svga: better path for generating white pixels for xor logicop workaround

6 files changed, 75 insertions, 106 deletions

diff --git a/src/gallium/drivers/svga/svga_context.c b/src/gallium/drivers/svga/svga_context.c
index af99c9de37c..1e76592ec60 100644
--- a/src/gallium/drivers/svga/svga_context.c
+++ b/src/gallium/drivers/svga/svga_context.c
@@ -215,7 +215,6 @@ struct pipe_context *svga_context_create( struct pipe_screen *screen )
    svga->state.hw_draw.num_views = 0;
 
    svga->dirty = ~0;
-   svga->state.white_fs_id = SVGA3D_INVALID_ID;
 
    LIST_INITHEAD(&svga->dirty_buffers);
 
diff --git a/src/gallium/drivers/svga/svga_context.h b/src/gallium/drivers/svga/svga_context.h
index 32e9304f819..52d47046e74 100644
--- a/src/gallium/drivers/svga/svga_context.h
+++ b/src/gallium/drivers/svga/svga_context.h
@@ -329,10 +329,6 @@ struct svga_context
 
       unsigned texture_timestamp;
 
-      /* Internally generated shaders:
-       */
-      unsigned white_fs_id;
-
       /* 
        */
       struct svga_sw_state          sw;
diff --git a/src/gallium/drivers/svga/svga_state_fs.c b/src/gallium/drivers/svga/svga_state_fs.c
index 4fe91416778..3b700cd0227 100644
--- a/src/gallium/drivers/svga/svga_state_fs.c
+++ b/src/gallium/drivers/svga/svga_state_fs.c
@@ -103,70 +103,6 @@ fail:
    return ret;
 }
 
-/* The blend workaround for simulating logicop xor behaviour requires
- * that the incoming fragment color be white.  This change achieves
- * that by hooking up a hard-wired fragment shader that just emits
- * color 1,1,1,1
- *   
- * This is a slightly incomplete solution as it assumes that the
- * actual bound shader has no other effects beyond generating a
- * fragment color.  In particular shaders containing TEXKIL and/or
- * depth-write will not have the correct behaviour, nor will those
- * expecting to use alphatest.
- *   
- * These are avoidable issues, but they are not much worse than the
- * unavoidable ones associated with this technique, so it's not clear
- * how much effort should be expended trying to resolve them - the
- * ultimate result will still not be correct in most cases.
- *
- * Shader below was generated with:
- *   SVGA_DEBUG=tgsi ./mesa/progs/fp/fp-tri white.txt
- */
-static int emit_white_fs( struct svga_context *svga )
-{
-   int ret = PIPE_ERROR;
-
-   /* ps_3_0
-    * def c0, 1.000000, 0.000000, 0.000000, 1.000000
-    * mov oC0, c0.x
-    * end
-    */
-   static const unsigned white_tokens[] = {
-      0xffff0300,
-      0x05000051,
-      0xa00f0000,
-      0x3f800000,
-      0x00000000,
-      0x00000000,
-      0x3f800000,
-      0x02000001,
-      0x800f0800,
-      0xa0000000,
-      0x0000ffff,
-   };
-
-   assert(SVGA3D_INVALID_ID == UTIL_BITMASK_INVALID_INDEX);
-   svga->state.white_fs_id = util_bitmask_add(svga->fs_bm);
-   if(svga->state.white_fs_id == SVGA3D_INVALID_ID)
-      goto no_fs_id;
-
-   ret = SVGA3D_DefineShader(svga->swc, 
-                             svga->state.white_fs_id,
-                             SVGA3D_SHADERTYPE_PS,
-                             white_tokens, 
-                             sizeof(white_tokens));
-   if (ret)
-      goto no_definition;
-
-   return 0;
-
-no_definition:
-   util_bitmask_clear(svga->fs_bm, svga->state.white_fs_id);
-   svga->state.white_fs_id = SVGA3D_INVALID_ID;
-no_fs_id:
-   return ret;
-}
-
 
 /* SVGA_NEW_TEXTURE_BINDING
  * SVGA_NEW_RAST
@@ -194,6 +130,23 @@ static int make_fs_key( const struct svga_context *svga,
                        PIPE_WINDING_CW);
    }
 
+   /* The blend workaround for simulating logicop xor behaviour
+    * requires that the incoming fragment color be white.  This change
+    * achieves that by creating a varient of the current fragment
+    * shader that overrides all output colors with 1,1,1,1
+    *   
+    * This will work for most shaders, including those containing
+    * TEXKIL and/or depth-write.  However, it will break on the
+    * combination of xor-logicop plus alphatest.
+    *
+    * Ultimately, we could implement alphatest in the shader using
+    * texkil prior to overriding the outgoing fragment color.
+    *   
+    * SVGA_NEW_BLEND
+    */
+   if (svga->curr.blend->need_white_fragments) {
+      key->white_fragments = 1;
+   }
    
    /* XXX: want to limit this to the textures that the shader actually
     * refers to.
@@ -233,40 +186,29 @@ static int emit_hw_fs( struct svga_context *svga,
    unsigned id = SVGA3D_INVALID_ID;
    int ret = 0;
 
+   struct svga_fragment_shader *fs = svga->curr.fs;
+   struct svga_fs_compile_key key;
+
    /* SVGA_NEW_BLEND
+    * SVGA_NEW_TEXTURE_BINDING
+    * SVGA_NEW_RAST
+    * SVGA_NEW_NEED_SWTNL
+    * SVGA_NEW_SAMPLER
     */
-   if (svga->curr.blend->need_white_fragments) {
-      if (svga->state.white_fs_id == SVGA3D_INVALID_ID) {
-         ret = emit_white_fs( svga );
-         if (ret)
-            return ret;
-      }
-      id = svga->state.white_fs_id;
-   }
-   else {
-      struct svga_fragment_shader *fs = svga->curr.fs;
-      struct svga_fs_compile_key key;
-
-      /* SVGA_NEW_TEXTURE_BINDING
-       * SVGA_NEW_RAST
-       * SVGA_NEW_NEED_SWTNL
-       * SVGA_NEW_SAMPLER
-       */
-      ret = make_fs_key( svga, &key );
+   ret = make_fs_key( svga, &key );
+   if (ret)
+      return ret;
+
+   result = search_fs_key( fs, &key );
+   if (!result) {
+      ret = compile_fs( svga, fs, &key, &result );
       if (ret)
          return ret;
-
-      result = search_fs_key( fs, &key );
-      if (!result) {
-         ret = compile_fs( svga, fs, &key, &result );
-         if (ret)
-            return ret;
-      }
-
-      assert (result);
-      id = result->id;
    }
 
+   assert (result);
+   id = result->id;
+
    assert(id != SVGA3D_INVALID_ID);
 
    if (result != svga->state.hw_draw.fs) {
diff --git a/src/gallium/drivers/svga/svga_tgsi.h b/src/gallium/drivers/svga/svga_tgsi.h
index 896c90a89ae..25f60c78b7c 100644
--- a/src/gallium/drivers/svga/svga_tgsi.h
+++ b/src/gallium/drivers/svga/svga_tgsi.h
@@ -49,6 +49,7 @@ struct svga_fs_compile_key
 {
    boolean light_twoside:1;
    boolean front_cw:1;
+   boolean white_fragments:1;
    ubyte num_textures;
    ubyte num_unnormalized_coords;
    struct {
diff --git a/src/gallium/drivers/svga/svga_tgsi_decl_sm30.c b/src/gallium/drivers/svga/svga_tgsi_decl_sm30.c
index 2291cf116dd..4c078ba9117 100644
--- a/src/gallium/drivers/svga/svga_tgsi_decl_sm30.c
+++ b/src/gallium/drivers/svga/svga_tgsi_decl_sm30.c
@@ -194,8 +194,19 @@ static boolean ps30_output( struct svga_shader_emitter *emit,
 
    switch (semantic.SemanticName) {
    case TGSI_SEMANTIC_COLOR:
-      emit->output_map[idx] = dst_register( SVGA3DREG_COLOROUT, 
-                                            semantic.SemanticIndex );
+      if (emit->unit == PIPE_SHADER_FRAGMENT &&
+          emit->key.fkey.white_fragments) {
+
+         emit->output_map[idx] = dst_register( SVGA3DREG_TEMP,
+                                               emit->nr_hw_temp++ );
+         emit->temp_col[idx] = emit->output_map[idx];
+         emit->true_col[idx] = dst_register( SVGA3DREG_COLOROUT, 
+                                              semantic.SemanticIndex );
+      }
+      else {
+         emit->output_map[idx] = dst_register( SVGA3DREG_COLOROUT, 
+                                               semantic.SemanticIndex );
+      }
       break;
    case TGSI_SEMANTIC_POSITION:
       emit->output_map[idx] = dst_register( SVGA3DREG_TEMP,
diff --git a/src/gallium/drivers/svga/svga_tgsi_insn.c b/src/gallium/drivers/svga/svga_tgsi_insn.c
index ea409b7e165..b5424dacf7d 100644
--- a/src/gallium/drivers/svga/svga_tgsi_insn.c
+++ b/src/gallium/drivers/svga/svga_tgsi_insn.c
@@ -1011,10 +1011,10 @@ static boolean emit_kilp(struct svga_shader_emitter *emit,
 {
    SVGA3dShaderInstToken inst;
    SVGA3dShaderDestToken temp;
-   struct src_register one = get_zero_immediate( emit );
+   struct src_register one = scalar( get_zero_immediate( emit ),
+                                     TGSI_SWIZZLE_W );
 
    inst = inst_token( SVGA3DOP_TEXKILL );
-   one = scalar( one, TGSI_SWIZZLE_W );
 
    /* texkill doesn't allow negation on the operand so lets move
     * negation of {1} to a temp register */
@@ -2254,11 +2254,28 @@ static boolean emit_ps_postamble( struct svga_shader_emitter *emit )
    for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
       if (SVGA3dShaderGetRegType(emit->true_col[i].value) != 0) {
 
-         if (!submit_op1( emit,
-                          inst_token(SVGA3DOP_MOV),
-                          emit->true_col[i],
-                          src(emit->temp_col[i]) ))
-            return FALSE;
+         /* Potentially override output colors with white for XOR
+          * logicop workaround.
+          */
+         if (emit->unit == PIPE_SHADER_FRAGMENT &&
+             emit->key.fkey.white_fragments) {
+
+            struct src_register one = scalar( get_zero_immediate( emit ),
+                                              TGSI_SWIZZLE_W );
+
+            if (!submit_op1( emit,
+                             inst_token(SVGA3DOP_MOV),
+                             emit->true_col[i],
+                             one ))
+               return FALSE;
+         }
+         else {
+            if (!submit_op1( emit,
+                             inst_token(SVGA3DOP_MOV),
+                             emit->true_col[i],
+                             src(emit->temp_col[i]) ))
+               return FALSE;
+         }
       }
    }
 
@@ -2467,6 +2484,9 @@ needs_to_create_zero( struct svga_shader_emitter *emit )
       if (emit->key.fkey.light_twoside)
          return TRUE;
 
+      if (emit->key.fkey.white_fragments)
+         return TRUE;
+
       if (emit->emit_frontface)
          return TRUE;