Merge branch 'master' into glsl-pp-rework-2

Conflicts:
	progs/perf/drawoverhead.c
	progs/perf/teximage.c
	progs/perf/vbo.c
	progs/perf/vertexrate.c
	src/mesa/shader/slang/library/slang_common_builtin_gc.h

1 files changed, 87 insertions, 239 deletions

diff --git a/src/mesa/state_tracker/st_atom_shader.c b/src/mesa/state_tracker/st_atom_shader.c
index ee649be885..09baff875b 100644
--- a/src/mesa/state_tracker/st_atom_shader.c
+++ b/src/mesa/state_tracker/st_atom_shader.c
@@ -56,82 +56,18 @@
 #include "st_mesa_to_tgsi.h"
 
 
-/**
- * This represents a vertex program, especially translated to match
- * the inputs of a particular fragment shader.
- */
-struct translated_vertex_program
-{
-   struct st_vertex_program *master;
-
-   /** The fragment shader "signature" this vertex shader is meant for: */
-   GLbitfield frag_inputs;
 
-   /** Compared against master vertex program's serialNo: */
-   GLuint serialNo;
 
-   /** Maps VERT_RESULT_x to slot */
-   GLuint output_to_slot[VERT_RESULT_MAX];
-   ubyte output_to_semantic_name[VERT_RESULT_MAX];
-   ubyte output_to_semantic_index[VERT_RESULT_MAX];
-
-   /** Pointer to the translated vertex program */
-   struct st_vertex_program *vp;
-
-   struct translated_vertex_program *next;  /**< next in linked list */
-};
 
-
-
-/**
- * Given a vertex program output attribute, return the corresponding
- * fragment program input attribute.
- * \return -1 for vertex outputs that have no corresponding fragment input
+/*
+ * Translate fragment program if needed.
  */
-static GLint
-vp_out_to_fp_in(GLuint vertResult)
-{
-   if (vertResult >= VERT_RESULT_TEX0 &&
-       vertResult < VERT_RESULT_TEX0 + MAX_TEXTURE_COORD_UNITS)
-      return FRAG_ATTRIB_TEX0 + (vertResult - VERT_RESULT_TEX0);
-
-   if (vertResult >= VERT_RESULT_VAR0 &&
-       vertResult < VERT_RESULT_VAR0 + MAX_VARYING)
-      return FRAG_ATTRIB_VAR0 + (vertResult - VERT_RESULT_VAR0);
-
-   switch (vertResult) {
-   case VERT_RESULT_HPOS:
-      return FRAG_ATTRIB_WPOS;
-   case VERT_RESULT_COL0:
-      return FRAG_ATTRIB_COL0;
-   case VERT_RESULT_COL1:
-      return FRAG_ATTRIB_COL1;
-   case VERT_RESULT_FOGC:
-      return FRAG_ATTRIB_FOGC;
-   default:
-      /* Back-face colors, edge flags, etc */
-      return -1;
-   }
-}
-
-
-/**
- * Find a translated vertex program that corresponds to stvp and
- * has outputs matched to stfp's inputs.
- * This performs vertex and fragment translation (to TGSI) when needed.
- */
-static struct translated_vertex_program *
-find_translated_vp(struct st_context *st,
-                   struct st_vertex_program *stvp,
-                   struct st_fragment_program *stfp)
+static void
+translate_fp(struct st_context *st,
+             struct st_fragment_program *stfp)
 {
-   static const GLuint UNUSED = ~0;
-   struct translated_vertex_program *xvp;
    const GLbitfield fragInputsRead = stfp->Base.Base.InputsRead;
 
-   /*
-    * Translate fragment program if needed.
-    */
    if (!stfp->state.tokens) {
       GLuint inAttr, numIn = 0;
 
@@ -141,7 +77,7 @@ find_translated_vp(struct st_context *st,
             numIn++;
          }
          else {
-            stfp->input_to_slot[inAttr] = UNUSED;
+            stfp->input_to_slot[inAttr] = -1;
          }
       }
 
@@ -151,170 +87,63 @@ find_translated_vp(struct st_context *st,
 
       st_translate_fragment_program(st, stfp, stfp->input_to_slot);
    }
+}
 
 
-   /* See if we've got a translated vertex program whose outputs match
-    * the fragment program's inputs.
-    * XXX This could be a hash lookup, using InputsRead as the key.
-    */
-   for (xvp = stfp->vertex_programs; xvp; xvp = xvp->next) {
-      if (xvp->master == stvp && xvp->frag_inputs == fragInputsRead) {
-         break;
-      }
-   }
 
-   /* No?  Allocate translated vp object now */
-   if (!xvp) {
-      xvp = ST_CALLOC_STRUCT(translated_vertex_program);
-      xvp->frag_inputs = fragInputsRead;
-      xvp->master = stvp;
+/**
+ * Find a translated vertex program that corresponds to stvp and
+ * has outputs matched to stfp's inputs.
+ * This performs vertex and fragment translation (to TGSI) when needed.
+ */
+static struct st_vp_varient *
+find_translated_vp(struct st_context *st,
+                   struct st_vertex_program *stvp )
+{
+   struct st_vp_varient *vpv;
+   struct st_vp_varient_key key;
 
-      xvp->next = stfp->vertex_programs;
-      stfp->vertex_programs = xvp;
-   }
+   /* Nothing in our key yet.  This will change:
+    */
+   memset(&key, 0, sizeof key);
+   key.dummy = 0;
 
-   /* See if we need to translate vertex program to TGSI form */
-   if (xvp->serialNo != stvp->serialNo) {
-      GLuint outAttr;
-      const GLbitfield outputsWritten = stvp->Base.Base.OutputsWritten;
-      GLuint numVpOuts = 0;
-      GLboolean emitPntSize = GL_FALSE, emitBFC0 = GL_FALSE, emitBFC1 = GL_FALSE;
-      GLbitfield usedGenerics = 0x0;
-      GLbitfield usedOutputSlots = 0x0;
-
-      /* Compute mapping of vertex program outputs to slots, which depends
-       * on the fragment program's input->slot mapping.
+   /* Do we need to throw away old translations after a change in the
+    * GL program string?
+    */
+   if (stvp->serialNo != stvp->lastSerialNo) {
+      /* These may have changed if the program string changed.
        */
-      for (outAttr = 0; outAttr < VERT_RESULT_MAX; outAttr++) {
-         /* set defaults: */
-         xvp->output_to_slot[outAttr] = UNUSED;
-         xvp->output_to_semantic_name[outAttr] = TGSI_SEMANTIC_COUNT;
-         xvp->output_to_semantic_index[outAttr] = 99;
-
-         if (outAttr == VERT_RESULT_HPOS) {
-            /* always put xformed position into slot zero */
-            GLuint slot = 0;
-            xvp->output_to_slot[VERT_RESULT_HPOS] = slot;
-            xvp->output_to_semantic_name[outAttr] = TGSI_SEMANTIC_POSITION;
-            xvp->output_to_semantic_index[outAttr] = 0;
-            numVpOuts++;
-            usedOutputSlots |= (1 << slot);
-         }
-         else if (outputsWritten & (1 << outAttr)) {
-            /* see if the frag prog wants this vert output */
-            GLint fpInAttrib = vp_out_to_fp_in(outAttr);
-            if (fpInAttrib >= 0) {
-               GLuint fpInSlot = stfp->input_to_slot[fpInAttrib];
-               if (fpInSlot != ~0) {
-                  /* match this vp output to the fp input */
-                  GLuint vpOutSlot = stfp->input_map[fpInSlot];
-                  xvp->output_to_slot[outAttr] = vpOutSlot;
-                  xvp->output_to_semantic_name[outAttr] = stfp->input_semantic_name[fpInSlot];
-                  xvp->output_to_semantic_index[outAttr] = stfp->input_semantic_index[fpInSlot];
-                  numVpOuts++;
-                  usedOutputSlots |= (1 << vpOutSlot);
-               }
-               else {
-#if 0 /*debug*/
-                  printf("VP output %d not used by FP\n", outAttr);
-#endif
-               }
-            }
-            else if (outAttr == VERT_RESULT_PSIZ)
-               emitPntSize = GL_TRUE;
-            else if (outAttr == VERT_RESULT_BFC0)
-               emitBFC0 = GL_TRUE;
-            else if (outAttr == VERT_RESULT_BFC1)
-               emitBFC1 = GL_TRUE;
-         }
-#if 0 /*debug*/
-         printf("assign vp output_to_slot[%d] = %d\n", outAttr, 
-                xvp->output_to_slot[outAttr]);
-#endif
-      }
-
-      /* must do these last */
-      if (emitPntSize) {
-         GLuint slot = numVpOuts++;
-         xvp->output_to_slot[VERT_RESULT_PSIZ] = slot;
-         xvp->output_to_semantic_name[VERT_RESULT_PSIZ] = TGSI_SEMANTIC_PSIZE;
-         xvp->output_to_semantic_index[VERT_RESULT_PSIZ] = 0;
-         usedOutputSlots |= (1 << slot);
-      }
-      if (emitBFC0) {
-         GLuint slot = numVpOuts++;
-         xvp->output_to_slot[VERT_RESULT_BFC0] = slot;
-         xvp->output_to_semantic_name[VERT_RESULT_BFC0] = TGSI_SEMANTIC_COLOR;
-         xvp->output_to_semantic_index[VERT_RESULT_BFC0] = 0;
-         usedOutputSlots |= (1 << slot);
-      }
-      if (emitBFC1) {
-         GLuint slot = numVpOuts++;
-         xvp->output_to_slot[VERT_RESULT_BFC1] = slot;
-         xvp->output_to_semantic_name[VERT_RESULT_BFC1] = TGSI_SEMANTIC_COLOR;
-         xvp->output_to_semantic_index[VERT_RESULT_BFC1] = 1;
-         usedOutputSlots |= (1 << slot);
-      }
-
-      /* build usedGenerics mask */
-      usedGenerics = 0x0;
-      for (outAttr = 0; outAttr < VERT_RESULT_MAX; outAttr++) {
-         if (xvp->output_to_semantic_name[outAttr] == TGSI_SEMANTIC_GENERIC) {
-            usedGenerics |= (1 << xvp->output_to_semantic_index[outAttr]);
-         }
-      }
+      st_prepare_vertex_program( st, stvp );
 
-      /* For each vertex program output that doesn't match up to a fragment
-       * program input, map the vertex program output to a free slot and
-       * free generic attribute.
+      /* We are now up-to-date:
        */
-      for (outAttr = 0; outAttr < VERT_RESULT_MAX; outAttr++) {
-         if (outputsWritten & (1 << outAttr)) {
-            if (xvp->output_to_slot[outAttr] == UNUSED) {
-               GLint freeGeneric = _mesa_ffs(~usedGenerics) - 1;
-               GLint freeSlot = _mesa_ffs(~usedOutputSlots) - 1;
-               usedGenerics |= (1 << freeGeneric);
-               usedOutputSlots |= (1 << freeSlot);
-               xvp->output_to_slot[outAttr] = freeSlot;
-               xvp->output_to_semantic_name[outAttr] = TGSI_SEMANTIC_GENERIC;
-               xvp->output_to_semantic_index[outAttr] = freeGeneric;
-            }
-         }
-
-#if 0 /*debug*/
-         printf("vp output_to_slot[%d] = %d\n", outAttr, 
-                xvp->output_to_slot[outAttr]);
-#endif
+      stvp->lastSerialNo = stvp->serialNo;
+   }
+   
+   /* See if we've got a translated vertex program whose outputs match
+    * the fragment program's inputs.
+    */
+   for (vpv = stvp->varients; vpv; vpv = vpv->next) {
+      if (memcmp(&vpv->key, &key, sizeof key) == 0) {
+         break;
       }
+   }
 
-      assert(stvp->Base.Base.NumInstructions > 1);
-
-      st_translate_vertex_program(st, stvp, xvp->output_to_slot,
-                                  xvp->output_to_semantic_name,
-                                  xvp->output_to_semantic_index);
-
-      xvp->vp = stvp;
-
-      /* translated VP is up to date now */
-      xvp->serialNo = stvp->serialNo;
+   /* No?  Perform new translation here. */
+   if (!vpv) {
+      vpv = st_translate_vertex_program(st, stvp, &key);
+      if (!vpv)
+         return NULL;
+      
+      vpv->next = stvp->varients;
+      stvp->varients = vpv;
    }
 
-   return xvp;
+   return vpv;
 }
 
 
-void
-st_free_translated_vertex_programs(struct st_context *st,
-                                   struct translated_vertex_program *xvp)
-{
-   struct translated_vertex_program *next;
-
-   while (xvp) {
-      next = xvp->next;
-      _mesa_free(xvp);
-      xvp = next;
-   }
-}
 
 
 static void *
@@ -328,32 +157,19 @@ get_passthrough_fs(struct st_context *st)
    return st->passthrough_fs;
 }
 
-
 static void
-update_linkage( struct st_context *st )
+update_fp( struct st_context *st )
 {
-   struct st_vertex_program *stvp;
    struct st_fragment_program *stfp;
-   struct translated_vertex_program *xvp;
-
-   /* find active shader and params -- Should be covered by
-    * ST_NEW_VERTEX_PROGRAM
-    */
-   assert(st->ctx->VertexProgram._Current);
-   stvp = st_vertex_program(st->ctx->VertexProgram._Current);
-   assert(stvp->Base.Base.Target == GL_VERTEX_PROGRAM_ARB);
 
    assert(st->ctx->FragmentProgram._Current);
    stfp = st_fragment_program(st->ctx->FragmentProgram._Current);
    assert(stfp->Base.Base.Target == GL_FRAGMENT_PROGRAM_ARB);
 
-   xvp = find_translated_vp(st, stvp, stfp);
+   translate_fp(st, stfp);
 
-   st_reference_vertprog(st, &st->vp, stvp);
    st_reference_fragprog(st, &st->fp, stfp);
 
-   cso_set_vertex_shader_handle(st->cso_context, stvp->driver_shader);
-
    if (st->missing_textures) {
       /* use a pass-through frag shader that uses no textures */
       void *fs = get_passthrough_fs(st);
@@ -362,16 +178,48 @@ update_linkage( struct st_context *st )
    else {
       cso_set_fragment_shader_handle(st->cso_context, stfp->driver_shader);
    }
+}
+
+const struct st_tracked_state st_update_fp = {
+   "st_update_fp",					/* name */
+   {							/* dirty */
+      0,						/* mesa */
+      ST_NEW_FRAGMENT_PROGRAM                           /* st */
+   },
+   update_fp  					/* update */
+};
+
+
+
+
+static void
+update_vp( struct st_context *st )
+{
+   struct st_vertex_program *stvp;
+
+   /* find active shader and params -- Should be covered by
+    * ST_NEW_VERTEX_PROGRAM
+    */
+   assert(st->ctx->VertexProgram._Current);
+   stvp = st_vertex_program(st->ctx->VertexProgram._Current);
+   assert(stvp->Base.Base.Target == GL_VERTEX_PROGRAM_ARB);
+
+   st->vp_varient = find_translated_vp(st, stvp);
+
+   st_reference_vertprog(st, &st->vp, stvp);
+
+   cso_set_vertex_shader_handle(st->cso_context, 
+                                st->vp_varient->driver_shader);
 
-   st->vertex_result_to_slot = xvp->output_to_slot;
+   st->vertex_result_to_slot = stvp->result_to_output;
 }
 
 
-const struct st_tracked_state st_update_shader = {
-   "st_update_shader",					/* name */
+const struct st_tracked_state st_update_vp = {
+   "st_update_vp",					/* name */
    {							/* dirty */
       0,						/* mesa */
-      ST_NEW_VERTEX_PROGRAM | ST_NEW_FRAGMENT_PROGRAM	/* st */
+      ST_NEW_VERTEX_PROGRAM                             /* st */
    },
-   update_linkage					/* update */
+   update_vp					/* update */
 };