1 files changed, 39 insertions, 22 deletions

diff --git a/src/mesa/drivers/dri/i915/i915_fragprog.c b/src/mesa/drivers/dri/i915/i915_fragprog.c
index 7a9fb7f088..2bfe665cb6 100644
--- a/src/mesa/drivers/dri/i915/i915_fragprog.c
+++ b/src/mesa/drivers/dri/i915/i915_fragprog.c
@@ -269,7 +269,7 @@ translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
 #define EMIT_TEX( OP )						\
 do {								\
    GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget );	\
-   const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current; \
+   const struct gl_fragment_program *program = &p->FragProg;	\
    GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit];	\
    GLuint sampler = i915_emit_decl(p, REG_TYPE_S,		\
 				   unit, dim);			\
@@ -304,8 +304,9 @@ do {									\
  */
 static void calc_live_regs( struct i915_fragment_program *p )
 {
-    const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current;
+    const struct gl_fragment_program *program = &p->FragProg;
     GLuint regsUsed = 0xffff0000;
+    uint8_t live_components[16] = { 0, };
     GLint i;
    
     for (i = program->Base.NumInstructions - 1; i >= 0; i--) {
@@ -314,13 +315,26 @@ static void calc_live_regs( struct i915_fragment_program *p )
         int a;
 
         /* Register is written to: unmark as live for this and preceeding ops */ 
-        if (inst->DstReg.File == PROGRAM_TEMPORARY)
-            regsUsed &= ~(1 << inst->DstReg.Index);
+        if (inst->DstReg.File == PROGRAM_TEMPORARY) {
+            live_components[inst->DstReg.Index] &= ~inst->DstReg.WriteMask;
+            if (live_components[inst->DstReg.Index] == 0)
+                regsUsed &= ~(1 << inst->DstReg.Index);
+        }
 
         for (a = 0; a < opArgs; a++) {
             /* Register is read from: mark as live for this and preceeding ops */ 
-            if (inst->SrcReg[a].File == PROGRAM_TEMPORARY)
+            if (inst->SrcReg[a].File == PROGRAM_TEMPORARY) {
+                unsigned c;
+
                 regsUsed |= 1 << inst->SrcReg[a].Index;
+
+                for (c = 0; c < 4; c++) {
+                    const unsigned field = GET_SWZ(inst->SrcReg[a].Swizzle, c);
+
+                    if (field <= SWIZZLE_W)
+                        live_components[inst->SrcReg[a].Index] |= (1U << field);
+                }
+            }
         }
 
         p->usedRegs[i] = regsUsed;
@@ -330,7 +344,7 @@ static void calc_live_regs( struct i915_fragment_program *p )
 static GLuint get_live_regs( struct i915_fragment_program *p, 
                              const struct prog_instruction *inst )
 {
-    const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current;
+    const struct gl_fragment_program *program = &p->FragProg;
     GLuint nr = inst - program->Base.Instructions;
 
     return p->usedRegs[nr];
@@ -351,8 +365,7 @@ static GLuint get_live_regs( struct i915_fragment_program *p,
 static void
 upload_program(struct i915_fragment_program *p)
 {
-   const struct gl_fragment_program *program =
-      p->ctx->FragmentProgram._Current;
+   const struct gl_fragment_program *program = &p->FragProg;
    const struct prog_instruction *inst = program->Base.Instructions;
 
    if (INTEL_DEBUG & DEBUG_WM)
@@ -795,18 +808,18 @@ upload_program(struct i915_fragment_program *p)
 	 flags = get_result_flags(inst);
 	 dst = get_result_vector(p, inst);
 
-	 /* dst = src1 >= src2 */
+	 /* tmp = src1 >= src2 */
 	 i915_emit_arith(p,
 			 A0_SGE,
-			 dst,
+			 tmp,
 			 flags, 0,
 			 src_vector(p, &inst->SrcReg[0], program),
 			 src_vector(p, &inst->SrcReg[1], program),
 			 0);
-	 /* tmp = src1 <= src2 */
+	 /* dst = src1 <= src2 */
 	 i915_emit_arith(p,
 			 A0_SGE,
-			 tmp,
+			 dst,
 			 flags, 0,
 			 negate(src_vector(p, &inst->SrcReg[0], program),
 				1, 1, 1, 1),
@@ -944,18 +957,18 @@ upload_program(struct i915_fragment_program *p)
 	 flags = get_result_flags(inst);
 	 dst = get_result_vector(p, inst);
 
-	 /* dst = src1 < src2 */
+	 /* tmp = src1 < src2 */
 	 i915_emit_arith(p,
 			 A0_SLT,
-			 dst,
+			 tmp,
 			 flags, 0,
 			 src_vector(p, &inst->SrcReg[0], program),
 			 src_vector(p, &inst->SrcReg[1], program),
 			 0);
-	 /* tmp = src1 > src2 */
+	 /* dst = src1 > src2 */
 	 i915_emit_arith(p,
 			 A0_SLT,
-			 tmp,
+			 dst,
 			 flags, 0,
 			 negate(src_vector(p, &inst->SrcReg[0], program),
 				1, 1, 1, 1),
@@ -1152,7 +1165,7 @@ translate_program(struct i915_fragment_program *p)
 
    if (INTEL_DEBUG & DEBUG_WM) {
       printf("fp:\n");
-      _mesa_print_program(&p->ctx->FragmentProgram._Current->Base);
+      _mesa_print_program(&p->FragProg.Base);
       printf("\n");
    }
 
@@ -1162,11 +1175,6 @@ translate_program(struct i915_fragment_program *p)
    fixup_depth_write(p);
    i915_fini_program(p);
 
-   if (INTEL_DEBUG & DEBUG_WM) {
-      printf("i915:\n");
-      i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
-   }
-
    p->translated = 1;
 }
 
@@ -1413,6 +1421,10 @@ i915ValidateFragmentProgram(struct i915_context *i915)
                                               intel->vertex_attr_count,
                                               intel->ViewportMatrix.m, 0);
 
+      assert(intel->prim.current_offset == intel->prim.start_offset);
+      intel->prim.start_offset = (intel->prim.current_offset + intel->vertex_size-1) / intel->vertex_size * intel->vertex_size;
+      intel->prim.current_offset = intel->prim.start_offset;
+
       intel->vertex_size >>= 2;
 
       i915->state.Ctx[I915_CTXREG_LIS2] = s2;
@@ -1427,6 +1439,11 @@ i915ValidateFragmentProgram(struct i915_context *i915)
 
    if (!p->on_hardware)
       i915_upload_program(i915, p);
+
+   if (INTEL_DEBUG & DEBUG_WM) {
+      printf("i915:\n");
+      i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
+   }
 }
 
 void