st/mesa: checkpoint: implement fragment program variants

The fragment programs variants are per-context, as the vertex programs.

NOTE: glDrawPixels is totally broken at this point.

8 files changed, 394 insertions, 217 deletions

diff --git a/src/mesa/state_tracker/st_atom_shader.c b/src/mesa/state_tracker/st_atom_shader.c
index f0981629b5..94890e9fea 100644
--- a/src/mesa/state_tracker/st_atom_shader.c
+++ b/src/mesa/state_tracker/st_atom_shader.c
@@ -51,20 +51,6 @@
 
 
 
-/**
- * Translate fragment program if needed.
- */
-static void
-translate_fp(struct st_context *st,
-             struct st_fragment_program *stfp)
-{
-   if (!stfp->tgsi.tokens) {
-      assert(stfp->Base.Base.NumInstructions > 0);
-
-      st_translate_fragment_program(st, stfp);
-   }
-}
-
 /*
  * Translate geometry program if needed.
  */
@@ -168,12 +154,17 @@ static void
 update_fp( struct st_context *st )
 {
    struct st_fragment_program *stfp;
+   struct st_fp_varient_key key;
 
    assert(st->ctx->FragmentProgram._Current);
    stfp = st_fragment_program(st->ctx->FragmentProgram._Current);
    assert(stfp->Base.Base.Target == GL_FRAGMENT_PROGRAM_ARB);
 
-   translate_fp(st, stfp);
+
+   memset(&key, 0, sizeof(key));
+   key.st = st;
+
+   st->fp_varient = st_get_fp_varient(st, stfp, &key);
 
    st_reference_fragprog(st, &st->fp, stfp);
 
@@ -183,7 +174,8 @@ update_fp( struct st_context *st )
       cso_set_fragment_shader_handle(st->cso_context, fs);
    }
    else {
-      cso_set_fragment_shader_handle(st->cso_context, stfp->driver_shader);
+      cso_set_fragment_shader_handle(st->cso_context,
+                                     st->fp_varient->driver_shader);
    }
 }
 
diff --git a/src/mesa/state_tracker/st_cb_bitmap.c b/src/mesa/state_tracker/st_cb_bitmap.c
index f08697fe23..a2ab743358 100644
--- a/src/mesa/state_tracker/st_cb_bitmap.c
+++ b/src/mesa/state_tracker/st_cb_bitmap.c
@@ -185,48 +185,47 @@ find_free_bit(uint bitfield)
 
 /**
  * Combine basic bitmap fragment program with the user-defined program.
+ * \param st  current context
+ * \param fpIn  the incoming fragment program
+ * \param fpOut  the new fragment program which does fragment culling
+ * \param bitmap_sampler  sampler number for the bitmap texture
  */
-static struct st_fragment_program *
-combined_bitmap_fragment_program(struct gl_context *ctx)
+void
+st_make_combined_bitmap_fragment_program(struct st_context *st,
+                                         struct gl_fragment_program *fpIn,
+                                         struct gl_fragment_program **fpOut,
+                                         GLuint *bitmap_sampler)
 {
-   struct st_context *st = st_context(ctx);
-   struct st_fragment_program *stfp = st->fp;
-
-   if (!stfp->bitmap_program) {
-      /*
-       * Generate new program which is the user-defined program prefixed
-       * with the bitmap sampler/kill instructions.
-       */
-      struct st_fragment_program *bitmap_prog;
-      uint sampler;
+   struct st_fragment_program *bitmap_prog;
+   struct gl_program *newProg;
+   uint sampler;
 
-      sampler = find_free_bit(st->fp->Base.Base.SamplersUsed);
-      bitmap_prog = make_bitmap_fragment_program(ctx, sampler);
-
-      stfp->bitmap_program = (struct st_fragment_program *)
-         _mesa_combine_programs(ctx,
-                                &bitmap_prog->Base.Base, &stfp->Base.Base);
-      stfp->bitmap_program->bitmap_sampler = sampler;
+   /*
+    * Generate new program which is the user-defined program prefixed
+    * with the bitmap sampler/kill instructions.
+    */
+   sampler = find_free_bit(fpIn->Base.SamplersUsed);
+   bitmap_prog = make_bitmap_fragment_program(st->ctx, sampler);
 
-      /* done with this after combining */
-      st_reference_fragprog(st, &bitmap_prog, NULL);
+   newProg = _mesa_combine_programs(st->ctx,
+                                    &bitmap_prog->Base.Base,
+                                    &fpIn->Base);
+   /* done with this after combining */
+   st_reference_fragprog(st, &bitmap_prog, NULL);
 
 #if 0
-      {
-         struct gl_program *p = &stfp->bitmap_program->Base.Base;
-         printf("Combined bitmap program:\n");
-         _mesa_print_program(p);
-         printf("InputsRead: 0x%x\n", p->InputsRead);
-         printf("OutputsWritten: 0x%x\n", p->OutputsWritten);
-         _mesa_print_parameter_list(p->Parameters);
-      }
-#endif
-
-      /* translate to TGSI tokens */
-      st_translate_fragment_program(st, stfp->bitmap_program);
+   {
+      printf("Combined bitmap program:\n");
+      _mesa_print_program(newProg);
+      printf("InputsRead: 0x%x\n", newProg->InputsRead);
+      printf("OutputsWritten: 0x%x\n", newProg->OutputsWritten);
+      _mesa_print_parameter_list(newProg->Parameters);
    }
+#endif
 
-   return stfp->bitmap_program;
+   /* return results */
+   *fpOut = (struct gl_fragment_program *) newProg;
+   *bitmap_sampler = sampler;
 }
 
 
@@ -411,11 +410,16 @@ draw_bitmap_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
    struct st_context *st = st_context(ctx);
    struct pipe_context *pipe = st->pipe;
    struct cso_context *cso = st->cso_context;
-   struct st_fragment_program *stfp;
+   struct st_fp_varient *fpv;
+   struct st_fp_varient_key key;
    GLuint maxSize;
    GLuint offset;
 
-   stfp = combined_bitmap_fragment_program(ctx);
+   memset(&key, 0, sizeof(key));
+   key.st = st;
+   key.bitmap = GL_TRUE;
+
+   fpv = st_get_fp_varient(st, st->fp, &key);
 
    /* As an optimization, Mesa's fragment programs will sometimes get the
     * primary color from a statevar/constant rather than a varying variable.
@@ -428,7 +432,7 @@ draw_bitmap_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
       GLfloat colorSave[4];
       COPY_4V(colorSave, ctx->Current.Attrib[VERT_ATTRIB_COLOR0]);
       COPY_4V(ctx->Current.Attrib[VERT_ATTRIB_COLOR0], color);
-      st_upload_constants(st, stfp->Base.Base.Parameters, PIPE_SHADER_FRAGMENT);
+      st_upload_constants(st, fpv->parameters, PIPE_SHADER_FRAGMENT);
       COPY_4V(ctx->Current.Attrib[VERT_ATTRIB_COLOR0], colorSave);
    }
 
@@ -454,7 +458,7 @@ draw_bitmap_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
    cso_set_rasterizer(cso, &st->bitmap.rasterizer);
 
    /* fragment shader state: TEX lookup program */
-   cso_set_fragment_shader_handle(cso, stfp->driver_shader);
+   cso_set_fragment_shader_handle(cso, fpv->driver_shader);
 
    /* vertex shader state: position + texcoord pass-through */
    cso_set_vertex_shader_handle(cso, st->bitmap.vs);
@@ -462,21 +466,21 @@ draw_bitmap_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
    /* user samplers, plus our bitmap sampler */
    {
       struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS];
-      uint num = MAX2(stfp->bitmap_sampler + 1, st->state.num_samplers);
+      uint num = MAX2(fpv->bitmap_sampler + 1, st->state.num_samplers);
       uint i;
       for (i = 0; i < st->state.num_samplers; i++) {
          samplers[i] = &st->state.samplers[i];
       }
-      samplers[stfp->bitmap_sampler] = &st->bitmap.samplers[sv->texture->target != PIPE_TEXTURE_RECT];
+      samplers[fpv->bitmap_sampler] = &st->bitmap.samplers[sv->texture->target != PIPE_TEXTURE_RECT];
       cso_set_samplers(cso, num, (const struct pipe_sampler_state **) samplers);
    }
 
    /* user textures, plus the bitmap texture */
    {
       struct pipe_sampler_view *sampler_views[PIPE_MAX_SAMPLERS];
-      uint num = MAX2(stfp->bitmap_sampler + 1, st->state.num_textures);
+      uint num = MAX2(fpv->bitmap_sampler + 1, st->state.num_textures);
       memcpy(sampler_views, st->state.sampler_views, sizeof(sampler_views));
-      sampler_views[stfp->bitmap_sampler] = sv;
+      sampler_views[fpv->bitmap_sampler] = sv;
       cso_set_fragment_sampler_views(cso, num, sampler_views);
    }
 
diff --git a/src/mesa/state_tracker/st_cb_bitmap.h b/src/mesa/state_tracker/st_cb_bitmap.h
index d04b2b6779..041dda7dd4 100644
--- a/src/mesa/state_tracker/st_cb_bitmap.h
+++ b/src/mesa/state_tracker/st_cb_bitmap.h
@@ -34,6 +34,8 @@
 
 struct dd_function_table;
 struct st_context;
+struct gl_fragment_program;
+struct st_fragment_program;
 
 #if FEATURE_drawpix
 
@@ -47,6 +49,12 @@ extern void
 st_destroy_bitmap(struct st_context *st);
 
 extern void
+st_make_combined_bitmap_fragment_program(struct st_context *st,
+                                         struct gl_fragment_program *fpIn,
+                                         struct gl_fragment_program **fpOut,
+                                         GLuint *bitmap_sampler);
+
+extern void
 st_flush_bitmap_cache(struct st_context *st);
 
 /* Flush bitmap cache and release vertex buffer.  Needed at end of
diff --git a/src/mesa/state_tracker/st_cb_drawpixels.c b/src/mesa/state_tracker/st_cb_drawpixels.c
index c978602457..e2b1946f15 100644
--- a/src/mesa/state_tracker/st_cb_drawpixels.c
+++ b/src/mesa/state_tracker/st_cb_drawpixels.c
@@ -142,7 +142,9 @@ combined_drawpix_fragment_program(struct gl_context *ctx)
 #endif
 
       /* translate to TGSI tokens */
+#if 0000000000000
       st_translate_fragment_program(st, stfp);
+#endif
 
       /* save new program, update serial numbers */
       st->pixel_xfer.xfer_prog_sn = st->pixel_xfer.program->serialNo;
@@ -158,7 +160,11 @@ combined_drawpix_fragment_program(struct gl_context *ctx)
     */
    st_upload_constants(st, stfp->Base.Base.Parameters, PIPE_SHADER_FRAGMENT);
 
+#if 00000000000
    return stfp->driver_shader;
+#else
+   return 0;
+#endif
 }
 
 
@@ -183,7 +189,11 @@ make_fragment_shader_z_stencil(struct st_context *st, GLboolean write_depth,
 
    if (st->drawpix.shaders[shaderIndex]) {
       /* already have the proper shader */
+#if 00000
       return st->drawpix.shaders[shaderIndex]->driver_shader;
+#else
+      return 0;
+#endif
    }
 
    /*
@@ -247,12 +257,16 @@ make_fragment_shader_z_stencil(struct st_context *st, GLboolean write_depth,
 
    stp = st_fragment_program((struct gl_fragment_program *) p);
 
+#if 0000
    /* save the new shader */
    st->drawpix.shaders[shaderIndex] = stp;
 
    st_translate_fragment_program(st, stp);
 
    return stp->driver_shader;
+#else
+   return 000;
+#endif
 }
 
 
diff --git a/src/mesa/state_tracker/st_cb_program.c b/src/mesa/state_tracker/st_cb_program.c
index 4d83fcc6cc..c803f84915 100644
--- a/src/mesa/state_tracker/st_cb_program.c
+++ b/src/mesa/state_tracker/st_cb_program.c
@@ -167,7 +167,10 @@ st_delete_program(struct gl_context *ctx, struct gl_program *prog)
       break;
    case GL_FRAGMENT_PROGRAM_ARB:
       {
-         struct st_fragment_program *stfp = (struct st_fragment_program *) prog;
+         struct st_fragment_program *stfp =
+            (struct st_fragment_program *) prog;
+         st_fp_release_varients(st, stfp);
+#if 0
 
          if (stfp->driver_shader) {
             cso_delete_fragment_shader(st->cso_context, stfp->driver_shader);
@@ -184,6 +187,8 @@ st_delete_program(struct gl_context *ctx, struct gl_program *prog)
             _mesa_reference_program(ctx, &prg, NULL);
             stfp->bitmap_program = NULL;
          }
+#endif
+
       }
       break;
    default:
@@ -214,6 +219,8 @@ static GLboolean st_program_string_notify( struct gl_context *ctx,
 
       stfp->serialNo++;
 
+      st_fp_release_varients(st, stfp);
+#if 00000
       if (stfp->driver_shader) {
          cso_delete_fragment_shader(st->cso_context, stfp->driver_shader);
          stfp->driver_shader = NULL;
@@ -223,6 +230,7 @@ static GLboolean st_program_string_notify( struct gl_context *ctx,
          st_free_tokens(stfp->tgsi.tokens);
          stfp->tgsi.tokens = NULL;
       }
+#endif
 
       if (st->fp == stfp)
 	 st->dirty.st |= ST_NEW_FRAGMENT_PROGRAM;
diff --git a/src/mesa/state_tracker/st_context.h b/src/mesa/state_tracker/st_context.h
index d342c0cff1..dbc6744eed 100644
--- a/src/mesa/state_tracker/st_context.h
+++ b/src/mesa/state_tracker/st_context.h
@@ -131,6 +131,7 @@ struct st_context
    struct st_geometry_program *gp;  /**< Currently bound geometry program */
 
    struct st_vp_varient *vp_varient;
+   struct st_fp_varient *fp_varient;
 
    struct gl_texture_object *default_texture;
 
diff --git a/src/mesa/state_tracker/st_program.c b/src/mesa/state_tracker/st_program.c
index d7d9e0e36f..322eec880b 100644
--- a/src/mesa/state_tracker/st_program.c
+++ b/src/mesa/state_tracker/st_program.c
@@ -34,6 +34,7 @@
 #include "main/imports.h"
 #include "main/hash.h"
 #include "main/mtypes.h"
+#include "program/prog_parameter.h"
 #include "program/prog_print.h"
 #include "program/programopt.h"
 
@@ -45,6 +46,7 @@
 #include "tgsi/tgsi_ureg.h"
 
 #include "st_debug.h"
+#include "st_cb_bitmap.h"
 #include "st_context.h"
 #include "st_program.h"
 #include "st_mesa_to_tgsi.h"
@@ -95,6 +97,39 @@ st_vp_release_varients( struct st_context *st,
 
 
 
+/**
+ * Delete a fragment program varient.  Note the caller must unlink
+ * the varient from the linked list.
+ */
+static void
+delete_fp_varient(struct st_context *st, struct st_fp_varient *fpv)
+{
+   if (fpv->driver_shader) 
+      cso_delete_fragment_shader(st->cso_context, fpv->driver_shader);
+      
+   FREE(fpv);
+}
+
+
+/**
+ * Free all varients of a fragment program.
+ */
+void
+st_fp_release_varients(struct st_context *st, struct st_fragment_program *stfp)
+{
+   struct st_fp_varient *fpv;
+
+   for (fpv = stfp->varients; fpv; ) {
+      struct st_fp_varient *next = fpv->next;
+      delete_fp_varient(st, fpv);
+      fpv = next;
+   }
+
+   stfp->varients = NULL;
+}
+
+
+
 
 /**
  * Translate a Mesa vertex shader into a TGSI shader.
@@ -290,196 +325,253 @@ fail:
 
 
 /**
- * Translate a Mesa fragment shader into a TGSI shader.
+ * Translate a Mesa fragment shader into a TGSI shader using extra info in
+ * the key.
  * \return  pointer to cached pipe_shader object.
  */
-void
+struct st_fp_varient *
 st_translate_fragment_program(struct st_context *st,
-                              struct st_fragment_program *stfp )
+                              struct st_fragment_program *stfp,
+                              const struct st_fp_varient_key *key)
 {
    struct pipe_context *pipe = st->pipe;
-   GLuint outputMapping[FRAG_RESULT_MAX];
-   GLuint inputMapping[FRAG_ATTRIB_MAX];
-   GLuint interpMode[PIPE_MAX_SHADER_INPUTS];  /* XXX size? */
-   GLuint attr;
-   enum pipe_error error;
-   const GLbitfield inputsRead = stfp->Base.Base.InputsRead;
-   struct ureg_program *ureg;
+   struct st_fp_varient *varient = MALLOC_STRUCT(st_fp_varient);
 
-   ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
-   ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
-   uint fs_num_inputs = 0;
+   if (!varient)
+      return NULL;
 
-   ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
-   ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
-   uint fs_num_outputs = 0;
+   if (key->bitmap) {
+      struct gl_fragment_program *fp;
+      st_make_combined_bitmap_fragment_program(st, &stfp->Base,
+                                               &fp, &varient->bitmap_sampler);
 
-   _mesa_remove_output_reads(&stfp->Base.Base, PROGRAM_OUTPUT);
+      varient->parameters = _mesa_clone_parameter_list(fp->Base.Parameters);
+      stfp = (struct st_fragment_program *) fp;
+   }
 
-   /*
-    * Convert Mesa program inputs to TGSI input register semantics.
-    */
-   for (attr = 0; attr < FRAG_ATTRIB_MAX; attr++) {
-      if (inputsRead & (1 << attr)) {
-         const GLuint slot = fs_num_inputs++;
+   if (!stfp->tgsi.tokens) {
+      /* need to translate Mesa instructions to TGSI now */
+      GLuint outputMapping[FRAG_RESULT_MAX];
+      GLuint inputMapping[FRAG_ATTRIB_MAX];
+      GLuint interpMode[PIPE_MAX_SHADER_INPUTS];  /* XXX size? */
+      GLuint attr;
+      enum pipe_error error;
+      const GLbitfield inputsRead = stfp->Base.Base.InputsRead;
+      struct ureg_program *ureg;
 
-         inputMapping[attr] = slot;
+      ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
+      ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
+      uint fs_num_inputs = 0;
 
-         switch (attr) {
-         case FRAG_ATTRIB_WPOS:
-            input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
-            input_semantic_index[slot] = 0;
-            interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
-            break;
-         case FRAG_ATTRIB_COL0:
-            input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
-            input_semantic_index[slot] = 0;
-            interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
-            break;
-         case FRAG_ATTRIB_COL1:
-            input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
-            input_semantic_index[slot] = 1;
-            interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
-            break;
-         case FRAG_ATTRIB_FOGC:
-            input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
-            input_semantic_index[slot] = 0;
-            interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
-            break;
-         case FRAG_ATTRIB_FACE:
-            input_semantic_name[slot] = TGSI_SEMANTIC_FACE;
-            input_semantic_index[slot] = 0;
-            interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
-            break;
-            /* In most cases, there is nothing special about these
-             * inputs, so adopt a convention to use the generic
-             * semantic name and the mesa FRAG_ATTRIB_ number as the
-             * index. 
-             * 
-             * All that is required is that the vertex shader labels
-             * its own outputs similarly, and that the vertex shader
-             * generates at least every output required by the
-             * fragment shader plus fixed-function hardware (such as
-             * BFC).
-             * 
-             * There is no requirement that semantic indexes start at
-             * zero or be restricted to a particular range -- nobody
-             * should be building tables based on semantic index.
-             */
-         case FRAG_ATTRIB_PNTC:
-         case FRAG_ATTRIB_TEX0:
-         case FRAG_ATTRIB_TEX1:
-         case FRAG_ATTRIB_TEX2:
-         case FRAG_ATTRIB_TEX3:
-         case FRAG_ATTRIB_TEX4:
-         case FRAG_ATTRIB_TEX5:
-         case FRAG_ATTRIB_TEX6:
-         case FRAG_ATTRIB_TEX7:
-         case FRAG_ATTRIB_VAR0:
-         default:
-            /* Actually, let's try and zero-base this just for
-             * readability of the generated TGSI.
-             */
-            assert(attr >= FRAG_ATTRIB_TEX0);
-            input_semantic_index[slot] = (attr - FRAG_ATTRIB_TEX0);
-            input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
-            if (attr == FRAG_ATTRIB_PNTC)
-               interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
-            else
-               interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
-            break;
-         }
-      }
-      else {
-	 inputMapping[attr] = -1;
-      }
-   }
+      ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
+      ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
+      uint fs_num_outputs = 0;
 
-   /*
-    * Semantics and mapping for outputs
-    */
-   {
-      uint numColors = 0;
-      GLbitfield64 outputsWritten = stfp->Base.Base.OutputsWritten;
-
-      /* if z is written, emit that first */
-      if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
-         fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
-         fs_output_semantic_index[fs_num_outputs] = 0;
-         outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs;
-         fs_num_outputs++;
-         outputsWritten &= ~(1 << FRAG_RESULT_DEPTH);
-      }
 
-      if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) {
-         fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_STENCIL;
-         fs_output_semantic_index[fs_num_outputs] = 0;
-         outputMapping[FRAG_RESULT_STENCIL] = fs_num_outputs;
-         fs_num_outputs++;
-         outputsWritten &= ~(1 << FRAG_RESULT_STENCIL);
-      }
+      _mesa_remove_output_reads(&stfp->Base.Base, PROGRAM_OUTPUT);
+
+      /*
+       * Convert Mesa program inputs to TGSI input register semantics.
+       */
+      for (attr = 0; attr < FRAG_ATTRIB_MAX; attr++) {
+         if (inputsRead & (1 << attr)) {
+            const GLuint slot = fs_num_inputs++;
+
+            inputMapping[attr] = slot;
 
-      /* handle remaning outputs (color) */
-      for (attr = 0; attr < FRAG_RESULT_MAX; attr++) {
-         if (outputsWritten & BITFIELD64_BIT(attr)) {
             switch (attr) {
-            case FRAG_RESULT_DEPTH:
-            case FRAG_RESULT_STENCIL:
-               /* handled above */
-               assert(0);
+            case FRAG_ATTRIB_WPOS:
+               input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
+               input_semantic_index[slot] = 0;
+               interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
                break;
+            case FRAG_ATTRIB_COL0:
+               input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
+               input_semantic_index[slot] = 0;
+               interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
+               break;
+            case FRAG_ATTRIB_COL1:
+               input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
+               input_semantic_index[slot] = 1;
+               interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
+               break;
+            case FRAG_ATTRIB_FOGC:
+               input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
+               input_semantic_index[slot] = 0;
+               interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
+               break;
+            case FRAG_ATTRIB_FACE:
+               input_semantic_name[slot] = TGSI_SEMANTIC_FACE;
+               input_semantic_index[slot] = 0;
+               interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
+               break;
+               /* In most cases, there is nothing special about these
+                * inputs, so adopt a convention to use the generic
+                * semantic name and the mesa FRAG_ATTRIB_ number as the
+                * index. 
+                * 
+                * All that is required is that the vertex shader labels
+                * its own outputs similarly, and that the vertex shader
+                * generates at least every output required by the
+                * fragment shader plus fixed-function hardware (such as
+                * BFC).
+                * 
+                * There is no requirement that semantic indexes start at
+                * zero or be restricted to a particular range -- nobody
+                * should be building tables based on semantic index.
+                */
+            case FRAG_ATTRIB_PNTC:
+            case FRAG_ATTRIB_TEX0:
+            case FRAG_ATTRIB_TEX1:
+            case FRAG_ATTRIB_TEX2:
+            case FRAG_ATTRIB_TEX3:
+            case FRAG_ATTRIB_TEX4:
+            case FRAG_ATTRIB_TEX5:
+            case FRAG_ATTRIB_TEX6:
+            case FRAG_ATTRIB_TEX7:
+            case FRAG_ATTRIB_VAR0:
             default:
-               assert(attr == FRAG_RESULT_COLOR ||
-                      (FRAG_RESULT_DATA0 <= attr && attr < FRAG_RESULT_MAX));
-               fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
-               fs_output_semantic_index[fs_num_outputs] = numColors;
-               outputMapping[attr] = fs_num_outputs;
-               numColors++;
+               /* Actually, let's try and zero-base this just for
+                * readability of the generated TGSI.
+                */
+               assert(attr >= FRAG_ATTRIB_TEX0);
+               input_semantic_index[slot] = (attr - FRAG_ATTRIB_TEX0);
+               input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
+               if (attr == FRAG_ATTRIB_PNTC)
+                  interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
+               else
+                  interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
                break;
             }
+         }
+         else {
+            inputMapping[attr] = -1;
+         }
+      }
+
+      /*
+       * Semantics and mapping for outputs
+       */
+      {
+         uint numColors = 0;
+         GLbitfield64 outputsWritten = stfp->Base.Base.OutputsWritten;
+
+         /* if z is written, emit that first */
+         if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
+            fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
+            fs_output_semantic_index[fs_num_outputs] = 0;
+            outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs;
+            fs_num_outputs++;
+            outputsWritten &= ~(1 << FRAG_RESULT_DEPTH);
+         }
 
+         if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) {
+            fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_STENCIL;
+            fs_output_semantic_index[fs_num_outputs] = 0;
+            outputMapping[FRAG_RESULT_STENCIL] = fs_num_outputs;
             fs_num_outputs++;
+            outputsWritten &= ~(1 << FRAG_RESULT_STENCIL);
+         }
+
+         /* handle remaning outputs (color) */
+         for (attr = 0; attr < FRAG_RESULT_MAX; attr++) {
+            if (outputsWritten & BITFIELD64_BIT(attr)) {
+               switch (attr) {
+               case FRAG_RESULT_DEPTH:
+               case FRAG_RESULT_STENCIL:
+                  /* handled above */
+                  assert(0);
+                  break;
+               default:
+                  assert(attr == FRAG_RESULT_COLOR ||
+                         (FRAG_RESULT_DATA0 <= attr && attr < FRAG_RESULT_MAX));
+                  fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
+                  fs_output_semantic_index[fs_num_outputs] = numColors;
+                  outputMapping[attr] = fs_num_outputs;
+                  numColors++;
+                  break;
+               }
+
+               fs_num_outputs++;
+            }
          }
       }
-   }
 
-   ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT );
-   if (ureg == NULL)
-      return;
+      ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT );
+      if (ureg == NULL)
+         return NULL;
 
-   if (ST_DEBUG & DEBUG_MESA) {
-      _mesa_print_program(&stfp->Base.Base);
-      _mesa_print_program_parameters(st->ctx, &stfp->Base.Base);
-      debug_printf("\n");
-   }
+      if (ST_DEBUG & DEBUG_MESA) {
+         _mesa_print_program(&stfp->Base.Base);
+         _mesa_print_program_parameters(st->ctx, &stfp->Base.Base);
+         debug_printf("\n");
+      }
 
-   error = 
-      st_translate_mesa_program(st->ctx,
-                                TGSI_PROCESSOR_FRAGMENT,
-                                ureg,
-                                &stfp->Base.Base,
-                                /* inputs */
-                                fs_num_inputs,
-                                inputMapping,
-                                input_semantic_name,
-                                input_semantic_index,
-                                interpMode,
-                                /* outputs */
-                                fs_num_outputs,
-                                outputMapping,
-                                fs_output_semantic_name,
-                                fs_output_semantic_index, FALSE );
+      error = st_translate_mesa_program(st->ctx,
+                                        TGSI_PROCESSOR_FRAGMENT,
+                                        ureg,
+                                        &stfp->Base.Base,
+                                        /* inputs */
+                                        fs_num_inputs,
+                                        inputMapping,
+                                        input_semantic_name,
+                                        input_semantic_index,
+                                        interpMode,
+                                        /* outputs */
+                                        fs_num_outputs,
+                                        outputMapping,
+                                        fs_output_semantic_name,
+                                        fs_output_semantic_index, FALSE );
+
+      stfp->tgsi.tokens = ureg_get_tokens( ureg, NULL );
+      ureg_destroy( ureg );
+   }
 
-   stfp->tgsi.tokens = ureg_get_tokens( ureg, NULL );
-   ureg_destroy( ureg );
-   stfp->driver_shader = pipe->create_fs_state(pipe, &stfp->tgsi);
+   /* fill in varient */
+   varient->driver_shader = pipe->create_fs_state(pipe, &stfp->tgsi);
+   varient->key = *key;
 
    if (ST_DEBUG & DEBUG_TGSI) {
       tgsi_dump( stfp->tgsi.tokens, 0/*TGSI_DUMP_VERBOSE*/ );
       debug_printf("\n");
    }
+
+   return varient;
 }
 
+
+/**
+ * Translate fragment program if needed.
+ */
+struct st_fp_varient *
+st_get_fp_varient(struct st_context *st,
+                  struct st_fragment_program *stfp,
+                  const struct st_fp_varient_key *key)
+{
+   struct st_fp_varient *fpv;
+
+   /* Search for existing varient */
+   for (fpv = stfp->varients; fpv; fpv = fpv->next) {
+      if (memcmp(&fpv->key, key, sizeof(*key)) == 0) {
+         break;
+      }
+   }
+
+   if (!fpv) {
+      /* create new */
+      fpv = st_translate_fragment_program(st, stfp, key);
+      if (fpv) {
+         /* insert into list */
+         fpv->next = stfp->varients;
+         stfp->varients = fpv;
+      }
+   }
+
+   return fpv;
+}
+
+
+
 void
 st_translate_geometry_program(struct st_context *st,
                               struct st_geometry_program *stgp)
@@ -807,6 +899,21 @@ destroy_program_variants(struct st_context *st, struct gl_program *program)
       {
          struct st_fragment_program *stfp =
             (struct st_fragment_program *) program;
+         struct st_fp_varient *fpv, **prev = &stfp->varients;
+
+         for (fpv = stfp->varients; fpv; ) {
+            struct st_fp_varient *next = fpv->next;
+            if (fpv->key.st == st) {
+               /* unlink from list */
+               *prev = next;
+               /* destroy this variant */
+               delete_fp_varient(st, fpv);
+            }
+            else {
+               prev = &fpv;
+            }
+            fpv = next;
+         }
       }
       break;
    default:
diff --git a/src/mesa/state_tracker/st_program.h b/src/mesa/state_tracker/st_program.h
index 10362ed93b..642a256527 100644
--- a/src/mesa/state_tracker/st_program.h
+++ b/src/mesa/state_tracker/st_program.h
@@ -40,6 +40,34 @@
 #include "st_context.h"
 
 
+/** Fragment program variant key */
+struct st_fp_varient_key
+{
+   struct st_context *st;          /**< variants are per-context */
+   GLboolean bitmap;               /**< glBitmap varient? */
+};
+
+
+/**
+ * Variant of a fragment program.
+ */
+struct st_fp_varient
+{
+   /** Parameters which generated this version of fragment program */
+   struct st_fp_varient_key key;
+
+   /** Driver's compiled shader */
+   void *driver_shader;
+
+   /** For glBitmap variants */
+   struct gl_program_parameter_list *parameters;
+   uint bitmap_sampler;
+
+   /** next in linked list */
+   struct st_fp_varient *next;
+};
+
+
 /**
  * Derived from Mesa gl_fragment_program:
  */
@@ -49,7 +77,11 @@ struct st_fragment_program
    GLuint serialNo;
 
    struct pipe_shader_state tgsi;
+#if 0
    void *driver_shader;
+#else
+   struct st_fp_varient *varients;
+#endif
 
    /** Program prefixed with glBitmap prologue */
    struct st_fragment_program *bitmap_program;
@@ -199,9 +231,16 @@ st_reference_fragprog(struct st_context *st,
 }
 
 
-extern void
+extern struct st_fp_varient *
 st_translate_fragment_program(struct st_context *st,
-                              struct st_fragment_program *fp);
+                              struct st_fragment_program *fp,
+                              const struct st_fp_varient_key *key);
+
+extern struct st_fp_varient *
+st_get_fp_varient(struct st_context *st,
+                  struct st_fragment_program *stfp,
+                  const struct st_fp_varient_key *key);
+
 
 extern void
 st_translate_geometry_program(struct st_context *st,
@@ -224,6 +263,10 @@ st_vp_release_varients( struct st_context *st,
                         struct st_vertex_program *stvp );
 
 extern void
+st_fp_release_varients( struct st_context *st,
+                        struct st_fragment_program *stfp );
+
+extern void
 st_print_shaders(struct gl_context *ctx);
 
 extern void