9 files changed, 122 insertions, 105 deletions

diff --git a/src/gallium/drivers/cell/ppu/cell_state_shader.c b/src/gallium/drivers/cell/ppu/cell_state_shader.c
index fb2e940348..c3a3fbd066 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_shader.c
@@ -30,33 +30,50 @@
 #include "pipe/p_inlines.h"
 #include "pipe/p_winsys.h"
 #include "draw/draw_context.h"
-#if 0
-#include "pipe/p_shader_tokens.h"
-#include "gallivm/gallivm.h"
-#include "tgsi/util/tgsi_dump.h"
-#include "tgsi/exec/tgsi_sse2.h"
-#endif
+#include "tgsi/util/tgsi_parse.h"
 
 #include "cell_context.h"
 #include "cell_state.h"
 
 
+
+/** cast wrapper */
+static INLINE struct cell_fragment_shader_state *
+cell_fragment_shader_state(void *shader)
+{
+   return (struct pipe_shader_state *) shader;
+}
+
+
+/** cast wrapper */
+static INLINE struct cell_vertex_shader_state *
+cell_vertex_shader_state(void *shader)
+{
+   return (struct pipe_shader_state *) shader;
+}
+
+
+
 static void *
 cell_create_fs_state(struct pipe_context *pipe,
                      const struct pipe_shader_state *templ)
 {
    /*struct cell_context *cell = cell_context(pipe);*/
-   struct cell_fragment_shader_state *state;
+   struct cell_fragment_shader_state *cfs;
 
-   state = CALLOC_STRUCT(cell_fragment_shader_state);
-   if (!state)
+   cfs = CALLOC_STRUCT(cell_fragment_shader_state);
+   if (!cfs)
       return NULL;
 
-   state->shader = *templ;
+   cfs->shader.tokens = tgsi_dup_tokens(templ->tokens);
+   if (!cfs->shader.tokens) {
+      FREE(cfs);
+      return NULL;
+   }
 
-   tgsi_scan_shader(templ->tokens, &state->info);
+   tgsi_scan_shader(templ->tokens, &cfs->info);
 
-   return state;
+   return cfs;
 }
 
 
@@ -65,7 +82,7 @@ cell_bind_fs_state(struct pipe_context *pipe, void *fs)
 {
    struct cell_context *cell = cell_context(pipe);
 
-   cell->fs = (struct cell_fragment_shader_state *) fs;
+   cell->fs = cell_fragment_shader_state(fs);
 
    cell->dirty |= CELL_NEW_FS;
 }
@@ -74,10 +91,10 @@ cell_bind_fs_state(struct pipe_context *pipe, void *fs)
 static void
 cell_delete_fs_state(struct pipe_context *pipe, void *fs)
 {
-   struct cell_fragment_shader_state *state =
-      (struct cell_fragment_shader_state *) fs;
+   struct cell_fragment_shader_state *cfs = cell_fragment_shader_state(fs);
 
-   FREE( state );
+   FREE((void *) cfs->shader.tokens);
+   FREE(cfs);
 }
 
 
@@ -86,22 +103,28 @@ cell_create_vs_state(struct pipe_context *pipe,
                      const struct pipe_shader_state *templ)
 {
    struct cell_context *cell = cell_context(pipe);
-   struct cell_vertex_shader_state *state;
+   struct cell_vertex_shader_state *cvs;
 
-   state = CALLOC_STRUCT(cell_vertex_shader_state);
-   if (!state)
+   cvs = CALLOC_STRUCT(cell_vertex_shader_state);
+   if (!cvs)
       return NULL;
 
-   state->shader = *templ;
-   tgsi_scan_shader(templ->tokens, &state->info);
+   cvs->shader.tokens = tgsi_dup_tokens(templ->tokens);
+   if (!cvs->shader.tokens) {
+      FREE(cvs);
+      return NULL;
+   }
 
-   state->draw_data = draw_create_vertex_shader(cell->draw, &state->shader);
-   if (state->draw_data == NULL) {
-      FREE( state );
+   tgsi_scan_shader(templ->tokens, &cvs->info);
+
+   cvs->draw_data = draw_create_vertex_shader(cell->draw, &cvs->shader);
+   if (cvs->draw_data == NULL) {
+      FREE( (void *) cvs->shader.tokens );
+      FREE( cvs );
       return NULL;
    }
 
-   return state;
+   return cvs;
 }
 
 
@@ -110,7 +133,7 @@ cell_bind_vs_state(struct pipe_context *pipe, void *vs)
 {
    struct cell_context *cell = cell_context(pipe);
 
-   cell->vs = (const struct cell_vertex_shader_state *) vs;
+   cell->vs = cell_vertex_shader_state(vs);
 
    draw_bind_vertex_shader(cell->draw,
                            (cell->vs ? cell->vs->draw_data : NULL));
@@ -123,12 +146,11 @@ static void
 cell_delete_vs_state(struct pipe_context *pipe, void *vs)
 {
    struct cell_context *cell = cell_context(pipe);
+   struct cell_vertex_shader_state *cvs = cell_vertex_shader_state(vs);
 
-   struct cell_vertex_shader_state *state =
-      (struct cell_vertex_shader_state *) vs;
-
-   draw_delete_vertex_shader(cell->draw, state->draw_data);
-   FREE( state );
+   draw_delete_vertex_shader(cell->draw, cvs->draw_data);
+   FREE( (void *) cvs->shader.tokens );
+   FREE( cvs );
 }
 
 
diff --git a/src/gallium/drivers/i915simple/i915_state.c b/src/gallium/drivers/i915simple/i915_state.c
index 3d94b52366..4adeb37e86 100644
--- a/src/gallium/drivers/i915simple/i915_state.c
+++ b/src/gallium/drivers/i915simple/i915_state.c
@@ -33,6 +33,7 @@
 #include "pipe/p_winsys.h"
 #include "pipe/p_util.h"
 #include "pipe/p_inlines.h"
+#include "tgsi/util/tgsi_parse.h"
 
 #include "i915_context.h"
 #include "i915_reg.h"
@@ -436,7 +437,7 @@ i915_create_fs_state(struct pipe_context *pipe,
    if (!ifs)
       return NULL;
 
-   ifs->state = *templ;
+   ifs->state.tokens = tgsi_dup_tokens(templ->tokens);
 
    tgsi_scan_shader(templ->tokens, &ifs->info);
 
@@ -465,6 +466,8 @@ void i915_delete_fs_state(struct pipe_context *pipe, void *shader)
       FREE(ifs->program);
    ifs->program_len = 0;
 
+   FREE(ifs->state.tokens);
+
    FREE(ifs);
 }
 
diff --git a/src/gallium/drivers/i965simple/brw_state.c b/src/gallium/drivers/i965simple/brw_state.c
index 376f1487b2..ac243b7e4f 100644
--- a/src/gallium/drivers/i965simple/brw_state.c
+++ b/src/gallium/drivers/i965simple/brw_state.c
@@ -35,6 +35,7 @@
 #include "pipe/p_inlines.h"

 #include "pipe/p_shader_tokens.h"

 #include "tgsi/util/tgsi_dump.h"

+#include "tgsi/util/tgsi_parse.h"

 

 #include "brw_context.h"

 #include "brw_defines.h"

@@ -182,9 +183,7 @@ static void * brw_create_fs_state(struct pipe_context *pipe,
 {

    struct brw_fragment_program *brw_fp = CALLOC_STRUCT(brw_fragment_program);

 

-   /* XXX: Do I have to duplicate the tokens as well??

-    */

-   brw_fp->program = *shader;

+   brw_fp->program.tokens = tgsi_dup_tokens(shader->tokens);

    brw_fp->id = brw_context(pipe)->program_id++;

 

    tgsi_scan_shader(shader->tokens, &brw_fp->info);

@@ -210,7 +209,10 @@ static void brw_bind_fs_state(struct pipe_context *pipe, void *shader)
 

 static void brw_delete_fs_state(struct pipe_context *pipe, void *shader)

 {

-   FREE(shader);

+   struct brw_fragment_program *brw_fp = (struct brw_fragment_program *) shader;

+

+   FREE((void *) brw_fp->program.tokens);

+   FREE(brw_fp);

 }

 

 

@@ -223,9 +225,7 @@ static void *brw_create_vs_state(struct pipe_context *pipe,
 {

    struct brw_vertex_program *brw_vp = CALLOC_STRUCT(brw_vertex_program);

 

-   /* XXX: Do I have to duplicate the tokens as well??

-    */

-   brw_vp->program = *shader;

+   brw_vp->program.tokens = tgsi_dup_tokens(shader->tokens);

    brw_vp->id = brw_context(pipe)->program_id++;

 

    tgsi_scan_shader(shader->tokens, &brw_vp->info);

@@ -251,7 +251,10 @@ static void brw_bind_vs_state(struct pipe_context *pipe, void *vs)
 

 static void brw_delete_vs_state(struct pipe_context *pipe, void *shader)

 {

-   FREE(shader);

+   struct brw_vertex_program *brw_vp = (struct brw_vertex_program *) shader;

+

+   FREE((void *) brw_vp->program.tokens);

+   FREE(brw_vp);

 }

 

 

diff --git a/src/gallium/drivers/softpipe/sp_fs_exec.c b/src/gallium/drivers/softpipe/sp_fs_exec.c
index d5bd7a702f..0b199a2193 100644
--- a/src/gallium/drivers/softpipe/sp_fs_exec.c
+++ b/src/gallium/drivers/softpipe/sp_fs_exec.c
@@ -37,6 +37,7 @@
 #include "pipe/p_util.h"
 #include "pipe/p_inlines.h"
 #include "tgsi/exec/tgsi_exec.h"
+#include "tgsi/util/tgsi_parse.h"
 
 struct sp_exec_fragment_shader {
    struct sp_fragment_shader base;
@@ -116,6 +117,7 @@ exec_run( const struct sp_fragment_shader *base,
 static void 
 exec_delete( struct sp_fragment_shader *base )
 {
+   FREE((void *) base->shader.tokens);
    FREE(base);
 }
 
@@ -137,7 +139,8 @@ softpipe_create_fs_exec(struct softpipe_context *softpipe,
    if (!shader)
       return NULL;
 
-   shader->base.shader = *templ;
+   /* we need to keep a local copy of the tokens */
+   shader->base.shader.tokens = tgsi_dup_tokens(templ->tokens);
    shader->base.prepare = exec_prepare;
    shader->base.run = exec_run;
    shader->base.delete = exec_delete;
diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c
index 25fdfea491..55741cc1df 100644
--- a/src/gallium/drivers/softpipe/sp_fs_sse.c
+++ b/src/gallium/drivers/softpipe/sp_fs_sse.c
@@ -139,7 +139,11 @@ softpipe_create_fs_sse(struct softpipe_context *softpipe,
    }
 
    shader->func = (codegen_function) x86_get_func( &shader->sse2_program );
-   assert(shader->func);
+   if (!shader->func) {
+      x86_release_func( &shader->sse2_program );
+      FREE(shader);
+      return NULL;
+   }
 
    shader->base.shader = *templ;
    shader->base.prepare = fs_sse_prepare;
diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c
index 625d0f9b48..8c88c192f8 100644
--- a/src/gallium/drivers/softpipe/sp_quad_fs.c
+++ b/src/gallium/drivers/softpipe/sp_quad_fs.c
@@ -53,7 +53,6 @@ struct quad_shade_stage
    struct tgsi_sampler samplers[PIPE_MAX_SAMPLERS];
    struct tgsi_exec_machine machine;
    struct tgsi_exec_vector *inputs, *outputs;
-   int colorOutSlot, depthOutSlot;
 };
 
 
@@ -156,20 +155,6 @@ static void shade_begin(struct quad_stage *qs)
       qss->samplers[i].texture = softpipe->texture[i];
    }
 
-   /* find output slots for depth, color */
-   qss->colorOutSlot = -1;
-   qss->depthOutSlot = -1;
-   for (i = 0; i < qss->stage.softpipe->fs->info.num_outputs; i++) {
-      switch (qss->stage.softpipe->fs->info.output_semantic_name[i]) {
-      case TGSI_SEMANTIC_POSITION:
-         qss->depthOutSlot = i;
-         break;
-      case TGSI_SEMANTIC_COLOR:
-         qss->colorOutSlot = i;
-         break;
-      }
-   }
-   
    softpipe->fs->prepare( softpipe->fs, 
 			  &qss->machine,
 			  qss->samplers );
diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c
index 5370d85275..543d86a5cb 100644
--- a/src/gallium/drivers/softpipe/sp_setup.c
+++ b/src/gallium/drivers/softpipe/sp_setup.c
@@ -209,77 +209,76 @@ static INLINE int block( int x )
 
 
 /**
- * Compute mask which indicates which pixels in the 2x2 quad are actually inside
- * the triangle's bounds.
- *
- * this is pretty nasty...  may need to rework flush_spans again to
- * fix it, if possible.
- */
-static unsigned calculate_mask( struct setup_context *setup, int x )
-{
-   unsigned mask = 0x0;
-
-   if (x >= setup->span.left[0] && x < setup->span.right[0])
-      mask |= MASK_TOP_LEFT;
-
-   if (x >= setup->span.left[1] && x < setup->span.right[1])
-      mask |= MASK_BOTTOM_LEFT;
-
-   if (x+1 >= setup->span.left[0] && x+1 < setup->span.right[0])
-      mask |= MASK_TOP_RIGHT;
-
-   if (x+1 >= setup->span.left[1] && x+1 < setup->span.right[1])
-      mask |= MASK_BOTTOM_RIGHT;
-
-   return mask;
-}
-
-
-/**
  * Render a horizontal span of quads
  */
 static void flush_spans( struct setup_context *setup )
 {
+   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, maxright;
    int x;
 
    switch (setup->span.y_flags) {
    case 0x3:
       /* both odd and even lines written (both quad rows) */
-      minleft = MIN2(setup->span.left[0], setup->span.left[1]);
-      maxright = MAX2(setup->span.right[0], setup->span.right[1]);
+      minleft = block(MIN2(xleft0, xleft1));
+      maxright = block(MAX2(xright0, xright1));
+      for (x = minleft; x <= maxright; x += 2) {
+         /* determine which of the four pixels is inside the span bounds */
+         uint mask = 0x0;
+         if (x >= xleft0 && x < xright0)
+            mask |= MASK_TOP_LEFT;
+         if (x >= xleft1 && x < xright1)
+            mask |= MASK_BOTTOM_LEFT;
+         if (x+1 >= xleft0 && x+1 < xright0)
+            mask |= MASK_TOP_RIGHT;
+         if (x+1 >= xleft1 && x+1 < xright1)
+            mask |= MASK_BOTTOM_RIGHT;
+         emit_quad( setup, x, setup->span.y, mask );
+      }
       break;
 
    case 0x1:
       /* only even line written (quad top row) */
-      minleft = setup->span.left[0];
-      maxright = setup->span.right[0];
+      minleft = block(xleft0);
+      maxright = block(xright0);
+      for (x = minleft; x <= maxright; x += 2) {
+         uint mask = 0x0;
+         if (x >= xleft0 && x < xright0)
+            mask |= MASK_TOP_LEFT;
+         if (x+1 >= xleft0 && x+1 < xright0)
+            mask |= MASK_TOP_RIGHT;
+         emit_quad( setup, x, setup->span.y, mask );
+      }
       break;
 
    case 0x2:
       /* only odd line written (quad bottom row) */
-      minleft = setup->span.left[1];
-      maxright = setup->span.right[1];
+      minleft = block(xleft1);
+      maxright = block(xright1);
+      for (x = minleft; x <= maxright; x += 2) {
+         uint mask = 0x0;
+         if (x >= xleft1 && x < xright1)
+            mask |= MASK_BOTTOM_LEFT;
+         if (x+1 >= xleft1 && x+1 < xright1)
+            mask |= MASK_BOTTOM_RIGHT;
+         emit_quad( setup, x, setup->span.y, mask );
+      }
       break;
 
    default:
       return;
    }
 
-   /* XXX this loop could be moved into the above switch cases and
-    * calculate_mask() could be simplified a bit...
-    */
-   for (x = block(minleft); x <= block(maxright); x += 2) {
-      emit_quad( setup, x, setup->span.y,
-                 calculate_mask( setup, x ) );
-   }
-
    setup->span.y = 0;
    setup->span.y_flags = 0;
    setup->span.right[0] = 0;
    setup->span.right[1] = 0;
 }
 
+
 #if DEBUG_VERTS
 static void print_vertex(const struct setup_context *setup,
                          const float (*v)[4])
diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h
index 45056502b8..452e51fa79 100644
--- a/src/gallium/drivers/softpipe/sp_state.h
+++ b/src/gallium/drivers/softpipe/sp_state.h
@@ -54,9 +54,11 @@
 
 struct tgsi_sampler;
 struct tgsi_exec_machine;
+struct vertex_info;
 
 
-/** Subclass of pipe_shader_state (though it doesn't really need to be).
+/**
+ * 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...
  */
@@ -80,11 +82,10 @@ struct sp_fragment_shader {
    void (*delete)( struct sp_fragment_shader * );
 };
 
-struct vertex_info;
 
 /** Subclass of pipe_shader_state */
 struct sp_vertex_shader {
-   struct pipe_shader_state shader;
+   struct pipe_shader_state shader;  /* Note: this field not actually used */
    struct draw_vertex_shader *draw_data;
 };
 
diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c
index 9e77b7e91b..24b91fbc79 100644
--- a/src/gallium/drivers/softpipe/sp_state_fs.c
+++ b/src/gallium/drivers/softpipe/sp_state_fs.c
@@ -102,10 +102,7 @@ softpipe_create_vs_state(struct pipe_context *pipe,
       return NULL;
    }
 
-   state->shader = *templ;
-
-   state->draw_data = draw_create_vertex_shader(softpipe->draw,
-                                                &state->shader);
+   state->draw_data = draw_create_vertex_shader(softpipe->draw, templ);
    if (state->draw_data == NULL) {
       FREE( state );
       return NULL;