITION\n" + "DCL OUT[1], COLOR\n" + "DCL TEMP[0]\n" + "DCL CONST[0..1]\n" + "0: MUL TEMP[0], IN[0], CONST[0]\n" + "1: ADD TEMP[0], TEMP[0], CONST[1]\n" + "2: MOV OUT[0], TEMP[0]\n" + "3: MOV OUT[1], IN[1]\n" + "4: END\n"; + + +static const char vs_texture_asm[] = + "VERT1.1\n" + "DCL IN[0]\n" + "DCL IN[1]\n" + "DCL OUT[0], POSITION\n" + "DCL OUT[1], GENERIC\n" + "DCL TEMP[0]\n" + "DCL CONST[0..1]\n" + "0: MUL TEMP[0], IN[0], CONST[0]\n" + "1: ADD TEMP[0], TEMP[0], CONST[1]\n" + "2: MOV OUT[0], TEMP[0]\n" + "3: MOV OUT[1], IN[1]\n" + "4: END\n"; + +#endif diff --git a/src/gallium/state_trackers/vega/bezier.c b/src/gallium/state_trackers/vega/bezier.c new file mode 100644 index 00000000000..39a7ade0161 --- /dev/null +++ b/src/gallium/state_trackers/vega/bezier.c @@ -0,0 +1,704 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "bezier.h" + +#include "matrix.h" +#include "polygon.h" + +#include "pipe/p_compiler.h" +#include "util/u_debug.h" + +#include <stdlib.h> +#include <stdio.h> +#include <assert.h> +#include <math.h> + +static const float flatness = 0.5; + + +static INLINE void split_left(struct bezier *bez, VGfloat t, struct bezier* left) +{ + left->x1 = bez->x1; + left->y1 = bez->y1; + + left->x2 = bez->x1 + t * (bez->x2 - bez->x1); + left->y2 = bez->y1 + t * (bez->y2 - bez->y1); + + left->x3 = bez->x2 + t * (bez->x3 - bez->x2); + left->y3 = bez->y2 + t * (bez->y3 - bez->y2); + + bez->x3 = bez->x3 + t * (bez->x4 - bez->x3); + bez->y3 = bez->y3 + t * (bez->y4 - bez->y3); + + bez->x2 = left->x3 + t * (bez->x3 - left->x3); + bez->y2 = left->y3 + t * (bez->y3 - left->y3); + + left->x3 = left->x2 + t * (left->x3 - left->x2); + left->y3 = left->y2 + t * (left->y3 - left->y2); + + left->x4 = bez->x1 = left->x3 + t * (bez->x2 - left->x3); + left->y4 = bez->y1 = left->y3 + t * (bez->y2 - left->y3); +} + +static INLINE void split(struct bezier *bez, + struct bezier *first_half, + struct bezier *second_half) +{ + double c = (bez->x2 + bez->x3) * 0.5; + first_half->x2 = (bez->x1 + bez->x2) * 0.5; + second_half->x3 = (bez->x3 + bez->x4) * 0.5; + first_half->x1 = bez->x1; + second_half->x4 = bez->x4; + first_half->x3 = (first_half->x2 + c) * 0.5; + second_half->x2 = (second_half->x3 + c) * 0.5; + first_half->x4 = second_half->x1 = + (first_half->x3 + second_half->x2) * 0.5; + + c = (bez->y2 + bez->y3) / 2; + first_half->y2 = (bez->y1 + bez->y2) * 0.5; + second_half->y3 = (bez->y3 + bez->y4) * 0.5; + first_half->y1 = bez->y1; + second_half->y4 = bez->y4; + first_half->y3 = (first_half->y2 + c) * 0.5; + second_half->y2 = (second_half->y3 + c) * 0.5; + first_half->y4 = second_half->y1 = + (first_half->y3 + second_half->y2) * 0.5; +} + +struct polygon * bezier_to_polygon(struct bezier *bez) +{ + struct polygon *poly = polygon_create(64); + polygon_vertex_append(poly, bez->x1, bez->y1); + bezier_add_to_polygon(bez, poly); + return poly; +} + +void bezier_add_to_polygon(const struct bezier *bez, + struct polygon *poly) +{ + struct bezier beziers[32]; + struct bezier *b; + + beziers[0] = *bez; + b = beziers; + + while (b >= beziers) { + double y4y1 = b->y4 - b->y1; + double x4x1 = b->x4 - b->x1; + double l = ABS(x4x1) + ABS(y4y1); + double d; + if (l > 1.f) { + d = ABS((x4x1)*(b->y1 - b->y2) - (y4y1)*(b->x1 - b->x2)) + + ABS((x4x1)*(b->y1 - b->y3) - (y4y1)*(b->x1 - b->x3)); + } else { + d = ABS(b->x1 - b->x2) + ABS(b->y1 - b->y2) + + ABS(b->x1 - b->x3) + ABS(b->y1 - b->y3); + l = 1.; + } + if (d < flatness*l || b == beziers + 31) { + /* good enough, we pop it off and add the endpoint */ + polygon_vertex_append(poly, b->x4, b->y4); + --b; + } else { + /* split, second half of the bezier goes lower into the stack */ + split(b, b+1, b); + ++b; + } + } +} + +static void add_if_close(struct bezier *bez, VGfloat *length, VGfloat error) +{ + struct bezier left, right; /* bez poly splits */ + VGfloat len = 0.0; /* arc length */ + VGfloat chord; /* chord length */ + + len = len + line_length(bez->x1, bez->y1, bez->x2, bez->y2); + len = len + line_length(bez->x2, bez->y2, bez->x3, bez->y3); + len = len + line_length(bez->x3, bez->y3, bez->x4, bez->y4); + + chord = line_length(bez->x1, bez->y1, bez->x4, bez->y4); + + if ((len-chord) > error) { + split(bez, &left, &right); /* split in two */ + add_if_close(&left, length, error); /* try left side */ + add_if_close(&right, length, error); /* try right side */ + return; + } + + *length = *length + len; + + return; +} + +float bezier_length(struct bezier *bez, float error) +{ + VGfloat length = 0.f; + + add_if_close(bez, &length, error); + return length; +} + +void bezier_init(struct bezier *bez, + float x1, float y1, + float x2, float y2, + float x3, float y3, + float x4, float y4) +{ + bez->x1 = x1; + bez->y1 = y1; + bez->x2 = x2; + bez->y2 = y2; + bez->x3 = x3; + bez->y3 = y3; + bez->x4 = x4; + bez->y4 = y4; +#if 0 + debug_printf("bezier in [%f, %f, %f, %f, %f, %f]\n", + x1, y1, x2, y2, x3, y3, x4, y4); +#endif +} + + +static INLINE void bezier_init2v(struct bezier *bez, + float *pt1, + float *pt2, + float *pt3, + float *pt4) +{ + bez->x1 = pt1[0]; + bez->y1 = pt1[1]; + + bez->x2 = pt2[0]; + bez->y2 = pt2[1]; + + bez->x3 = pt3[0]; + bez->y3 = pt3[1]; + + bez->x4 = pt4[0]; + bez->y4 = pt4[1]; +} + + +void bezier_transform(struct bezier *bez, + struct matrix *matrix) +{ + assert(matrix_is_affine(matrix)); + matrix_map_point(matrix, bez->x1, bez->y1, &bez->x1, &bez->y1); + matrix_map_point(matrix, bez->x2, bez->y2, &bez->x2, &bez->y2); + matrix_map_point(matrix, bez->x3, bez->y3, &bez->x3, &bez->y3); + matrix_map_point(matrix, bez->x4, bez->y4, &bez->x4, &bez->y4); +} + +static INLINE void bezier_point_at(const struct bezier *bez, float t, float *pt) +{ + float a, b, c, d; + float m_t; + m_t = 1. - t; + b = m_t * m_t; + c = t * t; + d = c * t; + a = b * m_t; + b *= 3. * t; + c *= 3. * m_t; + pt[0] = a*bez->x1 + b*bez->x2 + c*bez->x3 + d*bez->x4; + pt[1] = a*bez->y1 + b*bez->y2 + c*bez->y3 + d*bez->y4; +} + +static INLINE void bezier_normal_at(const struct bezier *bez, float t, float *norm) +{ + float m_t = 1. - t; + float a = m_t * m_t; + float b = t * m_t; + float c = t * t; + + norm[0] = (bez->y2-bez->y1) * a + (bez->y3-bez->y2) * b + (bez->y4-bez->y3) * c; + norm[1] = -(bez->x2-bez->x1) * a - (bez->x3-bez->x2) * b - (bez->x4-bez->x3) * c; +} + +enum shift_result { + Ok, + Discard, + Split, + Circle +}; + +static enum shift_result good_offset(const struct bezier *b1, + const struct bezier *b2, + float offset, float threshold) +{ + const float o2 = offset*offset; + const float max_dist_line = threshold*offset*offset; + const float max_dist_normal = threshold*offset; + const float spacing = 0.25; + for (float i = spacing; i < 0.99; i += spacing) { + float p1[2],p2[2], d, l; + float normal[2]; + bezier_point_at(b1, i, p1); + bezier_point_at(b2, i, p2); + d = (p1[0] - p2[0])*(p1[0] - p2[0]) + (p1[1] - p2[1])*(p1[1] - p2[1]); + if (ABS(d - o2) > max_dist_line) + return Split; + + bezier_normal_at(b1, i, normal); + l = ABS(normal[0]) + ABS(normal[1]); + if (l != 0.) { + d = ABS(normal[0]*(p1[1] - p2[1]) - normal[1]*(p1[0] - p2[0]) ) / l; + if (d > max_dist_normal) + return Split; + } + } + return Ok; +} + +static INLINE void shift_line_by_normal(float *l, float offset) +{ + float norm[4]; + float tx, ty; + + line_normal(l, norm); + line_normalize(norm); + + tx = (norm[2] - norm[0]) * offset; + ty = (norm[3] - norm[1]) * offset; + l[0] += tx; l[1] += ty; + l[2] += tx; l[3] += ty; +} + +static INLINE VGboolean is_bezier_line(float (*points)[2], int count) +{ + float dx13 = points[2][0] - points[0][0]; + float dy13 = points[2][1] - points[0][1]; + + float dx12 = points[1][0] - points[0][0]; + float dy12 = points[1][1] - points[0][1]; + + debug_assert(count > 2); + + if (count == 3) { + return floatsEqual(dx12 * dy13, dx13 * dy12); + } else if (count == 4) { + float dx14 = points[3][0] - points[0][0]; + float dy14 = points[3][1] - points[0][1]; + + return (floatsEqual(dx12 * dy13, dx13 * dy12) && + floatsEqual(dx12 * dy14, dx14 * dy12)); + } + + return VG_FALSE; +} + +static INLINE void compute_pt_normal(float *pt1, float *pt2, float *res) +{ + float line[4]; + float normal[4]; + line[0] = 0.f; line[1] = 0.f; + line[2] = pt2[0] - pt1[0]; + line[3] = pt2[1] - pt1[1]; + line_normal(line, normal); + line_normalize(normal); + + res[0] = normal[2]; + res[1] = normal[3]; +} + +static enum shift_result shift(const struct bezier *orig, + struct bezier *shifted, + float offset, float threshold) +{ + int map[4]; + VGboolean p1_p2_equal = (orig->x1 == orig->x2 && orig->y1 == orig->y2); + VGboolean p2_p3_equal = (orig->x2 == orig->x3 && orig->y2 == orig->y3); + VGboolean p3_p4_equal = (orig->x3 == orig->x4 && orig->y3 == orig->y4); + + float points[4][2]; + int np = 0; + float bounds[4]; + float points_shifted[4][2]; + float prev_normal[2]; + + points[np][0] = orig->x1; + points[np][1] = orig->y1; + map[0] = 0; + ++np; + if (!p1_p2_equal) { + points[np][0] = orig->x2; + points[np][1] = orig->y2; + ++np; + } + map[1] = np - 1; + if (!p2_p3_equal) { + points[np][0] = orig->x3; + points[np][1] = orig->y3; + ++np; + } + map[2] = np - 1; + if (!p3_p4_equal) { + points[np][0] = orig->x4; + points[np][1] = orig->y4; + ++np; + } + map[3] = np - 1; + if (np == 1) + return Discard; + + /* We need to specialcase lines of 3 or 4 points due to numerical + instability in intersection code below */ + if (np > 2 && is_bezier_line(points, np)) { + float l[4] = { points[0][0], points[0][1], + points[np-1][0], points[np-1][1] }; + float ctrl1[2], ctrl2[2]; + if (floatsEqual(points[0][0], points[np-1][0]) && + floatsEqual(points[0][1], points[np-1][1])) + return Discard; + + shift_line_by_normal(l, offset); + line_point_at(l, 0.33, ctrl1); + line_point_at(l, 0.66, ctrl2); + bezier_init(shifted, l[0], l[1], + ctrl1[0], ctrl1[1], ctrl2[0], ctrl2[1], + l[2], l[3]); + return Ok; + } + + bezier_bounds(orig, bounds); + if (np == 4 && bounds[2] < .1*offset && bounds[3] < .1*offset) { + float l = (orig->x1 - orig->x2)*(orig->x1 - orig->x2) + + (orig->y1 - orig->y2)*(orig->y1 - orig->y1) * + (orig->x3 - orig->x4)*(orig->x3 - orig->x4) + + (orig->y3 - orig->y4)*(orig->y3 - orig->y4); + float dot = (orig->x1 - orig->x2)*(orig->x3 - orig->x4) + + (orig->y1 - orig->y2)*(orig->y3 - orig->y4); + if (dot < 0 && dot*dot < 0.8*l) + /* the points are close and reverse dirction. Approximate the whole + thing by a semi circle */ + return Circle; + } + + compute_pt_normal(points[0], points[1], prev_normal); + + points_shifted[0][0] = points[0][0] + offset * prev_normal[0]; + points_shifted[0][1] = points[0][1] + offset * prev_normal[1]; + + for (int i = 1; i < np - 1; ++i) { + float normal_sum[2], r; + float next_normal[2]; + compute_pt_normal(points[i], points[i + 1], next_normal); + + normal_sum[0] = prev_normal[0] + next_normal[0]; + normal_sum[1] = prev_normal[1] + next_normal[1]; + + r = 1.0 + prev_normal[0] * next_normal[0] + + prev_normal[1] * next_normal[1]; + + if (floatsEqual(r + 1, 1)) { + points_shifted[i][0] = points[i][0] + offset * prev_normal[0]; + points_shifted[i][1] = points[i][1] + offset * prev_normal[1]; + } else { + float k = offset / r; + points_shifted[i][0] = points[i][0] + k * normal_sum[0]; + points_shifted[i][1] = points[i][1] + k * normal_sum[1]; + } + + prev_normal[0] = next_normal[0]; + prev_normal[1] = next_normal[1]; + } + + points_shifted[np - 1][0] = points[np - 1][0] + offset * prev_normal[0]; + points_shifted[np - 1][1] = points[np - 1][1] + offset * prev_normal[1]; + + bezier_init2v(shifted, + points_shifted[map[0]], points_shifted[map[1]], + points_shifted[map[2]], points_shifted[map[3]]); + + return good_offset(orig, shifted, offset, threshold); +} + +static VGboolean make_circle(const struct bezier *b, float offset, struct bezier *o) +{ + float normals[3][2]; + float dist; + float angles[2]; + float sign = 1.f; + int i; + float circle[3][2]; + + normals[0][0] = b->y2 - b->y1; + normals[0][1] = b->x1 - b->x2; + dist = sqrt(normals[0][0]*normals[0][0] + normals[0][1]*normals[0][1]); + if (floatsEqual(dist + 1, 1.f)) + return VG_FALSE; + normals[0][0] /= dist; + normals[0][1] /= dist; + + normals[2][0] = b->y4 - b->y3; + normals[2][1] = b->x3 - b->x4; + dist = sqrt(normals[2][0]*normals[2][0] + normals[2][1]*normals[2][1]); + if (floatsEqual(dist + 1, 1.f)) + return VG_FALSE; + normals[2][0] /= dist; + normals[2][1] /= dist; + + normals[1][0] = b->x1 - b->x2 - b->x3 + b->x4; + normals[1][1] = b->y1 - b->y2 - b->y3 + b->y4; + dist = -1*sqrt(normals[1][0]*normals[1][0] + normals[1][1]*normals[1][1]); + normals[1][0] /= dist; + normals[1][1] /= dist; + + for (i = 0; i < 2; ++i) { + float cos_a = normals[i][0]*normals[i+1][0] + normals[i][1]*normals[i+1][1]; + if (cos_a > 1.) + cos_a = 1.; + if (cos_a < -1.) + cos_a = -1; + angles[i] = acos(cos_a)/M_PI; + } + + if (angles[0] + angles[1] > 1.) { + /* more than 180 degrees */ + normals[1][0] = -normals[1][0]; + normals[1][1] = -normals[1][1]; + angles[0] = 1. - angles[0]; + angles[1] = 1. - angles[1]; + sign = -1.; + } + + circle[0][0] = b->x1 + normals[0][0]*offset; + circle[0][1] = b->y1 + normals[0][1]*offset; + + circle[1][0] = 0.5*(b->x1 + b->x4) + normals[1][0]*offset; + circle[1][1] = 0.5*(b->y1 + b->y4) + normals[1][1]*offset; + + circle[2][0] = b->x4 + normals[2][0]*offset; + circle[2][1] = b->y4 + normals[2][1]*offset; + + for (i = 0; i < 2; ++i) { + float kappa = 2.*KAPPA * sign * offset * angles[i]; + + o->x1 = circle[i][0]; + o->y1 = circle[i][1]; + o->x2 = circle[i][0] - normals[i][1]*kappa; + o->y2 = circle[i][1] + normals[i][0]*kappa; + o->x3 = circle[i+1][0] + normals[i+1][1]*kappa; + o->y3 = circle[i+1][1] - normals[i+1][0]*kappa; + o->x4 = circle[i+1][0]; + o->y4 = circle[i+1][1]; + + ++o; + } + return VG_TRUE; +} + +int bezier_translate_by_normal(struct bezier *bez, + struct bezier *curves, + int max_curves, + float normal_len, + float threshold) +{ + struct bezier beziers[10]; + struct bezier *b, *o; + + /* fixme: this should really be floatsEqual */ + if (bez->x1 == bez->x2 && bez->x1 == bez->x3 && bez->x1 == bez->x4 && + bez->y1 == bez->y2 && bez->y1 == bez->y3 && bez->y1 == bez->y4) + return 0; + + --max_curves; +redo: + beziers[0] = *bez; + b = beziers; + o = curves; + + while (b >= beziers) { + int stack_segments = b - beziers + 1; + enum shift_result res; + if ((stack_segments == 10) || (o - curves == max_curves - stack_segments)) { + threshold *= 1.5; + if (threshold > 2.) + goto give_up; + goto redo; + } + res = shift(b, o, normal_len, threshold); + if (res == Discard) { + --b; + } else if (res == Ok) { + ++o; + --b; + continue; + } else if (res == Circle && max_curves - (o - curves) >= 2) { + /* add semi circle */ + if (make_circle(b, normal_len, o)) + o += 2; + --b; + } else { + split(b, b+1, b); + ++b; + } + } + +give_up: + while (b >= beziers) { + enum shift_result res = shift(b, o, normal_len, threshold); + + /* if res isn't Ok or Split then *o is undefined */ + if (res == Ok || res == Split) + ++o; + + --b; + } + + debug_assert(o - curves <= max_curves); + return o - curves; +} + +void bezier_bounds(const struct bezier *bez, + float *bounds/*x/y/width/height*/) +{ + float xmin = bez->x1; + float xmax = bez->x1; + float ymin = bez->y1; + float ymax = bez->y1; + + if (bez->x2 < xmin) + xmin = bez->x2; + else if (bez->x2 > xmax) + xmax = bez->x2; + if (bez->x3 < xmin) + xmin = bez->x3; + else if (bez->x3 > xmax) + xmax = bez->x3; + if (bez->x4 < xmin) + xmin = bez->x4; + else if (bez->x4 > xmax) + xmax = bez->x4; + + if (bez->y2 < ymin) + ymin = bez->y2; + else if (bez->y2 > ymax) + ymax = bez->y2; + if (bez->y3 < ymin) + ymin = bez->y3; + else if (bez->y3 > ymax) + ymax = bez->y3; + if (bez->y4 < ymin) + ymin = bez->y4; + else if (bez->y4 > ymax) + ymax = bez->y4; + + bounds[0] = xmin; /* x */ + bounds[1] = ymin; /* y */ + bounds[2] = xmax - xmin; /* width */ + bounds[3] = ymax - ymin; /* height */ +} + +void bezier_start_tangent(const struct bezier *bez, + float *tangent) +{ + tangent[0] = bez->x1; + tangent[1] = bez->y1; + tangent[2] = bez->x2; + tangent[3] = bez->y2; + + if (null_line(tangent)) { + tangent[0] = bez->x1; + tangent[1] = bez->y1; + tangent[2] = bez->x3; + tangent[3] = bez->y3; + } + if (null_line(tangent)) { + tangent[0] = bez->x1; + tangent[1] = bez->y1; + tangent[2] = bez->x4; + tangent[3] = bez->y4; + } +} + + +static INLINE VGfloat bezier_t_at_length(struct bezier *bez, + VGfloat at_length, + VGfloat error) +{ + VGfloat len = bezier_length(bez, error); + VGfloat t = 1.0; + VGfloat last_bigger = 1.; + + if (at_length > len || floatsEqual(at_length, len)) + return t; + + if (floatIsZero(at_length)) + return 0.f; + + t *= 0.5; + while (1) { + struct bezier right = *bez; + struct bezier left; + VGfloat tmp_len; + split_left(&right, t, &left); + tmp_len = bezier_length(&left, error); + if (ABS(tmp_len - at_length) < error) + break; + + if (tmp_len < at_length) { + t += (last_bigger - t)*.5; + } else { + last_bigger = t; + t -= t*.5; + } + } + return t; +} + +void bezier_point_at_length(struct bezier *bez, + float length, + float *point, + float *normal) +{ + /* ~0.000001 seems to be required to pass G2080x tests */ + VGfloat t = bezier_t_at_length(bez, length, 0.000001); + bezier_point_at(bez, t, point); + bezier_normal_at(bez, t, normal); + vector_unit(normal); +} + +void bezier_point_at_t(struct bezier *bez, float t, + float *point, float *normal) +{ + bezier_point_at(bez, t, point); + bezier_normal_at(bez, t, normal); + vector_unit(normal); +} + +void bezier_exact_bounds(const struct bezier *bez, + float *bounds/*x/y/width/height*/) +{ + struct polygon *poly = polygon_create(64); + polygon_vertex_append(poly, bez->x1, bez->y1); + bezier_add_to_polygon(bez, poly); + polygon_bounding_rect(poly, bounds); + polygon_destroy(poly); +} + diff --git a/src/gallium/state_trackers/vega/bezier.h b/src/gallium/state_trackers/vega/bezier.h new file mode 100644 index 00000000000..e06666051ca --- /dev/null +++ b/src/gallium/state_trackers/vega/bezier.h @@ -0,0 +1,81 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef BEZIER_H +#define BEZIER_H + +struct polygon; +struct matrix; + +struct bezier { + float x1, y1; + float x2, y2; + float x3, y3; + float x4, y4; +}; + + +#define BEZIER_DEFAULT_ERROR 0.01 + +/* kappa as being l of a circle with r = 1, we can emulate any + * circle of radius r by using the formula + * l = r . kappa + * More at: + * http://www.whizkidtech.redprince.net/bezier/circle/ */ +#define KAPPA 0.5522847498 + +void bezier_init(struct bezier *bez, + float x1, float y1, + float x2, float y2, + float x3, float y3, + float x4, float y4); + +struct polygon *bezier_to_polygon(struct bezier *bez); +void bezier_add_to_polygon(const struct bezier *bez, + struct polygon *poly); +float bezier_length(struct bezier *bez, float error); +void bezier_transform(struct bezier *bez, + struct matrix *mat); + +int bezier_translate_by_normal(struct bezier *b, + struct bezier *curves, + int max_curves, + float normal_len, + float threshold); +void bezier_bounds(const struct bezier *bez, + float *bounds/*x/y/width/height*/); +void bezier_exact_bounds(const struct bezier *bez, + float *bounds/*x/y/width/height*/); + +void bezier_start_tangent(const struct bezier *bez, + float *tangent); + +void bezier_point_at_length(struct bezier *bez, float length, + float *point, float *normal); +void bezier_point_at_t(struct bezier *bez, float t, + float *point, float *normal); + +#endif diff --git a/src/gallium/state_trackers/vega/image.c b/src/gallium/state_trackers/vega/image.c new file mode 100644 index 00000000000..9a722980d52 --- /dev/null +++ b/src/gallium/state_trackers/vega/image.c @@ -0,0 +1,654 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "image.h" + +#include "vg_translate.h" +#include "vg_context.h" +#include "matrix.h" +#include "renderer.h" +#include "util_array.h" +#include "api_consts.h" +#include "shaders_cache.h" +#include "shader.h" + +#include "pipe/p_context.h" +#include "pipe/p_screen.h" +#include "pipe/p_inlines.h" +#include "util/u_blit.h" +#include "util/u_tile.h" +#include "util/u_memory.h" +#include "util/u_math.h" + +static enum pipe_format vg_format_to_pipe(VGImageFormat format) +{ + switch(format) { + case VG_sRGB_565: + return PIPE_FORMAT_R5G6B5_UNORM; + case VG_sRGBA_5551: + return PIPE_FORMAT_A1R5G5B5_UNORM; + case VG_sRGBA_4444: + return PIPE_FORMAT_A4R4G4B4_UNORM; + case VG_sL_8: + case VG_lL_8: + return PIPE_FORMAT_L8_UNORM; + case VG_BW_1: + return PIPE_FORMAT_A8R8G8B8_UNORM; + case VG_A_8: + return PIPE_FORMAT_A8_UNORM; +#ifdef OPENVG_VERSION_1_1 + case VG_A_1: + case VG_A_4: + return PIPE_FORMAT_A8_UNORM; +#endif + default: + return PIPE_FORMAT_A8R8G8B8_UNORM; + } +} + +static INLINE void vg_sync_size(VGfloat *src_loc, VGfloat *dst_loc) +{ + src_loc[2] = MIN2(src_loc[2], dst_loc[2]); + src_loc[3] = MIN2(src_loc[3], dst_loc[3]); + dst_loc[2] = src_loc[2]; + dst_loc[3] = src_loc[3]; +} + + +static void vg_copy_texture(struct vg_context *ctx, + struct pipe_texture *dst, VGint dx, VGint dy, + struct pipe_texture *src, VGint sx, VGint sy, + VGint width, VGint height) +{ + VGfloat dst_loc[4], src_loc[4]; + VGfloat dst_bounds[4], src_bounds[4]; + VGfloat src_shift[4], dst_shift[4], shift[4]; + + dst_loc[0] = dx; + dst_loc[1] = dy; + dst_loc[2] = width; + dst_loc[3] = height; + dst_bounds[0] = 0.f; + dst_bounds[1] = 0.f; + dst_bounds[2] = dst->width[0]; + dst_bounds[3] = dst->height[0]; + + src_loc[0] = sx; + src_loc[1] = sy; + src_loc[2] = width; + src_loc[3] = height; + src_bounds[0] = 0.f; + src_bounds[1] = 0.f; + src_bounds[2] = src->width[0]; + src_bounds[3] = src->height[0]; + + vg_bound_rect(src_loc, src_bounds, src_shift); + vg_bound_rect(dst_loc, dst_bounds, dst_shift); + shift[0] = src_shift[0] - dst_shift[0]; + shift[1] = src_shift[1] - dst_shift[1]; + + if (shift[0] < 0) + vg_shift_rectx(src_loc, src_bounds, -shift[0]); + else + vg_shift_rectx(dst_loc, dst_bounds, shift[0]); + + if (shift[1] < 0) + vg_shift_recty(src_loc, src_bounds, -shift[1]); + else + vg_shift_recty(dst_loc, dst_bounds, shift[1]); + + vg_sync_size(src_loc, dst_loc); + + if (src_loc[2] >= 0 && src_loc[3] >= 0 && + dst_loc[2] >= 0 && dst_loc[3] >= 0) { + renderer_copy_texture(ctx->renderer, + src, + src_loc[0], + src_loc[1] + src_loc[3], + src_loc[0] + src_loc[2], + src_loc[1], + dst, + dst_loc[0], + dst_loc[1] + dst_loc[3], + dst_loc[0] + dst_loc[2], + dst_loc[1]); + } + +} + +void vg_copy_surface(struct vg_context *ctx, + struct pipe_surface *dst, VGint dx, VGint dy, + struct pipe_surface *src, VGint sx, VGint sy, + VGint width, VGint height) +{ + VGfloat dst_loc[4], src_loc[4]; + VGfloat dst_bounds[4], src_bounds[4]; + VGfloat src_shift[4], dst_shift[4], shift[4]; + + dst_loc[0] = dx; + dst_loc[1] = dy; + dst_loc[2] = width; + dst_loc[3] = height; + dst_bounds[0] = 0.f; + dst_bounds[1] = 0.f; + dst_bounds[2] = dst->width; + dst_bounds[3] = dst->height; + + src_loc[0] = sx; + src_loc[1] = sy; + src_loc[2] = width; + src_loc[3] = height; + src_bounds[0] = 0.f; + src_bounds[1] = 0.f; + src_bounds[2] = src->width; + src_bounds[3] = src->height; + + vg_bound_rect(src_loc, src_bounds, src_shift); + vg_bound_rect(dst_loc, dst_bounds, dst_shift); + shift[0] = src_shift[0] - dst_shift[0]; + shift[1] = src_shift[1] - dst_shift[1]; + + if (shift[0] < 0) + vg_shift_rectx(src_loc, src_bounds, -shift[0]); + else + vg_shift_rectx(dst_loc, dst_bounds, shift[0]); + + if (shift[1] < 0) + vg_shift_recty(src_loc, src_bounds, -shift[1]); + else + vg_shift_recty(dst_loc, dst_bounds, shift[1]); + + vg_sync_size(src_loc, dst_loc); + + if (src_loc[2] > 0 && src_loc[3] > 0 && + dst_loc[2] > 0 && dst_loc[3] > 0) { + if (src == dst) + renderer_copy_surface(ctx->renderer, + src, + src_loc[0], + src->height - (src_loc[1] + src_loc[3]), + src_loc[0] + src_loc[2], + src->height - src_loc[1], + dst, + dst_loc[0], + dst->height - (dst_loc[1] + dst_loc[3]), + dst_loc[0] + dst_loc[2], + dst->height - dst_loc[1], + 0, 0); + else + renderer_copy_surface(ctx->renderer, + src, + src_loc[0], + src->height - src_loc[1], + src_loc[0] + src_loc[2], + src->height - (src_loc[1] + src_loc[3]), + dst, + dst_loc[0], + dst->height - (dst_loc[1] + dst_loc[3]), + dst_loc[0] + dst_loc[2], + dst->height - dst_loc[1], + 0, 0); + } + +} + +static struct pipe_texture *image_texture(struct vg_image *img) +{ + struct pipe_texture *tex = img->texture; + return tex; +} + + +static void image_cleari(struct vg_image *img, VGint clear_colori, + VGint x, VGint y, VGint width, VGint height) +{ + VGint *clearbuf; + VGint i; + VGfloat dwidth, dheight; + + clearbuf = malloc(sizeof(VGint)*width*height); + for (i = 0; i < width*height; ++i) + clearbuf[i] = clear_colori; + + dwidth = MIN2(width, img->width); + dheight = MIN2(height, img->height); + + image_sub_data(img, clearbuf, width * sizeof(VGint), + VG_sRGBA_8888, + x, y, dwidth, dheight); + free(clearbuf); +} + +struct vg_image * image_create(VGImageFormat format, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *image = CALLOC_STRUCT(vg_image); + enum pipe_format pformat = vg_format_to_pipe(format); + struct pipe_texture pt, *newtex; + struct pipe_screen *screen = ctx->pipe->screen; + + vg_init_object(&image->base, ctx, VG_OBJECT_IMAGE); + + image->format = format; + image->width = width; + image->height = height; + + image->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + image->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + image->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + image->sampler.min_img_filter = PIPE_TEX_MIPFILTER_NEAREST; + image->sampler.mag_img_filter = PIPE_TEX_MIPFILTER_NEAREST; + image->sampler.normalized_coords = 1; + + assert(screen->is_format_supported(screen, pformat, PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_SAMPLER, 0)); + + memset(&pt, 0, sizeof(pt)); + pt.target = PIPE_TEXTURE_2D; + pt.format = pformat; + pf_get_block(pformat, &pt.block); + pt.last_level = 0; + pt.width[0] = width; + pt.height[0] = height; + pt.depth[0] = 1; + pt.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER; + + newtex = screen->texture_create(screen, &pt); + + debug_assert(newtex); + + image->texture = newtex; + + vg_context_add_object(ctx, VG_OBJECT_IMAGE, image); + + image_cleari(image, 0, 0, 0, image->width, image->height); + return image; +} + +void image_destroy(struct vg_image *img) +{ + struct vg_context *ctx = vg_current_context(); + vg_context_remove_object(ctx, VG_OBJECT_IMAGE, img); + + + if (img->parent) { + /* remove img from the parent child array */ + int idx; + struct vg_image **array = + (struct vg_image **)img->parent->children_array->data; + + for (idx = 0; idx < img->parent->children_array->num_elements; ++idx) { + struct vg_image *child = array[idx]; + if (child == img) { + break; + } + } + debug_assert(idx < img->parent->children_array->num_elements); + array_remove_element(img->parent->children_array, idx); + } + + if (img->children_array && img->children_array->num_elements) { + /* reparent the children */ + VGint i; + struct vg_image *parent = img->parent; + struct vg_image **children = + (struct vg_image **)img->children_array->data; + if (!parent) { + VGint min_x = children[0]->x; + parent = children[0]; + + for (i = 1; i < img->children_array->num_elements; ++i) { + struct vg_image *child = children[i]; + if (child->x < min_x) { + parent = child; + } + } + } + + for (i = 0; i < img->children_array->num_elements; ++i) { + struct vg_image *child = children[i]; + if (child != parent) { + child->parent = parent; + if (!parent->children_array) { + parent->children_array = array_create( + sizeof(struct vg_image*)); + } + array_append_data(parent->children_array, + &child, 1); + } else + child->parent = NULL; + } + array_destroy(img->children_array); + } + + pipe_texture_reference(&img->texture, NULL); + free(img); +} + +void image_clear(struct vg_image *img, + VGint x, VGint y, VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + VGfloat *clear_colorf = ctx->state.vg.clear_color; + VGubyte r, g, b ,a; + VGint clear_colori; + /* FIXME: this is very nasty */ + r = float_to_ubyte(clear_colorf[0]); + g = float_to_ubyte(clear_colorf[1]); + b = float_to_ubyte(clear_colorf[2]); + a = float_to_ubyte(clear_colorf[3]); + clear_colori = r << 24 | g << 16 | b << 8 | a; + image_cleari(img, clear_colori, x, y, width, height); +} + +void image_sub_data(struct vg_image *image, + const void * data, + VGint dataStride, + VGImageFormat dataFormat, + VGint x, VGint y, + VGint width, VGint height) +{ + const VGint yStep = 1; + VGubyte *src = (VGubyte *)data; + VGfloat temp[VEGA_MAX_IMAGE_WIDTH][4]; + VGfloat *df = (VGfloat*)temp; + VGint i; + struct vg_context *ctx = vg_current_context(); + struct pipe_screen *screen = ctx->pipe->screen; + struct pipe_texture *texture = image_texture(image); + VGint xoffset = 0, yoffset = 0; + + if (x < 0) { + xoffset -= x; + width += x; + x = 0; + } + if (y < 0) { + yoffset -= y; + height += y; + y = 0; + } + + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (x > image->width || y > image->width) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + + if (x + width > image->width) { + width = image->width - x; + } + + if (y + height > image->height) { + height = image->height - y; + } + + { /* upload color_data */ + struct pipe_transfer *transfer = screen->get_tex_transfer( + screen, texture, 0, 0, 0, + PIPE_TRANSFER_WRITE, 0, 0, texture->width[0], texture->height[0]); + src += (dataStride * yoffset); + for (i = 0; i < height; i++) { + _vega_unpack_float_span_rgba(ctx, width, xoffset, src, dataFormat, temp); + pipe_put_tile_rgba(transfer, x+image->x, y+image->y, width, 1, df); + y += yStep; + src += dataStride; + } + screen->tex_transfer_destroy(transfer); + } +} + +void image_get_sub_data(struct vg_image * image, + void * data, + VGint dataStride, + VGImageFormat dataFormat, + VGint sx, VGint sy, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_context *pipe = ctx->pipe; + struct pipe_screen *screen = pipe->screen; + VGfloat temp[VEGA_MAX_IMAGE_WIDTH][4]; + VGfloat *df = (VGfloat*)temp; + VGint y = 0, yStep = 1; + VGint i; + VGubyte *dst = (VGubyte *)data; + + { + struct pipe_transfer *transfer = + screen->get_tex_transfer(screen, + image->texture, 0, 0, 0, + PIPE_TRANSFER_READ, + 0, 0, + image->x + image->width, + image->y + image->height); + /* Do a row at a time to flip image data vertically */ + for (i = 0; i < height; i++) { +#if 0 + debug_printf("%d-%d == %d\n", sy, height, y); +#endif + pipe_get_tile_rgba(transfer, sx+image->x, y, width, 1, df); + y += yStep; + _vega_pack_rgba_span_float(ctx, width, temp, dataFormat, dst); + dst += dataStride; + } + + screen->tex_transfer_destroy(transfer); + } +} + +struct vg_image * image_child_image(struct vg_image *parent, + VGint x, VGint y, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_image *image = CALLOC_STRUCT(vg_image); + + vg_init_object(&image->base, ctx, VG_OBJECT_IMAGE); + + image->x = parent->x + x; + image->y = parent->y + y; + image->width = width; + image->height = height; + image->parent = parent; + image->texture = 0; + pipe_texture_reference(&image->texture, + parent->texture); + + image->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + image->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + image->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + image->sampler.min_img_filter = PIPE_TEX_MIPFILTER_NEAREST; + image->sampler.mag_img_filter = PIPE_TEX_MIPFILTER_NEAREST; + image->sampler.normalized_coords = 1; + + if (!parent->children_array) + parent->children_array = array_create( + sizeof(struct vg_image*)); + + array_append_data(parent->children_array, + &image, 1); + + vg_context_add_object(ctx, VG_OBJECT_IMAGE, image); + + return image; +} + +void image_copy(struct vg_image *dst, VGint dx, VGint dy, + struct vg_image *src, VGint sx, VGint sy, + VGint width, VGint height, + VGboolean dither) +{ + struct vg_context *ctx = vg_current_context(); + + if (width <= 0 || height <= 0) { + vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR); + return; + } + /* make sure rendering has completed */ + ctx->pipe->flush(ctx->pipe, PIPE_FLUSH_RENDER_CACHE, NULL); + vg_copy_texture(ctx, dst->texture, dst->x + dx, dst->y + dy, + src->texture, src->x + sx, src->y + sy, width, height); +} + +void image_draw(struct vg_image *img) +{ + struct vg_context *ctx = vg_current_context(); + VGfloat x1, y1; + VGfloat x2, y2; + VGfloat x3, y3; + VGfloat x4, y4; + struct matrix *matrix; + + x1 = 0; + y1 = 0; + x2 = img->width; + y2 = 0; + x3 = img->width; + y3 = img->height; + x4 = 0; + y4 = img->height; + + matrix = &ctx->state.vg.image_user_to_surface_matrix; + + matrix_map_point(matrix, x1, y1, &x1, &y1); + matrix_map_point(matrix, x2, y2, &x2, &y2); + matrix_map_point(matrix, x3, y3, &x3, &y3); + matrix_map_point(matrix, x4, y4, &x4, &y4); + + shader_set_drawing_image(ctx->shader, VG_TRUE); + shader_set_paint(ctx->shader, ctx->state.vg.fill_paint); + shader_set_image(ctx->shader, img); + shader_bind(ctx->shader); + + renderer_texture_quad(ctx->renderer, image_texture(img), + img->x, img->y, img->x + img->width, img->y + img->height, + x1, y1, x2, y2, x3, y3, x4, y4); +} + +void image_set_pixels(VGint dx, VGint dy, + struct vg_image *src, VGint sx, VGint sy, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_context *pipe = ctx->pipe; + struct pipe_screen *screen = pipe->screen; + struct pipe_surface *surf; + struct st_renderbuffer *strb = ctx->draw_buffer->strb; + + /* make sure rendering has completed */ + pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); + + surf = screen->get_tex_surface(screen, image_texture(src), 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_READ); + + vg_copy_surface(ctx, strb->surface, dx, dy, + surf, sx+src->x, sy+src->y, width, height); + + screen->tex_surface_destroy(surf); +} + +void image_get_pixels(struct vg_image *dst, VGint dx, VGint dy, + VGint sx, VGint sy, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_context *pipe = ctx->pipe; + struct pipe_screen *screen = pipe->screen; + struct pipe_surface *surf; + struct st_renderbuffer *strb = ctx->draw_buffer->strb; + + /* flip the y coordinates */ + /*dy = dst->height - dy - height;*/ + + /* make sure rendering has completed */ + pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); + + surf = screen->get_tex_surface(screen, image_texture(dst), 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_WRITE | + PIPE_BUFFER_USAGE_GPU_READ); + vg_copy_surface(ctx, surf, dst->x + dx, dst->y + dy, + strb->surface, sx, sy, width, height); + + pipe_surface_reference(&surf, NULL); +} + + +VGboolean vg_image_overlaps(struct vg_image *dst, + struct vg_image *src) +{ + if (dst == src || dst->parent == src || + dst == src->parent) + return VG_TRUE; + if (dst->parent && dst->parent == src->parent) { + VGfloat left1 = dst->x; + VGfloat left2 = src->x; + VGfloat right1 = dst->x + dst->width; + VGfloat right2 = src->x + src->width; + VGfloat bottom1 = dst->y; + VGfloat bottom2 = src->y; + VGfloat top1 = dst->y + dst->height; + VGfloat top2 = src->y + src->height; + + return !(left2 > right1 || right2 < left1 || + top2 > bottom1 || bottom2 < top1); + } + return VG_FALSE; +} + +VGint image_bind_samplers(struct vg_image *img, struct pipe_sampler_state **samplers, + struct pipe_texture **textures) +{ + img->sampler.min_img_filter = image_sampler_filter(img->base.ctx); + img->sampler.mag_img_filter = image_sampler_filter(img->base.ctx); + samplers[3] = &img->sampler; + textures[3] = img->texture; + return 1; +} + +VGint image_sampler_filter(struct vg_context *ctx) +{ + switch(ctx->state.vg.image_quality) { + case VG_IMAGE_QUALITY_NONANTIALIASED: + return PIPE_TEX_FILTER_NEAREST; + break; + case VG_IMAGE_QUALITY_FASTER: + return PIPE_TEX_FILTER_NEAREST; + break; + case VG_IMAGE_QUALITY_BETTER: + /*return PIPE_TEX_FILTER_ANISO;*/ + return PIPE_TEX_FILTER_LINEAR; + break; + default: + debug_printf("Unknown image quality"); + } + return PIPE_TEX_FILTER_NEAREST; +} diff --git a/src/gallium/state_trackers/vega/image.h b/src/gallium/state_trackers/vega/image.h new file mode 100644 index 00000000000..78e17cffa64 --- /dev/null +++ b/src/gallium/state_trackers/vega/image.h @@ -0,0 +1,104 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef IMAGES_H +#define IMAGES_H + +#include "vg_context.h" +#include "pipe/p_state.h" + +struct pipe_texture; +struct array; +struct vg_context; +struct pipe_surface; + +struct vg_image { + struct vg_object base; + VGImageFormat format; + VGint x, y; + VGint width, height; + + struct vg_image *parent; + + struct pipe_texture *texture; + struct pipe_sampler_state sampler; + + struct array *children_array; +}; + +struct vg_image *image_create(VGImageFormat format, + VGint width, VGint height); +void image_destroy(struct vg_image *img); + +void image_clear(struct vg_image *img, + VGint x, VGint y, VGint width, VGint height); + +void image_sub_data(struct vg_image *image, + const void * data, + VGint dataStride, + VGImageFormat dataFormat, + VGint x, VGint y, + VGint width, VGint height); + +void image_get_sub_data(struct vg_image * image, + void * data, + VGint dataStride, + VGImageFormat dataFormat, + VGint x, VGint y, + VGint width, VGint height); + +struct vg_image *image_child_image(struct vg_image *parent, + VGint x, VGint y, + VGint width, VGint height); + +void image_copy(struct vg_image *dst, VGint dx, VGint dy, + struct vg_image *src, VGint sx, VGint sy, + VGint width, VGint height, + VGboolean dither); + +void image_draw(struct vg_image *img); + +void image_set_pixels(VGint dx, VGint dy, + struct vg_image *src, VGint sx, VGint sy, + VGint width, VGint height); +void image_get_pixels(struct vg_image *dst, VGint dx, VGint dy, + VGint sx, VGint sy, + VGint width, VGint height); + +VGint image_bind_samplers(struct vg_image *dst, struct pipe_sampler_state **samplers, + struct pipe_texture **textures); + +VGboolean vg_image_overlaps(struct vg_image *dst, + struct vg_image *src); + +VGint image_sampler_filter(struct vg_context *ctx); + +void vg_copy_surface(struct vg_context *ctx, + struct pipe_surface *dst, VGint dx, VGint dy, + struct pipe_surface *src, VGint sx, VGint sy, + VGint width, VGint height); + +#endif diff --git a/src/gallium/state_trackers/vega/mask.c b/src/gallium/state_trackers/vega/mask.c new file mode 100644 index 00000000000..24650a37d50 --- /dev/null +++ b/src/gallium/state_trackers/vega/mask.c @@ -0,0 +1,690 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "mask.h" + +#include "path.h" +#include "image.h" +#include "shaders_cache.h" +#include "renderer.h" +#include "asm_util.h" +#include "st_inlines.h" + +#include "pipe/p_context.h" +#include "pipe/p_screen.h" +#include "pipe/p_inlines.h" +#include "util/u_memory.h" + +struct vg_mask_layer { + struct vg_object base; + + VGint width; + VGint height; + + struct pipe_texture *texture; +}; + +static INLINE struct pipe_surface * +alpha_mask_surface(struct vg_context *ctx, int usage) +{ + struct pipe_screen *screen = ctx->pipe->screen; + struct st_framebuffer *stfb = ctx->draw_buffer; + return screen->get_tex_surface(screen, + stfb->alpha_mask, + 0, 0, 0, + usage); +} + +static INLINE VGboolean +intersect_rectangles(VGint dwidth, VGint dheight, + VGint swidth, VGint sheight, + VGint tx, VGint ty, + VGint twidth, VGint theight, + VGint *offsets, + VGint *location) +{ + if (tx + twidth <= 0 || tx >= dwidth) + return VG_FALSE; + if (ty + theight <= 0 || ty >= dheight) + return VG_FALSE; + + offsets[0] = 0; + offsets[1] = 0; + location[0] = tx; + location[1] = ty; + + if (tx < 0) { + offsets[0] -= tx; + location[0] = 0; + + location[2] = MIN2(tx + swidth, MIN2(dwidth, tx + twidth)); + offsets[2] = location[2]; + } else { + offsets[2] = MIN2(twidth, MIN2(dwidth - tx, swidth )); + location[2] = offsets[2]; + } + + if (ty < 0) { + offsets[1] -= ty; + location[1] = 0; + + location[3] = MIN2(ty + sheight, MIN2(dheight, ty + theight)); + offsets[3] = location[3]; + } else { + offsets[3] = MIN2(theight, MIN2(dheight - ty, sheight)); + location[3] = offsets[3]; + } + + return VG_TRUE; +} + +#if DEBUG_MASKS +static void read_alpha_mask(void * data, VGint dataStride, + VGImageFormat dataFormat, + VGint sx, VGint sy, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_context *pipe = ctx->pipe; + struct pipe_screen *screen = pipe->screen; + + struct st_framebuffer *stfb = ctx->draw_buffer; + struct st_renderbuffer *strb = stfb->alpha_mask; + struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb; + + VGfloat temp[VEGA_MAX_IMAGE_WIDTH][4]; + VGfloat *df = (VGfloat*)temp; + VGint y = (fb->height - sy) - 1, yStep = -1; + VGint i; + VGubyte *dst = (VGubyte *)data; + VGint xoffset = 0, yoffset = 0; + + /* make sure rendering has completed */ + pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); + if (sx < 0) { + xoffset = -sx; + xoffset *= _vega_size_for_format(dataFormat); + width += sx; + sx = 0; + } + if (sy < 0) { + yoffset = -sy; + height += sy; + sy = 0; + y = (fb->height - sy) - 1; + yoffset *= dataStride; + } + + { + struct pipe_surface *surf; + + surf = screen->get_tex_surface(screen, strb->texture, 0, 0, 0, + PIPE_BUFFER_USAGE_CPU_READ); + + /* Do a row at a time to flip image data vertically */ + for (i = 0; i < height; i++) { +#if 0 + debug_printf("%d-%d == %d\n", sy, height, y); +#endif + pipe_get_tile_rgba(surf, sx, y, width, 1, df); + y += yStep; + _vega_pack_rgba_span_float(ctx, width, temp, dataFormat, + dst + yoffset + xoffset); + dst += dataStride; + } + + pipe_surface_reference(&surf, NULL); + } +} + +void save_alpha_to_file(const char *filename) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb; + VGint *data; + int i, j; + + data = malloc(sizeof(int) * fb->width * fb->height); + read_alpha_mask(data, fb->width * sizeof(int), + VG_sRGBA_8888, + 0, 0, fb->width, fb->height); + fprintf(stderr, "/*---------- start */\n"); + fprintf(stderr, "const int image_width = %d;\n", + fb->width); + fprintf(stderr, "const int image_height = %d;\n", + fb->height); + fprintf(stderr, "const int image_data = {\n"); + for (i = 0; i < fb->height; ++i) { + for (j = 0; j < fb->width; ++j) { + int rgba = data[i * fb->height + j]; + int argb = 0; + argb = (rgba >> 8); + argb |= ((rgba & 0xff) << 24); + fprintf(stderr, "0x%x, ", argb); + } + fprintf(stderr, "\n"); + } + fprintf(stderr, "};\n"); + fprintf(stderr, "/*---------- end */\n"); +} +#endif + +static void setup_mask_framebuffer(struct pipe_surface *surf, + VGint surf_width, VGint surf_height) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_framebuffer_state fb; + + memset(&fb, 0, sizeof(fb)); + fb.width = surf_width; + fb.height = surf_height; + fb.nr_cbufs = 1; + fb.cbufs[0] = surf; + { + VGint i; + for (i = 1; i < PIPE_MAX_COLOR_BUFS; ++i) + fb.cbufs[i] = 0; + } + cso_set_framebuffer(ctx->cso_context, &fb); +} + + +/* setup shader constants */ +static void setup_mask_operation(VGMaskOperation operation) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_constant_buffer *cbuf = &ctx->mask.cbuf; + const VGint param_bytes = 4 * sizeof(VGfloat); + const VGfloat ones[4] = {1.f, 1.f, 1.f, 1.f}; + void *shader = 0; + + /* We always need to get a new buffer, to keep the drivers simple and + * avoid gratuitous rendering synchronization. + */ + pipe_buffer_reference(&cbuf->buffer, NULL); + + cbuf->buffer = pipe_buffer_create(ctx->pipe->screen, 1, + PIPE_BUFFER_USAGE_CONSTANT, + param_bytes); + if (cbuf->buffer) { + st_no_flush_pipe_buffer_write(ctx, cbuf->buffer, + 0, param_bytes, ones); + } + + ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, cbuf); + switch (operation) { + case VG_UNION_MASK: { + if (!ctx->mask.union_fs) { + ctx->mask.union_fs = shader_create_from_text(ctx->pipe, + union_mask_asm, + 200, + PIPE_SHADER_FRAGMENT); + } + shader = ctx->mask.union_fs->driver; + } + break; + case VG_INTERSECT_MASK: { + if (!ctx->mask.intersect_fs) { + ctx->mask.intersect_fs = shader_create_from_text(ctx->pipe, + intersect_mask_asm, + 200, + PIPE_SHADER_FRAGMENT); + } + shader = ctx->mask.intersect_fs->driver; + } + break; + case VG_SUBTRACT_MASK: { + if (!ctx->mask.subtract_fs) { + ctx->mask.subtract_fs = shader_create_from_text(ctx->pipe, + subtract_mask_asm, + 200, + PIPE_SHADER_FRAGMENT); + } + shader = ctx->mask.subtract_fs->driver; + } + break; + case VG_SET_MASK: { + if (!ctx->mask.set_fs) { + ctx->mask.set_fs = shader_create_from_text(ctx->pipe, + set_mask_asm, + 200, + PIPE_SHADER_FRAGMENT); + } + shader = ctx->mask.set_fs->driver; + } + break; + default: + assert(0); + break; + } + cso_set_fragment_shader_handle(ctx->cso_context, shader); +} + +static void setup_mask_samplers(struct pipe_texture *umask) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS]; + struct pipe_texture *textures[PIPE_MAX_SAMPLERS]; + struct st_framebuffer *fb_buffers = ctx->draw_buffer; + struct pipe_texture *uprev = NULL; + struct pipe_sampler_state sampler; + + uprev = fb_buffers->blend_texture; + sampler = ctx->mask.sampler; + sampler.normalized_coords = 1; + + samplers[0] = NULL; + samplers[1] = NULL; + samplers[2] = NULL; + textures[0] = NULL; + textures[1] = NULL; + textures[2] = NULL; + + samplers[0] = &sampler; + samplers[1] = &ctx->mask.sampler; + + textures[0] = umask; + textures[1] = uprev; + + cso_set_samplers(ctx->cso_context, 2, + (const struct pipe_sampler_state **)samplers); + cso_set_sampler_textures(ctx->cso_context, 2, textures); +} + + +/* setup shader constants */ +static void setup_mask_fill(const VGfloat color[4]) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_constant_buffer *cbuf = &ctx->mask.cbuf; + const VGint param_bytes = 4 * sizeof(VGfloat); + + /* We always need to get a new buffer, to keep the drivers simple and + * avoid gratuitous rendering synchronization. + */ + pipe_buffer_reference(&cbuf->buffer, NULL); + + cbuf->buffer = pipe_buffer_create(ctx->pipe->screen, 1, + PIPE_BUFFER_USAGE_CONSTANT, + param_bytes); + if (cbuf->buffer) { + st_no_flush_pipe_buffer_write(ctx, cbuf->buffer, 0, param_bytes, color); + } + + ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, cbuf); + cso_set_fragment_shader_handle(ctx->cso_context, + shaders_cache_fill(ctx->sc, + VEGA_SOLID_FILL_SHADER)); +} + +static void setup_mask_viewport() +{ + struct vg_context *ctx = vg_current_context(); + vg_set_viewport(ctx, VEGA_Y0_TOP); +} + +static void setup_mask_blend() +{ + struct vg_context *ctx = vg_current_context(); + + struct pipe_blend_state blend; + + memset(&blend, 0, sizeof(struct pipe_blend_state)); + blend.blend_enable = 1; + blend.colormask |= PIPE_MASK_R; + blend.colormask |= PIPE_MASK_G; + blend.colormask |= PIPE_MASK_B; + blend.colormask |= PIPE_MASK_A; + blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE; + blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE; + blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO; + blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO; + + cso_set_blend(ctx->cso_context, &blend); +} + + +static void surface_fill(struct pipe_surface *surf, + int surf_width, int surf_height, + int x, int y, int width, int height, + const VGfloat color[4]) +{ + struct vg_context *ctx = vg_current_context(); + + if (x < 0) { + width += x; + x = 0; + } + if (y < 0) { + height += y; + y = 0; + } + + cso_save_framebuffer(ctx->cso_context); + cso_save_blend(ctx->cso_context); + cso_save_fragment_shader(ctx->cso_context); + cso_save_viewport(ctx->cso_context); + + setup_mask_blend(); + setup_mask_fill(color); + setup_mask_framebuffer(surf, surf_width, surf_height); + setup_mask_viewport(); + + renderer_draw_quad(ctx->renderer, x, y, + x + width, y + height, 0.0f/*depth should be disabled*/); + + + /* make sure rendering has completed */ + ctx->pipe->flush(ctx->pipe, + PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME, + NULL); + +#if DEBUG_MASKS + save_alpha_to_file(0); +#endif + + cso_restore_blend(ctx->cso_context); + cso_restore_framebuffer(ctx->cso_context); + cso_restore_fragment_shader(ctx->cso_context); + cso_restore_viewport(ctx->cso_context); +} + + +static void mask_using_texture(struct pipe_texture *texture, + VGMaskOperation operation, + VGint x, VGint y, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct pipe_surface *surface = + alpha_mask_surface(ctx, PIPE_BUFFER_USAGE_GPU_WRITE); + VGint offsets[4], loc[4]; + + if (!surface) + return; + if (!intersect_rectangles(surface->width, surface->height, + texture->width[0], texture->height[0], + x, y, width, height, + offsets, loc)) + return; +#if 0 + debug_printf("Offset = [%d, %d, %d, %d]\n", offsets[0], + offsets[1], offsets[2], offsets[3]); + debug_printf("Locati = [%d, %d, %d, %d]\n", loc[0], + loc[1], loc[2], loc[3]); +#endif + + /* prepare our blend surface */ + vg_prepare_blend_surface_from_mask(ctx); + + cso_save_samplers(ctx->cso_context); + cso_save_sampler_textures(ctx->cso_context); + cso_save_framebuffer(ctx->cso_context); + cso_save_blend(ctx->cso_context); + cso_save_fragment_shader(ctx->cso_context); + cso_save_viewport(ctx->cso_context); + + setup_mask_samplers(texture); + setup_mask_blend(); + setup_mask_operation(operation); + setup_mask_framebuffer(surface, surface->width, surface->height); + setup_mask_viewport(); + + /* render the quad to propagate the rendering from stencil */ + renderer_draw_texture(ctx->renderer, texture, + offsets[0], offsets[1], + offsets[0] + offsets[2], offsets[1] + offsets[3], + loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3]); + + /* make sure rendering has completed */ + ctx->pipe->flush(ctx->pipe, PIPE_FLUSH_RENDER_CACHE, NULL); + cso_restore_blend(ctx->cso_context); + cso_restore_framebuffer(ctx->cso_context); + cso_restore_fragment_shader(ctx->cso_context); + cso_restore_samplers(ctx->cso_context); + cso_restore_sampler_textures(ctx->cso_context); + cso_restore_viewport(ctx->cso_context); + + pipe_surface_reference(&surface, NULL); +} + + +#ifdef OPENVG_VERSION_1_1 + +struct vg_mask_layer * mask_layer_create(VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct vg_mask_layer *mask = 0; + + mask = CALLOC_STRUCT(vg_mask_layer); + vg_init_object(&mask->base, ctx, VG_OBJECT_MASK); + mask->width = width; + mask->height = height; + + { + struct pipe_texture pt; + struct pipe_screen *screen = ctx->pipe->screen; + + memset(&pt, 0, sizeof(pt)); + pt.target = PIPE_TEXTURE_2D; + pt.format = PIPE_FORMAT_A8R8G8B8_UNORM; + pf_get_block(PIPE_FORMAT_A8R8G8B8_UNORM, &pt.block); + pt.last_level = 0; + pt.width[0] = width; + pt.height[0] = height; + pt.depth[0] = 1; + pt.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER; + pt.compressed = 0; + + mask->texture = screen->texture_create(screen, &pt); + } + + vg_context_add_object(ctx, VG_OBJECT_MASK, mask); + + return mask; +} + +void mask_layer_destroy(struct vg_mask_layer *layer) +{ + struct vg_context *ctx = vg_current_context(); + + vg_context_remove_object(ctx, VG_OBJECT_MASK, layer); + pipe_texture_release(&layer->texture); + free(layer); +} + +void mask_layer_fill(struct vg_mask_layer *layer, + VGint x, VGint y, + VGint width, VGint height, + VGfloat value) +{ + struct vg_context *ctx = vg_current_context(); + VGfloat alpha_color[4] = {0, 0, 0, 0}; + struct pipe_surface *surface; + + alpha_color[3] = value; + + surface = ctx->pipe->screen->get_tex_surface( + ctx->pipe->screen, layer->texture, + 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_WRITE); + + surface_fill(surface, + layer->width, layer->height, + x, y, width, height, alpha_color); + + ctx->pipe->screen->tex_surface_release(ctx->pipe->screen, &surface); +} + +void mask_copy(struct vg_mask_layer *layer, + VGint sx, VGint sy, + VGint dx, VGint dy, + VGint width, VGint height) +{ + struct vg_context *ctx = vg_current_context(); + struct st_framebuffer *fb_buffers = ctx->draw_buffer; + + renderer_copy_texture(ctx->renderer, + layer->texture, + sx, sy, + sx + width, sy + height, + fb_buffers->alpha_mask, + dx, dy, + dx + width, dy + height); +} + +static void mask_layer_render_to(struct vg_mask_layer *layer, + struct path *path, + VGbitfield paint_modes) +{ + struct vg_context *ctx = vg_current_context(); + const VGfloat fill_color[4] = {1.f, 1.f, 1.f, 1.f}; + struct pipe_screen *screen = ctx->pipe->screen; + struct pipe_surface *surface; + + surface = screen->get_tex_surface(screen, layer->texture, 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_WRITE); + + cso_save_framebuffer(ctx->cso_context); + cso_save_fragment_shader(ctx->cso_context); + cso_save_viewport(ctx->cso_context); + + setup_mask_blend(); + setup_mask_fill(fill_color); + setup_mask_framebuffer(surface, layer->width, layer->height); + setup_mask_viewport(); + + if (paint_modes & VG_FILL_PATH) { + struct matrix *mat = &ctx->state.vg.path_user_to_surface_matrix; + path_fill(path, mat); + } + + if (paint_modes & VG_STROKE_PATH){ + path_stroke(path); + } + + + /* make sure rendering has completed */ + ctx->pipe->flush(ctx->pipe, PIPE_FLUSH_RENDER_CACHE, NULL); + + cso_restore_framebuffer(ctx->cso_context); + cso_restore_fragment_shader(ctx->cso_context); + cso_restore_viewport(ctx->cso_context); + ctx->state.dirty |= BLEND_DIRTY; + + screen->tex_surface_release(ctx->pipe->screen, &surface); +} + +void mask_render_to(struct path *path, + VGbitfield paint_modes, + VGMaskOperation operation) +{ + struct vg_context *ctx = vg_current_context(); + struct st_framebuffer *fb_buffers = ctx->draw_buffer; + struct vg_mask_layer *temp_layer; + VGint width, height; + + width = fb_buffers->alpha_mask->width[0]; + height = fb_buffers->alpha_mask->width[0]; + + temp_layer = mask_layer_create(width, height); + + mask_layer_render_to(temp_layer, path, paint_modes); + + mask_using_layer(temp_layer, 0, 0, width, height, + operation); + + mask_layer_destroy(temp_layer); +} + +void mask_using_layer(struct vg_mask_layer *layer, + VGMaskOperation operation, + VGint x, VGint y, + VGint width, VGint height) +{ + mask_using_texture(layer->texture, operation, + x, y, width, height); +} + +VGint mask_layer_width(struct vg_mask_layer *layer) +{ + return layer->width; +} + +VGint mask_layer_height(struct vg_mask_layer *layer) +{ + return layer->height; +} + + +#endif + +void mask_using_image(struct vg_image *image, + VGMaskOperation operation, + VGint x, VGint y, + VGint width, VGint height) +{ + mask_using_texture(image->texture, operation, + x, y, width, height); +} + +void mask_fill(VGint x, VGint y, VGint width, VGint height, + VGfloat value) +{ + struct vg_context *ctx = vg_current_context(); + VGfloat alpha_color[4] = {.0f, .0f, .0f, value}; + struct pipe_surface *surf = alpha_mask_surface( + ctx, PIPE_BUFFER_USAGE_GPU_WRITE); + +#if DEBUG_MASKS + debug_printf("mask_fill(%d, %d, %d, %d) with rgba(%f, %f, %f, %f)\n", + x, y, width, height, + alpha_color[0], alpha_color[1], + alpha_color[2], alpha_color[3]); + debug_printf("XXX %f === %f \n", + alpha_color[3], value); +#endif + + surface_fill(surf, surf->width, surf->height, + x, y, width, height, alpha_color); + + pipe_surface_reference(&surf, NULL); +} + +VGint mask_bind_samplers(struct pipe_sampler_state **samplers, + struct pipe_texture **textures) +{ + struct vg_context *ctx = vg_current_context(); + + if (ctx->state.vg.masking) { + struct st_framebuffer *fb_buffers = ctx->draw_buffer; + + samplers[1] = &ctx->mask.sampler; + textures[1] = fb_buffers->alpha_mask; + return 1; + } else + return 0; +} diff --git a/src/gallium/state_trackers/vega/mask.h b/src/gallium/state_trackers/vega/mask.h new file mode 100644 index 00000000000..5eaaede0e3a --- /dev/null +++ b/src/gallium/state_trackers/vega/mask.h @@ -0,0 +1,68 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef MASK_H +#define MASK_H + +#include "vg_context.h" + +struct path; +struct vg_image; +struct pipe_texture; + +struct vg_mask_layer *mask_layer_create(VGint width, VGint height); +void mask_layer_destroy(struct vg_mask_layer *layer); +void mask_layer_fill(struct vg_mask_layer *layer, + VGint x, VGint y, + VGint width, VGint height, + VGfloat value); +VGint mask_layer_width(struct vg_mask_layer *layer); +VGint mask_layer_height(struct vg_mask_layer *layer); +void mask_copy(struct vg_mask_layer *layer, + VGint sx, VGint sy, + VGint dx, VGint dy, + VGint width, VGint height); + +void mask_render_to(struct path *path, + VGbitfield paint_modes, + VGMaskOperation operation); + +void mask_using_layer(struct vg_mask_layer *layer, + VGMaskOperation operation, + VGint x, VGint y, + VGint width, VGint height); +void mask_using_image(struct vg_image *image, + VGMaskOperation operation, + VGint x, VGint y, + VGint width, VGint height); +void mask_fill(VGint x, VGint y, + VGint width, VGint height, + VGfloat value); + +VGint mask_bind_samplers(struct pipe_sampler_state **samplers, + struct pipe_texture **textures); + +#endif diff --git a/src/gallium/state_trackers/vega/matrix.h b/src/gallium/state_trackers/vega/matrix.h new file mode 100644 index 00000000000..4c207f912a8 --- /dev/null +++ b/src/gallium/state_trackers/vega/matrix.h @@ -0,0 +1,462 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef MATRIX_H +#define MATRIX_H + +#include "VG/openvg.h" + +#include "pipe/p_compiler.h" +#include "util/u_math.h" + +#include <stdio.h> +#include <math.h> + +#define floatsEqual(x, y) (fabs(x - y) <= 0.00001f * MIN2(fabs(x), fabs(y))) +#define floatIsZero(x) (floatsEqual((x) + 1, 1)) +#define ABS(x) (fabsf(x)) + +#define DEGREES_TO_RADIANS(d) (0.0174532925199 * (d)) +#define FLT_TO_INT(flt) float_to_int_floor(((VGuint*)&(flt))[0]) + +static INLINE VGint float_to_int_floor(VGuint bits) +{ + int sign = (bits >> 31) ? -1 : 1; + int exp = ((bits >> 23) & 255) - 127; + int mant = bits & 0x007fffff; + int sh = 23 - exp; + + /* abs(value) >= 2^31 -> clamp. */ + + if (exp >= 31) + return (VGint)((sign < 0) ? 0x80000000u : 0x7fffffffu); + + /* abs(value) < 1 -> return -1 or 0. */ + + if (exp < 0) + return (sign < 0 && (exp > -127 || mant != 0)) ? -1 : 0; + + /* abs(value) >= 2^23 -> shift left. */ + + mant |= 0x00800000; + if (sh <= 0) + return sign * (mant << -sh); + + /* Negative -> add a rounding term. */ + + if (sign < 0) + mant += (1 << sh) - 1; + + /* Shift right to obtain the result. */ + + return sign * (mant >> sh); +} + + +struct matrix { + VGfloat m[9]; +}; + +static INLINE void matrix_init(struct matrix *mat, + const VGfloat *val) +{ + memcpy(mat->m, val, sizeof(VGfloat) * 9); +} + +static INLINE void matrix_inits(struct matrix *mat, + VGfloat m11, VGfloat m12, VGfloat m13, + VGfloat m21, VGfloat m22, VGfloat m23, + VGfloat m31, VGfloat m32, VGfloat m33) +{ + mat->m[0] = m11; mat->m[1] = m12; mat->m[2] = m13; + mat->m[3] = m21; mat->m[4] = m22; mat->m[5] = m23; + mat->m[6] = m31; mat->m[7] = m32; mat->m[8] = m33; +} + +static INLINE void matrix_load_identity(struct matrix *matrix) +{ + static const VGfloat identity[9] = {1.f, 0.f, 0.f, + 0.f, 1.f, 0.f, + 0.f, 0.f, 1.f}; + memcpy(matrix->m, identity, sizeof(identity)); +} + +static INLINE VGboolean matrix_is_identity(struct matrix *matrix) +{ + return floatsEqual(matrix->m[0], 1) && floatIsZero(matrix->m[1]) && + floatIsZero(matrix->m[2]) && + floatIsZero(matrix->m[3]) && floatsEqual(matrix->m[4], 1) && + floatIsZero(matrix->m[5]) && + floatIsZero(matrix->m[6]) && floatIsZero(matrix->m[7]) && + floatIsZero(matrix->m[8]); +} + +static INLINE VGboolean matrix_is_affine(struct matrix *matrix) +{ + return floatIsZero(matrix->m[2]) && floatIsZero(matrix->m[5]) + && floatsEqual(matrix->m[8], 1); +} + + +static INLINE void matrix_make_affine(struct matrix *matrix) +{ + matrix->m[2] = 0.f; + matrix->m[5] = 0.f; + matrix->m[8] = 1.f; +} + +static INLINE void matrix_mult(struct matrix *dst, + struct matrix *src) +{ + VGfloat m11 = dst->m[0]*src->m[0] + dst->m[3]*src->m[1] + dst->m[6]*src->m[2]; + VGfloat m12 = dst->m[0]*src->m[3] + dst->m[3]*src->m[4] + dst->m[6]*src->m[5]; + VGfloat m13 = dst->m[0]*src->m[6] + dst->m[3]*src->m[7] + dst->m[6]*src->m[8]; + + VGfloat m21 = dst->m[1]*src->m[0] + dst->m[4]*src->m[1] + dst->m[7]*src->m[2]; + VGfloat m22 = dst->m[1]*src->m[3] + dst->m[4]*src->m[4] + dst->m[7]*src->m[5]; + VGfloat m23 = dst->m[1]*src->m[6] + dst->m[4]*src->m[7] + dst->m[7]*src->m[8]; + + VGfloat m31 = dst->m[2]*src->m[0] + dst->m[5]*src->m[1] + dst->m[8]*src->m[2]; + VGfloat m32 = dst->m[2]*src->m[3] + dst->m[5]*src->m[4] + dst->m[8]*src->m[5]; + VGfloat m33 = dst->m[2]*src->m[6] + dst->m[5]*src->m[7] + dst->m[8]*src->m[8]; + + dst->m[0] = m11; dst->m[1] = m21; dst->m[2] = m31; + dst->m[3] = m12; dst->m[4] = m22; dst->m[5] = m32; + dst->m[6] = m13; dst->m[7] = m23; dst->m[8] = m33; +} + + +static INLINE void matrix_map_point(struct matrix *mat, + VGfloat x, VGfloat y, + VGfloat *out_x, VGfloat *out_y) +{ + /* to be able to do matrix_map_point(m, x, y, &x, &y) use + * temporaries */ + VGfloat tmp_x = x, tmp_y = y; + + *out_x = mat->m[0]*tmp_x + mat->m[3]*tmp_y + mat->m[6]; + *out_y = mat->m[1]*tmp_x + mat->m[4]*tmp_y + mat->m[7]; + if (!matrix_is_affine(mat)) { + VGfloat w = 1/(mat->m[2]*tmp_x + mat->m[5]*tmp_y + mat->m[8]); + *out_x *= w; + *out_y *= w; + } +} + +static INLINE void matrix_translate(struct matrix *dst, + VGfloat tx, VGfloat ty) +{ + if (!matrix_is_affine(dst)) { + struct matrix trans_matrix; + matrix_load_identity(&trans_matrix); + trans_matrix.m[6] = tx; + trans_matrix.m[7] = ty; + matrix_mult(dst, &trans_matrix); + } else { + dst->m[6] += tx*dst->m[0] + ty*dst->m[3]; + dst->m[7] += ty*dst->m[4] + tx*dst->m[1]; + } +} + +static INLINE void matrix_scale(struct matrix *dst, + VGfloat sx, VGfloat sy) +{ + if (!matrix_is_affine(dst)) { + struct matrix scale_matrix; + matrix_load_identity(&scale_matrix); + scale_matrix.m[0] = sx; + scale_matrix.m[4] = sy; + matrix_mult(dst, &scale_matrix); + } else { + dst->m[0] *= sx; dst->m[1] *= sx; + dst->m[3] *= sy; dst->m[4] *= sy; + } +} + +static INLINE void matrix_shear(struct matrix *dst, + VGfloat shx, VGfloat shy) +{ + struct matrix shear_matrix; + matrix_load_identity(&shear_matrix); + shear_matrix.m[1] = shy; + shear_matrix.m[3] = shx; + matrix_mult(dst, &shear_matrix); +} + +static INLINE void matrix_rotate(struct matrix *dst, + VGfloat angle) +{ + struct matrix mat; + float sin_val = 0; + float cos_val = 0; + + + if (floatsEqual(angle, 90) || floatsEqual(angle, -270)) + sin_val = 1.f; + else if (floatsEqual(angle, 270) || floatsEqual(angle, -90)) + sin_val = -1.f; + else if (floatsEqual(angle, 180)) + cos_val = -1.f; + else { + float radians = DEGREES_TO_RADIANS(angle); + sin_val = sin(radians); + cos_val = cos(radians); + } + + if (!matrix_is_affine(dst)) { + matrix_load_identity(&mat); + mat.m[0] = cos_val; mat.m[1] = sin_val; + mat.m[3] = -sin_val; mat.m[4] = cos_val; + + matrix_mult(dst, &mat); + } else { + VGfloat m11 = cos_val*dst->m[0] + sin_val*dst->m[3]; + VGfloat m12 = cos_val*dst->m[1] + sin_val*dst->m[4]; + VGfloat m21 = -sin_val*dst->m[0] + cos_val*dst->m[3]; + VGfloat m22 = -sin_val*dst->m[1] + cos_val*dst->m[4]; + dst->m[0] = m11; dst->m[1] = m12; + dst->m[3] = m21; dst->m[4] = m22; + } +} + + +static INLINE VGfloat matrix_determinant(struct matrix *mat) +{ + return mat->m[0]*(mat->m[8]*mat->m[4]-mat->m[7]*mat->m[5]) - + mat->m[3]*(mat->m[8]*mat->m[1]-mat->m[7]*mat->m[2])+ + mat->m[6]*(mat->m[5]*mat->m[1]-mat->m[4]*mat->m[2]); +} + + +static INLINE void matrix_adjoint(struct matrix *mat) +{ + VGfloat h[9]; + h[0] = mat->m[4]*mat->m[8] - mat->m[5]*mat->m[7]; + h[3] = mat->m[5]*mat->m[6] - mat->m[3]*mat->m[8]; + h[6] = mat->m[3]*mat->m[7] - mat->m[4]*mat->m[6]; + h[1] = mat->m[2]*mat->m[7] - mat->m[1]*mat->m[8]; + h[4] = mat->m[0]*mat->m[8] - mat->m[2]*mat->m[6]; + h[7] = mat->m[1]*mat->m[6] - mat->m[0]*mat->m[7]; + h[2] = mat->m[1]*mat->m[5] - mat->m[2]*mat->m[4]; + h[5] = mat->m[2]*mat->m[3] - mat->m[0]*mat->m[5]; + h[8] = mat->m[0]*mat->m[4] - mat->m[1]*mat->m[3]; + + + memcpy(mat->m, h, sizeof(VGfloat) * 9); +} + +static INLINE void matrix_divs(struct matrix *mat, + VGfloat s) +{ + mat->m[0] /= s; + mat->m[1] /= s; + mat->m[2] /= s; + mat->m[3] /= s; + mat->m[4] /= s; + mat->m[5] /= s; + mat->m[6] /= s; + mat->m[7] /= s; + mat->m[8] /= s; +} + +static INLINE VGboolean matrix_invert(struct matrix *mat) +{ + VGfloat det = matrix_determinant(mat); + + if (floatIsZero(det)) + return VG_FALSE; + + matrix_adjoint(mat); + matrix_divs(mat, det); + return VG_TRUE; +} + +static INLINE VGboolean matrix_is_invertible(struct matrix *mat) +{ + return !floatIsZero(matrix_determinant(mat)); +} + + +static INLINE VGboolean matrix_square_to_quad(VGfloat dx0, VGfloat dy0, + VGfloat dx1, VGfloat dy1, + VGfloat dx3, VGfloat dy3, + VGfloat dx2, VGfloat dy2, + struct matrix *mat) +{ + VGfloat ax = dx0 - dx1 + dx2 - dx3; + VGfloat ay = dy0 - dy1 + dy2 - dy3; + + if (floatIsZero(ax) && floatIsZero(ay)) { + /* affine case */ + matrix_inits(mat, + dx1 - dx0, dy1 - dy0, 0, + dx2 - dx1, dy2 - dy1, 0, + dx0, dy0, 1); + } else { + VGfloat a, b, c, d, e, f, g, h; + VGfloat ax1 = dx1 - dx2; + VGfloat ax2 = dx3 - dx2; + VGfloat ay1 = dy1 - dy2; + VGfloat ay2 = dy3 - dy2; + + /* determinants */ + VGfloat gtop = ax * ay2 - ax2 * ay; + VGfloat htop = ax1 * ay - ax * ay1; + VGfloat bottom = ax1 * ay2 - ax2 * ay1; + + if (!bottom) + return VG_FALSE; + + g = gtop / bottom; + h = htop / bottom; + + a = dx1 - dx0 + g * dx1; + b = dx3 - dx0 + h * dx3; + c = dx0; + d = dy1 - dy0 + g * dy1; + e = dy3 - dy0 + h * dy3; + f = dy0; + + matrix_inits(mat, + a, d, g, + b, e, h, + c, f, 1.f); + } + + return VG_TRUE; +} + +static INLINE VGboolean matrix_quad_to_square(VGfloat sx0, VGfloat sy0, + VGfloat sx1, VGfloat sy1, + VGfloat sx2, VGfloat sy2, + VGfloat sx3, VGfloat sy3, + struct matrix *mat) +{ + if (!matrix_square_to_quad(sx0, sy0, sx1, sy1, + sx2, sy2, sx3, sy3, + mat)) + return VG_FALSE; + + return matrix_invert(mat); +} + + +static INLINE VGboolean matrix_quad_to_quad(VGfloat dx0, VGfloat dy0, + VGfloat dx1, VGfloat dy1, + VGfloat dx2, VGfloat dy2, + VGfloat dx3, VGfloat dy3, + VGfloat sx0, VGfloat sy0, + VGfloat sx1, VGfloat sy1, + VGfloat sx2, VGfloat sy2, + VGfloat sx3, VGfloat sy3, + struct matrix *mat) +{ + struct matrix sqr_to_qd; + + if (!matrix_square_to_quad(dx0, dy0, dx1, dy1, + dx2, dy2, dx3, dy3, + mat)) + return VG_FALSE; + + if (!matrix_quad_to_square(sx0, sy0, sx1, sy1, + sx2, sy2, sx3, sy3, + &sqr_to_qd)) + return VG_FALSE; + + matrix_mult(mat, &sqr_to_qd); + + return VG_TRUE; +} + + +static INLINE VGboolean null_line(const VGfloat *l) +{ + return floatsEqual(l[0], l[2]) && floatsEqual(l[1], l[3]); +} + +static INLINE void line_normal(float *l, float *norm) +{ + norm[0] = l[0]; + norm[1] = l[1]; + + norm[2] = l[0] + (l[3] - l[1]); + norm[3] = l[1] - (l[2] - l[0]); +} + +static INLINE void line_normalize(float *l) +{ + float x = l[2] - l[0]; + float y = l[3] - l[1]; + float len = sqrt(x*x + y*y); + l[2] = l[0] + x/len; + l[3] = l[1] + y/len; +} + +static INLINE VGfloat line_length(VGfloat x1, VGfloat y1, + VGfloat x2, VGfloat y2) +{ + VGfloat x = x2 - x1; + VGfloat y = y2 - y1; + return sqrt(x*x + y*y); +} + +static INLINE VGfloat line_lengthv(const VGfloat *l) +{ + VGfloat x = l[2] - l[0]; + VGfloat y = l[3] - l[1]; + return sqrt(x*x + y*y); +} + + +static INLINE void line_point_at(float *l, float t, float *pt) +{ + float dx = l[2] - l[0]; + float dy = l[3] - l[1]; + + pt[0] = l[0] + dx * t; + pt[1] = l[1] + dy * t; +} + +static INLINE void vector_unit(float *vec) +{ + float len = sqrt(vec[0] * vec[0] + vec[1] * vec[1]); + vec[0] /= len; + vec[1] /= len; +} + +static INLINE void line_normal_vector(float *line, float *vec) +{ + VGfloat normal[4]; + + line_normal(line, normal); + + vec[0] = normal[2] - normal[0]; + vec[1] = normal[3] - normal[1]; + + vector_unit(vec); +} + +#endif diff --git a/src/gallium/state_trackers/vega/paint.c b/src/gallium/state_trackers/vega/paint.c new file mode 100644 index 00000000000..04a6ba9cdcd --- /dev/null +++ b/src/gallium/state_trackers/vega/paint.c @@ -0,0 +1,699 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "paint.h" + +#include "shaders_cache.h" +#include "matrix.h" +#include "image.h" +#include "st_inlines.h" + +#include "pipe/p_compiler.h" +#include "pipe/p_inlines.h" + +#include "util/u_memory.h" +#include "util/u_math.h" + +#include "cso_cache/cso_context.h" + +struct vg_paint { + struct vg_object base; + + VGPaintType type; + + struct { + VGfloat color[4]; + VGint colori[4]; + } solid; + + struct { + VGColorRampSpreadMode spread; + VGuint color_data[1024]; + struct { + VGfloat coords[4]; + VGint coordsi[4]; + } linear; + struct { + VGfloat vals[5]; + VGint valsi[5]; + } radial; + struct pipe_texture *texture; + struct pipe_sampler_state sampler; + + VGfloat *ramp_stops; + VGint *ramp_stopsi; + VGint num_stops; + + VGboolean color_ramps_premultiplied; + } gradient; + + struct { + struct pipe_texture *texture; + VGTilingMode tiling_mode; + struct pipe_sampler_state sampler; + } pattern; + + struct pipe_constant_buffer cbuf; + struct pipe_shader_state fs_state; + void *fs; +}; + +static INLINE VGuint mix_pixels(VGuint p1, VGuint a, VGuint p2, VGuint b) +{ + VGuint t = (p1 & 0xff00ff) * a + (p2 & 0xff00ff) * b; + t >>= 8; t &= 0xff00ff; + + p1 = ((p1 >> 8) & 0xff00ff) * a + ((p2 >> 8) & 0xff00ff) * b; + p1 &= 0xff00ff00; p1 |= t; + + return p1; +} + +static INLINE VGuint float4_to_argb(const VGfloat *clr) +{ + return float_to_ubyte(clr[3]) << 24 | + float_to_ubyte(clr[0]) << 16 | + float_to_ubyte(clr[1]) << 8 | + float_to_ubyte(clr[2]) << 0; +} + +static INLINE void create_gradient_data(const VGfloat *ramp_stops, + VGint num, + VGuint *data, + VGint size) +{ + VGint i; + VGint pos = 0; + VGfloat fpos = 0, incr = 1.f / size; + VGuint last_color; + + while (fpos < ramp_stops[0]) { + data[pos] = float4_to_argb(ramp_stops + 1); + fpos += incr; + ++pos; + } + + for (i = 0; i < num - 1; ++i) { + VGint rcur = 5 * i; + VGint rnext = 5 * (i + 1); + VGfloat delta = 1.f/(ramp_stops[rnext] - ramp_stops[rcur]); + while (fpos < ramp_stops[rnext] && pos < size) { + VGint dist = 256 * ((fpos - ramp_stops[rcur]) * delta); + VGint idist = 256 - dist; + VGuint current_color = float4_to_argb(ramp_stops + rcur + 1); + VGuint next_color = float4_to_argb(ramp_stops + rnext + 1); + data[pos] = mix_pixels(current_color, idist, + next_color, dist); + fpos += incr; + ++pos; + } + } + + last_color = float4_to_argb(ramp_stops + ((num - 1) * 5 + 1)); + while (pos < size) { + data[pos] = last_color; + ++pos; + } + data[size-1] = last_color; +} + +static INLINE struct pipe_texture *create_gradient_texture(struct vg_paint *p) +{ + struct pipe_context *pipe = p->base.ctx->pipe; + struct pipe_screen *screen = pipe->screen; + struct pipe_texture *tex = 0; + struct pipe_texture templ; + + memset(&templ, 0, sizeof(templ)); + templ.target = PIPE_TEXTURE_1D; + templ.format = PIPE_FORMAT_A8R8G8B8_UNORM; + templ.last_level = 0; + templ.width[0] = 1024; + templ.height[0] = 1; + templ.depth[0] = 1; + pf_get_block(PIPE_FORMAT_A8R8G8B8_UNORM, &templ.block); + templ.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER; + + tex = screen->texture_create(screen, &templ); + + { /* upload color_data */ + struct pipe_transfer *transfer = + st_no_flush_get_tex_transfer(p->base.ctx, tex, 0, 0, 0, + PIPE_TRANSFER_WRITE, 0, 0, 1024, 1); + void *map = screen->transfer_map(screen, transfer); + memcpy(map, p->gradient.color_data, sizeof(VGint)*1024); + screen->transfer_unmap(screen, transfer); + screen->tex_transfer_destroy(transfer); + } + + return tex; +} + +struct vg_paint * paint_create(struct vg_context *ctx) +{ + struct vg_paint *paint = CALLOC_STRUCT(vg_paint); + const VGfloat default_color[] = {0.0f, 0.0f, 0.0f, 1.0f}; + const VGfloat def_ling[] = {0.0f, 0.0f, 1.0f, 0.0f}; + const VGfloat def_radg[] = {0.0f, 0.0f, 0.0f, 0.0f, 1.0f}; + vg_init_object(&paint->base, ctx, VG_OBJECT_PAINT); + vg_context_add_object(ctx, VG_OBJECT_PAINT, paint); + + paint->type = VG_PAINT_TYPE_COLOR; + memcpy(paint->solid.color, default_color, + 4 * sizeof(VGfloat)); + paint->gradient.spread = VG_COLOR_RAMP_SPREAD_PAD; + memcpy(paint->gradient.linear.coords, def_ling, + 4 * sizeof(VGfloat)); + memcpy(paint->gradient.radial.vals, def_radg, + 5 * sizeof(VGfloat)); + + paint->gradient.sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + paint->gradient.sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + paint->gradient.sampler.min_img_filter = PIPE_TEX_MIPFILTER_NEAREST; + paint->gradient.sampler.mag_img_filter = PIPE_TEX_MIPFILTER_NEAREST; + paint->gradient.sampler.normalized_coords = 1; + + memcpy(&paint->pattern.sampler, + &paint->gradient.sampler, + sizeof(struct pipe_sampler_state)); + + return paint; +} + +void paint_destroy(struct vg_paint *paint) +{ + struct vg_context *ctx = paint->base.ctx; + if (paint->pattern.texture) + pipe_texture_reference(&paint->pattern.texture, NULL); + if (ctx) + vg_context_remove_object(ctx, VG_OBJECT_PAINT, paint); + + free(paint->gradient.ramp_stopsi); + free(paint->gradient.ramp_stops); + free(paint); +} + +void paint_set_color(struct vg_paint *paint, + const VGfloat *color) +{ + paint->solid.color[0] = color[0]; + paint->solid.color[1] = color[1]; + paint->solid.color[2] = color[2]; + paint->solid.color[3] = color[3]; + + paint->solid.colori[0] = FLT_TO_INT(color[0]); + paint->solid.colori[1] = FLT_TO_INT(color[1]); + paint->solid.colori[2] = FLT_TO_INT(color[2]); + paint->solid.colori[3] = FLT_TO_INT(color[3]); +} + +static INLINE void paint_color_buffer(struct vg_paint *paint, void *buffer) +{ + VGfloat *map = (VGfloat*)buffer; + memcpy(buffer, paint->solid.color, 4 * sizeof(VGfloat)); + map[4] = 0.f; + map[5] = 1.f; + map[6] = 2.f; + map[7] = 4.f; +} + +static INLINE void paint_linear_gradient_buffer(struct vg_paint *paint, void *buffer) +{ + struct vg_context *ctx = paint->base.ctx; + VGfloat *map = (VGfloat*)buffer; + + map[0] = paint->gradient.linear.coords[2] - paint->gradient.linear.coords[0]; + map[1] = paint->gradient.linear.coords[3] - paint->gradient.linear.coords[1]; + map[2] = 1.f / (map[0] * map[0] + map[1] * map[1]); + map[3] = 1.f; + + map[4] = 0.f; + map[5] = 1.f; + map[6] = 2.f; + map[7] = 4.f; + { + struct matrix mat; + struct matrix inv; + matrix_load_identity(&mat); + matrix_translate(&mat, -paint->gradient.linear.coords[0], -paint->gradient.linear.coords[1]); + memcpy(&inv, &ctx->state.vg.fill_paint_to_user_matrix, + sizeof(struct matrix)); + matrix_invert(&inv); + matrix_mult(&inv, &mat); + memcpy(&mat, &inv, + sizeof(struct matrix)); + + map[8] = mat.m[0]; map[9] = mat.m[3]; map[10] = mat.m[6]; map[11] = 0.f; + map[12] = mat.m[1]; map[13] = mat.m[4]; map[14] = mat.m[7]; map[15] = 0.f; + map[16] = mat.m[2]; map[17] = mat.m[5]; map[18] = mat.m[8]; map[19] = 0.f; + } +#if 0 + debug_printf("Coords (%f, %f, %f, %f)\n", + map[0], map[1], map[2], map[3]); +#endif +} + + +static INLINE void paint_radial_gradient_buffer(struct vg_paint *paint, void *buffer) +{ + VGfloat *radialCoords = paint->gradient.radial.vals; + struct vg_context *ctx = paint->base.ctx; + + VGfloat *map = (VGfloat*)buffer; + + map[0] = radialCoords[0] - radialCoords[2]; + map[1] = radialCoords[1] - radialCoords[3]; + map[2] = -map[0] * map[0] - map[1] * map[1] + + radialCoords[4] * radialCoords[4]; + map[3] = 1.f; + + map[4] = 0.f; + map[5] = 1.f; + map[6] = 2.f; + map[7] = 4.f; + + { + struct matrix mat; + struct matrix inv; + matrix_load_identity(&mat); + matrix_translate(&mat, -radialCoords[2], -radialCoords[3]); + memcpy(&inv, &ctx->state.vg.fill_paint_to_user_matrix, + sizeof(struct matrix)); + matrix_invert(&inv); + matrix_mult(&inv, &mat); + memcpy(&mat, &inv, + sizeof(struct matrix)); + + map[8] = mat.m[0]; map[9] = mat.m[3]; map[10] = mat.m[6]; map[11] = 0.f; + map[12] = mat.m[1]; map[13] = mat.m[4]; map[14] = mat.m[7]; map[15] = 0.f; + map[16] = mat.m[2]; map[17] = mat.m[5]; map[18] = mat.m[8]; map[19] = 0.f; + } + +#if 0 + debug_printf("Coords (%f, %f, %f, %f)\n", + map[0], map[1], map[2], map[3]); +#endif +} + + +static INLINE void paint_pattern_buffer(struct vg_paint *paint, void *buffer) +{ + struct vg_context *ctx = paint->base.ctx; + + VGfloat *map = (VGfloat *)buffer; + memcpy(map, paint->solid.color, 4 * sizeof(VGfloat)); + + map[4] = 0.f; + map[5] = 1.f; + map[6] = paint->pattern.texture->width[0]; + map[7] = paint->pattern.texture->height[0]; + { + struct matrix mat; + memcpy(&mat, &ctx->state.vg.fill_paint_to_user_matrix, + sizeof(struct matrix)); + matrix_invert(&mat); + { + struct matrix pm; + memcpy(&pm, &ctx->state.vg.path_user_to_surface_matrix, + sizeof(struct matrix)); + matrix_invert(&pm); + matrix_mult(&pm, &mat); + memcpy(&mat, &pm, sizeof(struct matrix)); + } + + map[8] = mat.m[0]; map[9] = mat.m[3]; map[10] = mat.m[6]; map[11] = 0.f; + map[12] = mat.m[1]; map[13] = mat.m[4]; map[14] = mat.m[7]; map[15] = 0.f; + map[16] = mat.m[2]; map[17] = mat.m[5]; map[18] = mat.m[8]; map[19] = 0.f; + } +} + +void paint_set_type(struct vg_paint *paint, VGPaintType type) +{ + paint->type = type; +} + +void paint_set_ramp_stops(struct vg_paint *paint, const VGfloat *stops, + int num) +{ + const VGfloat default_stops[] = {0.0f, 0.0f, 0.0f, 0.0f, 1.0f, + 1.0f, 1.0f, 1.0f, 1.0f, 1.0f}; + VGint i; + const VGint num_stops = num / 5; + VGfloat last_coord; + + paint->gradient.num_stops = num; + if (num) { + free(paint->gradient.ramp_stops); + paint->gradient.ramp_stops = malloc(sizeof(VGfloat)*num); + memcpy(paint->gradient.ramp_stops, stops, sizeof(VGfloat)*num); + } else + return; + + /* stops must be in increasing order. the last stop is 1.0. if the + * first one is bigger than 1 then the whole sequence is invalid*/ + if (stops[0] > 1) { + stops = default_stops; + num = 10; + } + last_coord = stops[0]; + for (i = 1; i < num_stops; ++i) { + VGint idx = 5 * i; + VGfloat coord = stops[idx]; + if (!floatsEqual(last_coord, coord) && coord < last_coord) { + stops = default_stops; + num = 10; + break; + } + last_coord = coord; + } + + create_gradient_data(stops, num / 5, paint->gradient.color_data, + 1024); + + if (paint->gradient.texture) { + pipe_texture_reference(&paint->gradient.texture, NULL); + paint->gradient.texture = 0; + } + + paint->gradient.texture = create_gradient_texture(paint); +} + +void paint_set_colori(struct vg_paint *p, + VGuint rgba) +{ + p->solid.color[0] = ((rgba >> 24) & 0xff) / 255.f; + p->solid.color[1] = ((rgba >> 16) & 0xff) / 255.f; + p->solid.color[2] = ((rgba >> 8) & 0xff) / 255.f; + p->solid.color[3] = ((rgba >> 0) & 0xff) / 255.f; +} + +VGuint paint_colori(struct vg_paint *p) +{ +#define F2B(f) (float_to_ubyte(f)) + + return ((F2B(p->solid.color[0]) << 24) | + (F2B(p->solid.color[1]) << 16) | + (F2B(p->solid.color[2]) << 8) | + (F2B(p->solid.color[3]) << 0)); +#undef F2B +} + +void paint_set_linear_gradient(struct vg_paint *paint, + const VGfloat *coords) +{ + memcpy(paint->gradient.linear.coords, coords, sizeof(VGfloat) * 4); +} + +void paint_set_spread_mode(struct vg_paint *paint, + VGint mode) +{ + paint->gradient.spread = mode; + switch(mode) { + case VG_COLOR_RAMP_SPREAD_PAD: + paint->gradient.sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + break; + case VG_COLOR_RAMP_SPREAD_REPEAT: + paint->gradient.sampler.wrap_s = PIPE_TEX_WRAP_REPEAT; + break; + case VG_COLOR_RAMP_SPREAD_REFLECT: + paint->gradient.sampler.wrap_s = PIPE_TEX_WRAP_MIRROR_REPEAT; + break; + } +} + +VGColorRampSpreadMode paint_spread_mode(struct vg_paint *paint) +{ + return paint->gradient.spread; +} + +void paint_set_radial_gradient(struct vg_paint *paint, + const VGfloat *values) +{ + memcpy(paint->gradient.radial.vals, values, sizeof(VGfloat) * 5); +} + +void paint_set_pattern(struct vg_paint *paint, + struct vg_image *img) +{ + if (paint->pattern.texture) + pipe_texture_reference(&paint->pattern.texture, NULL); + + paint->pattern.texture = 0; + pipe_texture_reference(&paint->pattern.texture, + img->texture); +} + +void paint_set_pattern_tiling(struct vg_paint *paint, + VGTilingMode mode) +{ + paint->pattern.tiling_mode = mode; + + switch(mode) { + case VG_TILE_FILL: + paint->pattern.sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_BORDER; + paint->pattern.sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_BORDER; + break; + case VG_TILE_PAD: + paint->pattern.sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + paint->pattern.sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + break; + case VG_TILE_REPEAT: + paint->pattern.sampler.wrap_s = PIPE_TEX_WRAP_REPEAT; + paint->pattern.sampler.wrap_t = PIPE_TEX_WRAP_REPEAT; + break; + case VG_TILE_REFLECT: + paint->pattern.sampler.wrap_s = PIPE_TEX_WRAP_MIRROR_REPEAT; + paint->pattern.sampler.wrap_t = PIPE_TEX_WRAP_MIRROR_REPEAT; + break; + default: + debug_assert("!Unknown tiling mode"); + } +} + +void paint_get_color(struct vg_paint *paint, + VGfloat *color) +{ + color[0] = paint->solid.color[0]; + color[1] = paint->solid.color[1]; + color[2] = paint->solid.color[2]; + color[3] = paint->solid.color[3]; +} + +void paint_ramp_stops(struct vg_paint *paint, VGfloat *stops, + int num) +{ + memcpy(stops, paint->gradient.ramp_stops, sizeof(VGfloat)*num); +} + +void paint_linear_gradient(struct vg_paint *paint, + VGfloat *coords) +{ + memcpy(coords, paint->gradient.linear.coords, sizeof(VGfloat)*4); +} + +void paint_radial_gradient(struct vg_paint *paint, + VGfloat *coords) +{ + memcpy(coords, paint->gradient.radial.vals, sizeof(VGfloat)*5); +} + +int paint_num_ramp_stops(struct vg_paint *paint) +{ + return paint->gradient.num_stops; +} + +VGPaintType paint_type(struct vg_paint *paint) +{ + return paint->type; +} + +void paint_set_coloriv(struct vg_paint *paint, + const VGint *color) +{ + paint->solid.color[0] = color[0]; + paint->solid.color[1] = color[1]; + paint->solid.color[2] = color[2]; + paint->solid.color[3] = color[3]; + + paint->solid.colori[0] = color[0]; + paint->solid.colori[1] = color[1]; + paint->solid.colori[2] = color[2]; + paint->solid.colori[3] = color[3]; +} + +void paint_get_coloriv(struct vg_paint *paint, + VGint *color) +{ + color[0] = paint->solid.colori[0]; + color[1] = paint->solid.colori[1]; + color[2] = paint->solid.colori[2]; + color[3] = paint->solid.colori[3]; +} + +void paint_set_color_ramp_premultiplied(struct vg_paint *paint, + VGboolean set) +{ + paint->gradient.color_ramps_premultiplied = set; +} + +VGboolean paint_color_ramp_premultiplied(struct vg_paint *paint) +{ + return paint->gradient.color_ramps_premultiplied; +} + +void paint_set_ramp_stopsi(struct vg_paint *paint, const VGint *stops, + int num) +{ + if (num) { + free(paint->gradient.ramp_stopsi); + paint->gradient.ramp_stopsi = malloc(sizeof(VGint)*num); + memcpy(paint->gradient.ramp_stopsi, stops, sizeof(VGint)*num); + } +} + +void paint_ramp_stopsi(struct vg_paint *paint, VGint *stops, + int num) +{ + memcpy(stops, paint->gradient.ramp_stopsi, sizeof(VGint)*num); +} + +void paint_set_linear_gradienti(struct vg_paint *paint, + const VGint *coords) +{ + memcpy(paint->gradient.linear.coordsi, coords, sizeof(VGint) * 4); +} + +void paint_linear_gradienti(struct vg_paint *paint, + VGint *coords) +{ + memcpy(coords, paint->gradient.linear.coordsi, sizeof(VGint)*4); +} + +void paint_set_radial_gradienti(struct vg_paint *paint, + const VGint *values) +{ + memcpy(paint->gradient.radial.valsi, values, sizeof(VGint) * 5); +} + +void paint_radial_gradienti(struct vg_paint *paint, + VGint *coords) +{ + memcpy(coords, paint->gradient.radial.valsi, sizeof(VGint)*5); +} + +VGTilingMode paint_pattern_tiling(struct vg_paint *paint) +{ + return paint->pattern.tiling_mode; +} + +VGint paint_bind_samplers(struct vg_paint *paint, struct pipe_sampler_state **samplers, + struct pipe_texture **textures) +{ + struct vg_context *ctx = vg_current_context(); + + switch(paint->type) { + case VG_PAINT_TYPE_LINEAR_GRADIENT: + case VG_PAINT_TYPE_RADIAL_GRADIENT: { + if (paint->gradient.texture) { + paint->gradient.sampler.min_img_filter = image_sampler_filter(ctx); + paint->gradient.sampler.mag_img_filter = image_sampler_filter(ctx); + samplers[0] = &paint->gradient.sampler; + textures[0] = paint->gradient.texture; + return 1; + } + } + break; + case VG_PAINT_TYPE_PATTERN: { + memcpy(paint->pattern.sampler.border_color, + ctx->state.vg.tile_fill_color, + sizeof(VGfloat) * 4); + paint->pattern.sampler.min_img_filter = image_sampler_filter(ctx); + paint->pattern.sampler.mag_img_filter = image_sampler_filter(ctx); + samplers[0] = &paint->pattern.sampler; + textures[0] = paint->pattern.texture; + return 1; + } + break; + default: + samplers[0] = &paint->pattern.sampler; /* dummy */ + textures[0] = 0; + return 0; + break; + } + return 0; +} + +void paint_resolve_type(struct vg_paint *paint) +{ + if (paint->type == VG_PAINT_TYPE_PATTERN && + !paint->pattern.texture) { + paint->type = VG_PAINT_TYPE_COLOR; + } +} + +VGint paint_constant_buffer_size(struct vg_paint *paint) +{ + switch(paint->type) { + case VG_PAINT_TYPE_COLOR: + return 8 * sizeof(VGfloat);/*4 color + 4 constants (0.f,1.f,2.f,4.f)*/ + break; + case VG_PAINT_TYPE_LINEAR_GRADIENT: + return 20 * sizeof(VGfloat); + break; + case VG_PAINT_TYPE_RADIAL_GRADIENT: + return 20 * sizeof(VGfloat); + break; + case VG_PAINT_TYPE_PATTERN: + return 20 * sizeof(VGfloat); + break; + default: + debug_printf("Uknown paint type: %d\n", paint->type); + } + + return 0; +} + +void paint_fill_constant_buffer(struct vg_paint *paint, + void *buffer) +{ + switch(paint->type) { + case VG_PAINT_TYPE_COLOR: + paint_color_buffer(paint, buffer); + break; + case VG_PAINT_TYPE_LINEAR_GRADIENT: + paint_linear_gradient_buffer(paint, buffer); + break; + case VG_PAINT_TYPE_RADIAL_GRADIENT: + paint_radial_gradient_buffer(paint, buffer); + break; + case VG_PAINT_TYPE_PATTERN: + paint_pattern_buffer(paint, buffer); + break; + + default: + abort(); + } +} diff --git a/src/gallium/state_trackers/vega/paint.h b/src/gallium/state_trackers/vega/paint.h new file mode 100644 index 00000000000..999b5c167ca --- /dev/null +++ b/src/gallium/state_trackers/vega/paint.h @@ -0,0 +1,118 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef PAINT_H +#define PAINT_H + +#include "vg_context.h" + +#include "VG/openvg.h" +#include "pipe/p_state.h" + +struct vg_paint; +struct vg_image; +struct pipe_sampler_state; +struct pipe_texture; + +struct vg_paint *paint_create(struct vg_context *ctx); +void paint_destroy(struct vg_paint *paint); + +void paint_set_color(struct vg_paint *paint, + const VGfloat *color); +void paint_get_color(struct vg_paint *paint, + VGfloat *color); + +void paint_set_coloriv(struct vg_paint *paint, + const VGint *color); +void paint_get_coloriv(struct vg_paint *paint, + VGint *color); + +void paint_set_colori(struct vg_paint *paint, + VGuint rgba); + +VGuint paint_colori(struct vg_paint *paint); + +void paint_set_type(struct vg_paint *paint, VGPaintType type); +VGPaintType paint_type(struct vg_paint *paint); +void paint_resolve_type(struct vg_paint *paint); + +void paint_set_linear_gradient(struct vg_paint *paint, + const VGfloat *coords); +void paint_linear_gradient(struct vg_paint *paint, + VGfloat *coords); +void paint_set_linear_gradienti(struct vg_paint *paint, + const VGint *coords); +void paint_linear_gradienti(struct vg_paint *paint, + VGint *coords); + + +void paint_set_radial_gradient(struct vg_paint *paint, + const VGfloat *values); +void paint_radial_gradient(struct vg_paint *paint, + VGfloat *coords); +void paint_set_radial_gradienti(struct vg_paint *paint, + const VGint *values); +void paint_radial_gradienti(struct vg_paint *paint, + VGint *coords); + + +void paint_set_ramp_stops(struct vg_paint *paint, const VGfloat *stops, + int num); +void paint_ramp_stops(struct vg_paint *paint, VGfloat *stops, + int num); + +void paint_set_ramp_stopsi(struct vg_paint *paint, const VGint *stops, + int num); +void paint_ramp_stopsi(struct vg_paint *paint, VGint *stops, + int num); + +int paint_num_ramp_stops(struct vg_paint *paint); + +void paint_set_spread_mode(struct vg_paint *paint, + VGint mode); +VGColorRampSpreadMode paint_spread_mode(struct vg_paint *paint); + + +void paint_set_pattern(struct vg_paint *paint, + struct vg_image *img); +void paint_set_pattern_tiling(struct vg_paint *paint, + VGTilingMode mode); +VGTilingMode paint_pattern_tiling(struct vg_paint *paint); + +void paint_set_color_ramp_premultiplied(struct vg_paint *paint, + VGboolean set); +VGboolean paint_color_ramp_premultiplied(struct vg_paint *paint); + + +VGint paint_bind_samplers(struct vg_paint *paint, struct pipe_sampler_state **samplers, + struct pipe_texture **textures); + +VGint paint_constant_buffer_size(struct vg_paint *paint); +void paint_fill_constant_buffer(struct vg_paint *paint, + void *buffer); + + +#endif diff --git a/src/gallium/state_trackers/vega/path.c b/src/gallium/state_trackers/vega/path.c new file mode 100644 index 00000000000..4fc23a7a278 --- /dev/null +++ b/src/gallium/state_trackers/vega/path.c @@ -0,0 +1,2034 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "path.h" + +#include "stroker.h" +#include "polygon.h" +#include "bezier.h" +#include "matrix.h" +#include "vg_context.h" +#include "util_array.h" +#include "arc.h" +#include "path_utils.h" +#include "paint.h" +#include "shader.h" + +#include "util/u_memory.h" + +#include <assert.h> + +#define DEBUG_PATH 0 + +struct path { + struct vg_object base; + VGbitfield caps; + VGboolean dirty; + VGboolean dirty_stroke; + + VGPathDatatype datatype; + + VGfloat scale; + VGfloat bias; + + VGint num_segments; + + struct array * segments; + struct array * control_points; + + struct { + struct polygon_array polygon_array; + struct matrix matrix; + } fill_polys; + + struct { + struct path *path; + struct matrix matrix; + VGfloat stroke_width; + VGfloat miter_limit; + VGCapStyle cap_style; + VGJoinStyle join_style; + } stroked; +}; + + +static INLINE void data_at(void **data, + struct path *p, + VGint start, VGint count, + VGfloat *out) +{ + VGPathDatatype dt = p->datatype; + VGint i; + VGint end = start + count; + VGfloat *itr = out; + + switch(dt) { + case VG_PATH_DATATYPE_S_8: { + VGbyte **bdata = (VGbyte **)data; + for (i = start; i < end; ++i) { + *itr = (*bdata)[i]; + ++itr; + } + *bdata += count; + } + break; + case VG_PATH_DATATYPE_S_16: { + VGshort **bdata = (VGshort **)data; + for (i = start; i < end; ++i) { + *itr = (*bdata)[i]; + ++itr; + } + *bdata += count; + } + break; + case VG_PATH_DATATYPE_S_32: { + VGint **bdata = (VGint **)data; + for (i = start; i < end; ++i) { + *itr = (*bdata)[i]; + ++itr; + } + *bdata += count; + } + break; + case VG_PATH_DATATYPE_F: { + VGfloat **fdata = (VGfloat **)data; + for (i = start; i < end; ++i) { + *itr = (*fdata)[i]; + ++itr; + } + *fdata += count; + } + break; + default: + debug_assert(!"Unknown path datatype!"); + } +} + + +void vg_float_to_datatype(VGPathDatatype datatype, + VGubyte *common_data, + const VGfloat *data, + VGint num_coords) +{ + VGint i; + switch(datatype) { + case VG_PATH_DATATYPE_S_8: { + for (i = 0; i < num_coords; ++i) { + common_data[i] = (VGubyte)data[i]; + } + } + break; + case VG_PATH_DATATYPE_S_16: { + VGshort *buf = (VGshort*)common_data; + for (i = 0; i < num_coords; ++i) { + buf[i] = (VGshort)data[i]; + } + } + break; + case VG_PATH_DATATYPE_S_32: { + VGint *buf = (VGint*)common_data; + for (i = 0; i < num_coords; ++i) { + buf[i] = (VGint)data[i]; + } + } + break; + case VG_PATH_DATATYPE_F: { + memcpy(common_data, data, sizeof(VGfloat) * num_coords); + } + break; + default: + debug_assert(!"Unknown path datatype!"); + } +} + +static void coords_adjust_by_scale_bias(struct path *p, + void *pdata, VGint num_coords, + VGfloat scale, VGfloat bias, + VGPathDatatype datatype) +{ + VGfloat data[8]; + void *coords = (VGfloat *)pdata; + VGubyte *common_data = (VGubyte *)pdata; + VGint size_dst = size_for_datatype(datatype); + VGint i; + + for (i = 0; i < num_coords; ++i) { + data_at(&coords, p, 0, 1, data); + data[0] = data[0] * scale + bias; + vg_float_to_datatype(datatype, common_data, data, 1); + common_data += size_dst; + } +} + +struct path * path_create(VGPathDatatype dt, VGfloat scale, VGfloat bias, + VGint segmentCapacityHint, + VGint coordCapacityHint, + VGbitfield capabilities) +{ + struct path *path = CALLOC_STRUCT(path); + + vg_init_object(&path->base, vg_current_context(), VG_OBJECT_PATH); + path->caps = capabilities & VG_PATH_CAPABILITY_ALL; + vg_context_add_object(vg_current_context(), VG_OBJECT_PATH, path); + + path->datatype = dt; + path->scale = scale; + path->bias = bias; + + path->segments = array_create(size_for_datatype(VG_PATH_DATATYPE_S_8)); + path->control_points = array_create(size_for_datatype(dt)); + + path->dirty = VG_TRUE; + path->dirty_stroke = VG_TRUE; + + return path; +} + +void path_destroy(struct path *p) +{ + vg_context_remove_object(vg_current_context(), VG_OBJECT_PATH, p); + + array_destroy(p->segments); + array_destroy(p->control_points); + array_destroy(p->fill_polys.polygon_array.array); + + if (p->stroked.path) + path_destroy(p->stroked.path); + + free(p); +} + +VGbitfield path_capabilities(struct path *p) +{ + return p->caps; +} + +void path_set_capabilities(struct path *p, VGbitfield bf) +{ + p->caps = (bf & VG_PATH_CAPABILITY_ALL); +} + +void path_append_data(struct path *p, + VGint numSegments, + const VGubyte * pathSegments, + const void * pathData) +{ + VGint old_segments = p->num_segments; + VGint num_new_coords = num_elements_for_segments(pathSegments, numSegments); + array_append_data(p->segments, pathSegments, numSegments); + array_append_data(p->control_points, pathData, num_new_coords); + + p->num_segments += numSegments; + if (!floatsEqual(p->scale, 1.f) || !floatsEqual(p->bias, 0.f)) { + VGubyte *coords = (VGubyte*)p->control_points->data; + coords_adjust_by_scale_bias(p, + coords + old_segments * p->control_points->datatype_size, + num_new_coords, + p->scale, p->bias, p->datatype); + } + p->dirty = VG_TRUE; + p->dirty_stroke = VG_TRUE; +} + +VGint path_num_segments(struct path *p) +{ + return p->num_segments; +} + +static INLINE void map_if_relative(VGfloat ox, VGfloat oy, + VGboolean relative, + VGfloat *x, VGfloat *y) +{ + if (relative) { + if (x) + *x += ox; + if (y) + *y += oy; + } +} + +static INLINE void close_polygon(struct polygon *current, + VGfloat sx, VGfloat sy, + VGfloat ox, VGfloat oy, + struct matrix *matrix) +{ + if (!floatsEqual(sx, ox) || + !floatsEqual(sy, oy)) { + VGfloat x0 = sx; + VGfloat y0 = sy; + matrix_map_point(matrix, x0, y0, &x0, &y0); + polygon_vertex_append(current, x0, y0); + } +} + +static void convert_path(struct path *p, + VGPathDatatype to, + void *dst, + VGint num_coords) +{ + VGfloat data[8]; + void *coords = (VGfloat *)p->control_points->data; + VGubyte *common_data = (VGubyte *)dst; + VGint size_dst = size_for_datatype(to); + VGint i; + + for (i = 0; i < num_coords; ++i) { + data_at(&coords, p, 0, 1, data); + vg_float_to_datatype(to, common_data, data, 1); + common_data += size_dst; + } +} + + +static void polygon_array_calculate_bounds( struct polygon_array *polyarray ) +{ + struct array *polys = polyarray->array; + VGfloat min_x, max_x; + VGfloat min_y, max_y; + VGfloat bounds[4]; + unsigned i; + + assert(polys); + assert(polys->num_elements); + polygon_bounding_rect((((struct polygon**)polys->data)[0]), bounds); + min_x = bounds[0]; + min_y = bounds[1]; + max_x = bounds[0] + bounds[2]; + max_y = bounds[1] + bounds[3]; + for (i = 1; i < polys->num_elements; ++i) { + struct polygon *p = (((struct polygon**)polys->data)[i]); + polygon_bounding_rect(p, bounds); + min_x = MIN2(min_x, bounds[0]); + min_y = MIN2(min_y, bounds[1]); + max_x = MAX2(max_x, bounds[0] + bounds[2]); + max_y = MAX2(max_y, bounds[1] + bounds[3]); + } + + polyarray->min_x = min_x; + polyarray->min_y = min_y; + polyarray->max_x = max_x; + polyarray->max_y = max_y; +} + + +static struct polygon_array * path_get_fill_polygons(struct path *p, struct matrix *matrix) +{ + VGint i; + struct polygon *current = 0; + VGfloat sx, sy, px, py, ox, oy; + VGfloat x0, y0, x1, y1, x2, y2, x3, y3; + VGfloat data[8]; + void *coords = (VGfloat *)p->control_points->data; + struct array *array; + + if (p->fill_polys.polygon_array.array) + { + if (memcmp( &p->fill_polys.matrix, + matrix, + sizeof *matrix ) == 0 && p->dirty == VG_FALSE) + { + return &p->fill_polys.polygon_array; + } + else { + array_destroy( p->fill_polys.polygon_array.array ); + p->fill_polys.polygon_array.array = NULL; + } + } + + array = array_create(sizeof(struct array*)); + + sx = sy = px = py = ox = oy = 0.f; + + current = polygon_create(32); + + for (i = 0; i < p->num_segments; ++i) { + VGubyte segment = ((VGubyte*)(p->segments->data))[i]; + VGint command = SEGMENT_COMMAND(segment); + VGboolean relative = SEGMENT_ABS_REL(segment); + + switch(command) { + case VG_CLOSE_PATH: + close_polygon(current, sx, sy, ox, oy, matrix); + ox = sx; + oy = sy; + break; + case VG_MOVE_TO: + if (current && polygon_vertex_count(current) > 0) { + /* add polygon */ + close_polygon(current, sx, sy, ox, oy, matrix); + array_append_data(array, ¤t, 1); + current = polygon_create(32); + } + data_at(&coords, p, 0, 2, data); + x0 = data[0]; + y0 = data[1]; + map_if_relative(ox, oy, relative, &x0, &y0); + sx = x0; + sy = y0; + ox = x0; + oy = y0; + px = x0; + py = y0; + matrix_map_point(matrix, x0, y0, &x0, &y0); + polygon_vertex_append(current, x0, y0); + break; + case VG_LINE_TO: + data_at(&coords, p, 0, 2, data); + x0 = data[0]; + y0 = data[1]; + map_if_relative(ox, oy, relative, &x0, &y0); + ox = x0; + oy = y0; + px = x0; + py = y0; + matrix_map_point(matrix, x0, y0, &x0, &y0); + polygon_vertex_append(current, x0, y0); + break; + case VG_HLINE_TO: + data_at(&coords, p, 0, 1, data); + x0 = data[0]; + y0 = oy; + map_if_relative(ox, oy, relative, &x0, 0); + ox = x0; + px = x0; + py = y0; + matrix_map_point(matrix, x0, y0, &x0, &y0); + polygon_vertex_append(current, x0, y0); + break; + case VG_VLINE_TO: + data_at(&coords, p, 0, 1, data); + x0 = ox; + y0 = data[0]; + map_if_relative(ox, oy, relative, 0, &y0); + oy = y0; + px = x0; + py = y0; + matrix_map_point(matrix, x0, y0, &x0, &y0); + polygon_vertex_append(current, x0, y0); + break; + case VG_CUBIC_TO: { + struct bezier bezier; + data_at(&coords, p, 0, 6, data); + x0 = ox; + y0 = oy; + x1 = data[0]; + y1 = data[1]; + x2 = data[2]; + y2 = data[3]; + x3 = data[4]; + y3 = data[5]; + map_if_relative(ox, oy, relative, &x1, &y1); + map_if_relative(ox, oy, relative, &x2, &y2); + map_if_relative(ox, oy, relative, &x3, &y3); + ox = x3; + oy = y3; + px = x2; + py = y2; + assert(matrix_is_affine(matrix)); + matrix_map_point(matrix, x0, y0, &x0, &y0); + matrix_map_point(matrix, x1, y1, &x1, &y1); + matrix_map_point(matrix, x2, y2, &x2, &y2); + matrix_map_point(matrix, x3, y3, &x3, &y3); + bezier_init(&bezier, x0, y0, x1, y1, + x2, y2, x3, y3); + bezier_add_to_polygon(&bezier, current); + } + break; + case VG_QUAD_TO: { + struct bezier bezier; + data_at(&coords, p, 0, 4, data); + x0 = ox; + y0 = oy; + x1 = data[0]; + y1 = data[1]; + x3 = data[2]; + y3 = data[3]; + map_if_relative(ox, oy, relative, &x1, &y1); + map_if_relative(ox, oy, relative, &x3, &y3); + px = x1; + py = y1; + { /* form a cubic out of it */ + x2 = (x3 + 2*x1) / 3.f; + y2 = (y3 + 2*y1) / 3.f; + x1 = (x0 + 2*x1) / 3.f; + y1 = (y0 + 2*y1) / 3.f; + } + ox = x3; + oy = y3; + assert(matrix_is_affine(matrix)); + matrix_map_point(matrix, x0, y0, &x0, &y0); + matrix_map_point(matrix, x1, y1, &x1, &y1); + matrix_map_point(matrix, x2, y2, &x2, &y2); + matrix_map_point(matrix, x3, y3, &x3, &y3); + bezier_init(&bezier, x0, y0, x1, y1, + x2, y2, x3, y3); + bezier_add_to_polygon(&bezier, current); + } + break; + case VG_SQUAD_TO: { + struct bezier bezier; + data_at(&coords, p, 0, 2, data); + x0 = ox; + y0 = oy; + x1 = 2*ox-px; + y1 = 2*oy-py; + x3 = data[0]; + y3 = data[1]; + map_if_relative(ox, oy, relative, &x3, &y3); + px = x1; + py = y1; + { /* form a cubic out of it */ + x2 = (x3 + 2*x1) / 3.f; + y2 = (y3 + 2*y1) / 3.f; + x1 = (x0 + 2*x1) / 3.f; + y1 = (y0 + 2*y1) / 3.f; + } + ox = x3; + oy = y3; + assert(matrix_is_affine(matrix)); + matrix_map_point(matrix, x0, y0, &x0, &y0); + matrix_map_point(matrix, x1, y1, &x1, &y1); + matrix_map_point(matrix, x2, y2, &x2, &y2); + matrix_map_point(matrix, x3, y3, &x3, &y3); + bezier_init(&bezier, x0, y0, x1, y1, + x2, y2, x3, y3); + bezier_add_to_polygon(&bezier, current); + } + break; + case VG_SCUBIC_TO: { + struct bezier bezier; + data_at(&coords, p, 0, 4, data); + x0 = ox; + y0 = oy; + x1 = 2*ox-px; + y1 = 2*oy-py; + x2 = data[0]; + y2 = data[1]; + x3 = data[2]; + y3 = data[3]; + map_if_relative(ox, oy, relative, &x2, &y2); + map_if_relative(ox, oy, relative, &x3, &y3); + ox = x3; + oy = y3; + px = x2; + py = y2; + assert(matrix_is_affine(matrix)); + matrix_map_point(matrix, x0, y0, &x0, &y0); + matrix_map_point(matrix, x1, y1, &x1, &y1); + matrix_map_point(matrix, x2, y2, &x2, &y2); + matrix_map_point(matrix, x3, y3, &x3, &y3); + bezier_init(&bezier, x0, y0, x1, y1, + x2, y2, x3, y3); + bezier_add_to_polygon(&bezier, current); + } + break; + case VG_SCCWARC_TO: + case VG_SCWARC_TO: + case VG_LCCWARC_TO: + case VG_LCWARC_TO: { + VGfloat rh, rv, rot; + struct arc arc; + + data_at(&coords, p, 0, 5, data); + x0 = ox; + y0 = oy; + rh = data[0]; + rv = data[1]; + rot = data[2]; + x1 = data[3]; + y1 = data[4]; + map_if_relative(ox, oy, relative, &x1, &y1); +#if 0 + debug_printf("------- ARC (%f, %f), (%f, %f) %f, %f, %f\n", + x0, y0, x1, y1, rh, rv, rot); +#endif + arc_init(&arc, command, x0, y0, x1, y1, + rh, rv, rot); + arc_add_to_polygon(&arc, current, + matrix); + ox = x1; + oy = y1; + px = x1; + py = y1; + } + break; + default: + abort(); + assert(!"Unknown segment!"); + } + } + if (current) { + if (polygon_vertex_count(current) > 0) { + close_polygon(current, sx, sy, ox, oy, matrix); + array_append_data(array, ¤t, 1); + } else + polygon_destroy(current); + } + + p->fill_polys.polygon_array.array = array; + p->fill_polys.matrix = *matrix; + + polygon_array_calculate_bounds( &p->fill_polys.polygon_array ); + + p->dirty = VG_FALSE; + + return &p->fill_polys.polygon_array; +} + +VGbyte path_datatype_size(struct path *p) +{ + return size_for_datatype(p->datatype); +} + +VGPathDatatype path_datatype(struct path *p) +{ + return p->datatype; +} + +VGfloat path_scale(struct path *p) +{ + return p->scale; +} + +VGfloat path_bias(struct path *p) +{ + return p->bias; +} + +VGint path_num_coords(struct path *p) +{ + return num_elements_for_segments((VGubyte*)p->segments->data, + p->num_segments); +} + +void path_modify_coords(struct path *p, + VGint startIndex, + VGint numSegments, + const void * pathData) +{ + VGubyte *segments = (VGubyte*)(p->segments->data); + VGint count = num_elements_for_segments(&segments[startIndex], numSegments); + VGint start_cp = num_elements_for_segments(segments, startIndex); + + array_change_data(p->control_points, pathData, start_cp, count); + coords_adjust_by_scale_bias(p, + ((VGubyte*)p->control_points->data) + + (startIndex * p->control_points->datatype_size), + path_num_coords(p), + p->scale, p->bias, p->datatype); + p->dirty = VG_TRUE; + p->dirty_stroke = VG_TRUE; +} + +void path_for_each_segment(struct path *path, + path_for_each_cb cb, + void *user_data) +{ + VGint i; + struct path_for_each_data p; + VGfloat data[8]; + void *coords = (VGfloat *)path->control_points->data; + + p.coords = data; + p.sx = p.sy = p.px = p.py = p.ox = p.oy = 0.f; + p.user_data = user_data; + + for (i = 0; i < path->num_segments; ++i) { + VGint command; + VGboolean relative; + + p.segment = ((VGubyte*)(path->segments->data))[i]; + command = SEGMENT_COMMAND(p.segment); + relative = SEGMENT_ABS_REL(p.segment); + + switch(command) { + case VG_CLOSE_PATH: + cb(path, &p); + break; + case VG_MOVE_TO: + data_at(&coords, path, 0, 2, data); + map_if_relative(p.ox, p.oy, relative, &data[0], &data[1]); + cb(path, &p); + p.sx = data[0]; + p.sy = data[1]; + p.ox = data[0]; + p.oy = data[1]; + p.px = data[0]; + p.py = data[1]; + break; + case VG_LINE_TO: + data_at(&coords, path, 0, 2, data); + map_if_relative(p.ox, p.oy, relative, &data[0], &data[1]); + cb(path, &p); + p.ox = data[0]; + p.oy = data[1]; + p.px = data[0]; + p.py = data[1]; + break; + case VG_HLINE_TO: + data_at(&coords, path, 0, 1, data); + map_if_relative(p.ox, p.oy, relative, &data[0], 0); + p.segment = VG_LINE_TO; + data[1] = p.oy; + cb(path, &p); + p.ox = data[0]; + p.oy = data[1]; + p.px = data[0]; + p.py = data[1]; + break; + case VG_VLINE_TO: + data_at(&coords, path, 0, 1, data); + map_if_relative(p.ox, p.oy, relative, 0, &data[0]); + p.segment = VG_LINE_TO; + data[1] = data[0]; + data[0] = p.ox; + cb(path, &p); + p.ox = data[0]; + p.oy = data[1]; + p.px = data[0]; + p.py = data[1]; + break; + case VG_CUBIC_TO: { + data_at(&coords, path, 0, 6, data); + map_if_relative(p.ox, p.oy, relative, &data[0], &data[1]); + map_if_relative(p.ox, p.oy, relative, &data[2], &data[3]); + map_if_relative(p.ox, p.oy, relative, &data[4], &data[5]); + cb(path, &p); + p.px = data[2]; + p.py = data[3]; + p.ox = data[4]; + p.oy = data[5]; + } + break; + case VG_QUAD_TO: { + data_at(&coords, path, 0, 4, data); + map_if_relative(p.ox, p.oy, relative, &data[0], &data[1]); + map_if_relative(p.ox, p.oy, relative, &data[2], &data[3]); + cb(path, &p); + p.px = data[0]; + p.py = data[1]; + p.ox = data[2]; + p.oy = data[3]; + } + break; + case VG_SQUAD_TO: { + data_at(&coords, path, 0, 2, data); + map_if_relative(p.ox, p.oy, relative, &data[0], &data[1]); + cb(path, &p); + p.px = 2*p.ox-p.px; + p.py = 2*p.oy-p.py; + p.ox = data[2]; + p.oy = data[3]; + } + break; + case VG_SCUBIC_TO: { + data_at(&coords, path, 0, 4, data); + map_if_relative(p.ox, p.oy, relative, &data[0], &data[1]); + map_if_relative(p.ox, p.oy, relative, &data[2], &data[3]); + cb(path, &p); + p.px = data[0]; + p.py = data[1]; + p.ox = data[2]; + p.oy = data[3]; + } + break; + case VG_SCCWARC_TO: + case VG_SCWARC_TO: + case VG_LCCWARC_TO: + case VG_LCWARC_TO: { + data_at(&coords, path, 0, 5, data); + map_if_relative(p.ox, p.oy, relative, &data[3], &data[4]); +#if 0 + debug_printf("------- ARC (%f, %f), (%f, %f) %f, %f, %f\n", + p.ox, p.oy, data[3], data[4], data[0], data[1], data[2]); +#endif + cb(path, &p); + p.ox = data[3]; + p.oy = data[4]; + p.px = data[3]; + p.py = data[4]; + } + break; + default: + abort(); + assert(!"Unknown segment!"); + } + } +} + +struct transform_data { + struct array *segments; + struct array *coords; + + struct matrix *matrix; + + VGPathDatatype datatype; +}; + +static VGboolean transform_cb(struct path *p, + struct path_for_each_data *pd) +{ + struct transform_data *td = (struct transform_data *)pd->user_data; + VGint num_coords = num_elements_for_segments(&pd->segment, 1); + VGubyte segment = SEGMENT_COMMAND(pd->segment);/* abs bit is 0 */ + VGfloat data[8]; + VGubyte common_data[sizeof(VGfloat)*8]; + + memcpy(data, pd->coords, sizeof(VGfloat) * num_coords); + + switch(segment) { + case VG_CLOSE_PATH: + break; + case VG_MOVE_TO: + matrix_map_point(td->matrix, + data[0], data[1], &data[0], &data[1]); + break; + case VG_LINE_TO: + matrix_map_point(td->matrix, + data[0], data[1], &data[0], &data[1]); + break; + case VG_HLINE_TO: + case VG_VLINE_TO: + assert(0); + break; + case VG_QUAD_TO: + matrix_map_point(td->matrix, + data[0], data[1], &data[0], &data[1]); + matrix_map_point(td->matrix, + data[2], data[3], &data[2], &data[3]); + break; + case VG_CUBIC_TO: + matrix_map_point(td->matrix, + data[0], data[1], &data[0], &data[1]); + matrix_map_point(td->matrix, + data[2], data[3], &data[2], &data[3]); + matrix_map_point(td->matrix, + data[4], data[5], &data[4], &data[5]); + break; + case VG_SQUAD_TO: + matrix_map_point(td->matrix, + data[0], data[1], &data[0], &data[1]); + break; + case VG_SCUBIC_TO: + matrix_map_point(td->matrix, + data[0], data[1], &data[0], &data[1]); + matrix_map_point(td->matrix, + data[2], data[3], &data[2], &data[3]); + break; + case VG_SCCWARC_TO: + case VG_SCWARC_TO: + case VG_LCCWARC_TO: + case VG_LCWARC_TO: { + struct arc arc; + struct path *path = path_create(td->datatype, + 1, 0, 0, 0, VG_PATH_CAPABILITY_ALL); + arc_init(&arc, segment, + pd->ox, pd->oy, data[3], data[4], + data[0], data[1], data[2]); + + arc_to_path(&arc, path, td->matrix); + + num_coords = path_num_coords(path); + + array_append_data(td->segments, path->segments->data, + path->num_segments); + array_append_data(td->coords, path->control_points->data, + num_coords); + path_destroy(path); + + return VG_TRUE; + } + break; + default: + break; + } + + vg_float_to_datatype(td->datatype, common_data, data, num_coords); + + array_append_data(td->segments, &pd->segment, 1); + array_append_data(td->coords, common_data, num_coords); + return VG_TRUE; +} + +void path_transform(struct path *dst, struct path *src) +{ + struct transform_data data; + struct vg_context *ctx = dst->base.ctx; + + data.segments = dst->segments; + data.coords = dst->control_points; + data.matrix = &ctx->state.vg.path_user_to_surface_matrix; + data.datatype = dst->datatype; + + path_for_each_segment(src, transform_cb, (void*)&data); + + dst->num_segments = dst->segments->num_elements; + dst->dirty = VG_TRUE; + dst->dirty_stroke = VG_TRUE; +} + +void path_append_path(struct path *dst, + struct path *src) +{ + VGint num_coords = path_num_coords(src); + void *dst_data = malloc(size_for_datatype(dst->datatype) * num_coords); + array_append_data(dst->segments, + src->segments->data, + src->num_segments); + convert_path(src, dst->datatype, + dst_data, num_coords); + array_append_data(dst->control_points, + dst_data, + num_coords); + free(dst_data); + + dst->num_segments += src->num_segments; + dst->dirty = VG_TRUE; + dst->dirty_stroke = VG_TRUE; +} + +static INLINE VGboolean is_segment_arc(VGubyte segment) +{ + VGubyte scommand = SEGMENT_COMMAND(segment); + return (scommand == VG_SCCWARC_TO || + scommand == VG_SCWARC_TO || + scommand == VG_LCCWARC_TO || + scommand == VG_LCWARC_TO); +} + +struct path_iter_data { + struct path *path; + VGubyte segment; + void *coords; + VGfloat px, py, ox, oy, sx, sy; +}; +static INLINE VGubyte normalize_coords(struct path_iter_data *pd, + VGint *num_coords, + VGfloat *data) +{ + VGint command = SEGMENT_COMMAND(pd->segment); + VGboolean relative = SEGMENT_ABS_REL(pd->segment); + + switch(command) { + case VG_CLOSE_PATH: + *num_coords = 0; + pd->ox = pd->sx; + pd->oy = pd->sy; + return VG_CLOSE_PATH; + break; + case VG_MOVE_TO: + data_at(&pd->coords, pd->path, 0, 2, data); + map_if_relative(pd->ox, pd->oy, relative, &data[0], &data[1]); + pd->sx = data[0]; + pd->sy = data[1]; + pd->ox = data[0]; + pd->oy = data[1]; + pd->px = data[0]; + pd->py = data[1]; + *num_coords = 2; + return VG_MOVE_TO_ABS; + break; + case VG_LINE_TO: + data_at(&pd->coords, pd->path, 0, 2, data); + map_if_relative(pd->ox, pd->oy, relative, &data[0], &data[1]); + pd->ox = data[0]; + pd->oy = data[1]; + pd->px = data[0]; + pd->py = data[1]; + *num_coords = 2; + return VG_LINE_TO_ABS; + break; + case VG_HLINE_TO: + data_at(&pd->coords, pd->path, 0, 1, data); + map_if_relative(pd->ox, pd->oy, relative, &data[0], 0); + data[1] = pd->oy; + pd->ox = data[0]; + pd->oy = data[1]; + pd->px = data[0]; + pd->py = data[1]; + *num_coords = 2; + return VG_LINE_TO_ABS; + break; + case VG_VLINE_TO: + data_at(&pd->coords, pd->path, 0, 1, data); + map_if_relative(pd->ox, pd->oy, relative, 0, &data[0]); + data[1] = data[0]; + data[0] = pd->ox; + pd->ox = data[0]; + pd->oy = data[1]; + pd->px = data[0]; + pd->py = data[1]; + *num_coords = 2; + return VG_LINE_TO_ABS; + break; + case VG_CUBIC_TO: { + data_at(&pd->coords, pd->path, 0, 6, data); + map_if_relative(pd->ox, pd->oy, relative, &data[0], &data[1]); + map_if_relative(pd->ox, pd->oy, relative, &data[2], &data[3]); + map_if_relative(pd->ox, pd->oy, relative, &data[4], &data[5]); + pd->px = data[2]; + pd->py = data[3]; + pd->ox = data[4]; + pd->oy = data[5]; + *num_coords = 6; + return VG_CUBIC_TO_ABS; + } + break; + case VG_QUAD_TO: { + VGfloat x0, y0, x1, y1, x2, y2, x3, y3; + data_at(&pd->coords, pd->path, 0, 4, data); + x0 = pd->ox; + y0 = pd->oy; + x1 = data[0]; + y1 = data[1]; + x3 = data[2]; + y3 = data[3]; + map_if_relative(pd->ox, pd->oy, relative, &x1, &y1); + map_if_relative(pd->ox, pd->oy, relative, &x3, &y3); + pd->px = x1; + pd->py = y1; + { /* form a cubic out of it */ + x2 = (x3 + 2*x1) / 3.f; + y2 = (y3 + 2*y1) / 3.f; + x1 = (x0 + 2*x1) / 3.f; + y1 = (y0 + 2*y1) / 3.f; + } + pd->ox = x3; + pd->oy = y3; + data[0] = x1; + data[1] = y1; + data[2] = x2; + data[3] = y2; + data[4] = x3; + data[5] = y3; + *num_coords = 6; + return VG_CUBIC_TO_ABS; + } + break; + case VG_SQUAD_TO: { + VGfloat x0, y0, x1, y1, x2, y2, x3, y3; + data_at(&pd->coords, pd->path, 0, 2, data); + x0 = pd->ox; + y0 = pd->oy; + x1 = 2 * pd->ox - pd->px; + y1 = 2 * pd->oy - pd->py; + x3 = data[0]; + y3 = data[1]; + map_if_relative(pd->ox, pd->oy, relative, &x3, &y3); + pd->px = x1; + pd->py = y1; + { /* form a cubic out of it */ + x2 = (x3 + 2*x1) / 3.f; + y2 = (y3 + 2*y1) / 3.f; + x1 = (x0 + 2*x1) / 3.f; + y1 = (y0 + 2*y1) / 3.f; + } + pd->ox = x3; + pd->oy = y3; + data[0] = x1; + data[1] = y1; + data[2] = x2; + data[3] = y2; + data[4] = x3; + data[5] = y3; + *num_coords = 6; + return VG_CUBIC_TO_ABS; + } + break; + case VG_SCUBIC_TO: { + VGfloat x0, y0, x1, y1, x2, y2, x3, y3; + data_at(&pd->coords, pd->path, 0, 4, data); + x0 = pd->ox; + y0 = pd->oy; + x1 = 2*pd->ox-pd->px; + y1 = 2*pd->oy-pd->py; + x2 = data[0]; + y2 = data[1]; + x3 = data[2]; + y3 = data[3]; + map_if_relative(pd->ox, pd->oy, relative, &x2, &y2); + map_if_relative(pd->ox, pd->oy, relative, &x3, &y3); + pd->ox = x3; + pd->oy = y3; + pd->px = x2; + pd->py = y2; + data[0] = x1; + data[1] = y1; + data[2] = x2; + data[3] = y2; + data[4] = x3; + data[5] = y3; + *num_coords = 6; + return VG_CUBIC_TO_ABS; + } + break; + case VG_SCCWARC_TO: + case VG_SCWARC_TO: + case VG_LCCWARC_TO: + case VG_LCWARC_TO: { + data_at(&pd->coords, pd->path, 0, 5, data); + map_if_relative(pd->ox, pd->oy, relative, &data[3], &data[4]); + pd->ox = data[3]; + pd->oy = data[4]; + pd->px = data[3]; + pd->py = data[4]; + *num_coords = 5; + return command | VG_ABSOLUTE; + } + break; + default: + abort(); + assert(!"Unknown segment!"); + } +} + +static void linearly_interpolate(VGfloat *result, + const VGfloat *start, + const VGfloat *end, + VGfloat amount, + VGint number) +{ + VGint i; + for (i = 0; i < number; ++i) { + result[i] = start[i] + (end[i] - start[i]) * amount; + } +} + +VGboolean path_interpolate(struct path *dst, + struct path *start, struct path *end, + VGfloat amount) +{ + /* temporary path that we can discard if it will turn + * out that start is not compatible with end */ + struct path *res_path = path_create(dst->datatype, + 1.0, 0.0, + 0, 0, dst->caps); + VGint i; + VGfloat start_coords[8]; + VGfloat end_coords[8]; + VGfloat results[8]; + VGubyte common_data[sizeof(VGfloat)*8]; + struct path_iter_data start_iter, end_iter; + + memset(&start_iter, 0, sizeof(struct path_iter_data)); + memset(&end_iter, 0, sizeof(struct path_iter_data)); + + start_iter.path = start; + start_iter.coords = start->control_points->data; + end_iter.path = end; + end_iter.coords = end->control_points->data; + + for (i = 0; i < start->num_segments; ++i) { + VGubyte segment; + VGubyte ssegment, esegment; + VGint snum_coords, enum_coords; + start_iter.segment = ((VGubyte*)(start->segments->data))[i]; + end_iter.segment = ((VGubyte*)(end->segments->data))[i]; + + ssegment = normalize_coords(&start_iter, &snum_coords, + start_coords); + esegment = normalize_coords(&end_iter, &enum_coords, + end_coords); + + if (is_segment_arc(ssegment)) { + if (!is_segment_arc(esegment)) { + path_destroy(res_path); + return VG_FALSE; + } + if (amount > 0.5) + segment = esegment; + else + segment = ssegment; + } else if (is_segment_arc(esegment)) { + path_destroy(res_path); + return VG_FALSE; + } + else if (ssegment != esegment) { + path_destroy(res_path); + return VG_FALSE; + } + else + segment = ssegment; + + linearly_interpolate(results, start_coords, end_coords, + amount, snum_coords); + vg_float_to_datatype(dst->datatype, common_data, results, snum_coords); + path_append_data(res_path, 1, &segment, common_data); + } + + path_append_path(dst, res_path); + path_destroy(res_path); + + dst->dirty = VG_TRUE; + dst->dirty_stroke = VG_TRUE; + + return VG_TRUE; +} + +void path_clear(struct path *p, VGbitfield capabilities) +{ + path_set_capabilities(p, capabilities); + array_destroy(p->segments); + array_destroy(p->control_points); + p->segments = array_create(size_for_datatype(VG_PATH_DATATYPE_S_8)); + p->control_points = array_create(size_for_datatype(p->datatype)); + p->num_segments = 0; + p->dirty = VG_TRUE; + p->dirty_stroke = VG_TRUE; +} + +struct path * path_create_stroke(struct path *p, + struct matrix *matrix) +{ + VGint i; + VGfloat sx, sy, px, py, ox, oy; + VGfloat x0, y0, x1, y1, x2, y2, x3, y3; + VGfloat data[8]; + void *coords = (VGfloat *)p->control_points->data; + int dashed = (p->base.ctx->state.vg.stroke.dash_pattern_num ? 1 : 0); + struct dash_stroker stroker; + struct vg_state *vg_state = &p->base.ctx->state.vg; + + if (p->stroked.path) + { + /* ### compare the dash patterns to see if we can cache them. + * for now we simply always bail out if the path is dashed. + */ + if (memcmp( &p->stroked.matrix, + matrix, + sizeof *matrix ) == 0 && + !dashed && !p->dirty_stroke && + floatsEqual(p->stroked.stroke_width, vg_state->stroke.line_width.f) && + floatsEqual(p->stroked.miter_limit, vg_state->stroke.miter_limit.f) && + p->stroked.cap_style == vg_state->stroke.cap_style && + p->stroked.join_style == vg_state->stroke.join_style) + { + return p->stroked.path; + } + else { + path_destroy( p->stroked.path ); + p->stroked.path = NULL; + } + } + + + sx = sy = px = py = ox = oy = 0.f; + + if (dashed) + dash_stroker_init((struct stroker *)&stroker, vg_state); + else + stroker_init((struct stroker *)&stroker, vg_state); + + stroker_begin((struct stroker *)&stroker); + + for (i = 0; i < p->num_segments; ++i) { + VGubyte segment = ((VGubyte*)(p->segments->data))[i]; + VGint command = SEGMENT_COMMAND(segment); + VGboolean relative = SEGMENT_ABS_REL(segment); + + switch(command) { + case VG_CLOSE_PATH: { + VGfloat x0 = sx; + VGfloat y0 = sy; + matrix_map_point(matrix, x0, y0, &x0, &y0); + stroker_line_to((struct stroker *)&stroker, x0, y0); + } + break; + case VG_MOVE_TO: + data_at(&coords, p, 0, 2, data); + x0 = data[0]; + y0 = data[1]; + map_if_relative(ox, oy, relative, &x0, &y0); + sx = x0; + sy = y0; + ox = x0; + oy = y0; + px = x0; + py = y0; + matrix_map_point(matrix, x0, y0, &x0, &y0); + stroker_move_to((struct stroker *)&stroker, x0, y0); + break; + case VG_LINE_TO: + data_at(&coords, p, 0, 2, data); + x0 = data[0]; + y0 = data[1]; + map_if_relative(ox, oy, relative, &x0, &y0); + ox = x0; + oy = y0; + px = x0; + py = y0; + matrix_map_point(matrix, x0, y0, &x0, &y0); + stroker_line_to((struct stroker *)&stroker, x0, y0); + break; + case VG_HLINE_TO: + data_at(&coords, p, 0, 1, data); + x0 = data[0]; + y0 = oy; + map_if_relative(ox, oy, relative, &x0, 0); + ox = x0; + px = x0; + py = y0; + matrix_map_point(matrix, x0, y0, &x0, &y0); + stroker_line_to((struct stroker *)&stroker, x0, y0); + break; + case VG_VLINE_TO: + data_at(&coords, p, 0, 1, data); + x0 = ox; + y0 = data[0]; + map_if_relative(ox, oy, relative, 0, &y0); + oy = y0; + px = x0; + py = y0; + matrix_map_point(matrix, x0, y0, &x0, &y0); + stroker_line_to((struct stroker *)&stroker, x0, y0); + break; + case VG_CUBIC_TO: { + data_at(&coords, p, 0, 6, data); + x0 = ox; + y0 = oy; + x1 = data[0]; + y1 = data[1]; + x2 = data[2]; + y2 = data[3]; + x3 = data[4]; + y3 = data[5]; + map_if_relative(ox, oy, relative, &x1, &y1); + map_if_relative(ox, oy, relative, &x2, &y2); + map_if_relative(ox, oy, relative, &x3, &y3); + if (floatsEqual(x1, ox) && floatsEqual(y1, oy) && + floatsEqual(x1, x2) && floatsEqual(y1, y2) && + floatsEqual(x2, x3) && floatsEqual(y2, y3)) { + /*ignore the empty segment */ + continue; + } else if (floatsEqual(x3, ox) && floatsEqual(y3, oy)) { + /* if dup vertex, emit a line */ + ox = x3; + oy = y3; + matrix_map_point(matrix, x3, y3, &x3, &y3); + stroker_line_to((struct stroker *)&stroker, x3, y3); + continue; + } + ox = x3; + oy = y3; + px = x2; + py = y2; + assert(matrix_is_affine(matrix)); + matrix_map_point(matrix, x0, y0, &x0, &y0); + matrix_map_point(matrix, x1, y1, &x1, &y1); + matrix_map_point(matrix, x2, y2, &x2, &y2); + matrix_map_point(matrix, x3, y3, &x3, &y3); + stroker_curve_to((struct stroker *)&stroker, x1, y1, x2, y2, x3, y3); + } + break; + case VG_QUAD_TO: { + data_at(&coords, p, 0, 4, data); + x0 = ox; + y0 = oy; + x1 = data[0]; + y1 = data[1]; + x3 = data[2]; + y3 = data[3]; + map_if_relative(ox, oy, relative, &x1, &y1); + map_if_relative(ox, oy, relative, &x3, &y3); + px = x1; + py = y1; + { /* form a cubic out of it */ + x2 = (x3 + 2*x1) / 3.f; + y2 = (y3 + 2*y1) / 3.f; + x1 = (x0 + 2*x1) / 3.f; + y1 = (y0 + 2*y1) / 3.f; + } + if (floatsEqual(x1, ox) && floatsEqual(y1, oy) && + floatsEqual(x1, x2) && floatsEqual(y1, y2) && + floatsEqual(x2, x3) && floatsEqual(y2, y3)) { + /*ignore the empty segment */ + continue; + } else if (floatsEqual(x3, ox) && floatsEqual(y3, oy)) { + /* if dup vertex, emit a line */ + ox = x3; + oy = y3; + matrix_map_point(matrix, x3, y3, &x3, &y3); + stroker_line_to((struct stroker *)&stroker, x3, y3); + continue; + } + ox = x3; + oy = y3; + assert(matrix_is_affine(matrix)); + matrix_map_point(matrix, x0, y0, &x0, &y0); + matrix_map_point(matrix, x1, y1, &x1, &y1); + matrix_map_point(matrix, x2, y2, &x2, &y2); + matrix_map_point(matrix, x3, y3, &x3, &y3); + stroker_curve_to((struct stroker *)&stroker, x1, y1, x2, y2, x3, y3); + } + break; + case VG_SQUAD_TO: { + data_at(&coords, p, 0, 2, data); + x0 = ox; + y0 = oy; + x1 = 2*ox-px; + y1 = 2*oy-py; + x3 = data[0]; + y3 = data[1]; + map_if_relative(ox, oy, relative, &x3, &y3); + px = x1; + py = y1; + { /* form a cubic out of it */ + x2 = (x3 + 2*x1) / 3.f; + y2 = (y3 + 2*y1) / 3.f; + x1 = (x0 + 2*x1) / 3.f; + y1 = (y0 + 2*y1) / 3.f; + } + if (floatsEqual(x1, ox) && floatsEqual(y1, oy) && + floatsEqual(x1, x2) && floatsEqual(y1, y2) && + floatsEqual(x2, x3) && floatsEqual(y2, y3)) { + /*ignore the empty segment */ + continue; + } else if (floatsEqual(x3, ox) && floatsEqual(y3, oy)) { + /* if dup vertex, emit a line */ + ox = x3; + oy = y3; + matrix_map_point(matrix, x3, y3, &x3, &y3); + stroker_line_to((struct stroker *)&stroker, x3, y3); + continue; + } + ox = x3; + oy = y3; + assert(matrix_is_affine(matrix)); + matrix_map_point(matrix, x0, y0, &x0, &y0); + matrix_map_point(matrix, x1, y1, &x1, &y1); + matrix_map_point(matrix, x2, y2, &x2, &y2); + matrix_map_point(matrix, x3, y3, &x3, &y3); + stroker_curve_to((struct stroker *)&stroker, x1, y1, x2, y2, x3, y3); + } + break; + case VG_SCUBIC_TO: { + data_at(&coords, p, 0, 4, data); + x0 = ox; + y0 = oy; + x1 = 2*ox-px; + y1 = 2*oy-py; + x2 = data[0]; + y2 = data[1]; + x3 = data[2]; + y3 = data[3]; + map_if_relative(ox, oy, relative, &x2, &y2); + map_if_relative(ox, oy, relative, &x3, &y3); + if (floatsEqual(x1, ox) && floatsEqual(y1, oy) && + floatsEqual(x1, x2) && floatsEqual(y1, y2) && + floatsEqual(x2, x3) && floatsEqual(y2, y3)) { + /*ignore the empty segment */ + continue; + } else if (floatsEqual(x3, ox) && floatsEqual(y3, oy)) { + /* if dup vertex, emit a line */ + ox = x3; + oy = y3; + matrix_map_point(matrix, x3, y3, &x3, &y3); + stroker_line_to((struct stroker *)&stroker, x3, y3); + continue; + } + ox = x3; + oy = y3; + px = x2; + py = y2; + assert(matrix_is_affine(matrix)); + matrix_map_point(matrix, x0, y0, &x0, &y0); + matrix_map_point(matrix, x1, y1, &x1, &y1); + matrix_map_point(matrix, x2, y2, &x2, &y2); + matrix_map_point(matrix, x3, y3, &x3, &y3); + stroker_curve_to((struct stroker *)&stroker, x1, y1, x2, y2, x3, y3); + } + break; + case VG_SCCWARC_TO: + case VG_SCWARC_TO: + case VG_LCCWARC_TO: + case VG_LCWARC_TO: { + VGfloat rh, rv, rot; + struct arc arc; + + data_at(&coords, p, 0, 5, data); + x0 = ox; + y0 = oy; + rh = data[0]; + rv = data[1]; + rot = data[2]; + x1 = data[3]; + y1 = data[4]; + map_if_relative(ox, oy, relative, &x1, &y1); + if (floatsEqual(x1, ox) && floatsEqual(y1, oy)) { + /* if dup vertex, emit a line */ + ox = x1; + oy = y1; + matrix_map_point(matrix, x1, y1, &x1, &y1); + stroker_line_to((struct stroker *)&stroker, x1, y1); + continue; + } + arc_init(&arc, command, x0, y0, x1, y1, + rh, rv, rot); + arc_stroke_cb(&arc, (struct stroker *)&stroker, + matrix); + ox = x1; + oy = y1; + px = x1; + py = y1; + } + break; + default: + abort(); + assert(!"Unknown segment!"); + } + } + + stroker_end((struct stroker *)&stroker); + + if (dashed) + dash_stroker_cleanup((struct dash_stroker *)&stroker); + else + stroker_cleanup((struct stroker *)&stroker); + + p->stroked.path = stroker.base.path; + p->stroked.matrix = *matrix; + p->dirty_stroke = VG_FALSE; + p->stroked.stroke_width = vg_state->stroke.line_width.f; + p->stroked.miter_limit = vg_state->stroke.miter_limit.f; + p->stroked.cap_style = vg_state->stroke.cap_style; + p->stroked.join_style = vg_state->stroke.join_style; + + return stroker.base.path; +} + +void path_render(struct path *p, VGbitfield paintModes) +{ + struct vg_context *ctx = vg_current_context(); + struct matrix *mat = &ctx->state.vg.path_user_to_surface_matrix; + + vg_validate_state(ctx); + + shader_set_drawing_image(ctx->shader, VG_FALSE); + shader_set_image(ctx->shader, 0); +#if 0 + fprintf(stderr, "Matrix(11=%f 12=%f 13=%f 21=%f 22=%f 23=%f 31=%f 32=%f 33=%f)\n", + mat->m[0], mat->m[1], mat->m[2], + mat->m[3], mat->m[4], mat->m[5], + mat->m[6], mat->m[7], mat->m[8]); +#endif + if (paintModes & VG_FILL_PATH) { + /* First the fill */ + shader_set_paint(ctx->shader, ctx->state.vg.fill_paint); + shader_bind(ctx->shader); + path_fill(p, mat); + } + + if (paintModes & VG_STROKE_PATH){ + /* 8.7.5: "line width less than or equal to 0 prevents stroking from + * taking place."*/ + if (ctx->state.vg.stroke.line_width.f <= 0) + return; + shader_set_paint(ctx->shader, ctx->state.vg.stroke_paint); + shader_bind(ctx->shader); + path_stroke(p); + } +} + +void path_fill(struct path *p, struct matrix *mat) +{ + struct vg_context *ctx = vg_current_context(); + { + struct polygon_array *polygon_array = path_get_fill_polygons(p, mat); + struct array *polys = polygon_array->array; + + if (!polygon_array || !polys || !polys->num_elements) { + return; + } + polygon_array_fill(polygon_array, ctx); + } +} + +void path_stroke(struct path *p) +{ + struct vg_context *ctx = vg_current_context(); + struct matrix *mat = &ctx->state.vg.path_user_to_surface_matrix; + VGFillRule old_fill = ctx->state.vg.fill_rule; + struct matrix identity; + struct path *stroke; + + matrix_load_identity(&identity); + stroke = path_create_stroke(p, &identity); + if (stroke && !path_is_empty(stroke)) { + ctx->state.vg.fill_rule = VG_NON_ZERO; + + path_fill(stroke, mat); + + ctx->state.vg.fill_rule = old_fill; + } +} + +void path_move_to(struct path *p, float x, float y) +{ + VGubyte segment = VG_MOVE_TO_ABS; + VGubyte common_data[sizeof(VGfloat) * 2]; + VGfloat data[2] = {x, y}; + + vg_float_to_datatype(p->datatype, common_data, data, 2); + path_append_data(p, 1, &segment, common_data); +} + +void path_line_to(struct path *p, float x, float y) +{ + VGubyte segment = VG_LINE_TO_ABS; + VGubyte common_data[sizeof(VGfloat) * 2]; + VGfloat data[2] = {x, y}; + + vg_float_to_datatype(p->datatype, common_data, data, 2); + + path_append_data(p, 1, &segment, common_data); +} + +void path_cubic_to(struct path *p, float px1, float py1, + float px2, float py2, + float x, float y) +{ + VGubyte segment = VG_CUBIC_TO_ABS; + VGubyte common_data[sizeof(VGfloat) * 6]; + VGfloat data[6]; + + data[0] = px1; data[1] = py1; + data[2] = px2; data[3] = py2; + data[4] = x; data[5] = y; + + vg_float_to_datatype(p->datatype, common_data, data, 6); + + path_append_data(p, 1, &segment, common_data); +} + +static INLINE void line_bounds(VGfloat *line /*x1,y1,x2,y2*/, + VGfloat *bounds) +{ + bounds[0] = MIN2(line[0], line[2]); + bounds[1] = MIN2(line[1], line[3]); + bounds[2] = MAX2(line[0], line[2]) - bounds[0]; + bounds[3] = MAX2(line[1], line[3]) - bounds[1]; +} + +static INLINE void unite_bounds(VGfloat *bounds, + VGfloat *el) +{ + VGfloat cx1, cy1, cx2, cy2; + VGfloat nx1, ny1, nx2, ny2; + + cx1 = bounds[0]; + cy1 = bounds[1]; + cx2 = bounds[0] + bounds[2]; + cy2 = bounds[1] + bounds[3]; + + nx1 = el[0]; + ny1 = el[1]; + nx2 = el[0] + el[2]; + ny2 = el[1] + el[3]; + + bounds[0] = MIN2(cx1, nx1); + bounds[1] = MIN2(cy1, ny1); + bounds[2] = MAX2(cx2, nx2) - bounds[0]; + bounds[3] = MAX2(cy2, ny2) - bounds[1]; +} + +static INLINE void set_bounds(VGfloat *bounds, + VGfloat *element_bounds, + VGboolean *initialized) +{ + if (!(*initialized)) { + memcpy(bounds, element_bounds, 4 * sizeof(VGfloat)); + *initialized = VG_TRUE; + } else + unite_bounds(bounds, element_bounds); +} + +void path_bounding_rect(struct path *p, float *x, float *y, + float *w, float *h) +{ + VGint i; + VGfloat coords[8]; + struct path_iter_data iter; + VGint num_coords; + VGfloat bounds[4]; + VGfloat element_bounds[4]; + VGfloat ox, oy; + VGboolean bounds_inited = VG_FALSE; + + memset(&iter, 0, sizeof(struct path_iter_data)); + memset(&bounds, 0, sizeof(bounds)); + + if (!p->num_segments) { + bounds[2] = -1; + bounds[3] = -1; + } + + + iter.path = p; + iter.coords = p->control_points->data; + + for (i = 0; i < p->num_segments; ++i) { + VGubyte segment; + iter.segment = ((VGubyte*)(p->segments->data))[i]; + + ox = iter.ox; + oy = iter.oy; + + segment = normalize_coords(&iter, &num_coords, coords); + + switch(segment) { + case VG_CLOSE_PATH: + case VG_MOVE_TO_ABS: + break; + case VG_LINE_TO_ABS: { + VGfloat line[4] = {ox, oy, coords[0], coords[1]}; + line_bounds(line, element_bounds); + set_bounds(bounds, element_bounds, &bounds_inited); + } + break; + case VG_CUBIC_TO_ABS: { + struct bezier bezier; + bezier_init(&bezier, ox, oy, + coords[0], coords[1], + coords[2], coords[3], + coords[4], coords[5]); + bezier_exact_bounds(&bezier, element_bounds); + set_bounds(bounds, element_bounds, &bounds_inited); + } + break; + case VG_SCCWARC_TO: + case VG_SCWARC_TO: + case VG_LCCWARC_TO: + case VG_LCWARC_TO: { + struct arc arc; + struct matrix identity; + struct path *path = path_create(VG_PATH_DATATYPE_F, + 1, 0, 0, 0, VG_PATH_CAPABILITY_ALL); + + matrix_load_identity(&identity); + arc_init(&arc, segment, + ox, oy, coords[3], coords[4], + coords[0], coords[1], coords[2]); + + arc_to_path(&arc, path, &identity); + + path_bounding_rect(path, element_bounds + 0, element_bounds + 1, + element_bounds + 2, element_bounds + 3); + set_bounds(bounds, element_bounds, &bounds_inited); + } + break; + default: + assert(0); + } + } + + *x = bounds[0]; + *y = bounds[1]; + *w = bounds[2]; + *h = bounds[3]; +} + +float path_length(struct path *p, int start_segment, int num_segments) +{ + VGint i; + VGfloat coords[8]; + struct path_iter_data iter; + VGint num_coords; + VGfloat length = 0; + VGfloat ox, oy; + VGboolean in_range = VG_FALSE; + + memset(&iter, 0, sizeof(struct path_iter_data)); + + iter.path = p; + iter.coords = p->control_points->data; + + for (i = 0; i < (start_segment + num_segments); ++i) { + VGubyte segment; + + iter.segment = ((VGubyte*)(p->segments->data))[i]; + + ox = iter.ox; + oy = iter.oy; + + segment = normalize_coords(&iter, &num_coords, coords); + + in_range = (i >= start_segment) && i <= (start_segment + num_segments); + if (!in_range) + continue; + + switch(segment) { + case VG_MOVE_TO_ABS: + break; + case VG_CLOSE_PATH: { + VGfloat line[4] = {ox, oy, iter.sx, iter.sy}; + length += line_lengthv(line); + } + break; + case VG_LINE_TO_ABS: { + VGfloat line[4] = {ox, oy, coords[0], coords[1]}; + length += line_lengthv(line); + } + break; + case VG_CUBIC_TO_ABS: { + struct bezier bezier; + bezier_init(&bezier, ox, oy, + coords[0], coords[1], + coords[2], coords[3], + coords[4], coords[5]); + length += bezier_length(&bezier, BEZIER_DEFAULT_ERROR); + } + break; + case VG_SCCWARC_TO: + case VG_SCWARC_TO: + case VG_LCCWARC_TO: + case VG_LCWARC_TO: { + struct arc arc; + struct matrix identity; + struct path *path = path_create(VG_PATH_DATATYPE_F, + 1, 0, 0, 0, VG_PATH_CAPABILITY_ALL); + + matrix_load_identity(&identity); + arc_init(&arc, segment, + ox, oy, coords[3], coords[4], + coords[0], coords[1], coords[2]); + + arc_to_path(&arc, path, &identity); + + length += path_length(path, 0, path_num_segments(path)); + } + break; + default: + assert(0); + } + } + + return length; +} + +static INLINE VGboolean point_on_current_segment(VGfloat distance, + VGfloat length, + VGfloat segment_length) +{ + return + (((floatIsZero(distance) || distance < 0) && floatIsZero(length)) || + ((distance > length || floatsEqual(distance, length)) && + (floatsEqual(distance, length + segment_length) || + distance < (length + segment_length)))); +} + +static VGboolean path_point_segment(struct path_iter_data iter, + struct path_iter_data prev_iter, + VGfloat coords[8], + VGfloat distance, + VGfloat length, VGfloat *current_length, + VGfloat *point, VGfloat *normal) +{ + switch (iter.segment) { + case VG_MOVE_TO_ABS: + break; + case VG_CLOSE_PATH: { + VGfloat line[4] = {prev_iter.ox, prev_iter.oy, iter.sx, iter.sy}; + VGboolean on_current_segment = VG_FALSE; + *current_length = line_lengthv(line); + on_current_segment = point_on_current_segment(distance, + length, + *current_length); + if (on_current_segment) { + VGfloat at = (distance - length) / line_lengthv(line); + line_normal_vector(line, normal); + line_point_at(line, at, point); + return VG_TRUE; + } + } + break; + case VG_LINE_TO_ABS: { + VGfloat line[4] = {prev_iter.ox, prev_iter.oy, coords[0], coords[1]}; + VGboolean on_current_segment = VG_FALSE; + *current_length = line_lengthv(line); + on_current_segment = point_on_current_segment(distance, + length, + *current_length); + if (on_current_segment) { + VGfloat at = (distance - length) / line_lengthv(line); + line_normal_vector(line, normal); + line_point_at(line, at, point); + return VG_TRUE; + } + } + break; + case VG_CUBIC_TO_ABS: { + struct bezier bezier; + bezier_init(&bezier, prev_iter.ox, prev_iter.oy, + coords[0], coords[1], + coords[2], coords[3], + coords[4], coords[5]); + *current_length = bezier_length(&bezier, BEZIER_DEFAULT_ERROR); + if (point_on_current_segment(distance, length, *current_length)) { + bezier_point_at_length(&bezier, distance - length, + point, normal); + return VG_TRUE; + } + } + break; + case VG_SCCWARC_TO: + case VG_SCWARC_TO: + case VG_LCCWARC_TO: + case VG_LCWARC_TO: { + struct arc arc; + struct matrix identity; + struct path *path = path_create(VG_PATH_DATATYPE_F, + 1, 0, 0, 0, VG_PATH_CAPABILITY_ALL); + + matrix_load_identity(&identity); + arc_init(&arc, iter.segment, + prev_iter.ox, prev_iter.oy, coords[3], coords[4], + coords[0], coords[1], coords[2]); + + arc_to_path(&arc, path, &identity); + + *current_length = path_length(path, 0, path_num_segments(path)); + if (point_on_current_segment(distance, length, *current_length)) { + path_point(path, 0, path_num_segments(path), + distance - length, point, normal); + return VG_TRUE; + } + } + break; + default: + assert(0); + } + return VG_FALSE; +} + +void path_point(struct path *p, VGint start_segment, VGint num_segments, + VGfloat distance, VGfloat *point, VGfloat *normal) +{ + VGint i; + VGfloat coords[8]; + struct path_iter_data iter, prev_iter; + VGint num_coords; + VGfloat length = 0; + VGfloat current_length = 0; + + memset(&iter, 0, sizeof(struct path_iter_data)); + memset(&prev_iter, 0, sizeof(struct path_iter_data)); + + point[0] = 0; + point[1] = 0; + + normal[0] = 0; + normal[1] = -1; + + iter.path = p; + iter.coords = p->control_points->data; + if (distance < 0) + distance = 0; + + for (i = 0; i < (start_segment + num_segments); ++i) { + VGboolean outside_range = (i < start_segment || + i >= (start_segment + num_segments)); + + prev_iter = iter; + + iter.segment = ((VGubyte*)(p->segments->data))[i]; + iter.segment = normalize_coords(&iter, &num_coords, coords); + + if (outside_range) + continue; + + if (path_point_segment(iter, prev_iter, coords, + distance, length, ¤t_length, + point, normal)) + return; + + length += current_length; + } + + /* + *OpenVG 1.0 - 8.6.11 vgPointAlongPath + * + * If distance is greater than or equal to the path length + *(i.e., the value returned by vgPathLength when called with the same + *startSegment and numSegments parameters), the visual ending point of + *the path is used. + */ + { + switch (iter.segment) { + case VG_MOVE_TO_ABS: + break; + case VG_CLOSE_PATH: { + VGfloat line[4] = {prev_iter.ox, prev_iter.oy, iter.sx, iter.sy}; + line_normal_vector(line, normal); + line_point_at(line, 1.f, point); + } + break; + case VG_LINE_TO_ABS: { + VGfloat line[4] = {prev_iter.ox, prev_iter.oy, coords[0], coords[1]}; + line_normal_vector(line, normal); + line_point_at(line, 1.f, point); + } + break; + case VG_CUBIC_TO_ABS: { + struct bezier bezier; + bezier_init(&bezier, prev_iter.ox, prev_iter.oy, + coords[0], coords[1], + coords[2], coords[3], + coords[4], coords[5]); + bezier_point_at_t(&bezier, 1.f, point, normal); + } + break; + case VG_SCCWARC_TO: + case VG_SCWARC_TO: + case VG_LCCWARC_TO: + case VG_LCWARC_TO: { + struct arc arc; + struct matrix identity; + struct path *path = path_create(VG_PATH_DATATYPE_F, + 1, 0, 0, 0, VG_PATH_CAPABILITY_ALL); + + matrix_load_identity(&identity); + arc_init(&arc, iter.segment, + prev_iter.ox, prev_iter.oy, coords[3], coords[4], + coords[0], coords[1], coords[2]); + + arc_to_path(&arc, path, &identity); + + path_point(path, 0, path_num_segments(path), + /* to make sure we're bigger than len * 2 it */ + 2 * path_length(path, 0, path_num_segments(path)), + point, normal); + } + break; + default: + assert(0); + } + } +} + +VGboolean path_is_empty(struct path *p) +{ + return p->segments->num_elements == 0; +} diff --git a/src/gallium/state_trackers/vega/path.h b/src/gallium/state_trackers/vega/path.h new file mode 100644 index 00000000000..e34538b7368 --- /dev/null +++ b/src/gallium/state_trackers/vega/path.h @@ -0,0 +1,126 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef _PATH_H +#define _PATH_H + +#include "VG/openvg.h" + +struct path; +struct polygon; +struct matrix; + +enum fill_rule { + ODD_EVEN_FILL, + WINDING_FILL +}; + + +struct path_for_each_data { + VGubyte segment; + /* all coords are absolute, even if segment is relative */ + const VGfloat *coords; + VGfloat sx, sy, ox, oy, px, py; + void *user_data; +}; + +typedef VGboolean (*path_for_each_cb)(struct path *p, + struct path_for_each_data *data); + + +struct path *path_create(VGPathDatatype dt, VGfloat scale, VGfloat bias, + VGint segmentCapacityHint, + VGint coordCapacityHint, + VGbitfield capabilities); +void path_destroy(struct path *p); + +VGbitfield path_capabilities(struct path *p); +void path_set_capabilities(struct path *p, VGbitfield bf); + +void path_append_data(struct path *p, + VGint numSegments, + const VGubyte * pathSegments, + const void * pathData); + +void path_append_path(struct path *dst, + struct path *src); + +VGint path_num_segments(struct path *p); + +void path_bounding_rect(struct path *p, float *x, float *y, + float *w, float *h); +float path_length(struct path *p, int start_segment, int num_segments); + +void path_set_fill_rule(enum fill_rule fill); +enum fill_rule path_fill_rule(enum fill_rule fill); + +VGboolean path_is_empty(struct path *p); + +VGbyte path_datatype_size(struct path *p); + +VGPathDatatype path_datatype(struct path *p); +VGfloat path_scale(struct path *p); +VGfloat path_bias(struct path *p); +VGint path_num_coords(struct path *p); + +void path_modify_coords(struct path *p, + VGint startIndex, + VGint numSegments, + const void * pathData); + +struct path *path_create_stroke(struct path *p, + struct matrix *m); + +void path_for_each_segment(struct path *path, + path_for_each_cb cb, + void *user_data); + +void path_transform(struct path *dst, struct path *src); +VGboolean path_interpolate(struct path *dst, + struct path *start, struct path *end, + VGfloat amount); + +void path_clear(struct path *p, VGbitfield capabilities); +void path_render(struct path *p, VGbitfield paintModes); +void path_fill(struct path *p, struct matrix *mat); +void path_stroke(struct path *p); + +void path_move_to(struct path *p, float x, float y); +void path_line_to(struct path *p, float x, float y); +void path_cubic_to(struct path *p, float px1, float py1, + float px2, float py2, + float x, float y); + +void path_point(struct path *p, VGint startSegment, VGint numSegments, + VGfloat distance, VGfloat *point, VGfloat *normal); + + + +void vg_float_to_datatype(VGPathDatatype datatype, + VGubyte *common_data, + const VGfloat *data, + VGint num_coords); +#endif diff --git a/src/gallium/state_trackers/vega/path_utils.h b/src/gallium/state_trackers/vega/path_utils.h new file mode 100644 index 00000000000..c2b3221dc53 --- /dev/null +++ b/src/gallium/state_trackers/vega/path_utils.h @@ -0,0 +1,109 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef PATH_UTILS_H +#define PATH_UTILS_H + +#include "VG/openvg.h" + +#define SEGMENT_COMMAND(command) /* Extract segment type */ \ + ((command) & 0x1e) +#define SEGMENT_ABS_REL(command) /* Extract absolute/relative bit */ \ + ((command) & 0x1) + +static INLINE VGint size_for_datatype(VGPathDatatype datatype) +{ + switch(datatype) { + case VG_PATH_DATATYPE_S_8: + return 1; + case VG_PATH_DATATYPE_S_16: + return 2; + case VG_PATH_DATATYPE_S_32: + return 4; + case VG_PATH_DATATYPE_F: + return 4; + default: + assert(!"unknown datatype"); + } + return 0; +} + +static INLINE VGint num_elements_for_segments(const VGubyte *segments, + VGint num_segments) +{ + VGint i; + VGint count = 0; + + for (i = 0; i < num_segments; ++i) { + VGubyte segment = segments[i]; + VGint command = SEGMENT_COMMAND(segment); + switch(command) { + case VG_CLOSE_PATH: + break; + case VG_MOVE_TO: + count += 2; + break; + case VG_LINE_TO: + count += 2; + break; + case VG_HLINE_TO: + count += 1; + break; + case VG_VLINE_TO: + count += 1; + break; + case VG_QUAD_TO: + count += 4; + break; + case VG_CUBIC_TO: + count += 6; + break; + case VG_SQUAD_TO: + count += 2; + break; + case VG_SCUBIC_TO: + count += 4; + break; + case VG_SCCWARC_TO: + count += 5; + break; + case VG_SCWARC_TO: + count += 5; + break; + case VG_LCCWARC_TO: + count += 5; + break; + case VG_LCWARC_TO: + count += 5; + break; + default: + assert(!"Unknown segment!"); + } + } + return count; +} + +#endif diff --git a/src/gallium/state_trackers/vega/polygon.c b/src/gallium/state_trackers/vega/polygon.c new file mode 100644 index 00000000000..b6d282d803b --- /dev/null +++ b/src/gallium/state_trackers/vega/polygon.c @@ -0,0 +1,550 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "polygon.h" + +#include "matrix.h" /*for floatsEqual*/ +#include "vg_context.h" +#include "vg_state.h" +#include "paint.h" +#include "renderer.h" +#include "util_array.h" +#include "VG/openvg.h" + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_state.h" +#include "pipe/p_inlines.h" +#include "pipe/p_screen.h" + +#include "util/u_draw_quad.h" +#include "util/u_math.h" + +#include <string.h> +#include <stdlib.h> + +#define DEBUG_POLYGON 0 + +#define COMPONENTS 2 + +struct polygon +{ + VGfloat *data; + VGint size; + + VGint num_verts; + + VGboolean dirty; + struct pipe_buffer *vbuf; + struct pipe_screen *screen; +}; + +static float *ptr_to_vertex(float *data, int idx) +{ + return data + (idx * COMPONENTS); +} + +#if 0 +static void polygon_print(struct polygon *poly) +{ + int i; + float *vert; + debug_printf("Polygon %p, size = %d\n", poly, poly->num_verts); + for (i = 0; i < poly->num_verts; ++i) { + vert = ptr_to_vertex(poly->data, i); + debug_printf("%f, %f, ", vert[0], vert[1]); + } + debug_printf("\nend\n"); +} +#endif + + +struct polygon * polygon_create(int size) +{ + struct polygon *poly = (struct polygon*)malloc(sizeof(struct polygon)); + + poly->data = malloc(sizeof(float) * COMPONENTS * size); + poly->size = size; + poly->num_verts = 0; + poly->dirty = VG_TRUE; + poly->vbuf = NULL; + + return poly; +} + +struct polygon * polygon_create_from_data(float *data, int size) +{ + struct polygon *poly = polygon_create(size); + + memcpy(poly->data, data, sizeof(float) * COMPONENTS * size); + poly->num_verts = size; + poly->dirty = VG_TRUE; + poly->vbuf = NULL; + + return poly; +} + +void polygon_destroy(struct polygon *poly) +{ + if (poly->vbuf) + pipe_buffer_reference(&poly->vbuf, NULL); + + free(poly->data); + free(poly); +} + +void polygon_resize(struct polygon *poly, int new_size) +{ + float *data = (float*)malloc(sizeof(float) * COMPONENTS * new_size); + int size = MIN2(sizeof(float) * COMPONENTS * new_size, + sizeof(float) * COMPONENTS * poly->size); + memcpy(data, poly->data, size); + free(poly->data); + poly->data = data; + poly->size = new_size; + poly->dirty = VG_TRUE; +} + +int polygon_size(struct polygon *poly) +{ + return poly->size; +} + +int polygon_vertex_count(struct polygon *poly) +{ + return poly->num_verts; +} + +float * polygon_data(struct polygon *poly) +{ + return poly->data; +} + +void polygon_vertex_append(struct polygon *p, + float x, float y) +{ + float *vert; +#if DEBUG_POLYGON + debug_printf("Append vertex [%f, %f]\n", x, y); +#endif + if (p->num_verts >= p->size) { + polygon_resize(p, p->size * 2); + } + + vert = ptr_to_vertex(p->data, p->num_verts); + vert[0] = x; + vert[1] = y; + ++p->num_verts; + p->dirty = VG_TRUE; +} + +void polygon_set_vertex(struct polygon *p, int idx, + float x, float y) +{ + float *vert; + if (idx >= p->num_verts) { + /*fixme: error reporting*/ + abort(); + return; + } + + vert = ptr_to_vertex(p->data, idx); + vert[0] = x; + vert[1] = y; + p->dirty = VG_TRUE; +} + +void polygon_vertex(struct polygon *p, int idx, + float *vertex) +{ + float *vert; + if (idx >= p->num_verts) { + /*fixme: error reporting*/ + abort(); + return; + } + + vert = ptr_to_vertex(p->data, idx); + vertex[0] = vert[0]; + vertex[1] = vert[1]; +} + +void polygon_bounding_rect(struct polygon *p, + float *rect) +{ + int i; + float minx, miny, maxx, maxy; + float *vert = ptr_to_vertex(p->data, 0); + minx = vert[0]; + maxx = vert[0]; + miny = vert[1]; + maxy = vert[1]; + + for (i = 1; i < p->num_verts; ++i) { + vert = ptr_to_vertex(p->data, i); + minx = MIN2(vert[0], minx); + miny = MIN2(vert[1], miny); + + maxx = MAX2(vert[0], maxx); + maxy = MAX2(vert[1], maxy); + } + + rect[0] = minx; + rect[1] = miny; + rect[2] = maxx - minx; + rect[3] = maxy - miny; +} + +int polygon_contains_point(struct polygon *p, + float x, float y) +{ + return 0; +} + +void polygon_append_polygon(struct polygon *dst, + struct polygon *src) +{ + if (dst->num_verts + src->num_verts >= dst->size) { + polygon_resize(dst, dst->num_verts + src->num_verts * 1.5); + } + memcpy(ptr_to_vertex(dst->data, dst->num_verts), + src->data, src->num_verts * COMPONENTS * sizeof(VGfloat)); + dst->num_verts += src->num_verts; +} + +VGboolean polygon_is_closed(struct polygon *p) +{ + VGfloat start[2], end[2]; + + polygon_vertex(p, 0, start); + polygon_vertex(p, p->num_verts - 1, end); + + return floatsEqual(start[0], end[0]) && floatsEqual(start[1], end[1]); +} + +static void set_blend_for_fill(struct pipe_blend_state *blend) +{ + memset(blend, 0, sizeof(struct pipe_blend_state)); + blend->colormask = 0; /*disable colorwrites*/ + + blend->rgb_src_factor = PIPE_BLENDFACTOR_ONE; + blend->alpha_src_factor = PIPE_BLENDFACTOR_ONE; + blend->rgb_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA; + blend->alpha_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA; +} + +static void draw_polygon(struct vg_context *ctx, + struct polygon *poly) +{ + int vert_size; + struct pipe_context *pipe; + struct pipe_vertex_buffer vbuffer; + struct pipe_vertex_element velement; + + vert_size = poly->num_verts * COMPONENTS * sizeof(float); + + /*polygon_print(poly);*/ + + pipe = ctx->pipe; + + if (poly->vbuf == NULL || poly->dirty) { + if (poly->vbuf) { + pipe_buffer_reference(&poly->vbuf, + NULL); + } + poly->screen = pipe->screen; + poly->vbuf= pipe_user_buffer_create(poly->screen, + poly->data, + vert_size); + poly->dirty = VG_FALSE; + } + + + /* tell pipe about the vertex buffer */ + memset(&vbuffer, 0, sizeof(vbuffer)); + vbuffer.buffer = poly->vbuf; + vbuffer.stride = COMPONENTS * sizeof(float); /* vertex size */ + vbuffer.buffer_offset = 0; + vbuffer.max_index = poly->num_verts - 1; + pipe->set_vertex_buffers(pipe, 1, &vbuffer); + + /* tell pipe about the vertex attributes */ + velement.src_offset = 0; + velement.vertex_buffer_index = 0; + velement.src_format = PIPE_FORMAT_R32G32_FLOAT; + velement.nr_components = COMPONENTS; + pipe->set_vertex_elements(pipe, 1, &velement); + + /* draw */ + pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLE_FAN, + 0, poly->num_verts); +} + +void polygon_fill(struct polygon *poly, struct vg_context *ctx) +{ + struct pipe_depth_stencil_alpha_state dsa; + struct pipe_blend_state blend; + VGfloat bounds[4]; + VGfloat min_x, min_y, max_x, max_y; + assert(poly); + polygon_bounding_rect(poly, bounds); + min_x = bounds[0]; + min_y = bounds[1]; + max_x = bounds[0] + bounds[2]; + max_y = bounds[1] + bounds[3]; + +#if DEBUG_POLYGON + debug_printf("Poly bounds are [%f, %f], [%f, %f]\n", + min_x, min_y, max_x, max_y); +#endif + + set_blend_for_fill(&blend); + + memset(&dsa, 0, sizeof(struct pipe_depth_stencil_alpha_state)); + + cso_save_blend(ctx->cso_context); + cso_save_depth_stencil_alpha(ctx->cso_context); + + dsa.stencil[0].enabled = 1; + if (ctx->state.vg.fill_rule == VG_EVEN_ODD) { + dsa.stencil[0].writemask = 1; + dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INVERT; + dsa.stencil[0].func = PIPE_FUNC_ALWAYS; + dsa.stencil[0].ref_value = 0; + dsa.stencil[0].valuemask = ~0; + + cso_set_blend(ctx->cso_context, &blend); + cso_set_depth_stencil_alpha(ctx->cso_context, &dsa); + draw_polygon(ctx, poly); + } else if (ctx->state.vg.fill_rule == VG_NON_ZERO) { + struct pipe_screen *screen = ctx->pipe->screen; + + if (screen->get_param(screen, PIPE_CAP_TWO_SIDED_STENCIL)) { + /* front */ + dsa.stencil[0].writemask = ~0; + dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INCR_WRAP; + dsa.stencil[0].func = PIPE_FUNC_ALWAYS; + dsa.stencil[0].ref_value = 0; + dsa.stencil[0].valuemask = ~0; + + /* back */ + dsa.stencil[1].enabled = 1; + dsa.stencil[1].writemask = ~0; + dsa.stencil[1].fail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[1].zfail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[1].zpass_op = PIPE_STENCIL_OP_DECR_WRAP; + dsa.stencil[1].func = PIPE_FUNC_ALWAYS; + dsa.stencil[1].ref_value = 0; + dsa.stencil[1].valuemask = ~0; + + cso_set_blend(ctx->cso_context, &blend); + cso_set_depth_stencil_alpha(ctx->cso_context, &dsa); + draw_polygon(ctx, poly); + } else { + struct pipe_rasterizer_state raster; + + memcpy(&raster, &ctx->state.g3d.rasterizer, sizeof(struct pipe_rasterizer_state)); + + cso_save_rasterizer(ctx->cso_context); + dsa.stencil[0].func = PIPE_FUNC_ALWAYS; + dsa.stencil[0].ref_value = 0; + dsa.stencil[0].valuemask = ~0; + + raster.cull_mode = raster.front_winding ^ PIPE_WINDING_BOTH; + dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INCR_WRAP; + + cso_set_blend(ctx->cso_context, &blend); + cso_set_depth_stencil_alpha(ctx->cso_context, &dsa); + cso_set_rasterizer(ctx->cso_context, &raster); + draw_polygon(ctx, poly); + + raster.cull_mode = raster.front_winding; + dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_DECR_WRAP; + cso_set_depth_stencil_alpha(ctx->cso_context, &dsa); + cso_set_rasterizer(ctx->cso_context, &raster); + draw_polygon(ctx, poly); + + cso_restore_rasterizer(ctx->cso_context); + } + } + + /* restore color writes */ + cso_restore_blend(ctx->cso_context); + /* setup stencil ops */ + dsa.stencil[0].func = PIPE_FUNC_NOTEQUAL; + dsa.stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE; + dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE; + dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE; + dsa.stencil[0].ref_value = 0; + dsa.stencil[0].valuemask = dsa.stencil[0].writemask; + dsa.stencil[1].enabled = 0; + memcpy(&dsa.depth, &ctx->state.g3d.dsa.depth, + sizeof(struct pipe_depth_state)); + cso_set_depth_stencil_alpha(ctx->cso_context, &dsa); + + /* render the quad to propagate the rendering from stencil */ + renderer_draw_quad(ctx->renderer, min_x, min_y, + max_x, max_y, 0.0f/*depth should be disabled*/); + + cso_restore_depth_stencil_alpha(ctx->cso_context); +} + +void polygon_array_fill(struct polygon_array *polyarray, struct vg_context *ctx) +{ + struct array *polys = polyarray->array; + struct pipe_depth_stencil_alpha_state dsa; + struct pipe_blend_state blend; + VGfloat min_x = polyarray->min_x; + VGfloat min_y = polyarray->min_y; + VGfloat max_x = polyarray->max_x; + VGfloat max_y = polyarray->max_y; + VGint i; + + +#if DEBUG_POLYGON + debug_printf("%s: Poly bounds are [%f, %f], [%f, %f]\n", + __FUNCTION__, + min_x, min_y, max_x, max_y); +#endif + + set_blend_for_fill(&blend); + + memset(&dsa, 0, sizeof(struct pipe_depth_stencil_alpha_state)); + + cso_save_blend(ctx->cso_context); + cso_save_depth_stencil_alpha(ctx->cso_context); + + dsa.stencil[0].enabled = 1; + if (ctx->state.vg.fill_rule == VG_EVEN_ODD) { + dsa.stencil[0].writemask = 1; + dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INVERT; + dsa.stencil[0].func = PIPE_FUNC_ALWAYS; + dsa.stencil[0].ref_value = 0; + dsa.stencil[0].valuemask = ~0; + + cso_set_blend(ctx->cso_context, &blend); + cso_set_depth_stencil_alpha(ctx->cso_context, &dsa); + for (i = 0; i < polys->num_elements; ++i) { + struct polygon *poly = (((struct polygon**)polys->data)[i]); + draw_polygon(ctx, poly); + } + } else if (ctx->state.vg.fill_rule == VG_NON_ZERO) { + struct pipe_screen *screen = ctx->pipe->screen; + + if (screen->get_param(screen, PIPE_CAP_TWO_SIDED_STENCIL)) { + /* front */ + dsa.stencil[0].writemask = ~0; + dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INCR_WRAP; + dsa.stencil[0].func = PIPE_FUNC_ALWAYS; + dsa.stencil[0].ref_value = 0; + dsa.stencil[0].valuemask = ~0; + + /* back */ + dsa.stencil[1].enabled = 1; + dsa.stencil[1].writemask = ~0; + dsa.stencil[1].fail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[1].zfail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[1].zpass_op = PIPE_STENCIL_OP_DECR_WRAP; + dsa.stencil[1].func = PIPE_FUNC_ALWAYS; + dsa.stencil[1].ref_value = 0; + dsa.stencil[1].valuemask = ~0; + + cso_set_blend(ctx->cso_context, &blend); + cso_set_depth_stencil_alpha(ctx->cso_context, &dsa); + for (i = 0; i < polys->num_elements; ++i) { + struct polygon *poly = (((struct polygon**)polys->data)[i]); + draw_polygon(ctx, poly); + } + } else { + struct pipe_rasterizer_state raster; + + memcpy(&raster, &ctx->state.g3d.rasterizer, sizeof(struct pipe_rasterizer_state)); + + cso_save_rasterizer(ctx->cso_context); + dsa.stencil[0].func = PIPE_FUNC_ALWAYS; + dsa.stencil[0].ref_value = 0; + dsa.stencil[0].valuemask = ~0; + + raster.cull_mode = raster.front_winding ^ PIPE_WINDING_BOTH; + dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INCR_WRAP; + + cso_set_blend(ctx->cso_context, &blend); + cso_set_depth_stencil_alpha(ctx->cso_context, &dsa); + cso_set_rasterizer(ctx->cso_context, &raster); + for (i = 0; i < polys->num_elements; ++i) { + struct polygon *poly = (((struct polygon**)polys->data)[i]); + draw_polygon(ctx, poly); + } + + raster.cull_mode = raster.front_winding; + dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP; + dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_DECR_WRAP; + cso_set_depth_stencil_alpha(ctx->cso_context, &dsa); + cso_set_rasterizer(ctx->cso_context, &raster); + for (i = 0; i < polys->num_elements; ++i) { + struct polygon *poly = (((struct polygon**)polys->data)[i]); + draw_polygon(ctx, poly); + } + + cso_restore_rasterizer(ctx->cso_context); + } + } + + /* restore color writes */ + cso_restore_blend(ctx->cso_context); + /* setup stencil ops */ + dsa.stencil[0].func = PIPE_FUNC_NOTEQUAL; + dsa.stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE; + dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE; + dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE; + dsa.stencil[0].ref_value = 0; + dsa.stencil[0].valuemask = dsa.stencil[0].writemask; + dsa.stencil[1].enabled = 0; + memcpy(&dsa.depth, &ctx->state.g3d.dsa.depth, + sizeof(struct pipe_depth_state)); + cso_set_depth_stencil_alpha(ctx->cso_context, &dsa); + + /* render the quad to propagate the rendering from stencil */ + renderer_draw_quad(ctx->renderer, min_x, min_y, + max_x, max_y, 0.0f/*depth should be disabled*/); + + cso_restore_depth_stencil_alpha(ctx->cso_context); +} diff --git a/src/gallium/state_trackers/vega/polygon.h b/src/gallium/state_trackers/vega/polygon.h new file mode 100644 index 00000000000..22672b728ed --- /dev/null +++ b/src/gallium/state_trackers/vega/polygon.h @@ -0,0 +1,75 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef POLYGON_H +#define POLYGON_H + +#include "VG/openvg.h" + +struct polygon; +struct vg_context; +struct vg_paint; +struct array; + +struct polygon *polygon_create(int size); +struct polygon *polygon_create_from_data(float *data, int size); +void polygon_destroy(struct polygon *poly); + +void polygon_resize(struct polygon *poly, int new_size); +int polygon_size(struct polygon *poly); + +int polygon_vertex_count(struct polygon *poly); +float * polygon_data(struct polygon *poly); + +void polygon_vertex_append(struct polygon *p, + float x, float y); +void polygon_append_polygon(struct polygon *dst, + struct polygon *src); +void polygon_set_vertex(struct polygon *p, int idx, + float x, float y); +void polygon_vertex(struct polygon *p, int idx, + float *vertex); + +void polygon_bounding_rect(struct polygon *p, + float *rect); +int polygon_contains_point(struct polygon *p, + float x, float y); + +VGboolean polygon_is_closed(struct polygon *p); + +void polygon_fill(struct polygon *p, struct vg_context *pipe); + +/* TODO: make a file/module around this struct + */ +struct polygon_array { + struct array *array; + VGfloat min_x, max_x; + VGfloat min_y, max_y; +}; + +void polygon_array_fill(struct polygon_array *polyarray, struct vg_context *ctx); + +#endif diff --git a/src/gallium/state_trackers/vega/renderer.c b/src/gallium/state_trackers/vega/renderer.c new file mode 100644 index 00000000000..f7c5f2f0cdf --- /dev/null +++ b/src/gallium/state_trackers/vega/renderer.c @@ -0,0 +1,592 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "renderer.h" + +#include "vg_context.h" + +#include "pipe/p_context.h" +#include "pipe/p_state.h" +#include "pipe/p_inlines.h" +#include "pipe/p_screen.h" +#include "pipe/p_shader_tokens.h" + +#include "util/u_draw_quad.h" +#include "util/u_simple_shaders.h" +#include "util/u_memory.h" + +#include "cso_cache/cso_context.h" + +struct renderer { + struct pipe_context *pipe; + struct vg_context *owner; + + struct cso_context *cso; + + void *fs; + + VGfloat vertices[4][2][4]; +}; + +static void setup_shaders(struct renderer *ctx) +{ + struct pipe_context *pipe = ctx->pipe; + /* fragment shader */ + ctx->fs = util_make_fragment_tex_shader(pipe); +} + +static struct pipe_buffer * +setup_vertex_data(struct renderer *ctx, + float x0, float y0, float x1, float y1, float z) +{ + ctx->vertices[0][0][0] = x0; + ctx->vertices[0][0][1] = y0; + ctx->vertices[0][0][2] = z; + ctx->vertices[0][1][0] = 0.0f; /*s*/ + ctx->vertices[0][1][1] = 0.0f; /*t*/ + + ctx->vertices[1][0][0] = x1; + ctx->vertices[1][0][1] = y0; + ctx->vertices[1][0][2] = z; + ctx->vertices[1][1][0] = 1.0f; /*s*/ + ctx->vertices[1][1][1] = 0.0f; /*t*/ + + ctx->vertices[2][0][0] = x1; + ctx->vertices[2][0][1] = y1; + ctx->vertices[2][0][2] = z; + ctx->vertices[2][1][0] = 1.0f; + ctx->vertices[2][1][1] = 1.0f; + + ctx->vertices[3][0][0] = x0; + ctx->vertices[3][0][1] = y1; + ctx->vertices[3][0][2] = z; + ctx->vertices[3][1][0] = 0.0f; + ctx->vertices[3][1][1] = 1.0f; + + return pipe_user_buffer_create( ctx->pipe->screen, + ctx->vertices, + sizeof(ctx->vertices) ); +} + +static struct pipe_buffer * +setup_vertex_data_tex(struct renderer *ctx, + float x0, float y0, float x1, float y1, + float s0, float t0, float s1, float t1, + float z) +{ + ctx->vertices[0][0][0] = x0; + ctx->vertices[0][0][1] = y0; + ctx->vertices[0][0][2] = z; + ctx->vertices[0][1][0] = s0; /*s*/ + ctx->vertices[0][1][1] = t0; /*t*/ + + ctx->vertices[1][0][0] = x1; + ctx->vertices[1][0][1] = y0; + ctx->vertices[1][0][2] = z; + ctx->vertices[1][1][0] = s1; /*s*/ + ctx->vertices[1][1][1] = t0; /*t*/ + + ctx->vertices[2][0][0] = x1; + ctx->vertices[2][0][1] = y1; + ctx->vertices[2][0][2] = z; + ctx->vertices[2][1][0] = s1; + ctx->vertices[2][1][1] = t1; + + ctx->vertices[3][0][0] = x0; + ctx->vertices[3][0][1] = y1; + ctx->vertices[3][0][2] = z; + ctx->vertices[3][1][0] = s0; + ctx->vertices[3][1][1] = t1; + + return pipe_user_buffer_create( ctx->pipe->screen, + ctx->vertices, + sizeof(ctx->vertices) ); +} + + +static struct pipe_buffer * +setup_vertex_data_qtex(struct renderer *ctx, + float x0, float y0, float x1, float y1, + float x2, float y2, float x3, float y3, + float s0, float t0, float s1, float t1, + float z) +{ + ctx->vertices[0][0][0] = x0; + ctx->vertices[0][0][1] = y0; + ctx->vertices[0][0][2] = z; + ctx->vertices[0][1][0] = s0; /*s*/ + ctx->vertices[0][1][1] = t0; /*t*/ + + ctx->vertices[1][0][0] = x1; + ctx->vertices[1][0][1] = y1; + ctx->vertices[1][0][2] = z; + ctx->vertices[1][1][0] = s1; /*s*/ + ctx->vertices[1][1][1] = t0; /*t*/ + + ctx->vertices[2][0][0] = x2; + ctx->vertices[2][0][1] = y2; + ctx->vertices[2][0][2] = z; + ctx->vertices[2][1][0] = s1; + ctx->vertices[2][1][1] = t1; + + ctx->vertices[3][0][0] = x3; + ctx->vertices[3][0][1] = y3; + ctx->vertices[3][0][2] = z; + ctx->vertices[3][1][0] = s0; + ctx->vertices[3][1][1] = t1; + + return pipe_user_buffer_create( ctx->pipe->screen, + ctx->vertices, + sizeof(ctx->vertices) ); +} + +struct renderer * renderer_create(struct vg_context *owner) +{ + VGint i; + struct renderer *renderer = CALLOC_STRUCT(renderer); + + if (!renderer) + return NULL; + + renderer->owner = owner; + renderer->pipe = owner->pipe; + renderer->cso = owner->cso_context; + + setup_shaders(renderer); + + /* init vertex data that doesn't change */ + for (i = 0; i < 4; i++) { + renderer->vertices[i][0][3] = 1.0f; /* w */ + renderer->vertices[i][1][2] = 0.0f; /* r */ + renderer->vertices[i][1][3] = 1.0f; /* q */ + } + + return renderer; +} + +void renderer_destroy(struct renderer *ctx) +{ +#if 0 + if (ctx->fs) { + cso_delete_fragment_shader(ctx->cso, ctx->fs); + ctx->fs = NULL; + } +#endif + free(ctx); +} + +void renderer_draw_quad(struct renderer *r, + VGfloat x1, VGfloat y1, + VGfloat x2, VGfloat y2, + VGfloat depth) +{ + struct pipe_buffer *buf; + + buf = setup_vertex_data(r, x1, y1, x2, y2, depth); + + if (buf) { + util_draw_vertex_buffer(r->pipe, buf, 0, + PIPE_PRIM_TRIANGLE_FAN, + 4, /* verts */ + 2); /* attribs/vert */ + + pipe_buffer_reference( &buf, + NULL ); + } +} + +void renderer_draw_texture(struct renderer *r, + struct pipe_texture *tex, + VGfloat x1offset, VGfloat y1offset, + VGfloat x2offset, VGfloat y2offset, + VGfloat x1, VGfloat y1, + VGfloat x2, VGfloat y2) +{ + struct pipe_context *pipe = r->pipe; + struct pipe_buffer *buf; + VGfloat s0, t0, s1, t1; + + assert(tex->width[0] != 0); + assert(tex->height[0] != 0); + + s0 = x1offset / tex->width[0]; + s1 = x2offset / tex->width[0]; + t0 = y1offset / tex->height[0]; + t1 = y2offset / tex->height[0]; + + cso_save_vertex_shader(r->cso); + /* shaders */ + cso_set_vertex_shader_handle(r->cso, vg_texture_vs(r->owner)); + + /* draw quad */ + buf = setup_vertex_data_tex(r, x1, y1, x2, y2, + s0, t0, s1, t1, 0.0f); + + if (buf) { + util_draw_vertex_buffer(pipe, buf, 0, + PIPE_PRIM_TRIANGLE_FAN, + 4, /* verts */ + 2); /* attribs/vert */ + + pipe_buffer_reference( &buf, + NULL ); + } + + cso_restore_vertex_shader(r->cso); +} + +void renderer_copy_texture(struct renderer *ctx, + struct pipe_texture *src, + VGfloat sx1, VGfloat sy1, + VGfloat sx2, VGfloat sy2, + struct pipe_texture *dst, + VGfloat dx1, VGfloat dy1, + VGfloat dx2, VGfloat dy2) +{ + struct pipe_context *pipe = ctx->pipe; + struct pipe_screen *screen = pipe->screen; + struct pipe_buffer *buf; + struct pipe_surface *dst_surf = screen->get_tex_surface( + screen, dst, 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_WRITE); + struct pipe_framebuffer_state fb; + float s0, t0, s1, t1; + + assert(src->width[0] != 0); + assert(src->height[0] != 0); + assert(dst->width[0] != 0); + assert(dst->height[0] != 0); + +#if 0 + debug_printf("copy texture [%f, %f, %f, %f], [%f, %f, %f, %f]\n", + sx1, sy1, sx2, sy2, dx1, dy1, dx2, dy2); +#endif + +#if 1 + s0 = sx1 / src->width[0]; + s1 = sx2 / src->width[0]; + t0 = sy1 / src->height[0]; + t1 = sy2 / src->height[0]; +#else + s0 = 0; + s1 = 1; + t0 = 0; + t1 = 1; +#endif + + assert(screen->is_format_supported(screen, dst_surf->format, PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_RENDER_TARGET, 0)); + + /* save state (restored below) */ + cso_save_blend(ctx->cso); + cso_save_samplers(ctx->cso); + cso_save_sampler_textures(ctx->cso); + cso_save_framebuffer(ctx->cso); + cso_save_fragment_shader(ctx->cso); + cso_save_vertex_shader(ctx->cso); + + cso_save_viewport(ctx->cso); + + + /* set misc state we care about */ + { + struct pipe_blend_state blend; + memset(&blend, 0, sizeof(blend)); + blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE; + blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE; + blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO; + blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO; + blend.colormask = PIPE_MASK_RGBA; + cso_set_blend(ctx->cso, &blend); + } + + /* sampler */ + { + struct pipe_sampler_state sampler; + memset(&sampler, 0, sizeof(sampler)); + sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE; + sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST; + sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST; + sampler.normalized_coords = 1; + cso_single_sampler(ctx->cso, 0, &sampler); + cso_single_sampler_done(ctx->cso); + } + + vg_set_viewport(ctx->owner, VEGA_Y0_TOP); + + /* texture */ + cso_set_sampler_textures(ctx->cso, 1, &src); + + /* shaders */ + cso_set_vertex_shader_handle(ctx->cso, vg_texture_vs(ctx->owner)); + cso_set_fragment_shader_handle(ctx->cso, ctx->fs); + + /* drawing dest */ + memset(&fb, 0, sizeof(fb)); + fb.width = dst_surf->width; + fb.height = dst_surf->height; + fb.nr_cbufs = 1; + fb.cbufs[0] = dst_surf; + { + VGint i; + for (i = 1; i < PIPE_MAX_COLOR_BUFS; ++i) + fb.cbufs[i] = 0; + } + cso_set_framebuffer(ctx->cso, &fb); + + /* draw quad */ + buf = setup_vertex_data_tex(ctx, + dx1, dy1, + dx2, dy2, + s0, t0, s1, t1, + 0.0f); + + if (buf) { + util_draw_vertex_buffer(ctx->pipe, buf, 0, + PIPE_PRIM_TRIANGLE_FAN, + 4, /* verts */ + 2); /* attribs/vert */ + + pipe_buffer_reference( &buf, + NULL ); + } + + /* restore state we changed */ + cso_restore_blend(ctx->cso); + cso_restore_samplers(ctx->cso); + cso_restore_sampler_textures(ctx->cso); + cso_restore_framebuffer(ctx->cso); + cso_restore_vertex_shader(ctx->cso); + cso_restore_fragment_shader(ctx->cso); + cso_restore_viewport(ctx->cso); + + pipe_surface_reference(&dst_surf, NULL); +} + +void renderer_copy_surface(struct renderer *ctx, + struct pipe_surface *src, + int srcX0, int srcY0, + int srcX1, int srcY1, + struct pipe_surface *dst, + int dstX0, int dstY0, + int dstX1, int dstY1, + float z, unsigned filter) +{ + struct pipe_context *pipe = ctx->pipe; + struct pipe_screen *screen = pipe->screen; + struct pipe_buffer *buf; + struct pipe_texture texTemp, *tex; + struct pipe_surface *texSurf; + struct pipe_framebuffer_state fb; + struct st_framebuffer *stfb = ctx->owner->draw_buffer; + const int srcW = abs(srcX1 - srcX0); + const int srcH = abs(srcY1 - srcY0); + const int srcLeft = MIN2(srcX0, srcX1); + const int srcTop = MIN2(srcY0, srcY1); + + assert(filter == PIPE_TEX_MIPFILTER_NEAREST || + filter == PIPE_TEX_MIPFILTER_LINEAR); + + if (srcLeft != srcX0) { + /* left-right flip */ + int tmp = dstX0; + dstX0 = dstX1; + dstX1 = tmp; + } + + if (srcTop != srcY0) { + /* up-down flip */ + int tmp = dstY0; + dstY0 = dstY1; + dstY1 = tmp; + } + + assert(screen->is_format_supported(screen, src->format, PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_SAMPLER, 0)); + assert(screen->is_format_supported(screen, dst->format, PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_SAMPLER, 0)); + assert(screen->is_format_supported(screen, dst->format, PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_RENDER_TARGET, 0)); + + /* + * XXX for now we're always creating a temporary texture. + * Strictly speaking that's not always needed. + */ + + /* create temp texture */ + memset(&texTemp, 0, sizeof(texTemp)); + texTemp.target = PIPE_TEXTURE_2D; + texTemp.format = src->format; + texTemp.last_level = 0; + texTemp.width[0] = srcW; + texTemp.height[0] = srcH; + texTemp.depth[0] = 1; + pf_get_block(src->format, &texTemp.block); + + tex = screen->texture_create(screen, &texTemp); + if (!tex) + return; + + texSurf = screen->get_tex_surface(screen, tex, 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_WRITE); + + /* load temp texture */ + pipe->surface_copy(pipe, + texSurf, 0, 0, /* dest */ + src, srcLeft, srcTop, /* src */ + srcW, srcH); /* size */ + + /* free the surface, update the texture if necessary.*/ + screen->tex_surface_destroy(texSurf); + + /* save state (restored below) */ + cso_save_blend(ctx->cso); + cso_save_samplers(ctx->cso); + cso_save_sampler_textures(ctx->cso); + cso_save_framebuffer(ctx->cso); + cso_save_fragment_shader(ctx->cso); + cso_save_vertex_shader(ctx->cso); + cso_save_viewport(ctx->cso); + + /* set misc state we care about */ + { + struct pipe_blend_state blend; + memset(&blend, 0, sizeof(blend)); + blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE; + blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE; + blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO; + blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO; + blend.colormask = PIPE_MASK_RGBA; + cso_set_blend(ctx->cso, &blend); + } + + vg_set_viewport(ctx->owner, VEGA_Y0_TOP); + + /* sampler */ + { + struct pipe_sampler_state sampler; + memset(&sampler, 0, sizeof(sampler)); + sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE; + sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST; + sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST; + sampler.normalized_coords = 1; + cso_single_sampler(ctx->cso, 0, &sampler); + cso_single_sampler_done(ctx->cso); + } + + /* texture */ + cso_set_sampler_textures(ctx->cso, 1, &tex); + + /* shaders */ + cso_set_fragment_shader_handle(ctx->cso, ctx->fs); + cso_set_vertex_shader_handle(ctx->cso, vg_texture_vs(ctx->owner)); + + /* drawing dest */ + if (stfb->strb->surface != dst) { + memset(&fb, 0, sizeof(fb)); + fb.width = dst->width; + fb.height = dst->height; + fb.nr_cbufs = 1; + fb.cbufs[0] = dst; + fb.zsbuf = stfb->dsrb->surface; + cso_set_framebuffer(ctx->cso, &fb); + } + + /* draw quad */ + buf = setup_vertex_data(ctx, + (float) dstX0, (float) dstY0, + (float) dstX1, (float) dstY1, z); + + if (buf) { + util_draw_vertex_buffer(ctx->pipe, buf, 0, + PIPE_PRIM_TRIANGLE_FAN, + 4, /* verts */ + 2); /* attribs/vert */ + + pipe_buffer_reference( &buf, + NULL ); + } + + + /* restore state we changed */ + cso_restore_blend(ctx->cso); + cso_restore_samplers(ctx->cso); + cso_restore_sampler_textures(ctx->cso); + cso_restore_framebuffer(ctx->cso); + cso_restore_fragment_shader(ctx->cso); + cso_restore_vertex_shader(ctx->cso); + cso_restore_viewport(ctx->cso); + + pipe_texture_reference(&tex, NULL); +} + +void renderer_texture_quad(struct renderer *r, + struct pipe_texture *tex, + VGfloat x1offset, VGfloat y1offset, + VGfloat x2offset, VGfloat y2offset, + VGfloat x1, VGfloat y1, + VGfloat x2, VGfloat y2, + VGfloat x3, VGfloat y3, + VGfloat x4, VGfloat y4) +{ + struct pipe_context *pipe = r->pipe; + struct pipe_buffer *buf; + VGfloat s0, t0, s1, t1; + + assert(tex->width[0] != 0); + assert(tex->height[0] != 0); + + s0 = x1offset / tex->width[0]; + s1 = x2offset / tex->width[0]; + t0 = y1offset / tex->height[0]; + t1 = y2offset / tex->height[0]; + + cso_save_vertex_shader(r->cso); + /* shaders */ + cso_set_vertex_shader_handle(r->cso, vg_texture_vs(r->owner)); + + /* draw quad */ + buf = setup_vertex_data_qtex(r, x1, y1, x2, y2, x3, y3, x4, y4, + s0, t0, s1, t1, 0.0f); + + if (buf) { + util_draw_vertex_buffer(pipe, buf, 0, + PIPE_PRIM_TRIANGLE_FAN, + 4, /* verts */ + 2); /* attribs/vert */ + + pipe_buffer_reference(&buf, + NULL); + } + + cso_restore_vertex_shader(r->cso); +} diff --git a/src/gallium/state_trackers/vega/renderer.h b/src/gallium/state_trackers/vega/renderer.h new file mode 100644 index 00000000000..990cd32c313 --- /dev/null +++ b/src/gallium/state_trackers/vega/renderer.h @@ -0,0 +1,76 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef RENDERER_H +#define RENDERER_H + +#include "VG/openvg.h" + +struct renderer; + +struct vg_context; +struct pipe_texture; +struct pipe_surface; + +struct renderer *renderer_create(struct vg_context *owner); +void renderer_destroy(struct renderer *); + +void renderer_draw_quad(struct renderer *, + VGfloat x1, VGfloat y1, + VGfloat x2, VGfloat y2, + VGfloat depth); +void renderer_draw_texture(struct renderer *, + struct pipe_texture *texture, + VGfloat x1offset, VGfloat y1offset, + VGfloat x2offset, VGfloat y2offset, + VGfloat x1, VGfloat y1, + VGfloat x2, VGfloat y2); +void renderer_texture_quad(struct renderer *, + struct pipe_texture *texture, + VGfloat x1offset, VGfloat y1offset, + VGfloat x2offset, VGfloat y2offset, + VGfloat x1, VGfloat y1, + VGfloat x2, VGfloat y2, + VGfloat x3, VGfloat y3, + VGfloat x4, VGfloat y4); +void renderer_copy_texture(struct renderer *r, + struct pipe_texture *src, + VGfloat sx1, VGfloat sy1, + VGfloat sx2, VGfloat sy2, + struct pipe_texture *dst, + VGfloat dx1, VGfloat dy1, + VGfloat dx2, VGfloat dy2); +void renderer_copy_surface(struct renderer *r, + struct pipe_surface *src, + int sx1, int sy1, + int sx2, int sy2, + struct pipe_surface *dst, + int dx1, int dy1, + int dx2, int dy2, + float z, unsigned filter); + + +#endif diff --git a/src/gallium/state_trackers/vega/shader.c b/src/gallium/state_trackers/vega/shader.c new file mode 100644 index 00000000000..d9074a377b3 --- /dev/null +++ b/src/gallium/state_trackers/vega/shader.c @@ -0,0 +1,310 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "shader.h" + +#include "vg_context.h" +#include "shaders_cache.h" +#include "paint.h" +#include "mask.h" +#include "image.h" + +#include "pipe/p_context.h" +#include "pipe/p_screen.h" +#include "pipe/p_state.h" +#include "pipe/p_inlines.h" +#include "util/u_memory.h" + +#define MAX_CONSTANTS 20 + +struct shader { + struct vg_context *context; + + VGboolean masking; + struct vg_paint *paint; + struct vg_image *image; + + VGboolean drawing_image; + VGImageMode image_mode; + + float constants[MAX_CONSTANTS]; + struct pipe_constant_buffer cbuf; + struct pipe_shader_state fs_state; + void *fs; +}; + +struct shader * shader_create(struct vg_context *ctx) +{ + struct shader *shader = 0; + + shader = CALLOC_STRUCT(shader); + shader->context = ctx; + + return shader; +} + +void shader_destroy(struct shader *shader) +{ + free(shader); +} + +void shader_set_masking(struct shader *shader, VGboolean set) +{ + shader->masking = set; +} + +VGboolean shader_is_masking(struct shader *shader) +{ + return shader->masking; +} + +void shader_set_paint(struct shader *shader, struct vg_paint *paint) +{ + shader->paint = paint; +} + +struct vg_paint * shader_paint(struct shader *shader) +{ + return shader->paint; +} + + +static void setup_constant_buffer(struct shader *shader) +{ + struct vg_context *ctx = shader->context; + struct pipe_context *pipe = shader->context->pipe; + struct pipe_constant_buffer *cbuf = &shader->cbuf; + VGint param_bytes = paint_constant_buffer_size(shader->paint); + float temp_buf[MAX_CONSTANTS]; + + assert(param_bytes <= sizeof(temp_buf)); + paint_fill_constant_buffer(shader->paint, temp_buf); + + if (cbuf->buffer == NULL || + memcmp(temp_buf, shader->constants, param_bytes) != 0) + { + pipe_buffer_reference(&cbuf->buffer, NULL); + + memcpy(shader->constants, temp_buf, param_bytes); + cbuf->buffer = pipe_user_buffer_create(pipe->screen, + &shader->constants, + sizeof(shader->constants)); + } + + ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, cbuf); +} + +static VGint blend_bind_samplers(struct vg_context *ctx, + struct pipe_sampler_state **samplers, + struct pipe_texture **textures) +{ + VGBlendMode bmode = ctx->state.vg.blend_mode; + + if (bmode == VG_BLEND_MULTIPLY || + bmode == VG_BLEND_SCREEN || + bmode == VG_BLEND_DARKEN || + bmode == VG_BLEND_LIGHTEN) { + struct st_framebuffer *stfb = ctx->draw_buffer; + + vg_prepare_blend_surface(ctx); + + samplers[2] = &ctx->blend_sampler; + textures[2] = stfb->blend_texture; + + if (!samplers[0] || !textures[0]) { + samplers[1] = samplers[2]; + textures[1] = textures[2]; + } + if (!samplers[1] || !textures[1]) { + samplers[1] = samplers[0]; + textures[1] = textures[0]; + } + + return 1; + } + return 0; +} + +static void setup_samplers(struct shader *shader) +{ + struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS]; + struct pipe_texture *textures[PIPE_MAX_SAMPLERS]; + struct vg_context *ctx = shader->context; + /* a little wonky: we use the num as a boolean that just says + * whether any sampler/textures have been set. the actual numbering + * for samplers is always the same: + * 0 - paint sampler/texture for gradient/pattern + * 1 - mask sampler/texture + * 2 - blend sampler/texture + * 3 - image sampler/texture + * */ + VGint num = 0; + + samplers[0] = NULL; + samplers[1] = NULL; + samplers[2] = NULL; + samplers[3] = NULL; + textures[0] = NULL; + textures[1] = NULL; + textures[2] = NULL; + textures[3] = NULL; + + num += paint_bind_samplers(shader->paint, samplers, textures); + num += mask_bind_samplers(samplers, textures); + num += blend_bind_samplers(ctx, samplers, textures); + if (shader->drawing_image && shader->image) + num += image_bind_samplers(shader->image, samplers, textures); + + if (num) { + cso_set_samplers(ctx->cso_context, 4, (const struct pipe_sampler_state **)samplers); + cso_set_sampler_textures(ctx->cso_context, 4, textures); + } +} + +static INLINE VGboolean is_format_bw(struct shader *shader) +{ +#if 0 + struct vg_context *ctx = shader->context; + struct st_framebuffer *stfb = ctx->draw_buffer; +#endif + + if (shader->drawing_image && shader->image) { + if (shader->image->format == VG_BW_1) + return VG_TRUE; + } + + return VG_FALSE; +} + +static void setup_shader_program(struct shader *shader) +{ + struct vg_context *ctx = shader->context; + VGint shader_id = 0; + VGBlendMode blend_mode = ctx->state.vg.blend_mode; + VGboolean black_white = is_format_bw(shader); + + /* 1st stage: fill */ + if (!shader->drawing_image || + (shader->image_mode == VG_DRAW_IMAGE_MULTIPLY || shader->image_mode == VG_DRAW_IMAGE_STENCIL)) { + switch(paint_type(shader->paint)) { + case VG_PAINT_TYPE_COLOR: + shader_id |= VEGA_SOLID_FILL_SHADER; + break; + case VG_PAINT_TYPE_LINEAR_GRADIENT: + shader_id |= VEGA_LINEAR_GRADIENT_SHADER; + break; + case VG_PAINT_TYPE_RADIAL_GRADIENT: + shader_id |= VEGA_RADIAL_GRADIENT_SHADER; + break; + case VG_PAINT_TYPE_PATTERN: + shader_id |= VEGA_PATTERN_SHADER; + break; + + default: + abort(); + } + } + + /* second stage image */ + if (shader->drawing_image) { + switch(shader->image_mode) { + case VG_DRAW_IMAGE_NORMAL: + shader_id |= VEGA_IMAGE_NORMAL_SHADER; + break; + case VG_DRAW_IMAGE_MULTIPLY: + shader_id |= VEGA_IMAGE_MULTIPLY_SHADER; + break; + case VG_DRAW_IMAGE_STENCIL: + shader_id |= VEGA_IMAGE_STENCIL_SHADER; + break; + default: + debug_printf("Unknown image mode!"); + } + } + + if (shader->masking) + shader_id |= VEGA_MASK_SHADER; + + switch(blend_mode) { + case VG_BLEND_MULTIPLY: + shader_id |= VEGA_BLEND_MULTIPLY_SHADER; + break; + case VG_BLEND_SCREEN: + shader_id |= VEGA_BLEND_SCREEN_SHADER; + break; + case VG_BLEND_DARKEN: + shader_id |= VEGA_BLEND_DARKEN_SHADER; + break; + case VG_BLEND_LIGHTEN: + shader_id |= VEGA_BLEND_LIGHTEN_SHADER; + break; + default: + /* handled by pipe_blend_state */ + break; + } + + if (black_white) + shader_id |= VEGA_BW_SHADER; + + shader->fs = shaders_cache_fill(ctx->sc, shader_id); + cso_set_fragment_shader_handle(ctx->cso_context, shader->fs); +} + + +void shader_bind(struct shader *shader) +{ + /* first resolve the real paint type */ + paint_resolve_type(shader->paint); + + setup_constant_buffer(shader); + setup_samplers(shader); + setup_shader_program(shader); +} + +void shader_set_image_mode(struct shader *shader, VGImageMode image_mode) +{ + shader->image_mode = image_mode; +} + +VGImageMode shader_image_mode(struct shader *shader) +{ + return shader->image_mode; +} + +void shader_set_drawing_image(struct shader *shader, VGboolean drawing_image) +{ + shader->drawing_image = drawing_image; +} + +VGboolean shader_drawing_image(struct shader *shader) +{ + return shader->drawing_image; +} + +void shader_set_image(struct shader *shader, struct vg_image *img) +{ + shader->image = img; +} diff --git a/src/gallium/state_trackers/vega/shader.h b/src/gallium/state_trackers/vega/shader.h new file mode 100644 index 00000000000..847eee6a310 --- /dev/null +++ b/src/gallium/state_trackers/vega/shader.h @@ -0,0 +1,56 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef SHADER_H +#define SHADER_H + +#include "VG/openvg.h" + +struct shader; +struct vg_paint; +struct vg_context; +struct vg_image; + +struct shader *shader_create(struct vg_context *context); +void shader_destroy(struct shader *shader); + +void shader_set_masking(struct shader *shader, VGboolean set); +VGboolean shader_is_masking(struct shader *shader); + +void shader_set_paint(struct shader *shader, struct vg_paint *paint); +struct vg_paint *shader_paint(struct shader *shader); + +void shader_set_image_mode(struct shader *shader, VGImageMode image_mode); +VGImageMode shader_image_mode(struct shader *shader); + +void shader_set_drawing_image(struct shader *shader, VGboolean drawing_image); +VGboolean shader_drawing_image(struct shader *shader); + +void shader_set_image(struct shader *shader, struct vg_image *img); + +void shader_bind(struct shader *shader); + +#endif diff --git a/src/gallium/state_trackers/vega/shaders_cache.c b/src/gallium/state_trackers/vega/shaders_cache.c new file mode 100644 index 00000000000..fd0831fab1f --- /dev/null +++ b/src/gallium/state_trackers/vega/shaders_cache.c @@ -0,0 +1,439 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "shaders_cache.h" + +#include "vg_context.h" + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_inlines.h" +#include "pipe/p_screen.h" +#include "pipe/p_shader_tokens.h" + +#include "tgsi/tgsi_build.h" +#include "tgsi/tgsi_dump.h" +#include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_util.h" +#include "tgsi/tgsi_text.h" + +#include "util/u_memory.h" +#include "util/u_math.h" +#include "util/u_debug.h" +#include "cso_cache/cso_hash.h" +#include "cso_cache/cso_context.h" + +#include "VG/openvg.h" + +#include "asm_fill.h" + +/* Essentially we construct an ubber-shader based on the state + * of the pipeline. The stages are: + * 1) Fill (mandatory, solid color/gradient/pattern/image draw) + * 2) Image composition (image mode multiply and stencil) + * 3) Mask + * 4) Extended blend (multiply/screen/darken/lighten) + * 5) Premultiply/Unpremultiply + * 6) Color transform (to black and white) + */ +#define SHADER_STAGES 6 + +struct cached_shader { + void *driver_shader; + struct pipe_shader_state state; +}; + +struct shaders_cache { + struct vg_context *pipe; + + struct cso_hash *hash; +}; + + +static INLINE struct tgsi_token *tokens_from_assembly(const char *txt, int num_tokens) +{ + struct tgsi_token *tokens; + + tokens = (struct tgsi_token *) MALLOC(num_tokens * sizeof(tokens[0])); + + tgsi_text_translate(txt, tokens, num_tokens); + +#if DEBUG_SHADERS + tgsi_dump(tokens, 0); +#endif + + return tokens; +} + +#define ALL_FILLS (VEGA_SOLID_FILL_SHADER | \ + VEGA_LINEAR_GRADIENT_SHADER | \ + VEGA_RADIAL_GRADIENT_SHADER | \ + VEGA_PATTERN_SHADER | \ + VEGA_IMAGE_NORMAL_SHADER) + + +/* +static const char max_shader_preamble[] = + "FRAG1.1\n" + "DCL IN[0], POSITION, LINEAR\n" + "DCL IN[1], GENERIC[0], PERSPECTIVE\n" + "DCL OUT[0], COLOR, CONSTANT\n" + "DCL CONST[0..9], CONSTANT\n" + "DCL TEMP[0..9], CONSTANT\n" + "DCL SAMP[0..9], CONSTANT\n"; + + max_shader_preamble strlen == 175 +*/ +#define MAX_PREAMBLE 175 + +static INLINE VGint range_min(VGint min, VGint current) +{ + if (min < 0) + min = current; + else + min = MIN2(min, current); + return min; +} + +static INLINE VGint range_max(VGint max, VGint current) +{ + return MAX2(max, current); +} + +static void +create_preamble(char *txt, + const struct shader_asm_info *shaders[SHADER_STAGES], + int num_shaders) +{ + VGboolean declare_input = VG_FALSE; + VGint start_const = -1, end_const = 0; + VGint start_temp = -1, end_temp = 0; + VGint start_sampler = -1, end_sampler = 0; + VGint i; + VGint num_consts, num_temps, num_samplers; + + for (i = 0; i < num_shaders; ++i) { + if (shaders[i]->num_consts) + start_const = range_min(start_const, shaders[i]->start_const); + if (shaders[i]->num_temps) + start_temp = range_min(start_temp, shaders[i]->start_temp); + if (shaders[i]->num_samplers) + start_sampler = range_min(start_sampler, shaders[i]->start_sampler); + + end_const = range_max(end_const, shaders[i]->start_const + + shaders[i]->num_consts); + end_temp = range_max(end_temp, shaders[i]->start_temp + + shaders[i]->num_temps); + end_sampler = range_max(end_sampler, shaders[i]->start_sampler + + shaders[i]->num_samplers); + if (shaders[i]->needs_position) + declare_input = VG_TRUE; + } + /* if they're still unitialized, initialize them */ + if (start_const < 0) + start_const = 0; + if (start_temp < 0) + start_temp = 0; + if (start_sampler < 0) + start_sampler = 0; + + num_consts = end_const - start_const; + num_temps = end_temp - start_temp; + num_samplers = end_sampler - start_sampler; + /* end exclusive */ + --end_const; + --end_temp; + --end_sampler; + + sprintf(txt, "FRAG1.1\n"); + + if (declare_input) { + sprintf(txt + strlen(txt), "DCL IN[0], POSITION, LINEAR\n"); + sprintf(txt + strlen(txt), "DCL IN[1], GENERIC[0], PERSPECTIVE\n"); + } + + /* we always have a color output */ + sprintf(txt + strlen(txt), "DCL OUT[0], COLOR, CONSTANT\n"); + + if (num_consts > 1) + sprintf(txt + strlen(txt), "DCL CONST[%d..%d], CONSTANT\n", start_const, end_const); + else if (num_consts == 1) + sprintf(txt + strlen(txt), "DCL CONST[%d], CONSTANT\n", start_const); + + if (num_temps > 1) + sprintf(txt + strlen(txt), "DCL TEMP[%d..%d], CONSTANT\n", start_temp, end_temp); + else if (num_temps > 1) + sprintf(txt + strlen(txt), "DCL TEMP[%d], CONSTANT\n", start_temp); + + if (num_samplers > 1) + sprintf(txt + strlen(txt), "DCL SAMP[%d..%d], CONSTANT\n", start_sampler, end_sampler); + else if (num_samplers == 1) + sprintf(txt + strlen(txt), "DCL SAMP[%d], CONSTANT\n", start_sampler); +} + +static void * +combine_shaders(const struct shader_asm_info *shaders[SHADER_STAGES], int num_shaders, + struct pipe_context *pipe, + struct pipe_shader_state *shader) +{ + char *combined_txt; + int combined_len = MAX_PREAMBLE; + int combined_tokens = 0; + int i = 0; + int current_shader = 0; + int current_len; + + for (i = 0; i < num_shaders; ++i) { + combined_len += strlen(shaders[i]->txt); + combined_tokens += shaders[i]->num_tokens; + } + /* add for the %s->TEMP[0] substitutions */ + combined_len += num_shaders * 7 /*TEMP[0]*/ + 4 /*"END\n"*/; + + combined_txt = (char*)malloc(combined_len); + combined_txt[0] = '\0'; + + create_preamble(combined_txt, shaders, num_shaders); + + while (current_shader < num_shaders) { + const char temp[] = "TEMP[0]"; + const char out[] = "OUT[0]"; + const char *subst = temp; + + current_len = strlen(combined_txt); + + /* if the last shader then output */ + if (current_shader + 1 == num_shaders) + subst = out; + + snprintf(combined_txt + current_len, + combined_len - current_len, + shaders[current_shader]->txt, + subst); + ++current_shader; + } + + + current_len = strlen(combined_txt); + snprintf(combined_txt + current_len, + combined_len - current_len, + "END\n"); + + debug_printf("Combined shader is : \n%s\n", + combined_txt); + + shader->tokens = tokens_from_assembly( + combined_txt, combined_tokens); + + free(combined_txt); + + return pipe->create_fs_state(pipe, shader); +} + +static void * +create_shader(struct pipe_context *pipe, + int id, + struct pipe_shader_state *shader) +{ + int idx = 0; + const struct shader_asm_info * shaders[SHADER_STAGES]; + + /* the shader has to have a fill */ + debug_assert(id & ALL_FILLS); + + /* first stage */ + if (id & VEGA_SOLID_FILL_SHADER) { + debug_assert(idx == 0); + shaders[idx] = &shaders_asm[0]; + debug_assert(shaders_asm[0].id == VEGA_SOLID_FILL_SHADER); + ++idx; + } + if ((id & VEGA_LINEAR_GRADIENT_SHADER)) { + debug_assert(idx == 0); + shaders[idx] = &shaders_asm[1]; + debug_assert(shaders_asm[1].id == VEGA_LINEAR_GRADIENT_SHADER); + ++idx; + } + if ((id & VEGA_RADIAL_GRADIENT_SHADER)) { + debug_assert(idx == 0); + shaders[idx] = &shaders_asm[2]; + debug_assert(shaders_asm[2].id == VEGA_RADIAL_GRADIENT_SHADER); + ++idx; + } + if ((id & VEGA_PATTERN_SHADER)) { + debug_assert(idx == 0); + debug_assert(shaders_asm[3].id == VEGA_PATTERN_SHADER); + shaders[idx] = &shaders_asm[3]; + ++idx; + } + if ((id & VEGA_IMAGE_NORMAL_SHADER)) { + debug_assert(idx == 0); + debug_assert(shaders_asm[4].id == VEGA_IMAGE_NORMAL_SHADER); + shaders[idx] = &shaders_asm[4]; + ++idx; + } + + /* second stage */ + if ((id & VEGA_IMAGE_MULTIPLY_SHADER)) { + debug_assert(shaders_asm[5].id == VEGA_IMAGE_MULTIPLY_SHADER); + shaders[idx] = &shaders_asm[5]; + ++idx; + } else if ((id & VEGA_IMAGE_STENCIL_SHADER)) { + debug_assert(shaders_asm[6].id == VEGA_IMAGE_STENCIL_SHADER); + shaders[idx] = &shaders_asm[6]; + ++idx; + } + + /* third stage */ + if ((id & VEGA_MASK_SHADER)) { + debug_assert(idx == 1); + debug_assert(shaders_asm[7].id == VEGA_MASK_SHADER); + shaders[idx] = &shaders_asm[7]; + ++idx; + } + + /* fourth stage */ + if ((id & VEGA_BLEND_MULTIPLY_SHADER)) { + debug_assert(shaders_asm[8].id == VEGA_BLEND_MULTIPLY_SHADER); + shaders[idx] = &shaders_asm[8]; + ++idx; + } else if ((id & VEGA_BLEND_SCREEN_SHADER)) { + debug_assert(shaders_asm[9].id == VEGA_BLEND_SCREEN_SHADER); + shaders[idx] = &shaders_asm[9]; + ++idx; + } else if ((id & VEGA_BLEND_DARKEN_SHADER)) { + debug_assert(shaders_asm[10].id == VEGA_BLEND_DARKEN_SHADER); + shaders[idx] = &shaders_asm[10]; + ++idx; + } else if ((id & VEGA_BLEND_LIGHTEN_SHADER)) { + debug_assert(shaders_asm[11].id == VEGA_BLEND_LIGHTEN_SHADER); + shaders[idx] = &shaders_asm[11]; + ++idx; + } + + /* fifth stage */ + if ((id & VEGA_PREMULTIPLY_SHADER)) { + debug_assert(shaders_asm[12].id == VEGA_PREMULTIPLY_SHADER); + shaders[idx] = &shaders_asm[12]; + ++idx; + } else if ((id & VEGA_UNPREMULTIPLY_SHADER)) { + debug_assert(shaders_asm[13].id == VEGA_UNPREMULTIPLY_SHADER); + shaders[idx] = &shaders_asm[13]; + ++idx; + } + + /* sixth stage */ + if ((id & VEGA_BW_SHADER)) { + debug_assert(shaders_asm[14].id == VEGA_BW_SHADER); + shaders[idx] = &shaders_asm[14]; + ++idx; + } + + return combine_shaders(shaders, idx, pipe, shader); +} + +/*************************************************/ + +struct shaders_cache * shaders_cache_create(struct vg_context *vg) +{ + struct shaders_cache *sc = CALLOC_STRUCT(shaders_cache); + + sc->pipe = vg; + sc->hash = cso_hash_create(); + + return sc; +} + +void shaders_cache_destroy(struct shaders_cache *sc) +{ + struct cso_hash_iter iter = cso_hash_first_node(sc->hash); + + while (!cso_hash_iter_is_null(iter)) { + struct cached_shader *cached = + (struct cached_shader *)cso_hash_iter_data(iter); + cso_delete_fragment_shader(sc->pipe->cso_context, + cached->driver_shader); + iter = cso_hash_erase(sc->hash, iter); + } + + cso_hash_delete(sc->hash); + free(sc); +} + +void * shaders_cache_fill(struct shaders_cache *sc, + int shader_key) +{ + VGint key = shader_key; + struct cached_shader *cached; + struct cso_hash_iter iter = cso_hash_find(sc->hash, key); + + if (cso_hash_iter_is_null(iter)) { + cached = CALLOC_STRUCT(cached_shader); + cached->driver_shader = create_shader(sc->pipe->pipe, key, &cached->state); + + cso_hash_insert(sc->hash, key, cached); + + return cached->driver_shader; + } + + cached = (struct cached_shader *)cso_hash_iter_data(iter); + + assert(cached->driver_shader); + return cached->driver_shader; +} + +struct vg_shader * shader_create_from_text(struct pipe_context *pipe, + const char *txt, int num_tokens, + int type) +{ + struct vg_shader *shader = (struct vg_shader *)malloc( + sizeof(struct vg_shader)); + struct tgsi_token *tokens = tokens_from_assembly(txt, num_tokens); + struct pipe_shader_state state; + + debug_assert(type == PIPE_SHADER_VERTEX || + type == PIPE_SHADER_FRAGMENT); + + state.tokens = tokens; + shader->type = type; + shader->tokens = tokens; + + if (type == PIPE_SHADER_FRAGMENT) + shader->driver = pipe->create_fs_state(pipe, &state); + else + shader->driver = pipe->create_vs_state(pipe, &state); + return shader; +} + +void vg_shader_destroy(struct vg_context *ctx, struct vg_shader *shader) +{ + if (shader->type == PIPE_SHADER_FRAGMENT) + cso_delete_fragment_shader(ctx->cso_context, shader->driver); + else + cso_delete_vertex_shader(ctx->cso_context, shader->driver); + free(shader->tokens); + free(shader); +} diff --git a/src/gallium/state_trackers/vega/shaders_cache.h b/src/gallium/state_trackers/vega/shaders_cache.h new file mode 100644 index 00000000000..feca58b61a9 --- /dev/null +++ b/src/gallium/state_trackers/vega/shaders_cache.h @@ -0,0 +1,77 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef SHADERS_CACHE_H +#define SHADERS_CACHE_H + + +struct vg_context; +struct pipe_context; +struct tgsi_token; +struct shaders_cache; + +enum VegaShaderType { + VEGA_SOLID_FILL_SHADER = 1 << 0, + VEGA_LINEAR_GRADIENT_SHADER = 1 << 1, + VEGA_RADIAL_GRADIENT_SHADER = 1 << 2, + VEGA_PATTERN_SHADER = 1 << 3, + VEGA_IMAGE_NORMAL_SHADER = 1 << 4, + VEGA_IMAGE_MULTIPLY_SHADER = 1 << 5, + VEGA_IMAGE_STENCIL_SHADER = 1 << 6, + + VEGA_MASK_SHADER = 1 << 7, + + VEGA_BLEND_MULTIPLY_SHADER = 1 << 8, + VEGA_BLEND_SCREEN_SHADER = 1 << 9, + VEGA_BLEND_DARKEN_SHADER = 1 << 10, + VEGA_BLEND_LIGHTEN_SHADER = 1 << 11, + + VEGA_PREMULTIPLY_SHADER = 1 << 12, + VEGA_UNPREMULTIPLY_SHADER = 1 << 13, + + VEGA_BW_SHADER = 1 << 14 +}; + +struct vg_shader { + void *driver; + struct tgsi_token *tokens; + int type;/* PIPE_SHADER_VERTEX, PIPE_SHADER_FRAGMENT */ +}; + +struct shaders_cache *shaders_cache_create(struct vg_context *pipe); +void shaders_cache_destroy(struct shaders_cache *sc); +void *shaders_cache_fill(struct shaders_cache *sc, + int shader_key); + +struct vg_shader *shader_create_from_text(struct pipe_context *pipe, + const char *txt, int num_tokens, + int type); + +void vg_shader_destroy(struct vg_context *ctx, struct vg_shader *shader); + + + +#endif diff --git a/src/gallium/state_trackers/vega/st_inlines.h b/src/gallium/state_trackers/vega/st_inlines.h new file mode 100644 index 00000000000..1f331dfcdb7 --- /dev/null +++ b/src/gallium/state_trackers/vega/st_inlines.h @@ -0,0 +1,159 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Functions for checking if buffers/textures are referenced when we need + * to read/write from/to them. Flush when needed. + */ + +#ifndef ST_INLINES_H +#define ST_INLINES_H + +#include "vg_context.h" + +#include "pipe/p_context.h" +#include "pipe/p_screen.h" +#include "pipe/p_defines.h" +#include "pipe/p_inlines.h" +#include "pipe/p_state.h" + +static INLINE struct pipe_transfer * +st_cond_flush_get_tex_transfer(struct vg_context *st, + struct pipe_texture *pt, + unsigned int face, + unsigned int level, + unsigned int zslice, + enum pipe_transfer_usage usage, + unsigned int x, unsigned int y, + unsigned int w, unsigned int h) +{ + struct pipe_screen *screen = st->pipe->screen; + struct pipe_context *pipe = st->pipe; + unsigned referenced = + pipe->is_texture_referenced(pipe, pt, face, level); + + if (referenced && ((referenced & PIPE_REFERENCED_FOR_WRITE) || + usage == PIPE_TRANSFER_WRITE || + usage == PIPE_TRANSFER_READ_WRITE)) + vgFlush(); + + return screen->get_tex_transfer(screen, pt, face, level, zslice, usage, + x, y, w, h); +} + +static INLINE struct pipe_transfer * +st_no_flush_get_tex_transfer(struct vg_context *st, + struct pipe_texture *pt, + unsigned int face, + unsigned int level, + unsigned int zslice, + enum pipe_transfer_usage usage, + unsigned int x, unsigned int y, + unsigned int w, unsigned int h) +{ + struct pipe_screen *screen = st->pipe->screen; + + return screen->get_tex_transfer(screen, pt, face, level, + zslice, usage, x, y, w, h); +} + +static INLINE void * +st_cond_flush_pipe_buffer_map(struct vg_context *st, + struct pipe_buffer *buf, + unsigned int map_flags) +{ + struct pipe_context *pipe = st->pipe; + unsigned int referenced = pipe->is_buffer_referenced(pipe, buf); + + if (referenced && ((referenced & PIPE_REFERENCED_FOR_WRITE) || + (map_flags & PIPE_BUFFER_USAGE_CPU_WRITE))) + vgFlush(); + + return pipe_buffer_map(pipe->screen, buf, map_flags); +} + +static INLINE void * +st_no_flush_pipe_buffer_map(struct vg_context *st, + struct pipe_buffer *buf, + unsigned int map_flags) +{ + return pipe_buffer_map(st->pipe->screen, buf, map_flags); +} + + +static INLINE void +st_cond_flush_pipe_buffer_write(struct vg_context *st, + struct pipe_buffer *buf, + unsigned int offset, + unsigned int size, + const void * data) +{ + struct pipe_context *pipe = st->pipe; + + if (pipe->is_buffer_referenced(pipe, buf)) + vgFlush(); + + pipe_buffer_write(pipe->screen, buf, offset, size, data); +} + +static INLINE void +st_no_flush_pipe_buffer_write(struct vg_context *st, + struct pipe_buffer *buf, + unsigned int offset, + unsigned int size, + const void * data) +{ + pipe_buffer_write(st->pipe->screen, buf, offset, size, data); +} + +static INLINE void +st_cond_flush_pipe_buffer_read(struct vg_context *st, + struct pipe_buffer *buf, + unsigned int offset, + unsigned int size, + void * data) +{ + struct pipe_context *pipe = st->pipe; + + if (pipe->is_buffer_referenced(pipe, buf) & PIPE_REFERENCED_FOR_WRITE) + vgFlush(); + + pipe_buffer_read(pipe->screen, buf, offset, size, data); +} + +static INLINE void +st_no_flush_pipe_buffer_read(struct vg_context *st, + struct pipe_buffer *buf, + unsigned int offset, + unsigned int size, + void * data) +{ + pipe_buffer_read(st->pipe->screen, buf, offset, size, data); +} + +#endif + diff --git a/src/gallium/state_trackers/vega/stroker.c b/src/gallium/state_trackers/vega/stroker.c new file mode 100644 index 00000000000..1b92d2b5c62 --- /dev/null +++ b/src/gallium/state_trackers/vega/stroker.c @@ -0,0 +1,1349 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "stroker.h" + +#include "path.h" +#include "vg_state.h" +#include "util_array.h" +#include "arc.h" +#include "bezier.h" +#include "matrix.h" +#include "path_utils.h" +#include "polygon.h" + +#include "math.h" + +#ifndef M_2PI +#define M_2PI 6.28318530717958647692528676655900576 +#endif + +#define STROKE_SEGMENTS 0 +#define STROKE_DEBUG 0 +#define DEBUG_EMITS 0 + +static const VGfloat curve_threshold = 0.25f; + +static const VGfloat zero_coords[] = {0.f, 0.f}; + +enum intersection_type { + NoIntersections, + BoundedIntersection, + UnboundedIntersection, +}; + +enum line_join_mode { + FlatJoin, + SquareJoin, + MiterJoin, + RoundJoin, + RoundCap +}; + +struct stroke_iterator { + void (*next)(struct stroke_iterator *); + VGboolean (*has_next)(struct stroke_iterator *); + + VGPathCommand (*current_command)(struct stroke_iterator *it); + void (*current_coords)(struct stroke_iterator *it, VGfloat *coords); + + VGint position; + VGint coord_position; + + const VGubyte *cmds; + const VGfloat *coords; + VGint num_commands; + VGint num_coords; + + struct polygon *curve_poly; + VGint curve_index; +}; + +static VGPathCommand stroke_itr_command(struct stroke_iterator *itr) +{ + return itr->current_command(itr); +} + +static void stroke_itr_coords(struct stroke_iterator *itr, VGfloat *coords) +{ + itr->current_coords(itr, coords); +} + +static void stroke_fw_itr_coords(struct stroke_iterator *itr, VGfloat *coords) +{ + if (itr->position >= itr->num_commands) + return; + switch (stroke_itr_command(itr)) { + case VG_MOVE_TO_ABS: + coords[0] = itr->coords[itr->coord_position]; + coords[1] = itr->coords[itr->coord_position + 1]; + break; + case VG_LINE_TO_ABS: + coords[0] = itr->coords[itr->coord_position]; + coords[1] = itr->coords[itr->coord_position + 1]; + break; + case VG_CUBIC_TO_ABS: + coords[0] = itr->coords[itr->coord_position]; + coords[1] = itr->coords[itr->coord_position + 1]; + coords[2] = itr->coords[itr->coord_position + 2]; + coords[3] = itr->coords[itr->coord_position + 3]; + coords[4] = itr->coords[itr->coord_position + 4]; + coords[5] = itr->coords[itr->coord_position + 5]; + break; + default: + debug_assert(!"invalid command!\n"); + } +} + + +static void stroke_bw_itr_coords(struct stroke_iterator *itr, VGfloat *coords) +{ + if (itr->position >= itr->num_commands) + return; + switch (stroke_itr_command(itr)) { + case VG_MOVE_TO_ABS: + coords[0] = itr->coords[itr->coord_position]; + coords[1] = itr->coords[itr->coord_position + 1]; + break; + case VG_LINE_TO_ABS: + coords[0] = itr->coords[itr->coord_position]; + coords[1] = itr->coords[itr->coord_position + 1]; + break; + case VG_CUBIC_TO_ABS: + coords[0] = itr->coords[itr->coord_position + 4]; + coords[1] = itr->coords[itr->coord_position + 5]; + coords[2] = itr->coords[itr->coord_position + 2]; + coords[3] = itr->coords[itr->coord_position + 3]; + coords[4] = itr->coords[itr->coord_position + 0]; + coords[5] = itr->coords[itr->coord_position + 1]; + break; + default: + debug_assert(!"invalid command!\n"); + } +} + + +static VGPathCommand stroke_fw_current_command(struct stroke_iterator *it) +{ + return it->cmds[it->position]; +} + +static VGPathCommand stroke_bw_current_command(struct stroke_iterator *it) +{ + VGPathCommand prev_cmd; + if (it->position == it->num_commands -1) + return VG_MOVE_TO_ABS; + + prev_cmd = it->cmds[it->position + 1]; + return prev_cmd; +} + +static VGboolean stroke_fw_has_next(struct stroke_iterator *itr) +{ + return itr->position < (itr->num_commands - 1); +} + +static VGboolean stroke_bw_has_next(struct stroke_iterator *itr) +{ + return itr->position > 0; +} + +static void stroke_fw_next(struct stroke_iterator *itr) +{ + VGubyte cmd; + debug_assert(stroke_fw_has_next(itr)); + + cmd = stroke_itr_command(itr); + + itr->coord_position += num_elements_for_segments(&cmd, 1); + ++itr->position; +} + +static void stroke_bw_next(struct stroke_iterator *itr) +{ + VGubyte cmd; + debug_assert(stroke_bw_has_next(itr)); + + --itr->position; + cmd = stroke_itr_command(itr); + + itr->coord_position -= num_elements_for_segments(&cmd, 1); +} + +static void stroke_itr_common_init(struct stroke_iterator *itr, + struct array *cmds, + struct array *coords) +{ + itr->cmds = (VGubyte*)cmds->data; + itr->num_commands = cmds->num_elements; + + itr->coords = (VGfloat*)coords->data; + itr->num_coords = coords->num_elements; +} + +static void stroke_forward_iterator(struct stroke_iterator *itr, + struct array *cmds, + struct array *coords) +{ + stroke_itr_common_init(itr, cmds, coords); + itr->position = 0; + itr->coord_position = 0; + + itr->next = stroke_fw_next; + itr->has_next = stroke_fw_has_next; + itr->current_command = stroke_fw_current_command; + itr->current_coords = stroke_fw_itr_coords; +} + +static void stroke_backward_iterator(struct stroke_iterator *itr, + struct array *cmds, + struct array *coords) +{ + VGubyte cmd; + stroke_itr_common_init(itr, cmds, coords); + itr->position = itr->num_commands - 1; + + cmd = stroke_bw_current_command(itr); + itr->coord_position = itr->num_coords - + num_elements_for_segments(&cmd, 1); + + itr->next = stroke_bw_next; + itr->has_next = stroke_bw_has_next; + itr->current_command = stroke_bw_current_command; + itr->current_coords = stroke_bw_itr_coords; +} + + + +static void stroke_flat_next(struct stroke_iterator *itr) +{ + VGubyte cmd; + + if (itr->curve_index >= 0) { + ++itr->curve_index; + if (itr->curve_index >= polygon_vertex_count(itr->curve_poly)) { + itr->curve_index = -1; + polygon_destroy(itr->curve_poly); + itr->curve_poly = 0; + } else + return; + } + debug_assert(stroke_fw_has_next(itr)); + + cmd = itr->cmds[itr->position]; + itr->coord_position += num_elements_for_segments(&cmd, 1); + ++itr->position; + + cmd = itr->cmds[itr->position]; + + if (cmd == VG_CUBIC_TO_ABS) { + struct bezier bezier; + VGfloat bez[8]; + + bez[0] = itr->coords[itr->coord_position - 2]; + bez[1] = itr->coords[itr->coord_position - 1]; + bez[2] = itr->coords[itr->coord_position]; + bez[3] = itr->coords[itr->coord_position + 1]; + bez[4] = itr->coords[itr->coord_position + 2]; + bez[5] = itr->coords[itr->coord_position + 3]; + bez[6] = itr->coords[itr->coord_position + 4]; + bez[7] = itr->coords[itr->coord_position + 5]; + + bezier_init(&bezier, + bez[0], bez[1], + bez[2], bez[3], + bez[4], bez[5], + bez[6], bez[7]); + /* skip the first one, it's the same as the prev point */ + itr->curve_index = 1; + if (itr->curve_poly) { + polygon_destroy(itr->curve_poly); + itr->curve_poly = 0; + } + itr->curve_poly = bezier_to_polygon(&bezier); + } +} + +static VGboolean stroke_flat_has_next(struct stroke_iterator *itr) +{ + return (itr->curve_index >= 0 && + itr->curve_index < (polygon_vertex_count(itr->curve_poly)-1)) + || itr->position < (itr->num_commands - 1); +} + +static VGPathCommand stroke_flat_current_command(struct stroke_iterator *it) +{ + if (it->cmds[it->position] == VG_CUBIC_TO_ABS) { + return VG_LINE_TO_ABS; + } + return it->cmds[it->position]; +} + +static void stroke_flat_itr_coords(struct stroke_iterator *itr, VGfloat *coords) +{ + if (itr->curve_index <= -1 && itr->position >= itr->num_commands) + return; + + if (itr->curve_index >= 0) { + polygon_vertex(itr->curve_poly, itr->curve_index, + coords); + return; + } + + switch (stroke_itr_command(itr)) { + case VG_MOVE_TO_ABS: + coords[0] = itr->coords[itr->coord_position]; + coords[1] = itr->coords[itr->coord_position + 1]; + break; + case VG_LINE_TO_ABS: + coords[0] = itr->coords[itr->coord_position]; + coords[1] = itr->coords[itr->coord_position + 1]; + break; + case VG_CUBIC_TO_ABS: + default: + debug_assert(!"invalid command!\n"); + } +} + +static void stroke_flat_iterator(struct stroke_iterator *itr, + struct array *cmds, + struct array *coords) +{ + stroke_itr_common_init(itr, cmds, coords); + itr->position = 0; + itr->coord_position = 0; + + itr->next = stroke_flat_next; + itr->has_next = stroke_flat_has_next; + itr->current_command = stroke_flat_current_command; + itr->current_coords = stroke_flat_itr_coords; + itr->curve_index = -1; + itr->curve_poly = 0; +} + + +static INLINE VGboolean finite_coords4(const VGfloat *c) +{ + return + isfinite(c[0]) && isfinite(c[1]) && + isfinite(c[2]) && isfinite(c[3]); +} + +/* from Graphics Gems II */ +#define SAME_SIGNS(a, b) ((a) * (b) >= 0) +static VGboolean do_lines_intersect(VGfloat x1, VGfloat y1, VGfloat x2, VGfloat y2, + VGfloat x3, VGfloat y3, VGfloat x4, VGfloat y4) +{ + VGfloat a1, a2, b1, b2, c1, c2; /* Coefficients of line eqns */ + VGfloat r1, r2, r3, r4; /* 'sign' values */ + + a1 = y2 - y1; + b1 = x1 - x2; + c1 = x2 * y1 - x1 * y2; + + r3 = a1 * x3 + b1 * y3 + c1; + r4 = a1 * x4 + b1 * y4 + c1; + + if (r3 != 0 && r4 != 0 && SAME_SIGNS(r3, r4)) + return VG_FALSE; + + a2 = y4 - y3; + b2 = x3 - x4; + c2 = x4 * y3 - x3 * y4; + + r1 = a2 * x1 + b2 * y1 + c2; + r2 = a2 * x2 + b2 * y2 + c2; + + if (r1 != 0 && r2 != 0 && SAME_SIGNS(r1, r2)) + return VG_FALSE; + + return VG_TRUE; +} + +static INLINE VGfloat line_dx(const VGfloat *l) +{ + return l[2] - l[0]; +} + +static INLINE VGfloat line_dy(const VGfloat *l) +{ + return l[3] - l[1]; +} + +static INLINE VGfloat line_angle(const VGfloat *l) +{ + const VGfloat dx = line_dx(l); + const VGfloat dy = line_dy(l); + + const VGfloat theta = atan2(-dy, dx) * 360.0 / M_2PI; + + const VGfloat theta_normalized = theta < 0 ? theta + 360 : theta; + + if (floatsEqual(theta_normalized, 360.f)) + return 0; + else + return theta_normalized; +} + +static INLINE void line_set_length(VGfloat *l, VGfloat len) +{ + VGfloat uv[] = {l[0], l[1], l[2], l[3]}; + if (null_line(l)) + return; + line_normalize(uv); + l[2] = l[0] + line_dx(uv) * len; + l[3] = l[1] + line_dy(uv) * len; +} + +static INLINE void line_translate(VGfloat *l, VGfloat x, VGfloat y) +{ + l[0] += x; + l[1] += y; + l[2] += x; + l[3] += y; +} + +static INLINE VGfloat line_angle_to(const VGfloat *l1, + const VGfloat *l2) +{ + VGfloat a1, a2, delta, delta_normalized; + if (null_line(l1) || null_line(l1)) + return 0; + + a1 = line_angle(l1); + a2 = line_angle(l2); + + delta = a2 - a1; + delta_normalized = delta < 0 ? delta + 360 : delta; + + if (floatsEqual(delta, 360.f)) + return 0; + else + return delta_normalized; +} + +static INLINE VGfloat line_angles(const VGfloat *l1, + const VGfloat *l2) +{ + VGfloat cos_line, rad = 0; + + if (null_line(l1) || null_line(l2)) + return 0; + + cos_line = (line_dx(l1)*line_dx(l2) + line_dy(l1)*line_dy(l2)) / + (line_lengthv(l1)*line_lengthv(l2)); + rad = 0; + + if (cos_line >= -1.0 && cos_line <= 1.0) + rad = acos(cos_line); + return rad * 360 / M_2PI; +} + + +static INLINE VGfloat adapted_angle_on_x(const VGfloat *line) +{ + const VGfloat identity[] = {0, 0, 1, 0}; + VGfloat angle = line_angles(line, identity); + if (line_dy(line) > 0) + angle = 360 - angle; + return angle; +} + +static enum intersection_type line_intersect(const VGfloat *l1, + const VGfloat *l2, + float *intersection_point) +{ + VGfloat isect[2]; + enum intersection_type type; + VGboolean dx_zero, ldx_zero; + + if (null_line(l1) || null_line(l2) || + !finite_coords4(l1) || !finite_coords4(l2)) + return NoIntersections; + + type = do_lines_intersect(l1[0], l1[1], l1[2], l1[3], l2[0], l2[1], l2[2], l2[3]) + ? BoundedIntersection : UnboundedIntersection; + + dx_zero = floatsEqual(line_dx(l1) + 1, 1); + ldx_zero = floatsEqual(line_dx(l2) + 1, 1); + + /* one of the lines is vertical */ + if (dx_zero && ldx_zero) { + type = NoIntersections; + } else if (dx_zero) { + VGfloat la = line_dy(l2) / line_dx(l2); + isect[0] = l1[0]; + isect[1] = la * l1[0] + l2[1] - la * l2[0]; + } else if (ldx_zero) { + VGfloat ta = line_dy(l1) / line_dx(l1); + isect[0] = l2[0]; + isect[1] = ta * l2[0] + l1[1] - ta*l1[0]; + } else { + VGfloat x; + VGfloat ta = line_dy(l1) / line_dx(l1); + VGfloat la = line_dy(l2) / line_dx(l2); + if (ta == la) + return NoIntersections; + + x = ( - l2[1] + la * l2[0] + l1[1] - ta * l1[0] ) / (la - ta); + isect[0] = x; + isect[1] = ta*(x - l1[0]) + l1[1]; + } + if (intersection_point) { + intersection_point[0] = isect[0]; + intersection_point[1] = isect[1]; + } + return type; +} + +static INLINE enum line_join_mode stroker_join_mode(struct stroker *s) +{ + switch(s->join_style) { + case VG_JOIN_MITER: + return MiterJoin; + case VG_JOIN_ROUND: + return RoundJoin; + case VG_JOIN_BEVEL: + return FlatJoin; + default: + return FlatJoin; + } +} + +static INLINE enum line_join_mode stroker_cap_mode(struct stroker *s) +{ + switch(s->cap_style) { + case VG_CAP_BUTT: + return FlatJoin; + case VG_CAP_ROUND: + return RoundCap; + case VG_CAP_SQUARE: + return SquareJoin; + default: + return FlatJoin; + } +} + +void stroker_emit_move_to(struct stroker *stroker, VGfloat x, VGfloat y) +{ + VGubyte cmds = VG_MOVE_TO_ABS; + VGfloat coords[2] = {x, y}; +#if DEBUG_EMITS + debug_printf("emit move %f, %f\n", x, y); +#endif + stroker->back2_x = stroker->back1_x; + stroker->back2_y = stroker->back1_y; + stroker->back1_x = x; + stroker->back1_y = y; + + path_append_data(stroker->path, + 1, + &cmds, &coords); +} + +void stroker_emit_line_to(struct stroker *stroker, VGfloat x, VGfloat y) +{ + VGubyte cmds = VG_LINE_TO_ABS; + VGfloat coords[2] = {x, y}; +#if DEBUG_EMITS + debug_printf("emit line %f, %f\n", x, y); +#endif + stroker->back2_x = stroker->back1_x; + stroker->back2_y = stroker->back1_y; + stroker->back1_x = x; + stroker->back1_y = y; + path_append_data(stroker->path, + 1, + &cmds, &coords); +} + +void stroker_emit_curve_to(struct stroker *stroker, VGfloat px1, VGfloat py1, + VGfloat px2, VGfloat py2, + VGfloat x, VGfloat y) +{ + VGubyte cmds = VG_CUBIC_TO_ABS; + VGfloat coords[6] = {px1, py1, px2, py2, x, y}; +#if DEBUG_EMITS + debug_printf("emit curve %f, %f, %f, %f, %f, %f\n", px1, py1, + px2, py2, x, y); +#endif + + if (px2 == x && py2 == y) { + if (px1 == x && py1 == y) { + stroker->back2_x = stroker->back1_x; + stroker->back2_y = stroker->back1_y; + } else { + stroker->back2_x = px1; + stroker->back2_y = py1; + } + } else { + stroker->back2_x = px2; + stroker->back2_y = py2; + } + stroker->back1_x = x; + stroker->back1_y = y; + + path_append_data(stroker->path, + 1, + &cmds, &coords); +} + +static INLINE void create_round_join(struct stroker *stroker, + VGfloat x1, VGfloat y1, + VGfloat x2, VGfloat y2, + VGfloat width, VGfloat height) +{ + struct arc arc; + struct matrix matrix; + + matrix_load_identity(&matrix); + + /*stroker_emit_line_to(stroker, nx, ny);*/ + + arc_init(&arc, VG_SCCWARC_TO_ABS, + x1, y1, x2, y2, width/2, height/2, 0); + arc_stroker_emit(&arc, stroker, &matrix); +} + + +static void create_joins(struct stroker *stroker, + VGfloat focal_x, VGfloat focal_y, + const VGfloat *next_line, enum line_join_mode join) +{ +#if DEBUG_EMITS + debug_printf("create_joins: focal=[%f, %f], next_line=[%f, %f,%f, %f]\n", + focal_x, focal_y, + next_line[0], next_line[1], next_line[2], next_line[3]); +#endif + /* if we're alredy connected do nothing */ + if (floatsEqual(stroker->back1_x, next_line[0]) && + floatsEqual(stroker->back1_y, next_line[1])) + return; + + if (join == FlatJoin) { + stroker_emit_line_to(stroker, next_line[0], next_line[1]); + } else { + VGfloat prev_line[] = {stroker->back2_x, stroker->back2_y, + stroker->back1_x, stroker->back1_y}; + + VGfloat isect[2]; + enum intersection_type type = line_intersect(prev_line, next_line, isect); + + if (join == SquareJoin) { + VGfloat offset = stroker->stroke_width / 2; + VGfloat l1[4] = {prev_line[0], + prev_line[1], + prev_line[2], + prev_line[3]}; + VGfloat l2[4] = {next_line[2], + next_line[3], + next_line[0], + next_line[1]}; + + line_translate(l1, line_dx(l1), line_dy(l1)); + line_set_length(l1, offset); + + line_translate(l2, line_dx(l2), line_dy(l2)); + line_set_length(l2, offset); + + stroker_emit_line_to(stroker, l1[2], l1[3]); + stroker_emit_line_to(stroker, l2[2], l2[3]); + stroker_emit_line_to(stroker, l2[0], l2[1]); + } else if (join == RoundJoin) { + VGfloat offset = stroker->stroke_width / 2; + VGfloat short_cut[4] = {prev_line[2], prev_line[3], + next_line[0], next_line[1]}; + VGfloat angle = line_angles(prev_line, short_cut); + + if (type == BoundedIntersection || + (angle > 90 && !floatsEqual(angle, 90.f))) { + stroker_emit_line_to(stroker, next_line[0], next_line[1]); + return; + } + create_round_join(stroker, prev_line[2], prev_line[3], + next_line[0], next_line[1], + offset * 2, offset * 2); + + stroker_emit_line_to(stroker, next_line[0], next_line[1]); + } else if (join == RoundCap) { + VGfloat offset = stroker->stroke_width / 2; + VGfloat l1[4] = { prev_line[0], prev_line[1], + prev_line[2], prev_line[3] }; + VGfloat l2[4] = {focal_x, focal_y, + prev_line[2], prev_line[3]}; + + line_translate(l1, line_dx(l1), line_dy(l1)); + line_set_length(l1, KAPPA * offset); + + /* normal between prev_line and focal */ + line_translate(l2, -line_dy(l2), line_dx(l2)); + line_set_length(l2, KAPPA * offset); + + stroker_emit_curve_to(stroker, l1[2], l1[3], + l2[2], l2[3], + l2[0], l2[1]); + + l2[0] = l2[0]; + l2[1] = l2[1]; + l2[2] = l2[0] - line_dx(l2); + l2[3] = l2[1] - line_dy(l2); + + line_translate(l1, next_line[0] - l1[0], next_line[1] - l1[1]); + + stroker_emit_curve_to(stroker, + l2[2], l2[3], + l1[2], l1[3], + l1[0], l1[1]); + } else if (join == MiterJoin) { + VGfloat miter_line[4] = {stroker->back1_x, stroker->back1_y, + isect[0], isect[1]}; + VGfloat sl = (stroker->stroke_width * stroker->miter_limit); + VGfloat inside_line[4] = {prev_line[2], prev_line[3], + next_line[0], next_line[1]}; + VGfloat angle = line_angle_to(inside_line, prev_line); + + if (type == BoundedIntersection || + (angle > 90 && !floatsEqual(angle, 90.f))) { + /* + debug_printf("f = %f, nl = %f, pl = %f, is = %f\n", + focal_x, next_line[0], + prev_line[2], isect[0]);*/ + stroker_emit_line_to(stroker, next_line[0], next_line[1]); + return; + } + + if (type == NoIntersections || line_lengthv(miter_line) > sl) { + stroker_emit_line_to(stroker, next_line[0], next_line[1]); + } else { + stroker_emit_line_to(stroker, isect[0], isect[1]); + stroker_emit_line_to(stroker, next_line[0], next_line[1]); + } + } else { + debug_assert(!"create_joins bad join style"); + } + } +} + +static void stroker_add_segment(struct stroker *stroker, + VGPathCommand cmd, + const VGfloat *coords, + VGint num_coords) +{ + /* skip duplicated points */ + if (stroker->segments->num_elements && + stroker->last_cmd == cmd) { + VGfloat *data = stroker->control_points->data; + data += stroker->control_points->num_elements; + data -= num_coords; + switch (cmd) { + case VG_MOVE_TO_ABS: + if (floatsEqual(coords[0], data[0]) && + floatsEqual(coords[1], data[1])) + return; + break; + case VG_LINE_TO_ABS: + if (floatsEqual(coords[0], data[0]) && + floatsEqual(coords[1], data[1])) + return; + break; + case VG_CUBIC_TO_ABS: + if (floatsEqual(coords[0], data[0]) && + floatsEqual(coords[1], data[1]) && + floatsEqual(coords[2], data[2]) && + floatsEqual(coords[3], data[3]) && + floatsEqual(coords[4], data[4]) && + floatsEqual(coords[5], data[5])) + return; + break; + default: + debug_assert(!"Invalid stroke segment"); + } + } else if (stroker->last_cmd == VG_CUBIC_TO_ABS && + cmd == VG_LINE_TO_ABS) { + VGfloat *data = stroker->control_points->data; + data += stroker->control_points->num_elements; + data -= 2; + if (floatsEqual(coords[0], data[0]) && + floatsEqual(coords[1], data[1])) + return; + } + stroker->last_cmd = cmd; + array_append_data(stroker->segments, &cmd, 1); + array_append_data(stroker->control_points, coords, num_coords); +} + +void stroker_move_to(struct stroker *stroker, VGfloat x, VGfloat y) +{ + VGfloat coords[2] = {x, y}; +#if STROKE_SEGMENTS + debug_printf("stroker_move_to(%f, %f)\n", x, y); +#endif + + if (stroker->segments->num_elements > 0) + stroker->process_subpath(stroker); + + array_reset(stroker->segments); + array_reset(stroker->control_points); + + stroker_add_segment(stroker, VG_MOVE_TO_ABS, coords, 2); +} + +void stroker_line_to(struct stroker *stroker, VGfloat x, VGfloat y) +{ + VGfloat coords[] = {x, y}; + +#if STROKE_SEGMENTS + debug_printf("stroker_line_to(%f, %f)\n", x, y); +#endif + if (!stroker->segments->num_elements) + stroker_add_segment(stroker, VG_MOVE_TO_ABS, zero_coords, 2); + + stroker_add_segment(stroker, VG_LINE_TO_ABS, coords, 2); +} + +void stroker_curve_to(struct stroker *stroker, VGfloat px1, VGfloat py1, + VGfloat px2, VGfloat py2, + VGfloat x, VGfloat y) +{ + VGfloat coords[] = {px1, py1, + px2, py2, + x, y}; +#if STROKE_SEGMENTS + debug_printf("stroker_curve_to(%f, %f, %f, %f, %f, %f)\n", + px1, py1, px2, py2, x, y); +#endif + if (!stroker->segments->num_elements) + stroker_add_segment(stroker, VG_MOVE_TO_ABS, zero_coords, 2); + + stroker_add_segment(stroker, VG_CUBIC_TO_ABS, coords, 6); +} + +static INLINE VGboolean is_segment_null(VGPathCommand cmd, + VGfloat *coords, + VGfloat *res) +{ + switch(cmd) { + case VG_MOVE_TO_ABS: + case VG_LINE_TO_ABS: + return floatsEqual(coords[0], res[0]) && + floatsEqual(coords[1], res[1]); + break; + case VG_CUBIC_TO_ABS: + return floatsEqual(coords[0], res[0]) && + floatsEqual(coords[1], res[1]) && + floatsEqual(coords[2], res[0]) && + floatsEqual(coords[3], res[1]) && + floatsEqual(coords[4], res[0]) && + floatsEqual(coords[5], res[1]); + break; + default: + assert(0); + } + return VG_FALSE; +} + +static VGboolean vg_stroke_outline(struct stroke_iterator *it, + struct stroker *stroker, + VGboolean cap_first, + VGfloat *start_tangent) +{ + const int MAX_OFFSET = 16; + struct bezier offset_curves[MAX_OFFSET]; + VGPathCommand first_element; + VGfloat start[2], prev[2]; + VGboolean first = VG_TRUE; + VGfloat offset; + + first_element = stroke_itr_command(it); + if (first_element != VG_MOVE_TO_ABS) { + stroker_emit_move_to(stroker, 0.f, 0.f); + prev[0] = 0.f; + prev[1] = 0.f; + } + stroke_itr_coords(it, start); +#if STROKE_DEBUG + debug_printf(" -> (side) [%.2f, %.2f]\n", + start[0], + start[1]); +#endif + + prev[0] = start[0]; + prev[1] = start[1]; + + offset = stroker->stroke_width / 2; + + if (!it->has_next(it)) { + /* single point */ + + return VG_TRUE; + } + + while (it->has_next(it)) { + VGPathCommand cmd; + VGfloat coords[8]; + + it->next(it); + cmd = stroke_itr_command(it); + stroke_itr_coords(it, coords); + + if (cmd == VG_LINE_TO_ABS) { + VGfloat line[4] = {prev[0], prev[1], coords[0], coords[1]}; + VGfloat normal[4]; + line_normal(line, normal); + +#if STROKE_DEBUG + debug_printf("\n ---> (side) lineto [%.2f, %.2f]\n", coords[0], coords[1]); +#endif + line_set_length(normal, offset); + line_translate(line, line_dx(normal), line_dy(normal)); + + /* if we are starting a new subpath, move to correct starting point */ + if (first) { + if (cap_first) + create_joins(stroker, prev[0], prev[1], line, + stroker_cap_mode(stroker)); + else + stroker_emit_move_to(stroker, line[0], line[1]); + memcpy(start_tangent, line, + sizeof(VGfloat) * 4); + first = VG_FALSE; + } else { + create_joins(stroker, prev[0], prev[1], line, + stroker_join_mode(stroker)); + } + + /* add the stroke for this line */ + stroker_emit_line_to(stroker, line[2], line[3]); + prev[0] = coords[0]; + prev[1] = coords[1]; + } else if (cmd == VG_CUBIC_TO_ABS) { +#if STROKE_DEBUG + debug_printf("\n ---> (side) cubicTo [%.2f, %.2f]\n", + coords[4], + coords[5]); +#endif + struct bezier bezier; + int count; + + bezier_init(&bezier, + prev[0], prev[1], coords[0], coords[1], + coords[2], coords[3], coords[4], coords[5]); + + count = bezier_translate_by_normal(&bezier, + offset_curves, + MAX_OFFSET, + offset, + curve_threshold); + + if (count) { + /* if we are starting a new subpath, move to correct starting point */ + VGfloat tangent[4]; + VGint i; + + bezier_start_tangent(&bezier, tangent); + line_translate(tangent, + offset_curves[0].x1 - bezier.x1, + offset_curves[0].y1 - bezier.y1); + if (first) { + VGfloat pt[2] = {offset_curves[0].x1, + offset_curves[0].y1}; + + if (cap_first) { + create_joins(stroker, prev[0], prev[1], tangent, + stroker_cap_mode(stroker)); + } else { + stroker_emit_move_to(stroker, pt[0], pt[1]); + } + start_tangent[0] = tangent[0]; + start_tangent[1] = tangent[1]; + start_tangent[2] = tangent[2]; + start_tangent[3] = tangent[3]; + first = VG_FALSE; + } else { + create_joins(stroker, prev[0], prev[1], tangent, + stroker_join_mode(stroker)); + } + + /* add these beziers */ + for (i = 0; i < count; ++i) { + struct bezier *bez = &offset_curves[i]; + stroker_emit_curve_to(stroker, + bez->x2, bez->y2, + bez->x3, bez->y3, + bez->x4, bez->y4); + } + } + + prev[0] = coords[4]; + prev[1] = coords[5]; + } + } + + if (floatsEqual(start[0], prev[0]) && + floatsEqual(start[1], prev[1])) { + /* closed subpath, join first and last point */ +#if STROKE_DEBUG + debug_printf("\n stroker: closed subpath\n"); +#endif + create_joins(stroker, prev[0], prev[1], start_tangent, + stroker_join_mode(stroker)); + return VG_TRUE; + } else { +#if STROKE_DEBUG + debug_printf("\n stroker: open subpath\n"); +#endif + return VG_FALSE; + } +} + +static void stroker_process_subpath(struct stroker *stroker) +{ + VGboolean fwclosed, bwclosed; + VGfloat fw_start_tangent[4], bw_start_tangent[4]; + struct stroke_iterator fwit; + struct stroke_iterator bwit; + debug_assert(stroker->segments->num_elements > 0); + + memset(fw_start_tangent, 0, + sizeof(VGfloat)*4); + memset(bw_start_tangent, 0, + sizeof(VGfloat)*4); + + stroke_forward_iterator(&fwit, stroker->segments, + stroker->control_points); + stroke_backward_iterator(&bwit, stroker->segments, + stroker->control_points); + + debug_assert(fwit.cmds[0] == VG_MOVE_TO_ABS); + + fwclosed = vg_stroke_outline(&fwit, stroker, VG_FALSE, fw_start_tangent); + bwclosed = vg_stroke_outline(&bwit, stroker, !fwclosed, bw_start_tangent); + + if (!bwclosed) + create_joins(stroker, + fwit.coords[0], fwit.coords[1], fw_start_tangent, + stroker_cap_mode(stroker)); + else { + /* hack to handle the requirement of the VG spec that says that strokes + * of len==0 that have butt cap or round cap still need + * to be rendered. (8.7.4 Stroke Generation) */ + if (stroker->segments->num_elements <= 3) { + VGPathCommand cmd; + VGfloat data[8], coords[8]; + struct stroke_iterator *it = &fwit; + + stroke_forward_iterator(it, stroker->segments, + stroker->control_points); + cmd = stroke_itr_command(it); + stroke_itr_coords(it, coords); + if (cmd != VG_MOVE_TO_ABS) { + memset(data, 0, sizeof(VGfloat) * 8); + if (!is_segment_null(cmd, coords, data)) + return; + } else { + data[0] = coords[0]; + data[1] = coords[1]; + } + while (it->has_next(it)) { + it->next(it); + cmd = stroke_itr_command(it); + stroke_itr_coords(it, coords); + if (!is_segment_null(cmd, coords, data)) + return; + } + /* generate the square/round cap */ + if (stroker->cap_style == VG_CAP_SQUARE) { + VGfloat offset = stroker->stroke_width / 2; + stroker_emit_move_to(stroker, data[0] - offset, + data[1] - offset); + stroker_emit_line_to(stroker, data[0] + offset, + data[1] - offset); + stroker_emit_line_to(stroker, data[0] + offset, + data[1] + offset); + stroker_emit_line_to(stroker, data[0] - offset, + data[1] + offset); + stroker_emit_line_to(stroker, data[0] - offset, + data[1] - offset); + } else if (stroker->cap_style == VG_CAP_ROUND) { + VGfloat offset = stroker->stroke_width / 2; + VGfloat cx = data[0], cy = data[1]; + { /*circle */ + struct arc arc; + struct matrix matrix; + matrix_load_identity(&matrix); + + stroker_emit_move_to(stroker, cx + offset, cy); + arc_init(&arc, VG_SCCWARC_TO_ABS, + cx + offset, cy, + cx - offset, cy, + offset, offset, 0); + arc_stroker_emit(&arc, stroker, &matrix); + arc_init(&arc, VG_SCCWARC_TO_ABS, + cx - offset, cy, + cx + offset, cy, + offset, offset, 0); + arc_stroker_emit(&arc, stroker, &matrix); + } + } + } + } +} + +static INLINE VGfloat dash_pattern(struct dash_stroker *stroker, + VGint idx) +{ + if (stroker->dash_pattern[idx] < 0) + return 0.f; + return stroker->dash_pattern[idx]; +} + +static void dash_stroker_process_subpath(struct stroker *str) +{ + struct dash_stroker *stroker = (struct dash_stroker *)str; + VGfloat sum_length = 0; + VGint i; + VGint idash = 0; + VGfloat pos = 0; + VGfloat elen = 0; + VGfloat doffset = stroker->dash_phase; + VGfloat estart = 0; + VGfloat estop = 0; + VGfloat cline[4]; + struct stroke_iterator it; + VGfloat prev[2]; + VGfloat move_to_pos[2]; + VGfloat line_to_pos[2]; + + VGboolean has_move_to = VG_FALSE; + + stroke_flat_iterator(&it, stroker->base.segments, + stroker->base.control_points); + + stroke_itr_coords(&it, prev); + move_to_pos[0] = prev[0]; + move_to_pos[1] = prev[1]; + + debug_assert(stroker->dash_pattern_num > 0); + + for (i = 0; i < stroker->dash_pattern_num; ++i) { + sum_length += dash_pattern(stroker, i); + } + + if (floatIsZero(sum_length)) { + return; + } + + doffset -= floorf(doffset / sum_length) * sum_length; + + while (!floatIsZero(doffset) && doffset >= dash_pattern(stroker, idash)) { + doffset -= dash_pattern(stroker, idash); + idash = (idash + 1) % stroker->dash_pattern_num; + } + + while (it.has_next(&it)) { + VGPathCommand cmd; + VGfloat coords[8]; + VGboolean done; + + it.next(&it); + cmd = stroke_itr_command(&it); + stroke_itr_coords(&it, coords); + + debug_assert(cmd == VG_LINE_TO_ABS); + cline[0] = prev[0]; + cline[1] = prev[1]; + cline[2] = coords[0]; + cline[3] = coords[1]; + + elen = line_lengthv(cline); + + estop = estart + elen; + + done = pos >= estop; + while (!done) { + VGfloat p2[2]; + + VGint idash_incr = 0; + VGboolean has_offset = doffset > 0; + VGfloat dpos = pos + dash_pattern(stroker, idash) - doffset - estart; + + debug_assert(dpos >= 0); + + if (dpos > elen) { /* dash extends this line */ + doffset = dash_pattern(stroker, idash) - (dpos - elen); + pos = estop; + done = VG_TRUE; + p2[0] = cline[2]; + p2[1] = cline[3]; + } else { /* Dash is on this line */ + line_point_at(cline, dpos/elen, p2); + pos = dpos + estart; + done = pos >= estop; + idash_incr = 1; + doffset = 0; + } + + if (idash % 2 == 0) { + line_to_pos[0] = p2[0]; + line_to_pos[1] = p2[1]; + + if (!has_offset || !has_move_to) { + stroker_move_to(&stroker->stroker, move_to_pos[0], move_to_pos[1]); + has_move_to = VG_TRUE; + } + stroker_line_to(&stroker->stroker, line_to_pos[0], line_to_pos[1]); + } else { + move_to_pos[0] = p2[0]; + move_to_pos[1] = p2[1]; + } + + idash = (idash + idash_incr) % stroker->dash_pattern_num; + } + + estart = estop; + prev[0] = coords[0]; + prev[1] = coords[1]; + } + + if (it.curve_poly) { + polygon_destroy(it.curve_poly); + it.curve_poly = 0; + } + + stroker->base.path = stroker->stroker.path; +} + +static void default_begin(struct stroker *stroker) +{ + array_reset(stroker->segments); + array_reset(stroker->control_points); +} + +static void default_end(struct stroker *stroker) +{ + if (stroker->segments->num_elements > 0) + stroker->process_subpath(stroker); +} + + +static void dash_stroker_begin(struct stroker *stroker) +{ + struct dash_stroker *dasher = + (struct dash_stroker *)stroker; + + default_begin(&dasher->stroker); + default_begin(stroker); +} + +static void dash_stroker_end(struct stroker *stroker) +{ + struct dash_stroker *dasher = + (struct dash_stroker *)stroker; + + default_end(stroker); + default_end(&dasher->stroker); +} + +void stroker_init(struct stroker *stroker, + struct vg_state *state) +{ + stroker->stroke_width = state->stroke.line_width.f; + stroker->miter_limit = state->stroke.miter_limit.f; + stroker->cap_style = state->stroke.cap_style; + stroker->join_style = state->stroke.join_style; + + stroker->begin = default_begin; + stroker->process_subpath = stroker_process_subpath; + stroker->end = default_end; + + stroker->segments = array_create(sizeof(VGubyte)); + stroker->control_points = array_create(sizeof(VGfloat)); + + stroker->back1_x = 0; + stroker->back1_y = 0; + stroker->back2_x = 0; + stroker->back2_y = 0; + + stroker->path = path_create(VG_PATH_DATATYPE_F, 1.0f, 0.0f, + 0, 0, VG_PATH_CAPABILITY_ALL); + + stroker->last_cmd = VG_CLOSE_PATH; +} + +void dash_stroker_init(struct stroker *str, + struct vg_state *state) +{ + struct dash_stroker *stroker = (struct dash_stroker *)str; + int i; + + stroker_init(str, state); + stroker_init(&stroker->stroker, state); + + { + int real_num = state->stroke.dash_pattern_num; + if (real_num % 2)/* if odd, ignore the last one */ + --real_num; + for (i = 0; i < real_num; ++i) + stroker->dash_pattern[i] = state->stroke.dash_pattern[i].f; + stroker->dash_pattern_num = real_num; + } + + stroker->dash_phase = state->stroke.dash_phase.f; + stroker->dash_phase_reset = state->stroke.dash_phase_reset; + + stroker->base.begin = dash_stroker_begin; + stroker->base.process_subpath = dash_stroker_process_subpath; + stroker->base.end = dash_stroker_end; + path_destroy(stroker->base.path); + stroker->base.path = NULL; +} + +void stroker_begin(struct stroker *stroker) +{ + stroker->begin(stroker); +} + +void stroker_end(struct stroker *stroker) +{ + stroker->end(stroker); +} + +void stroker_cleanup(struct stroker *stroker) +{ + array_destroy(stroker->segments); + array_destroy(stroker->control_points); +} + +void dash_stroker_cleanup(struct dash_stroker *stroker) +{ + /* if stroker->base.path is null means we never + * processed a valid path so delete the temp one + * we already created */ + if (!stroker->base.path) + path_destroy(stroker->stroker.path); + stroker_cleanup(&stroker->stroker); + stroker_cleanup((struct stroker*)stroker); +} diff --git a/src/gallium/state_trackers/vega/stroker.h b/src/gallium/state_trackers/vega/stroker.h new file mode 100644 index 00000000000..36543cd9237 --- /dev/null +++ b/src/gallium/state_trackers/vega/stroker.h @@ -0,0 +1,89 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef STROKER_H +#define STROKER_H + +#include "VG/openvg.h" +#include "api_consts.h" + +struct path; +struct vg_state; +struct array; + +struct stroker { + void (*begin)(struct stroker *stroker); + void (*process_subpath)(struct stroker *stroker); + void (*end)(struct stroker *stroker); + + struct array *segments; + struct array *control_points; + struct path *path; + + VGfloat back1_x, back1_y; + VGfloat back2_x, back2_y; + + VGfloat stroke_width; + VGfloat miter_limit; + VGCapStyle cap_style; + VGJoinStyle join_style; + + VGPathCommand last_cmd; +}; + +struct dash_stroker { + struct stroker base; + + struct stroker stroker; + + VGfloat dash_pattern[VEGA_MAX_DASH_COUNT]; + VGint dash_pattern_num; + VGfloat dash_phase; + VGboolean dash_phase_reset; +}; + +void stroker_init(struct stroker *stroker, + struct vg_state *state); +void dash_stroker_init(struct stroker *stroker, + struct vg_state *state); +void dash_stroker_cleanup(struct dash_stroker *stroker); +void stroker_cleanup(struct stroker *stroker); + +void stroker_begin(struct stroker *stroker); +void stroker_move_to(struct stroker *stroker, VGfloat x, VGfloat y); +void stroker_line_to(struct stroker *stroker, VGfloat x, VGfloat y); +void stroker_curve_to(struct stroker *stroker, VGfloat px1, VGfloat py1, + VGfloat px2, VGfloat py2, + VGfloat x, VGfloat y); +void stroker_end(struct stroker *stroker); + +void stroker_emit_move_to(struct stroker *stroker, VGfloat x, VGfloat y); +void stroker_emit_line_to(struct stroker *stroker, VGfloat x, VGfloat y); +void stroker_emit_curve_to(struct stroker *stroker, VGfloat px1, VGfloat py1, + VGfloat px2, VGfloat py2, + VGfloat x, VGfloat y); + +#endif diff --git a/src/gallium/state_trackers/vega/util_array.h b/src/gallium/state_trackers/vega/util_array.h new file mode 100644 index 00000000000..4343dfe36e2 --- /dev/null +++ b/src/gallium/state_trackers/vega/util_array.h @@ -0,0 +1,122 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef UTIL_ARRAY_H +#define UTIL_ARRAY_H + +#include "util/u_memory.h" + +#define DEFAULT_ARRAY_SIZE 256 + +struct array { + VGint datatype_size; + void *data; + VGint size; + VGint num_elements; +}; + +static INLINE struct array *array_create(VGint datatype_size) +{ + struct array *array = CALLOC_STRUCT(array); + array->datatype_size = datatype_size; + + array->size = DEFAULT_ARRAY_SIZE; + array->data = malloc(array->size * array->datatype_size); + + return array; +} + + +static INLINE struct array *array_create_size(VGint datatype_size, VGint size) +{ + struct array *array = CALLOC_STRUCT(array); + array->datatype_size = datatype_size; + + array->size = size; + array->data = malloc(array->size * array->datatype_size); + + return array; +} + +static INLINE void array_destroy(struct array *array) +{ + if (array) + free(array->data); + free(array); +} + +static INLINE void array_resize(struct array *array, int num) +{ + VGint size = array->datatype_size * num; + void *data = malloc(size); + memcpy(data, array->data, array->size * array->datatype_size); + free(array->data); + array->data = data; + array->size = num; + array->num_elements = (array->num_elements > num) ? num : + array->num_elements; +} + +static INLINE void array_append_data(struct array *array, + const void *data, int num_elements) +{ + VGbyte *adata; + + while (array->num_elements + num_elements > array->size) { + array_resize(array, (array->num_elements + num_elements) * 1.5); + } + adata = (VGbyte *)array->data; + memcpy(adata + (array->num_elements * array->datatype_size), data, + num_elements * array->datatype_size); + array->num_elements += num_elements; +} + +static INLINE void array_change_data(struct array *array, + const void *data, + int start_idx, + int num_elements) +{ + VGbyte *adata = (VGbyte *)array->data; + memcpy(adata + (start_idx * array->datatype_size), data, + num_elements * array->datatype_size); +} + +static INLINE void array_remove_element(struct array *array, + int idx) +{ + VGbyte *adata = (VGbyte *)array->data; + memmove(adata + (idx * array->datatype_size), + adata + ((idx + 1) * array->datatype_size), + (array->num_elements - idx - 1) * array->datatype_size); + --array->num_elements; +} + +static INLINE void array_reset(struct array *array) +{ + array->num_elements = 0; +} + +#endif diff --git a/src/gallium/state_trackers/vega/vg_context.c b/src/gallium/state_trackers/vega/vg_context.c new file mode 100644 index 00000000000..e0ff02f3a99 --- /dev/null +++ b/src/gallium/state_trackers/vega/vg_context.c @@ -0,0 +1,543 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "vg_context.h" + +#include "paint.h" +#include "renderer.h" +#include "shaders_cache.h" +#include "shader.h" +#include "asm_util.h" +#include "st_inlines.h" + +#include "pipe/p_context.h" +#include "pipe/p_inlines.h" +#include "pipe/p_shader_tokens.h" + +#include "cso_cache/cso_context.h" + +#include "util/u_simple_shaders.h" +#include "util/u_memory.h" +#include "util/u_blit.h" + +struct vg_context *_vg_context = 0; + +struct vg_context * vg_current_context(void) +{ + return _vg_context; +} + +static void init_clear(struct vg_context *st) +{ + struct pipe_context *pipe = st->pipe; + + /* rasterizer state: bypass clipping */ + memset(&st->clear.raster, 0, sizeof(st->clear.raster)); + st->clear.raster.gl_rasterization_rules = 1; + + /* fragment shader state: color pass-through program */ + st->clear.fs = + util_make_fragment_passthrough_shader(pipe); +} +void vg_set_current_context(struct vg_context *ctx) +{ + _vg_context = ctx; +} + +struct vg_context * vg_create_context(struct pipe_context *pipe, + const void *visual, + struct vg_context *share) +{ + struct vg_context *ctx; + + ctx = CALLOC_STRUCT(vg_context); + + ctx->pipe = pipe; + + vg_init_state(&ctx->state.vg); + ctx->state.dirty = ALL_DIRTY; + + ctx->cso_context = cso_create_context(pipe); + + init_clear(ctx); + + ctx->default_paint = paint_create(ctx); + ctx->state.vg.stroke_paint = ctx->default_paint; + ctx->state.vg.fill_paint = ctx->default_paint; + + + ctx->mask.sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + ctx->mask.sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + ctx->mask.sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE; + ctx->mask.sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST; + ctx->mask.sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST; + ctx->mask.sampler.normalized_coords = 0; + + ctx->blend_sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + ctx->blend_sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + ctx->blend_sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE; + ctx->blend_sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST; + ctx->blend_sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST; + ctx->blend_sampler.normalized_coords = 0; + + vg_set_error(ctx, VG_NO_ERROR); + + ctx->owned_objects[VG_OBJECT_PAINT] = cso_hash_create(); + ctx->owned_objects[VG_OBJECT_IMAGE] = cso_hash_create(); + ctx->owned_objects[VG_OBJECT_MASK] = cso_hash_create(); + ctx->owned_objects[VG_OBJECT_FONT] = cso_hash_create(); + ctx->owned_objects[VG_OBJECT_PATH] = cso_hash_create(); + + ctx->renderer = renderer_create(ctx); + ctx->sc = shaders_cache_create(ctx); + ctx->shader = shader_create(ctx); + + ctx->blit = util_create_blit(ctx->pipe, ctx->cso_context); + + return ctx; +} + +void vg_destroy_context(struct vg_context *ctx) +{ + struct pipe_constant_buffer *cbuf = &ctx->mask.cbuf; + struct pipe_constant_buffer *vsbuf = &ctx->vs_const_buffer; + + util_destroy_blit(ctx->blit); + renderer_destroy(ctx->renderer); + shaders_cache_destroy(ctx->sc); + shader_destroy(ctx->shader); + paint_destroy(ctx->default_paint); + + if (cbuf && cbuf->buffer) + pipe_buffer_reference(&cbuf->buffer, NULL); + + if (vsbuf && vsbuf->buffer) + pipe_buffer_reference(&vsbuf->buffer, NULL); + + if (ctx->clear.fs) { + cso_delete_fragment_shader(ctx->cso_context, ctx->clear.fs); + ctx->clear.fs = NULL; + } + + if (ctx->plain_vs) { + vg_shader_destroy(ctx, ctx->plain_vs); + ctx->plain_vs = NULL; + } + if (ctx->clear_vs) { + vg_shader_destroy(ctx, ctx->clear_vs); + ctx->clear_vs = NULL; + } + if (ctx->texture_vs) { + vg_shader_destroy(ctx, ctx->texture_vs); + ctx->texture_vs = NULL; + } + + if (ctx->pass_through_depth_fs) + vg_shader_destroy(ctx, ctx->pass_through_depth_fs); + if (ctx->mask.union_fs) + vg_shader_destroy(ctx, ctx->mask.union_fs); + if (ctx->mask.intersect_fs) + vg_shader_destroy(ctx, ctx->mask.intersect_fs); + if (ctx->mask.subtract_fs) + vg_shader_destroy(ctx, ctx->mask.subtract_fs); + if (ctx->mask.set_fs) + vg_shader_destroy(ctx, ctx->mask.set_fs); + + cso_release_all(ctx->cso_context); + cso_destroy_context(ctx->cso_context); + + cso_hash_delete(ctx->owned_objects[VG_OBJECT_PAINT]); + cso_hash_delete(ctx->owned_objects[VG_OBJECT_IMAGE]); + cso_hash_delete(ctx->owned_objects[VG_OBJECT_MASK]); + cso_hash_delete(ctx->owned_objects[VG_OBJECT_FONT]); + cso_hash_delete(ctx->owned_objects[VG_OBJECT_PATH]); + + free(ctx); +} + +void vg_init_object(struct vg_object *obj, struct vg_context *ctx, enum vg_object_type type) +{ + obj->type = type; + obj->ctx = ctx; +} + +VGboolean vg_context_is_object_valid(struct vg_context *ctx, + enum vg_object_type type, + void *ptr) +{ + if (ctx) { + struct cso_hash *hash = ctx->owned_objects[type]; + if (!hash) + return VG_FALSE; + return cso_hash_contains(hash, (unsigned)(long)ptr); + } + return VG_FALSE; +} + +void vg_context_add_object(struct vg_context *ctx, + enum vg_object_type type, + void *ptr) +{ + if (ctx) { + struct cso_hash *hash = ctx->owned_objects[type]; + if (!hash) + return; + cso_hash_insert(hash, (unsigned)(long)ptr, ptr); + } +} + +void vg_context_remove_object(struct vg_context *ctx, + enum vg_object_type type, + void *ptr) +{ + if (ctx) { + struct cso_hash *hash = ctx->owned_objects[type]; + if (!hash) + return; + cso_hash_take(hash, (unsigned)(long)ptr); + } +} + +static void update_clip_state(struct vg_context *ctx) +{ + struct pipe_depth_stencil_alpha_state *dsa = &ctx->state.g3d.dsa; + struct vg_state *state = &ctx->state.vg; + + memset(dsa, 0, sizeof(struct pipe_depth_stencil_alpha_state)); + + if (state->scissoring) { + struct pipe_blend_state *blend = &ctx->state.g3d.blend; + struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb; + dsa->depth.writemask = 1;/*glDepthMask(TRUE);*/ + dsa->depth.func = PIPE_FUNC_ALWAYS; + dsa->depth.enabled = 1; + + cso_save_blend(ctx->cso_context); + cso_save_fragment_shader(ctx->cso_context); + /* set a passthrough shader */ + if (!ctx->pass_through_depth_fs) + ctx->pass_through_depth_fs = shader_create_from_text(ctx->pipe, + pass_through_depth_asm, + 40, + PIPE_SHADER_FRAGMENT); + cso_set_fragment_shader_handle(ctx->cso_context, + ctx->pass_through_depth_fs->driver); + cso_set_depth_stencil_alpha(ctx->cso_context, dsa); + + ctx->pipe->clear(ctx->pipe, PIPE_CLEAR_DEPTHSTENCIL, NULL, 1.0, 0); + + /* disable color writes */ + blend->colormask = 0; /*disable colorwrites*/ + cso_set_blend(ctx->cso_context, blend); + + /* enable scissoring */ + for (int i = 0; i < state->scissor_rects_num; ++i) { + const float x = state->scissor_rects[i * 4 + 0].f; + const float y = state->scissor_rects[i * 4 + 1].f; + const float width = state->scissor_rects[i * 4 + 2].f; + const float height = state->scissor_rects[i * 4 + 3].f; + VGfloat minx, miny, maxx, maxy; + + minx = 0; + miny = 0; + maxx = fb->width; + maxy = fb->height; + + if (x > minx) + minx = x; + if (y > miny) + miny = y; + + if (x + width < maxx) + maxx = x + width; + if (y + height < maxy) + maxy = y + height; + + /* check for null space */ + if (minx >= maxx || miny >= maxy) + minx = miny = maxx = maxy = 0; + + /*glClear(GL_DEPTH_BUFFER_BIT);*/ + renderer_draw_quad(ctx->renderer, minx, miny, maxx, maxy, 0.0f); + } + + blend->colormask = 1; /*enable colorwrites*/ + cso_restore_blend(ctx->cso_context); + cso_restore_fragment_shader(ctx->cso_context); + + dsa->depth.enabled = 1; /* glEnable(GL_DEPTH_TEST); */ + dsa->depth.writemask = 0;/*glDepthMask(FALSE);*/ + dsa->depth.func = PIPE_FUNC_GEQUAL; + } +} + +void vg_validate_state(struct vg_context *ctx) +{ + if ((ctx->state.dirty & BLEND_DIRTY)) { + struct pipe_blend_state *blend = &ctx->state.g3d.blend; + memset(blend, 0, sizeof(struct pipe_blend_state)); + blend->blend_enable = 1; + blend->colormask |= PIPE_MASK_R; + blend->colormask |= PIPE_MASK_G; + blend->colormask |= PIPE_MASK_B; + blend->colormask |= PIPE_MASK_A; + + switch (ctx->state.vg.blend_mode) { + case VG_BLEND_SRC: + blend->rgb_src_factor = PIPE_BLENDFACTOR_ONE; + blend->alpha_src_factor = PIPE_BLENDFACTOR_ONE; + blend->rgb_dst_factor = PIPE_BLENDFACTOR_ZERO; + blend->alpha_dst_factor = PIPE_BLENDFACTOR_ZERO; + break; + case VG_BLEND_SRC_OVER: + blend->rgb_src_factor = PIPE_BLENDFACTOR_SRC_ALPHA; + blend->alpha_src_factor = PIPE_BLENDFACTOR_ONE; + blend->rgb_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA; + blend->alpha_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA; + break; + case VG_BLEND_DST_OVER: + blend->rgb_src_factor = PIPE_BLENDFACTOR_INV_DST_ALPHA; + blend->alpha_src_factor = PIPE_BLENDFACTOR_INV_DST_ALPHA; + blend->rgb_dst_factor = PIPE_BLENDFACTOR_DST_ALPHA; + blend->alpha_dst_factor = PIPE_BLENDFACTOR_DST_ALPHA; + break; + case VG_BLEND_SRC_IN: + blend->rgb_src_factor = PIPE_BLENDFACTOR_DST_ALPHA; + blend->alpha_src_factor = PIPE_BLENDFACTOR_DST_ALPHA; + blend->rgb_dst_factor = PIPE_BLENDFACTOR_ZERO; + blend->alpha_dst_factor = PIPE_BLENDFACTOR_ZERO; + break; + case VG_BLEND_DST_IN: + blend->rgb_src_factor = PIPE_BLENDFACTOR_ZERO; + blend->alpha_src_factor = PIPE_BLENDFACTOR_ZERO; + blend->rgb_dst_factor = PIPE_BLENDFACTOR_SRC_ALPHA; + blend->alpha_dst_factor = PIPE_BLENDFACTOR_SRC_ALPHA; + break; + case VG_BLEND_MULTIPLY: + case VG_BLEND_SCREEN: + case VG_BLEND_DARKEN: + case VG_BLEND_LIGHTEN: + blend->rgb_src_factor = PIPE_BLENDFACTOR_ONE; + blend->alpha_src_factor = PIPE_BLENDFACTOR_ONE; + blend->rgb_dst_factor = PIPE_BLENDFACTOR_ZERO; + blend->alpha_dst_factor = PIPE_BLENDFACTOR_ZERO; + break; + case VG_BLEND_ADDITIVE: + blend->rgb_src_factor = PIPE_BLENDFACTOR_ONE; + blend->alpha_src_factor = PIPE_BLENDFACTOR_ONE; + blend->rgb_dst_factor = PIPE_BLENDFACTOR_ONE; + blend->alpha_dst_factor = PIPE_BLENDFACTOR_ONE; + break; + default: + assert(!"not implemented blend mode"); + } + cso_set_blend(ctx->cso_context, &ctx->state.g3d.blend); + } + if ((ctx->state.dirty & RASTERIZER_DIRTY)) { + struct pipe_rasterizer_state *raster = &ctx->state.g3d.rasterizer; + memset(raster, 0, sizeof(struct pipe_rasterizer_state)); + raster->gl_rasterization_rules = 1; + cso_set_rasterizer(ctx->cso_context, &ctx->state.g3d.rasterizer); + } + if ((ctx->state.dirty & VIEWPORT_DIRTY)) { + struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb; + const VGint param_bytes = 8 * sizeof(VGfloat); + VGfloat vs_consts[8] = { + 2.f/fb->width, 2.f/fb->height, 1, 1, + -1, -1, 0, 0 + }; + struct pipe_constant_buffer *cbuf = &ctx->vs_const_buffer; + + vg_set_viewport(ctx, VEGA_Y0_BOTTOM); + + pipe_buffer_reference(&cbuf->buffer, NULL); + cbuf->buffer = pipe_buffer_create(ctx->pipe->screen, 16, + PIPE_BUFFER_USAGE_CONSTANT, + param_bytes); + + if (cbuf->buffer) { + st_no_flush_pipe_buffer_write(ctx, cbuf->buffer, + 0, param_bytes, vs_consts); + } + ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_VERTEX, 0, cbuf); + } + if ((ctx->state.dirty & VS_DIRTY)) { + cso_set_vertex_shader_handle(ctx->cso_context, + vg_plain_vs(ctx)); + } + + /* must be last because it renders to the depth buffer*/ + if ((ctx->state.dirty & DEPTH_STENCIL_DIRTY)) { + update_clip_state(ctx); + cso_set_depth_stencil_alpha(ctx->cso_context, &ctx->state.g3d.dsa); + } + + shader_set_masking(ctx->shader, ctx->state.vg.masking); + shader_set_image_mode(ctx->shader, ctx->state.vg.image_mode); + + ctx->state.dirty = NONE_DIRTY; +} + +VGboolean vg_object_is_valid(void *ptr, enum vg_object_type type) +{ + struct vg_object *obj = ptr; + if (ptr && is_aligned(obj) && obj->type == type) + return VG_TRUE; + else + return VG_FALSE; +} + +void vg_set_error(struct vg_context *ctx, + VGErrorCode code) +{ + /*vgGetError returns the oldest error code provided by + * an API call on the current context since the previous + * call to vgGetError on that context (or since the creation + of the context).*/ + if (ctx->_error == VG_NO_ERROR) + ctx->_error = code; +} + +void vg_prepare_blend_surface(struct vg_context *ctx) +{ + struct pipe_surface *dest_surface = NULL; + struct pipe_context *pipe = ctx->pipe; + struct st_framebuffer *stfb = ctx->draw_buffer; + struct st_renderbuffer *strb = stfb->strb; + + /* first finish all pending rendering */ + vgFinish(); + + dest_surface = pipe->screen->get_tex_surface(pipe->screen, + stfb->blend_texture, + 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_WRITE); + /* flip it, because we want to use it as a sampler */ + util_blit_pixels_tex(ctx->blit, + strb->texture, + 0, strb->height, + strb->width, 0, + dest_surface, + 0, 0, + strb->width, strb->height, + 0.0, PIPE_TEX_MIPFILTER_NEAREST); + + if (dest_surface) + pipe_surface_reference(&dest_surface, NULL); + + /* make sure it's complete */ + vgFinish(); +} + + +void vg_prepare_blend_surface_from_mask(struct vg_context *ctx) +{ + struct pipe_surface *dest_surface = NULL; + struct pipe_context *pipe = ctx->pipe; + struct st_framebuffer *stfb = ctx->draw_buffer; + struct st_renderbuffer *strb = stfb->strb; + + vg_validate_state(ctx); + + /* first finish all pending rendering */ + vgFinish(); + + dest_surface = pipe->screen->get_tex_surface(pipe->screen, + stfb->blend_texture, + 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_WRITE); + + /* flip it, because we want to use it as a sampler */ + util_blit_pixels_tex(ctx->blit, + stfb->alpha_mask, + 0, strb->height, + strb->width, 0, + dest_surface, + 0, 0, + strb->width, strb->height, + 0.0, PIPE_TEX_MIPFILTER_NEAREST); + + /* make sure it's complete */ + vgFinish(); + + if (dest_surface) + pipe_surface_reference(&dest_surface, NULL); +} + +void * vg_plain_vs(struct vg_context *ctx) +{ + if (!ctx->plain_vs) { + ctx->plain_vs = shader_create_from_text(ctx->pipe, + vs_plain_asm, + 200, + PIPE_SHADER_VERTEX); + } + + return ctx->plain_vs->driver; +} + + +void * vg_clear_vs(struct vg_context *ctx) +{ + if (!ctx->clear_vs) { + ctx->clear_vs = shader_create_from_text(ctx->pipe, + vs_clear_asm, + 200, + PIPE_SHADER_VERTEX); + } + + return ctx->clear_vs->driver; +} + +void * vg_texture_vs(struct vg_context *ctx) +{ + if (!ctx->texture_vs) { + ctx->texture_vs = shader_create_from_text(ctx->pipe, + vs_texture_asm, + 200, + PIPE_SHADER_VERTEX); + } + + return ctx->texture_vs->driver; +} + +void vg_set_viewport(struct vg_context *ctx, VegaOrientation orientation) +{ + struct pipe_viewport_state viewport; + struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb; + VGfloat y_scale = (orientation == VEGA_Y0_BOTTOM) ? -2.f : 2.f; + + viewport.scale[0] = fb->width / 2.f; + viewport.scale[1] = fb->height / y_scale; + viewport.scale[2] = 1.0; + viewport.scale[3] = 1.0; + viewport.translate[0] = fb->width / 2.f; + viewport.translate[1] = fb->height / 2.f; + viewport.translate[2] = 0.0; + viewport.translate[3] = 0.0; + + cso_set_viewport(ctx->cso_context, &viewport); +} diff --git a/src/gallium/state_trackers/vega/vg_context.h b/src/gallium/state_trackers/vega/vg_context.h new file mode 100644 index 00000000000..ccc8889c8c5 --- /dev/null +++ b/src/gallium/state_trackers/vega/vg_context.h @@ -0,0 +1,292 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef VG_CONTEXT_H +#define VG_CONTEXT_H + +#include "vg_state.h" + +#include "pipe/p_format.h" +#include "pipe/p_state.h" +#include "util/u_pointer.h" +#include "util/u_math.h" + +#include "cso_cache/cso_hash.h" +#include "cso_cache/cso_context.h" + +struct renderer; +struct shaders_cache; +struct shader; +struct vg_shader; + +struct st_renderbuffer { + enum pipe_format format; + struct pipe_surface *surface; + struct pipe_texture *texture; + VGint width, height; +}; + +struct st_framebuffer { + VGint init_width, init_height; + struct st_renderbuffer *strb; + struct st_renderbuffer *dsrb; + + struct pipe_texture *alpha_mask; + + struct pipe_texture *blend_texture; + + void *privateData; +}; + +enum vg_object_type { + VG_OBJECT_UNKNOWN = 0, + VG_OBJECT_PAINT, + VG_OBJECT_IMAGE, + VG_OBJECT_MASK, + VG_OBJECT_FONT, + VG_OBJECT_PATH, + + VG_OBJECT_LAST +}; +enum dirty_state { + NONE_DIRTY = 0<<0, + BLEND_DIRTY = 1<<1, + RASTERIZER_DIRTY = 1<<2, + VIEWPORT_DIRTY = 1<<3, + VS_DIRTY = 1<<4, + DEPTH_STENCIL_DIRTY = 1<<5, + ALL_DIRTY = BLEND_DIRTY | RASTERIZER_DIRTY | + VIEWPORT_DIRTY | VS_DIRTY | DEPTH_STENCIL_DIRTY +}; + +struct vg_context +{ + struct pipe_context *pipe; + + struct { + struct vg_state vg; + struct { + struct pipe_blend_state blend; + struct pipe_rasterizer_state rasterizer; + struct pipe_shader_state vs_state; + struct pipe_depth_stencil_alpha_state dsa; + struct pipe_framebuffer_state fb; + } g3d; + VGbitfield dirty; + } state; + + VGErrorCode _error; + + struct st_framebuffer *draw_buffer; + + struct cso_hash *owned_objects[VG_OBJECT_LAST]; + + struct { + struct pipe_shader_state vert_shader; + struct pipe_shader_state frag_shader; + struct pipe_rasterizer_state raster; + void *fs; + float vertices[4][2][4]; /**< vertex pos + color */ + } clear; + + struct { + struct pipe_constant_buffer cbuf; + struct pipe_sampler_state sampler; + + struct vg_shader *union_fs; + struct vg_shader *intersect_fs; + struct vg_shader *subtract_fs; + struct vg_shader *set_fs; + } mask; + + struct vg_shader *pass_through_depth_fs; + + struct cso_context *cso_context; + + struct pipe_buffer *stencil_quad; + VGfloat stencil_vertices[4][2][4]; + + struct renderer *renderer; + struct shaders_cache *sc; + struct shader *shader; + + struct pipe_sampler_state blend_sampler; + struct { + struct pipe_constant_buffer buffer; + void *color_matrix_fs; + } filter; + struct vg_paint *default_paint; + + struct blit_state *blit; + + struct vg_shader *plain_vs; + struct vg_shader *clear_vs; + struct vg_shader *texture_vs; + struct pipe_constant_buffer vs_const_buffer; +}; + +struct vg_object { + enum vg_object_type type; + struct vg_context *ctx; +}; +void vg_init_object(struct vg_object *obj, struct vg_context *ctx, enum vg_object_type type); +VGboolean vg_object_is_valid(void *ptr, enum vg_object_type type); + +struct vg_context *vg_create_context(struct pipe_context *pipe, + const void *visual, + struct vg_context *share); +void vg_destroy_context(struct vg_context *ctx); +struct vg_context *vg_current_context(void); +void vg_set_current_context(struct vg_context *ctx); + +VGboolean vg_context_is_object_valid(struct vg_context *ctx, + enum vg_object_type type, + void *ptr); +void vg_context_add_object(struct vg_context *ctx, + enum vg_object_type type, + void *ptr); +void vg_context_remove_object(struct vg_context *ctx, + enum vg_object_type type, + void *ptr); + +void vg_validate_state(struct vg_context *ctx); + +void vg_set_error(struct vg_context *ctx, + VGErrorCode code); + +void vg_prepare_blend_surface(struct vg_context *ctx); +void vg_prepare_blend_surface_from_mask(struct vg_context *ctx); + + +static INLINE VGboolean is_aligned_to(const void *ptr, VGbyte alignment) +{ + void *aligned = align_pointer(ptr, alignment); + return (ptr == aligned) ? VG_TRUE : VG_FALSE; +} + +static INLINE VGboolean is_aligned(const void *ptr) +{ + return is_aligned_to(ptr, 4); +} + +static INLINE void vg_shift_rectx(VGfloat coords[4], + const VGfloat *bounds, + const VGfloat shift) +{ + coords[0] += shift; + coords[2] -= shift; + if (bounds) { + coords[2] = MIN2(coords[2], bounds[2]); + /* bound x/y + width/height */ + if ((coords[0] + coords[2]) > (bounds[0] + bounds[2])) { + coords[2] = (bounds[0] + bounds[2]) - coords[0]; + } + } +} + +static INLINE void vg_shift_recty(VGfloat coords[4], + const VGfloat *bounds, + const VGfloat shift) +{ + coords[1] += shift; + coords[3] -= shift; + if (bounds) { + coords[3] = MIN2(coords[3], bounds[3]); + if ((coords[1] + coords[3]) > (bounds[1] + bounds[3])) { + coords[3] = (bounds[1] + bounds[3]) - coords[1]; + } + } +} + +static INLINE void vg_bound_rect(VGfloat coords[4], + const VGfloat bounds[4], + VGfloat shift[4]) +{ + /* if outside the bounds */ + if (coords[0] > (bounds[0] + bounds[2]) || + coords[1] > (bounds[1] + bounds[3]) || + (coords[0] + coords[2]) < bounds[0] || + (coords[1] + coords[3]) < bounds[1]) { + coords[0] = 0.f; + coords[1] = 0.f; + coords[2] = 0.f; + coords[3] = 0.f; + shift[0] = 0.f; + shift[1] = 0.f; + return; + } + + /* bound x */ + if (coords[0] < bounds[0]) { + shift[0] = bounds[0] - coords[0]; + coords[2] -= shift[0]; + coords[0] = bounds[0]; + } else + shift[0] = 0.f; + + /* bound y */ + if (coords[1] < bounds[1]) { + shift[1] = bounds[1] - coords[1]; + coords[3] -= shift[1]; + coords[1] = bounds[1]; + } else + shift[1] = 0.f; + + shift[2] = bounds[2] - coords[2]; + shift[3] = bounds[3] - coords[3]; + /* bound width/height */ + coords[2] = MIN2(coords[2], bounds[2]); + coords[3] = MIN2(coords[3], bounds[3]); + + /* bound x/y + width/height */ + if ((coords[0] + coords[2]) > (bounds[0] + bounds[2])) { + coords[2] = (bounds[0] + bounds[2]) - coords[0]; + } + if ((coords[1] + coords[3]) > (bounds[1] + bounds[3])) { + coords[3] = (bounds[1] + bounds[3]) - coords[1]; + } + + /* if outside the bounds */ + if ((coords[0] + coords[2]) < bounds[0] || + (coords[1] + coords[3]) < bounds[1]) { + coords[0] = 0.f; + coords[1] = 0.f; + coords[2] = 0.f; + coords[3] = 0.f; + return; + } +} + +void *vg_plain_vs(struct vg_context *ctx); +void *vg_clear_vs(struct vg_context *ctx); +void *vg_texture_vs(struct vg_context *ctx); +typedef enum { + VEGA_Y0_TOP, + VEGA_Y0_BOTTOM +} VegaOrientation; +void vg_set_viewport(struct vg_context *ctx, VegaOrientation orientation); + +#endif diff --git a/src/gallium/state_trackers/vega/vg_state.c b/src/gallium/state_trackers/vega/vg_state.c new file mode 100644 index 00000000000..6f6bfdaf7a2 --- /dev/null +++ b/src/gallium/state_trackers/vega/vg_state.c @@ -0,0 +1,124 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "vg_state.h" + +#include <string.h> + +void vg_init_state(struct vg_state *state) +{ + state->matrix_mode = VG_MATRIX_PATH_USER_TO_SURFACE; + state->fill_rule = VG_EVEN_ODD; + state->image_quality = VG_IMAGE_QUALITY_FASTER; + state->rendering_quality = VG_RENDERING_QUALITY_BETTER; + state->blend_mode = VG_BLEND_SRC_OVER; + state->image_mode = VG_DRAW_IMAGE_NORMAL; + + memset(state->scissor_rects, 0, sizeof(state->scissor_rects)); + state->scissor_rects_num = 0; + + state->color_transform = VG_FALSE; + state->color_transform_values[0] = 1.0f; + state->color_transform_values[1] = 1.0f; + state->color_transform_values[2] = 1.0f; + state->color_transform_values[3] = 1.0f; + state->color_transform_values[4] = 0.0f; + state->color_transform_values[5] = 0.0f; + state->color_transform_values[6] = 0.0f; + state->color_transform_values[7] = 0.0f; + + /* Stroke parameters */ + state->stroke.line_width.f = 1.0f; + state->stroke.line_width.i = 1; + state->stroke.cap_style = VG_CAP_BUTT; + state->stroke.join_style = VG_JOIN_MITER; + state->stroke.miter_limit.f = 4.0f; + state->stroke.miter_limit.i = 4; + state->stroke.dash_pattern_num = 0; + state->stroke.dash_phase.f = 0.0f; + state->stroke.dash_phase.i = 0; + state->stroke.dash_phase_reset = VG_FALSE; + + /* Edge fill color for VG_TILE_FILL tiling mode */ + state->tile_fill_color[0] = 0.0f; + state->tile_fill_color[1] = 0.0f; + state->tile_fill_color[2] = 0.0f; + state->tile_fill_color[3] = 0.0f; + + /* Color for vgClear */ + state->clear_color[0] = 0.0f; + state->clear_color[1] = 0.0f; + state->clear_color[2] = 0.0f; + state->clear_color[3] = 0.0f; + + /* Glyph origin */ + state->glyph_origin[0].f = 0.0f; + state->glyph_origin[1].f = 0.0f; + state->glyph_origin[0].i = 0; + state->glyph_origin[1].i = 0; + + /* Enable/disable alpha masking and scissoring */ + state->masking = VG_FALSE; + state->scissoring = VG_FALSE; + + /* Pixel layout information */ + state->pixel_layout = VG_PIXEL_LAYOUT_UNKNOWN; + state->screen_layout = VG_PIXEL_LAYOUT_UNKNOWN; + + /* Source format selection for image filters */ + state->filter_format_linear = VG_FALSE; + state->filter_format_premultiplied = VG_FALSE; + + /* Destination write enable mask for image filters */ + state->filter_channel_mask = (VG_RED | VG_GREEN | VG_BLUE | VG_ALPHA); + + matrix_load_identity(&state->path_user_to_surface_matrix); + matrix_load_identity(&state->image_user_to_surface_matrix); + matrix_load_identity(&state->fill_paint_to_user_matrix); + matrix_load_identity(&state->stroke_paint_to_user_matrix); + matrix_load_identity(&state->glyph_user_to_surface_matrix); +} + +struct matrix *vg_state_matrix(struct vg_state *state) +{ + switch(state->matrix_mode) { + case VG_MATRIX_PATH_USER_TO_SURFACE: + return &state->path_user_to_surface_matrix; + case VG_MATRIX_IMAGE_USER_TO_SURFACE: + return &state->image_user_to_surface_matrix; + case VG_MATRIX_FILL_PAINT_TO_USER: + return &state->fill_paint_to_user_matrix; + case VG_MATRIX_STROKE_PAINT_TO_USER: + return &state->stroke_paint_to_user_matrix; +#ifdef OPENVG_VERSION_1_1 + case VG_MATRIX_GLYPH_USER_TO_SURFACE: + return &state->glyph_user_to_surface_matrix; +#endif + default: + break; + } + return NULL; +} diff --git a/src/gallium/state_trackers/vega/vg_state.h b/src/gallium/state_trackers/vega/vg_state.h new file mode 100644 index 00000000000..ed90689f917 --- /dev/null +++ b/src/gallium/state_trackers/vega/vg_state.h @@ -0,0 +1,109 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef VG_STATE_H +#define VG_STATE_H + +#include "VG/openvg.h" + +#include "api_consts.h" +#include "matrix.h" + +struct vg_value +{ + VGfloat f; + VGint i; +}; + +struct vg_state { + /* Mode settings */ + VGMatrixMode matrix_mode; + VGFillRule fill_rule; + VGImageQuality image_quality; + VGRenderingQuality rendering_quality; + VGBlendMode blend_mode; + VGImageMode image_mode; + + /* Scissoring rectangles */ + struct vg_value scissor_rects[32*4]; + VGint scissor_rects_num; + + /* Color Transformation */ + VGboolean color_transform; + VGfloat color_transform_values[8]; + + /* Stroke parameters */ + struct { + struct vg_value line_width; + VGCapStyle cap_style; + VGJoinStyle join_style; + struct vg_value miter_limit; + struct vg_value dash_pattern[VEGA_MAX_DASH_COUNT]; + VGint dash_pattern_num; + struct vg_value dash_phase; + VGboolean dash_phase_reset; + } stroke; + + /* Edge fill color for VG_TILE_FILL tiling mode */ + VGfloat tile_fill_color[4]; + VGint tile_fill_colori[4]; + + /* Color for vgClear */ + VGfloat clear_color[4]; + VGint clear_colori[4]; + + /* Glyph origin */ + struct vg_value glyph_origin[2]; + + /* Enable/disable alpha masking and scissoring */ + VGboolean masking; + VGboolean scissoring; + + /* Pixel layout information */ + VGPixelLayout pixel_layout; + VGPixelLayout screen_layout; + + /* Source format selection for image filters */ + VGboolean filter_format_linear; + VGboolean filter_format_premultiplied; + + /* Destination write enable mask for image filters */ + VGbitfield filter_channel_mask; + + struct matrix path_user_to_surface_matrix; + struct matrix image_user_to_surface_matrix; + struct matrix fill_paint_to_user_matrix; + struct matrix stroke_paint_to_user_matrix; + struct matrix glyph_user_to_surface_matrix; + + struct vg_paint *stroke_paint; + struct vg_paint *fill_paint; +}; + +void vg_init_state(struct vg_state *state); +struct matrix * vg_state_matrix(struct vg_state *state); + +#endif diff --git a/src/gallium/state_trackers/vega/vg_tracker.c b/src/gallium/state_trackers/vega/vg_tracker.c new file mode 100644 index 00000000000..56cc60aebe1 --- /dev/null +++ b/src/gallium/state_trackers/vega/vg_tracker.c @@ -0,0 +1,416 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "vg_context.h" +#include "vg_tracker.h" +#include "mask.h" + +#include "pipe/p_context.h" +#include "pipe/p_inlines.h" +#include "pipe/p_screen.h" +#include "util/u_memory.h" +#include "util/u_math.h" + +static struct pipe_texture * +create_texture(struct pipe_context *pipe, enum pipe_format format, + VGint width, VGint height) +{ + struct pipe_texture templ; + + memset(&templ, 0, sizeof(templ)); + + if (format != PIPE_FORMAT_NONE) { + templ.format = format; + } + else { + templ.format = PIPE_FORMAT_A8R8G8B8_UNORM; + } + + templ.target = PIPE_TEXTURE_2D; + pf_get_block(templ.format, &templ.block); + templ.width[0] = width; + templ.height[0] = height; + templ.depth[0] = 1; + templ.last_level = 0; + + if (pf_get_component_bits(format, PIPE_FORMAT_COMP_S)) { + templ.tex_usage = PIPE_TEXTURE_USAGE_DEPTH_STENCIL; + } else { + templ.tex_usage = (PIPE_TEXTURE_USAGE_DISPLAY_TARGET | + PIPE_TEXTURE_USAGE_RENDER_TARGET | + PIPE_TEXTURE_USAGE_SAMPLER); + } + + return pipe->screen->texture_create(pipe->screen, &templ); +} + +/** + * Allocate a renderbuffer for a an on-screen window (not a user-created + * renderbuffer). The window system code determines the format. + */ +static struct st_renderbuffer * +st_new_renderbuffer_fb(enum pipe_format format) +{ + struct st_renderbuffer *strb; + + strb = CALLOC_STRUCT(st_renderbuffer); + if (!strb) { + /*_vega_error(NULL, VG_OUT_OF_MEMORY, "creating renderbuffer");*/ + return NULL; + } + + strb->format = format; + + return strb; +} + + +/** + * This is called to allocate the original drawing surface, and + * during window resize. + */ +static VGboolean +st_renderbuffer_alloc_storage(struct vg_context * ctx, + struct st_renderbuffer *strb, + VGuint width, VGuint height) +{ + struct pipe_context *pipe = ctx->pipe; + unsigned surface_usage; + + /* Free the old surface and texture + */ + pipe_surface_reference(&strb->surface, NULL); + pipe_texture_reference(&strb->texture, NULL); + + + /* Probably need dedicated flags for surface usage too: + */ + surface_usage = (PIPE_BUFFER_USAGE_GPU_READ | + PIPE_BUFFER_USAGE_GPU_WRITE); + + strb->texture = create_texture(pipe, strb->format, + width, height); + + if (!strb->texture) + return FALSE; + + strb->surface = pipe->screen->get_tex_surface(pipe->screen, + strb->texture, + 0, 0, 0, + surface_usage); + strb->width = width; + strb->height = height; + + assert(strb->surface->width == width); + assert(strb->surface->height == height); + + return strb->surface != NULL; +} + +struct vg_context * st_create_context(struct pipe_context *pipe, + const void *visual, + struct vg_context *share) +{ + struct vg_context *ctx = vg_create_context(pipe, visual, share); + /*debug_printf("--------- CREATE CONTEXT %p\n", ctx);*/ + return ctx; +} + +void st_destroy_context(struct vg_context *st) +{ + /*debug_printf("--------- DESTROY CONTEXT %p\n", st);*/ + vg_destroy_context(st); +} + +void st_copy_context_state(struct vg_context *dst, struct vg_context *src, + uint mask) +{ + fprintf(stderr, "FIXME: %s\n", __FUNCTION__); +} + +void st_get_framebuffer_dimensions(struct st_framebuffer *stfb, + uint *width, + uint *height) +{ + *width = stfb->strb->width; + *height = stfb->strb->height; +} + +struct st_framebuffer * st_create_framebuffer(const void *visual, + enum pipe_format colorFormat, + enum pipe_format depthFormat, + enum pipe_format stencilFormat, + uint width, uint height, + void *privateData) +{ + struct st_framebuffer *stfb = CALLOC_STRUCT(st_framebuffer); + if (stfb) { + struct st_renderbuffer *rb = + st_new_renderbuffer_fb(colorFormat); + stfb->strb = rb; +#if 0 + if (doubleBuffer) { + struct st_renderbuffer *rb = + st_new_renderbuffer_fb(colorFormat); + } +#endif + + /* we want to combine the depth/stencil */ + if (stencilFormat == depthFormat) + stfb->dsrb = st_new_renderbuffer_fb(stencilFormat); + else + stfb->dsrb = st_new_renderbuffer_fb(PIPE_FORMAT_S8Z24_UNORM); + + /*### currently we always allocate it but it's possible it's + not necessary if EGL_ALPHA_MASK_SIZE was 0 + */ + stfb->alpha_mask = 0; + + stfb->init_width = width; + stfb->init_height = height; + stfb->privateData = privateData; + } + + return stfb; +} + +static void setup_new_alpha_mask(struct vg_context *ctx, + struct st_framebuffer *stfb, + uint width, uint height) +{ + struct pipe_context *pipe = ctx->pipe; + struct pipe_texture *old_texture = stfb->alpha_mask; + + /* + we use PIPE_FORMAT_A8R8G8B8_UNORM because we want to render to + this texture and use it as a sampler, so while this wastes some + space it makes both of those a lot simpler + */ + stfb->alpha_mask = + create_texture(pipe, PIPE_FORMAT_A8R8G8B8_UNORM, width, height); + + if (!stfb->alpha_mask) { + if (old_texture) + pipe_texture_reference(&old_texture, NULL); + return; + } + + vg_validate_state(ctx); + + /* alpha mask starts with 1.f alpha */ + mask_fill(0, 0, width, height, 1.f); + + /* if we had an old surface copy it over */ + if (old_texture) { + struct pipe_surface *surface = pipe->screen->get_tex_surface( + pipe->screen, + stfb->alpha_mask, + 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_WRITE); + struct pipe_surface *old_surface = pipe->screen->get_tex_surface( + pipe->screen, + old_texture, + 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_READ); + pipe->surface_copy(pipe, + surface, + 0, 0, + old_surface, + 0, 0, + MIN2(old_surface->width, width), + MIN2(old_surface->height, height)); + if (surface) + pipe_surface_reference(&surface, NULL); + if (old_surface) + pipe_surface_reference(&old_surface, NULL); + } + + /* Free the old texture + */ + if (old_texture) + pipe_texture_reference(&old_texture, NULL); +} + +void st_resize_framebuffer(struct st_framebuffer *stfb, + uint width, uint height) +{ + struct vg_context *ctx = vg_current_context(); + struct st_renderbuffer *strb = stfb->strb; + struct pipe_framebuffer_state *state; + + if (!ctx) + return; + + state = &ctx->state.g3d.fb; + + /* If this is a noop, exit early and don't do the clear, etc below. + */ + if (strb->width == width && + strb->height == height && + state->zsbuf) + return; + + if (strb->width != width || strb->height != height) + st_renderbuffer_alloc_storage(ctx, strb, + width, height); + + if (stfb->dsrb->width != width || stfb->dsrb->height != height) + st_renderbuffer_alloc_storage(ctx, stfb->dsrb, + width, height); + + { + VGuint i; + + memset(state, 0, sizeof(struct pipe_framebuffer_state)); + + state->width = width; + state->height = height; + + state->nr_cbufs = 1; + state->cbufs[0] = strb->surface; + for (i = 1; i < PIPE_MAX_COLOR_BUFS; ++i) + state->cbufs[i] = 0; + + state->zsbuf = stfb->dsrb->surface; + + cso_set_framebuffer(ctx->cso_context, state); + } + + ctx->state.dirty |= VIEWPORT_DIRTY; + ctx->state.dirty |= DEPTH_STENCIL_DIRTY;/*to reset the scissors*/ + + ctx->pipe->clear(ctx->pipe, PIPE_CLEAR_DEPTHSTENCIL, + NULL, 0.0, 0); + + /* we need all the other state already set */ + + setup_new_alpha_mask(ctx, stfb, width, height); + + pipe_texture_reference( &stfb->blend_texture, NULL ); + stfb->blend_texture = create_texture(ctx->pipe, PIPE_FORMAT_A8R8G8B8_UNORM, + width, height); +} + +void st_set_framebuffer_surface(struct st_framebuffer *stfb, + uint surfIndex, struct pipe_surface *surf) +{ + struct st_renderbuffer *rb = stfb->strb; + + /* unreference existing surfaces */ + pipe_surface_reference( &rb->surface, NULL ); + pipe_texture_reference( &rb->texture, NULL ); + + /* reference new ones */ + pipe_surface_reference( &rb->surface, surf ); + pipe_texture_reference( &rb->texture, surf->texture ); + + rb->width = surf->width; + rb->height = surf->height; +} + +int st_get_framebuffer_surface(struct st_framebuffer *stfb, + uint surfIndex, struct pipe_surface **surf) +{ + struct st_renderbuffer *rb = stfb->strb; + *surf = rb->surface; + return VG_TRUE; +} + +int st_get_framebuffer_texture(struct st_framebuffer *stfb, + uint surfIndex, struct pipe_texture **tex) +{ + struct st_renderbuffer *rb = stfb->strb; + *tex = rb->texture; + return VG_TRUE; +} + +void * st_framebuffer_private(struct st_framebuffer *stfb) +{ + return stfb->privateData; +} + +void st_unreference_framebuffer(struct st_framebuffer *stfb) +{ + /* FIXME */ +} + +void st_make_current(struct vg_context *st, + struct st_framebuffer *draw, + struct st_framebuffer *read) +{ + vg_set_current_context(st); + if (st) { + st->draw_buffer = draw; + } +} + +struct vg_context *st_get_current(void) +{ + return vg_current_context(); +} + +void st_flush(struct vg_context *st, uint pipeFlushFlags, + struct pipe_fence_handle **fence) +{ + st->pipe->flush(st->pipe, pipeFlushFlags, fence); +} + +void st_finish(struct vg_context *st) +{ + struct pipe_fence_handle *fence = NULL; + + st_flush(st, PIPE_FLUSH_RENDER_CACHE, &fence); + + st->pipe->screen->fence_finish(st->pipe->screen, fence, 0); + st->pipe->screen->fence_reference(st->pipe->screen, &fence, NULL); +} + +void st_notify_swapbuffers(struct st_framebuffer *stfb) +{ + struct vg_context *ctx = vg_current_context(); + if (ctx && ctx->draw_buffer == stfb) { + st_flush(ctx, + PIPE_FLUSH_RENDER_CACHE | + PIPE_FLUSH_SWAPBUFFERS | + PIPE_FLUSH_FRAME, + NULL); + } +} + +void st_notify_swapbuffers_complete(struct st_framebuffer *stfb) +{ +} + +int st_bind_texture_surface(struct pipe_surface *ps, int target, int level, + enum pipe_format format) +{ + return 0; +} + +int st_unbind_texture_surface(struct pipe_surface *ps, int target, int level) +{ + return 0; +} diff --git a/src/gallium/state_trackers/vega/vg_tracker.h b/src/gallium/state_trackers/vega/vg_tracker.h new file mode 100644 index 00000000000..5457631106f --- /dev/null +++ b/src/gallium/state_trackers/vega/vg_tracker.h @@ -0,0 +1,107 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef VG_TRACKER_H +#define VG_TRACKER_H + +#include "VG/openvg.h" + +#include "pipe/p_compiler.h" +#include "pipe/p_format.h" + +#define ST_SURFACE_FRONT_LEFT 0 +#define ST_SURFACE_BACK_LEFT 1 +#define ST_SURFACE_FRONT_RIGHT 2 +#define ST_SURFACE_BACK_RIGHT 3 +#define ST_SURFACE_DEPTH 8 + +struct vg_context; +struct st_framebuffer; +struct pipe_context; +struct pipe_fence_handle; +struct pipe_surface; + + +struct vg_context *st_create_context(struct pipe_context *pipe, + const void *visual, + struct vg_context *share); + +void st_destroy_context( struct vg_context *st ); + +void st_copy_context_state(struct vg_context *dst, struct vg_context *src, + uint mask); + +struct st_framebuffer *st_create_framebuffer(const void *visual, + enum pipe_format colorFormat, + enum pipe_format depthFormat, + enum pipe_format stencilFormat, + uint width, uint height, + void *privateData); + +void st_resize_framebuffer(struct st_framebuffer *stfb, + uint width, uint height); + +void st_set_framebuffer_surface(struct st_framebuffer *stfb, + uint surfIndex, struct pipe_surface *surf); + +void st_get_framebuffer_dimensions( struct st_framebuffer *stfb, + uint *width, uint *height); + +int st_bind_texture_surface(struct pipe_surface *ps, int target, int level, + enum pipe_format format); + +int st_unbind_texture_surface(struct pipe_surface *ps, int target, int level); + +int st_get_framebuffer_surface(struct st_framebuffer *stfb, + uint surfIndex, struct pipe_surface **surf); + +int st_get_framebuffer_texture(struct st_framebuffer *stfb, + uint surfIndex, struct pipe_texture **tex); + +void *st_framebuffer_private(struct st_framebuffer *stfb); + +void st_unreference_framebuffer(struct st_framebuffer *stfb); + +void st_make_current(struct vg_context *st, + struct st_framebuffer *draw, + struct st_framebuffer *read); + +struct vg_context *st_get_current(void); + +void st_flush(struct vg_context *st, uint pipeFlushFlags, + struct pipe_fence_handle **fence); +void st_finish(struct vg_context *st); + +void st_notify_swapbuffers(struct st_framebuffer *stfb); +void st_notify_swapbuffers_complete(struct st_framebuffer *stfb); + + +/** Generic function type */ +typedef void (*st_proc)(); + +st_proc st_get_proc_address(const char *procname); + +#endif diff --git a/src/gallium/state_trackers/vega/vg_translate.c b/src/gallium/state_trackers/vega/vg_translate.c new file mode 100644 index 00000000000..00e07647062 --- /dev/null +++ b/src/gallium/state_trackers/vega/vg_translate.c @@ -0,0 +1,1030 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "vg_translate.h" + +#include "pipe/p_format.h" +#include "util/u_pack_color.h" + +void _vega_pack_rgba_span_float(struct vg_context *ctx, + VGuint n, VGfloat rgba[][4], + VGImageFormat dstFormat, + void *dstAddr) +{ + VGint i; + + switch (dstFormat) { + case VG_sRGBX_8888: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = 255; + dst[i] = r << 24 | g << 16 | b << 8 | a; + } + return; + } + break; + case VG_sRGBA_8888: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = r << 24 | g << 16 | b << 8 | a; + } + return; + } + break; + case VG_sRGBA_8888_PRE: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = r << 24 | g << 16 | b << 8 | a; + } + return; + } + break; + case VG_sRGB_565: { + VGshort *dst = (VGshort *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + r = (r / 255.0) * 32; + g = (g / 255.0) * 32; + b = (b / 255.0) * 32; + + dst[i] = b | g << 5 | r << 11; + } + return; + } + break; + case VG_sRGBA_5551: { + VGshort *dst = (VGshort *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b, a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + r = (r / 255.0) * 32; + g = (g / 255.0) * 32; + b = (b / 255.0) * 32; + a = (a / 255.0); + + dst[i] = a | b << 1 | g << 6 | r << 11; + } + return; + } + break; + case VG_sRGBA_4444: { + VGshort *dst = (VGshort *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b, a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + r = (r / 255.0) * 16; + g = (g / 255.0) * 16; + b = (b / 255.0) * 16; + a = (a / 255.0) * 16; + + dst[i] = a | b << 4 | g << 8 | r << 12; + } + return; + } + break; + case VG_sL_8: { + VGubyte *dst = (VGubyte *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b, a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + + dst[i] = a; + } + return; + } + break; + case VG_lRGBX_8888: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = 255; + dst[i] = r << 24 | g << 16 | b << 8 | a; + } + return; + } + break; + case VG_lRGBA_8888: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = r << 24 | g << 16 | b << 8 | a; + } + return; + } + case VG_lRGBA_8888_PRE: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = r << 24 | g << 16 | b << 8 | a; + } + return; + } + break; + case VG_lL_8: { + VGubyte *dst = (VGubyte *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = a; + } + return; + } + break; + case VG_A_8: { + VGubyte *dst = (VGubyte *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b, a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + + dst[i] = a; + } + return; + } + break; + case VG_BW_1: { + VGshort *dst = (VGshort *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b, a; + VGubyte res; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + + res = (r + g + b + a)/4; + dst[i] = (res & (128)); + } + return; + } + break; +#ifdef OPENVG_VERSION_1_1 + case VG_A_1: { + VGshort *dst = (VGshort *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b, a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + + dst[i] = (a & (128)); + } + return; + } + break; + case VG_A_4: { + VGshort *dst = (VGshort *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b, a; + VGubyte res; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + + res = a/4; + dst[i] = (res & (128)); + } + return; + } + break; +#endif + case VG_sXRGB_8888: + break; + case VG_sARGB_8888: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = a << 24 | r << 16 | g << 8 | b; + } + return; + } + break; + case VG_sARGB_8888_PRE: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = a << 24 | r << 16 | g << 8 | b; + } + return; + } + break; + case VG_sARGB_1555: + break; + case VG_sARGB_4444: + break; + case VG_lXRGB_8888: + break; + case VG_lARGB_8888: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = a << 24 | r << 16 | g << 8 | b; + } + return; + } + break; + case VG_lARGB_8888_PRE: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = a << 24 | r << 16 | g << 8 | b; + } + return; + } + break; + case VG_sBGRX_8888: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = 0xff; + dst[i] = b << 24 | g << 16 | r << 8 | a; + } + return; + } + break; + case VG_sBGRA_8888: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = b << 24 | g << 16 | r << 8 | a; + } + return; + } + break; + case VG_sBGRA_8888_PRE: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = b << 24 | g << 16 | r << 8 | a; + } + return; + } + break; + case VG_sBGR_565: + break; + case VG_sBGRA_5551: + break; + case VG_sBGRA_4444: + break; + case VG_lBGRX_8888: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = 0xff; + dst[i] = b << 24 | g << 16 | r << 8 | a; + } + return; + } + break; + case VG_lBGRA_8888: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = b << 24 | g << 16 | r << 8 | a; + } + return; + } + break; + case VG_lBGRA_8888_PRE: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = b << 24 | g << 16 | r << 8 | a; + } + return; + } + break; + case VG_sXBGR_8888: + break; + case VG_sABGR_8888: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = a << 24 | b << 16 | g << 8 | r; + } + return; + } + break; + case VG_sABGR_8888_PRE: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = a << 24 | b << 16 | g << 8 | r; + } + return; + } + break; + case VG_sABGR_1555: + break; + case VG_sABGR_4444: + break; + case VG_lXBGR_8888: + break; + case VG_lABGR_8888: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = a << 24 | b << 16 | g << 8 | r; + } + return; + } + break; + case VG_lABGR_8888_PRE: { + VGint *dst = (VGint *)dstAddr; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = float_to_ubyte(rgba[i][0]); + g = float_to_ubyte(rgba[i][1]); + b = float_to_ubyte(rgba[i][2]); + a = float_to_ubyte(rgba[i][3]); + dst[i] = a << 24 | b << 16 | g << 8 | r; + } + return; + } + break; + default: + assert(!"Unknown ReadPixels format"); + break; + } + assert(!"Not implemented ReadPixels format"); +} + +void _vega_unpack_float_span_rgba(struct vg_context *ctx, + VGuint n, + VGuint offset, + const void * data, + VGImageFormat dataFormat, + VGfloat rgba[][4]) +{ + VGint i; + + switch (dataFormat) { + case VG_sRGBX_8888: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = (*src >> 24) & 0xff; + g = (*src >> 16) & 0xff; + b = (*src >> 8) & 0xff; + a = 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + } + return; + case VG_sRGBA_8888: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = (*src >> 24) & 0xff; + g = (*src >> 16) & 0xff; + b = (*src >> 8) & 0xff; + a = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_sRGBA_8888_PRE: { + VGint *src = (VGint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = (*src >> 24) & 0xff; + g = (*src >> 16) & 0xff; + b = (*src >> 8) & 0xff; + a = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_sRGB_565: { + VGshort *src = (VGshort *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGfloat clr[4]; + clr[0] = ((*src >> 10) & 31)/31.; + clr[1] = ((*src >> 5) & 95)/95.; + clr[2] = ((*src >> 0) & 31)/31.; + clr[3] = 1.f; + + util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + } + return; + case VG_sRGBA_5551: { + VGshort *src = (VGshort *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGfloat clr[4]; + clr[0] = ((*src >> 10) & 31)/31.; + clr[1] = ((*src >> 5) & 31)/31.; + clr[2] = ((*src >> 1) & 31)/31.; + clr[3] = ((*src >> 0) & 1)/1.; + + util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + } + return; + case VG_sRGBA_4444: { + VGshort *src = (VGshort *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGfloat clr[4]; + clr[0] = ((*src >> 12) & 15)/15.; + clr[1] = ((*src >> 8) & 15)/15.; + clr[2] = ((*src >> 4) & 15)/15.; + clr[3] = ((*src >> 0) & 15)/15.; + + util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + } + return; + case VG_sL_8: { + VGubyte *src = (VGubyte *)data; + src += offset; + for (i = 0; i < n; ++i) { + util_pack_color_ub(0xff, 0xff, 0xff, *src, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + } + return; + case VG_lRGBX_8888: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = (*src >> 24) & 0xff; + g = (*src >> 16) & 0xff; + b = (*src >> 8) & 0xff; + a = 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + } + return; + case VG_lRGBA_8888: { + VGint *src = (VGint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = (*src >> 24) & 0xff; + g = (*src >> 16) & 0xff; + b = (*src >> 8) & 0xff; + a = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_lRGBA_8888_PRE: { + VGint *src = (VGint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + r = (*src >> 24) & 0xff; + g = (*src >> 16) & 0xff; + b = (*src >> 8) & 0xff; + a = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_lL_8: { + VGubyte *src = (VGubyte *)data; + src += offset; + for (i = 0; i < n; ++i) { + util_pack_color_ub(0xff, 0xff, 0xff, *src, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + } + return; + case VG_A_8: { + VGubyte *src = (VGubyte *)data; + src += offset; + for (i = 0; i < n; ++i) { + util_pack_color_ub(0xff, 0xff, 0xff, *src, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + } + return; + case VG_BW_1: { + VGubyte *src = (VGubyte *)data; + src += offset; + for (i = 0; i < n; i += 8) { + VGfloat clr[4]; + VGint j; + for (j = 0; j < 8 && j < n ; ++j) { + VGint shift = j; + clr[0] = (((*src) & (1<<shift)) >> shift); + clr[1] = clr[0]; + clr[2] = clr[0]; + clr[3] = 1.f; + + util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i+j]); + } + ++src; + } + } + return; +#ifdef OPENVG_VERSION_1_1 + case VG_A_1: { + VGubyte *src = (VGubyte *)data; + src += offset; + for (i = 0; i < n; i += 8) { + VGfloat clr[4]; + VGint j; + for (j = 0; j < 8 && j < n ; ++j) { + VGint shift = j; + clr[0] = 0.f; + clr[1] = 0.f; + clr[2] = 0.f; + clr[3] = (((*src) & (1<<shift)) >> shift); + + util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i+j]); + } + ++src; + } + } + return; + case VG_A_4: { + VGubyte *src = (VGubyte *)data; + src += offset/2; + for (i = 0; i < n; i += 2) { + VGfloat clr[4]; + VGint j; + for (j = 0; j < n && j < 2; ++j) { + VGint bitter, shift; + if (j == 0) { + bitter = 0x0f; + shift = 0; + } else { + bitter = 0xf0; + shift = 4; + } + clr[0] = 0.f; + clr[1] = 0.f; + clr[2] = 0.f; + clr[3] = ((*src) & (bitter)) >> shift; + + util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i +j]); + } + ++src; + } + } + return; +#endif + case VG_sXRGB_8888: + break; + case VG_sARGB_8888: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + a = (*src >> 24) & 0xff; + r = (*src >> 16) & 0xff; + g = (*src >> 8) & 0xff; + b = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_sARGB_8888_PRE: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + a = (*src >> 24) & 0xff; + r = (*src >> 16) & 0xff; + g = (*src >> 8) & 0xff; + b = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_sARGB_1555: + break; + case VG_sARGB_4444: + break; + case VG_lXRGB_8888: + break; + case VG_lARGB_8888: { + VGint *src = (VGint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + a = (*src >> 24) & 0xff; + r = (*src >> 16) & 0xff; + g = (*src >> 8) & 0xff; + b = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_lARGB_8888_PRE: { + VGint *src = (VGint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + a = (*src >> 24) & 0xff; + r = (*src >> 16) & 0xff; + g = (*src >> 8) & 0xff; + b = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_sBGRX_8888: + break; + case VG_sBGRA_8888: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + b = (*src >> 24) & 0xff; + g = (*src >> 16) & 0xff; + r = (*src >> 8) & 0xff; + a = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_sBGRA_8888_PRE: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + b = (*src >> 24) & 0xff; + g = (*src >> 16) & 0xff; + r = (*src >> 8) & 0xff; + a = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_sBGR_565: + break; + case VG_sBGRA_5551: + break; + case VG_sBGRA_4444: + break; + case VG_lBGRX_8888: + break; + case VG_lBGRA_8888: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + b = (*src >> 24) & 0xff; + g = (*src >> 16) & 0xff; + r = (*src >> 8) & 0xff; + a = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_lBGRA_8888_PRE: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + b = (*src >> 24) & 0xff; + g = (*src >> 16) & 0xff; + r = (*src >> 8) & 0xff; + a = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_sXBGR_8888: + break; + case VG_sABGR_8888: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + a = (*src >> 24) & 0xff; + b = (*src >> 16) & 0xff; + g = (*src >> 8) & 0xff; + r = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_sABGR_8888_PRE: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + a = (*src >> 24) & 0xff; + b = (*src >> 16) & 0xff; + g = (*src >> 8) & 0xff; + r = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_sABGR_1555: + break; + case VG_sABGR_4444: + break; + case VG_lXBGR_8888: + break; + case VG_lABGR_8888: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + a = (*src >> 24) & 0xff; + b = (*src >> 16) & 0xff; + g = (*src >> 8) & 0xff; + r = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + case VG_lABGR_8888_PRE: { + VGuint *src = (VGuint *)data; + src += offset; + for (i = 0; i < n; ++i) { + VGubyte r, g, b ,a; + a = (*src >> 24) & 0xff; + b = (*src >> 16) & 0xff; + g = (*src >> 8) & 0xff; + r = (*src >> 0) & 0xff; + + util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, + rgba[i]); + ++src; + } + return; + } + break; + default: + assert(!"Unknown ReadPixels format"); + break; + } + assert(!"Not implemented ReadPixels format"); +} + +VGint _vega_size_for_format(VGImageFormat dataFormat) +{ + switch (dataFormat) { + case VG_sRGBX_8888: + case VG_sRGBA_8888: + case VG_sRGBA_8888_PRE: + return 4; + case VG_sRGB_565: + case VG_sRGBA_5551: + case VG_sRGBA_4444: + return 2; + case VG_sL_8: + return 1; + case VG_lRGBX_8888: + case VG_lRGBA_8888: + case VG_lRGBA_8888_PRE: + return 4; + case VG_lL_8: + return 1; + case VG_A_8: + return 1; + case VG_BW_1: + return 1; +#ifdef OPENVG_VERSION_1_1 + case VG_A_1: + break; + case VG_A_4: + break; +#endif + case VG_sXRGB_8888: + case VG_sARGB_8888: + case VG_sARGB_8888_PRE: + return 4; + case VG_sARGB_1555: + case VG_sARGB_4444: + return 2; + case VG_lXRGB_8888: + case VG_lARGB_8888: + case VG_lARGB_8888_PRE: + case VG_sBGRX_8888: + case VG_sBGRA_8888: + case VG_sBGRA_8888_PRE: + return 4; + case VG_sBGR_565: + case VG_sBGRA_5551: + case VG_sBGRA_4444: + return 2; + case VG_lBGRX_8888: + case VG_lBGRA_8888: + case VG_lBGRA_8888_PRE: + case VG_sXBGR_8888: + case VG_sABGR_8888: + case VG_sABGR_8888_PRE: + return 4; + case VG_sABGR_1555: + case VG_sABGR_4444: + return 2; + case VG_lXBGR_8888: + case VG_lABGR_8888: + case VG_lABGR_8888_PRE: + return 4; + default: + assert(!"Unknown ReadPixels format"); + break; + } + assert(!"Not implemented ReadPixels format"); + return 0; +} diff --git a/src/gallium/state_trackers/vega/vg_translate.h b/src/gallium/state_trackers/vega/vg_translate.h new file mode 100644 index 00000000000..70815bacbcb --- /dev/null +++ b/src/gallium/state_trackers/vega/vg_translate.h @@ -0,0 +1,49 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef VG_TRANSLATE_H +#define VG_TRANSLATE_H + +#include "VG/openvg.h" +#include "vg_context.h" + +/*FIXME: we really should be using translate module + * but pipe_format can't express some of the VG formats + * (the premultiplied ones) so currently it won't work */ + +void _vega_pack_rgba_span_float(struct vg_context *ctx, + VGuint n, VGfloat rgba[][4], + VGImageFormat dstFormat, + void *dstAddr); +void _vega_unpack_float_span_rgba(struct vg_context *ctx, + VGuint n, + VGuint offset, + const void * data, + VGImageFormat dataFormat, + VGfloat rgba[][4]); +VGint _vega_size_for_format(VGImageFormat format); + +#endif diff --git a/src/gallium/state_trackers/vega/vgu.c b/src/gallium/state_trackers/vega/vgu.c new file mode 100644 index 00000000000..7dc51c5599c --- /dev/null +++ b/src/gallium/state_trackers/vega/vgu.c @@ -0,0 +1,450 @@ +/************************************************************************** + * + * Copyright 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "VG/openvg.h" +#include "VG/vgu.h" + +#include "matrix.h" +#include "path.h" + +#include "util/u_debug.h" +#include "util/u_pointer.h" + +#include <math.h> +#include <assert.h> + +static VGboolean is_aligned_to(const void *ptr, VGbyte alignment) +{ + void *aligned = align_pointer(ptr, alignment); + return (ptr == aligned) ? VG_TRUE : VG_FALSE; +} + +static VGboolean is_aligned(const void *ptr) +{ + return is_aligned_to(ptr, 4); +} + +static void vgu_append_float_coords(VGPath path, + const VGubyte *cmds, + VGint num_cmds, + const VGfloat *coords, + VGint num_coords) +{ + VGubyte common_data[40 * sizeof(VGfloat)]; + struct path *p = (struct path *)path; + + vg_float_to_datatype(path_datatype(p), common_data, coords, num_coords); + vgAppendPathData(path, num_cmds, cmds, common_data); +} + +VGUErrorCode vguLine(VGPath path, + VGfloat x0, VGfloat y0, + VGfloat x1, VGfloat y1) +{ + static const VGubyte cmds[] = {VG_MOVE_TO_ABS, VG_LINE_TO_ABS}; + VGfloat coords[4]; + VGbitfield caps; + + if (path == VG_INVALID_HANDLE) { + return VGU_BAD_HANDLE_ERROR; + } + caps = vgGetPathCapabilities(path); + if (!(caps & VG_PATH_CAPABILITY_APPEND_TO)) { + return VGU_PATH_CAPABILITY_ERROR; + } + + coords[0] = x0; + coords[1] = y0; + coords[2] = x1; + coords[3] = y1; + + vgu_append_float_coords(path, cmds, 2, coords, 4); + + return VGU_NO_ERROR; +} + +VGUErrorCode vguPolygon(VGPath path, + const VGfloat * points, + VGint count, + VGboolean closed) +{ + VGubyte *cmds; + VGfloat *coords; + VGbitfield caps; + VGint i; + + if (path == VG_INVALID_HANDLE) { + return VGU_BAD_HANDLE_ERROR; + } + + if (!points || count <= 0 || !is_aligned(points)) { + return VGU_ILLEGAL_ARGUMENT_ERROR; + } + + caps = vgGetPathCapabilities(path); + if (!(caps & VG_PATH_CAPABILITY_APPEND_TO)) { + return VGU_PATH_CAPABILITY_ERROR; + } + + cmds = malloc(sizeof(VGubyte) * count + 1); + coords = malloc(sizeof(VGfloat) * count * 2); + + cmds[0] = VG_MOVE_TO_ABS; + coords[0] = points[0]; + coords[1] = points[1]; + for (i = 1; i < count; ++i) { + cmds[i] = VG_LINE_TO_ABS; + coords[2*i + 0] = points[2*i + 0]; + coords[2*i + 1] = points[2*i + 1]; + } + + if (closed) { + cmds[i] = VG_CLOSE_PATH; + ++i; + } + + vgu_append_float_coords(path, cmds, i, coords, 2*i); + + free(cmds); + free(coords); + + return VGU_NO_ERROR; +} + +VGUErrorCode vguRect(VGPath path, + VGfloat x, VGfloat y, + VGfloat width, VGfloat height) +{ + static const VGubyte cmds[] = {VG_MOVE_TO_ABS, + VG_HLINE_TO_REL, + VG_VLINE_TO_REL, + VG_HLINE_TO_REL, + VG_CLOSE_PATH + }; + VGfloat coords[5]; + VGbitfield caps; + + if (path == VG_INVALID_HANDLE) { + return VGU_BAD_HANDLE_ERROR; + } + caps = vgGetPathCapabilities(path); + if (!(caps & VG_PATH_CAPABILITY_APPEND_TO)) { + return VGU_PATH_CAPABILITY_ERROR; + } + if (width <= 0 || height <= 0) { + return VGU_ILLEGAL_ARGUMENT_ERROR; + } + + coords[0] = x; + coords[1] = y; + coords[2] = width; + coords[3] = height; + coords[4] = -width; + + vgu_append_float_coords(path, cmds, 5, coords, 5); + + return VGU_NO_ERROR; +} + +VGUErrorCode vguRoundRect(VGPath path, + VGfloat x, VGfloat y, + VGfloat width, + VGfloat height, + VGfloat arcWidth, + VGfloat arcHeight) +{ + static const VGubyte cmds[] = {VG_MOVE_TO_ABS, + VG_HLINE_TO_REL, + VG_SCCWARC_TO_REL, + VG_VLINE_TO_REL, + VG_SCCWARC_TO_REL, + VG_HLINE_TO_REL, + VG_SCCWARC_TO_REL, + VG_VLINE_TO_REL, + VG_SCCWARC_TO_REL, + VG_CLOSE_PATH + }; + VGfloat c[26]; + VGbitfield caps; + + if (path == VG_INVALID_HANDLE) { + return VGU_BAD_HANDLE_ERROR; + } + caps = vgGetPathCapabilities(path); + if (!(caps & VG_PATH_CAPABILITY_APPEND_TO)) { + return VGU_PATH_CAPABILITY_ERROR; + } + if (width <= 0 || height <= 0) { + return VGU_ILLEGAL_ARGUMENT_ERROR; + } + + c[0] = x + arcWidth/2; c[1] = y; + + c[2] = width - arcWidth; + + c[3] = arcWidth/2; c[4] = arcHeight/2; c[5] = 0; + c[6] = arcWidth/2; c[7] = arcHeight/2; + + c[8] = height - arcHeight; + + c[9] = arcWidth/2; c[10] = arcHeight/2; c[11] = 0; + c[12] = -arcWidth/2; c[13] = arcHeight/2; + + c[14] = -(width - arcWidth); + + c[15] = arcWidth/2; c[16] = arcHeight/2; c[17] = 0; + c[18] = -arcWidth/2; c[19] = -arcHeight/2; + + c[20] = -(height - arcHeight); + + c[21] = arcWidth/2; c[22] = arcHeight/2; c[23] = 0; + c[24] = arcWidth/2; c[25] = -arcHeight/2; + + vgu_append_float_coords(path, cmds, 10, c, 26); + + return VGU_NO_ERROR; +} + +VGUErrorCode vguEllipse(VGPath path, + VGfloat cx, VGfloat cy, + VGfloat width, + VGfloat height) +{ + static const VGubyte cmds[] = {VG_MOVE_TO_ABS, + VG_SCCWARC_TO_REL, + VG_SCCWARC_TO_REL, + VG_CLOSE_PATH + }; + VGfloat coords[12]; + VGbitfield caps; + + if (path == VG_INVALID_HANDLE) { + return VGU_BAD_HANDLE_ERROR; + } + caps = vgGetPathCapabilities(path); + if (!(caps & VG_PATH_CAPABILITY_APPEND_TO)) { + return VGU_PATH_CAPABILITY_ERROR; + } + if (width <= 0 || height <= 0) { + return VGU_ILLEGAL_ARGUMENT_ERROR; + } + + coords[0] = cx + width/2; coords[1] = cy; + + coords[2] = width/2; coords[3] = height/2; coords[4] = 0; + coords[5] = -width; coords[6] = 0; + + coords[7] = width/2; coords[8] = height/2; coords[9] = 0; + coords[10] = width; coords[11] = 0; + + vgu_append_float_coords(path, cmds, 4, coords, 11); + + return VGU_NO_ERROR; +} + +VGUErrorCode vguArc(VGPath path, + VGfloat x, VGfloat y, + VGfloat width, VGfloat height, + VGfloat startAngle, + VGfloat angleExtent, + VGUArcType arcType) +{ + VGubyte cmds[11]; + VGfloat coords[40]; + VGbitfield caps; + VGfloat last = startAngle + angleExtent; + VGint i, c = 0; + + if (path == VG_INVALID_HANDLE) { + return VGU_BAD_HANDLE_ERROR; + } + caps = vgGetPathCapabilities(path); + if (!(caps & VG_PATH_CAPABILITY_APPEND_TO)) { + return VGU_PATH_CAPABILITY_ERROR; + } + if (width <= 0 || height <= 0) { + return VGU_ILLEGAL_ARGUMENT_ERROR; + } + if (arcType != VGU_ARC_OPEN && + arcType != VGU_ARC_CHORD && + arcType != VGU_ARC_PIE) { + return VGU_ILLEGAL_ARGUMENT_ERROR; + } + + cmds[c] = VG_MOVE_TO_ABS; ++c; + coords[0] = x+cos(DEGREES_TO_RADIANS(startAngle))*width/2; + coords[1] = y+sin(DEGREES_TO_RADIANS(startAngle))*height/2; +#ifdef DEBUG_VGUARC + debug_printf("start [%f, %f]\n", coords[0], coords[1]); +#endif + i = 2; + if (angleExtent > 0) { + VGfloat angle = startAngle + 180; + while (angle < last) { + cmds[c] = VG_SCCWARC_TO_ABS; ++c; + coords[i] = width/2; coords[i+1] = height/2; coords[i+2] = 0; + coords[i+3] = x + cos(DEGREES_TO_RADIANS(angle))*width/2; + coords[i+4] = y + sin(DEGREES_TO_RADIANS(angle))*height/2; +#ifdef DEBUG_VGUARC + debug_printf("1 [%f, %f]\n", coords[i+3], + coords[i+4]); +#endif + i += 5; + angle += 180; + } + cmds[c] = VG_SCCWARC_TO_ABS; ++c; + coords[i] = width/2; coords[i+1] = height/2; coords[i+2] = 0; + coords[i+3] = x+cos(DEGREES_TO_RADIANS(last))*width/2; + coords[i+4] = y+sin(DEGREES_TO_RADIANS(last))*height/2; +#ifdef DEBUG_VGUARC + debug_printf("2 [%f, %f]\n", coords[i+3], + coords[i+4]); +#endif + i += 5; + } else { + VGfloat angle = startAngle - 180; + while (angle > last) { + cmds[c] = VG_SCWARC_TO_ABS; ++c; + coords[i] = width/2; coords[i+1] = height/2; coords[i+2] = 0; + coords[i+3] = x + cos(DEGREES_TO_RADIANS(angle)) * width/2; + coords[i+4] = y + sin(DEGREES_TO_RADIANS(angle)) * height/2; +#ifdef DEBUG_VGUARC + debug_printf("3 [%f, %f]\n", coords[i+3], + coords[i+4]); +#endif + angle -= 180; + i += 5; + } + cmds[c] = VG_SCWARC_TO_ABS; ++c; + coords[i] = width/2; coords[i+1] = height/2; coords[i+2] = 0; + coords[i+3] = x + cos(DEGREES_TO_RADIANS(last)) * width/2; + coords[i+4] = y + sin(DEGREES_TO_RADIANS(last)) * height/2; +#ifdef DEBUG_VGUARC + debug_printf("4 [%f, %f]\n", coords[i+3], + coords[i+4]); +#endif + i += 5; + } + + if (arcType == VGU_ARC_PIE) { + cmds[c] = VG_LINE_TO_ABS; ++c; + coords[i] = x; coords[i + 1] = y; + i += 2; + } + if (arcType == VGU_ARC_PIE || arcType == VGU_ARC_CHORD) { + cmds[c] = VG_CLOSE_PATH; + ++c; + } + + assert(c < 11); + + vgu_append_float_coords(path, cmds, c, coords, i); + + return VGU_NO_ERROR; +} + +VGUErrorCode vguComputeWarpQuadToSquare(VGfloat sx0, VGfloat sy0, + VGfloat sx1, VGfloat sy1, + VGfloat sx2, VGfloat sy2, + VGfloat sx3, VGfloat sy3, + VGfloat * matrix) +{ + struct matrix mat; + + if (!matrix || !is_aligned(matrix)) + return VGU_ILLEGAL_ARGUMENT_ERROR; + + if (!matrix_quad_to_square(sx0, sy0, + sx1, sy1, + sx2, sy2, + sx3, sy3, + &mat)) + return VGU_BAD_WARP_ERROR; + + if (!matrix_is_invertible(&mat)) + return VGU_BAD_WARP_ERROR; + + memcpy(matrix, mat.m, sizeof(VGfloat) * 9); + + return VGU_NO_ERROR; +} + +VGUErrorCode vguComputeWarpSquareToQuad(VGfloat dx0, VGfloat dy0, + VGfloat dx1, VGfloat dy1, + VGfloat dx2, VGfloat dy2, + VGfloat dx3, VGfloat dy3, + VGfloat * matrix) +{ + struct matrix mat; + + if (!matrix || !is_aligned(matrix)) + return VGU_ILLEGAL_ARGUMENT_ERROR; + + if (!matrix_square_to_quad(dx0, dy0, + dx1, dy1, + dx2, dy2, + dx3, dy3, + &mat)) + return VGU_BAD_WARP_ERROR; + + if (!matrix_is_invertible(&mat)) + return VGU_BAD_WARP_ERROR; + + memcpy(matrix, mat.m, sizeof(VGfloat) * 9); + + return VGU_NO_ERROR; +} + +VGUErrorCode vguComputeWarpQuadToQuad(VGfloat dx0, VGfloat dy0, + VGfloat dx1, VGfloat dy1, + VGfloat dx2, VGfloat dy2, + VGfloat dx3, VGfloat dy3, + VGfloat sx0, VGfloat sy0, + VGfloat sx1, VGfloat sy1, + VGfloat sx2, VGfloat sy2, + VGfloat sx3, VGfloat sy3, + VGfloat * matrix) +{ + struct matrix mat; + + if (!matrix || !is_aligned(matrix)) + return VGU_ILLEGAL_ARGUMENT_ERROR; + + if (!matrix_quad_to_quad(dx0, dy0, + dx1, dy1, + dx2, dy2, + dx3, dy3, + sx0, sy0, + sx1, sy1, + sx2, sy2, + sx3, sy3, + &mat)) + return VGU_BAD_WARP_ERROR; + + memcpy(matrix, mat.m, sizeof(VGfloat) * 9); + + return VGU_NO_ERROR; +} diff --git a/src/gallium/state_trackers/xorg/Makefile b/src/gallium/state_trackers/xorg/Makefile index a00ea3e2a4e..27a1990724d 100644 --- a/src/gallium/state_trackers/xorg/Makefile +++ b/src/gallium/state_trackers/xorg/Makefile @@ -1,29 +1,18 @@ -TARGET = libxorgtracker.a -CFILES = $(wildcard ./*.c) -OBJECTS = $(patsubst ./%.c,./%.o,$(CFILES)) -GALLIUMDIR = ../.. -TOP = ../../../.. - +TOP = ../../../.. include $(TOP)/configs/current -CFLAGS = -DHAVE_CONFIG_H \ - -g -Wall -Wimplicit-function-declaration -fPIC \ - $(shell pkg-config --cflags pixman-1 xorg-server libdrm xproto) \ - -I$(GALLIUMDIR)/include \ - -I$(GALLIUMDIR)/auxiliary \ - -I$(TOP)/src/mesa/drivers/dri/common \ - -I$(TOP)/src/mesa \ - -I$(TOP)/include \ - -I$(TOP)/src/egl/main - -############################################# - -.PHONY = all clean +LIBNAME = xorgtracker -all: $(TARGET) +LIBRARY_INCLUDES = \ + -DHAVE_CONFIG_H \ + $(shell pkg-config --cflags-only-I pixman-1 xorg-server libdrm xproto) \ + -I$(TOP)/src/gallium/include \ + -I$(TOP)/src/gallium/auxiliary \ + -I$(TOP)/include \ + -I$(TOP)/src/mesa \ + -I$(TOP)/src/mesa/drivers/dri/common \ + -I$(TOP)/src/mesa/main -$(TARGET): $(OBJECTS) - ar rcs $(TARGET) $(OBJECTS) +C_SOURCES = $(wildcard ./*.c) -clean: - rm -rf $(OBJECTS) $(TARGET) +include ../../Makefile.template diff --git a/src/gallium/state_trackers/xorg/SConscript b/src/gallium/state_trackers/xorg/SConscript new file mode 100644 index 00000000000..65f55ea378a --- /dev/null +++ b/src/gallium/state_trackers/xorg/SConscript @@ -0,0 +1,27 @@ +####################################################################### +# SConscript for xorg state_tracker + +Import('*') + +if 'xorg' in env['statetrackers']: + + env = env.Clone() + + env.Append(CPPPATH = [ + '#/src/mesa', + ]) + + env.ParseConfig('pkg-config --cflags --libs xorg-server') + + st_xorg = env.ConvenienceLibrary( + target = 'st_xorg', + source = [ 'xorg_composite.c', + 'xorg_crtc.c', + 'xorg_dri2.c', + 'xorg_driver.c', + 'xorg_exa.c', + 'xorg_exa_tgsi.c', + 'xorg_output.c', + ] + ) + Export('st_xorg') diff --git a/src/gallium/state_trackers/xorg/xorg_composite.c b/src/gallium/state_trackers/xorg/xorg_composite.c new file mode 100644 index 00000000000..c708ac31702 --- /dev/null +++ b/src/gallium/state_trackers/xorg/xorg_composite.c @@ -0,0 +1,584 @@ +#include "xorg_composite.h" + +#include "xorg_exa_tgsi.h" + +#include "cso_cache/cso_context.h" +#include "util/u_draw_quad.h" + +#include "pipe/p_inlines.h" + +struct xorg_composite_blend { + int op:8; + + unsigned rgb_src_factor:5; /**< PIPE_BLENDFACTOR_x */ + unsigned rgb_dst_factor:5; /**< PIPE_BLENDFACTOR_x */ + + unsigned alpha_src_factor:5; /**< PIPE_BLENDFACTOR_x */ + unsigned alpha_dst_factor:5; /**< PIPE_BLENDFACTOR_x */ +}; + +#define BLEND_OP_OVER 3 +static const struct xorg_composite_blend xorg_blends[] = { + { PictOpClear, + PIPE_BLENDFACTOR_CONST_COLOR, PIPE_BLENDFACTOR_CONST_ALPHA, + PIPE_BLENDFACTOR_ZERO, PIPE_BLENDFACTOR_ZERO }, + + { PictOpSrc, + PIPE_BLENDFACTOR_ONE, PIPE_BLENDFACTOR_ONE, + PIPE_BLENDFACTOR_ZERO, PIPE_BLENDFACTOR_ZERO }, + + { PictOpDst, + PIPE_BLENDFACTOR_ZERO, PIPE_BLENDFACTOR_ZERO, + PIPE_BLENDFACTOR_ONE, PIPE_BLENDFACTOR_ONE }, + + { PictOpOver, + PIPE_BLENDFACTOR_SRC_ALPHA, PIPE_BLENDFACTOR_ONE, + PIPE_BLENDFACTOR_INV_SRC_ALPHA, PIPE_BLENDFACTOR_INV_SRC_ALPHA }, + + { PictOpOverReverse, + PIPE_BLENDFACTOR_SRC_ALPHA, PIPE_BLENDFACTOR_ONE, + PIPE_BLENDFACTOR_INV_SRC_ALPHA, PIPE_BLENDFACTOR_INV_SRC_ALPHA }, +}; + +static INLINE void +pixel_to_float4(PictFormatPtr format, + CARD32 pixel, float *color) +{ + CARD32 r, g, b, a; + + debug_assert(format->type == PictTypeDirect); + + r = (pixel >> format->direct.red) & format->direct.redMask; + g = (pixel >> format->direct.green) & format->direct.greenMask; + b = (pixel >> format->direct.blue) & format->direct.blueMask; + a = (pixel >> format->direct.alpha) & format->direct.alphaMask; + color[0] = ((float)r) / ((float)format->direct.redMask); + color[1] = ((float)g) / ((float)format->direct.greenMask); + color[2] = ((float)b) / ((float)format->direct.blueMask); + color[3] = ((float)a) / ((float)format->direct.alphaMask); +} + +struct acceleration_info { + int op : 16; + int with_mask : 1; + int component_alpha : 1; +}; +static const struct acceleration_info accelerated_ops[] = { + {PictOpClear, 1, 0}, + {PictOpSrc, 1, 0}, + {PictOpDst, 1, 0}, + {PictOpOver, 1, 0}, + {PictOpOverReverse, 1, 0}, + {PictOpIn, 1, 0}, + {PictOpInReverse, 1, 0}, + {PictOpOut, 1, 0}, + {PictOpOutReverse, 1, 0}, + {PictOpAtop, 1, 0}, + {PictOpAtopReverse, 1, 0}, + {PictOpXor, 1, 0}, + {PictOpAdd, 1, 0}, + {PictOpSaturate, 1, 0}, +}; + +static struct xorg_composite_blend +blend_for_op(int op) +{ + const int num_blends = + sizeof(xorg_blends)/sizeof(struct xorg_composite_blend); + int i; + + for (i = 0; i < num_blends; ++i) { + if (xorg_blends[i].op == op) + return xorg_blends[i]; + } + return xorg_blends[BLEND_OP_OVER]; +} + +static INLINE int +render_repeat_to_gallium(int mode) +{ + switch(mode) { + case RepeatNone: + return PIPE_TEX_WRAP_CLAMP; + case RepeatNormal: + return PIPE_TEX_WRAP_REPEAT; + case RepeatReflect: + return PIPE_TEX_WRAP_MIRROR_REPEAT; + case RepeatPad: + return PIPE_TEX_WRAP_CLAMP_TO_EDGE; + default: + debug_printf("Unsupported repeat mode\n"); + } + return PIPE_TEX_WRAP_REPEAT; +} + + +static INLINE void +setup_vertex0(float vertex[2][4], float x, float y, + float color[4]) +{ + vertex[0][0] = x; + vertex[0][1] = y; + vertex[0][2] = 0.f; /*z*/ + vertex[0][3] = 1.f; /*w*/ + + vertex[1][0] = color[0]; /*r*/ + vertex[1][1] = color[1]; /*g*/ + vertex[1][2] = color[2]; /*b*/ + vertex[1][3] = color[3]; /*a*/ +} + +static struct pipe_buffer * +setup_vertex_data0(struct exa_context *ctx, + int srcX, int srcY, int maskX, int maskY, + int dstX, int dstY, int width, int height) +{ + float vertices[4][2][4]; + + /* 1st vertex */ + setup_vertex0(vertices[0], dstX, dstY, + ctx->solid_color); + /* 2nd vertex */ + setup_vertex0(vertices[1], dstX + width, dstY, + ctx->solid_color); + /* 3rd vertex */ + setup_vertex0(vertices[2], dstX + width, dstY + height, + ctx->solid_color); + /* 4th vertex */ + setup_vertex0(vertices[3], dstX, dstY + height, + ctx->solid_color); + + return pipe_user_buffer_create(ctx->ctx->screen, + vertices, + sizeof(vertices)); +} + +static INLINE void +setup_vertex1(float vertex[2][4], float x, float y, float s, float t) +{ + vertex[0][0] = x; + vertex[0][1] = y; + vertex[0][2] = 0.f; /*z*/ + vertex[0][3] = 1.f; /*w*/ + + vertex[1][0] = s; /*s*/ + vertex[1][1] = t; /*t*/ + vertex[1][2] = 0.f; /*r*/ + vertex[1][3] = 1.f; /*q*/ +} + +static struct pipe_buffer * +setup_vertex_data1(struct exa_context *ctx, + int srcX, int srcY, int maskX, int maskY, + int dstX, int dstY, int width, int height) +{ + float vertices[4][2][4]; + float s0, t0, s1, t1; + struct pipe_texture *src = ctx->bound_textures[0]; + + s0 = srcX / src->width[0]; + s1 = srcX + width / src->width[0]; + t0 = srcY / src->height[0]; + t1 = srcY + height / src->height[0]; + + /* 1st vertex */ + setup_vertex1(vertices[0], dstX, dstY, + s0, t0); + /* 2nd vertex */ + setup_vertex1(vertices[1], dstX + width, dstY, + s1, t0); + /* 3rd vertex */ + setup_vertex1(vertices[2], dstX + width, dstY + height, + s1, t1); + /* 4th vertex */ + setup_vertex1(vertices[3], dstX, dstY + height, + s0, t1); + + return pipe_user_buffer_create(ctx->ctx->screen, + vertices, + sizeof(vertices)); +} + + +static INLINE void +setup_vertex2(float vertex[3][4], float x, float y, + float s0, float t0, float s1, float t1) +{ + vertex[0][0] = x; + vertex[0][1] = y; + vertex[0][2] = 0.f; /*z*/ + vertex[0][3] = 1.f; /*w*/ + + vertex[1][0] = s0; /*s*/ + vertex[1][1] = t0; /*t*/ + vertex[1][2] = 0.f; /*r*/ + vertex[1][3] = 1.f; /*q*/ + + vertex[2][0] = s1; /*s*/ + vertex[2][1] = t1; /*t*/ + vertex[2][2] = 0.f; /*r*/ + vertex[2][3] = 1.f; /*q*/ +} + +static struct pipe_buffer * +setup_vertex_data2(struct exa_context *ctx, + int srcX, int srcY, int maskX, int maskY, + int dstX, int dstY, int width, int height) +{ + float vertices[4][3][4]; + float st0[4], st1[4]; + struct pipe_texture *src = ctx->bound_textures[0]; + struct pipe_texture *mask = ctx->bound_textures[0]; + + st0[0] = srcX / src->width[0]; + st0[1] = srcY / src->height[0]; + st0[2] = srcX + width / src->width[0]; + st0[3] = srcY + height / src->height[0]; + + st1[0] = maskX / mask->width[0]; + st1[1] = maskY / mask->height[0]; + st1[2] = maskX + width / mask->width[0]; + st1[3] = maskY + height / mask->height[0]; + + /* 1st vertex */ + setup_vertex2(vertices[0], dstX, dstY, + st0[0], st0[1], st1[0], st1[1]); + /* 2nd vertex */ + setup_vertex2(vertices[1], dstX + width, dstY, + st0[2], st0[1], st1[2], st1[1]); + /* 3rd vertex */ + setup_vertex2(vertices[2], dstX + width, dstY + height, + st0[2], st0[3], st1[2], st1[3]); + /* 4th vertex */ + setup_vertex2(vertices[3], dstX, dstY + height, + st0[0], st0[3], st1[0], st1[3]); + + return pipe_user_buffer_create(ctx->ctx->screen, + vertices, + sizeof(vertices)); +} + +boolean xorg_composite_accelerated(int op, + PicturePtr pSrcPicture, + PicturePtr pMaskPicture, + PicturePtr pDstPicture) +{ + unsigned i; + unsigned accel_ops_count = + sizeof(accelerated_ops)/sizeof(struct acceleration_info); + + + /*FIXME: currently accel is disabled */ + return FALSE; + + if (pSrcPicture) { + /* component alpha not supported */ + if (pSrcPicture->componentAlpha) + return FALSE; + /* fills not supported */ + if (pSrcPicture->pSourcePict) + return FALSE; + } + + for (i = 0; i < accel_ops_count; ++i) { + if (op == accelerated_ops[i].op) { + if (pMaskPicture && !accelerated_ops[i].with_mask) + return FALSE; + return TRUE; + } + } + return FALSE; +} + +static void +bind_framebuffer_state(struct exa_context *exa, PicturePtr pDstPicture, + struct exa_pixmap_priv *pDst) +{ + unsigned i; + struct pipe_framebuffer_state state; + struct pipe_surface *surface = exa_gpu_surface(exa, pDst); + memset(&state, 0, sizeof(struct pipe_framebuffer_state)); + + state.width = pDstPicture->pDrawable->width; + state.height = pDstPicture->pDrawable->height; + + state.nr_cbufs = 1; + state.cbufs[0] = surface; + for (i = 1; i < PIPE_MAX_COLOR_BUFS; ++i) + state.cbufs[i] = 0; + + /* currently we don't use depth/stencil */ + state.zsbuf = 0; + + cso_set_framebuffer(exa->cso, &state); +} + +enum AxisOrientation { + Y0_BOTTOM, + Y0_TOP +}; + +static void +set_viewport(struct exa_context *exa, int width, int height, + enum AxisOrientation orientation) +{ + struct pipe_viewport_state viewport; + float y_scale = (orientation == Y0_BOTTOM) ? -2.f : 2.f; + + viewport.scale[0] = width / 2.f; + viewport.scale[1] = height / y_scale; + viewport.scale[2] = 1.0; + viewport.scale[3] = 1.0; + viewport.translate[0] = width / 2.f; + viewport.translate[1] = height / 2.f; + viewport.translate[2] = 0.0; + viewport.translate[3] = 0.0; + + cso_set_viewport(exa->cso, &viewport); +} + +static void +bind_viewport_state(struct exa_context *exa, PicturePtr pDstPicture) +{ + int width = pDstPicture->pDrawable->width; + int height = pDstPicture->pDrawable->height; + + set_viewport(exa, width, height, Y0_TOP); +} + +static void +bind_blend_state(struct exa_context *exa, int op, + PicturePtr pSrcPicture, PicturePtr pMaskPicture) +{ + boolean component_alpha = pSrcPicture->componentAlpha; + struct xorg_composite_blend blend_opt; + struct pipe_blend_state blend; + + if (component_alpha) { + op = PictOpOver; + } + blend_opt = blend_for_op(op); + + memset(&blend, 0, sizeof(struct pipe_blend_state)); + blend.blend_enable = 1; + blend.colormask |= PIPE_MASK_R; + blend.colormask |= PIPE_MASK_G; + blend.colormask |= PIPE_MASK_B; + blend.colormask |= PIPE_MASK_A; + + blend.rgb_src_factor = blend_opt.rgb_src_factor; + blend.alpha_src_factor = blend_opt.alpha_src_factor; + blend.rgb_dst_factor = blend_opt.rgb_dst_factor; + blend.alpha_dst_factor = blend_opt.alpha_dst_factor; + + cso_set_blend(exa->cso, &blend); +} + +static void +bind_rasterizer_state(struct exa_context *exa) +{ + struct pipe_rasterizer_state raster; + memset(&raster, 0, sizeof(struct pipe_rasterizer_state)); + raster.gl_rasterization_rules = 1; + cso_set_rasterizer(exa->cso, &raster); +} + +static void +bind_shaders(struct exa_context *exa, int op, + PicturePtr pSrcPicture, PicturePtr pMaskPicture) +{ + unsigned vs_traits = 0, fs_traits = 0; + struct xorg_shader shader; + + if (pSrcPicture) { + if (pSrcPicture->pSourcePict) { + if (pSrcPicture->pSourcePict->type == SourcePictTypeSolidFill) { + fs_traits |= FS_SOLID_FILL; + vs_traits |= VS_SOLID_FILL; + pixel_to_float4(pSrcPicture->pFormat, + pSrcPicture->pSourcePict->solidFill.color, + exa->solid_color); + } else { + debug_assert("!gradients not supported"); + } + } else { + fs_traits |= FS_COMPOSITE; + vs_traits |= VS_COMPOSITE; + } + } + + if (pMaskPicture) { + vs_traits |= VS_MASK; + fs_traits |= FS_MASK; + } + + shader = xorg_shaders_get(exa->shaders, vs_traits, fs_traits); + cso_set_vertex_shader_handle(exa->cso, shader.vs); + cso_set_fragment_shader_handle(exa->cso, shader.fs); +} + + +static void +bind_samplers(struct exa_context *exa, int op, + PicturePtr pSrcPicture, PicturePtr pMaskPicture, + PicturePtr pDstPicture, + struct exa_pixmap_priv *pSrc, + struct exa_pixmap_priv *pMask, + struct exa_pixmap_priv *pDst) +{ + struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS]; + struct pipe_sampler_state src_sampler, mask_sampler; + + exa->num_bound_samplers = 0; + + memset(&src_sampler, 0, sizeof(struct pipe_sampler_state)); + memset(&mask_sampler, 0, sizeof(struct pipe_sampler_state)); + + if (pSrcPicture && pSrc) { + unsigned src_wrap = render_repeat_to_gallium( + pSrcPicture->repeatType); + src_sampler.wrap_s = src_wrap; + src_sampler.wrap_t = src_wrap; + src_sampler.min_img_filter = PIPE_TEX_MIPFILTER_NEAREST; + src_sampler.mag_img_filter = PIPE_TEX_MIPFILTER_NEAREST; + src_sampler.normalized_coords = 1; + samplers[0] = &src_sampler; + exa->bound_textures[0] = pSrc->tex; + ++exa->num_bound_samplers; + } + + if (pMaskPicture && pMask) { + unsigned mask_wrap = render_repeat_to_gallium( + pMaskPicture->repeatType); + mask_sampler.wrap_s = mask_wrap; + mask_sampler.wrap_t = mask_wrap; + mask_sampler.min_img_filter = PIPE_TEX_MIPFILTER_NEAREST; + mask_sampler.mag_img_filter = PIPE_TEX_MIPFILTER_NEAREST; + mask_sampler.normalized_coords = 1; + samplers[1] = &mask_sampler; + exa->bound_textures[1] = pMask->tex; + ++exa->num_bound_samplers; + } + + cso_set_samplers(exa->cso, exa->num_bound_samplers, + (const struct pipe_sampler_state **)samplers); + cso_set_sampler_textures(exa->cso, exa->num_bound_samplers, + exa->bound_textures); +} + +static void +setup_vs_constant_buffer(struct exa_context *exa, + int width, int height) +{ + const int param_bytes = 8 * sizeof(float); + float vs_consts[8] = { + 2.f/width, 2.f/height, 1, 1, + -1, -1, 0, 0 + }; + struct pipe_constant_buffer *cbuf = &exa->vs_const_buffer; + + pipe_buffer_reference(&cbuf->buffer, NULL); + cbuf->buffer = pipe_buffer_create(exa->ctx->screen, 16, + PIPE_BUFFER_USAGE_CONSTANT, + param_bytes); + + if (cbuf->buffer) { + pipe_buffer_write(exa->ctx->screen, cbuf->buffer, + 0, param_bytes, vs_consts); + } + exa->ctx->set_constant_buffer(exa->ctx, PIPE_SHADER_VERTEX, 0, cbuf); +} + + +static void +setup_fs_constant_buffer(struct exa_context *exa) +{ + const int param_bytes = 4 * sizeof(float); + float fs_consts[8] = { + 0, 0, 0, 1, + }; + struct pipe_constant_buffer *cbuf = &exa->fs_const_buffer; + + pipe_buffer_reference(&cbuf->buffer, NULL); + cbuf->buffer = pipe_buffer_create(exa->ctx->screen, 16, + PIPE_BUFFER_USAGE_CONSTANT, + param_bytes); + + if (cbuf->buffer) { + pipe_buffer_write(exa->ctx->screen, cbuf->buffer, + 0, param_bytes, fs_consts); + } + exa->ctx->set_constant_buffer(exa->ctx, PIPE_SHADER_FRAGMENT, 0, cbuf); +} + +static void +setup_constant_buffers(struct exa_context *exa, PicturePtr pDstPicture) +{ + int width = pDstPicture->pDrawable->width; + int height = pDstPicture->pDrawable->height; + + setup_vs_constant_buffer(exa, width, height); + setup_fs_constant_buffer(exa); +} + +boolean xorg_composite_bind_state(struct exa_context *exa, + int op, + PicturePtr pSrcPicture, + PicturePtr pMaskPicture, + PicturePtr pDstPicture, + struct exa_pixmap_priv *pSrc, + struct exa_pixmap_priv *pMask, + struct exa_pixmap_priv *pDst) +{ + bind_framebuffer_state(exa, pDstPicture, pDst); + bind_viewport_state(exa, pDstPicture); + bind_blend_state(exa, op, pSrcPicture, pMaskPicture); + bind_rasterizer_state(exa); + bind_shaders(exa, op, pSrcPicture, pMaskPicture); + bind_samplers(exa, op, pSrcPicture, pMaskPicture, + pDstPicture, pSrc, pMask, pDst); + + setup_constant_buffers(exa, pDstPicture); + + return FALSE; +} + +void xorg_composite(struct exa_context *exa, + struct exa_pixmap_priv *dst, + int srcX, int srcY, int maskX, int maskY, + int dstX, int dstY, int width, int height) +{ + struct pipe_context *pipe = exa->ctx; + struct pipe_buffer *buf = 0; + + if (exa->num_bound_samplers == 0 ) { /* solid fill */ + buf = setup_vertex_data0(exa, + srcX, srcY, maskX, maskY, + dstX, dstY, width, height); + } else if (exa->num_bound_samplers == 1 ) /* src */ + buf = setup_vertex_data1(exa, + srcX, srcY, maskX, maskY, + dstX, dstY, width, height); + else if (exa->num_bound_samplers == 2) /* src + mask */ + buf = setup_vertex_data2(exa, + srcX, srcY, maskX, maskY, + dstX, dstY, width, height); + else if (exa->num_bound_samplers == 3) { /* src + mask + dst */ + debug_assert(!"src/mask/dst not handled right now"); +#if 0 + buf = setup_vertex_data2(exa, + srcX, srcY, maskX, maskY, + dstX, dstY, width, height); +#endif + } + + if (buf) { + util_draw_vertex_buffer(pipe, buf, 0, + PIPE_PRIM_TRIANGLE_FAN, + 4, /* verts */ + 1 + exa->num_bound_samplers); /* attribs/vert */ + + pipe_buffer_reference(&buf, NULL); + } +} + diff --git a/src/gallium/state_trackers/xorg/xorg_composite.h b/src/gallium/state_trackers/xorg/xorg_composite.h new file mode 100644 index 00000000000..17dfcb199ea --- /dev/null +++ b/src/gallium/state_trackers/xorg/xorg_composite.h @@ -0,0 +1,25 @@ +#ifndef XORG_COMPOSITE_H +#define XORG_COMPOSITE_H + +#include "xorg_exa.h" + +boolean xorg_composite_accelerated(int op, + PicturePtr pSrcPicture, + PicturePtr pMaskPicture, + PicturePtr pDstPicture); + +boolean xorg_composite_bind_state(struct exa_context *exa, + int op, + PicturePtr pSrcPicture, + PicturePtr pMaskPicture, + PicturePtr pDstPicture, + struct exa_pixmap_priv *pSrc, + struct exa_pixmap_priv *pMask, + struct exa_pixmap_priv *pDst); + +void xorg_composite(struct exa_context *exa, + struct exa_pixmap_priv *dst, + int srcX, int srcY, int maskX, int maskY, + int dstX, int dstY, int width, int height); + +#endif diff --git a/src/gallium/state_trackers/xorg/xorg_crtc.c b/src/gallium/state_trackers/xorg/xorg_crtc.c index 7304113a659..fe08bde9efc 100644 --- a/src/gallium/state_trackers/xorg/xorg_crtc.c +++ b/src/gallium/state_trackers/xorg/xorg_crtc.c @@ -46,13 +46,14 @@ #include <X11/extensions/dpms.h> #include "pipe/p_inlines.h" +#include "util/u_rect.h" struct crtc_private { drmModeCrtcPtr drm_crtc; /* hwcursor */ - struct pipe_buffer *cursor_buf; + struct pipe_texture *cursor_tex; unsigned cursor_handle; }; @@ -171,11 +172,10 @@ crtc_shadow_destroy(xf86CrtcPtr crtc, PixmapPtr rotate_pixmap, void *data) static void crtc_destroy(xf86CrtcPtr crtc) { - modesettingPtr ms = modesettingPTR(crtc->scrn); struct crtc_private *crtcp = crtc->driver_private; - if (crtcp->cursor_buf) - pipe_buffer_reference(&crtcp->cursor_buf, NULL); + if (crtcp->cursor_tex) + pipe_texture_reference(&crtcp->cursor_tex, NULL); drmModeFreeCrtc(crtcp->drm_crtc); xfree(crtcp); @@ -187,24 +187,42 @@ crtc_load_cursor_argb(xf86CrtcPtr crtc, CARD32 * image) unsigned char *ptr; modesettingPtr ms = modesettingPTR(crtc->scrn); struct crtc_private *crtcp = crtc->driver_private; - - if (!crtcp->cursor_buf) { - crtcp->cursor_buf = pipe_buffer_create(ms->screen, - 0, - PIPE_BUFFER_USAGE_CPU_WRITE | - PIPE_BUFFER_USAGE_GPU_READ, - 64*64*4); - drm_api_hooks.handle_from_buffer(ms->screen, - crtcp->cursor_buf, - &crtcp->cursor_handle); + struct pipe_transfer *transfer; + + if (!crtcp->cursor_tex) { + struct pipe_texture templat; + unsigned pitch; + + memset(&templat, 0, sizeof(templat)); + templat.tex_usage |= PIPE_TEXTURE_USAGE_RENDER_TARGET; + templat.tex_usage |= PIPE_TEXTURE_USAGE_PRIMARY; + templat.target = PIPE_TEXTURE_2D; + templat.last_level = 0; + templat.depth[0] = 1; + templat.format = PIPE_FORMAT_A8R8G8B8_UNORM; + templat.width[0] = 64; + templat.height[0] = 64; + pf_get_block(templat.format, &templat.block); + + crtcp->cursor_tex = ms->screen->texture_create(ms->screen, + &templat); + ms->api->local_handle_from_texture(ms->api, + ms->screen, + crtcp->cursor_tex, + &pitch, + &crtcp->cursor_handle); } - ptr = pipe_buffer_map(ms->screen, crtcp->cursor_buf, PIPE_BUFFER_USAGE_CPU_WRITE); - - if (ptr) - memcpy(ptr, image, 64 * 64 * 4); - - pipe_buffer_unmap(ms->screen, crtcp->cursor_buf); + transfer = ms->screen->get_tex_transfer(ms->screen, crtcp->cursor_tex, + 0, 0, 0, + PIPE_TRANSFER_WRITE, + 0, 0, 64, 64); + ptr = ms->screen->transfer_map(ms->screen, transfer); + util_copy_rect(ptr, &crtcp->cursor_tex->block, + transfer->stride, 0, 0, + 64, 64, (void*)image, 64 * 4, 0, 0); + ms->screen->transfer_unmap(ms->screen, transfer); + ms->screen->tex_transfer_destroy(transfer); } static void @@ -222,7 +240,7 @@ crtc_show_cursor(xf86CrtcPtr crtc) modesettingPtr ms = modesettingPTR(crtc->scrn); struct crtc_private *crtcp = crtc->driver_private; - if (crtcp->cursor_buf) + if (crtcp->cursor_tex) drmModeSetCursor(ms->fd, crtcp->drm_crtc->crtc_id, crtcp->cursor_handle, 64, 64); } @@ -260,13 +278,12 @@ static const xf86CrtcFuncsRec crtc_funcs = { }; void -cursor_destroy(xf86CrtcPtr crtc) +crtc_cursor_destroy(xf86CrtcPtr crtc) { - modesettingPtr ms = modesettingPTR(crtc->scrn); struct crtc_private *crtcp = crtc->driver_private; - if (crtcp->cursor_buf) { - pipe_buffer_reference(&crtcp->cursor_buf, NULL); + if (crtcp->cursor_tex) { + pipe_texture_reference(&crtcp->cursor_tex, NULL); } } diff --git a/src/gallium/state_trackers/xorg/xorg_dri2.c b/src/gallium/state_trackers/xorg/xorg_dri2.c index 401bd39dac0..6431a0fe254 100644 --- a/src/gallium/state_trackers/xorg/xorg_dri2.c +++ b/src/gallium/state_trackers/xorg/xorg_dri2.c @@ -33,97 +33,193 @@ #include "xf86_OSproc.h" #include "xorg_tracker.h" +#include "xorg_exa.h" #include "dri2.h" #include "pipe/p_state.h" #include "pipe/p_inlines.h" +#include "util/u_rect.h" + typedef struct { PixmapPtr pPixmap; struct pipe_texture *tex; - struct pipe_buffer *buf; + struct pipe_fence_handle *fence; } *BufferPrivatePtr; -static DRI2BufferPtr -driCreateBuffers(DrawablePtr pDraw, unsigned int *attachments, int count) +static Bool +driDoCreateBuffer(DrawablePtr pDraw, DRI2BufferPtr buffer, unsigned int format) { - struct pipe_texture *depth, *tex; - struct pipe_buffer *buf; + struct pipe_texture *tex = NULL; ScreenPtr pScreen = pDraw->pScreen; ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum]; modesettingPtr ms = modesettingPTR(pScrn); - BufferPrivatePtr privates; - DRI2BufferPtr buffers; + struct exa_pixmap_priv *exa_priv; + BufferPrivatePtr private = buffer->driverPrivate; PixmapPtr pPixmap; unsigned stride, handle; - int i; - - buffers = xcalloc(count, sizeof *buffers); - if (!buffers) - goto fail_buffers; - privates = xcalloc(count, sizeof *privates); - if (!privates) - goto fail_privates; + if (pDraw->type == DRAWABLE_PIXMAP) + pPixmap = (PixmapPtr) pDraw; + else + pPixmap = (*pScreen->GetWindowPixmap)((WindowPtr) pDraw); + exa_priv = exaGetPixmapDriverPrivate(pPixmap); - depth = NULL; - for (i = 0; i < count; i++) { - pPixmap = NULL; - tex = NULL; - buf = NULL; - if (attachments[i] == DRI2BufferFrontLeft) { - if (pDraw->type == DRAWABLE_PIXMAP) - pPixmap = (PixmapPtr) pDraw; - else - pPixmap = (*pScreen->GetWindowPixmap)((WindowPtr) pDraw); - pPixmap->refcnt++; - tex = xorg_exa_get_texture(pPixmap); - } else if (attachments[i] == DRI2BufferStencil) { - pipe_texture_reference(&tex, depth); - } else if (attachments[i] == DRI2BufferDepth) { - struct pipe_texture template; - - memset(&template, 0, sizeof(template)); - template.target = PIPE_TEXTURE_2D; - template.format = PIPE_FORMAT_S8Z24_UNORM; - pf_get_block(template.format, &template.block); - template.width[0] = pDraw->width; - template.height[0] = pDraw->height; - template.depth[0] = 1; - template.last_level = 0; - template.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET; - tex = ms->screen->texture_create(ms->screen, &template); - } else { + switch (buffer->attachment) { + default: + if (buffer->attachment != DRI2BufferFakeFrontLeft || + pDraw->type != DRAWABLE_PIXMAP) { + private->pPixmap = (*pScreen->CreatePixmap)(pScreen, pDraw->width, + pDraw->height, + pDraw->depth, + 0); + } + break; + case DRI2BufferFrontLeft: + break; + case DRI2BufferStencil: +#if defined(DRI2INFOREC_VERSION) && DRI2INFOREC_VERSION > 2 + case DRI2BufferDepthStencil: + if (exa_priv->depth_stencil_tex && + !pf_is_depth_stencil(exa_priv->depth_stencil_tex->format)) + exa_priv->depth_stencil_tex = NULL; + /* Fall through */ +#endif + case DRI2BufferDepth: + if (exa_priv->depth_stencil_tex) + pipe_texture_reference(&tex, exa_priv->depth_stencil_tex); + else { struct pipe_texture template; memset(&template, 0, sizeof(template)); template.target = PIPE_TEXTURE_2D; - template.format = PIPE_FORMAT_A8R8G8B8_UNORM; + if (buffer->attachment == DRI2BufferDepth) + template.format = ms->ds_depth_bits_last ? + PIPE_FORMAT_X8Z24_UNORM : PIPE_FORMAT_Z24X8_UNORM; + else + template.format = ms->ds_depth_bits_last ? + PIPE_FORMAT_S8Z24_UNORM : PIPE_FORMAT_Z24S8_UNORM; pf_get_block(template.format, &template.block); template.width[0] = pDraw->width; template.height[0] = pDraw->height; template.depth[0] = 1; template.last_level = 0; - template.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET; + template.tex_usage = PIPE_TEXTURE_USAGE_DEPTH_STENCIL | + PIPE_TEXTURE_USAGE_DISPLAY_TARGET; tex = ms->screen->texture_create(ms->screen, &template); + pipe_texture_reference(&exa_priv->depth_stencil_tex, tex); } + break; + } + + if (!private->pPixmap) { + private->pPixmap = pPixmap; + pPixmap->refcnt++; + } + + if (!tex) { + xorg_exa_set_shared_usage(private->pPixmap); + pScreen->ModifyPixmapHeader(private->pPixmap, 0, 0, 0, 0, 0, NULL); + tex = xorg_exa_get_texture(private->pPixmap); + } + + if (!tex) + FatalError("NO TEXTURE IN DRI2\n"); + + ms->api->shared_handle_from_texture(ms->api, ms->screen, tex, &stride, &handle); + + buffer->name = handle; + buffer->pitch = stride; + buffer->cpp = 4; + buffer->driverPrivate = private; + buffer->flags = 0; /* not tiled */ + private->tex = tex; + + return TRUE; +} + +static void +driDoDestroyBuffer(DrawablePtr pDraw, DRI2BufferPtr buffer) +{ + ScreenPtr pScreen = pDraw->pScreen; + ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum]; + modesettingPtr ms = modesettingPTR(pScrn); + BufferPrivatePtr private = buffer->driverPrivate; + struct exa_pixmap_priv *exa_priv = exaGetPixmapDriverPrivate(private->pPixmap); + + pipe_texture_reference(&private->tex, NULL); + ms->screen->fence_reference(ms->screen, &private->fence, NULL); + pipe_texture_reference(&exa_priv->depth_stencil_tex, NULL); + (*pScreen->DestroyPixmap)(private->pPixmap); +} + +#if defined(DRI2INFOREC_VERSION) && DRI2INFOREC_VERSION > 2 + +static DRI2BufferPtr +driCreateBuffer(DrawablePtr pDraw, unsigned int attachment, unsigned int format) +{ + DRI2BufferPtr buffer; + BufferPrivatePtr private; + + buffer = xcalloc(1, sizeof *buffer); + if (!buffer) + return NULL; + + private = xcalloc(1, sizeof *private); + if (!private) { + goto fail; + } + + buffer->attachment = attachment; + buffer->driverPrivate = private; + + if (driDoCreateBuffer(pDraw, buffer, format)) + return buffer; + + xfree(private); +fail: + xfree(buffer); + return NULL; +} + +static void +driDestroyBuffer(DrawablePtr pDraw, DRI2BufferPtr buffer) +{ + driDoDestroyBuffer(pDraw, buffer); + + xfree(buffer->driverPrivate); + xfree(buffer); +} - drm_api_hooks.buffer_from_texture(tex, &buf, &stride); - drm_api_hooks.global_handle_from_buffer(ms->screen, buf, &handle); +#else /* DRI2INFOREC_VERSION <= 2 */ - buffers[i].name = handle; +static DRI2BufferPtr +driCreateBuffers(DrawablePtr pDraw, unsigned int *attachments, int count) +{ + BufferPrivatePtr privates; + DRI2BufferPtr buffers; + int i; + + buffers = xcalloc(count, sizeof *buffers); + if (!buffers) + goto fail_buffers; + + privates = xcalloc(count, sizeof *privates); + if (!privates) + goto fail_privates; + + for (i = 0; i < count; i++) { buffers[i].attachment = attachments[i]; - buffers[i].pitch = stride; - buffers[i].cpp = 4; buffers[i].driverPrivate = &privates[i]; - buffers[i].flags = 0; /* not tiled */ - privates[i].pPixmap = pPixmap; - privates[i].buf = buf; - privates[i].tex = tex; + + if (!driDoCreateBuffer(pDraw, &buffers[i], 0)) + goto fail; } return buffers; +fail: + xfree(privates); fail_privates: xfree(buffers); fail_buffers: @@ -133,20 +229,10 @@ fail_buffers: static void driDestroyBuffers(DrawablePtr pDraw, DRI2BufferPtr buffers, int count) { - ScreenPtr pScreen = pDraw->pScreen; - ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum]; - modesettingPtr ms = modesettingPTR(pScrn); - BufferPrivatePtr private; int i; for (i = 0; i < count; i++) { - private = buffers[i].driverPrivate; - - if (private->pPixmap) - (*pScreen->DestroyPixmap)(private->pPixmap); - - pipe_texture_reference(&private->tex, NULL); - pipe_buffer_reference(&private->buf, NULL); + driDoDestroyBuffer(pDraw, &buffers[i]); } if (buffers) { @@ -155,6 +241,8 @@ driDestroyBuffers(DrawablePtr pDraw, DRI2BufferPtr buffers, int count) } } +#endif /* DRI2INFOREC_VERSION */ + static void driCopyRegion(DrawablePtr pDraw, RegionPtr pRegion, DRI2BufferPtr pDestBuffer, DRI2BufferPtr pSrcBuffer) @@ -164,19 +252,83 @@ driCopyRegion(DrawablePtr pDraw, RegionPtr pRegion, modesettingPtr ms = modesettingPTR(pScrn); BufferPrivatePtr dst_priv = pDestBuffer->driverPrivate; BufferPrivatePtr src_priv = pSrcBuffer->driverPrivate; + PixmapPtr src_pixmap; + PixmapPtr dst_pixmap; + GCPtr gc; + RegionPtr copy_clip; + + /* + * In driCreateBuffers we dewrap windows into the + * backing pixmaps in order to get to the texture. + * We need to use the real drawable in CopyArea + * so that cliprects and offsets are correct. + */ + src_pixmap = src_priv->pPixmap; + dst_pixmap = dst_priv->pPixmap; + if (pSrcBuffer->attachment == DRI2BufferFrontLeft) + src_pixmap = (PixmapPtr)pDraw; + if (pDestBuffer->attachment == DRI2BufferFrontLeft) + dst_pixmap = (PixmapPtr)pDraw; + + /* + * The clients implements glXWaitX with a copy front to fake and then + * waiting on the server to signal its completion of it. While + * glXWaitGL is a client side flush and a copy from fake to front. + * This is how it is done in the DRI2 protocol, how ever depending + * which type of drawables the server does things a bit differently + * then what the protocol says as the fake and front are the same. + * + * for pixmaps glXWaitX is a server flush. + * for pixmaps glXWaitGL is a client flush. + * for windows glXWaitX is a copy from front to fake then a server flush. + * for windows glXWaitGL is a client flush then a copy from fake to front. + * + * XXX in the windows case this code always flushes but that isn't a + * must in the glXWaitGL case but we don't know if this is a glXWaitGL + * or a glFlush/glFinish call. + */ + if (dst_pixmap == src_pixmap) { + /* pixmap glXWaitX */ + if (pSrcBuffer->attachment == DRI2BufferFrontLeft && + pDestBuffer->attachment == DRI2BufferFakeFrontLeft) { + ms->ctx->flush(ms->ctx, PIPE_FLUSH_SWAPBUFFERS, NULL); + return; + } + /* pixmap glXWaitGL */ + if (pDestBuffer->attachment == DRI2BufferFrontLeft && + pSrcBuffer->attachment == DRI2BufferFakeFrontLeft) { + return; + } else { + xf86DrvMsg(pScrn->scrnIndex, X_WARNING, + "copying between the same pixmap\n"); + } + } - struct pipe_surface *dst_surf = - ms->screen->get_tex_surface(ms->screen, dst_priv->tex, 0, 0, 0, - PIPE_BUFFER_USAGE_GPU_WRITE); - struct pipe_surface *src_surf = - ms->screen->get_tex_surface(ms->screen, src_priv->tex, 0, 0, 0, - PIPE_BUFFER_USAGE_GPU_READ); + gc = GetScratchGC(pDraw->depth, pScreen); + copy_clip = REGION_CREATE(pScreen, NULL, 0); + REGION_COPY(pScreen, copy_clip, pRegion); + (*gc->funcs->ChangeClip) (gc, CT_REGION, copy_clip, 0); + ValidateGC(&dst_pixmap->drawable, gc); - ms->ctx->surface_copy(ms->ctx, dst_surf, 0, 0, src_surf, - 0, 0, pDraw->width, pDraw->height); + /* If this is a full buffer swap, throttle on the previous one */ + if (dst_priv->fence && REGION_NUM_RECTS(pRegion) == 1) { + BoxPtr extents = REGION_EXTENTS(pScreen, pRegion); - pipe_surface_reference(&dst_surf, NULL); - pipe_surface_reference(&src_surf, NULL); + if (extents->x1 == 0 && extents->y1 == 0 && + extents->x2 == pDraw->width && extents->y2 == pDraw->height) { + ms->screen->fence_finish(ms->screen, dst_priv->fence, 0); + ms->screen->fence_reference(ms->screen, &dst_priv->fence, NULL); + } + } + + (*gc->ops->CopyArea)(&src_pixmap->drawable, &dst_pixmap->drawable, gc, + 0, 0, pDraw->width, pDraw->height, 0, 0); + + FreeScratchGC(gc); + + ms->ctx->flush(ms->ctx, PIPE_FLUSH_SWAPBUFFERS, + pDestBuffer->attachment == DRI2BufferFrontLeft ? + &dst_priv->fence : NULL); } Bool @@ -186,18 +338,34 @@ driScreenInit(ScreenPtr pScreen) modesettingPtr ms = modesettingPTR(pScrn); DRI2InfoRec dri2info; - dri2info.version = 1; - dri2info.fd = ms->fd; -#if 0 - dri2info.driverName = pScrn->name; +#if defined(DRI2INFOREC_VERSION) + dri2info.version = DRI2INFOREC_VERSION; #else - dri2info.driverName = "i915"; /* FIXME */ + dri2info.version = 1; #endif + dri2info.fd = ms->fd; + + dri2info.driverName = pScrn->driverName; dri2info.deviceName = "/dev/dri/card0"; /* FIXME */ +#if defined(DRI2INFOREC_VERSION) && DRI2INFOREC_VERSION > 2 + dri2info.CreateBuffer = driCreateBuffer; + dri2info.DestroyBuffer = driDestroyBuffer; +#else dri2info.CreateBuffers = driCreateBuffers; dri2info.DestroyBuffers = driDestroyBuffers; +#endif dri2info.CopyRegion = driCopyRegion; + dri2info.Wait = NULL; + + ms->d_depth_bits_last = + ms->screen->is_format_supported(ms->screen, PIPE_FORMAT_X8Z24_UNORM, + PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0); + ms->ds_depth_bits_last = + ms->screen->is_format_supported(ms->screen, PIPE_FORMAT_S8Z24_UNORM, + PIPE_TEXTURE_2D, + PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0); return DRI2ScreenInit(pScreen, &dri2info); } diff --git a/src/gallium/state_trackers/xorg/xorg_driver.c b/src/gallium/state_trackers/xorg/xorg_driver.c index 45e831f0c28..643b6b3b9e4 100644 --- a/src/gallium/state_trackers/xorg/xorg_driver.c +++ b/src/gallium/state_trackers/xorg/xorg_driver.c @@ -33,10 +33,8 @@ #include "xf86.h" #include "xf86_OSproc.h" #include "compiler.h" -#include "xf86RAC.h" #include "xf86PciInfo.h" #include "xf86Pci.h" -#include "xf86Resources.h" #include "mipointer.h" #include "micmap.h" #include <X11/extensions/randr.h> @@ -54,6 +52,7 @@ #include <pciaccess.h> +#include "pipe/p_context.h" #include "xorg_tracker.h" #include "xorg_winsys.h" @@ -84,41 +83,9 @@ static const OptionInfoRec Options[] = { }; /* - * Functions that might be needed - */ - -static const char *exaSymbols[] = { - "exaGetVersion", - "exaDriverInit", - "exaDriverFini", - "exaOffscreenAlloc", - "exaOffscreenFree", - "exaWaitSync", - NULL -}; - -static const char *fbSymbols[] = { - "fbPictureInit", - "fbScreenInit", - NULL -}; - -static const char *ddcSymbols[] = { - "xf86PrintEDID", - "xf86SetDDCproperties", - NULL -}; - -/* * Exported Xorg driver functions to winsys */ -void -xorg_tracker_loader_ref_sym_lists() -{ - LoaderRefSymLists(exaSymbols, fbSymbols, ddcSymbols, NULL); -} - const OptionInfoRec * xorg_tracker_available_options(int chipid, int busid) { @@ -179,8 +146,10 @@ CreateFrontBuffer(ScrnInfoPtr pScrn) modesettingPtr ms = modesettingPTR(pScrn); ScreenPtr pScreen = pScrn->pScreen; PixmapPtr rootPixmap = pScreen->GetScreenPixmap(pScreen); + unsigned handle, stride; ms->noEvict = TRUE; + xorg_exa_set_displayed_usage(rootPixmap); pScreen->ModifyPixmapHeader(rootPixmap, pScrn->virtualX, pScrn->virtualY, pScrn->depth, pScrn->bitsPerPixel, @@ -188,13 +157,16 @@ CreateFrontBuffer(ScrnInfoPtr pScrn) NULL); ms->noEvict = FALSE; + handle = xorg_exa_get_pixmap_handle(rootPixmap, &stride); + drmModeAddFB(ms->fd, pScrn->virtualX, pScrn->virtualY, pScrn->depth, pScrn->bitsPerPixel, - pScrn->displayWidth * pScrn->bitsPerPixel / 8, - xorg_exa_get_pixmap_handle(rootPixmap), &ms->fb_id); + stride, + handle, + &ms->fb_id); pScrn->frameX0 = 0; pScrn->frameY0 = 0; @@ -282,10 +254,6 @@ PreInit(ScrnInfoPtr pScrn, int flags) } else ms->entityPrivate = NULL; - if (xf86RegisterResources(ms->pEnt->index, NULL, ResNone)) { - return FALSE; - } - if (xf86IsEntityShared(pScrn->entityList[0])) { if (xf86IsPrimInitDone(pScrn->entityList[0])) { /* do something */ @@ -300,12 +268,12 @@ PreInit(ScrnInfoPtr pScrn, int flags) ms->PciInfo->dev, ms->PciInfo->func ); + ms->api = drm_api_create(); ms->fd = drmOpen(NULL, BusID); if (ms->fd < 0) return FALSE; - pScrn->racMemFlags = RAC_FB | RAC_COLORMAP; pScrn->monitor = pScrn->confScreen->monitor; pScrn->progClock = TRUE; pScrn->rgbBits = 8; @@ -391,8 +359,6 @@ PreInit(ScrnInfoPtr pScrn, int flags) return FALSE; } - xf86LoaderReqSymLists(fbSymbols, NULL); - xf86LoadSubModule(pScrn, "exa"); #ifdef DRI2 @@ -418,6 +384,44 @@ RestoreHWState(ScrnInfoPtr pScrn) return TRUE; } +static void xorgBlockHandler(int i, pointer blockData, pointer pTimeout, + pointer pReadmask) +{ + ScreenPtr pScreen = screenInfo.screens[i]; + modesettingPtr ms = modesettingPTR(xf86Screens[pScreen->myNum]); + + pScreen->BlockHandler = ms->blockHandler; + pScreen->BlockHandler(i, blockData, pTimeout, pReadmask); + pScreen->BlockHandler = xorgBlockHandler; + + ms->ctx->flush(ms->ctx, PIPE_FLUSH_RENDER_CACHE, NULL); + +#ifdef DRM_MODE_FEATURE_DIRTYFB + { + RegionPtr dirty = DamageRegion(ms->damage); + unsigned num_cliprects = REGION_NUM_RECTS(dirty); + + if (num_cliprects) { + drmModeClip *clip = alloca(num_cliprects * sizeof(drmModeClip)); + BoxPtr rect = REGION_RECTS(dirty); + int i; + + for (i = 0; i < num_cliprects; i++, rect++) { + clip[i].x = rect->x1; + clip[i].y = rect->y1; + clip[i].width = rect->x2 - rect->x1; + clip[i].height = rect->y2 - rect->y1; + } + + /* TODO query connector property to see if this is needed */ + drmModeDirtyFB(ms->fd, ms->fb_id, clip, num_cliprects); + + DamageEmpty(ms->damage); + } + } +#endif +} + static Bool CreateScreenResources(ScreenPtr pScreen) { @@ -425,6 +429,7 @@ CreateScreenResources(ScreenPtr pScreen) modesettingPtr ms = modesettingPTR(pScrn); PixmapPtr rootPixmap; Bool ret; + unsigned handle, stride; ms->noEvict = TRUE; @@ -434,21 +439,41 @@ CreateScreenResources(ScreenPtr pScreen) rootPixmap = pScreen->GetScreenPixmap(pScreen); + xorg_exa_set_displayed_usage(rootPixmap); + xorg_exa_set_shared_usage(rootPixmap); if (!pScreen->ModifyPixmapHeader(rootPixmap, -1, -1, -1, -1, -1, NULL)) FatalError("Couldn't adjust screen pixmap\n"); ms->noEvict = FALSE; + handle = xorg_exa_get_pixmap_handle(rootPixmap, &stride); + drmModeAddFB(ms->fd, pScrn->virtualX, pScrn->virtualY, pScrn->depth, pScrn->bitsPerPixel, - pScrn->displayWidth * pScrn->bitsPerPixel / 8, - xorg_exa_get_pixmap_handle(rootPixmap), &ms->fb_id); + stride, + handle, + &ms->fb_id); AdjustFrame(pScrn->scrnIndex, pScrn->frameX0, pScrn->frameY0, 0); +#ifdef DRM_MODE_FEATURE_DIRTYFB + ms->damage = DamageCreate(NULL, NULL, DamageReportNone, TRUE, + pScreen, rootPixmap); + + if (ms->damage) { + DamageRegister(&rootPixmap->drawable, ms->damage); + + xf86DrvMsg(pScrn->scrnIndex, X_INFO, "Damage tracking initialized\n"); + } else { + xf86DrvMsg(pScrn->scrnIndex, X_ERROR, + "Failed to create screen damage record\n"); + return FALSE; + } +#endif + return ret; } @@ -476,7 +501,7 @@ ScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv) } if (!ms->screen) { - ms->screen = drm_api_hooks.create_screen(ms->fd, NULL); + ms->screen = ms->api->create_screen(ms->api, ms->fd, NULL); if (!ms->screen) { FatalError("Could not init pipe_screen\n"); @@ -525,6 +550,8 @@ ScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv) fbPictureInit(pScreen, NULL, 0); + ms->blockHandler = pScreen->BlockHandler; + pScreen->BlockHandler = xorgBlockHandler; ms->createScreenResources = pScreen->CreateScreenResources; pScreen->CreateScreenResources = CreateScreenResources; @@ -593,10 +620,6 @@ FreeScreen(int scrnIndex, int flags) FreeRec(xf86Screens[scrnIndex]); } -/* HACK */ -void -cursor_destroy(xf86CrtcPtr crtc); - static void LeaveVT(int scrnIndex, int flags) { @@ -608,7 +631,7 @@ LeaveVT(int scrnIndex, int flags) for (o = 0; o < config->num_crtc; o++) { xf86CrtcPtr crtc = config->crtc[o]; - cursor_destroy(crtc); + crtc_cursor_destroy(crtc); if (crtc->rotatedPixmap || crtc->rotatedData) { crtc->funcs->shadow_destroy(crtc, crtc->rotatedPixmap, @@ -622,6 +645,10 @@ LeaveVT(int scrnIndex, int flags) RestoreHWState(pScrn); + if (drmDropMaster(ms->fd)) + xf86DrvMsg(pScrn->scrnIndex, X_WARNING, + "drmDropMaster failed: %s\n", strerror(errno)); + pScrn->vtSema = FALSE; } @@ -634,6 +661,17 @@ EnterVT(int scrnIndex, int flags) ScrnInfoPtr pScrn = xf86Screens[scrnIndex]; modesettingPtr ms = modesettingPTR(pScrn); + if (drmSetMaster(ms->fd)) { + if (errno == EINVAL) { + xf86DrvMsg(pScrn->scrnIndex, X_WARNING, + "drmSetMaster failed: 2.6.29 or newer kernel required for " + "multi-server DRI\n"); + } else { + xf86DrvMsg(pScrn->scrnIndex, X_WARNING, + "drmSetMaster failed: %s\n", strerror(errno)); + } + } + /* * Only save state once per server generation since that's what most * drivers do. Could change this to save state at each VT enter. @@ -673,11 +711,22 @@ CloseScreen(int scrnIndex, ScreenPtr pScreen) driCloseScreen(pScreen); #endif + pScreen->BlockHandler = ms->blockHandler; pScreen->CreateScreenResources = ms->createScreenResources; +#ifdef DRM_MODE_FEATURE_DIRTYFB + if (ms->damage) { + DamageUnregister(&pScreen->GetScreenPixmap(pScreen)->drawable, ms->damage); + DamageDestroy(ms->damage); + ms->damage = NULL; + } +#endif + if (ms->exa) xorg_exa_close(pScrn); + ms->api->destroy(ms->api); + ms->api = NULL; drmClose(ms->fd); ms->fd = -1; diff --git a/src/gallium/state_trackers/xorg/xorg_exa.c b/src/gallium/state_trackers/xorg/xorg_exa.c index 79131743542..a17a71f23a8 100644 --- a/src/gallium/state_trackers/xorg/xorg_exa.c +++ b/src/gallium/state_trackers/xorg/xorg_exa.c @@ -28,34 +28,24 @@ * */ -#include "xorg-server.h" -#include "xf86.h" +#include "xorg_exa.h" #include "xorg_tracker.h" +#include "xorg_composite.h" +#include "xorg_exa_tgsi.h" + +#include <xorg-server.h> +#include <xf86.h> +#include <picturestr.h> +#include <picture.h> #include "pipe/p_format.h" #include "pipe/p_context.h" #include "pipe/p_state.h" #include "pipe/p_inlines.h" -#include "util/u_rect.h" - -struct exa_entity -{ - ExaDriverPtr pExa; - struct pipe_context *ctx; - struct pipe_screen *scrn; -}; - -struct PixmapPriv -{ - int flags; - struct pipe_texture *tex; - unsigned int color; - struct pipe_surface *src_surf; /* for copies */ +#include "cso_cache/cso_context.h" - struct pipe_transfer *map_transfer; - unsigned map_count; -}; +#include "util/u_rect.h" /* * Helper functions @@ -108,13 +98,75 @@ ExaMarkSync(ScreenPtr pScreen) } static Bool +ExaDownloadFromScreen(PixmapPtr pPix, int x, int y, int w, int h, char *dst, + int dst_pitch) +{ + ScreenPtr pScreen = pPix->drawable.pScreen; + ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum]; + modesettingPtr ms = modesettingPTR(pScrn); + struct exa_context *exa = ms->exa; + struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pPix); + struct pipe_transfer *transfer; + + if (!priv || !priv->tex) + return FALSE; + + if (exa->ctx->is_texture_referenced(exa->ctx, priv->tex, 0, 0) & + PIPE_REFERENCED_FOR_WRITE) + exa->ctx->flush(exa->ctx, 0, NULL); + + transfer = exa->scrn->get_tex_transfer(exa->scrn, priv->tex, 0, 0, 0, + PIPE_TRANSFER_READ, x, y, w, h); + if (!transfer) + return FALSE; + + util_copy_rect((unsigned char*)dst, &priv->tex->block, dst_pitch, 0, 0, + w, h, exa->scrn->transfer_map(exa->scrn, transfer), + transfer->stride, 0, 0); + + exa->scrn->transfer_unmap(exa->scrn, transfer); + exa->scrn->tex_transfer_destroy(transfer); + + return TRUE; +} + +static Bool +ExaUploadToScreen(PixmapPtr pPix, int x, int y, int w, int h, char *src, + int src_pitch) +{ + ScreenPtr pScreen = pPix->drawable.pScreen; + ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum]; + modesettingPtr ms = modesettingPTR(pScrn); + struct exa_context *exa = ms->exa; + struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pPix); + struct pipe_transfer *transfer; + + if (!priv || !priv->tex) + return FALSE; + + transfer = exa->scrn->get_tex_transfer(exa->scrn, priv->tex, 0, 0, 0, + PIPE_TRANSFER_WRITE, x, y, w, h); + if (!transfer) + return FALSE; + + util_copy_rect(exa->scrn->transfer_map(exa->scrn, transfer), + &priv->tex->block, transfer->stride, 0, 0, w, h, + (unsigned char*)src, src_pitch, 0, 0); + + exa->scrn->transfer_unmap(exa->scrn, transfer); + exa->scrn->tex_transfer_destroy(transfer); + + return TRUE; +} + +static Bool ExaPrepareAccess(PixmapPtr pPix, int index) { ScreenPtr pScreen = pPix->drawable.pScreen; ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum]; modesettingPtr ms = modesettingPTR(pScrn); - struct exa_entity *exa = ms->exa; - struct PixmapPriv *priv; + struct exa_context *exa = ms->exa; + struct exa_pixmap_priv *priv; priv = exaGetPixmapDriverPrivate(pPix); @@ -126,6 +178,10 @@ ExaPrepareAccess(PixmapPtr pPix, int index) if (priv->map_count++ == 0) { + if (exa->ctx->is_texture_referenced(exa->ctx, priv->tex, 0, 0) & + PIPE_REFERENCED_FOR_WRITE) + exa->ctx->flush(exa->ctx, 0, NULL); + priv->map_transfer = exa->scrn->get_tex_transfer(exa->scrn, priv->tex, 0, 0, 0, PIPE_TRANSFER_READ_WRITE, @@ -145,8 +201,8 @@ ExaFinishAccess(PixmapPtr pPix, int index) ScreenPtr pScreen = pPix->drawable.pScreen; ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum]; modesettingPtr ms = modesettingPTR(pScrn); - struct exa_entity *exa = ms->exa; - struct PixmapPriv *priv; + struct exa_context *exa = ms->exa; + struct exa_pixmap_priv *priv; priv = exaGetPixmapDriverPrivate(pPix); if (!priv) @@ -160,6 +216,7 @@ ExaFinishAccess(PixmapPtr pPix, int index) exa->scrn->transfer_unmap(exa->scrn, priv->map_transfer); exa->scrn->tex_transfer_destroy(priv->map_transfer); priv->map_transfer = NULL; + pPix->devPrivate.ptr = NULL; } } @@ -168,8 +225,8 @@ ExaDone(PixmapPtr pPixmap) { ScrnInfoPtr pScrn = xf86Screens[pPixmap->drawable.pScreen->myNum]; modesettingPtr ms = modesettingPTR(pScrn); - struct PixmapPriv *priv = exaGetPixmapDriverPrivate(pPixmap); - struct exa_entity *exa = ms->exa; + struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pPixmap); + struct exa_context *exa = ms->exa; if (!priv) return; @@ -190,11 +247,8 @@ ExaPrepareSolid(PixmapPtr pPixmap, int alu, Pixel planeMask, Pixel fg) { ScrnInfoPtr pScrn = xf86Screens[pPixmap->drawable.pScreen->myNum]; modesettingPtr ms = modesettingPTR(pScrn); - struct PixmapPriv *priv = exaGetPixmapDriverPrivate(pPixmap); - struct exa_entity *exa = ms->exa; - - if (1) - return FALSE; + struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pPixmap); + struct exa_context *exa = ms->exa; if (pPixmap->drawable.depth < 15) return FALSE; @@ -221,12 +275,9 @@ ExaSolid(PixmapPtr pPixmap, int x0, int y0, int x1, int y1) { ScrnInfoPtr pScrn = xf86Screens[pPixmap->drawable.pScreen->myNum]; modesettingPtr ms = modesettingPTR(pScrn); - struct exa_entity *exa = ms->exa; - struct PixmapPriv *priv = exaGetPixmapDriverPrivate(pPixmap); - struct pipe_surface *surf = - exa->scrn->get_tex_surface(exa->scrn, priv->tex, 0, 0, 0, - PIPE_BUFFER_USAGE_GPU_READ | - PIPE_BUFFER_USAGE_GPU_WRITE); + struct exa_context *exa = ms->exa; + struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pPixmap); + struct pipe_surface *surf = exa_gpu_surface(exa, priv); exa->ctx->surface_fill(exa->ctx, surf, x0, y0, x1 - x0, y1 - y0, priv->color); @@ -240,12 +291,9 @@ ExaPrepareCopy(PixmapPtr pSrcPixmap, PixmapPtr pDstPixmap, int xdir, { ScrnInfoPtr pScrn = xf86Screens[pDstPixmap->drawable.pScreen->myNum]; modesettingPtr ms = modesettingPTR(pScrn); - struct exa_entity *exa = ms->exa; - struct PixmapPriv *priv = exaGetPixmapDriverPrivate(pDstPixmap); - struct PixmapPriv *src_priv = exaGetPixmapDriverPrivate(pSrcPixmap); - - if (1) - return FALSE; + struct exa_context *exa = ms->exa; + struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pDstPixmap); + struct exa_pixmap_priv *src_priv = exaGetPixmapDriverPrivate(pSrcPixmap); if (alu != GXcopy) return FALSE; @@ -265,10 +313,7 @@ ExaPrepareCopy(PixmapPtr pSrcPixmap, PixmapPtr pDstPixmap, int xdir, if (!exa->ctx || !exa->ctx->surface_copy) return FALSE; - priv->src_surf = - exa->scrn->get_tex_surface(exa->scrn, src_priv->tex, 0, 0, 0, - PIPE_BUFFER_USAGE_GPU_READ | - PIPE_BUFFER_USAGE_GPU_WRITE); + priv->src_surf = exa_gpu_surface(exa, src_priv); return TRUE; } @@ -279,12 +324,9 @@ ExaCopy(PixmapPtr pDstPixmap, int srcX, int srcY, int dstX, int dstY, { ScrnInfoPtr pScrn = xf86Screens[pDstPixmap->drawable.pScreen->myNum]; modesettingPtr ms = modesettingPTR(pScrn); - struct exa_entity *exa = ms->exa; - struct PixmapPriv *priv = exaGetPixmapDriverPrivate(pDstPixmap); - struct pipe_surface *surf = - exa->scrn->get_tex_surface(exa->scrn, priv->tex, 0, 0, 0, - PIPE_BUFFER_USAGE_GPU_READ | - PIPE_BUFFER_USAGE_GPU_WRITE); + struct exa_context *exa = ms->exa; + struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pDstPixmap); + struct pipe_surface *surf = exa_gpu_surface(exa, priv); exa->ctx->surface_copy(exa->ctx, surf, dstX, dstY, priv->src_surf, srcX, srcY, width, height); @@ -296,13 +338,28 @@ ExaPrepareComposite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture, PicturePtr pDstPicture, PixmapPtr pSrc, PixmapPtr pMask, PixmapPtr pDst) { - return FALSE; + ScrnInfoPtr pScrn = xf86Screens[pDst->drawable.pScreen->myNum]; + modesettingPtr ms = modesettingPTR(pScrn); + struct exa_context *exa = ms->exa; + + return xorg_composite_bind_state(exa, op, pSrcPicture, pMaskPicture, + pDstPicture, + exaGetPixmapDriverPrivate(pSrc), + exaGetPixmapDriverPrivate(pMask), + exaGetPixmapDriverPrivate(pDst)); } static void ExaComposite(PixmapPtr pDst, int srcX, int srcY, int maskX, int maskY, int dstX, int dstY, int width, int height) { + ScrnInfoPtr pScrn = xf86Screens[pDst->drawable.pScreen->myNum]; + modesettingPtr ms = modesettingPTR(pScrn); + struct exa_context *exa = ms->exa; + struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pDst); + + xorg_composite(exa, priv, srcX, srcY, maskX, maskY, + dstX, dstY, width, height); } static Bool @@ -310,15 +367,18 @@ ExaCheckComposite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture, PicturePtr pDstPicture) { - return FALSE; + return xorg_composite_accelerated(op, + pSrcPicture, + pMaskPicture, + pDstPicture); } static void * ExaCreatePixmap(ScreenPtr pScreen, int size, int align) { - struct PixmapPriv *priv; + struct exa_pixmap_priv *priv; - priv = xcalloc(1, sizeof(struct PixmapPriv)); + priv = xcalloc(1, sizeof(struct exa_pixmap_priv)); if (!priv) return NULL; @@ -328,7 +388,7 @@ ExaCreatePixmap(ScreenPtr pScreen, int size, int align) static void ExaDestroyPixmap(ScreenPtr pScreen, void *dPriv) { - struct PixmapPriv *priv = (struct PixmapPriv *)dPriv; + struct exa_pixmap_priv *priv = (struct exa_pixmap_priv *)dPriv; ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum]; modesettingPtr ms = modesettingPTR(pScrn); @@ -344,7 +404,7 @@ ExaDestroyPixmap(ScreenPtr pScreen, void *dPriv) static Bool ExaPixmapIsOffscreen(PixmapPtr pPixmap) { - struct PixmapPriv *priv; + struct exa_pixmap_priv *priv; priv = exaGetPixmapDriverPrivate(pPixmap); @@ -357,14 +417,45 @@ ExaPixmapIsOffscreen(PixmapPtr pPixmap) return FALSE; } +int +xorg_exa_set_displayed_usage(PixmapPtr pPixmap) +{ + struct exa_pixmap_priv *priv; + priv = exaGetPixmapDriverPrivate(pPixmap); + + if (!priv) { + FatalError("NO PIXMAP PRIVATE\n"); + return 0; + } + + priv->flags |= PIPE_TEXTURE_USAGE_PRIMARY; + + return 0; +} + +int +xorg_exa_set_shared_usage(PixmapPtr pPixmap) +{ + struct exa_pixmap_priv *priv; + priv = exaGetPixmapDriverPrivate(pPixmap); + + if (!priv) { + FatalError("NO PIXMAP PRIVATE\n"); + return 0; + } + + priv->flags |= PIPE_TEXTURE_USAGE_DISPLAY_TARGET; + + return 0; +} + unsigned -xorg_exa_get_pixmap_handle(PixmapPtr pPixmap) +xorg_exa_get_pixmap_handle(PixmapPtr pPixmap, unsigned *stride_out) { ScreenPtr pScreen = pPixmap->drawable.pScreen; ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum]; modesettingPtr ms = modesettingPTR(pScrn); - struct PixmapPriv *priv; - struct pipe_buffer *buffer = NULL; + struct exa_pixmap_priv *priv; unsigned handle; unsigned stride; @@ -380,9 +471,10 @@ xorg_exa_get_pixmap_handle(PixmapPtr pPixmap) return 0; } - drm_api_hooks.buffer_from_texture(priv->tex, &buffer, &stride); - drm_api_hooks.handle_from_buffer(ms->screen, buffer, &handle); - pipe_buffer_reference(&buffer, NULL); + ms->api->local_handle_from_texture(ms->api, ms->screen, priv->tex, &stride, &handle); + if (stride_out) + *stride_out = stride; + return handle; } @@ -393,11 +485,11 @@ ExaModifyPixmapHeader(PixmapPtr pPixmap, int width, int height, { ScreenPtr pScreen = pPixmap->drawable.pScreen; ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum]; - struct PixmapPriv *priv = exaGetPixmapDriverPrivate(pPixmap); + struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pPixmap); modesettingPtr ms = modesettingPTR(pScrn); - struct exa_entity *exa = ms->exa; + struct exa_context *exa = ms->exa; - if (!priv) + if (!priv || pPixData) return FALSE; if (depth <= 0) @@ -419,35 +511,69 @@ ExaModifyPixmapHeader(PixmapPtr pPixmap, int width, int height, bitsPerPixel, devKind, NULL); /* Deal with screen resize */ - if (priv->tex && (priv->tex->width[0] != width || priv->tex->height[0] != height)) { - pipe_texture_reference(&priv->tex, NULL); - } - - if (!priv->tex) { - struct pipe_texture template; - - memset(&template, 0, sizeof(template)); - template.target = PIPE_TEXTURE_2D; - exa_get_pipe_format(depth, &template.format, &bitsPerPixel); - pf_get_block(template.format, &template.block); - template.width[0] = width; - template.height[0] = height; - template.depth[0] = 1; - template.last_level = 0; - template.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET; - priv->tex = exa->scrn->texture_create(exa->scrn, &template); - } - - if (pPixData) { - struct pipe_transfer *transfer = - exa->scrn->get_tex_transfer(exa->scrn, priv->tex, 0, 0, 0, - PIPE_TRANSFER_WRITE, - 0, 0, width, height); - pipe_copy_rect(exa->scrn->transfer_map(exa->scrn, transfer), - &priv->tex->block, transfer->stride, 0, 0, - width, height, pPixData, pPixmap->devKind, 0, 0); - exa->scrn->transfer_unmap(exa->scrn, transfer); - exa->scrn->tex_transfer_destroy(transfer); + if (!priv->tex || + (priv->tex->width[0] != width || + priv->tex->height[0] != height || + priv->tex_flags != priv->flags)) { + struct pipe_texture *texture = NULL; + +#ifdef DRM_MODE_FEATURE_DIRTYFB + if (priv->flags) +#endif + { + struct pipe_texture template; + + memset(&template, 0, sizeof(template)); + template.target = PIPE_TEXTURE_2D; + exa_get_pipe_format(depth, &template.format, &bitsPerPixel); + pf_get_block(template.format, &template.block); + template.width[0] = width; + template.height[0] = height; + template.depth[0] = 1; + template.last_level = 0; + template.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET | priv->flags; + priv->tex_flags = priv->flags; + texture = exa->scrn->texture_create(exa->scrn, &template); + + if (priv->tex) { + struct pipe_surface *dst_surf; + + dst_surf = exa->scrn->get_tex_surface(exa->scrn, texture, 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_WRITE); + priv->src_surf = exa_gpu_surface(exa, priv); + exa->ctx->surface_copy(exa->ctx, dst_surf, 0, 0, priv->src_surf, + 0, 0, min(width, texture->width[0]), + min(height, texture->height[0])); + exa->scrn->tex_surface_destroy(dst_surf); + exa->scrn->tex_surface_destroy(priv->src_surf); + priv->src_surf = NULL; + } else if (pPixmap->devPrivate.ptr) { + struct pipe_transfer *transfer; + + if (priv->map_count != 0) + FatalError("doing ExaModifyPixmapHeader on mapped buffer\n"); + + transfer = + exa->scrn->get_tex_transfer(exa->scrn, texture, 0, 0, 0, + PIPE_TRANSFER_WRITE, + 0, 0, width, height); + util_copy_rect(exa->scrn->transfer_map(exa->scrn, transfer), + &texture->block, transfer->stride, 0, 0, + width, height, pPixmap->devPrivate.ptr, + pPixmap->devKind, 0, 0); + exa->scrn->transfer_unmap(exa->scrn, transfer); + exa->scrn->tex_transfer_destroy(transfer); + } + } +#ifdef DRM_MODE_FEATURE_DIRTYFB + else { + xfree(pPixmap->devPrivate.ptr); + pPixmap->devPrivate.ptr = xalloc(pPixmap->drawable.height * + pPixmap->devKind); + } +#endif + + pipe_texture_reference(&priv->tex, texture); } return TRUE; @@ -456,87 +582,119 @@ ExaModifyPixmapHeader(PixmapPtr pPixmap, int width, int height, struct pipe_texture * xorg_exa_get_texture(PixmapPtr pPixmap) { - struct PixmapPriv *priv = exaGetPixmapDriverPrivate(pPixmap); - struct pipe_texture *tex = NULL; - pipe_texture_reference(&tex, priv->tex); - return tex; + struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pPixmap); + struct pipe_texture *tex = NULL; + pipe_texture_reference(&tex, priv->tex); + return tex; } void xorg_exa_close(ScrnInfoPtr pScrn) { - modesettingPtr ms = modesettingPTR(pScrn); - struct exa_entity *exa = ms->exa; - - if (exa->ctx) - exa->ctx->destroy(exa->ctx); + modesettingPtr ms = modesettingPTR(pScrn); + struct exa_context *exa = ms->exa; + struct pipe_constant_buffer *vsbuf = &exa->vs_const_buffer; + struct pipe_constant_buffer *fsbuf = &exa->fs_const_buffer; - exaDriverFini(pScrn->pScreen); - xfree(exa); - ms->exa = NULL; -} + if (exa->shaders) { + xorg_shaders_destroy(exa->shaders); + } -void * -xorg_exa_init(ScrnInfoPtr pScrn) -{ - modesettingPtr ms = modesettingPTR(pScrn); - struct exa_entity *exa; - ExaDriverPtr pExa; + if (vsbuf && vsbuf->buffer) + pipe_buffer_reference(&vsbuf->buffer, NULL); - exa = xcalloc(1, sizeof(struct exa_entity)); - if (!exa) - return NULL; + if (fsbuf && fsbuf->buffer) + pipe_buffer_reference(&fsbuf->buffer, NULL); - pExa = exaDriverAlloc(); - if (!pExa) { - goto out_err; - } + if (exa->cso) { + cso_release_all(exa->cso); + cso_destroy_context(exa->cso); + } - memset(pExa, 0, sizeof(*pExa)); - pExa->exa_major = 2; - pExa->exa_minor = 2; - pExa->memoryBase = 0; - pExa->memorySize = 0; - pExa->offScreenBase = 0; - pExa->pixmapOffsetAlign = 0; - pExa->pixmapPitchAlign = 1; - pExa->flags = EXA_OFFSCREEN_PIXMAPS | EXA_HANDLES_PIXMAPS; - pExa->maxX = 8191; /* FIXME */ - pExa->maxY = 8191; /* FIXME */ - pExa->WaitMarker = ExaWaitMarker; - pExa->MarkSync = ExaMarkSync; - pExa->PrepareSolid = ExaPrepareSolid; - pExa->Solid = ExaSolid; - pExa->DoneSolid = ExaDone; - pExa->PrepareCopy = ExaPrepareCopy; - pExa->Copy = ExaCopy; - pExa->DoneCopy = ExaDone; - pExa->CheckComposite = ExaCheckComposite; - pExa->PrepareComposite = ExaPrepareComposite; - pExa->Composite = ExaComposite; - pExa->DoneComposite = ExaDoneComposite; - pExa->PixmapIsOffscreen = ExaPixmapIsOffscreen; - pExa->PrepareAccess = ExaPrepareAccess; - pExa->FinishAccess = ExaFinishAccess; - pExa->CreatePixmap = ExaCreatePixmap; - pExa->DestroyPixmap = ExaDestroyPixmap; - pExa->ModifyPixmapHeader = ExaModifyPixmapHeader; - - if (!exaDriverInit(pScrn->pScreen, pExa)) { - goto out_err; - } + if (exa->ctx) + exa->ctx->destroy(exa->ctx); - exa->scrn = ms->screen; - exa->ctx = drm_api_hooks.create_context(exa->scrn); - /* Share context with DRI */ - ms->ctx = exa->ctx; + exaDriverFini(pScrn->pScreen); + xfree(exa); + ms->exa = NULL; +} - return (void *)exa; +void * +xorg_exa_init(ScrnInfoPtr pScrn) +{ + modesettingPtr ms = modesettingPTR(pScrn); + struct exa_context *exa; + ExaDriverPtr pExa; + + exa = xcalloc(1, sizeof(struct exa_context)); + if (!exa) + return NULL; + + pExa = exaDriverAlloc(); + if (!pExa) { + goto out_err; + } + + memset(pExa, 0, sizeof(*pExa)); + + pExa->exa_major = 2; + pExa->exa_minor = 2; + pExa->memoryBase = 0; + pExa->memorySize = 0; + pExa->offScreenBase = 0; + pExa->pixmapOffsetAlign = 0; + pExa->pixmapPitchAlign = 1; + pExa->flags = EXA_OFFSCREEN_PIXMAPS | EXA_HANDLES_PIXMAPS; + pExa->maxX = 8191; /* FIXME */ + pExa->maxY = 8191; /* FIXME */ + + pExa->WaitMarker = ExaWaitMarker; + pExa->MarkSync = ExaMarkSync; + pExa->PrepareSolid = ExaPrepareSolid; + pExa->Solid = ExaSolid; + pExa->DoneSolid = ExaDone; + pExa->PrepareCopy = ExaPrepareCopy; + pExa->Copy = ExaCopy; + pExa->DoneCopy = ExaDone; + pExa->CheckComposite = ExaCheckComposite; + pExa->PrepareComposite = ExaPrepareComposite; + pExa->Composite = ExaComposite; + pExa->DoneComposite = ExaDoneComposite; + pExa->PixmapIsOffscreen = ExaPixmapIsOffscreen; + pExa->DownloadFromScreen = ExaDownloadFromScreen; + pExa->UploadToScreen = ExaUploadToScreen; + pExa->PrepareAccess = ExaPrepareAccess; + pExa->FinishAccess = ExaFinishAccess; + pExa->CreatePixmap = ExaCreatePixmap; + pExa->DestroyPixmap = ExaDestroyPixmap; + pExa->ModifyPixmapHeader = ExaModifyPixmapHeader; + + if (!exaDriverInit(pScrn->pScreen, pExa)) { + goto out_err; + } + + exa->scrn = ms->screen; + exa->ctx = ms->api->create_context(ms->api, exa->scrn); + /* Share context with DRI */ + ms->ctx = exa->ctx; + + exa->cso = cso_create_context(exa->ctx); + exa->shaders = xorg_shaders_create(exa); + + return (void *)exa; + +out_err: + xorg_exa_close(pScrn); + + return NULL; +} - out_err: - xorg_exa_close(pScrn); +struct pipe_surface * +exa_gpu_surface(struct exa_context *exa, struct exa_pixmap_priv *priv) +{ + return exa->scrn->get_tex_surface(exa->scrn, priv->tex, 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_READ | + PIPE_BUFFER_USAGE_GPU_WRITE); - return NULL; } -/* vim: set sw=4 ts=8 sts=4: */ diff --git a/src/gallium/state_trackers/xorg/xorg_exa.h b/src/gallium/state_trackers/xorg/xorg_exa.h new file mode 100644 index 00000000000..5b515be1397 --- /dev/null +++ b/src/gallium/state_trackers/xorg/xorg_exa.h @@ -0,0 +1,50 @@ +#ifndef XORG_EXA_H +#define XORG_EXA_H + +#include "xorg_tracker.h" + +#include <pipe/p_state.h> + +struct cso_context; +struct xorg_shaders; + +/* src + mask + dst */ +#define MAX_EXA_SAMPLERS 3 + +struct exa_context +{ + ExaDriverPtr pExa; + struct pipe_context *ctx; + struct pipe_screen *scrn; + struct cso_context *cso; + struct xorg_shaders *shaders; + + struct pipe_constant_buffer vs_const_buffer; + struct pipe_constant_buffer fs_const_buffer; + + struct pipe_texture *bound_textures[MAX_EXA_SAMPLERS]; + int num_bound_samplers; + + float solid_color[4]; +}; + + +struct exa_pixmap_priv +{ + int flags; + int tex_flags; + + struct pipe_texture *tex; + struct pipe_texture *depth_stencil_tex; + unsigned int color; + struct pipe_surface *src_surf; /* for copies */ + + struct pipe_transfer *map_transfer; + unsigned map_count; +}; + +struct pipe_surface * +exa_gpu_surface(struct exa_context *exa, struct exa_pixmap_priv *priv); + + +#endif diff --git a/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c b/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c new file mode 100644 index 00000000000..cfee10c3b30 --- /dev/null +++ b/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c @@ -0,0 +1,489 @@ +#include "xorg_exa_tgsi.h" + +/*### stupidity defined in X11/extensions/XI.h */ +#undef Absolute + +#include "pipe/p_format.h" +#include "pipe/p_context.h" +#include "pipe/p_state.h" +#include "pipe/p_inlines.h" +#include "pipe/p_shader_tokens.h" + +#include "util/u_memory.h" +#include "util/u_simple_shaders.h" + +#include "tgsi/tgsi_ureg.h" + +#include "cso_cache/cso_context.h" +#include "cso_cache/cso_hash.h" + +/* Vertex shader: + * IN[0] = vertex pos + * IN[1] = src tex coord | solid fill color + * IN[2] = mask tex coord + * IN[3] = dst tex coord + * CONST[0] = (2/dst_width, 2/dst_height, 1, 1) + * CONST[1] = (-1, -1, 0, 0) + * + * OUT[0] = vertex pos + * OUT[1] = src tex coord | solid fill color + * OUT[2] = mask tex coord + * OUT[3] = dst tex coord + */ + +/* Fragment shader: + * SAMP[0] = src + * SAMP[1] = mask + * SAMP[2] = dst + * IN[0] = pos src | solid fill color + * IN[1] = pos mask + * IN[2] = pos dst + * CONST[0] = (0, 0, 0, 1) + * + * OUT[0] = color + */ + +struct xorg_shaders { + struct exa_context *exa; + + struct cso_hash *vs_hash; + struct cso_hash *fs_hash; +}; + +static const char over_op[] = + "SUB TEMP[3], CONST[0].wwww, TEMP[1].wwww\n" + "MUL TEMP[3], TEMP[0], TEMP[3]\n" + "ADD TEMP[0], TEMP[3], TEMP[0]\n"; + + +static INLINE void +create_preamble(struct ureg_program *ureg) +{ +} + + +static INLINE void +src_in_mask(struct ureg_program *ureg, + struct ureg_dst dst, + struct ureg_src src, + struct ureg_src mask) +{ + /* MUL dst, src, mask.wwww */ + ureg_MUL(ureg, dst, src, + ureg_scalar(mask, TGSI_SWIZZLE_W)); +} + +static struct ureg_src +vs_normalize_coords(struct ureg_program *ureg, struct ureg_src coords, + struct ureg_src const0, struct ureg_src const1) +{ + struct ureg_dst tmp = ureg_DECL_temporary(ureg); + struct ureg_src ret; + ureg_MUL(ureg, tmp, coords, const0); + ureg_ADD(ureg, tmp, ureg_src(tmp), const1); + ret = ureg_src(tmp); + ureg_release_temporary(ureg, tmp); + return ret; +} + +static void +linear_gradient(struct ureg_program *ureg, + struct ureg_dst out, + struct ureg_src pos, + struct ureg_src sampler, + struct ureg_src coords, + struct ureg_src const0124, + struct ureg_src matrow0, + struct ureg_src matrow1, + struct ureg_src matrow2) +{ + struct ureg_dst temp0 = ureg_DECL_temporary(ureg); + struct ureg_dst temp1 = ureg_DECL_temporary(ureg); + struct ureg_dst temp2 = ureg_DECL_temporary(ureg); + struct ureg_dst temp3 = ureg_DECL_temporary(ureg); + struct ureg_dst temp4 = ureg_DECL_temporary(ureg); + struct ureg_dst temp5 = ureg_DECL_temporary(ureg); + + ureg_MOV(ureg, + ureg_writemask(temp0, TGSI_WRITEMASK_XY), pos); + ureg_MOV(ureg, + ureg_writemask(temp0, TGSI_WRITEMASK_Z), + ureg_scalar(const0124, TGSI_SWIZZLE_Y)); + + ureg_DP3(ureg, temp1, matrow0, ureg_src(temp0)); + ureg_DP3(ureg, temp2, matrow1, ureg_src(temp0)); + ureg_DP3(ureg, temp3, matrow2, ureg_src(temp0)); + ureg_RCP(ureg, temp3, ureg_src(temp3)); + ureg_MUL(ureg, temp1, ureg_src(temp1), ureg_src(temp3)); + ureg_MUL(ureg, temp2, ureg_src(temp2), ureg_src(temp3)); + + ureg_MOV(ureg, ureg_writemask(temp4, TGSI_WRITEMASK_X), + ureg_src(temp1)); + ureg_MOV(ureg, ureg_writemask(temp4, TGSI_WRITEMASK_Y), + ureg_src(temp2)); + + ureg_MUL(ureg, temp0, + ureg_scalar(coords, TGSI_SWIZZLE_Y), + ureg_scalar(ureg_src(temp4), TGSI_SWIZZLE_Y)); + ureg_MAD(ureg, temp1, + ureg_scalar(coords, TGSI_SWIZZLE_X), + ureg_scalar(ureg_src(temp4), TGSI_SWIZZLE_X), + ureg_src(temp0)); + + ureg_MUL(ureg, temp2, + ureg_src(temp1), + ureg_scalar(coords, TGSI_SWIZZLE_Z)); + + ureg_TEX(ureg, out, + TGSI_TEXTURE_1D, ureg_src(temp2), sampler); + + ureg_release_temporary(ureg, temp0); + ureg_release_temporary(ureg, temp1); + ureg_release_temporary(ureg, temp2); + ureg_release_temporary(ureg, temp3); + ureg_release_temporary(ureg, temp4); + ureg_release_temporary(ureg, temp5); +} + + +static void +radial_gradient(struct ureg_program *ureg, + struct ureg_dst out, + struct ureg_src pos, + struct ureg_src sampler, + struct ureg_src coords, + struct ureg_src const0124, + struct ureg_src matrow0, + struct ureg_src matrow1, + struct ureg_src matrow2) +{ + struct ureg_dst temp0 = ureg_DECL_temporary(ureg); + struct ureg_dst temp1 = ureg_DECL_temporary(ureg); + struct ureg_dst temp2 = ureg_DECL_temporary(ureg); + struct ureg_dst temp3 = ureg_DECL_temporary(ureg); + struct ureg_dst temp4 = ureg_DECL_temporary(ureg); + struct ureg_dst temp5 = ureg_DECL_temporary(ureg); + + ureg_MOV(ureg, + ureg_writemask(temp0, TGSI_WRITEMASK_XY), + pos); + ureg_MOV(ureg, + ureg_writemask(temp0, TGSI_WRITEMASK_Z), + ureg_scalar(const0124, TGSI_SWIZZLE_Y)); + + ureg_DP3(ureg, temp1, matrow0, ureg_src(temp0)); + ureg_DP3(ureg, temp2, matrow1, ureg_src(temp0)); + ureg_DP3(ureg, temp3, matrow2, ureg_src(temp0)); + ureg_RCP(ureg, temp3, ureg_src(temp3)); + ureg_MUL(ureg, temp1, ureg_src(temp1), ureg_src(temp3)); + ureg_MUL(ureg, temp2, ureg_src(temp2), ureg_src(temp3)); + + ureg_MOV(ureg, ureg_writemask(temp5, TGSI_WRITEMASK_X), + ureg_src(temp1)); + ureg_MOV(ureg, ureg_writemask(temp5, TGSI_WRITEMASK_Y), + ureg_src(temp2)); + + ureg_MUL(ureg, temp0, ureg_scalar(coords, TGSI_SWIZZLE_Y), + ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_Y)); + ureg_MAD(ureg, temp1, + ureg_scalar(coords, TGSI_SWIZZLE_X), + ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_X), + ureg_src(temp0)); + ureg_ADD(ureg, temp1, + ureg_src(temp1), ureg_src(temp1)); + ureg_MUL(ureg, temp3, + ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_Y), + ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_Y)); + ureg_MAD(ureg, temp4, + ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_X), + ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_X), + ureg_src(temp3)); + ureg_MOV(ureg, temp4, ureg_negate(ureg_src(temp4))); + ureg_MUL(ureg, temp2, + ureg_scalar(coords, TGSI_SWIZZLE_Z), + ureg_src(temp4)); + ureg_MUL(ureg, temp0, + ureg_scalar(const0124, TGSI_SWIZZLE_W), + ureg_src(temp2)); + ureg_MUL(ureg, temp3, + ureg_src(temp1), ureg_src(temp1)); + ureg_SUB(ureg, temp2, + ureg_src(temp3), ureg_src(temp0)); + ureg_RSQ(ureg, temp2, ureg_abs(ureg_src(temp2))); + ureg_RCP(ureg, temp2, ureg_src(temp2)); + ureg_SUB(ureg, temp1, + ureg_src(temp2), ureg_src(temp1)); + ureg_ADD(ureg, temp0, + ureg_scalar(coords, TGSI_SWIZZLE_Z), + ureg_scalar(coords, TGSI_SWIZZLE_Z)); + ureg_RCP(ureg, temp0, ureg_src(temp0)); + ureg_MUL(ureg, temp2, + ureg_src(temp1), ureg_src(temp0)); + ureg_TEX(ureg, out, TGSI_TEXTURE_1D, + ureg_src(temp2), sampler); + + ureg_release_temporary(ureg, temp0); + ureg_release_temporary(ureg, temp1); + ureg_release_temporary(ureg, temp2); + ureg_release_temporary(ureg, temp3); + ureg_release_temporary(ureg, temp4); + ureg_release_temporary(ureg, temp5); +} + +static void * +create_vs(struct pipe_context *pipe, + unsigned vs_traits) +{ + struct ureg_program *ureg; + struct ureg_src src; + struct ureg_dst dst; + struct ureg_src const0, const1; + boolean is_fill = vs_traits & VS_FILL; + boolean is_composite = vs_traits & VS_COMPOSITE; + boolean has_mask = vs_traits & VS_MASK; + + ureg = ureg_create(TGSI_PROCESSOR_VERTEX); + if (ureg == NULL) + return 0; + + const0 = ureg_DECL_constant(ureg); + const1 = ureg_DECL_constant(ureg); + + /* it has to be either a fill or a composite op */ + debug_assert(is_fill ^ is_composite); + + src = ureg_DECL_vs_input(ureg, + TGSI_SEMANTIC_POSITION, 0); + dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0); + src = vs_normalize_coords(ureg, src, + const0, const1); + ureg_MOV(ureg, dst, src); + + + if (is_composite) { + src = ureg_DECL_vs_input(ureg, + TGSI_SEMANTIC_GENERIC, 1); + dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 1); + ureg_MOV(ureg, dst, src); + } + if (is_fill) { + src = ureg_DECL_vs_input(ureg, + TGSI_SEMANTIC_COLOR, 1); + dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 1); + ureg_MOV(ureg, dst, src); + } + + if (has_mask) { + src = ureg_DECL_vs_input(ureg, + TGSI_SEMANTIC_GENERIC, 2); + dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 2); + ureg_MOV(ureg, dst, src); + } + + ureg_END(ureg); + + return ureg_create_shader_and_destroy(ureg, pipe); +} + +static void * +create_fs(struct pipe_context *pipe, + unsigned fs_traits) +{ + struct ureg_program *ureg; + struct ureg_src /*dst_sampler,*/ src_sampler, mask_sampler; + struct ureg_src /*dst_pos,*/ src_input, mask_pos; + struct ureg_dst src, mask; + struct ureg_dst out; + boolean has_mask = fs_traits & FS_MASK; + boolean is_fill = fs_traits & FS_FILL; + boolean is_composite = fs_traits & FS_COMPOSITE; + boolean is_solid = fs_traits & FS_SOLID_FILL; + boolean is_lingrad = fs_traits & FS_LINGRAD_FILL; + boolean is_radgrad = fs_traits & FS_RADGRAD_FILL; + + ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT); + if (ureg == NULL) + return 0; + + /* it has to be either a fill or a composite op */ + debug_assert(is_fill ^ is_composite); + + out = ureg_DECL_output(ureg, + TGSI_SEMANTIC_COLOR, + 0); + + if (is_composite) { + src_sampler = ureg_DECL_sampler(ureg, 0); + src_input = ureg_DECL_fs_input(ureg, + TGSI_SEMANTIC_POSITION, + 0, + TGSI_INTERPOLATE_PERSPECTIVE); + } + if (is_fill) { + if (is_solid) + src_input = ureg_DECL_fs_input(ureg, + TGSI_SEMANTIC_COLOR, + 0, + TGSI_INTERPOLATE_PERSPECTIVE); + else + src_input = ureg_DECL_fs_input(ureg, + TGSI_SEMANTIC_POSITION, + 0, + TGSI_INTERPOLATE_PERSPECTIVE); + } + + if (has_mask) { + mask_sampler = ureg_DECL_sampler(ureg, 1); + mask_pos = ureg_DECL_fs_input(ureg, + TGSI_SEMANTIC_POSITION, + 1, + TGSI_INTERPOLATE_PERSPECTIVE); + } + +#if 0 /* unused right now */ + dst_sampler = ureg_DECL_sampler(ureg, 2); + dst_pos = ureg_DECL_fs_input(ureg, + TGSI_SEMANTIC_POSITION, + 2, + TGSI_INTERPOLATE_PERSPECTIVE); +#endif + + if (is_composite) { + if (has_mask) + src = ureg_DECL_temporary(ureg); + else + src = out; + ureg_TEX(ureg, src, + TGSI_TEXTURE_2D, src_input, src_sampler); + } else if (is_fill) { + if (is_solid) { + if (has_mask) + src = ureg_dst(src_input); + else + ureg_MOV(ureg, out, src_input); + } else if (is_lingrad || is_radgrad) { + struct ureg_src coords, const0124, + matrow0, matrow1, matrow2; + + if (has_mask) + src = ureg_DECL_temporary(ureg); + else + src = out; + + coords = ureg_DECL_constant(ureg); + const0124 = ureg_DECL_constant(ureg); + matrow0 = ureg_DECL_constant(ureg); + matrow1 = ureg_DECL_constant(ureg); + matrow2 = ureg_DECL_constant(ureg); + + if (is_lingrad) { + linear_gradient(ureg, src, + src_input, src_sampler, + coords, const0124, + matrow0, matrow1, matrow2); + } else if (is_radgrad) { + radial_gradient(ureg, src, + src_input, src_sampler, + coords, const0124, + matrow0, matrow1, matrow2); + } + } else + debug_assert(!"Unknown fill type!"); + } + + if (has_mask) { + mask = ureg_DECL_temporary(ureg); + ureg_TEX(ureg, mask, + TGSI_TEXTURE_2D, mask_pos, mask_sampler); + /* src IN mask */ + src_in_mask(ureg, out, ureg_src(src), ureg_src(mask)); + ureg_release_temporary(ureg, mask); + } + + ureg_END(ureg); + + return ureg_create_shader_and_destroy(ureg, pipe); +} + +struct xorg_shaders * xorg_shaders_create(struct exa_context *exa) +{ + struct xorg_shaders *sc = CALLOC_STRUCT(xorg_shaders); + + sc->exa = exa; + sc->vs_hash = cso_hash_create(); + sc->fs_hash = cso_hash_create(); + + return sc; +} + +static void +cache_destroy(struct cso_context *cso, + struct cso_hash *hash, + unsigned processor) +{ + struct cso_hash_iter iter = cso_hash_first_node(hash); + while (!cso_hash_iter_is_null(iter)) { + void *shader = (void *)cso_hash_iter_data(iter); + if (processor == PIPE_SHADER_FRAGMENT) { + cso_delete_fragment_shader(cso, shader); + } else if (processor == PIPE_SHADER_VERTEX) { + cso_delete_vertex_shader(cso, shader); + } + iter = cso_hash_erase(hash, iter); + } + cso_hash_delete(hash); +} + +void xorg_shaders_destroy(struct xorg_shaders *sc) +{ + cache_destroy(sc->exa->cso, sc->vs_hash, + PIPE_SHADER_VERTEX); + cache_destroy(sc->exa->cso, sc->fs_hash, + PIPE_SHADER_FRAGMENT); + + free(sc); +} + +static INLINE void * +shader_from_cache(struct pipe_context *pipe, + unsigned type, + struct cso_hash *hash, + unsigned key) +{ + void *shader = 0; + + struct cso_hash_iter iter = cso_hash_find(hash, key); + + if (cso_hash_iter_is_null(iter)) { + if (type == PIPE_SHADER_VERTEX) + shader = create_vs(pipe, key); + else + shader = create_fs(pipe, key); + cso_hash_insert(hash, key, shader); + } else + shader = (void *)cso_hash_iter_data(iter); + + return shader; +} + +struct xorg_shader xorg_shaders_get(struct xorg_shaders *sc, + unsigned vs_traits, + unsigned fs_traits) +{ + struct xorg_shader shader = {0}; + void *vs, *fs; + + vs = shader_from_cache(sc->exa->ctx, PIPE_SHADER_VERTEX, + sc->vs_hash, vs_traits); + fs = shader_from_cache(sc->exa->ctx, PIPE_SHADER_FRAGMENT, + sc->fs_hash, fs_traits); + + debug_assert(vs && fs); + if (!vs || !fs) + return shader; + + shader.vs = vs; + shader.fs = fs; + + return shader; +} diff --git a/src/gallium/state_trackers/xorg/xorg_exa_tgsi.h b/src/gallium/state_trackers/xorg/xorg_exa_tgsi.h new file mode 100644 index 00000000000..1535a0c8c30 --- /dev/null +++ b/src/gallium/state_trackers/xorg/xorg_exa_tgsi.h @@ -0,0 +1,43 @@ +#ifndef XORG_EXA_TGSI_H +#define XORG_EXA_TGSI_H + +#include "xorg_exa.h" + +enum xorg_vs_traits { + VS_COMPOSITE = 1 << 0, + VS_MASK = 1 << 1, + VS_SOLID_FILL = 1 << 2, + VS_LINGRAD_FILL = 1 << 3, + VS_RADGRAD_FILL = 1 << 4, + VS_FILL = (VS_SOLID_FILL | + VS_LINGRAD_FILL | + VS_RADGRAD_FILL) + /*VS_TRANSFORM = 1 << 5*/ +}; + +enum xorg_fs_traits { + FS_COMPOSITE = 1 << 0, + FS_MASK = 1 << 1, + FS_SOLID_FILL = 1 << 2, + FS_LINGRAD_FILL = 1 << 3, + FS_RADGRAD_FILL = 1 << 4, + FS_FILL = (FS_SOLID_FILL | + FS_LINGRAD_FILL | + FS_RADGRAD_FILL) +}; + +struct xorg_shader { + void *fs; + void *vs; +}; + +struct xorg_shaders; + +struct xorg_shaders *xorg_shaders_create(struct exa_context *exa); +void xorg_shaders_destroy(struct xorg_shaders *shaders); + +struct xorg_shader xorg_shaders_get(struct xorg_shaders *shaders, + unsigned vs_traits, + unsigned fs_traits); + +#endif diff --git a/src/gallium/state_trackers/xorg/xorg_tracker.h b/src/gallium/state_trackers/xorg/xorg_tracker.h index 82c3890dfbd..b1ab783a15a 100644 --- a/src/gallium/state_trackers/xorg/xorg_tracker.h +++ b/src/gallium/state_trackers/xorg/xorg_tracker.h @@ -31,11 +31,20 @@ #ifndef _XORG_TRACKER_H_ #define _XORG_TRACKER_H_ +#include <stddef.h> +#include <stdint.h> #include <errno.h> #include <drm.h> #include <xf86drm.h> #include <xf86drmMode.h> -#include "exa.h" +#include <xorg-server.h> +#include <xf86.h> +#include "xf86Crtc.h" +#include <exa.h> + +#ifdef DRM_MODE_FEATURE_DIRTYFB +#include <damage.h> +#endif #include "pipe/p_screen.h" #include "state_tracker/drm_api.h" @@ -72,16 +81,23 @@ typedef struct _modesettingRec unsigned int SaveGeneration; + void (*blockHandler)(int, pointer, pointer, pointer); CreateScreenResourcesProcPtr createScreenResources; /* gallium */ + struct drm_api *api; struct pipe_screen *screen; struct pipe_context *ctx; + boolean d_depth_bits_last; + boolean ds_depth_bits_last; /* exa */ void *exa; Bool noEvict; +#ifdef DRM_MODE_FEATURE_DIRTYFB + DamagePtr damage; +#endif } modesettingRec, *modesettingPtr; #define modesettingPTR(p) ((modesettingPtr)((p)->driverPrivate)) @@ -94,7 +110,13 @@ struct pipe_texture * xorg_exa_get_texture(PixmapPtr pPixmap); unsigned -xorg_exa_get_pixmap_handle(PixmapPtr pPixmap); +xorg_exa_get_pixmap_handle(PixmapPtr pPixmap, unsigned *stride); + +int +xorg_exa_set_displayed_usage(PixmapPtr pPixmap); + +int +xorg_exa_set_shared_usage(PixmapPtr pPixmap); void * xorg_exa_init(ScrnInfoPtr pScrn); @@ -119,6 +141,9 @@ driCloseScreen(ScreenPtr pScreen); void crtc_init(ScrnInfoPtr pScrn); +void +crtc_cursor_destroy(xf86CrtcPtr crtc); + /*********************************************************************** * xorg_output.c diff --git a/src/gallium/state_trackers/xorg/xorg_winsys.h b/src/gallium/state_trackers/xorg/xorg_winsys.h index d523080e90f..47ee4b9ffd8 100644 --- a/src/gallium/state_trackers/xorg/xorg_winsys.h +++ b/src/gallium/state_trackers/xorg/xorg_winsys.h @@ -37,7 +37,6 @@ #include "xorg-server.h" #include "xf86.h" -#include "xf86Resources.h" #include "pciaccess.h" #ifndef XSERVER_LIBPCIACCESS @@ -46,6 +45,5 @@ void xorg_tracker_set_functions(ScrnInfoPtr scrn); const OptionInfoRec * xorg_tracker_available_options(int chipid, int busid); -void xorg_tracker_loader_ref_sym_lists(void); #endif diff --git a/src/gallium/winsys/drm/Makefile.template b/src/gallium/winsys/drm/Makefile.template index 9f92cb42072..9635c3c50e9 100644 --- a/src/gallium/winsys/drm/Makefile.template +++ b/src/gallium/winsys/drm/Makefile.template @@ -83,7 +83,9 @@ default: depend symlinks $(TOP)/$(LIB_DIR)/gallium/$(LIBNAME) $(LIBNAME): $(OBJECTS) $(MESA_MODULES) $(PIPE_DRIVERS) $(WINOBJ) Makefile $(TOP)/src/mesa/drivers/dri/Makefile.template $(MKLIB) -noprefix -o $@ \ - $(OBJECTS) $(PIPE_DRIVERS) $(MESA_MODULES) $(WINOBJ) $(DRI_LIB_DEPS) $(DRIVER_EXTRAS) + $(OBJECTS) $(PIPE_DRIVERS) \ + -Wl,--start-group $(MESA_MODULES) -Wl,--end-group \ + $(WINOBJ) $(DRI_LIB_DEPS) $(DRIVER_EXTRAS) $(LIBNAME_EGL): $(WINSYS_OBJECTS) $(LIBS) $(MKLIB) -o $(LIBNAME_EGL) \ @@ -118,7 +120,7 @@ clean: install: $(LIBNAME) $(INSTALL) -d $(DESTDIR)$(DRI_DRIVER_INSTALL_DIR) - $(INSTALL) -m 755 $(LIBNAME) $(DESTDIR)$(DRI_DRIVER_INSTALL_DIR) + $(MINSTALL) -m 755 $(LIBNAME) $(DESTDIR)$(DRI_DRIVER_INSTALL_DIR) include depend diff --git a/src/gallium/winsys/drm/intel/SConscript b/src/gallium/winsys/drm/intel/SConscript new file mode 100644 index 00000000000..50d7b75ed68 --- /dev/null +++ b/src/gallium/winsys/drm/intel/SConscript @@ -0,0 +1,7 @@ +Import('*') + +SConscript(['gem/SConscript',]) + +if 'mesa' in env['statetrackers']: + + SConscript(['dri/SConscript']) diff --git a/src/gallium/winsys/drm/intel/dri/Makefile b/src/gallium/winsys/drm/intel/dri/Makefile index ac0891a6463..5e212b62a46 100644 --- a/src/gallium/winsys/drm/intel/dri/Makefile +++ b/src/gallium/winsys/drm/intel/dri/Makefile @@ -6,7 +6,9 @@ LIBNAME = i915_dri.so PIPE_DRIVERS = \ $(TOP)/src/gallium/state_trackers/dri/libdridrm.a \ $(TOP)/src/gallium/winsys/drm/intel/gem/libinteldrm.a \ + $(TOP)/src/gallium/drivers/trace/libtrace.a \ $(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \ + $(TOP)/src/gallium/drivers/identity/libidentity.a \ $(TOP)/src/gallium/drivers/i915simple/libi915simple.a diff --git a/src/gallium/winsys/drm/intel/dri/SConscript b/src/gallium/winsys/drm/intel/dri/SConscript new file mode 100644 index 00000000000..6c00861f517 --- /dev/null +++ b/src/gallium/winsys/drm/intel/dri/SConscript @@ -0,0 +1,19 @@ +Import('*') + +env = drienv.Clone() + +env.ParseConfig('pkg-config --cflags --libs libdrm_intel') + +drivers = [ + st_dri, + inteldrm, + softpipe, + i915simple, + trace, +] + +env.SharedLibrary( + target ='i915_dri.so', + source = COMMON_GALLIUM_SOURCES, + LIBS = drivers + mesa + auxiliaries + env['LIBS'], +) diff --git a/src/gallium/winsys/drm/intel/egl/Makefile b/src/gallium/winsys/drm/intel/egl/Makefile index c5217ad2d63..490baded66b 100644 --- a/src/gallium/winsys/drm/intel/egl/Makefile +++ b/src/gallium/winsys/drm/intel/egl/Makefile @@ -8,6 +8,7 @@ PIPE_DRIVERS = \ $(TOP)/src/gallium/state_trackers/egl/libegldrm.a \ $(GALLIUMDIR)/winsys/drm/intel/gem/libinteldrm.a \ $(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \ + $(TOP)/src/gallium/drivers/trace/libtrace.a \ $(TOP)/src/gallium/drivers/i915simple/libi915simple.a DRIVER_SOURCES = diff --git a/src/gallium/winsys/drm/intel/gem/Makefile b/src/gallium/winsys/drm/intel/gem/Makefile index 7ab1a2a771b..0d6d4e37dbd 100644 --- a/src/gallium/winsys/drm/intel/gem/Makefile +++ b/src/gallium/winsys/drm/intel/gem/Makefile @@ -4,10 +4,10 @@ include $(TOP)/configs/current LIBNAME = inteldrm C_SOURCES = \ - intel_be_batchbuffer.c \ - intel_be_context.c \ - intel_be_device.c \ - intel_be_api.c + intel_drm_batchbuffer.c \ + intel_drm_buffer.c \ + intel_drm_fence.c \ + intel_drm_api.c LIBRARY_INCLUDES = $(shell pkg-config libdrm --cflags-only-I) diff --git a/src/gallium/winsys/drm/intel/gem/SConscript b/src/gallium/winsys/drm/intel/gem/SConscript new file mode 100644 index 00000000000..26717f391fa --- /dev/null +++ b/src/gallium/winsys/drm/intel/gem/SConscript @@ -0,0 +1,17 @@ +Import('*') + +env = drienv.Clone() + +inteldrm_sources = [ + 'intel_drm_api.c', + 'intel_drm_batchbuffer.c', + 'intel_drm_buffer.c', + 'intel_drm_fence.c', +] + +inteldrm = env.ConvenienceLibrary( + target ='inteldrm', + source = inteldrm_sources, +) + +Export('inteldrm') diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_api.c b/src/gallium/winsys/drm/intel/gem/intel_be_api.c deleted file mode 100644 index f4ef7c2d88b..00000000000 --- a/src/gallium/winsys/drm/intel/gem/intel_be_api.c +++ /dev/null @@ -1,15 +0,0 @@ - -#include "intel_be_api.h" -#include "i915simple/i915_winsys.h" - -struct drm_api drm_api_hooks = -{ - /* intel_be_context.c */ - .create_context = intel_be_create_context, - /* intel_be_device.c */ - .create_screen = intel_be_create_screen, - .buffer_from_texture = i915_get_texture_buffer, - .buffer_from_handle = intel_be_buffer_from_handle, - .handle_from_buffer = intel_be_handle_from_buffer, - .global_handle_from_buffer = intel_be_global_handle_from_buffer, -}; diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_api.h b/src/gallium/winsys/drm/intel/gem/intel_be_api.h deleted file mode 100644 index 1c622f3b978..00000000000 --- a/src/gallium/winsys/drm/intel/gem/intel_be_api.h +++ /dev/null @@ -1,15 +0,0 @@ - -#ifndef _INTEL_BE_API_H_ -#define _INTEL_BE_API_H_ - -#include "pipe/p_compiler.h" - -#include "state_tracker/drm_api.h" - -#include "intel_be_device.h" - -struct pipe_screen *intel_be_create_screen(int drmFD, - struct drm_create_screen_arg *arg); -struct pipe_context *intel_be_create_context(struct pipe_screen *screen); - -#endif diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.c b/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.c deleted file mode 100644 index d5e63c3bae5..00000000000 --- a/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.c +++ /dev/null @@ -1,139 +0,0 @@ - -#include "i915simple/i915_debug.h" -#include "intel_be_batchbuffer.h" -#include "intel_be_context.h" -#include "intel_be_device.h" -#include "intel_be_fence.h" -#include <errno.h> - -#include "util/u_memory.h" - -struct intel_be_batchbuffer * -intel_be_batchbuffer_alloc(struct intel_be_context *intel) -{ - struct intel_be_batchbuffer *batch = CALLOC_STRUCT(intel_be_batchbuffer); - - - batch->base.buffer = NULL; - batch->base.winsys = &intel->base; - batch->base.map = NULL; - batch->base.ptr = NULL; - batch->base.size = 0; - batch->base.actual_size = intel->device->max_batch_size; - batch->base.relocs = 0; - batch->base.max_relocs = 500;/*INTEL_DEFAULT_RELOCS;*/ - - batch->base.map = malloc(batch->base.actual_size); - memset(batch->base.map, 0, batch->base.actual_size); - - batch->base.ptr = batch->base.map; - - intel_be_batchbuffer_reset(batch); - - return batch; -} - -void -intel_be_batchbuffer_reset(struct intel_be_batchbuffer *batch) -{ - struct intel_be_context *intel = intel_be_context(batch->base.winsys); - struct intel_be_device *dev = intel->device; - - if (batch->bo) - drm_intel_bo_unreference(batch->bo); - - memset(batch->base.map, 0, batch->base.actual_size); - batch->base.ptr = batch->base.map; - batch->base.size = batch->base.actual_size - BATCH_RESERVED; - - batch->base.relocs = 0; - - batch->bo = drm_intel_bo_alloc(dev->pools.gem, - "gallium3d_batch_buffer", - batch->base.actual_size, 0); -} - -int -intel_be_offset_relocation(struct intel_be_batchbuffer *batch, - unsigned pre_add, - drm_intel_bo *bo, - uint32_t read_domains, - uint32_t write_domain) -{ - unsigned offset; - int ret = 0; - - assert(batch->base.relocs < batch->base.max_relocs); - - offset = (unsigned)(batch->base.ptr - batch->base.map); - - ret = drm_intel_bo_emit_reloc(batch->bo, offset, - bo, pre_add, - read_domains, - write_domain); - - ((uint32_t*)batch->base.ptr)[0] = bo->offset + pre_add; - batch->base.ptr += 4; - - if (!ret) - batch->base.relocs++; - - return ret; -} - -void -intel_be_batchbuffer_flush(struct intel_be_batchbuffer *batch, - struct intel_be_fence **fence) -{ - struct i915_batchbuffer *i915 = &batch->base; - unsigned used = 0; - int ret = 0; - - assert(i915_batchbuffer_space(i915) >= 0); - - used = batch->base.ptr - batch->base.map; - assert((used & 3) == 0); - - if (used & 4) { - i915_batchbuffer_dword(i915, (0x0<<29)|(0x4<<23)|(1<<0)); // MI_FLUSH | FLUSH_MAP_CACHE; - i915_batchbuffer_dword(i915, (0x0<<29)|(0x0<<23)); // MI_NOOP - i915_batchbuffer_dword(i915, (0x0<<29)|(0xA<<23)); // MI_BATCH_BUFFER_END; - } else { - i915_batchbuffer_dword(i915, (0x0<<29)|(0x4<<23)|(1<<0)); //MI_FLUSH | FLUSH_MAP_CACHE; - i915_batchbuffer_dword(i915, (0x0<<29)|(0xA<<23)); // MI_BATCH_BUFFER_END; - } - - used = batch->base.ptr - batch->base.map; - - drm_intel_bo_subdata(batch->bo, 0, used, batch->base.map); - ret = drm_intel_bo_exec(batch->bo, used, NULL, 0, 0); - - assert(ret == 0); - - intel_be_batchbuffer_reset(batch); - - if (fence) { - if (*fence) - intel_be_fence_reference(fence, NULL); - - (*fence) = CALLOC_STRUCT(intel_be_fence); - pipe_reference_init(&(*fence)->reference, 1); - (*fence)->bo = NULL; - } -} - -void -intel_be_batchbuffer_finish(struct intel_be_batchbuffer *batch) -{ - -} - -void -intel_be_batchbuffer_free(struct intel_be_batchbuffer *batch) -{ - if (batch->bo) - drm_intel_bo_unreference(batch->bo); - - free(batch->base.map); - free(batch); -} diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.h b/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.h deleted file mode 100644 index 195bf8dee77..00000000000 --- a/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.h +++ /dev/null @@ -1,55 +0,0 @@ - -#ifndef INTEL_BE_BATCHBUFFER_H -#define INTEL_BE_BATCHBUFFER_H - -#include "i915simple/i915_batch.h" - -#include "drm.h" -#include "intel_bufmgr.h" - -#define BATCH_RESERVED 16 - -#define INTEL_DEFAULT_RELOCS 100 -#define INTEL_MAX_RELOCS 400 - -#define INTEL_BATCH_NO_CLIPRECTS 0x1 -#define INTEL_BATCH_CLIPRECTS 0x2 - -struct intel_be_context; -struct intel_be_device; -struct intel_be_fence; - -struct intel_be_batchbuffer -{ - struct i915_batchbuffer base; - - struct intel_be_context *intel; - struct intel_be_device *device; - - drm_intel_bo *bo; -}; - -struct intel_be_batchbuffer * -intel_be_batchbuffer_alloc(struct intel_be_context *intel); - -void -intel_be_batchbuffer_free(struct intel_be_batchbuffer *batch); - -void -intel_be_batchbuffer_finish(struct intel_be_batchbuffer *batch); - -void -intel_be_batchbuffer_flush(struct intel_be_batchbuffer *batch, - struct intel_be_fence **fence); - -void -intel_be_batchbuffer_reset(struct intel_be_batchbuffer *batch); - -int -intel_be_offset_relocation(struct intel_be_batchbuffer *batch, - unsigned pre_add, - drm_intel_bo *bo, - uint32_t read_domains, - uint32_t write_doman); - -#endif diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_context.c b/src/gallium/winsys/drm/intel/gem/intel_be_context.c deleted file mode 100644 index fe0b138fbea..00000000000 --- a/src/gallium/winsys/drm/intel/gem/intel_be_context.c +++ /dev/null @@ -1,120 +0,0 @@ - -#include "pipe/p_screen.h" - -#include "softpipe/sp_winsys.h" - -#include "intel_be_device.h" -#include "intel_be_context.h" -#include "intel_be_batchbuffer.h" - -#include "i915_drm.h" - -#include "intel_be_api.h" - -static struct i915_batchbuffer * -intel_be_batch_get(struct i915_winsys *sws) -{ - struct intel_be_context *intel = intel_be_context(sws); - return &intel->batch->base; -} - -static void -intel_be_batch_reloc(struct i915_winsys *sws, - struct pipe_buffer *buf, - unsigned access_flags, - unsigned delta) -{ - struct intel_be_context *intel = intel_be_context(sws); - drm_intel_bo *bo = intel_bo(buf); - int ret; - uint32_t read = 0; - uint32_t write = 0; - - if (access_flags & I915_BUFFER_ACCESS_WRITE) { - write = I915_GEM_DOMAIN_RENDER; - read = I915_GEM_DOMAIN_RENDER; - } - - if (access_flags & I915_BUFFER_ACCESS_READ) { - read |= I915_GEM_DOMAIN_VERTEX; - } - - ret = intel_be_offset_relocation(intel->batch, - delta, - bo, - read, - write); - assert(ret == 0); - - /* TODO change return type */ - /* return ret; */ -} - -static void -intel_be_batch_flush(struct i915_winsys *sws, - struct pipe_fence_handle **fence) -{ - struct intel_be_context *intel = intel_be_context(sws); - struct intel_be_fence **f = (struct intel_be_fence **)fence; - - if (fence && *fence) - assert(0); - - intel_be_batchbuffer_flush(intel->batch, f); -} - - -/* - * Misc functions. - */ - -static void -intel_be_destroy_context(struct i915_winsys *winsys) -{ - struct intel_be_context *intel = intel_be_context(winsys); - - intel_be_batchbuffer_free(intel->batch); - - free(intel); -} - -boolean -intel_be_init_context(struct intel_be_context *intel, struct intel_be_device *device) -{ - assert(intel); - assert(device); - intel->device = device; - - intel->base.batch_get = intel_be_batch_get; - intel->base.batch_reloc = intel_be_batch_reloc; - intel->base.batch_flush = intel_be_batch_flush; - - intel->base.destroy = intel_be_destroy_context; - - intel->batch = intel_be_batchbuffer_alloc(intel); - - return true; -} - -struct pipe_context * -intel_be_create_context(struct pipe_screen *screen) -{ - struct intel_be_context *intel; - struct pipe_context *pipe; - struct intel_be_device *device = intel_be_device(screen->winsys); - - intel = (struct intel_be_context *)malloc(sizeof(*intel)); - memset(intel, 0, sizeof(*intel)); - - intel_be_init_context(intel, device); - - if (device->softpipe) - pipe = softpipe_create(screen); - else - pipe = i915_create_context(screen, &device->base, &intel->base); - - if (pipe) - pipe->priv = intel; - - return pipe; -} diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_context.h b/src/gallium/winsys/drm/intel/gem/intel_be_context.h deleted file mode 100644 index 5a369669c0d..00000000000 --- a/src/gallium/winsys/drm/intel/gem/intel_be_context.h +++ /dev/null @@ -1,31 +0,0 @@ - -#ifndef INTEL_BE_CONTEXT_H -#define INTEL_BE_CONTEXT_H - -#include "i915simple/i915_winsys.h" - -struct intel_be_context -{ - /** Interface to i915simple driver */ - struct i915_winsys base; - - struct intel_be_device *device; - struct intel_be_batchbuffer *batch; -}; - -static INLINE struct intel_be_context * -intel_be_context(struct i915_winsys *sws) -{ - return (struct intel_be_context *)sws; -} - -/** - * Intialize a allocated intel_be_context struct. - * - * Remember to set the hardware_* functions. - */ -boolean -intel_be_init_context(struct intel_be_context *intel, - struct intel_be_device *device); - -#endif diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_device.c b/src/gallium/winsys/drm/intel/gem/intel_be_device.c deleted file mode 100644 index 907ac86637f..00000000000 --- a/src/gallium/winsys/drm/intel/gem/intel_be_device.c +++ /dev/null @@ -1,347 +0,0 @@ - -#include "intel_be_device.h" - -#include "pipe/internal/p_winsys_screen.h" -#include "pipe/p_defines.h" -#include "pipe/p_state.h" -#include "pipe/p_inlines.h" -#include "util/u_memory.h" -#include "util/u_debug.h" - -#include "intel_be_fence.h" - -#include "i915simple/i915_winsys.h" -#include "softpipe/sp_winsys.h" - -#include "intel_be_api.h" -#include <stdio.h> - -/* - * Buffer - */ - -static void * -intel_be_buffer_map(struct pipe_winsys *winsys, - struct pipe_buffer *buf, - unsigned flags) -{ - drm_intel_bo *bo = intel_bo(buf); - int write = 0; - int ret; - - if (flags & PIPE_BUFFER_USAGE_DONTBLOCK) { - /* Remove this when drm_intel_bo_map supports DONTBLOCK - */ - return NULL; - } - - if (flags & PIPE_BUFFER_USAGE_CPU_WRITE) - write = 1; - - ret = drm_intel_bo_map(bo, write); - - if (ret) - return NULL; - - return bo->virtual; -} - -static void -intel_be_buffer_unmap(struct pipe_winsys *winsys, - struct pipe_buffer *buf) -{ - drm_intel_bo_unmap(intel_bo(buf)); -} - -static void -intel_be_buffer_destroy(struct pipe_buffer *buf) -{ - drm_intel_bo_unreference(intel_bo(buf)); - free(buf); -} - -static struct pipe_buffer * -intel_be_buffer_create(struct pipe_winsys *winsys, - unsigned alignment, - unsigned usage, - unsigned size) -{ - struct intel_be_buffer *buffer = CALLOC_STRUCT(intel_be_buffer); - struct intel_be_device *dev = intel_be_device(winsys); - drm_intel_bufmgr *pool; - char *name; - - if (!buffer) - return NULL; - - pipe_reference_init(&buffer->base.reference, 1); - buffer->base.alignment = alignment; - buffer->base.usage = usage; - buffer->base.size = size; - buffer->flinked = FALSE; - buffer->flink = 0; - - if (usage & (PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_CONSTANT)) { - /* Local buffer */ - name = "gallium3d_local"; - pool = dev->pools.gem; - } else if (usage & PIPE_BUFFER_USAGE_CUSTOM) { - /* For vertex buffers */ - name = "gallium3d_internal_vertex"; - pool = dev->pools.gem; - } else { - /* Regular buffers */ - name = "gallium3d_regular"; - pool = dev->pools.gem; - } - - buffer->bo = drm_intel_bo_alloc(pool, name, size, alignment); - - if (!buffer->bo) - goto err; - - return &buffer->base; - -err: - free(buffer); - return NULL; -} - -static struct pipe_buffer * -intel_be_user_buffer_create(struct pipe_winsys *winsys, void *ptr, unsigned bytes) -{ - struct intel_be_buffer *buffer = CALLOC_STRUCT(intel_be_buffer); - struct intel_be_device *dev = intel_be_device(winsys); - int ret; - - if (!buffer) - return NULL; - - pipe_reference_init(&buffer->base.reference, 1); - buffer->base.alignment = 0; - buffer->base.usage = 0; - buffer->base.size = bytes; - - buffer->bo = drm_intel_bo_alloc(dev->pools.gem, - "gallium3d_user_buffer", - bytes, 0); - - if (!buffer->bo) - goto err; - - ret = drm_intel_bo_subdata(buffer->bo, - 0, bytes, ptr); - - if (ret) - goto err; - - return &buffer->base; - -err: - free(buffer); - return NULL; -} - -struct pipe_buffer * -intel_be_buffer_from_handle(struct pipe_screen *screen, - const char* name, unsigned handle) -{ - struct intel_be_device *dev = intel_be_device(screen->winsys); - struct intel_be_buffer *buffer = CALLOC_STRUCT(intel_be_buffer); - - if (!buffer) - return NULL; - - buffer->bo = drm_intel_bo_gem_create_from_name(dev->pools.gem, name, handle); - - if (!buffer->bo) - goto err; - - pipe_reference_init(&buffer->base.reference, 1); - buffer->base.screen = screen; - buffer->base.alignment = buffer->bo->align; - buffer->base.usage = PIPE_BUFFER_USAGE_GPU_READ | - PIPE_BUFFER_USAGE_GPU_WRITE | - PIPE_BUFFER_USAGE_CPU_READ | - PIPE_BUFFER_USAGE_CPU_WRITE; - buffer->base.size = buffer->bo->size; - - return &buffer->base; - -err: - free(buffer); - return NULL; -} - -boolean -intel_be_handle_from_buffer(struct pipe_screen *screen, - struct pipe_buffer *buffer, - unsigned *handle) -{ - if (!buffer) - return FALSE; - - *handle = intel_bo(buffer)->handle; - return TRUE; -} - -boolean -intel_be_global_handle_from_buffer(struct pipe_screen *screen, - struct pipe_buffer *buffer, - unsigned *handle) -{ - struct intel_be_buffer *buf = intel_be_buffer(buffer); - - if (!buffer) - return FALSE; - - if (!buf->flinked) { - if (drm_intel_bo_flink(intel_bo(buffer), &buf->flink)) - return FALSE; - buf->flinked = TRUE; - } - - *handle = buf->flink; - return TRUE; -} -/* - * Fence - */ - -static void -intel_be_fence_refunref(struct pipe_winsys *sws, - struct pipe_fence_handle **ptr, - struct pipe_fence_handle *fence) -{ - struct intel_be_fence **p = (struct intel_be_fence **)ptr; - struct intel_be_fence *f = (struct intel_be_fence *)fence; - - intel_be_fence_reference(p, f); -} - -static int -intel_be_fence_signalled(struct pipe_winsys *sws, - struct pipe_fence_handle *fence, - unsigned flag) -{ - assert(0); - - return 0; -} - -static int -intel_be_fence_finish(struct pipe_winsys *sws, - struct pipe_fence_handle *fence, - unsigned flag) -{ - struct intel_be_fence *f = (struct intel_be_fence *)fence; - - /* fence already expired */ - if (!f->bo) - return 0; - - drm_intel_bo_wait_rendering(f->bo); - drm_intel_bo_unreference(f->bo); - f->bo = NULL; - - return 0; -} - -/* - * Misc functions - */ - -static void -intel_be_destroy_winsys(struct pipe_winsys *winsys) -{ - struct intel_be_device *dev = intel_be_device(winsys); - - drm_intel_bufmgr_destroy(dev->pools.gem); - - free(dev); -} - -boolean -intel_be_init_device(struct intel_be_device *dev, int fd, unsigned id) -{ - dev->fd = fd; - dev->id = id; - dev->max_batch_size = 16 * 4096; - dev->max_vertex_size = 128 * 4096; - - dev->base.buffer_create = intel_be_buffer_create; - dev->base.user_buffer_create = intel_be_user_buffer_create; - dev->base.buffer_map = intel_be_buffer_map; - dev->base.buffer_unmap = intel_be_buffer_unmap; - dev->base.buffer_destroy = intel_be_buffer_destroy; - - /* Not used anymore */ - dev->base.surface_buffer_create = NULL; - - dev->base.fence_reference = intel_be_fence_refunref; - dev->base.fence_signalled = intel_be_fence_signalled; - dev->base.fence_finish = intel_be_fence_finish; - - dev->base.destroy = intel_be_destroy_winsys; - - dev->pools.gem = drm_intel_bufmgr_gem_init(dev->fd, dev->max_batch_size); - - dev->softpipe = debug_get_bool_option("INTEL_SOFTPIPE", FALSE); - - return true; -} - -static void -intel_be_get_device_id(unsigned int *device_id) -{ - char path[512]; - FILE *file; - - /* - * FIXME: Fix this up to use a drm ioctl or whatever. - */ - - snprintf(path, sizeof(path), "/sys/class/drm/card0/device/device"); - file = fopen(path, "r"); - if (!file) { - return; - } - - fgets(path, sizeof(path), file); - sscanf(path, "%x", device_id); - fclose(file); -} - -struct pipe_screen * -intel_be_create_screen(int drmFD, struct drm_create_screen_arg *arg) -{ - struct intel_be_device *dev; - struct pipe_screen *screen; - unsigned int deviceID; - - if (arg != NULL) { - switch(arg->mode) { - case DRM_CREATE_NORMAL: - break; - default: - return NULL; - } - } - - /* Allocate the private area */ - dev = malloc(sizeof(*dev)); - if (!dev) - return NULL; - memset(dev, 0, sizeof(*dev)); - - intel_be_get_device_id(&deviceID); - intel_be_init_device(dev, drmFD, deviceID); - - if (dev->softpipe) { - screen = softpipe_create_screen(&dev->base); - drm_api_hooks.buffer_from_texture = softpipe_get_texture_buffer; - } else - screen = i915_create_screen(&dev->base, deviceID); - - return screen; -} diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_device.h b/src/gallium/winsys/drm/intel/gem/intel_be_device.h deleted file mode 100644 index b32637ece29..00000000000 --- a/src/gallium/winsys/drm/intel/gem/intel_be_device.h +++ /dev/null @@ -1,94 +0,0 @@ - -#ifndef INTEL_DRM_DEVICE_H -#define INTEL_DRM_DEVICE_H - -#include "pipe/internal/p_winsys_screen.h" -#include "pipe/p_context.h" - -#include "drm.h" -#include "intel_bufmgr.h" - -/* - * Device - */ - -struct intel_be_device -{ - struct pipe_winsys base; - - boolean softpipe; - - int fd; /**< Drm file discriptor */ - - unsigned id; - - size_t max_batch_size; - size_t max_vertex_size; - - struct { - drm_intel_bufmgr *gem; - } pools; -}; - -static INLINE struct intel_be_device * -intel_be_device(struct pipe_winsys *winsys) -{ - return (struct intel_be_device *)winsys; -} - -boolean -intel_be_init_device(struct intel_be_device *device, int fd, unsigned id); - -/* - * Buffer - */ - -struct intel_be_buffer { - struct pipe_buffer base; - drm_intel_bo *bo; - boolean flinked; - unsigned flink; -}; - -/** - * Create a be buffer from a drm bo handle. - * - * Takes a reference. - */ -struct pipe_buffer * -intel_be_buffer_from_handle(struct pipe_screen *screen, - const char* name, unsigned handle); - -/** - * Gets a handle from a buffer. - * - * If buffer is destroyed handle may become invalid. - */ -boolean -intel_be_handle_from_buffer(struct pipe_screen *screen, - struct pipe_buffer *buffer, - unsigned *handle); - -/** - * Gets the global handle from a buffer. - * - * If buffer is destroyed handle may become invalid. - */ -boolean -intel_be_global_handle_from_buffer(struct pipe_screen *screen, - struct pipe_buffer *buffer, - unsigned *handle); - -static INLINE struct intel_be_buffer * -intel_be_buffer(struct pipe_buffer *buf) -{ - return (struct intel_be_buffer *)buf; -} - -static INLINE drm_intel_bo * -intel_bo(struct pipe_buffer *buf) -{ - return intel_be_buffer(buf)->bo; -} - -#endif diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_fence.h b/src/gallium/winsys/drm/intel/gem/intel_be_fence.h deleted file mode 100644 index a8abb01a9ed..00000000000 --- a/src/gallium/winsys/drm/intel/gem/intel_be_fence.h +++ /dev/null @@ -1,34 +0,0 @@ - -#ifndef INTEL_BE_FENCE_H -#define INTEL_BE_FENCE_H - -#include "pipe/p_defines.h" - -#include "drm.h" -#include "intel_bufmgr.h" - -/** - * Because gem does not have fence's we have to create our own fences. - * - * They work by keeping the batchbuffer around and checking if that has - * been idled. If bo is NULL fence has expired. - */ -struct intel_be_fence -{ - struct pipe_reference reference; - drm_intel_bo *bo; -}; - -static INLINE void -intel_be_fence_reference(struct intel_be_fence **ptr, struct intel_be_fence *f) -{ - struct intel_be_fence *old_fence = *ptr; - - if (pipe_reference((struct pipe_reference**)ptr, &f->reference)) { - if (old_fence->bo) - drm_intel_bo_unreference(old_fence->bo); - free(old_fence); - } -} - -#endif diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_api.c b/src/gallium/winsys/drm/intel/gem/intel_drm_api.c new file mode 100644 index 00000000000..4c5a1d2ea89 --- /dev/null +++ b/src/gallium/winsys/drm/intel/gem/intel_drm_api.c @@ -0,0 +1,202 @@ + +#include "state_tracker/drm_api.h" + +#include "intel_drm_winsys.h" +#include "util/u_memory.h" + +#include "i915simple/i915_context.h" +#include "i915simple/i915_screen.h" + + +/* + * Helper functions + */ + + +static void +intel_drm_get_device_id(unsigned int *device_id) +{ + char path[512]; + FILE *file; + void *shutup_gcc; + + /* + * FIXME: Fix this up to use a drm ioctl or whatever. + */ + + snprintf(path, sizeof(path), "/sys/class/drm/card0/device/device"); + file = fopen(path, "r"); + if (!file) { + return; + } + + shutup_gcc = fgets(path, sizeof(path), file); + sscanf(path, "%x", device_id); + fclose(file); +} + +static struct intel_buffer * +intel_drm_buffer_from_handle(struct intel_drm_winsys *idws, + const char* name, unsigned handle) +{ + struct intel_drm_buffer *buf = CALLOC_STRUCT(intel_drm_buffer); + + if (!buf) + return NULL; + + buf->magic = 0xDEAD1337; + buf->bo = drm_intel_bo_gem_create_from_name(idws->pools.gem, name, handle); + buf->flinked = TRUE; + buf->flink = handle; + + if (!buf->bo) + goto err; + + return (struct intel_buffer *)buf; + +err: + FREE(buf); + return NULL; +} + + +/* + * Exported functions + */ + + +static struct pipe_texture * +intel_drm_texture_from_shared_handle(struct drm_api *api, + struct pipe_screen *screen, + struct pipe_texture *templ, + const char* name, + unsigned pitch, + unsigned handle) +{ + struct intel_drm_winsys *idws = intel_drm_winsys(i915_screen(screen)->iws); + struct intel_buffer *buffer; + + buffer = intel_drm_buffer_from_handle(idws, name, handle); + if (!buffer) + return NULL; + + return i915_texture_blanket_intel(screen, templ, pitch, buffer); +} + +static boolean +intel_drm_shared_handle_from_texture(struct drm_api *api, + struct pipe_screen *screen, + struct pipe_texture *texture, + unsigned *pitch, + unsigned *handle) +{ + struct intel_drm_buffer *buf = NULL; + struct intel_buffer *buffer = NULL; + if (!i915_get_texture_buffer_intel(texture, &buffer, pitch)) + return FALSE; + + buf = intel_drm_buffer(buffer); + if (!buf->flinked) { + if (drm_intel_bo_flink(buf->bo, &buf->flink)) + return FALSE; + buf->flinked = TRUE; + } + + *handle = buf->flink; + + return TRUE; +} + +static boolean +intel_drm_local_handle_from_texture(struct drm_api *api, + struct pipe_screen *screen, + struct pipe_texture *texture, + unsigned *pitch, + unsigned *handle) +{ + struct intel_buffer *buffer = NULL; + if (!i915_get_texture_buffer_intel(texture, &buffer, pitch)) + return FALSE; + + *handle = intel_drm_buffer(buffer)->bo->handle; + + return TRUE; +} + +static void +intel_drm_winsys_destroy(struct intel_winsys *iws) +{ + struct intel_drm_winsys *idws = intel_drm_winsys(iws); + + drm_intel_bufmgr_destroy(idws->pools.gem); + + FREE(idws); +} + +static struct pipe_screen * +intel_drm_create_screen(struct drm_api *api, int drmFD, + struct drm_create_screen_arg *arg) +{ + struct intel_drm_winsys *idws; + unsigned int deviceID; + + if (arg != NULL) { + switch(arg->mode) { + case DRM_CREATE_NORMAL: + break; + default: + return NULL; + } + } + + idws = CALLOC_STRUCT(intel_drm_winsys); + if (!idws) + return NULL; + + intel_drm_get_device_id(&deviceID); + + intel_drm_winsys_init_batchbuffer_functions(idws); + intel_drm_winsys_init_buffer_functions(idws); + intel_drm_winsys_init_fence_functions(idws); + + idws->fd = drmFD; + idws->id = deviceID; + idws->max_batch_size = 16 * 4096; + + idws->base.destroy = intel_drm_winsys_destroy; + + idws->pools.gem = drm_intel_bufmgr_gem_init(idws->fd, idws->max_batch_size); + + idws->softpipe = FALSE; + idws->dump_cmd = debug_get_bool_option("INTEL_DUMP_CMD", FALSE); + + return i915_create_screen(&idws->base, deviceID); +} + +static struct pipe_context * +intel_drm_create_context(struct drm_api *api, struct pipe_screen *screen) +{ + return i915_create_context(screen); +} + +static void +destroy(struct drm_api *api) +{ + +} + +struct drm_api intel_drm_api = +{ + .create_context = intel_drm_create_context, + .create_screen = intel_drm_create_screen, + .texture_from_shared_handle = intel_drm_texture_from_shared_handle, + .shared_handle_from_texture = intel_drm_shared_handle_from_texture, + .local_handle_from_texture = intel_drm_local_handle_from_texture, + .destroy = destroy, +}; + +struct drm_api * +drm_api_create() +{ + return &intel_drm_api; +} diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_batchbuffer.c b/src/gallium/winsys/drm/intel/gem/intel_drm_batchbuffer.c new file mode 100644 index 00000000000..5ca3ad9762d --- /dev/null +++ b/src/gallium/winsys/drm/intel/gem/intel_drm_batchbuffer.c @@ -0,0 +1,214 @@ + +#include "intel_drm_winsys.h" +#include "util/u_memory.h" + +#include "i915_drm.h" + +#define BATCH_RESERVED 16 + +#define INTEL_DEFAULT_RELOCS 100 +#define INTEL_MAX_RELOCS 400 + +#define INTEL_BATCH_NO_CLIPRECTS 0x1 +#define INTEL_BATCH_CLIPRECTS 0x2 + +#undef INTEL_RUN_SYNC + +struct intel_drm_batchbuffer +{ + struct intel_batchbuffer base; + + size_t actual_size; + + drm_intel_bo *bo; +}; + +static INLINE struct intel_drm_batchbuffer * +intel_drm_batchbuffer(struct intel_batchbuffer *batch) +{ + return (struct intel_drm_batchbuffer *)batch; +} + +static void +intel_drm_batchbuffer_reset(struct intel_drm_batchbuffer *batch) +{ + struct intel_drm_winsys *idws = intel_drm_winsys(batch->base.iws); + + if (batch->bo) + drm_intel_bo_unreference(batch->bo); + batch->bo = drm_intel_bo_alloc(idws->pools.gem, + "gallium3d_batchbuffer", + batch->actual_size, + 4096); + drm_intel_bo_map(batch->bo, TRUE); + batch->base.map = batch->bo->virtual; + + memset(batch->base.map, 0, batch->actual_size); + batch->base.ptr = batch->base.map; + batch->base.size = batch->actual_size - BATCH_RESERVED; + batch->base.relocs = 0; +} + +static struct intel_batchbuffer * +intel_drm_batchbuffer_create(struct intel_winsys *iws) +{ + struct intel_drm_winsys *idws = intel_drm_winsys(iws); + struct intel_drm_batchbuffer *batch = CALLOC_STRUCT(intel_drm_batchbuffer); + + batch->base.map = NULL; + batch->base.ptr = NULL; + batch->base.size = 0; + + batch->base.relocs = 0; + batch->base.max_relocs = 300;/*INTEL_DEFAULT_RELOCS;*/ + + batch->base.iws = iws; + + batch->actual_size = idws->max_batch_size; + + intel_drm_batchbuffer_reset(batch); + + return &batch->base; +} + +static int +intel_drm_batchbuffer_reloc(struct intel_batchbuffer *ibatch, + struct intel_buffer *buffer, + enum intel_buffer_usage usage, + unsigned pre_add) +{ + struct intel_drm_batchbuffer *batch = intel_drm_batchbuffer(ibatch); + unsigned write_domain = 0; + unsigned read_domain = 0; + unsigned offset; + int ret = 0; + + assert(batch->base.relocs < batch->base.max_relocs); + + if (usage == INTEL_USAGE_SAMPLER) { + write_domain = 0; + read_domain = I915_GEM_DOMAIN_SAMPLER; + + } else if (usage == INTEL_USAGE_RENDER) { + write_domain = I915_GEM_DOMAIN_RENDER; + read_domain = I915_GEM_DOMAIN_RENDER; + + } else if (usage == INTEL_USAGE_2D_TARGET) { + write_domain = I915_GEM_DOMAIN_RENDER; + read_domain = I915_GEM_DOMAIN_RENDER; + + } else if (usage == INTEL_USAGE_2D_SOURCE) { + write_domain = 0; + read_domain = I915_GEM_DOMAIN_RENDER; + + } else if (usage == INTEL_USAGE_VERTEX) { + write_domain = 0; + read_domain = I915_GEM_DOMAIN_VERTEX; + + } else { + assert(0); + return -1; + } + + offset = (unsigned)(batch->base.ptr - batch->base.map); + + ret = drm_intel_bo_emit_reloc(batch->bo, offset, + intel_bo(buffer), pre_add, + read_domain, + write_domain); + + ((uint32_t*)batch->base.ptr)[0] = intel_bo(buffer)->offset + pre_add; + batch->base.ptr += 4; + + if (!ret) + batch->base.relocs++; + + return ret; +} + +static void +intel_drm_batchbuffer_flush(struct intel_batchbuffer *ibatch, + struct pipe_fence_handle **fence) +{ + struct intel_drm_batchbuffer *batch = intel_drm_batchbuffer(ibatch); + unsigned used = 0; + int ret = 0; + int i; + + assert(intel_batchbuffer_space(ibatch) >= 0); + + used = batch->base.ptr - batch->base.map; + assert((used & 3) == 0); + + if (used & 4) { + // MI_FLUSH | FLUSH_MAP_CACHE; + intel_batchbuffer_dword(ibatch, (0x0<<29)|(0x4<<23)|(1<<0)); + // MI_NOOP + intel_batchbuffer_dword(ibatch, (0x0<<29)|(0x0<<23)); + // MI_BATCH_BUFFER_END; + intel_batchbuffer_dword(ibatch, (0x0<<29)|(0xA<<23)); + } else { + //MI_FLUSH | FLUSH_MAP_CACHE; + intel_batchbuffer_dword(ibatch, (0x0<<29)|(0x4<<23)|(1<<0)); + // MI_BATCH_BUFFER_END; + intel_batchbuffer_dword(ibatch, (0x0<<29)|(0xA<<23)); + } + + used = batch->base.ptr - batch->base.map; + + drm_intel_bo_unmap(batch->bo); + + /* Do the sending to HW */ + ret = drm_intel_bo_exec(batch->bo, used, NULL, 0, 0); + assert(ret == 0); + + if (intel_drm_winsys(ibatch->iws)->dump_cmd) { + unsigned *ptr; + drm_intel_bo_map(batch->bo, FALSE); + ptr = (unsigned*)batch->bo->virtual; + + debug_printf("%s:\n", __func__); + for (i = 0; i < used / 4; i++, ptr++) { + debug_printf("\t%08x: %08x\n", i*4, *ptr); + } + + drm_intel_bo_unmap(batch->bo); + } else { +#ifdef INTEL_RUN_SYNC + drm_intel_bo_map(batch->bo, FALSE); + drm_intel_bo_unmap(batch->bo); +#endif + } + + if (fence) { + ibatch->iws->fence_reference(ibatch->iws, fence, NULL); + +#ifdef INTEL_RUN_SYNC + /* we run synced to GPU so just pass null */ + (*fence) = intel_drm_fence_create(NULL); +#else + (*fence) = intel_drm_fence_create(batch->bo); +#endif + } + + intel_drm_batchbuffer_reset(batch); +} + +static void +intel_drm_batchbuffer_destroy(struct intel_batchbuffer *ibatch) +{ + struct intel_drm_batchbuffer *batch = intel_drm_batchbuffer(ibatch); + + if (batch->bo) + drm_intel_bo_unreference(batch->bo); + + free(batch); +} + +void intel_drm_winsys_init_batchbuffer_functions(struct intel_drm_winsys *idws) +{ + idws->base.batchbuffer_create = intel_drm_batchbuffer_create; + idws->base.batchbuffer_reloc = intel_drm_batchbuffer_reloc; + idws->base.batchbuffer_flush = intel_drm_batchbuffer_flush; + idws->base.batchbuffer_destroy = intel_drm_batchbuffer_destroy; +} diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c b/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c new file mode 100644 index 00000000000..e017cd2e982 --- /dev/null +++ b/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c @@ -0,0 +1,134 @@ + +#include "intel_drm_winsys.h" +#include "util/u_memory.h" + +#include "i915_drm.h" + +static struct intel_buffer * +intel_drm_buffer_create(struct intel_winsys *iws, + unsigned size, unsigned alignment, + enum intel_buffer_type type) +{ + struct intel_drm_buffer *buf = CALLOC_STRUCT(intel_drm_buffer); + struct intel_drm_winsys *idws = intel_drm_winsys(iws); + drm_intel_bufmgr *pool; + char *name; + + if (!buf) + return NULL; + + buf->magic = 0xDEAD1337; + buf->flinked = FALSE; + buf->flink = 0; + buf->map_gtt = FALSE; + + if (type == INTEL_NEW_TEXTURE) { + name = "gallium3d_texture"; + pool = idws->pools.gem; + } else if (type == INTEL_NEW_VERTEX) { + name = "gallium3d_vertex"; + pool = idws->pools.gem; + } else if (type == INTEL_NEW_SCANOUT) { + name = "gallium3d_scanout"; + pool = idws->pools.gem; + buf->map_gtt = TRUE; + } else { + assert(0); + name = "gallium3d_unknown"; + pool = idws->pools.gem; + } + + buf->bo = drm_intel_bo_alloc(pool, name, size, alignment); + + if (!buf->bo) + goto err; + + return (struct intel_buffer *)buf; + +err: + assert(0); + FREE(buf); + return NULL; +} + +static int +intel_drm_buffer_set_fence_reg(struct intel_winsys *iws, + struct intel_buffer *buffer, + unsigned stride, + enum intel_buffer_tile tile) +{ + assert(I915_TILING_NONE == INTEL_TILE_NONE); + assert(I915_TILING_X == INTEL_TILE_X); + assert(I915_TILING_Y == INTEL_TILE_Y); + + return drm_intel_bo_set_tiling(intel_bo(buffer), &tile, stride); +} + +static void * +intel_drm_buffer_map(struct intel_winsys *iws, + struct intel_buffer *buffer, + boolean write) +{ + struct intel_drm_buffer *buf = intel_drm_buffer(buffer); + drm_intel_bo *bo = intel_bo(buffer); + int ret = 0; + + assert(bo); + + if (buf->map_count) + goto out; + + if (buf->map_gtt) + ret = drm_intel_gem_bo_map_gtt(bo); + else + ret = drm_intel_bo_map(bo, write); + + buf->ptr = bo->virtual; + + assert(ret == 0); +out: + if (ret) + return NULL; + + buf->map_count++; + return buf->ptr; +} + +static void +intel_drm_buffer_unmap(struct intel_winsys *iws, + struct intel_buffer *buffer) +{ + struct intel_drm_buffer *buf = intel_drm_buffer(buffer); + + if (--buf->map_count) + return; + + if (buf->map_gtt) + drm_intel_gem_bo_unmap_gtt(intel_bo(buffer)); + else + drm_intel_bo_unmap(intel_bo(buffer)); +} + +static void +intel_drm_buffer_destroy(struct intel_winsys *iws, + struct intel_buffer *buffer) +{ + drm_intel_bo_unreference(intel_bo(buffer)); + +#ifdef DEBUG + intel_drm_buffer(buffer)->magic = 0; + intel_drm_buffer(buffer)->bo = NULL; +#endif + + FREE(buffer); +} + +void +intel_drm_winsys_init_buffer_functions(struct intel_drm_winsys *idws) +{ + idws->base.buffer_create = intel_drm_buffer_create; + idws->base.buffer_set_fence_reg = intel_drm_buffer_set_fence_reg; + idws->base.buffer_map = intel_drm_buffer_map; + idws->base.buffer_unmap = intel_drm_buffer_unmap; + idws->base.buffer_destroy = intel_drm_buffer_destroy; +} diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_fence.c b/src/gallium/winsys/drm/intel/gem/intel_drm_fence.c new file mode 100644 index 00000000000..e70bfe7b44e --- /dev/null +++ b/src/gallium/winsys/drm/intel/gem/intel_drm_fence.c @@ -0,0 +1,81 @@ + +#include "intel_drm_winsys.h" +#include "util/u_memory.h" +#include "pipe/p_refcnt.h" + +/** + * Because gem does not have fence's we have to create our own fences. + * + * They work by keeping the batchbuffer around and checking if that has + * been idled. If bo is NULL fence has expired. + */ +struct intel_drm_fence +{ + struct pipe_reference reference; + drm_intel_bo *bo; +}; + + +struct pipe_fence_handle * +intel_drm_fence_create(drm_intel_bo *bo) +{ + struct intel_drm_fence *fence = CALLOC_STRUCT(intel_drm_fence); + + pipe_reference_init(&fence->reference, 1); + /* bo is null if fence already expired */ + if (bo) { + drm_intel_bo_reference(bo); + fence->bo = bo; + } + + return (struct pipe_fence_handle *)fence; +} + +static void +intel_drm_fence_reference(struct intel_winsys *iws, + struct pipe_fence_handle **ptr, + struct pipe_fence_handle *fence) +{ + struct intel_drm_fence *old = (struct intel_drm_fence *)*ptr; + struct intel_drm_fence *f = (struct intel_drm_fence *)fence; + + if (pipe_reference((struct pipe_reference**)ptr, &f->reference)) { + if (old->bo) + drm_intel_bo_unreference(old->bo); + FREE(old); + } +} + +static int +intel_drm_fence_signalled(struct intel_winsys *iws, + struct pipe_fence_handle *fence) +{ + assert(0); + + return 0; +} + +static int +intel_drm_fence_finish(struct intel_winsys *iws, + struct pipe_fence_handle *fence) +{ + struct intel_drm_fence *f = (struct intel_drm_fence *)fence; + + /* fence already expired */ + if (!f->bo) + return 0; + + drm_intel_bo_wait_rendering(f->bo); + drm_intel_bo_unreference(f->bo); + f->bo = NULL; + + return 0; +} + +void +intel_drm_winsys_init_fence_functions(struct intel_drm_winsys *idws) +{ + idws->base.fence_reference = intel_drm_fence_reference; + idws->base.fence_signalled = intel_drm_fence_signalled; + idws->base.fence_finish = intel_drm_fence_finish; +} diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_winsys.h b/src/gallium/winsys/drm/intel/gem/intel_drm_winsys.h new file mode 100644 index 00000000000..415c45feea0 --- /dev/null +++ b/src/gallium/winsys/drm/intel/gem/intel_drm_winsys.h @@ -0,0 +1,78 @@ + +#ifndef INTEL_DRM_WINSYS_H +#define INTEL_DRM_WINSYS_H + +#include "i915simple/intel_batchbuffer.h" + +#include "drm.h" +#include "intel_bufmgr.h" + + +/* + * Winsys + */ + + +struct intel_drm_winsys +{ + struct intel_winsys base; + + boolean softpipe; + boolean dump_cmd; + + int fd; /**< Drm file discriptor */ + + unsigned id; + + size_t max_batch_size; + + struct { + drm_intel_bufmgr *gem; + } pools; +}; + +static INLINE struct intel_drm_winsys * +intel_drm_winsys(struct intel_winsys *iws) +{ + return (struct intel_drm_winsys *)iws; +} + +struct intel_drm_winsys * intel_drm_winsys_create(int fd, unsigned pci_id); +struct pipe_fence_handle * intel_drm_fence_create(drm_intel_bo *bo); + +void intel_drm_winsys_init_batchbuffer_functions(struct intel_drm_winsys *idws); +void intel_drm_winsys_init_buffer_functions(struct intel_drm_winsys *idws); +void intel_drm_winsys_init_fence_functions(struct intel_drm_winsys *idws); + + +/* + * Buffer + */ + + +struct intel_drm_buffer { + unsigned magic; + + drm_intel_bo *bo; + + void *ptr; + unsigned map_count; + boolean map_gtt; + + boolean flinked; + unsigned flink; +}; + +static INLINE struct intel_drm_buffer * +intel_drm_buffer(struct intel_buffer *buffer) +{ + return (struct intel_drm_buffer *)buffer; +} + +static INLINE drm_intel_bo * +intel_bo(struct intel_buffer *buffer) +{ + return intel_drm_buffer(buffer)->bo; +} + +#endif diff --git a/src/gallium/winsys/drm/intel/xorg/Makefile b/src/gallium/winsys/drm/intel/xorg/Makefile index b1b6b9362b6..9e56853b021 100644 --- a/src/gallium/winsys/drm/intel/xorg/Makefile +++ b/src/gallium/winsys/drm/intel/xorg/Makefile @@ -1,36 +1,39 @@ TARGET = modesetting_drv.so CFILES = $(wildcard ./*.c) OBJECTS = $(patsubst ./%.c,./%.o,$(CFILES)) -GALLIUMDIR = ../../../.. TOP = ../../../../../.. -include ${TOP}/configs/current +include $(TOP)/configs/current -CFLAGS = -DHAVE_CONFIG_H \ - -g -Wall -Wimplicit-function-declaration -fPIC \ - $(shell pkg-config --cflags pixman-1 xorg-server libdrm xproto) \ - -I../gem \ - -I${GALLIUMDIR}/include \ - -I${GALLIUMDIR}/drivers \ - -I${GALLIUMDIR}/auxiliary \ - -I${TOP}/src/mesa \ - -I$(TOP)/include \ - -I$(TOP)/src/egl/main +INCLUDES = \ + $(shell pkg-config --cflags-only-I pixman-1 xorg-server libdrm xproto) \ + -I../gem \ + -I$(TOP)/src/gallium/include \ + -I$(TOP)/src/gallium/drivers \ + -I$(TOP)/src/gallium/auxiliary \ + -I$(TOP)/src/mesa \ + -I$(TOP)/include \ + -I$(TOP)/src/egl/main LIBS = \ - $(GALLIUMDIR)/state_trackers/xorg/libxorgtracker.a \ - $(GALLIUMDIR)/winsys/drm/intel/gem/libinteldrm.a \ + $(TOP)/src/gallium/state_trackers/xorg/libxorgtracker.a \ + $(TOP)/src/gallium/winsys/drm/intel/gem/libinteldrm.a \ $(TOP)/src/gallium/drivers/i915simple/libi915simple.a \ + $(TOP)/src/gallium/drivers/trace/libtrace.a \ $(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \ $(GALLIUM_AUXILIARIES) +DRIVER_DEFINES = \ + -DHAVE_CONFIG_H + + ############################################# all default: $(TARGET) -$(TARGET): $(OBJECTS) Makefile $(GALLIUMDIR)/state_trackers/xorg/libxorgtracker.a $(LIBS) +$(TARGET): $(OBJECTS) Makefile $(TOP)/src/gallium/state_trackers/xorg/libxorgtracker.a $(LIBS) $(TOP)/bin/mklib -noprefix -o $@ \ $(OBJECTS) $(LIBS) $(shell pkg-config --libs libdrm) -ldrm_intel @@ -39,6 +42,16 @@ clean: install: $(INSTALL) -d $(DESTDIR)/$(XORG_DRIVER_INSTALL_DIR) - $(INSTALL) -m 755 $(TARGET) $(DESTDIR)/$(XORG_DRIVER_INSTALL_DIR) + $(MINSTALL) -m 755 $(TARGET) $(DESTDIR)/$(XORG_DRIVER_INSTALL_DIR) + + +############################################## + + +.c.o: + $(CC) -c $(CFLAGS) $(INCLUDES) $(DRIVER_DEFINES) $< -o $@ + + +############################################## .PHONY = all clean install diff --git a/src/gallium/winsys/drm/intel/xorg/intel_xorg.c b/src/gallium/winsys/drm/intel/xorg/intel_xorg.c index 46a7971f609..369dc356cf8 100644 --- a/src/gallium/winsys/drm/intel/xorg/intel_xorg.c +++ b/src/gallium/winsys/drm/intel/xorg/intel_xorg.c @@ -47,8 +47,8 @@ static SymTabRec intel_xorg_chipsets[] = { }; static PciChipsets intel_xorg_pci_devices[] = { - {PCI_MATCH_ANY, PCI_MATCH_ANY, RES_SHARED_VGA}, - {-1, -1, RES_UNDEFINED} + {PCI_MATCH_ANY, PCI_MATCH_ANY, NULL}, + {-1, -1, NULL} }; static XF86ModuleVersionInfo intel_xorg_version = { @@ -105,12 +105,6 @@ intel_xorg_setup(pointer module, pointer opts, int *errmaj, int *errmin) xf86AddDriver(&modesetting, module, HaveDriverFuncs); /* - * Tell the loader about symbols from other modules that this module - * might refer to. - */ - xorg_tracker_loader_ref_sym_lists(); - - /* * The return value must be non-NULL on success even though there * is no TearDownProc. */ @@ -140,7 +134,7 @@ intel_xorg_pci_probe(DriverPtr driver, NULL, NULL, NULL, NULL, NULL); if (scrn != NULL) { scrn->driverVersion = 1; - scrn->driverName = "modesetting"; + scrn->driverName = "i915"; scrn->name = "modesetting"; scrn->Probe = NULL; diff --git a/src/gallium/winsys/drm/nouveau/Makefile b/src/gallium/winsys/drm/nouveau/Makefile index f8c81358544..6c9cbef26df 100644 --- a/src/gallium/winsys/drm/nouveau/Makefile +++ b/src/gallium/winsys/drm/nouveau/Makefile @@ -2,7 +2,7 @@ TOP = ../../../../.. include $(TOP)/configs/current -SUBDIRS = drm dri dri2 +SUBDIRS = drm $(GALLIUM_STATE_TRACKERS_DIRS) default install clean: @for dir in $(SUBDIRS) ; do \ diff --git a/src/gallium/winsys/drm/nouveau/dri/Makefile b/src/gallium/winsys/drm/nouveau/dri/Makefile index f7db6201fea..0937f68c34b 100644 --- a/src/gallium/winsys/drm/nouveau/dri/Makefile +++ b/src/gallium/winsys/drm/nouveau/dri/Makefile @@ -3,32 +3,25 @@ include $(TOP)/configs/current LIBNAME = nouveau_dri.so -MINIGLX_SOURCES = - PIPE_DRIVERS = \ + $(TOP)/src/gallium/state_trackers/dri/libdridrm.a \ $(TOP)/src/gallium/winsys/drm/nouveau/drm/libnouveaudrm.a \ $(TOP)/src/gallium/drivers/nv04/libnv04.a \ $(TOP)/src/gallium/drivers/nv10/libnv10.a \ $(TOP)/src/gallium/drivers/nv20/libnv20.a \ $(TOP)/src/gallium/drivers/nv30/libnv30.a \ $(TOP)/src/gallium/drivers/nv40/libnv40.a \ - $(TOP)/src/gallium/drivers/nv50/libnv50.a - -DRIVER_SOURCES = \ - nouveau_context.c \ - nouveau_screen.c \ - nouveau_swapbuffers.c \ - nouveau_lock.c + $(TOP)/src/gallium/drivers/nv50/libnv50.a \ + $(TOP)/src/gallium/drivers/nouveau/libnouveau.a + +DRIVER_SOURCES = C_SOURCES = \ $(COMMON_GALLIUM_SOURCES) \ $(DRIVER_SOURCES) -ASM_SOURCES = +include ../../Makefile.template -DRIVER_DEFINES = $(shell pkg-config libdrm_nouveau --cflags) DRI_LIB_DEPS += $(shell pkg-config libdrm_nouveau --libs) -include ../../Makefile.template - symlinks: diff --git a/src/gallium/winsys/drm/nouveau/dri/nouveau_context.c b/src/gallium/winsys/drm/nouveau/dri/nouveau_context.c deleted file mode 100644 index deb6ffcff1c..00000000000 --- a/src/gallium/winsys/drm/nouveau/dri/nouveau_context.c +++ /dev/null @@ -1,118 +0,0 @@ -#include <main/glheader.h> -#include <glapi/glthread.h> -#include <GL/internal/glcore.h> -#include <utils.h> - -#include <state_tracker/st_public.h> -#include <state_tracker/st_context.h> -#include <state_tracker/drm_api.h> -#include <pipe/p_defines.h> -#include <pipe/p_context.h> -#include <pipe/p_screen.h> - -#include "nouveau_context.h" -#include "nouveau_screen.h" - -#include "nouveau_drmif.h" - -GLboolean -nouveau_context_create(const __GLcontextModes *glVis, - __DRIcontextPrivate *driContextPriv, - void *sharedContextPrivate) -{ - __DRIscreenPrivate *driScrnPriv = driContextPriv->driScreenPriv; - struct nouveau_screen *nv_screen = driScrnPriv->private; - struct nouveau_context *nv; - struct pipe_context *pipe; - struct st_context *st_share = NULL; - - if (sharedContextPrivate) - st_share = ((struct nouveau_context *)sharedContextPrivate)->st; - - nv = CALLOC_STRUCT(nouveau_context); - if (!nv) - return GL_FALSE; - - { - struct nouveau_device_priv *nvdev = - nouveau_device(nv_screen->device); - - nvdev->ctx = driContextPriv->hHWContext; - nvdev->lock = (drmLock *)&driScrnPriv->pSAREA->lock; - } - - pipe = drm_api_hooks.create_context(nv_screen->pscreen); - if (!pipe) { - FREE(nv); - return GL_FALSE; - } - pipe->priv = nv; - - driContextPriv->driverPrivate = nv; - nv->dri_screen = driScrnPriv; - - driParseConfigFiles(&nv->dri_option_cache, &nv_screen->option_cache, - nv->dri_screen->myNum, "nouveau"); - - nv->st = st_create_context(pipe, glVis, st_share); - return GL_TRUE; -} - -void -nouveau_context_destroy(__DRIcontextPrivate *driContextPriv) -{ - struct nouveau_context *nv = driContextPriv->driverPrivate; - - assert(nv); - - st_finish(nv->st); - st_destroy_context(nv->st); - - FREE(nv); -} - -GLboolean -nouveau_context_bind(__DRIcontextPrivate *driContextPriv, - __DRIdrawablePrivate *driDrawPriv, - __DRIdrawablePrivate *driReadPriv) -{ - struct nouveau_context *nv; - struct nouveau_framebuffer *draw, *read; - - if (!driContextPriv) { - st_make_current(NULL, NULL, NULL); - return GL_TRUE; - } - - nv = driContextPriv->driverPrivate; - draw = driDrawPriv->driverPrivate; - read = driReadPriv->driverPrivate; - - st_make_current(nv->st, draw->stfb, read->stfb); - - if ((nv->dri_drawable != driDrawPriv) || - (nv->last_stamp != driDrawPriv->lastStamp)) { - nv->dri_drawable = driDrawPriv; - st_resize_framebuffer(draw->stfb, driDrawPriv->w, - driDrawPriv->h); - nv->last_stamp = driDrawPriv->lastStamp; - } - - if (driDrawPriv != driReadPriv) { - st_resize_framebuffer(read->stfb, driReadPriv->w, - driReadPriv->h); - } - - return GL_TRUE; -} - -GLboolean -nouveau_context_unbind(__DRIcontextPrivate *driContextPriv) -{ - struct nouveau_context *nv = driContextPriv->driverPrivate; - (void)nv; - - st_flush(nv->st, 0, NULL); - return GL_TRUE; -} - diff --git a/src/gallium/winsys/drm/nouveau/dri/nouveau_context.h b/src/gallium/winsys/drm/nouveau/dri/nouveau_context.h deleted file mode 100644 index 2779b092e64..00000000000 --- a/src/gallium/winsys/drm/nouveau/dri/nouveau_context.h +++ /dev/null @@ -1,53 +0,0 @@ -#ifndef __NOUVEAU_CONTEXT_DRI_H__ -#define __NOUVEAU_CONTEXT_DRI_H__ - -#include <dri_util.h> -#include <xmlconfig.h> - -#include "nouveau/nouveau_winsys.h" - -#define NOUVEAU_ERR(fmt, args...) debug_printf("%s: "fmt, __func__, ##args) - -struct nouveau_framebuffer { - struct st_framebuffer *stfb; -}; - -struct nouveau_context { - struct st_context *st; - - /* DRI stuff */ - __DRIscreenPrivate *dri_screen; - __DRIdrawablePrivate *dri_drawable; - unsigned int last_stamp; - driOptionCache dri_option_cache; - drm_context_t drm_context; - drmLock drm_lock; - int locked; -}; - -extern GLboolean nouveau_context_create(const __GLcontextModes *, - __DRIcontextPrivate *, void *); -extern void nouveau_context_destroy(__DRIcontextPrivate *); -extern GLboolean nouveau_context_bind(__DRIcontextPrivate *, - __DRIdrawablePrivate *draw, - __DRIdrawablePrivate *read); -extern GLboolean nouveau_context_unbind(__DRIcontextPrivate *); - -extern void nouveau_contended_lock(struct nouveau_context *nv); -extern void LOCK_HARDWARE(struct nouveau_context *nv); -extern void UNLOCK_HARDWARE(struct nouveau_context *nv); - -#ifdef DEBUG -extern int __nouveau_debug; - -#define DEBUG_BO (1 << 0) - -#define DBG(flag, ...) do { \ - if (__nouveau_debug & (DEBUG_##flag)) \ - NOUVEAU_ERR(__VA_ARGS__); \ -} while(0) -#else -#define DBG(flag, ...) -#endif - -#endif diff --git a/src/gallium/winsys/drm/nouveau/dri/nouveau_screen.c b/src/gallium/winsys/drm/nouveau/dri/nouveau_screen.c deleted file mode 100644 index 4e9b76a9090..00000000000 --- a/src/gallium/winsys/drm/nouveau/dri/nouveau_screen.c +++ /dev/null @@ -1,330 +0,0 @@ -#include <utils.h> -#include <vblank.h> -#include <xmlpool.h> - -#include <pipe/p_context.h> -#include <state_tracker/st_public.h> -#include <state_tracker/st_cb_fbo.h> -#include <state_tracker/drm_api.h> - -#include "nouveau_context.h" -#include "nouveau_screen.h" -#include "nouveau_swapbuffers.h" -#include "nouveau_dri.h" - -#include "nouveau_drm.h" -#include "nouveau_drmif.h" - -#if NOUVEAU_DRM_HEADER_PATCHLEVEL != 12 -#error nouveau_drm.h version does not match expected version -#endif - -/* Extension stuff, enabling of extensions handled by Gallium's GL state - * tracker. But, we still need to define the entry points we want. - */ -#define need_GL_ARB_fragment_program -#define need_GL_ARB_multisample -#define need_GL_ARB_occlusion_query -#define need_GL_ARB_point_parameters -#define need_GL_ARB_shader_objects -#define need_GL_ARB_texture_compression -#define need_GL_ARB_vertex_program -#define need_GL_ARB_vertex_shader -#define need_GL_ARB_vertex_buffer_object -#define need_GL_EXT_compiled_vertex_array -#define need_GL_EXT_fog_coord -#define need_GL_EXT_secondary_color -#define need_GL_EXT_framebuffer_object -#define need_GL_VERSION_2_0 -#define need_GL_VERSION_2_1 -#include "extension_helper.h" - -const struct dri_extension card_extensions[] = -{ - { "GL_ARB_multisample", GL_ARB_multisample_functions }, - { "GL_ARB_occlusion_query", GL_ARB_occlusion_query_functions }, - { "GL_ARB_point_parameters", GL_ARB_point_parameters_functions }, - { "GL_ARB_shader_objects", GL_ARB_shader_objects_functions }, - { "GL_ARB_shading_language_100", GL_VERSION_2_0_functions }, - { "GL_ARB_shading_language_120", GL_VERSION_2_1_functions }, - { "GL_ARB_texture_compression", GL_ARB_texture_compression_functions }, - { "GL_ARB_vertex_program", GL_ARB_vertex_program_functions }, - { "GL_ARB_vertex_shader", GL_ARB_vertex_shader_functions }, - { "GL_ARB_vertex_buffer_object", GL_ARB_vertex_buffer_object_functions }, - { "GL_EXT_compiled_vertex_array", GL_EXT_compiled_vertex_array_functions }, - { "GL_EXT_fog_coord", GL_EXT_fog_coord_functions }, - { "GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions }, - { "GL_EXT_secondary_color", GL_EXT_secondary_color_functions }, - { NULL, 0 } -}; - -PUBLIC const char __driConfigOptions[] = -DRI_CONF_BEGIN -DRI_CONF_END; -static const GLuint __driNConfigOptions = 0; - -extern const struct dri_extension common_extensions[]; -extern const struct dri_extension nv40_extensions[]; - -static GLboolean -nouveau_create_buffer(__DRIscreenPrivate * driScrnPriv, - __DRIdrawablePrivate * driDrawPriv, - const __GLcontextModes *glVis, GLboolean pixmapBuffer) -{ - struct nouveau_framebuffer *nvfb; - enum pipe_format colour, depth, stencil; - - if (pixmapBuffer) - return GL_FALSE; - - nvfb = CALLOC_STRUCT(nouveau_framebuffer); - if (!nvfb) - return GL_FALSE; - - if (glVis->redBits == 5) - colour = PIPE_FORMAT_R5G6B5_UNORM; - else - colour = PIPE_FORMAT_A8R8G8B8_UNORM; - - if (glVis->depthBits == 16) - depth = PIPE_FORMAT_Z16_UNORM; - else if (glVis->depthBits == 24) - depth = PIPE_FORMAT_Z24S8_UNORM; - else - depth = PIPE_FORMAT_NONE; - - if (glVis->stencilBits == 8) - stencil = PIPE_FORMAT_Z24S8_UNORM; - else - stencil = PIPE_FORMAT_NONE; - - nvfb->stfb = st_create_framebuffer(glVis, colour, depth, stencil, - driDrawPriv->w, driDrawPriv->h, - (void*)nvfb); - if (!nvfb->stfb) { - free(nvfb); - return GL_FALSE; - } - - driDrawPriv->driverPrivate = (void *)nvfb; - return GL_TRUE; -} - -static void -nouveau_destroy_buffer(__DRIdrawablePrivate * driDrawPriv) -{ - struct nouveau_framebuffer *nvfb; - - nvfb = (struct nouveau_framebuffer *)driDrawPriv->driverPrivate; - st_unreference_framebuffer(nvfb->stfb); - free(nvfb); -} - -static __DRIconfig ** -nouveau_fill_in_modes(__DRIscreenPrivate *psp, - unsigned pixel_bits, unsigned depth_bits, - unsigned stencil_bits, GLboolean have_back_buffer) -{ - __DRIconfig **configs; - unsigned depth_buffer_factor; - unsigned back_buffer_factor; - GLenum fb_format; - GLenum fb_type; - - static const GLenum back_buffer_modes[] = { - GLX_NONE, GLX_SWAP_UNDEFINED_OML, - }; - - uint8_t depth_bits_array[3]; - uint8_t stencil_bits_array[3]; - uint8_t msaa_samples_array[1]; - - depth_bits_array[0] = 0; - depth_bits_array[1] = depth_bits; - depth_bits_array[2] = depth_bits; - - /* Just like with the accumulation buffer, always provide some modes - * with a stencil buffer. It will be a sw fallback, but some apps won't - * care about that. - */ - stencil_bits_array[0] = 0; - stencil_bits_array[1] = 0; - if (depth_bits == 24) - stencil_bits_array[1] = (stencil_bits == 0) ? 8 : stencil_bits; - stencil_bits_array[2] = (stencil_bits == 0) ? 8 : stencil_bits; - - msaa_samples_array[0] = 0; - - depth_buffer_factor = - ((depth_bits != 0) || (stencil_bits != 0)) ? 3 : 1; - back_buffer_factor = (have_back_buffer) ? 3 : 1; - - if (pixel_bits == 16) { - fb_format = GL_RGB; - fb_type = GL_UNSIGNED_SHORT_5_6_5; - } - else { - fb_format = GL_BGRA; - fb_type = GL_UNSIGNED_INT_8_8_8_8_REV; - } - - configs = driCreateConfigs(fb_format, fb_type, - depth_bits_array, stencil_bits_array, - depth_buffer_factor, back_buffer_modes, - back_buffer_factor, msaa_samples_array, 1); - if (configs == NULL) { - fprintf(stderr, "[%s:%u] Error creating FBConfig!\n", - __func__, __LINE__); - return NULL; - } - - return configs; -} - -static struct pipe_surface * -dri_surface_from_handle(struct pipe_screen *screen, - unsigned handle, - enum pipe_format format, - unsigned width, - unsigned height, - unsigned pitch) -{ - struct pipe_surface *surface = NULL; - struct pipe_texture *texture = NULL; - struct pipe_texture templat; - struct pipe_buffer *buf = NULL; - - buf = drm_api_hooks.buffer_from_handle(screen, "front buffer", handle); - if (!buf) - return NULL; - - memset(&templat, 0, sizeof(templat)); - templat.tex_usage = PIPE_TEXTURE_USAGE_PRIMARY | - NOUVEAU_TEXTURE_USAGE_LINEAR; - templat.target = PIPE_TEXTURE_2D; - templat.last_level = 0; - templat.depth[0] = 1; - templat.format = format; - templat.width[0] = width; - templat.height[0] = height; - pf_get_block(templat.format, &templat.block); - - texture = screen->texture_blanket(screen, - &templat, - &pitch, - buf); - - /* we don't need the buffer from this point on */ - pipe_buffer_reference(&buf, NULL); - - if (!texture) - return NULL; - - surface = screen->get_tex_surface(screen, texture, 0, 0, 0, - PIPE_BUFFER_USAGE_GPU_READ | - PIPE_BUFFER_USAGE_GPU_WRITE); - - /* we don't need the texture from this point on */ - pipe_texture_reference(&texture, NULL); - return surface; -} - -static const __DRIconfig ** -nouveau_screen_create(__DRIscreenPrivate *psp) -{ - struct nouveau_dri *nv_dri = psp->pDevPriv; - struct nouveau_screen *nv_screen; - static const __DRIversion ddx_expected = - { 0, 0, NOUVEAU_DRM_HEADER_PATCHLEVEL }; - static const __DRIversion dri_expected = { 4, 0, 0 }; - static const __DRIversion drm_expected = - { 0, 0, NOUVEAU_DRM_HEADER_PATCHLEVEL }; - - if (!driCheckDriDdxDrmVersions2("nouveau", - &psp->dri_version, &dri_expected, - &psp->ddx_version, &ddx_expected, - &psp->drm_version, &drm_expected)) { - return NULL; - } - - if (drm_expected.patch != psp->drm_version.patch) { - fprintf(stderr, "Incompatible DRM patch level.\n" - "Expected: %d\n" "Current : %d\n", - drm_expected.patch, psp->drm_version.patch); - return NULL; - } - - driInitExtensions(NULL, card_extensions, GL_FALSE); - - if (psp->devPrivSize != sizeof(struct nouveau_dri)) { - NOUVEAU_ERR("DRI struct mismatch between DDX/DRI\n"); - return NULL; - } - - nv_screen = CALLOC_STRUCT(nouveau_screen); - if (!nv_screen) - return NULL; - - nouveau_device_open_existing(&nv_screen->device, 0, psp->fd, 0); - - nv_screen->pscreen = drm_api_hooks.create_screen(psp->fd, NULL); - if (!nv_screen->pscreen) { - FREE(nv_screen); - return NULL; - } - nv_screen->pscreen->flush_frontbuffer = nouveau_flush_frontbuffer; - - { - enum pipe_format format; - - if (nv_dri->bpp == 16) - format = PIPE_FORMAT_R5G6B5_UNORM; - else - format = PIPE_FORMAT_A8R8G8B8_UNORM; - - nv_screen->fb = dri_surface_from_handle(nv_screen->pscreen, - nv_dri->front_offset, - format, - nv_dri->width, - nv_dri->height, - nv_dri->front_pitch * - nv_dri->bpp / 8); - } - - driParseOptionInfo(&nv_screen->option_cache, - __driConfigOptions, __driNConfigOptions); - - nv_screen->driScrnPriv = psp; - psp->private = (void *)nv_screen; - - return (const __DRIconfig **) - nouveau_fill_in_modes(psp, nv_dri->bpp, - (nv_dri->bpp == 16) ? 16 : 24, - (nv_dri->bpp == 16) ? 0 : 8, 1); -} - -static void -nouveau_screen_destroy(__DRIscreenPrivate *driScrnPriv) -{ - struct nouveau_screen *nv_screen = driScrnPriv->private; - - driScrnPriv->private = NULL; - FREE(nv_screen); -} - -const struct __DriverAPIRec -driDriverAPI = { - .InitScreen = nouveau_screen_create, - .DestroyScreen = nouveau_screen_destroy, - .CreateContext = nouveau_context_create, - .DestroyContext = nouveau_context_destroy, - .CreateBuffer = nouveau_create_buffer, - .DestroyBuffer = nouveau_destroy_buffer, - .SwapBuffers = nouveau_swap_buffers, - .MakeCurrent = nouveau_context_bind, - .UnbindContext = nouveau_context_unbind, - .CopySubBuffer = nouveau_copy_sub_buffer, - - .InitScreen2 = NULL, /* one day, I promise! */ -}; - diff --git a/src/gallium/winsys/drm/nouveau/dri/nouveau_screen.h b/src/gallium/winsys/drm/nouveau/dri/nouveau_screen.h deleted file mode 100644 index ac078f3c638..00000000000 --- a/src/gallium/winsys/drm/nouveau/dri/nouveau_screen.h +++ /dev/null @@ -1,16 +0,0 @@ -#ifndef __NOUVEAU_SCREEN_DRI_H__ -#define __NOUVEAU_SCREEN_DRI_H__ - -#include "xmlconfig.h" - -struct nouveau_screen { - __DRIscreenPrivate *driScrnPriv; - driOptionCache option_cache; - - struct nouveau_device *device; - - struct pipe_screen *pscreen; - struct pipe_surface *fb; -}; - -#endif diff --git a/src/gallium/winsys/drm/nouveau/dri/nouveau_swapbuffers.c b/src/gallium/winsys/drm/nouveau/dri/nouveau_swapbuffers.c deleted file mode 100644 index 9c841a0b2d0..00000000000 --- a/src/gallium/winsys/drm/nouveau/dri/nouveau_swapbuffers.c +++ /dev/null @@ -1,115 +0,0 @@ -#include <main/glheader.h> -#include <glapi/glthread.h> -#include <GL/internal/glcore.h> - -#include <pipe/p_context.h> -#include <state_tracker/st_public.h> -#include <state_tracker/st_context.h> -#include <state_tracker/st_cb_fbo.h> - -#include "nouveau_context.h" -#include "nouveau_screen.h" -#include "nouveau_swapbuffers.h" - -#include "nouveau_pushbuf.h" - -void -nouveau_copy_buffer(__DRIdrawablePrivate *dPriv, struct pipe_surface *surf, - const drm_clip_rect_t *rect) -{ - struct nouveau_context *nv = dPriv->driContextPriv->driverPrivate; - struct nouveau_screen *nv_screen = nv->dri_screen->private; - struct pipe_context *pipe = nv->st->pipe; - drm_clip_rect_t *pbox; - int nbox, i; - - LOCK_HARDWARE(nv); - if (!dPriv->numClipRects) { - UNLOCK_HARDWARE(nv); - return; - } - pbox = dPriv->pClipRects; - nbox = dPriv->numClipRects; - - for (i = 0; i < nbox; i++, pbox++) { - int sx, sy, dx, dy, w, h; - - sx = pbox->x1 - dPriv->x; - sy = pbox->y1 - dPriv->y; - dx = pbox->x1; - dy = pbox->y1; - w = pbox->x2 - pbox->x1; - h = pbox->y2 - pbox->y1; - - pipe->surface_copy(pipe, nv_screen->fb, dx, dy, surf, - sx, sy, w, h); - } - - pipe->flush(pipe, 0, NULL); - UNLOCK_HARDWARE(nv); - - if (nv->last_stamp != dPriv->lastStamp) { - struct nouveau_framebuffer *nvfb = dPriv->driverPrivate; - st_resize_framebuffer(nvfb->stfb, dPriv->w, dPriv->h); - nv->last_stamp = dPriv->lastStamp; - } -} - -void -nouveau_copy_sub_buffer(__DRIdrawablePrivate *dPriv, int x, int y, int w, int h) -{ - struct nouveau_framebuffer *nvfb = dPriv->driverPrivate; - struct pipe_surface *surf; - - st_get_framebuffer_surface(nvfb->stfb, ST_SURFACE_BACK_LEFT, &surf); - if (surf) { - drm_clip_rect_t rect; - rect.x1 = x; - rect.y1 = y; - rect.x2 = x + w; - rect.y2 = y + h; - - st_notify_swapbuffers(nvfb->stfb); - nouveau_copy_buffer(dPriv, surf, &rect); - } -} - -void -nouveau_swap_buffers(__DRIdrawablePrivate *dPriv) -{ - struct nouveau_framebuffer *nvfb = dPriv->driverPrivate; - struct pipe_surface *surf; - - st_get_framebuffer_surface(nvfb->stfb, ST_SURFACE_BACK_LEFT, &surf); - if (surf) { - st_notify_swapbuffers(nvfb->stfb); - nouveau_copy_buffer(dPriv, surf, NULL); - } -} - -void -nouveau_flush_frontbuffer(struct pipe_screen *pscreen, struct pipe_surface *ps, - void *context_private) -{ - struct nouveau_context *nv = context_private; - __DRIdrawablePrivate *dPriv = nv->dri_drawable; - - nouveau_copy_buffer(dPriv, ps, NULL); -} - -void -nouveau_contended_lock(struct nouveau_context *nv) -{ - struct nouveau_context *nv_sub = (struct nouveau_context*)nv; - __DRIdrawablePrivate *dPriv = nv_sub->dri_drawable; - __DRIscreenPrivate *sPriv = nv_sub->dri_screen; - - /* If the window moved, may need to set a new cliprect now. - * - * NOTE: This releases and regains the hw lock, so all state - * checking must be done *after* this call: - */ - if (dPriv) - DRI_VALIDATE_DRAWABLE_INFO(sPriv, dPriv); -} - diff --git a/src/gallium/winsys/drm/nouveau/dri/nouveau_swapbuffers.h b/src/gallium/winsys/drm/nouveau/dri/nouveau_swapbuffers.h deleted file mode 100644 index 4ca9cc22831..00000000000 --- a/src/gallium/winsys/drm/nouveau/dri/nouveau_swapbuffers.h +++ /dev/null @@ -1,11 +0,0 @@ -#ifndef __NOUVEAU_SWAPBUFFERS_H__ -#define __NOUVEAU_SWAPBUFFERS_H__ - -void nouveau_copy_buffer(__DRIdrawablePrivate *, struct pipe_surface *, - const drm_clip_rect_t *); -void nouveau_copy_sub_buffer(__DRIdrawablePrivate *, int x, int y, int w, int h); -void nouveau_swap_buffers(__DRIdrawablePrivate *); -void nouveau_flush_frontbuffer(struct pipe_screen *, struct pipe_surface *, - void *context_private); - -#endif diff --git a/src/gallium/winsys/drm/nouveau/dri2/Makefile b/src/gallium/winsys/drm/nouveau/dri2/Makefile deleted file mode 100644 index 377a80d5188..00000000000 --- a/src/gallium/winsys/drm/nouveau/dri2/Makefile +++ /dev/null @@ -1,26 +0,0 @@ -TOP = ../../../../../.. -include $(TOP)/configs/current - -LIBNAME = nouveau_dri2.so - -PIPE_DRIVERS = \ - $(TOP)/src/gallium/state_trackers/dri/libdridrm.a \ - $(TOP)/src/gallium/winsys/drm/nouveau/drm/libnouveaudrm.a \ - $(TOP)/src/gallium/drivers/nv04/libnv04.a \ - $(TOP)/src/gallium/drivers/nv10/libnv10.a \ - $(TOP)/src/gallium/drivers/nv20/libnv20.a \ - $(TOP)/src/gallium/drivers/nv30/libnv30.a \ - $(TOP)/src/gallium/drivers/nv40/libnv40.a \ - $(TOP)/src/gallium/drivers/nv50/libnv50.a - -DRIVER_SOURCES = - -C_SOURCES = \ - $(COMMON_GALLIUM_SOURCES) \ - $(DRIVER_SOURCES) - -include ../../Makefile.template - -DRI_LIB_DEPS += $(shell pkg-config libdrm_nouveau --libs) - -symlinks: diff --git a/src/gallium/winsys/drm/nouveau/drm/Makefile b/src/gallium/winsys/drm/nouveau/drm/Makefile index 2da78d8690f..54c3b26c755 100644 --- a/src/gallium/winsys/drm/nouveau/drm/Makefile +++ b/src/gallium/winsys/drm/nouveau/drm/Makefile @@ -3,9 +3,7 @@ include $(TOP)/configs/current LIBNAME = nouveaudrm -C_SOURCES = nouveau_drm_api.c \ - nouveau_winsys_pipe.c \ - nouveau_winsys.c +C_SOURCES = nouveau_drm_api.c LIBRARY_INCLUDES = $(shell pkg-config libdrm libdrm_nouveau --cflags-only-I) LIBRARY_DEFINES = $(shell pkg-config libdrm libdrm_nouveau --cflags-only-other) diff --git a/src/gallium/winsys/drm/nouveau/dri/nouveau_dri.h b/src/gallium/winsys/drm/nouveau/drm/nouveau_dri.h index 1207c2d609c..1207c2d609c 100644 --- a/src/gallium/winsys/drm/nouveau/dri/nouveau_dri.h +++ b/src/gallium/winsys/drm/nouveau/drm/nouveau_dri.h diff --git a/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c b/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c index a558fda1401..091cbbcfed1 100644 --- a/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c +++ b/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c @@ -1,20 +1,70 @@ +#include "pipe/p_context.h" +#include "pipe/p_state.h" #include "util/u_memory.h" #include "nouveau_drm_api.h" -#include "nouveau_winsys_pipe.h" #include "nouveau_drmif.h" #include "nouveau_channel.h" #include "nouveau_bo.h" +#include "nouveau/nouveau_winsys.h" +#include "nouveau/nouveau_screen.h" + +static struct pipe_surface * +dri_surface_from_handle(struct drm_api *api, struct pipe_screen *pscreen, + unsigned handle, enum pipe_format format, + unsigned width, unsigned height, unsigned pitch) +{ + struct pipe_surface *ps = NULL; + struct pipe_texture *pt = NULL; + struct pipe_texture tmpl; + + memset(&tmpl, 0, sizeof(tmpl)); + tmpl.tex_usage = PIPE_TEXTURE_USAGE_PRIMARY | + NOUVEAU_TEXTURE_USAGE_LINEAR; + tmpl.target = PIPE_TEXTURE_2D; + tmpl.last_level = 0; + tmpl.depth[0] = 1; + tmpl.format = format; + tmpl.width[0] = width; + tmpl.height[0] = height; + pf_get_block(tmpl.format, &tmpl.block); + + pt = api->texture_from_shared_handle(api, pscreen, &tmpl, + "front buffer", pitch, handle); + if (!pt) + return NULL; + + ps = pscreen->get_tex_surface(pscreen, pt, 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_READ | + PIPE_BUFFER_USAGE_GPU_WRITE); + + /* we don't need the texture from this point on */ + pipe_texture_reference(&pt, NULL); + return ps; +} + +static struct pipe_surface * +nouveau_dri1_front_surface(struct pipe_context *pipe) +{ + return nouveau_winsys_screen(pipe->screen)->front; +} + +static struct dri1_api nouveau_dri1_api = { + nouveau_dri1_front_surface, +}; + static struct pipe_screen * -nouveau_drm_create_screen(int fd, struct drm_create_screen_arg *arg) +nouveau_drm_create_screen(struct drm_api *api, int fd, + struct drm_create_screen_arg *arg) { - struct pipe_winsys *ws; + struct dri1_create_screen_arg *dri1 = (void *)arg; struct nouveau_winsys *nvws; + struct pipe_winsys *ws; struct nouveau_device *dev = NULL; struct pipe_screen *(*init)(struct pipe_winsys *, - struct nouveau_winsys *); + struct nouveau_device *); int ret; ret = nouveau_device_open_existing(&dev, 0, fd, 0); @@ -49,33 +99,53 @@ nouveau_drm_create_screen(int fd, struct drm_create_screen_arg *arg) return NULL; } - ws = nouveau_pipe_winsys_new(dev); - if (!ws) { + nvws = CALLOC_STRUCT(nouveau_winsys); + if (!nvws) { nouveau_device_close(&dev); return NULL; } + ws = &nvws->base; - nvws = nouveau_winsys_new(ws); - if (!nvws) { + nvws->pscreen = init(ws, dev); + if (!nvws->pscreen) { ws->destroy(ws); return NULL; } - nouveau_pipe_winsys(ws)->pscreen = init(ws, nvws); - if (!nouveau_pipe_winsys(ws)->pscreen) { - ws->destroy(ws); - return NULL; + if (arg->mode == DRM_CREATE_DRI1) { + struct nouveau_dri *nvdri = dri1->ddx_info; + enum pipe_format format; + + if (nvdri->bpp == 16) + format = PIPE_FORMAT_R5G6B5_UNORM; + else + format = PIPE_FORMAT_A8R8G8B8_UNORM; + + nvws->front = dri_surface_from_handle(api, nvws->pscreen, + nvdri->front_offset, + format, nvdri->width, + nvdri->height, + nvdri->front_pitch * + (nvdri->bpp / 8)); + if (!nvws->front) { + debug_printf("%s: error referencing front buffer\n", + __func__); + ws->destroy(ws); + return NULL; + } + + dri1->api = &nouveau_dri1_api; } - return nouveau_pipe_winsys(ws)->pscreen; + return nvws->pscreen; } static struct pipe_context * -nouveau_drm_create_context(struct pipe_screen *pscreen) +nouveau_drm_create_context(struct drm_api *api, struct pipe_screen *pscreen) { - struct nouveau_pipe_winsys *nvpws = nouveau_screen(pscreen); + struct nouveau_winsys *nvws = nouveau_winsys_screen(pscreen); struct pipe_context *(*init)(struct pipe_screen *, unsigned); - unsigned chipset = nvpws->channel->device->chipset; + unsigned chipset = nouveau_screen(pscreen)->device->chipset; int i; switch (chipset & 0xf0) { @@ -106,89 +176,88 @@ nouveau_drm_create_context(struct pipe_screen *pscreen) } /* Find a free slot for a pipe context, allocate a new one if needed */ - for (i = 0; i < nvpws->nr_pctx; i++) { - if (nvpws->pctx[i] == NULL) + for (i = 0; i < nvws->nr_pctx; i++) { + if (nvws->pctx[i] == NULL) break; } - if (i == nvpws->nr_pctx) { - nvpws->nr_pctx++; - nvpws->pctx = realloc(nvpws->pctx, - sizeof(*nvpws->pctx) * nvpws->nr_pctx); + if (i == nvws->nr_pctx) { + nvws->nr_pctx++; + nvws->pctx = realloc(nvws->pctx, + sizeof(*nvws->pctx) * nvws->nr_pctx); } - nvpws->pctx[i] = init(pscreen, i); - return nvpws->pctx[i]; -} - -static boolean -nouveau_drm_pb_from_pt(struct pipe_texture *pt, struct pipe_buffer **ppb, - unsigned *stride) -{ - return false; + nvws->pctx[i] = init(pscreen, i); + return nvws->pctx[i]; } -static struct pipe_buffer * -nouveau_drm_pb_from_handle(struct pipe_screen *pscreen, const char *name, - unsigned handle) +static struct pipe_texture * +nouveau_drm_pt_from_name(struct drm_api *api, struct pipe_screen *pscreen, + struct pipe_texture *templ, const char *name, + unsigned stride, unsigned handle) { - struct nouveau_pipe_winsys *nvpws = nouveau_screen(pscreen); - struct nouveau_device *dev = nvpws->channel->device; - struct nouveau_pipe_buffer *nvpb; + struct nouveau_device *dev = nouveau_screen(pscreen)->device; + struct pipe_buffer *pb; int ret; - nvpb = CALLOC_STRUCT(nouveau_pipe_buffer); - if (!nvpb) + pb = CALLOC(1, sizeof(struct pipe_buffer) + sizeof(struct nouveau_bo*)); + if (!pb) return NULL; - ret = nouveau_bo_handle_ref(dev, handle, &nvpb->bo); + ret = nouveau_bo_handle_ref(dev, handle, (struct nouveau_bo**)(pb+1)); if (ret) { debug_printf("%s: ref name 0x%08x failed with %d\n", __func__, handle, ret); - FREE(nvpb); + FREE(pb); return NULL; } - pipe_reference_init(&nvpb->base.reference, 1); - nvpb->base.screen = pscreen; - nvpb->base.alignment = 0; - nvpb->base.usage = PIPE_BUFFER_USAGE_GPU_READ_WRITE | - PIPE_BUFFER_USAGE_CPU_READ_WRITE; - nvpb->base.size = nvpb->bo->size; - return &nvpb->base; + pipe_reference_init(&pb->reference, 1); + pb->screen = pscreen; + pb->alignment = 0; + pb->usage = PIPE_BUFFER_USAGE_GPU_READ_WRITE | + PIPE_BUFFER_USAGE_CPU_READ_WRITE; + pb->size = nouveau_bo(pb)->size; + return pscreen->texture_blanket(pscreen, templ, &stride, pb); } static boolean -nouveau_drm_handle_from_pb(struct pipe_screen *pscreen, struct pipe_buffer *pb, - unsigned *handle) +nouveau_drm_name_from_pt(struct drm_api *api, struct pipe_screen *pscreen, + struct pipe_texture *pt, unsigned *stride, + unsigned *handle) { - struct nouveau_pipe_buffer *nvpb = nouveau_pipe_buffer(pb); + struct nouveau_miptree *mt = nouveau_miptree(pt); - if (!nvpb) - return FALSE; + if (!mt || !mt->bo) + return false; - *handle = nvpb->bo->handle; - return TRUE; + return nouveau_bo_handle_get(mt->bo, handle) == 0; } static boolean -nouveau_drm_name_from_pb(struct pipe_screen *pscreen, struct pipe_buffer *pb, - unsigned *handle) +nouveau_drm_handle_from_pt(struct drm_api *api, struct pipe_screen *pscreen, + struct pipe_texture *pt, unsigned *stride, + unsigned *handle) { - struct nouveau_pipe_buffer *nvpb = nouveau_pipe_buffer(pb); + struct nouveau_miptree *mt = nouveau_miptree(pt); - if (!nvpb) - return FALSE; + if (!mt || !mt->bo) + return false; - return nouveau_bo_handle_get(nvpb->bo, handle) == 0; + *handle = mt->bo->handle; + return true; } struct drm_api drm_api_hooks = { .create_screen = nouveau_drm_create_screen, .create_context = nouveau_drm_create_context, - .buffer_from_texture = nouveau_drm_pb_from_pt, - .buffer_from_handle = nouveau_drm_pb_from_handle, - .handle_from_buffer = nouveau_drm_handle_from_pb, - .global_handle_from_buffer = nouveau_drm_name_from_pb, + .texture_from_shared_handle = nouveau_drm_pt_from_name, + .shared_handle_from_texture = nouveau_drm_name_from_pt, + .local_handle_from_texture = nouveau_drm_handle_from_pt, }; +struct drm_api * +drm_api_create() { + return &drm_api_hooks; +} + diff --git a/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.h b/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.h index 2782c83c0e7..e61e0e0957a 100644 --- a/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.h +++ b/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.h @@ -1,5 +1,34 @@ #ifndef __NOUVEAU_DRM_API_H__ #define __NOUVEAU_DRM_API_H__ + #include "state_tracker/drm_api.h" +#include "state_tracker/dri1_api.h" + +#include "pipe/internal/p_winsys_screen.h" + +#include "nouveau_dri.h" + +struct nouveau_winsys { + struct pipe_winsys base; + + struct pipe_screen *pscreen; + + unsigned nr_pctx; + struct pipe_context **pctx; + + struct pipe_surface *front; +}; + +static INLINE struct nouveau_winsys * +nouveau_winsys(struct pipe_winsys *ws) +{ + return (struct nouveau_winsys *)ws; +} + +static INLINE struct nouveau_winsys * +nouveau_winsys_screen(struct pipe_screen *pscreen) +{ + return nouveau_winsys(pscreen->winsys); +} #endif diff --git a/src/gallium/winsys/drm/nouveau/drm/nouveau_winsys.c b/src/gallium/winsys/drm/nouveau/drm/nouveau_winsys.c deleted file mode 100644 index e3175fd7753..00000000000 --- a/src/gallium/winsys/drm/nouveau/drm/nouveau_winsys.c +++ /dev/null @@ -1,94 +0,0 @@ -#include "util/u_memory.h" - -#include "nouveau_winsys_pipe.h" - -static int -nouveau_pipe_notifier_alloc(struct nouveau_winsys *nvws, int count, - struct nouveau_notifier **notify) -{ - struct nouveau_pipe_winsys *nvpws = nouveau_pipe_winsys(nvws->ws); - - return nouveau_notifier_alloc(nvpws->channel, nvpws->next_handle++, - count, notify); -} - -static int -nouveau_pipe_grobj_alloc(struct nouveau_winsys *nvws, int grclass, - struct nouveau_grobj **grobj) -{ - struct nouveau_pipe_winsys *nvpws = nouveau_pipe_winsys(nvws->ws); - struct nouveau_channel *chan = nvpws->channel; - int ret; - - ret = nouveau_grobj_alloc(chan, nvpws->next_handle++, grclass, grobj); - if (ret) - return ret; - - BEGIN_RING(chan, *grobj, 0x0000, 1); - OUT_RING (chan, (*grobj)->handle); - (*grobj)->bound = NOUVEAU_GROBJ_BOUND_EXPLICIT; - return 0; -} - -static int -nouveau_pipe_push_reloc(struct nouveau_winsys *nvws, void *ptr, - struct pipe_buffer *buf, uint32_t data, - uint32_t flags, uint32_t vor, uint32_t tor) -{ - struct nouveau_bo *bo = nouveau_pipe_buffer(buf)->bo; - - return nouveau_pushbuf_emit_reloc(nvws->channel, ptr, bo, - data, flags, vor, tor); -} - -static int -nouveau_pipe_push_flush(struct nouveau_winsys *nvws, unsigned size, - struct pipe_fence_handle **fence) -{ - if (fence) - *fence = NULL; - - return nouveau_pushbuf_flush(nvws->channel, size); -} - -static struct nouveau_bo * -nouveau_pipe_get_bo(struct pipe_buffer *pb) -{ - return nouveau_pipe_buffer(pb)->bo; -} - -struct nouveau_winsys * -nouveau_winsys_new(struct pipe_winsys *ws) -{ - struct nouveau_pipe_winsys *nvpws = nouveau_pipe_winsys(ws); - struct nouveau_winsys *nvws; - - nvws = CALLOC_STRUCT(nouveau_winsys); - if (!nvws) - return NULL; - - nvws->ws = ws; - nvws->channel = nvpws->channel; - - nvws->res_init = nouveau_resource_init; - nvws->res_alloc = nouveau_resource_alloc; - nvws->res_free = nouveau_resource_free; - - nvws->push_reloc = nouveau_pipe_push_reloc; - nvws->push_flush = nouveau_pipe_push_flush; - - nvws->grobj_alloc = nouveau_pipe_grobj_alloc; - nvws->grobj_free = nouveau_grobj_free; - - nvws->notifier_alloc = nouveau_pipe_notifier_alloc; - nvws->notifier_free = nouveau_notifier_free; - nvws->notifier_reset = nouveau_notifier_reset; - nvws->notifier_status = nouveau_notifier_status; - nvws->notifier_retval = nouveau_notifier_return_val; - nvws->notifier_wait = nouveau_notifier_wait_status; - - nvws->get_bo = nouveau_pipe_get_bo; - - return nvws; -} - diff --git a/src/gallium/winsys/drm/nouveau/drm/nouveau_winsys_pipe.c b/src/gallium/winsys/drm/nouveau/drm/nouveau_winsys_pipe.c deleted file mode 100644 index 9e03a9f5dba..00000000000 --- a/src/gallium/winsys/drm/nouveau/drm/nouveau_winsys_pipe.c +++ /dev/null @@ -1,204 +0,0 @@ -#include "pipe/internal/p_winsys_screen.h" -#include <pipe/p_defines.h> -#include <pipe/p_inlines.h> -#include <util/u_memory.h> - -#include "nouveau_winsys_pipe.h" - -#include "nouveau_drmif.h" -#include "nouveau_bo.h" - -static const char * -nouveau_get_name(struct pipe_winsys *pws) -{ - return "Nouveau/DRI"; -} - -static uint32_t -nouveau_flags_from_usage(struct pipe_winsys *ws, unsigned usage) -{ - struct nouveau_pipe_winsys *nvpws = nouveau_pipe_winsys(ws); - struct pipe_screen *pscreen = nvpws->pscreen; - uint32_t flags = NOUVEAU_BO_LOCAL; - - if (usage & NOUVEAU_BUFFER_USAGE_TRANSFER) - flags |= NOUVEAU_BO_GART; - - if (usage & PIPE_BUFFER_USAGE_PIXEL) { - if (usage & NOUVEAU_BUFFER_USAGE_TEXTURE) - flags |= NOUVEAU_BO_GART; - if (!(usage & PIPE_BUFFER_USAGE_CPU_READ_WRITE)) - flags |= NOUVEAU_BO_VRAM; - - switch (nvpws->channel->device->chipset & 0xf0) { - case 0x50: - case 0x80: - case 0x90: - flags |= NOUVEAU_BO_TILED; - if (usage & NOUVEAU_BUFFER_USAGE_ZETA) - flags |= NOUVEAU_BO_ZTILE; - break; - default: - break; - } - } - - if (usage & PIPE_BUFFER_USAGE_VERTEX) { - if (pscreen->get_param(pscreen, NOUVEAU_CAP_HW_VTXBUF)) - flags |= NOUVEAU_BO_GART; - } - - if (usage & PIPE_BUFFER_USAGE_INDEX) { - if (pscreen->get_param(pscreen, NOUVEAU_CAP_HW_IDXBUF)) - flags |= NOUVEAU_BO_GART; - } - - return flags; -} - -static struct pipe_buffer * -nouveau_pipe_bo_create(struct pipe_winsys *ws, unsigned alignment, - unsigned usage, unsigned size) -{ - struct nouveau_pipe_winsys *nvpws = nouveau_pipe_winsys(ws); - struct nouveau_device *dev = nvpws->channel->device; - struct nouveau_pipe_buffer *nvbuf; - uint32_t flags; - - nvbuf = CALLOC_STRUCT(nouveau_pipe_buffer); - if (!nvbuf) - return NULL; - pipe_reference_init(&nvbuf->base.reference, 1); - nvbuf->base.alignment = alignment; - nvbuf->base.usage = usage; - nvbuf->base.size = size; - - flags = nouveau_flags_from_usage(ws, usage); - if (nouveau_bo_new(dev, flags, alignment, size, &nvbuf->bo)) { - FREE(nvbuf); - return NULL; - } - - return &nvbuf->base; -} - -static struct pipe_buffer * -nouveau_pipe_bo_user_create(struct pipe_winsys *ws, void *ptr, unsigned bytes) -{ - struct nouveau_pipe_winsys *nvpws = nouveau_pipe_winsys(ws); - struct nouveau_device *dev = nvpws->channel->device; - struct nouveau_pipe_buffer *nvbuf; - - nvbuf = CALLOC_STRUCT(nouveau_pipe_buffer); - if (!nvbuf) - return NULL; - pipe_reference_init(&nvbuf->base.reference, 1); - nvbuf->base.size = bytes; - - if (nouveau_bo_user(dev, ptr, bytes, &nvbuf->bo)) { - FREE(nvbuf); - return NULL; - } - - return &nvbuf->base; -} - -static void -nouveau_pipe_bo_del(struct pipe_buffer *buf) -{ - struct nouveau_pipe_buffer *nvbuf = nouveau_pipe_buffer(buf); - - nouveau_bo_ref(NULL, &nvbuf->bo); - FREE(nvbuf); -} - -static void * -nouveau_pipe_bo_map(struct pipe_winsys *pws, struct pipe_buffer *buf, - unsigned flags) -{ - struct nouveau_pipe_buffer *nvbuf = nouveau_pipe_buffer(buf); - uint32_t map_flags = 0; - - if (flags & PIPE_BUFFER_USAGE_CPU_READ) - map_flags |= NOUVEAU_BO_RD; - if (flags & PIPE_BUFFER_USAGE_CPU_WRITE) - map_flags |= NOUVEAU_BO_WR; - - if (nouveau_bo_map(nvbuf->bo, map_flags)) - return NULL; - return nvbuf->bo->map; -} - -static void -nouveau_pipe_bo_unmap(struct pipe_winsys *pws, struct pipe_buffer *buf) -{ - struct nouveau_pipe_buffer *nvbuf = nouveau_pipe_buffer(buf); - - nouveau_bo_unmap(nvbuf->bo); -} - -static void -nouveau_pipe_fence_reference(struct pipe_winsys *ws, - struct pipe_fence_handle **ptr, - struct pipe_fence_handle *pfence) -{ - *ptr = pfence; -} - -static int -nouveau_pipe_fence_signalled(struct pipe_winsys *ws, - struct pipe_fence_handle *pfence, unsigned flag) -{ - return 0; -} - -static int -nouveau_pipe_fence_finish(struct pipe_winsys *ws, - struct pipe_fence_handle *pfence, unsigned flag) -{ - return 0; -} - -static void -nouveau_destroy(struct pipe_winsys *ws) -{ - struct nouveau_pipe_winsys *nvpws = nouveau_pipe_winsys(ws); - - nouveau_device_close(&nvpws->channel->device); - FREE(nvpws); -} - -struct pipe_winsys * -nouveau_pipe_winsys_new(struct nouveau_device *dev) -{ - struct nouveau_pipe_winsys *nvpws; - int ret; - - nvpws = CALLOC_STRUCT(nouveau_pipe_winsys); - if (!nvpws) - return NULL; - - ret = nouveau_channel_alloc(dev, 0xbeef0201, 0xbeef0202, - &nvpws->channel); - if (ret) { - debug_printf("%s: error opening GPU channel: %d\n", - __func__, ret); - FREE(nvpws); - return NULL; - } - nvpws->next_handle = 0x77000000; - - nvpws->base.buffer_create = nouveau_pipe_bo_create; - nvpws->base.buffer_destroy = nouveau_pipe_bo_del; - nvpws->base.user_buffer_create = nouveau_pipe_bo_user_create; - nvpws->base.buffer_map = nouveau_pipe_bo_map; - nvpws->base.buffer_unmap = nouveau_pipe_bo_unmap; - - nvpws->base.fence_reference = nouveau_pipe_fence_reference; - nvpws->base.fence_signalled = nouveau_pipe_fence_signalled; - nvpws->base.fence_finish = nouveau_pipe_fence_finish; - - nvpws->base.get_name = nouveau_get_name; - nvpws->base.destroy = nouveau_destroy; - return &nvpws->base; -} diff --git a/src/gallium/winsys/drm/nouveau/drm/nouveau_winsys_pipe.h b/src/gallium/winsys/drm/nouveau/drm/nouveau_winsys_pipe.h deleted file mode 100644 index 10e1e269e8c..00000000000 --- a/src/gallium/winsys/drm/nouveau/drm/nouveau_winsys_pipe.h +++ /dev/null @@ -1,52 +0,0 @@ -#ifndef NOUVEAU_PIPE_WINSYS_H -#define NOUVEAU_PIPE_WINSYS_H - -#include "pipe/internal/p_winsys_screen.h" -#include "pipe/p_context.h" - -#include "nouveau/nouveau_winsys.h" - -#include "nouveau_device.h" - -struct nouveau_pipe_buffer { - struct pipe_buffer base; - struct nouveau_bo *bo; -}; - -static INLINE struct nouveau_pipe_buffer * -nouveau_pipe_buffer(struct pipe_buffer *buf) -{ - return (struct nouveau_pipe_buffer *)buf; -} - -struct nouveau_pipe_winsys { - struct pipe_winsys base; - - struct pipe_screen *pscreen; - - struct nouveau_channel *channel; - uint32_t next_handle; - - unsigned nr_pctx; - struct pipe_context **pctx; -}; - -static INLINE struct nouveau_pipe_winsys * -nouveau_pipe_winsys(struct pipe_winsys *ws) -{ - return (struct nouveau_pipe_winsys *)ws; -} - -static INLINE struct nouveau_pipe_winsys * -nouveau_screen(struct pipe_screen *pscreen) -{ - return nouveau_pipe_winsys(pscreen->winsys); -} - -struct pipe_winsys * -nouveau_pipe_winsys_new(struct nouveau_device *); - -struct nouveau_winsys * -nouveau_winsys_new(struct pipe_winsys *ws); - -#endif diff --git a/src/gallium/winsys/drm/radeon/SConscript b/src/gallium/winsys/drm/radeon/SConscript index 8f99055b2f7..b2dfd504d42 100644 --- a/src/gallium/winsys/drm/radeon/SConscript +++ b/src/gallium/winsys/drm/radeon/SConscript @@ -4,4 +4,4 @@ SConscript(['core/SConscript',]) if 'mesa' in env['statetrackers']: - SConscript(['dri2/SConscript']) + SConscript(['dri/SConscript']) diff --git a/src/gallium/winsys/drm/radeon/core/SConscript b/src/gallium/winsys/drm/radeon/core/SConscript index 578174e32ba..2ad68e403fe 100644 --- a/src/gallium/winsys/drm/radeon/core/SConscript +++ b/src/gallium/winsys/drm/radeon/core/SConscript @@ -11,7 +11,9 @@ radeon_sources = [ env.Append(CPPPATH = '#/src/gallium/drivers/r300') -env.ConvenienceLibrary( +radeonwinsys = env.ConvenienceLibrary( target ='radeonwinsys', source = radeon_sources, ) + +Export('radeonwinsys') diff --git a/src/gallium/winsys/drm/radeon/core/radeon_buffer.c b/src/gallium/winsys/drm/radeon/core/radeon_buffer.c index a15487352b2..07551e7cd16 100644 --- a/src/gallium/winsys/drm/radeon/core/radeon_buffer.c +++ b/src/gallium/winsys/drm/radeon/core/radeon_buffer.c @@ -1,8 +1,8 @@ -/* +/* * Copyright © 2008 Jérôme Glisse * 2009 Corbin Simpson * All Rights Reserved. - * + * * Permission is hereby granted, free of charge, to any person obtaining * a copy of this software and associated documentation files (the * "Software"), to deal in the Software without restriction, including @@ -10,14 +10,14 @@ * distribute, sub license, and/or sell copies of the Software, and to * permit persons to whom the Software is furnished to do so, subject to * the following conditions: - * + * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. * * The above copyright notice and this permission notice (including the @@ -72,6 +72,7 @@ static struct pipe_buffer *radeon_buffer_create(struct pipe_winsys *ws, alignment, domain, 0); if (radeon_buffer->bo == NULL) { FREE(radeon_buffer); + return NULL; } return &radeon_buffer->base; } @@ -93,6 +94,29 @@ static struct pipe_buffer *radeon_buffer_user_create(struct pipe_winsys *ws, return &radeon_buffer->base; } +static struct pipe_buffer *radeon_surface_buffer_create(struct pipe_winsys *ws, + unsigned width, + unsigned height, + enum pipe_format format, + unsigned usage, + unsigned *stride) +{ + struct pipe_format_block block; + unsigned nblocksx, nblocksy, size; + + pf_get_block(format, &block); + + nblocksx = pf_get_nblocksx(&block, width); + nblocksy = pf_get_nblocksy(&block, height); + + /* Radeons enjoy things in multiples of 32. */ + /* XXX this can be 32 when POT */ + *stride = (nblocksx * block.size + 63) & ~63; + size = *stride * nblocksy; + + return radeon_buffer_create(ws, 64, usage, size); +} + static void radeon_buffer_del(struct pipe_buffer *buffer) { struct radeon_pipe_buffer *radeon_buffer = @@ -110,16 +134,17 @@ static void *radeon_buffer_map(struct pipe_winsys *ws, (struct radeon_pipe_buffer*)buffer; int write = 0; - if (flags & PIPE_BUFFER_USAGE_DONTBLOCK) { - /* XXX Remove this when radeon_bo_map supports DONTBLOCK */ - return NULL; + if (!(flags & PIPE_BUFFER_USAGE_DONTBLOCK)) { + radeon_bo_wait(radeon_buffer->bo); } if (flags & PIPE_BUFFER_USAGE_CPU_WRITE) { write = 1; } - if (radeon_bo_map(radeon_buffer->bo, write)) + if (radeon_bo_map(radeon_buffer->bo, write)) { return NULL; + } + return radeon_buffer->bo->ptr; } @@ -175,15 +200,17 @@ struct radeon_winsys* radeon_pipe_winsys(int fd) return NULL; } + radeon_ws->priv->fd = fd; radeon_ws->priv->bom = radeon_bo_manager_gem_ctor(fd); radeon_ws->base.flush_frontbuffer = radeon_flush_frontbuffer; radeon_ws->base.buffer_create = radeon_buffer_create; - radeon_ws->base.buffer_destroy = radeon_buffer_del; radeon_ws->base.user_buffer_create = radeon_buffer_user_create; + radeon_ws->base.surface_buffer_create = radeon_surface_buffer_create; radeon_ws->base.buffer_map = radeon_buffer_map; radeon_ws->base.buffer_unmap = radeon_buffer_unmap; + radeon_ws->base.buffer_destroy = radeon_buffer_del; radeon_ws->base.fence_reference = radeon_fence_reference; radeon_ws->base.fence_signalled = radeon_fence_signalled; diff --git a/src/gallium/winsys/drm/radeon/core/radeon_buffer.h b/src/gallium/winsys/drm/radeon/core/radeon_buffer.h index ca8bbb3c117..f5153b06af5 100644 --- a/src/gallium/winsys/drm/radeon/core/radeon_buffer.h +++ b/src/gallium/winsys/drm/radeon/core/radeon_buffer.h @@ -1,7 +1,7 @@ -/* +/* * Copyright © 2008 Jérôme Glisse * All Rights Reserved. - * + * * Permission is hereby granted, free of charge, to any person obtaining * a copy of this software and associated documentation files (the * "Software"), to deal in the Software without restriction, including @@ -9,14 +9,14 @@ * distribute, sub license, and/or sell copies of the Software, and to * permit persons to whom the Software is furnished to do so, subject to * the following conditions: - * + * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. * * The above copyright notice and this permission notice (including the @@ -48,19 +48,19 @@ struct radeon_pipe_buffer { struct pipe_buffer base; struct radeon_bo *bo; + boolean flinked; + uint32_t flink; }; #define RADEON_MAX_BOS 24 struct radeon_winsys_priv { + /* DRM FD */ + int fd; + /* Radeon BO manager. */ struct radeon_bo_manager* bom; - /* Radeon BO space checker. */ - struct radeon_cs_space_check sc[RADEON_MAX_BOS]; - /* Current BO count. */ - unsigned bo_count; - /* Radeon CS manager. */ struct radeon_cs_manager* csm; diff --git a/src/gallium/winsys/drm/radeon/core/radeon_drm.c b/src/gallium/winsys/drm/radeon/core/radeon_drm.c index 428d3f65a16..47376a0f07b 100644 --- a/src/gallium/winsys/drm/radeon/core/radeon_drm.c +++ b/src/gallium/winsys/drm/radeon/core/radeon_drm.c @@ -1,7 +1,7 @@ -/* +/* * Copyright © 2009 Corbin Simpson * All Rights Reserved. - * + * * Permission is hereby granted, free of charge, to any person obtaining * a copy of this software and associated documentation files (the * "Software"), to deal in the Software without restriction, including @@ -9,14 +9,14 @@ * distribute, sub license, and/or sell copies of the Software, and to * permit persons to whom the Software is furnished to do so, subject to * the following conditions: - * + * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. * * The above copyright notice and this permission notice (including the @@ -26,13 +26,15 @@ /* * Authors: * Corbin Simpson <MostAwesomeDude@gmail.com> + * Joakim Sindholt <opensource@zhasha.com> */ #include "radeon_drm.h" /* Create a pipe_screen. */ -struct pipe_screen* radeon_create_screen(int drmFB, - struct drm_create_screen_arg *arg) +struct pipe_screen* radeon_create_screen(struct drm_api* api, + int drmFB, + struct drm_create_screen_arg *arg) { struct radeon_winsys* winsys = radeon_pipe_winsys(drmFB); @@ -46,25 +48,30 @@ struct pipe_screen* radeon_create_screen(int drmFB, } /* Create a pipe_context. */ -struct pipe_context* radeon_create_context(struct pipe_screen* screen) +struct pipe_context* radeon_create_context(struct drm_api* api, + struct pipe_screen* screen) { if (getenv("RADEON_SOFTPIPE")) { return radeon_create_softpipe(screen->winsys); } else { - return r300_create_context(screen, screen->winsys); + return r300_create_context(screen, + (struct r300_winsys*)screen->winsys); } } -boolean radeon_buffer_from_texture(struct pipe_texture* texture, +boolean radeon_buffer_from_texture(struct drm_api* api, + struct pipe_texture* texture, struct pipe_buffer** buffer, unsigned* stride) { - return FALSE; + /* XXX fix this */ + return r300_get_texture_buffer(texture, buffer, stride); } /* Create a buffer from a handle. */ /* XXX what's up with name? */ -struct pipe_buffer* radeon_buffer_from_handle(struct pipe_screen* screen, +struct pipe_buffer* radeon_buffer_from_handle(struct drm_api* api, + struct pipe_screen* screen, const char* name, unsigned handle) { @@ -91,33 +98,89 @@ struct pipe_buffer* radeon_buffer_from_handle(struct pipe_screen* screen, return &radeon_buffer->base; } -boolean radeon_handle_from_buffer(struct pipe_screen* screen, - struct pipe_buffer* buffer, - unsigned* handle) +struct pipe_texture* +radeon_texture_from_shared_handle(struct drm_api *api, + struct pipe_screen *screen, + struct pipe_texture *templ, + const char *name, + unsigned stride, + unsigned handle) +{ + struct pipe_buffer *buffer; + + buffer = radeon_buffer_from_handle(api, screen, name, handle); + if (!buffer) { + return NULL; + } + + return screen->texture_blanket(screen, templ, &stride, buffer); +} + +boolean radeon_shared_handle_from_texture(struct drm_api *api, + struct pipe_screen *screen, + struct pipe_texture *texture, + unsigned *stride, + unsigned *handle) { - struct radeon_pipe_buffer* radeon_buffer = - (struct radeon_pipe_buffer*)buffer; - *handle = radeon_buffer->bo->handle; + int retval, fd; + struct drm_gem_flink flink; + struct radeon_pipe_buffer* radeon_buffer; + struct pipe_buffer* buffer = &radeon_buffer->base; + if (!radeon_buffer_from_texture(api, texture, buffer, stride)) { + return FALSE; + } + + if (!radeon_buffer->flinked) { + fd = ((struct radeon_winsys*)screen->winsys)->priv->fd; + + flink.handle = radeon_buffer->bo->handle; + + retval = ioctl(fd, DRM_IOCTL_GEM_FLINK, &flink); + if (retval) { + debug_printf("radeon: DRM_IOCTL_GEM_FLINK failed, error %d\n", + retval); + return FALSE; + } + + radeon_buffer->flink = flink.name; + radeon_buffer->flinked = TRUE; + } + + *handle = radeon_buffer->flink; return TRUE; } -boolean radeon_global_handle_from_buffer(struct pipe_screen* screen, - struct pipe_buffer* buffer, - unsigned* handle) +boolean radeon_local_handle_from_texture(struct drm_api *api, + struct pipe_screen *screen, + struct pipe_texture *texture, + unsigned *stride, + unsigned *handle) { - /* XXX WTF is the difference here? global? */ - struct radeon_pipe_buffer* radeon_buffer = - (struct radeon_pipe_buffer*)buffer; - *handle = radeon_buffer->bo->handle; + struct pipe_buffer *buffer; + if (!radeon_buffer_from_texture(api, texture, &buffer, stride)) { + return FALSE; + } + + *handle = ((struct radeon_pipe_buffer*)buffer)->bo->handle; + + pipe_buffer_reference(&buffer, NULL); + return TRUE; } struct drm_api drm_api_hooks = { .create_screen = radeon_create_screen, .create_context = radeon_create_context, - /* XXX fix this */ - .buffer_from_texture = r300_get_texture_buffer, - .buffer_from_handle = radeon_buffer_from_handle, - .handle_from_buffer = radeon_handle_from_buffer, - .global_handle_from_buffer = radeon_global_handle_from_buffer, + .texture_from_shared_handle = radeon_texture_from_shared_handle, + .shared_handle_from_texture = radeon_shared_handle_from_texture, + .local_handle_from_texture = radeon_local_handle_from_texture, }; + +struct drm_api* drm_api_create() +{ +#ifdef DEBUG + return trace_drm_create(&drm_api_hooks); +#else + return &drm_api_hooks; +#endif +} diff --git a/src/gallium/winsys/drm/radeon/core/radeon_drm.h b/src/gallium/winsys/drm/radeon/core/radeon_drm.h index 049f9984dba..88a5c82b284 100644 --- a/src/gallium/winsys/drm/radeon/core/radeon_drm.h +++ b/src/gallium/winsys/drm/radeon/core/radeon_drm.h @@ -30,8 +30,13 @@ #ifndef RADEON_DRM_H #define RADEON_DRM_H +#include <sys/ioctl.h> + +#include "xf86drm.h" + #include "pipe/p_screen.h" +#include "trace/tr_drm.h" #include "util/u_memory.h" #include "state_tracker/drm_api.h" @@ -40,25 +45,35 @@ #include "radeon_r300.h" #include "radeon_winsys_softpipe.h" -struct pipe_screen* radeon_create_screen(int drmFB, +/* XXX */ +#include "r300_screen.h" + +struct pipe_screen* radeon_create_screen(struct drm_api* api, + int drmFB, struct drm_create_screen_arg *arg); -struct pipe_context* radeon_create_context(struct pipe_screen* screen); +struct pipe_context* radeon_create_context(struct drm_api* api, + struct pipe_screen* screen); -boolean radeon_buffer_from_texture(struct pipe_texture* texture, +boolean radeon_buffer_from_texture(struct drm_api* api, + struct pipe_texture* texture, struct pipe_buffer** buffer, unsigned* stride); -struct pipe_buffer* radeon_buffer_from_handle(struct pipe_screen* screen, +struct pipe_buffer* radeon_buffer_from_handle(struct drm_api* api, + struct pipe_screen* screen, const char* name, unsigned handle); -boolean radeon_handle_from_buffer(struct pipe_screen* screen, +boolean radeon_handle_from_buffer(struct drm_api* api, + struct pipe_screen* screen, struct pipe_buffer* buffer, unsigned* handle); -boolean radeon_global_handle_from_buffer(struct pipe_screen* screen, +boolean radeon_global_handle_from_buffer(struct drm_api* api, + struct pipe_screen* screen, struct pipe_buffer* buffer, unsigned* handle); +void radeon_destroy_drm_api(struct drm_api* api); #endif diff --git a/src/gallium/winsys/drm/radeon/core/radeon_r300.c b/src/gallium/winsys/drm/radeon/core/radeon_r300.c index da233203d7d..d7238762219 100644 --- a/src/gallium/winsys/drm/radeon/core/radeon_r300.c +++ b/src/gallium/winsys/drm/radeon/core/radeon_r300.c @@ -22,54 +22,30 @@ #include "radeon_r300.h" -static void radeon_r300_add_buffer(struct r300_winsys* winsys, - struct pipe_buffer* pbuffer, - uint32_t rd, - uint32_t wd) +static boolean radeon_r300_add_buffer(struct r300_winsys* winsys, + struct pipe_buffer* pbuffer, + uint32_t rd, + uint32_t wd) { - int i; struct radeon_winsys_priv* priv = (struct radeon_winsys_priv*)winsys->radeon_winsys; - struct radeon_cs_space_check* sc = priv->sc; struct radeon_bo* bo = ((struct radeon_pipe_buffer*)pbuffer)->bo; - /* Check to see if this BO is already in line for validation; - * find a slot for it otherwise. */ - for (i = 0; i < RADEON_MAX_BOS; i++) { - if (sc[i].bo == bo) { - return; - } else if (sc[i].bo == NULL) { - sc[i].bo = bo; - sc[i].read_domains = rd; - sc[i].write_domain = wd; - priv->bo_count = i + 1; - return; - } - } - - assert(FALSE && "Oh God too many BOs!"); + radeon_cs_space_add_persistent_bo(priv->cs, bo, rd, wd); + return TRUE; } static boolean radeon_r300_validate(struct r300_winsys* winsys) { - int retval; struct radeon_winsys_priv* priv = (struct radeon_winsys_priv*)winsys->radeon_winsys; - struct radeon_cs_space_check* sc = priv->sc; - retval = radeon_cs_space_check(priv->cs, sc, priv->bo_count); - - if (retval == RADEON_CS_SPACE_OP_TO_BIG) { - /* We might as well HCF, since this is not going to fit in the card, - * period. */ - exit(1); - } else if (retval == RADEON_CS_SPACE_FLUSH) { - /* We must flush before more rendering can commence. */ - return TRUE; + if (radeon_cs_space_check(priv->cs) < 0) { + return FALSE; } /* Things are fine, we can proceed as normal. */ - return FALSE; + return TRUE; } static boolean radeon_r300_check_cs(struct r300_winsys* winsys, int size) @@ -108,9 +84,15 @@ static void radeon_r300_write_cs_reloc(struct r300_winsys* winsys, { struct radeon_winsys_priv* priv = (struct radeon_winsys_priv*)winsys->radeon_winsys; + int retval = 0; - radeon_cs_write_reloc(priv->cs, + retval = radeon_cs_write_reloc(priv->cs, ((struct radeon_pipe_buffer*)pbuffer)->bo, rd, wd, flags); + + if (retval) { + debug_printf("radeon: Relocation of %p (%d, %d, %d) failed!\n", + pbuffer, rd, wd, flags); + } } static void radeon_r300_end_cs(struct r300_winsys* winsys, @@ -128,57 +110,66 @@ static void radeon_r300_flush_cs(struct r300_winsys* winsys) { struct radeon_winsys_priv* priv = (struct radeon_winsys_priv*)winsys->radeon_winsys; - int retval = 0; + int retval; + /* Emit the CS. */ retval = radeon_cs_emit(priv->cs); if (retval) { debug_printf("radeon: Bad CS, dumping...\n"); radeon_cs_print(priv->cs, stderr); } + + /* Clean out BOs. */ + radeon_cs_space_reset_bos(priv->cs); + + /* Reset CS. + * Someday, when we care about performance, we should really find a way + * to rotate between two or three CS objects so that the GPU can be + * spinning through one CS while another one is being filled. */ radeon_cs_erase(priv->cs); } /* Helper function to do the ioctls needed for setup and init. */ static void do_ioctls(struct r300_winsys* winsys, int fd) { - struct drm_radeon_gem_info info; - drm_radeon_getparam_t gp; - int target; + struct drm_radeon_gem_info gem_info = {0}; + struct drm_radeon_info info = {0}; + int target = 0; int retval; - gp.value = ⌖ + info.value = ⌖ /* First, get the number of pixel pipes */ - gp.param = RADEON_PARAM_NUM_GB_PIPES; - retval = drmCommandWriteRead(fd, DRM_RADEON_GETPARAM, &gp, sizeof(gp)); + info.request = RADEON_INFO_NUM_GB_PIPES; + retval = drmCommandWriteRead(fd, DRM_RADEON_INFO, &info, sizeof(info)); if (retval) { - fprintf(stderr, "%s: Failed to get GB pipe count, error number %d\n", - __FUNCTION__, retval); + fprintf(stderr, "%s: Failed to get GB pipe count, " + "error number %d\n", __FUNCTION__, retval); exit(1); } winsys->gb_pipes = target; /* Then, get PCI ID */ - gp.param = RADEON_PARAM_DEVICE_ID; - retval = drmCommandWriteRead(fd, DRM_RADEON_GETPARAM, &gp, sizeof(gp)); + info.request = RADEON_INFO_DEVICE_ID; + retval = drmCommandWriteRead(fd, DRM_RADEON_INFO, &info, sizeof(info)); if (retval) { - fprintf(stderr, "%s: Failed to get PCI ID, error number %d\n", - __FUNCTION__, retval); + fprintf(stderr, "%s: Failed to get PCI ID, " + "error number %d\n", __FUNCTION__, retval); exit(1); } winsys->pci_id = target; /* Finally, retrieve MM info */ retval = drmCommandWriteRead(fd, DRM_RADEON_GEM_INFO, - &info, sizeof(info)); + &gem_info, sizeof(gem_info)); if (retval) { fprintf(stderr, "%s: Failed to get MM info, error number %d\n", __FUNCTION__, retval); exit(1); } - winsys->gart_size = info.gart_size; + winsys->gart_size = gem_info.gart_size; /* XXX */ - winsys->vram_size = info.vram_visible; + winsys->vram_size = gem_info.vram_visible; } struct r300_winsys* diff --git a/src/gallium/winsys/drm/radeon/core/radeon_r300.h b/src/gallium/winsys/drm/radeon/core/radeon_r300.h index 5c373cd0848..775d7937fd8 100644 --- a/src/gallium/winsys/drm/radeon/core/radeon_r300.h +++ b/src/gallium/winsys/drm/radeon/core/radeon_r300.h @@ -20,16 +20,23 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ +#ifndef RADEON_R300_H +#define RADEON_R300_H + /* XXX WTF is this! I shouldn't have to include those first three! FUCK! */ #include <stdint.h> #include <stdlib.h> #include "drm.h" #include "radeon_drm.h" -#include "radeon_cs.h" +#include "radeon_cs_gem.h" #include "r300_winsys.h" #include "radeon_buffer.h" +struct radeon_winsys; + struct r300_winsys* radeon_create_r300_winsys(int fd, struct radeon_winsys* old_winsys); + +#endif /* RADEON_R300_H */ diff --git a/src/gallium/winsys/drm/radeon/dri/Makefile b/src/gallium/winsys/drm/radeon/dri/Makefile index c218ee9d010..a9889444de8 100644 --- a/src/gallium/winsys/drm/radeon/dri/Makefile +++ b/src/gallium/winsys/drm/radeon/dri/Makefile @@ -10,6 +10,7 @@ PIPE_DRIVERS = \ $(TOP)/src/gallium/state_trackers/dri/libdridrm.a \ $(TOP)/src/gallium/winsys/drm/radeon/core/libradeonwinsys.a \ $(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \ + $(TOP)/src/gallium/drivers/trace/libtrace.a \ $(TOP)/src/gallium/drivers/r300/libr300.a C_SOURCES = \ diff --git a/src/gallium/winsys/drm/radeon/dri/SConscript b/src/gallium/winsys/drm/radeon/dri/SConscript index f2cdee97d92..aea987a3aca 100644 --- a/src/gallium/winsys/drm/radeon/dri/SConscript +++ b/src/gallium/winsys/drm/radeon/dri/SConscript @@ -2,7 +2,10 @@ Import('*') env = drienv.Clone() +env.ParseConfig('pkg-config --cflags --libs libdrm_radeon') + drivers = [ + trace, softpipe, r300 ] @@ -10,5 +13,5 @@ drivers = [ env.SharedLibrary( target ='radeon_dri.so', source = COMMON_GALLIUM_SOURCES, - LIBS = drivers + mesa + auxiliaries + env['LIBS'], + LIBS = st_dri + radeonwinsys + mesa + drivers + auxiliaries + env['LIBS'], ) diff --git a/src/gallium/winsys/drm/radeon/egl/Makefile b/src/gallium/winsys/drm/radeon/egl/Makefile index d989b3aa935..6a1448d1b9b 100644 --- a/src/gallium/winsys/drm/radeon/egl/Makefile +++ b/src/gallium/winsys/drm/radeon/egl/Makefile @@ -8,6 +8,7 @@ PIPE_DRIVERS = \ $(TOP)/src/gallium/state_trackers/egl/libegldrm.a \ $(GALLIUMDIR)/winsys/drm/radeon/core/libradeonwinsys.a \ $(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \ + $(TOP)/src/gallium/drivers/trace/libtrace.a \ $(TOP)/src/gallium/drivers/r300/libr300.a DRIVER_SOURCES = diff --git a/src/gallium/winsys/drm/radeon/xorg/Makefile b/src/gallium/winsys/drm/radeon/xorg/Makefile index 6ffd4a3a541..0241625f69b 100644 --- a/src/gallium/winsys/drm/radeon/xorg/Makefile +++ b/src/gallium/winsys/drm/radeon/xorg/Makefile @@ -37,6 +37,6 @@ clean: install: $(INSTALL) -d $(DESTDIR)/$(XORG_DRIVER_INSTALL_DIR) - $(INSTALL) -m 755 $(TARGET) $(DESTDIR)/$(XORG_DRIVER_INSTALL_DIR) + $(MINSTALL) -m 755 $(TARGET) $(DESTDIR)/$(XORG_DRIVER_INSTALL_DIR) .PHONY = all clean install diff --git a/src/gallium/winsys/drm/radeon/xorg/radeon_xorg.c b/src/gallium/winsys/drm/radeon/xorg/radeon_xorg.c index 36824251f0c..837f2aa8fec 100644 --- a/src/gallium/winsys/drm/radeon/xorg/radeon_xorg.c +++ b/src/gallium/winsys/drm/radeon/xorg/radeon_xorg.c @@ -48,8 +48,8 @@ static SymTabRec radeon_xorg_chipsets[] = { }; static PciChipsets radeon_xorg_pci_devices[] = { - {PCI_MATCH_ANY, PCI_MATCH_ANY, RES_SHARED_VGA}, - {-1, -1, RES_UNDEFINED} + {PCI_MATCH_ANY, PCI_MATCH_ANY, NULL}, + {-1, -1, NULL} }; static XF86ModuleVersionInfo radeon_xorg_version = { @@ -106,12 +106,6 @@ radeon_xorg_setup(pointer module, pointer opts, int *errmaj, int *errmin) xf86AddDriver(&modesetting, module, HaveDriverFuncs); /* - * Tell the loader about symbols from other modules that this module - * might refer to. - */ - xorg_tracker_loader_ref_sym_lists(); - - /* * The return value must be non-NULL on success even though there * is no TearDownProc. */ @@ -141,7 +135,7 @@ radeon_xorg_pci_probe(DriverPtr driver, NULL, NULL, NULL, NULL, NULL); if (scrn != NULL) { scrn->driverVersion = 1; - scrn->driverName = "modesetting"; + scrn->driverName = "radeon"; scrn->name = "modesetting"; scrn->Probe = NULL; diff --git a/src/gallium/winsys/egl_xlib/Makefile b/src/gallium/winsys/egl_xlib/Makefile index 8646ee3b527..3efb7ed4afa 100644 --- a/src/gallium/winsys/egl_xlib/Makefile +++ b/src/gallium/winsys/egl_xlib/Makefile @@ -37,7 +37,7 @@ UNUSED_LIBS = \ $(TOP)/src/mesa/libmesagallium.a \ -LOCAL_CFLAGS = -D_EGL_PLATFORM_X=1 +LOCAL_CFLAGS = .c.o: @@ -74,7 +74,7 @@ depend: $(ALL_SOURCES) install: default $(INSTALL) -d $(INSTALL_DIR)/$(LIB_DIR) @if [ -e $(TOP)/$(LIB_DIR) ]; then \ - $(INSTALL) $(TOP)/$(LIB_DIR)/$(DRIVER_NAME) $(INSTALL_DIR)/$(LIB_DIR); \ + $(MINSTALL) $(TOP)/$(LIB_DIR)/$(DRIVER_NAME) $(INSTALL_DIR)/$(LIB_DIR); \ fi diff --git a/src/gallium/winsys/egl_xlib/egl_xlib.c b/src/gallium/winsys/egl_xlib/egl_xlib.c index b52f427e4a7..d02f8250478 100644 --- a/src/gallium/winsys/egl_xlib/egl_xlib.c +++ b/src/gallium/winsys/egl_xlib/egl_xlib.c @@ -33,6 +33,7 @@ #include <dlfcn.h> +#include <X11/Xlib.h> #include <X11/Xutil.h> #include "pipe/p_compiler.h" @@ -61,6 +62,14 @@ struct xlib_egl_driver { _EGLDriver Base; /**< base class */ + EGLint apis; +}; + + +/** driver data of _EGLDisplay */ +struct xlib_egl_display +{ + Display *Dpy; struct pipe_winsys *winsys; struct pipe_screen *screen; @@ -82,6 +91,7 @@ struct xlib_egl_surface { _EGLSurface Base; /**< base class */ + /* These are set for window surface */ Display *Dpy; /**< The X Display of the window */ Window Win; /**< The user-created window ID */ GC Gc; @@ -93,6 +103,12 @@ struct xlib_egl_surface }; +static void +flush_frontbuffer(struct pipe_winsys *pws, + struct pipe_surface *psurf, + void *context_private); + + /** cast wrapper */ static INLINE struct xlib_egl_driver * xlib_egl_driver(_EGLDriver *drv) @@ -101,19 +117,24 @@ xlib_egl_driver(_EGLDriver *drv) } -static struct xlib_egl_surface * -lookup_surface(EGLSurface surf) +static INLINE struct xlib_egl_display * +xlib_egl_display(_EGLDisplay *dpy) { - _EGLSurface *surface = _eglLookupSurface(surf); - return (struct xlib_egl_surface *) surface; + return (struct xlib_egl_display *) dpy->DriverData; } -static struct xlib_egl_context * -lookup_context(EGLContext surf) +static INLINE struct xlib_egl_surface * +lookup_surface(_EGLSurface *surf) { - _EGLContext *context = _eglLookupContext(surf); - return (struct xlib_egl_context *) context; + return (struct xlib_egl_surface *) surf; +} + + +static INLINE struct xlib_egl_context * +lookup_context(_EGLContext *ctx) +{ + return (struct xlib_egl_context *) ctx; } @@ -132,19 +153,18 @@ bitcount(unsigned int n) * Create the EGLConfigs. (one per X visual) */ static void -create_configs(_EGLDriver *drv, EGLDisplay dpy) +create_configs(struct xlib_egl_display *xdpy, _EGLDisplay *disp) { static const EGLint all_apis = (EGL_OPENGL_ES_BIT | EGL_OPENGL_ES2_BIT | EGL_OPENVG_BIT | EGL_OPENGL_BIT); - _EGLDisplay *disp = _eglLookupDisplay(dpy); XVisualInfo *visInfo, visTemplate; int num_visuals, i; /* get list of all X visuals, create an EGL config for each */ - visTemplate.screen = DefaultScreen(disp->Xdpy); - visInfo = XGetVisualInfo(disp->Xdpy, VisualScreenMask, + visTemplate.screen = DefaultScreen(xdpy->Dpy); + visInfo = XGetVisualInfo(xdpy->Dpy, VisualScreenMask, &visTemplate, &num_visuals); if (!visInfo) { printf("egl_xlib.c: couldn't get any X visuals\n"); @@ -180,10 +200,14 @@ create_configs(_EGLDriver *drv, EGLDisplay dpy) SET_CONFIG_ATTRIB(config, EGL_NATIVE_RENDERABLE, EGL_FALSE); SET_CONFIG_ATTRIB(config, EGL_CONFORMANT, all_apis); SET_CONFIG_ATTRIB(config, EGL_RENDERABLE_TYPE, all_apis); - SET_CONFIG_ATTRIB(config, EGL_SURFACE_TYPE, EGL_WINDOW_BIT); + SET_CONFIG_ATTRIB(config, EGL_SURFACE_TYPE, EGL_WINDOW_BIT | EGL_PBUFFER_BIT); + SET_CONFIG_ATTRIB(config, EGL_BIND_TO_TEXTURE_RGBA, EGL_TRUE); + SET_CONFIG_ATTRIB(config, EGL_BIND_TO_TEXTURE_RGB, EGL_TRUE); _eglAddConfig(disp, config); } + + XFree(visInfo); } @@ -191,16 +215,47 @@ create_configs(_EGLDriver *drv, EGLDisplay dpy) * Called via eglInitialize(), drv->API.Initialize(). */ static EGLBoolean -xlib_eglInitialize(_EGLDriver *drv, EGLDisplay dpy, - EGLint *minor, EGLint *major) +xlib_eglInitialize(_EGLDriver *drv, _EGLDisplay *dpy, + EGLint *major, EGLint *minor) { - create_configs(drv, dpy); + struct xlib_egl_driver *xdrv = xlib_egl_driver(drv); + struct xlib_egl_display *xdpy; + + xdpy = CALLOC_STRUCT(xlib_egl_display); + if (!xdpy) + return _eglError(EGL_BAD_ALLOC, "eglInitialize"); + + xdpy->Dpy = (Display *) dpy->NativeDisplay; + if (!xdpy->Dpy) { + xdpy->Dpy = XOpenDisplay(NULL); + if (!xdpy->Dpy) { + free(xdpy); + return EGL_FALSE; + } + } - drv->Initialized = EGL_TRUE; + /* create winsys and pipe screen */ + xdpy->winsys = create_sw_winsys(); + if (!xdpy->winsys) { + free(xdpy); + return _eglError(EGL_BAD_ALLOC, "eglInitialize"); + } + xdpy->winsys->flush_frontbuffer = flush_frontbuffer; + xdpy->screen = softpipe_create_screen(xdpy->winsys); + if (!xdpy->screen) { + free(xdpy->winsys); + free(xdpy); + return _eglError(EGL_BAD_ALLOC, "eglInitialize"); + } + + dpy->DriverData = (void *) xdpy; + dpy->ClientAPIsMask = xdrv->apis; + + create_configs(xdpy, dpy); /* we're supporting EGL 1.4 */ - *minor = 1; - *major = 4; + *major = 1; + *minor = 4; return EGL_TRUE; } @@ -210,8 +265,20 @@ xlib_eglInitialize(_EGLDriver *drv, EGLDisplay dpy, * Called via eglTerminate(), drv->API.Terminate(). */ static EGLBoolean -xlib_eglTerminate(_EGLDriver *drv, EGLDisplay dpy) +xlib_eglTerminate(_EGLDriver *drv, _EGLDisplay *dpy) { + struct xlib_egl_display *xdpy = xlib_egl_display(dpy); + + _eglReleaseDisplayResources(drv, dpy); + _eglCleanupDisplay(dpy); + + xdpy->screen->destroy(xdpy->screen); + free(xdpy->winsys); + + if (!dpy->NativeDisplay) + XCloseDisplay(xdpy->Dpy); + free(xdpy); + return EGL_TRUE; } @@ -264,7 +331,13 @@ static void check_and_update_buffer_size(struct xlib_egl_surface *surface) { uint width, height; - get_drawable_size(surface->Dpy, surface->Win, &width, &height); + if (surface->Base.Type == EGL_PBUFFER_BIT) { + width = surface->Base.Width; + height = surface->Base.Height; + } + else { + get_drawable_size(surface->Dpy, surface->Win, &width, &height); + } st_resize_framebuffer(surface->Framebuffer, width, height); surface->Base.Width = width; surface->Base.Height = height; @@ -281,6 +354,9 @@ display_surface(struct pipe_winsys *pws, XImage *ximage; void *data; + if (xsurf->Base.Type == EGL_PBUFFER_BIT) + return; + ximage = XCreateImage(xsurf->Dpy, xsurf->VisInfo.visual, xsurf->VisInfo.depth, @@ -330,24 +406,23 @@ flush_frontbuffer(struct pipe_winsys *pws, /** * Called via eglCreateContext(), drv->API.CreateContext(). */ -static EGLContext -xlib_eglCreateContext(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, - EGLContext share_list, const EGLint *attrib_list) +static _EGLContext * +xlib_eglCreateContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, + _EGLContext *share_list, const EGLint *attrib_list) { - struct xlib_egl_driver *xdrv = xlib_egl_driver(drv); - _EGLConfig *conf = _eglLookupConfig(drv, dpy, config); + struct xlib_egl_display *xdpy = xlib_egl_display(dpy); struct xlib_egl_context *ctx; struct st_context *share_ctx = NULL; /* XXX fix */ __GLcontextModes visual; ctx = CALLOC_STRUCT(xlib_egl_context); if (!ctx) - return EGL_NO_CONTEXT; + return NULL; /* let EGL lib init the common stuff */ - if (!_eglInitContext(drv, dpy, &ctx->Base, config, attrib_list)) { + if (!_eglInitContext(drv, &ctx->Base, conf, attrib_list)) { free(ctx); - return EGL_NO_CONTEXT; + return NULL; } /* API-dependent context creation */ @@ -359,7 +434,7 @@ xlib_eglCreateContext(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, /* fall-through */ case EGL_OPENGL_API: /* create a softpipe context */ - ctx->pipe = softpipe_create(xdrv->screen); + ctx->pipe = softpipe_create(xdpy->screen); /* Now do xlib / state tracker inits here */ _eglConfigToContextModesRec(conf, &visual); ctx->Context = st_create_context(ctx->pipe, &visual, share_ctx); @@ -367,43 +442,33 @@ xlib_eglCreateContext(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, default: _eglError(EGL_BAD_MATCH, "eglCreateContext(unsupported API)"); free(ctx); - return EGL_NO_CONTEXT; + return NULL; } - _eglSaveContext(&ctx->Base); - - return _eglGetContextHandle(&ctx->Base); + return &ctx->Base; } static EGLBoolean -xlib_eglDestroyContext(_EGLDriver *drv, EGLDisplay dpy, EGLContext ctx) +xlib_eglDestroyContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx) { struct xlib_egl_context *context = lookup_context(ctx); - if (context) { - if (context->Base.IsBound) { - context->Base.DeletePending = EGL_TRUE; - } - else { - /* API-dependent clean-up */ - switch (context->Base.ClientAPI) { - case EGL_OPENGL_ES_API: - case EGL_OPENVG_API: - /* fall-through */ - case EGL_OPENGL_API: - st_destroy_context(context->Context); - break; - default: - assert(0); - } - free(context); + + if (!_eglIsContextBound(&context->Base)) { + /* API-dependent clean-up */ + switch (context->Base.ClientAPI) { + case EGL_OPENGL_ES_API: + case EGL_OPENVG_API: + /* fall-through */ + case EGL_OPENGL_API: + st_destroy_context(context->Context); + break; + default: + assert(0); } - return EGL_TRUE; - } - else { - _eglError(EGL_BAD_CONTEXT, "eglDestroyContext"); - return EGL_TRUE; + free(context); } + return EGL_TRUE; } @@ -411,16 +476,25 @@ xlib_eglDestroyContext(_EGLDriver *drv, EGLDisplay dpy, EGLContext ctx) * Called via eglMakeCurrent(), drv->API.MakeCurrent(). */ static EGLBoolean -xlib_eglMakeCurrent(_EGLDriver *drv, EGLDisplay dpy, - EGLSurface draw, EGLSurface read, EGLContext ctx) +xlib_eglMakeCurrent(_EGLDriver *drv, _EGLDisplay *dpy, + _EGLSurface *draw, _EGLSurface *read, _EGLContext *ctx) { struct xlib_egl_context *context = lookup_context(ctx); struct xlib_egl_surface *draw_surf = lookup_surface(draw); struct xlib_egl_surface *read_surf = lookup_surface(read); + struct st_context *oldcontext = NULL; + _EGLContext *oldctx; + + oldctx = _eglGetCurrentContext(); + if (oldctx && _eglIsContextLinked(oldctx)) + oldcontext = st_get_current(); - if (!_eglMakeCurrent(drv, dpy, draw, read, context)) + if (!_eglMakeCurrent(drv, dpy, draw, read, ctx)) return EGL_FALSE; + /* Flush before switching context. Check client API? */ + if (oldcontext) + st_flush(oldcontext, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME, NULL); st_make_current((context ? context->Context : NULL), (draw_surf ? draw_surf->Framebuffer : NULL), (read_surf ? read_surf->Framebuffer : NULL)); @@ -474,39 +548,34 @@ choose_stencil_format(const __GLcontextModes *visual) /** * Called via eglCreateWindowSurface(), drv->API.CreateWindowSurface(). */ -static EGLSurface -xlib_eglCreateWindowSurface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, +static _EGLSurface * +xlib_eglCreateWindowSurface(_EGLDriver *drv, _EGLDisplay *disp, _EGLConfig *conf, NativeWindowType window, const EGLint *attrib_list) { - struct xlib_egl_driver *xdrv = xlib_egl_driver(drv); - _EGLDisplay *disp = _eglLookupDisplay(dpy); - _EGLConfig *conf = _eglLookupConfig(drv, dpy, config); - + struct xlib_egl_display *xdpy = xlib_egl_display(disp); struct xlib_egl_surface *surf; __GLcontextModes visual; uint width, height; surf = CALLOC_STRUCT(xlib_egl_surface); if (!surf) - return EGL_NO_SURFACE; + return NULL; /* Let EGL lib init the common stuff */ - if (!_eglInitSurface(drv, dpy, &surf->Base, EGL_WINDOW_BIT, - config, attrib_list)) { + if (!_eglInitSurface(drv, &surf->Base, EGL_WINDOW_BIT, + conf, attrib_list)) { free(surf); - return EGL_NO_SURFACE; + return NULL; } - _eglSaveSurface(&surf->Base); - /* * Now init the Xlib and gallium stuff */ surf->Win = (Window) window; /* The X window ID */ - surf->Dpy = disp->Xdpy; /* The X display */ + surf->Dpy = xdpy->Dpy; /* The X display */ surf->Gc = XCreateGC(surf->Dpy, surf->Win, 0, NULL); - surf->winsys = xdrv->winsys; + surf->winsys = xdpy->winsys; _eglConfigToContextModesRec(conf, &visual); get_drawable_size(surf->Dpy, surf->Win, &width, &height); @@ -525,35 +594,172 @@ xlib_eglCreateWindowSurface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, st_resize_framebuffer(surf->Framebuffer, width, height); - return _eglGetSurfaceHandle(&surf->Base); + return &surf->Base; +} + + +static _EGLSurface * +xlib_eglCreatePbufferSurface(_EGLDriver *drv, _EGLDisplay *disp, _EGLConfig *conf, + const EGLint *attrib_list) +{ + struct xlib_egl_display *xdpy = xlib_egl_display(disp); + struct xlib_egl_surface *surf; + __GLcontextModes visual; + uint width, height; + EGLBoolean bind_texture; + + surf = CALLOC_STRUCT(xlib_egl_surface); + if (!surf) { + _eglError(EGL_BAD_ALLOC, "eglCreatePbufferSurface"); + return NULL; + } + + if (!_eglInitSurface(drv, &surf->Base, EGL_PBUFFER_BIT, + conf, attrib_list)) { + free(surf); + return NULL; + } + if (surf->Base.Width < 0 || surf->Base.Height < 0) { + _eglError(EGL_BAD_PARAMETER, "eglCreatePbufferSurface"); + free(surf); + return NULL; + } + + bind_texture = (surf->Base.TextureFormat != EGL_NO_TEXTURE); + width = (uint) surf->Base.Width; + height = (uint) surf->Base.Height; + if ((surf->Base.TextureTarget == EGL_NO_TEXTURE && bind_texture) || + (surf->Base.TextureTarget != EGL_NO_TEXTURE && !bind_texture)) { + _eglError(EGL_BAD_MATCH, "eglCreatePbufferSurface"); + free(surf); + return NULL; + } + /* a framebuffer of zero width or height confuses st */ + if (width == 0 || height == 0) { + _eglError(EGL_BAD_MATCH, "eglCreatePbufferSurface"); + free(surf); + return NULL; + } + /* no mipmap generation */ + if (surf->Base.MipmapTexture) { + _eglError(EGL_BAD_MATCH, "eglCreatePbufferSurface"); + free(surf); + return NULL; + } + + surf->winsys = xdpy->winsys; + + _eglConfigToContextModesRec(conf, &visual); + + /* Create GL statetracker framebuffer */ + surf->Framebuffer = st_create_framebuffer(&visual, + choose_color_format(&visual), + choose_depth_format(&visual), + choose_stencil_format(&visual), + width, height, + (void *) surf); + st_resize_framebuffer(surf->Framebuffer, width, height); + + return &surf->Base; } static EGLBoolean -xlib_eglDestroySurface(_EGLDriver *drv, EGLDisplay dpy, EGLSurface surface) +xlib_eglDestroySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface) { struct xlib_egl_surface *surf = lookup_surface(surface); - if (surf) { - _eglHashRemove(_eglGlobal.Surfaces, (EGLuint) surface); - if (surf->Base.IsBound) { - surf->Base.DeletePending = EGL_TRUE; - } - else { + if (!_eglIsSurfaceBound(&surf->Base)) { + if (surf->Base.Type != EGL_PBUFFER_BIT) XFreeGC(surf->Dpy, surf->Gc); - st_unreference_framebuffer(surf->Framebuffer); - free(surf); - } - return EGL_TRUE; + st_unreference_framebuffer(surf->Framebuffer); + free(surf); } - else { - _eglError(EGL_BAD_SURFACE, "eglDestroySurface"); - return EGL_FALSE; + return EGL_TRUE; +} + + +static EGLBoolean +xlib_eglBindTexImage(_EGLDriver *drv, _EGLDisplay *dpy, + _EGLSurface *surface, EGLint buffer) +{ + struct xlib_egl_surface *xsurf = lookup_surface(surface); + struct xlib_egl_context *xctx; + struct pipe_surface *psurf; + enum pipe_format format; + int target; + + if (!xsurf || xsurf->Base.Type != EGL_PBUFFER_BIT) + return _eglError(EGL_BAD_SURFACE, "eglBindTexImage"); + if (buffer != EGL_BACK_BUFFER) + return _eglError(EGL_BAD_PARAMETER, "eglBindTexImage"); + if (xsurf->Base.BoundToTexture) + return _eglError(EGL_BAD_ACCESS, "eglBindTexImage"); + + /* this should be updated when choose_color_format is */ + switch (xsurf->Base.TextureFormat) { + case EGL_TEXTURE_RGB: + format = PIPE_FORMAT_R8G8B8_UNORM; + break; + case EGL_TEXTURE_RGBA: + format = PIPE_FORMAT_A8R8G8B8_UNORM; + break; + default: + return _eglError(EGL_BAD_MATCH, "eglBindTexImage"); + } + + switch (xsurf->Base.TextureTarget) { + case EGL_TEXTURE_2D: + target = ST_TEXTURE_2D; + break; + default: + return _eglError(EGL_BAD_MATCH, "eglBindTexImage"); + } + + /* flush properly */ + if (eglGetCurrentSurface(EGL_DRAW) == surface) { + xctx = lookup_context(_eglGetCurrentContext()); + st_flush(xctx->Context, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME, + NULL); + } + else if (_eglIsSurfaceBound(&xsurf->Base)) { + xctx = lookup_context(xsurf->Base.Binding); + if (xctx) + st_finish(xctx->Context); } + + st_get_framebuffer_surface(xsurf->Framebuffer, ST_SURFACE_BACK_LEFT, + &psurf); + st_bind_texture_surface(psurf, target, xsurf->Base.MipmapLevel, format); + xsurf->Base.BoundToTexture = EGL_TRUE; + + return EGL_TRUE; +} + + +static EGLBoolean +xlib_eglReleaseTexImage(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface, + EGLint buffer) +{ + struct xlib_egl_surface *xsurf = lookup_surface(surface); + struct pipe_surface *psurf; + + if (!xsurf || xsurf->Base.Type != EGL_PBUFFER_BIT || + !xsurf->Base.BoundToTexture) + return _eglError(EGL_BAD_SURFACE, "eglReleaseTexImage"); + if (buffer != EGL_BACK_BUFFER) + return _eglError(EGL_BAD_PARAMETER, "eglReleaseTexImage"); + + st_get_framebuffer_surface(xsurf->Framebuffer, ST_SURFACE_BACK_LEFT, + &psurf); + st_unbind_texture_surface(psurf, ST_TEXTURE_2D, xsurf->Base.MipmapLevel); + xsurf->Base.BoundToTexture = EGL_FALSE; + + return EGL_TRUE; } static EGLBoolean -xlib_eglSwapBuffers(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw) +xlib_eglSwapBuffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw) { /* error checking step: */ if (!_eglSwapBuffers(drv, dpy, draw)) @@ -588,7 +794,9 @@ find_supported_apis(void) EGLint mask = 0; void *handle; - handle = dlopen(NULL, 0); + handle = dlopen(NULL, RTLD_LAZY | RTLD_LOCAL); + if(!handle) + return mask; if (dlsym(handle, "st_api_OpenGL_ES1")) mask |= EGL_OPENGL_ES_BIT; @@ -608,12 +816,20 @@ find_supported_apis(void) } +static void +xlib_Unload(_EGLDriver *drv) +{ + struct xlib_egl_driver *xdrv = xlib_egl_driver(drv); + free(xdrv); +} + + /** * This is the main entrypoint into the driver. * Called by libEGL to instantiate an _EGLDriver object. */ _EGLDriver * -_eglMain(_EGLDisplay *dpy, const char *args) +_eglMain(const char *args) { struct xlib_egl_driver *xdrv; @@ -623,10 +839,6 @@ _eglMain(_EGLDisplay *dpy, const char *args) if (!xdrv) return NULL; - if (!dpy->Xdpy) { - dpy->Xdpy = XOpenDisplay(NULL); - } - _eglInitDriverFallbacks(&xdrv->Base); xdrv->Base.API.Initialize = xlib_eglInitialize; xdrv->Base.API.Terminate = xlib_eglTerminate; @@ -634,27 +846,24 @@ _eglMain(_EGLDisplay *dpy, const char *args) xdrv->Base.API.CreateContext = xlib_eglCreateContext; xdrv->Base.API.DestroyContext = xlib_eglDestroyContext; xdrv->Base.API.CreateWindowSurface = xlib_eglCreateWindowSurface; + xdrv->Base.API.CreatePbufferSurface = xlib_eglCreatePbufferSurface; xdrv->Base.API.DestroySurface = xlib_eglDestroySurface; + xdrv->Base.API.BindTexImage = xlib_eglBindTexImage; + xdrv->Base.API.ReleaseTexImage = xlib_eglReleaseTexImage; xdrv->Base.API.MakeCurrent = xlib_eglMakeCurrent; xdrv->Base.API.SwapBuffers = xlib_eglSwapBuffers; - xdrv->Base.ClientAPIsMask = find_supported_apis(); - if (xdrv->Base.ClientAPIsMask == 0x0) { + xdrv->apis = find_supported_apis(); + if (xdrv->apis == 0x0) { /* the app isn't directly linked with any EGL-supprted APIs * (such as libGLESv2.so) so use an EGL utility to see what * APIs might be loaded dynamically on this system. */ - xdrv->Base.ClientAPIsMask = _eglFindAPIs(); - } + xdrv->apis = _eglFindAPIs(); + } xdrv->Base.Name = "Xlib/softpipe"; - - /* create one winsys and use it for all contexts/surfaces */ - xdrv->winsys = create_sw_winsys(); - xdrv->winsys->flush_frontbuffer = flush_frontbuffer; - - xdrv->screen = softpipe_create_screen(xdrv->winsys); + xdrv->Base.Unload = xlib_Unload; return &xdrv->Base; } - diff --git a/src/gallium/winsys/egl_xlib/sw_winsys.c b/src/gallium/winsys/egl_xlib/sw_winsys.c index aa1bfa8e883..79ff2cc985d 100644 --- a/src/gallium/winsys/egl_xlib/sw_winsys.c +++ b/src/gallium/winsys/egl_xlib/sw_winsys.c @@ -166,6 +166,7 @@ surface_buffer_create(struct pipe_winsys *winsys, unsigned width, unsigned height, enum pipe_format format, unsigned usage, + unsigned tex_usage, unsigned *stride) { const unsigned alignment = 64; diff --git a/src/gallium/winsys/gdi/SConscript b/src/gallium/winsys/gdi/SConscript index aabab95f3a8..86eb9ef55ed 100644 --- a/src/gallium/winsys/gdi/SConscript +++ b/src/gallium/winsys/gdi/SConscript @@ -15,6 +15,7 @@ if env['platform'] == 'windows': 'gdi32', 'user32', 'kernel32', + 'ws2_32', ]) sources = [ diff --git a/src/gallium/winsys/xlib/Makefile b/src/gallium/winsys/xlib/Makefile index 04309e67eea..3a1945d92c5 100644 --- a/src/gallium/winsys/xlib/Makefile +++ b/src/gallium/winsys/xlib/Makefile @@ -34,6 +34,7 @@ XLIB_WINSYS_SOURCES = \ xlib_brw_aub.c \ xlib_brw_context.c \ xlib_brw_screen.c \ + xlib_llvmpipe.c \ xlib_softpipe.c \ xlib_trace.c @@ -90,7 +91,7 @@ install: default $(INSTALL) -d $(INSTALL_DIR)/$(LIB_DIR) $(INSTALL) -m 644 $(TOP)/include/GL/*.h $(INSTALL_DIR)/include/GL @if [ -e $(TOP)/$(LIB_DIR)/$(GL_LIB_NAME) ]; then \ - $(INSTALL) $(TOP)/$(LIB_DIR)/libGL* $(INSTALL_DIR)/$(LIB_DIR); \ + $(MINSTALL) $(TOP)/$(LIB_DIR)/libGL* $(INSTALL_DIR)/$(LIB_DIR); \ fi diff --git a/src/gallium/winsys/xlib/SConscript b/src/gallium/winsys/xlib/SConscript index 0fb4b50f634..467d595d33b 100644 --- a/src/gallium/winsys/xlib/SConscript +++ b/src/gallium/winsys/xlib/SConscript @@ -5,7 +5,7 @@ Import('*') if env['platform'] == 'linux' \ and 'mesa' in env['statetrackers'] \ - and ('softpipe' or 'i915simple' or 'trace') in env['drivers'] \ + and set(('softpipe', 'llvmpipe', 'i915simple', 'trace')).intersection(env['drivers']) \ and not env['dri']: env = env.Clone() @@ -29,6 +29,14 @@ if env['platform'] == 'linux' \ sources += ['xlib_softpipe.c'] drivers += [softpipe] + if 'llvmpipe' in env['drivers']: + env.Tool('llvm') + if 'LLVM_VERSION' in env: + env.Append(CPPDEFINES = 'GALLIUM_LLVMPIPE') + env.Tool('udis86') + sources += ['xlib_llvmpipe.c'] + drivers += [llvmpipe] + if 'i965simple' in env['drivers']: env.Append(CPPDEFINES = 'GALLIUM_I965SIMPLE') sources += [ diff --git a/src/gallium/winsys/xlib/xlib.c b/src/gallium/winsys/xlib/xlib.c index da722282156..4b71cf7ec38 100644 --- a/src/gallium/winsys/xlib/xlib.c +++ b/src/gallium/winsys/xlib/xlib.c @@ -45,6 +45,7 @@ enum mode { MODE_TRACE, MODE_BRW, MODE_CELL, + MODE_LLVMPIPE, MODE_SOFTPIPE }; @@ -62,7 +63,11 @@ static enum mode get_mode() return MODE_CELL; #endif +#if defined(GALLIUM_LLVMPIPE) + return MODE_LLVMPIPE; +#else return MODE_SOFTPIPE; +#endif } static void _init( void ) __attribute__((constructor)); @@ -87,6 +92,11 @@ static void _init( void ) xmesa_set_driver( &xlib_cell_driver ); #endif break; + case MODE_LLVMPIPE: +#if defined(GALLIUM_LLVMPIPE) + xmesa_set_driver( &xlib_llvmpipe_driver ); +#endif + break; case MODE_SOFTPIPE: #if defined(GALLIUM_SOFTPIPE) xmesa_set_driver( &xlib_softpipe_driver ); diff --git a/src/gallium/winsys/xlib/xlib.h b/src/gallium/winsys/xlib/xlib.h index d602ab0b133..347d45f4d66 100644 --- a/src/gallium/winsys/xlib/xlib.h +++ b/src/gallium/winsys/xlib/xlib.h @@ -7,6 +7,7 @@ extern struct xm_driver xlib_trace_driver; extern struct xm_driver xlib_softpipe_driver; +extern struct xm_driver xlib_llvmpipe_driver; extern struct xm_driver xlib_cell_driver; extern struct xm_driver xlib_brw_driver; diff --git a/src/gallium/winsys/xlib/xlib_brw_screen.c b/src/gallium/winsys/xlib/xlib_brw_screen.c index a78efb10fa8..ef545796f3c 100644 --- a/src/gallium/winsys/xlib/xlib_brw_screen.c +++ b/src/gallium/winsys/xlib/xlib_brw_screen.c @@ -249,6 +249,7 @@ aub_i915_surface_buffer_create(struct pipe_winsys *winsys, unsigned width, unsigned height, enum pipe_format format, unsigned usage, + unsigned tex_usage, unsigned *stride) { const unsigned alignment = 64; diff --git a/src/gallium/winsys/xlib/xlib_llvmpipe.c b/src/gallium/winsys/xlib/xlib_llvmpipe.c new file mode 100644 index 00000000000..3dd15e099b0 --- /dev/null +++ b/src/gallium/winsys/xlib/xlib_llvmpipe.c @@ -0,0 +1,464 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ + +/* + * Authors: + * Keith Whitwell + * Brian Paul + */ + + +#if defined(GALLIUM_LLVMPIPE) + +#include "xm_api.h" + +#undef ASSERT +#undef Elements + +#include "pipe/internal/p_winsys_screen.h" +#include "pipe/p_format.h" +#include "pipe/p_context.h" +#include "pipe/p_inlines.h" +#include "util/u_math.h" +#include "util/u_memory.h" +#include "llvmpipe/lp_winsys.h" +#include "llvmpipe/lp_texture.h" + +#include "xlib.h" + +/** + * Subclass of pipe_buffer for Xlib winsys. + * Low-level OS/window system memory buffer + */ +struct xm_displaytarget +{ + enum pipe_format format; + struct pipe_format_block block; + unsigned width; + unsigned height; + unsigned stride; + + void *data; + void *mapped; + + XImage *tempImage; +#ifdef USE_XSHM + int shm; + XShmSegmentInfo shminfo; +#endif +}; + + +/** + * Subclass of llvmpipe_winsys for Xlib winsys + */ +struct xmesa_llvmpipe_winsys +{ + struct llvmpipe_winsys base; +/* struct xmesa_visual *xm_visual; */ +}; + + + +/** Cast wrapper */ +static INLINE struct xm_displaytarget * +xm_displaytarget( struct llvmpipe_displaytarget *dt ) +{ + return (struct xm_displaytarget *)dt; +} + + +/** + * X Shared Memory Image extension code + */ + +#ifdef USE_XSHM + +static volatile int mesaXErrorFlag = 0; + +/** + * Catches potential Xlib errors. + */ +static int +mesaHandleXError(Display *dpy, XErrorEvent *event) +{ + (void) dpy; + (void) event; + mesaXErrorFlag = 1; + return 0; +} + + +static char *alloc_shm(struct xm_displaytarget *buf, unsigned size) +{ + XShmSegmentInfo *const shminfo = & buf->shminfo; + + shminfo->shmid = shmget(IPC_PRIVATE, size, IPC_CREAT|0777); + if (shminfo->shmid < 0) { + return NULL; + } + + shminfo->shmaddr = (char *) shmat(shminfo->shmid, 0, 0); + if (shminfo->shmaddr == (char *) -1) { + shmctl(shminfo->shmid, IPC_RMID, 0); + return NULL; + } + + shminfo->readOnly = False; + return shminfo->shmaddr; +} + + +/** + * Allocate a shared memory XImage back buffer for the given XMesaBuffer. + */ +static void +alloc_shm_ximage(struct xm_displaytarget *xm_buffer, + struct xmesa_buffer *xmb, + unsigned width, unsigned height) +{ + /* + * We have to do a _lot_ of error checking here to be sure we can + * really use the XSHM extension. It seems different servers trigger + * errors at different points if the extension won't work. Therefore + * we have to be very careful... + */ + int (*old_handler)(Display *, XErrorEvent *); + + xm_buffer->tempImage = XShmCreateImage(xmb->xm_visual->display, + xmb->xm_visual->visinfo->visual, + xmb->xm_visual->visinfo->depth, + ZPixmap, + NULL, + &xm_buffer->shminfo, + width, height); + if (xm_buffer->tempImage == NULL) { + xm_buffer->shm = 0; + return; + } + + + mesaXErrorFlag = 0; + old_handler = XSetErrorHandler(mesaHandleXError); + /* This may trigger the X protocol error we're ready to catch: */ + XShmAttach(xmb->xm_visual->display, &xm_buffer->shminfo); + XSync(xmb->xm_visual->display, False); + + if (mesaXErrorFlag) { + /* we are on a remote display, this error is normal, don't print it */ + XFlush(xmb->xm_visual->display); + mesaXErrorFlag = 0; + XDestroyImage(xm_buffer->tempImage); + xm_buffer->tempImage = NULL; + xm_buffer->shm = 0; + (void) XSetErrorHandler(old_handler); + return; + } + + xm_buffer->shm = 1; +} + +#endif /* USE_XSHM */ + +static boolean +xm_is_displaytarget_format_supported( struct llvmpipe_winsys *ws, + enum pipe_format format ) +{ + /* TODO: check visuals or other sensible thing here */ + return TRUE; +} + + +static void * +xm_displaytarget_map(struct llvmpipe_winsys *ws, + struct llvmpipe_displaytarget *dt, + unsigned flags) +{ + struct xm_displaytarget *xm_dt = xm_displaytarget(dt); + xm_dt->mapped = xm_dt->data; + return xm_dt->mapped; +} + +static void +xm_displaytarget_unmap(struct llvmpipe_winsys *ws, + struct llvmpipe_displaytarget *dt) +{ + struct xm_displaytarget *xm_dt = xm_displaytarget(dt); + xm_dt->mapped = NULL; +} + +static void +xm_displaytarget_destroy(struct llvmpipe_winsys *ws, + struct llvmpipe_displaytarget *dt) +{ + struct xm_displaytarget *xm_dt = xm_displaytarget(dt); + + if (xm_dt->data) { +#ifdef USE_XSHM + if (xm_dt->shminfo.shmid >= 0) { + shmdt(xm_dt->shminfo.shmaddr); + shmctl(xm_dt->shminfo.shmid, IPC_RMID, 0); + + xm_dt->shminfo.shmid = -1; + xm_dt->shminfo.shmaddr = (char *) -1; + } + else +#endif + FREE(xm_dt->data); + } + + FREE(xm_dt); +} + + +/** + * Display/copy the image in the surface into the X window specified + * by the XMesaBuffer. + */ +static void +xm_llvmpipe_display(struct xmesa_buffer *xm_buffer, + struct llvmpipe_displaytarget *dt) +{ + XImage *ximage; + struct xm_displaytarget *xm_dt = xm_displaytarget(dt); + static boolean no_swap = 0; + static boolean firsttime = 1; + + if (firsttime) { + no_swap = getenv("SP_NO_RAST") != NULL; + firsttime = 0; + } + + if (no_swap) + return; + +#ifdef USE_XSHM + if (xm_dt->shm) + { + if (xm_dt->tempImage == NULL) + { + assert(xm_dt->block.width == 1); + assert(xm_dt->block.height == 1); + alloc_shm_ximage(xm_dt, xm_buffer, + xm_dt->stride / xm_dt->block.size, + xm_dt->height); + } + + ximage = xm_dt->tempImage; + ximage->data = xm_dt->data; + + /* _debug_printf("XSHM\n"); */ + XShmPutImage(xm_buffer->xm_visual->display, xm_buffer->drawable, xm_buffer->gc, + ximage, 0, 0, 0, 0, xm_dt->width, xm_dt->height, False); + } + else +#endif + { + /* display image in Window */ + ximage = xm_dt->tempImage; + ximage->data = xm_dt->data; + + /* check that the XImage has been previously initialized */ + assert(ximage->format); + assert(ximage->bitmap_unit); + + /* update XImage's fields */ + ximage->width = xm_dt->width; + ximage->height = xm_dt->height; + ximage->bytes_per_line = xm_dt->stride; + + /* _debug_printf("XPUT\n"); */ + XPutImage(xm_buffer->xm_visual->display, xm_buffer->drawable, xm_buffer->gc, + ximage, 0, 0, 0, 0, xm_dt->width, xm_dt->height); + } +} + +/** + * Display/copy the image in the surface into the X window specified + * by the XMesaBuffer. + */ +static void +xm_displaytarget_display(struct llvmpipe_winsys *ws, + struct llvmpipe_displaytarget *dt, + void *context_private) +{ + XMesaContext xmctx = (XMesaContext) context_private; + struct xmesa_buffer *xm_buffer = xmctx->xm_buffer; + xm_llvmpipe_display(xm_buffer, dt); +} + + +static struct llvmpipe_displaytarget * +xm_displaytarget_create(struct llvmpipe_winsys *winsys, + enum pipe_format format, + unsigned width, unsigned height, + unsigned alignment, + unsigned *stride) +{ + struct xm_displaytarget *xm_dt = CALLOC_STRUCT(xm_displaytarget); + unsigned nblocksx, nblocksy, size; + + xm_dt = CALLOC_STRUCT(xm_displaytarget); + if(!xm_dt) + goto no_xm_dt; + + xm_dt->format = format; + xm_dt->width = width; + xm_dt->height = height; + + pf_get_block(format, &xm_dt->block); + nblocksx = pf_get_nblocksx(&xm_dt->block, width); + nblocksy = pf_get_nblocksy(&xm_dt->block, height); + xm_dt->stride = align(nblocksx * xm_dt->block.size, alignment); + size = xm_dt->stride * nblocksy; + +#ifdef USE_XSHM + if (!debug_get_bool_option("XLIB_NO_SHM", FALSE)) + { + xm_dt->shminfo.shmid = -1; + xm_dt->shminfo.shmaddr = (char *) -1; + xm_dt->shm = TRUE; + + xm_dt->data = alloc_shm(xm_dt, size); + if(!xm_dt->data) + goto no_data; + } +#endif + + if(!xm_dt->data) { + xm_dt->data = align_malloc(size, alignment); + if(!xm_dt->data) + goto no_data; + } + + *stride = xm_dt->stride; + return (struct llvmpipe_displaytarget *)xm_dt; + +no_data: + FREE(xm_dt); +no_xm_dt: + return NULL; +} + + +static void +xm_destroy( struct llvmpipe_winsys *ws ) +{ + FREE(ws); +} + + +static struct llvmpipe_winsys * +xlib_create_llvmpipe_winsys( void ) +{ + struct xmesa_llvmpipe_winsys *ws; + + ws = CALLOC_STRUCT(xmesa_llvmpipe_winsys); + if (!ws) + return NULL; + + ws->base.destroy = xm_destroy; + + ws->base.is_displaytarget_format_supported = xm_is_displaytarget_format_supported; + + ws->base.displaytarget_create = xm_displaytarget_create; + ws->base.displaytarget_map = xm_displaytarget_map; + ws->base.displaytarget_unmap = xm_displaytarget_unmap; + ws->base.displaytarget_destroy = xm_displaytarget_destroy; + + ws->base.displaytarget_display = xm_displaytarget_display; + + return &ws->base; +} + + +static struct pipe_screen * +xlib_create_llvmpipe_screen( void ) +{ + struct llvmpipe_winsys *winsys; + struct pipe_screen *screen; + + winsys = xlib_create_llvmpipe_winsys(); + if (winsys == NULL) + return NULL; + + screen = llvmpipe_create_screen(winsys); + if (screen == NULL) + goto fail; + + return screen; + +fail: + if (winsys) + winsys->destroy( winsys ); + + return NULL; +} + + +static struct pipe_context * +xlib_create_llvmpipe_context( struct pipe_screen *screen, + void *context_private ) +{ + struct pipe_context *pipe; + + pipe = llvmpipe_create(screen); + if (pipe == NULL) + goto fail; + + pipe->priv = context_private; + return pipe; + +fail: + /* Free stuff here */ + return NULL; +} + + +static void +xlib_llvmpipe_display_surface(struct xmesa_buffer *xm_buffer, + struct pipe_surface *surf) +{ + struct llvmpipe_texture *texture = llvmpipe_texture(surf->texture); + + assert(texture->dt); + if (texture->dt) + xm_llvmpipe_display(xm_buffer, texture->dt); +} + + +struct xm_driver xlib_llvmpipe_driver = +{ + .create_pipe_screen = xlib_create_llvmpipe_screen, + .create_pipe_context = xlib_create_llvmpipe_context, + .display_surface = xlib_llvmpipe_display_surface +}; + + + +#endif /* GALLIUM_LLVMPIPE */ diff --git a/src/gallium/winsys/xlib/xlib_softpipe.c b/src/gallium/winsys/xlib/xlib_softpipe.c index 2a08b829085..67fea023a3b 100644 --- a/src/gallium/winsys/xlib/xlib_softpipe.c +++ b/src/gallium/winsys/xlib/xlib_softpipe.c @@ -376,6 +376,7 @@ xm_surface_buffer_create(struct pipe_winsys *winsys, unsigned width, unsigned height, enum pipe_format format, unsigned usage, + unsigned tex_usage, unsigned *stride) { const unsigned alignment = 64; |