Make shaders operate on a block of memory instead of arrays of vertex_header's

7 files changed, 100 insertions, 77 deletions

diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index b3c65c90d6..cdca556fbd 100644
--- a/src/gallium/auxiliary/draw/draw_context.c
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -86,14 +86,12 @@ struct draw_context *draw_create( void )
    /* Statically allocate maximum sized vertices for the cache - could be cleverer...
     */
    {
-      uint i;
       const unsigned size = (MAX_VERTEX_SIZE + 0x0f) & ~0x0f;
-      char *tmp = align_malloc(Elements(draw->vs.queue) * size, 16);
+      char *tmp = align_malloc(VS_QUEUE_LENGTH * size, 16);
       if (!tmp)
          goto fail;
 
-      for (i = 0; i < Elements(draw->vs.queue); i++)
-	 draw->vs.queue[i].vertex = (struct vertex_header *)(tmp + i * size);
+      draw->vs.vertex_cache = tmp;
    }
 
    draw->shader_queue_flush = draw_vertex_shader_queue_flush;
@@ -156,8 +154,8 @@ void draw_destroy( struct draw_context *draw )
 
    tgsi_exec_machine_free_data(&draw->machine);
    
-   if (draw->vs.queue[0].vertex)
-      align_free( draw->vs.queue[0].vertex ); /* Frees all the vertices. */
+   if (draw->vs.vertex_cache)
+      align_free( draw->vs.vertex_cache ); /* Frees all the vertices. */
 
    /* Not so fast -- we're just borrowing this at the moment.
     * 
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index 5c710667fc..1115166192 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -152,7 +152,7 @@ struct draw_vertex_shader {
                    struct draw_context *draw,
                    const unsigned *elts,
                    unsigned count,
-                   struct vertex_header *vOut[] );
+                   void *out );
 
 
    void (*delete)( struct draw_vertex_shader * );
@@ -321,10 +321,8 @@ struct draw_context
    /* Vertex shader queue:
     */
    struct {
-      struct {
-	 unsigned elt;   /**< index into the user's vertex arrays */
-	 struct vertex_header *vertex;
-      } queue[VS_QUEUE_LENGTH];
+      unsigned elts[VS_QUEUE_LENGTH];   /**< index into the user's vertex arrays */
+      char *vertex_cache;
       unsigned queue_nr;
       unsigned post_nr;
    } vs;
@@ -450,4 +448,11 @@ dot4(const float *a, const float *b)
    return result;
 }
 
+static INLINE struct vertex_header *
+draw_header_from_block(char *block, int num)
+{
+   static const unsigned size = (MAX_VERTEX_SIZE + 0x0f) & ~0x0f;
+   return (struct vertex_header*)(block + num * size);
+}
+
 #endif /* DRAW_PRIVATE_H */
diff --git a/src/gallium/auxiliary/draw/draw_vertex_cache.c b/src/gallium/auxiliary/draw/draw_vertex_cache.c
index c0248979e2..9256dec4ea 100644
--- a/src/gallium/auxiliary/draw/draw_vertex_cache.c
+++ b/src/gallium/auxiliary/draw/draw_vertex_cache.c
@@ -71,25 +71,27 @@ static struct vertex_header *get_vertex( struct draw_context *draw,
       /* Cache hit?
        */
       if (draw->vcache.idx[slot].in == i) {
-//	 _mesa_printf("HIT %d %d\n", slot, i);
+	 /*debug_printf("HIT %d %d\n", slot, i);*/
 	 assert(draw->vcache.idx[slot].out < draw->vs.queue_nr);
-	 return draw->vs.queue[draw->vcache.idx[slot].out].vertex;
+	 return draw_header_from_block(draw->vs.vertex_cache,
+                                       draw->vcache.idx[slot].out);
       }
 
       /* Otherwise a collision
        */
       slot = VCACHE_SIZE + draw->vcache.overflow++;
-//      _mesa_printf("XXX %d --> %d\n", i, slot);
+      /*debug_printf("XXX %d --> %d\n", i, slot);*/
    }
 
    /* Deal with the cache miss: 
     */
    {
       unsigned out;
-      
+      struct vertex_header *header;
+
       assert(slot < Elements(draw->vcache.idx));
 
-//      _mesa_printf("NEW %d %d\n", slot, i);
+      /*debug_printf("NEW %d %d\n", slot, i);*/
       draw->vcache.idx[slot].in = i;
       draw->vcache.idx[slot].out = out = draw->vs.queue_nr++;
       draw->vcache.referenced |= (1 << slot);
@@ -99,17 +101,19 @@ static struct vertex_header *get_vertex( struct draw_context *draw,
        */
       assert(draw->vs.queue_nr < VS_QUEUE_LENGTH);
 
-      draw->vs.queue[out].elt = i;
-      draw->vs.queue[out].vertex->clipmask = 0;
-      draw->vs.queue[out].vertex->edgeflag = draw_get_edgeflag(draw, i);
-      draw->vs.queue[out].vertex->pad = 0;
-      draw->vs.queue[out].vertex->vertex_id = UNDEFINED_VERTEX_ID;
+      header = draw_header_from_block(draw->vs.vertex_cache, out);
+      draw->vs.elts[out] = i;
+      header->clipmask = 0;
+      header->edgeflag = draw_get_edgeflag(draw, i);
+      header->pad = 0;
+      header->vertex_id = UNDEFINED_VERTEX_ID;
 
       /* Need to set the vertex's edge flag here.  If we're being called
        * by do_ef_triangle(), that function needs edge flag info!
        */
 
-      return draw->vs.queue[draw->vcache.idx[slot].out].vertex;
+      return draw_header_from_block(draw->vs.vertex_cache,
+                                    draw->vcache.idx[slot].out);
    }
 }
 
@@ -142,8 +146,11 @@ void draw_vertex_cache_reset_vertex_ids( struct draw_context *draw )
 {
    unsigned i;
 
-   for (i = 0; i < draw->vs.post_nr; i++)
-      draw->vs.queue[i].vertex->vertex_id = UNDEFINED_VERTEX_ID;
+   for (i = 0; i < draw->vs.post_nr; i++) {
+      struct vertex_header * header =
+         draw_header_from_block(draw->vs.vertex_cache, i);
+      header->vertex_id = UNDEFINED_VERTEX_ID;
+   }
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vertex_shader.c b/src/gallium/auxiliary/draw/draw_vertex_shader.c
index 726921d77b..452d0175c3 100644
--- a/src/gallium/auxiliary/draw/draw_vertex_shader.c
+++ b/src/gallium/auxiliary/draw/draw_vertex_shader.c
@@ -57,18 +57,17 @@ draw_vertex_shader_queue_flush(struct draw_context *draw)
 
    /* run vertex shader on vertex cache entries, four per invokation */
    for (i = 0; i < draw->vs.queue_nr; i += MAX_SHADER_VERTICES) {
-      struct vertex_header *dests[MAX_SHADER_VERTICES];
       unsigned elts[MAX_SHADER_VERTICES];
       int j, n = MIN2(MAX_SHADER_VERTICES, draw->vs.queue_nr  - i);
+      struct vertex_header *dests =
+         draw_header_from_block(draw->vs.vertex_cache, i);
 
       for (j = 0; j < n; j++) {
-         elts[j] = draw->vs.queue[i + j].elt;
-         dests[j] = draw->vs.queue[i + j].vertex;
+         elts[j] = draw->vs.elts[i + j];
       }
 
       for ( ; j < MAX_SHADER_VERTICES; j++) {
 	 elts[j] = elts[0];
-         dests[j] = draw->vs.queue[i + j].vertex;
       }
 
       assert(n > 0);
diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c
index 09e0d0eaab..6fe4e554d5 100644
--- a/src/gallium/auxiliary/draw/draw_vs_exec.c
+++ b/src/gallium/auxiliary/draw/draw_vs_exec.c
@@ -69,7 +69,7 @@ vs_exec_run( struct draw_vertex_shader *shader,
 	     struct draw_context *draw,
 	     const unsigned *elts, 
 	     unsigned count,
-	     struct vertex_header *vOut[] )
+	     void *vOut )
 {
    struct tgsi_exec_machine *machine = &draw->machine;
    unsigned int i, j;
@@ -106,6 +106,8 @@ vs_exec_run( struct draw_vertex_shader *shader,
       for (j = 0; j < max_vertices; j++) {
          unsigned slot;
          float x, y, z, w;
+         struct vertex_header *out =
+            draw_header_from_block(vOut, i + j);
 
          /* Handle attr[0] (position) specially:
           *
@@ -113,15 +115,15 @@ vs_exec_run( struct draw_vertex_shader *shader,
           * program as a set of DP4 instructions appended to the
           * user-provided code.
           */
-         x = vOut[i + j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
-         y = vOut[i + j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
-         z = vOut[i + j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
-         w = vOut[i + j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
+         x = out->clip[0] = machine->Outputs[0].xyzw[0].f[j];
+         y = out->clip[1] = machine->Outputs[0].xyzw[1].f[j];
+         z = out->clip[2] = machine->Outputs[0].xyzw[2].f[j];
+         w = out->clip[3] = machine->Outputs[0].xyzw[3].f[j];
 
          if (!draw->rasterizer->bypass_clipping) {
-            vOut[i + j]->clipmask = compute_clipmask(vOut[i + j]->clip, draw->plane,
-                                                     draw->nr_planes);
-            clipped += vOut[i + j]->clipmask;
+            out->clipmask = compute_clipmask(out->clip, draw->plane,
+                                             draw->nr_planes);
+            clipped += out->clipmask;
 
             /* divide by w */
             w = 1.0f / w;
@@ -130,42 +132,42 @@ vs_exec_run( struct draw_vertex_shader *shader,
             z *= w;
          }
          else {
-            vOut[i + j]->clipmask = 0;
+            out->clipmask = 0;
          }
-         vOut[i + j]->edgeflag = 1;
+         out->edgeflag = 1;
 
          if (!draw->identity_viewport) {
             /* Viewport mapping */
-            vOut[i + j]->data[0][0] = x * scale[0] + trans[0];
-            vOut[i + j]->data[0][1] = y * scale[1] + trans[1];
-            vOut[i + j]->data[0][2] = z * scale[2] + trans[2];
-            vOut[i + j]->data[0][3] = w;
+            out->data[0][0] = x * scale[0] + trans[0];
+            out->data[0][1] = y * scale[1] + trans[1];
+            out->data[0][2] = z * scale[2] + trans[2];
+            out->data[0][3] = w;
          }
          else {
-            vOut[i + j]->data[0][0] = x;
-            vOut[i + j]->data[0][1] = y;
-            vOut[i + j]->data[0][2] = z;
-            vOut[i + j]->data[0][3] = w;
+            out->data[0][0] = x;
+            out->data[0][1] = y;
+            out->data[0][2] = z;
+            out->data[0][3] = w;
          }
 
          /* Remaining attributes are packed into sequential post-transform
           * vertex attrib slots.
           */
          for (slot = 1; slot < draw->num_vs_outputs; slot++) {
-            vOut[i + j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
-            vOut[i + j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
-            vOut[i + j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
-            vOut[i + j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
+            out->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
+            out->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
+            out->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
+            out->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
          }
 
 #if 0 /*DEBUG*/
          printf("%d) Post xform vert:\n", i + j);
          for (slot = 0; slot < draw->num_vs_outputs; slot++) {
             printf("\t%d: %f %f %f %f\n", slot,
-                   vOut[i + j]->data[slot][0],
-                   vOut[i + j]->data[slot][1],
-                   vOut[i + j]->data[slot][2],
-                   vOut[i + j]->data[slot][3]);
+                   out->data[slot][0],
+                   out->data[slot][1],
+                   out->data[slot][2],
+                   out->data[slot][3]);
          }
 #endif
       } /* loop over vertices */
diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c
index 6db1e65a2d..72317c67ae 100644
--- a/src/gallium/auxiliary/draw/draw_vs_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c
@@ -72,7 +72,7 @@ vs_llvm_run( struct draw_vertex_shader *base,
 	     struct draw_context *draw,
 	     const unsigned *elts,
 	     unsigned count,
-	     struct vertex_header *vOut[] )
+	     void *vOut )
 {
    struct draw_llvm_vertex_shader *shader =
       (struct draw_llvm_vertex_shader *)base;
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
index 6e8d2021f5..c877f5ee3a 100644
--- a/src/gallium/auxiliary/draw/draw_vs_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -83,7 +83,7 @@ vs_sse_run( struct draw_vertex_shader *base,
 	    struct draw_context *draw, 
 	    const unsigned *elts, 
 	    unsigned count,
-	    struct vertex_header *vOut[] )
+	    void *vOut )
 {
    struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
    struct tgsi_exec_machine *machine = &draw->machine;
@@ -135,16 +135,18 @@ vs_sse_run( struct draw_vertex_shader *base,
       for (j = 0; j < max_vertices; j++) {
          unsigned slot;
          float x, y, z, w;
+         struct vertex_header *out =
+            draw_header_from_block(vOut, i + j);
 
-         x = vOut[i + j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
-         y = vOut[i + j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
-         z = vOut[i + j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
-         w = vOut[i + j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
+         x = out->clip[0] = machine->Outputs[0].xyzw[0].f[j];
+         y = out->clip[1] = machine->Outputs[0].xyzw[1].f[j];
+         z = out->clip[2] = machine->Outputs[0].xyzw[2].f[j];
+         w = out->clip[3] = machine->Outputs[0].xyzw[3].f[j];
 
          if (!draw->rasterizer->bypass_clipping) {
-            vOut[i + j]->clipmask = compute_clipmask(vOut[i + j]->clip, draw->plane,
-                                                     draw->nr_planes);
-            clipped += vOut[i + j]->clipmask;
+            out->clipmask = compute_clipmask(out->clip, draw->plane,
+                                             draw->nr_planes);
+            clipped += out->clipmask;
 
             /* divide by w */
             w = 1.0f / w;
@@ -153,33 +155,43 @@ vs_sse_run( struct draw_vertex_shader *base,
             z *= w;
          }
          else {
-            vOut[i + j]->clipmask = 0;
+            out->clipmask = 0;
          }
-         vOut[j]->edgeflag = 1;
+         out->edgeflag = 1;
 
          if (!draw->identity_viewport) {
             /* Viewport mapping */
-            vOut[i + j]->data[0][0] = x * scale[0] + trans[0];
-            vOut[i + j]->data[0][1] = y * scale[1] + trans[1];
-            vOut[i + j]->data[0][2] = z * scale[2] + trans[2];
-            vOut[i + j]->data[0][3] = w;
+            out->data[0][0] = x * scale[0] + trans[0];
+            out->data[0][1] = y * scale[1] + trans[1];
+            out->data[0][2] = z * scale[2] + trans[2];
+            out->data[0][3] = w;
          }
          else {
-            vOut[i + j]->data[0][0] = x;
-            vOut[i + j]->data[0][1] = y;
-            vOut[i + j]->data[0][2] = z;
-            vOut[i + j]->data[0][3] = w;
+            out->data[0][0] = x;
+            out->data[0][1] = y;
+            out->data[0][2] = z;
+            out->data[0][3] = w;
          }
 
          /* Remaining attributes are packed into sequential post-transform
           * vertex attrib slots.
           */
          for (slot = 1; slot < draw->num_vs_outputs; slot++) {
-            vOut[i + j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
-            vOut[i + j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
-            vOut[i + j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
-            vOut[i + j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
+            out->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
+            out->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
+            out->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
+            out->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
          }
+#if 0 /*DEBUG*/
+         printf("%d) Post xform vert:\n", i + j);
+         for (slot = 0; slot < draw->num_vs_outputs; slot++) {
+            printf("\t%d: %f %f %f %f\n", slot,
+                   out->data[slot][0],
+                   out->data[slot][1],
+                   out->data[slot][2],
+                   out->data[slot][3]);
+         }
+#endif
       }
    }
    return clipped != 0;