Merge remote branch 'origin/master' into HEAD

15 files changed, 671 insertions, 65 deletions

diff --git a/src/mesa/drivers/dri/i915/i830_texstate.c b/src/mesa/drivers/dri/i915/i830_texstate.c
index c718bb0055d..df43b779a79 100644
--- a/src/mesa/drivers/dri/i915/i830_texstate.c
+++ b/src/mesa/drivers/dri/i915/i830_texstate.c
@@ -38,7 +38,7 @@
 
 
 static GLuint
-translate_texture_format(GLuint mesa_format)
+translate_texture_format(GLuint mesa_format, GLuint internal_format)
 {
    switch (mesa_format) {
    case MESA_FORMAT_L8:
@@ -56,7 +56,10 @@ translate_texture_format(GLuint mesa_format)
    case MESA_FORMAT_ARGB4444:
       return MAPSURF_16BIT | MT_16BIT_ARGB4444;
    case MESA_FORMAT_ARGB8888:
-      return MAPSURF_32BIT | MT_32BIT_ARGB8888;
+      if (internal_format == GL_RGB)
+	 return MAPSURF_32BIT | MT_32BIT_XRGB8888;
+      else
+	 return MAPSURF_32BIT | MT_32BIT_ARGB8888;
    case MESA_FORMAT_YCBCR_REV:
       return (MAPSURF_422 | MT_422_YCRCB_NORMAL);
    case MESA_FORMAT_YCBCR:
@@ -162,7 +165,8 @@ i830_update_tex_unit(struct intel_context *intel, GLuint unit, GLuint ss3)
 								0, intelObj->
 								firstLevel);
 
-      format = translate_texture_format(firstImage->TexFormat->MesaFormat);
+      format = translate_texture_format(firstImage->TexFormat->MesaFormat,
+					firstImage->InternalFormat);
       pitch = intelObj->mt->pitch * intelObj->mt->cpp;
    }
 
diff --git a/src/mesa/drivers/dri/i915/i915_texstate.c b/src/mesa/drivers/dri/i915/i915_texstate.c
index adbb52a3a3b..6d25f8dd8ef 100644
--- a/src/mesa/drivers/dri/i915/i915_texstate.c
+++ b/src/mesa/drivers/dri/i915/i915_texstate.c
@@ -37,7 +37,8 @@
 
 
 static GLuint
-translate_texture_format(GLuint mesa_format, GLenum DepthMode)
+translate_texture_format(GLuint mesa_format, GLuint internal_format,
+			 GLenum DepthMode)
 {
    switch (mesa_format) {
    case MESA_FORMAT_L8:
@@ -55,7 +56,10 @@ translate_texture_format(GLuint mesa_format, GLenum DepthMode)
    case MESA_FORMAT_ARGB4444:
       return MAPSURF_16BIT | MT_16BIT_ARGB4444;
    case MESA_FORMAT_ARGB8888:
-      return MAPSURF_32BIT | MT_32BIT_ARGB8888;
+      if (internal_format == GL_RGB)
+	 return MAPSURF_32BIT | MT_32BIT_XRGB8888;
+      else
+	 return MAPSURF_32BIT | MT_32BIT_ARGB8888;
    case MESA_FORMAT_YCBCR_REV:
       return (MAPSURF_422 | MT_422_YCRCB_NORMAL);
    case MESA_FORMAT_YCBCR:
@@ -173,7 +177,8 @@ i915_update_tex_unit(struct intel_context *intel, GLuint unit, GLuint ss3)
 								 firstLevel);
 
       format = translate_texture_format(firstImage->TexFormat->MesaFormat, 
-		tObj->DepthMode);
+					firstImage->InternalFormat,
+					tObj->DepthMode);
       pitch = intelObj->mt->pitch * intelObj->mt->cpp;
    }
 
diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
index 9b320480b6d..e6113eff87e 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
@@ -69,7 +69,8 @@ static GLuint translate_tex_target( GLenum target )
 }
 
 
-static GLuint translate_tex_format( GLuint mesa_format, GLenum depth_mode )
+static GLuint translate_tex_format( GLuint mesa_format, GLenum internal_format,
+				    GLenum depth_mode )
 {
    switch( mesa_format ) {
    case MESA_FORMAT_L8:
@@ -89,10 +90,16 @@ static GLuint translate_tex_format( GLuint mesa_format, GLenum depth_mode )
       return BRW_SURFACEFORMAT_R8G8B8_UNORM;      
 
    case MESA_FORMAT_ARGB8888:
-      return BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
+      if (internal_format == GL_RGB)
+	 return BRW_SURFACEFORMAT_B8G8R8X8_UNORM;
+      else
+	 return BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
 
    case MESA_FORMAT_RGBA8888_REV:
-      return BRW_SURFACEFORMAT_R8G8B8A8_UNORM;
+      if (internal_format == GL_RGB)
+	 return BRW_SURFACEFORMAT_R8G8B8X8_UNORM;
+      else
+	 return BRW_SURFACEFORMAT_R8G8B8A8_UNORM;
 
    case MESA_FORMAT_RGB565:
       return BRW_SURFACEFORMAT_B5G6R5_UNORM;
@@ -161,7 +168,7 @@ static GLuint translate_tex_format( GLuint mesa_format, GLenum depth_mode )
 struct brw_wm_surface_key {
    GLenum target, depthmode;
    dri_bo *bo;
-   GLint format;
+   GLint format, internal_format;
    GLint first_level, last_level;
    GLint width, height, depth;
    GLint pitch, cpp;
@@ -199,9 +206,11 @@ brw_create_texture_surface( struct brw_context *brw,
 
    surf.ss0.mipmap_layout_mode = BRW_SURFACE_MIPMAPLAYOUT_BELOW;
    surf.ss0.surface_type = translate_tex_target(key->target);
-
-   if (key->bo) 
-      surf.ss0.surface_format = translate_tex_format(key->format, key->depthmode);
+   if (key->bo) {
+      surf.ss0.surface_format = translate_tex_format(key->format,
+						     key->internal_format,
+						     key->depthmode);
+   }
    else {
       switch (key->depth) {
       case 32:
@@ -278,6 +287,7 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit )
       key.offset = intelObj->textureOffset;
    } else {
       key.format = firstImage->TexFormat->MesaFormat;
+      key.internal_format = firstImage->InternalFormat;
       key.pitch = intelObj->mt->pitch;
       key.depth = firstImage->Depth;
       key.bo = intelObj->mt->region->buffer;
diff --git a/src/mesa/drivers/dri/intel/intel_screen.c b/src/mesa/drivers/dri/intel/intel_screen.c
index e8c074712cd..d20ea151877 100644
--- a/src/mesa/drivers/dri/intel/intel_screen.c
+++ b/src/mesa/drivers/dri/intel/intel_screen.c
@@ -211,6 +211,7 @@ static const __DRItexOffsetExtension intelTexOffsetExtension = {
 static const __DRItexBufferExtension intelTexBufferExtension = {
     { __DRI_TEX_BUFFER, __DRI_TEX_BUFFER_VERSION },
    intelSetTexBuffer,
+   intelSetTexBuffer2,
 };
 
 static const __DRIextension *intelScreenExtensions[] = {
diff --git a/src/mesa/drivers/dri/intel/intel_tex.h b/src/mesa/drivers/dri/intel/intel_tex.h
index 742ccc043aa..f5372d82fb2 100644
--- a/src/mesa/drivers/dri/intel/intel_tex.h
+++ b/src/mesa/drivers/dri/intel/intel_tex.h
@@ -149,6 +149,8 @@ void intelSetTexOffset(__DRIcontext *pDRICtx, GLint texname,
 		       unsigned long long offset, GLint depth, GLuint pitch);
 void intelSetTexBuffer(__DRIcontext *pDRICtx,
 		       GLint target, __DRIdrawable *pDraw);
+void intelSetTexBuffer2(__DRIcontext *pDRICtx,
+			GLint target, GLint format, __DRIdrawable *pDraw);
 
 GLuint intel_finalize_mipmap_tree(struct intel_context *intel, GLuint unit);
 
diff --git a/src/mesa/drivers/dri/intel/intel_tex_image.c b/src/mesa/drivers/dri/intel/intel_tex_image.c
index 943636c37b2..e902187637d 100644
--- a/src/mesa/drivers/dri/intel/intel_tex_image.c
+++ b/src/mesa/drivers/dri/intel/intel_tex_image.c
@@ -714,7 +714,9 @@ intelSetTexOffset(__DRIcontext *pDRICtx, GLint texname,
 }
 
 void
-intelSetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv)
+intelSetTexBuffer2(__DRIcontext *pDRICtx, GLint target,
+		   GLint glx_texture_format,
+		   __DRIdrawable *dPriv)
 {
    struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
    struct intel_context *intel = pDRICtx->driverPrivate;
@@ -745,7 +747,10 @@ intelSetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv)
 
    type = GL_BGRA;
    format = GL_UNSIGNED_BYTE;
-   internalFormat = (rb->region->cpp == 3 ? 3 : 4);
+   if (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT)
+      internalFormat = GL_RGB;
+   else
+      internalFormat = GL_RGBA;
 
    mt = intel_miptree_create_for_region(intel, target,
 					internalFormat,
@@ -785,3 +790,12 @@ intelSetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv)
 
    _mesa_unlock_texture(&intel->ctx, texObj);
 }
+
+void
+intelSetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv)
+{
+   /* The old interface didn't have the format argument, so copy our
+    * implementation's behavior at the time.
+    */
+   intelSetTexBuffer2(pDRICtx, target, GLX_TEXTURE_FORMAT_RGBA_EXT, dPriv);
+}
diff --git a/src/mesa/shader/prog_optimize.c b/src/mesa/shader/prog_optimize.c
index ec06da141da..458a69f70b9 100644
--- a/src/mesa/shader/prog_optimize.c
+++ b/src/mesa/shader/prog_optimize.c
@@ -33,6 +33,9 @@
 #include "prog_print.h"
 
 
+#define MAX_LOOP_NESTING 50
+
+
 static GLboolean dbg = GL_FALSE;
 
 
@@ -76,6 +79,37 @@ remove_instructions(struct gl_program *prog, const GLboolean *removeFlags)
 
 
 /**
+ * Remap register indexes according to map.
+ * \param prog  the program to search/replace
+ * \param file  the type of register file to search/replace
+ * \param map  maps old register indexes to new indexes
+ */
+static void
+replace_regs(struct gl_program *prog, gl_register_file file, const GLint map[])
+{
+   GLuint i;
+
+   for (i = 0; i < prog->NumInstructions; i++) {
+      struct prog_instruction *inst = prog->Instructions + i;
+      const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode);
+      GLuint j;
+      for (j = 0; j < numSrc; j++) {
+         if (inst->SrcReg[j].File == file) {
+            GLuint index = inst->SrcReg[j].Index;
+            ASSERT(map[index] >= 0);
+            inst->SrcReg[j].Index = map[index];
+         }
+      }
+      if (inst->DstReg.File == file) {
+         const GLuint index = inst->DstReg.Index;
+         ASSERT(map[index] >= 0);
+         inst->DstReg.Index = map[index];
+      }
+   }
+}
+
+
+/**
  * Consolidate temporary registers to use low numbers.  For example, if the
  * shader only uses temps 4, 5, 8, replace them with 0, 1, 2.
  */
@@ -83,7 +117,7 @@ static void
 _mesa_consolidate_registers(struct gl_program *prog)
 {
    GLboolean tempUsed[MAX_PROGRAM_TEMPS];
-   GLuint tempMap[MAX_PROGRAM_TEMPS];
+   GLint tempMap[MAX_PROGRAM_TEMPS];
    GLuint tempMax = 0, i;
 
    if (dbg) {
@@ -92,6 +126,10 @@ _mesa_consolidate_registers(struct gl_program *prog)
 
    memset(tempUsed, 0, sizeof(tempUsed));
 
+   for (i = 0; i < MAX_PROGRAM_TEMPS; i++) {
+      tempMap[i] = -1;
+   }
+
    /* set tempUsed[i] if temporary [i] is referenced */
    for (i = 0; i < prog->NumInstructions; i++) {
       const struct prog_instruction *inst = prog->Instructions + i;
@@ -132,26 +170,8 @@ _mesa_consolidate_registers(struct gl_program *prog)
       }
    }
 
-   /* now replace occurances of old temp indexes with new indexes */
-   for (i = 0; i < prog->NumInstructions; i++) {
-      struct prog_instruction *inst = prog->Instructions + i;
-      const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode);
-      GLuint j;
-      for (j = 0; j < numSrc; j++) {
-         if (inst->SrcReg[j].File == PROGRAM_TEMPORARY) {
-            GLuint index = inst->SrcReg[j].Index;
-            assert(index <= tempMax);
-            assert(tempUsed[index]);
-            inst->SrcReg[j].Index = tempMap[index];
-         }
-      }
-      if (inst->DstReg.File == PROGRAM_TEMPORARY) {
-         const GLuint index = inst->DstReg.Index;
-         assert(tempUsed[index]);
-         assert(index <= tempMax);
-         inst->DstReg.Index = tempMap[index];
-      }
-   }
+   replace_regs(prog, PROGRAM_TEMPORARY, tempMap);
+
    if (dbg) {
       _mesa_printf("Optimize: End register consolidation\n");
    }
@@ -409,6 +429,370 @@ _mesa_remove_extra_moves(struct gl_program *prog)
 }
 
 
+/** A live register interval */
+struct interval
+{
+   GLuint Reg;         /** The temporary register index */
+   GLuint Start, End;  /** Start/end instruction numbers */
+};
+
+
+/** A list of register intervals */
+struct interval_list
+{
+   GLuint Num;
+   struct interval Intervals[MAX_PROGRAM_TEMPS];
+};
+
+
+static void
+append_interval(struct interval_list *list, const struct interval *inv)
+{
+   list->Intervals[list->Num++] = *inv;
+}
+
+
+/** Insert interval inv into list, sorted by interval end */
+static void
+insert_interval_by_end(struct interval_list *list, const struct interval *inv)
+{
+   /* XXX we could do a binary search insertion here since list is sorted */
+   GLint i = list->Num - 1;
+   while (i >= 0 && list->Intervals[i].End > inv->End) {
+      list->Intervals[i + 1] = list->Intervals[i];
+      i--;
+   }
+   list->Intervals[i + 1] = *inv;
+   list->Num++;
+
+#ifdef DEBUG
+   {
+      GLuint i;
+      for (i = 0; i + 1 < list->Num; i++) {
+         ASSERT(list->Intervals[i].End <= list->Intervals[i + 1].End);
+      }
+   }
+#endif
+}
+
+
+/** Remove the given interval from the interval list */
+static void
+remove_interval(struct interval_list *list, const struct interval *inv)
+{
+   /* XXX we could binary search since list is sorted */
+   GLuint k;
+   for (k = 0; k < list->Num; k++) {
+      if (list->Intervals[k].Reg == inv->Reg) {
+         /* found, remove it */
+         ASSERT(list->Intervals[k].Start == inv->Start);
+         ASSERT(list->Intervals[k].End == inv->End);
+         while (k < list->Num - 1) {
+            list->Intervals[k] = list->Intervals[k + 1];
+            k++;
+         }
+         list->Num--;
+         return;
+      }
+   }
+}
+
+
+/** called by qsort() */
+static int
+compare_start(const void *a, const void *b)
+{
+   const struct interval *ia = (const struct interval *) a;
+   const struct interval *ib = (const struct interval *) b;
+   if (ia->Start < ib->Start)
+      return -1;
+   else if (ia->Start > ib->Start)
+      return +1;
+   else
+      return 0;
+}
+
+/** sort the interval list according to interval starts */
+static void
+sort_interval_list_by_start(struct interval_list *list)
+{
+   qsort(list->Intervals, list->Num, sizeof(struct interval), compare_start);
+#ifdef DEBUG
+   {
+      GLuint i;
+      for (i = 0; i + 1 < list->Num; i++) {
+         ASSERT(list->Intervals[i].Start <= list->Intervals[i + 1].Start);
+      }
+   }
+#endif
+}
+
+
+/**
+ * Update the intermediate interval info for register 'index' and
+ * instruction 'ic'.
+ */
+static void
+update_interval(GLint intBegin[], GLint intEnd[], GLuint index, GLuint ic)
+{
+   ASSERT(index < MAX_PROGRAM_TEMPS);
+   if (intBegin[index] == -1) {
+      ASSERT(intEnd[index] == -1);
+      intBegin[index] = intEnd[index] = ic;
+   }
+   else {
+      intEnd[index] = ic;
+   }
+}
+
+
+/**
+ * Find the live intervals for each temporary register in the program.
+ * For register R, the interval [A,B] indicates that R is referenced
+ * from instruction A through instruction B.
+ * Special consideration is needed for loops and subroutines.
+ * \return GL_TRUE if success, GL_FALSE if we cannot proceed for some reason
+ */
+static GLboolean
+find_live_intervals(struct gl_program *prog,
+                    struct interval_list *liveIntervals)
+{
+   struct loop_info
+   {
+      GLuint Start, End;  /**< Start, end instructions of loop */
+   };
+   struct loop_info loopStack[MAX_LOOP_NESTING];
+   GLuint loopStackDepth = 0;
+   GLint intBegin[MAX_PROGRAM_TEMPS], intEnd[MAX_PROGRAM_TEMPS];
+   GLuint i;
+
+   /*
+    * Note: we'll return GL_FALSE below if we find relative indexing
+    * into the TEMP register file.  We can't handle that yet.
+    * We also give up on subroutines for now.
+    */
+
+   if (dbg) {
+      _mesa_printf("Optimize: Begin find intervals\n");
+   }
+
+   for (i = 0; i < MAX_PROGRAM_TEMPS; i++){
+      intBegin[i] = intEnd[i] = -1;
+   }
+
+   /* Scan instructions looking for temporary registers */
+   for (i = 0; i < prog->NumInstructions; i++) {
+      const struct prog_instruction *inst = prog->Instructions + i;
+      if (inst->Opcode == OPCODE_BGNLOOP) {
+         loopStack[loopStackDepth].Start = i;
+         loopStack[loopStackDepth].End = inst->BranchTarget;
+         loopStackDepth++;
+      }
+      else if (inst->Opcode == OPCODE_ENDLOOP) {
+         loopStackDepth--;
+      }
+      else if (inst->Opcode == OPCODE_CAL) {
+         return GL_FALSE;
+      }
+      else {
+         const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode);
+         GLuint j;
+         for (j = 0; j < numSrc; j++) {
+            if (inst->SrcReg[j].File == PROGRAM_TEMPORARY) {
+               const GLuint index = inst->SrcReg[j].Index;
+               if (inst->SrcReg[j].RelAddr)
+                  return GL_FALSE;
+               update_interval(intBegin, intEnd, index, i);
+               if (loopStackDepth > 0) {
+                  /* extend temp register's interval to end of loop */
+                  GLuint loopEnd = loopStack[loopStackDepth - 1].End;
+                  update_interval(intBegin, intEnd, index, loopEnd);
+               }
+            }
+         }
+         if (inst->DstReg.File == PROGRAM_TEMPORARY) {
+            const GLuint index = inst->DstReg.Index;
+            if (inst->DstReg.RelAddr)
+               return GL_FALSE;
+            update_interval(intBegin, intEnd, index, i);
+            if (loopStackDepth > 0) {
+               /* extend temp register's interval to end of loop */
+               GLuint loopEnd = loopStack[loopStackDepth - 1].End;
+               update_interval(intBegin, intEnd, index, loopEnd);
+            }
+         }
+      }
+   }
+
+   /* Build live intervals list from intermediate arrays */
+   liveIntervals->Num = 0;
+   for (i = 0; i < MAX_PROGRAM_TEMPS; i++) {
+      if (intBegin[i] >= 0) {
+         struct interval inv;
+         inv.Reg = i;
+         inv.Start = intBegin[i];
+         inv.End = intEnd[i];
+         append_interval(liveIntervals, &inv);
+      }
+   }
+
+   /* Sort the list according to interval starts */
+   sort_interval_list_by_start(liveIntervals);
+
+   if (dbg) {
+      /* print interval info */
+      for (i = 0; i < liveIntervals->Num; i++) {
+         const struct interval *inv = liveIntervals->Intervals + i;
+         _mesa_printf("Reg[%d] live [%d, %d]:",
+                      inv->Reg, inv->Start, inv->End);
+         if (1) {
+            int j;
+            for (j = 0; j < inv->Start; j++)
+               _mesa_printf(" ");
+            for (j = inv->Start; j <= inv->End; j++)
+               _mesa_printf("x");
+         }
+         _mesa_printf("\n");
+      }
+   }
+
+   return GL_TRUE;
+}
+
+
+static GLuint
+alloc_register(GLboolean usedRegs[MAX_PROGRAM_TEMPS])
+{
+   GLuint k;
+   for (k = 0; k < MAX_PROGRAM_TEMPS; k++) {
+      if (!usedRegs[k]) {
+         usedRegs[k] = GL_TRUE;
+         return k;
+      }
+   }
+   return MAX_PROGRAM_TEMPS;
+}
+
+
+/**
+ * This function implements "Linear Scan Register Allocation" to reduce
+ * the number of temporary registers used by the program.
+ *
+ * We compute the "live interval" for all temporary registers then
+ * examine the overlap of the intervals to allocate new registers.
+ * Basically, if two intervals do not overlap, they can use the same register.
+ */
+static void
+_mesa_reallocate_registers(struct gl_program *prog)
+{
+   struct interval_list liveIntervals;
+   GLint registerMap[MAX_PROGRAM_TEMPS];
+   GLboolean usedRegs[MAX_PROGRAM_TEMPS];
+   GLuint i;
+   GLuint maxTemp = 0;
+
+   if (dbg) {
+      _mesa_printf("Optimize: Begin live-interval register reallocation\n");
+      _mesa_print_program(prog);
+   }
+
+   for (i = 0; i < MAX_PROGRAM_TEMPS; i++){
+      registerMap[i] = -1;
+      usedRegs[i] = GL_FALSE;
+   }
+
+   if (!find_live_intervals(prog, &liveIntervals)) {
+      if (dbg)
+         _mesa_printf("Aborting register reallocation\n");
+      return;
+   }
+
+   {
+      struct interval_list activeIntervals;
+      activeIntervals.Num = 0;
+
+      /* loop over live intervals, allocating a new register for each */
+      for (i = 0; i < liveIntervals.Num; i++) {
+         const struct interval *live = liveIntervals.Intervals + i;
+
+         if (dbg)
+            _mesa_printf("Consider register %u\n", live->Reg);
+
+         /* Expire old intervals.  Intervals which have ended with respect
+          * to the live interval can have their remapped registers freed.
+          */
+         {
+            GLint j;
+            for (j = 0; j < activeIntervals.Num; j++) {
+               const struct interval *inv = activeIntervals.Intervals + j;
+               if (inv->End >= live->Start) {
+                  /* Stop now.  Since the activeInterval list is sorted
+                   * we know we don't have to go further.
+                   */
+                  break;
+               }
+               else {
+                  /* Interval 'inv' has expired */
+                  const GLint regNew = registerMap[inv->Reg];
+                  ASSERT(regNew >= 0);
+
+                  if (dbg)
+                     _mesa_printf("  expire interval for reg %u\n", inv->Reg);
+
+                  /* remove interval j from active list */
+                  remove_interval(&activeIntervals, inv);
+                  j--;  /* counter-act j++ in for-loop above */
+
+                  /* return register regNew to the free pool */
+                  if (dbg)
+                     _mesa_printf("  free reg %d\n", regNew);
+                  ASSERT(usedRegs[regNew] == GL_TRUE);
+                  usedRegs[regNew] = GL_FALSE;
+               }
+            }
+         }
+
+         /* find a free register for this live interval */
+         {
+            const GLuint k = alloc_register(usedRegs);
+            if (k == MAX_PROGRAM_TEMPS) {
+               /* out of registers, give up */
+               return;
+            }
+            registerMap[live->Reg] = k;
+            maxTemp = MAX2(maxTemp, k);
+            if (dbg)
+               _mesa_printf("  remap register %d -> %d\n", live->Reg, k);
+         }
+
+         /* Insert this live interval into the active list which is sorted
+          * by increasing end points.
+          */
+         insert_interval_by_end(&activeIntervals, live);
+      }
+   }
+
+   if (maxTemp + 1 < liveIntervals.Num) {
+      /* OK, we've reduced the number of registers needed.
+       * Scan the program and replace all the old temporary register
+       * indexes with the new indexes.
+       */
+      replace_regs(prog, PROGRAM_TEMPORARY, registerMap);
+
+      prog->NumTemporaries = maxTemp + 1;
+   }
+
+   if (dbg) {
+      _mesa_printf("Optimize: End live-interval register reallocation\n");
+      _mesa_printf("Num temp regs before: %u  after: %u\n",
+                   liveIntervals.Num, maxTemp + 1);
+      _mesa_print_program(prog);
+   }
+}
+
+
+
+
 /**
  * Apply optimizations to the given program to eliminate unnecessary
  * instructions, temp regs, etc.
@@ -424,4 +808,6 @@ _mesa_optimize_program(GLcontext *ctx, struct gl_program *program)
 
    if (1)
       _mesa_consolidate_registers(program);
+   else /*NEW*/
+      _mesa_reallocate_registers(program);
 }
diff --git a/src/mesa/shader/prog_statevars.c b/src/mesa/shader/prog_statevars.c
index f51d9e26512..aeb7cf6de20 100644
--- a/src/mesa/shader/prog_statevars.c
+++ b/src/mesa/shader/prog_statevars.c
@@ -506,6 +506,13 @@ _mesa_fetch_state(GLcontext *ctx, const gl_state_index state[],
             }
          }
          return;
+      case STATE_FB_SIZE:
+         value[0] = (GLfloat) (ctx->DrawBuffer->Width - 1);
+         value[1] = (GLfloat) (ctx->DrawBuffer->Height - 1);
+         value[2] = 0.0F;
+         value[3] = 0.0F;
+         return;
+
       case STATE_ROT_MATRIX_0:
          {
             const int unit = (int) state[2];
@@ -628,6 +635,9 @@ _mesa_program_state_flags(const gl_state_index state[STATE_LENGTH])
       case STATE_PCM_BIAS:
          return _NEW_PIXEL;
 
+      case STATE_FB_SIZE:
+         return _NEW_BUFFERS;
+
       default:
          /* unknown state indexes are silently ignored and
          *  no flag set, since it is handled by the driver.
@@ -828,6 +838,9 @@ append_token(char *dst, gl_state_index k)
    case STATE_SHADOW_AMBIENT:
       append(dst, "CompareFailValue");
       break;
+   case STATE_FB_SIZE:
+      append(dst, "FbSize");
+      break;
    case STATE_ROT_MATRIX_0:
       append(dst, "rotMatrixRow0");
       break;
diff --git a/src/mesa/shader/prog_statevars.h b/src/mesa/shader/prog_statevars.h
index d563080db1c..1180d9eaa4a 100644
--- a/src/mesa/shader/prog_statevars.h
+++ b/src/mesa/shader/prog_statevars.h
@@ -117,6 +117,7 @@ typedef enum gl_state_index_ {
    STATE_PCM_SCALE,             /**< Post color matrix RGBA scale */
    STATE_PCM_BIAS,              /**< Post color matrix RGBA bias */
    STATE_SHADOW_AMBIENT,        /**< ARB_shadow_ambient fail value; token[2] is texture unit index */
+   STATE_FB_SIZE,               /**< (width-1, height-1, 0, 0) */
    STATE_ROT_MATRIX_0,          /**< ATI_envmap_bumpmap, rot matrix row 0 */
    STATE_ROT_MATRIX_1,          /**< ATI_envmap_bumpmap, rot matrix row 1 */
    STATE_INTERNAL_DRIVER	/* first available state index for drivers (must be last) */
diff --git a/src/mesa/shader/slang/slang_codegen.c b/src/mesa/shader/slang/slang_codegen.c
index 8263aae3343..a7cfc45e6f0 100644
--- a/src/mesa/shader/slang/slang_codegen.c
+++ b/src/mesa/shader/slang/slang_codegen.c
@@ -1441,7 +1441,7 @@ _slang_gen_function_call(slang_assemble_ctx *A, slang_function *fun,
 
    if (A->pragmas->Debug) {
       char s[1000];
-      snprintf(s, sizeof(s), "Call/inline %s()", (char *) fun->header.a_name);
+      _mesa_snprintf(s, sizeof(s), "Call/inline %s()", (char *) fun->header.a_name);
       n->Comment = _slang_strdup(s);
    }
 
diff --git a/src/mesa/state_tracker/st_atom_rasterizer.c b/src/mesa/state_tracker/st_atom_rasterizer.c
index ea76487bcfc..61687fbc3e2 100644
--- a/src/mesa/state_tracker/st_atom_rasterizer.c
+++ b/src/mesa/state_tracker/st_atom_rasterizer.c
@@ -79,8 +79,6 @@ static void update_raster_state( struct st_context *st )
 
    memset(raster, 0, sizeof(*raster));
 
-   raster->origin_lower_left = 1; /* Always true for OpenGL */
-   
    /* _NEW_POLYGON, _NEW_BUFFERS
     */
    {
diff --git a/src/mesa/state_tracker/st_atom_stipple.c b/src/mesa/state_tracker/st_atom_stipple.c
index f395930ab40..31e124b3293 100644
--- a/src/mesa/state_tracker/st_atom_stipple.c
+++ b/src/mesa/state_tracker/st_atom_stipple.c
@@ -39,24 +39,52 @@
 #include "pipe/p_defines.h"
 
 
+/**
+ * OpenGL's polygon stipple is indexed with window coordinates in which
+ * the origin (0,0) is the lower-left corner of the window.
+ * With Gallium, the origin is the upper-left corner of the window.
+ * To convert GL's polygon stipple to what gallium expects we need to
+ * invert the pattern vertically and rotate the stipple rows according
+ * to the window height.
+ */
+static void
+invert_stipple(GLuint dest[32], const GLuint src[32], GLuint winHeight)
+{
+   GLuint i;
+
+   for (i = 0; i < 32; i++) {
+      dest[i] = src[(winHeight - 1 - i) & 0x1f];
+   }
+}
+
+
+
 static void 
 update_stipple( struct st_context *st )
 {
-   const GLuint sz = sizeof(st->state.poly_stipple.stipple);
+   const GLuint sz = sizeof(st->state.poly_stipple);
    assert(sz == sizeof(st->ctx->PolygonStipple));
 
-   if (memcmp(&st->state.poly_stipple.stipple, st->ctx->PolygonStipple, sz)) {
+   if (memcmp(st->state.poly_stipple, st->ctx->PolygonStipple, sz)) {
       /* state has changed */
-      memcpy(st->state.poly_stipple.stipple, st->ctx->PolygonStipple, sz);
-      st->pipe->set_polygon_stipple(st->pipe, &st->state.poly_stipple);
+      struct pipe_poly_stipple newStipple;
+
+      memcpy(st->state.poly_stipple, st->ctx->PolygonStipple, sz);
+
+      invert_stipple(newStipple.stipple, st->ctx->PolygonStipple,
+                     st->ctx->DrawBuffer->Height);
+
+      st->pipe->set_polygon_stipple(st->pipe, &newStipple);
    }
 }
 
 
+/** Update the stipple when the pattern or window height changes */
 const struct st_tracked_state st_update_polygon_stipple = {
    "st_update_polygon_stipple",				/* name */
    {							/* dirty */
-      (_NEW_POLYGONSTIPPLE),				/* mesa */
+      (_NEW_POLYGONSTIPPLE |
+       _NEW_BUFFERS),					/* mesa */
       0,						/* st */
    },
    update_stipple					/* update */
diff --git a/src/mesa/state_tracker/st_cb_texture.c b/src/mesa/state_tracker/st_cb_texture.c
index edfa8854d89..311d812ccfb 100644
--- a/src/mesa/state_tracker/st_cb_texture.c
+++ b/src/mesa/state_tracker/st_cb_texture.c
@@ -467,7 +467,7 @@ st_TexImage(GLcontext * ctx,
     */
    if (stObj->pt) {
       if (stObj->teximage_realloc ||
-          level > stObj->pt->last_level ||
+          level > (GLint) stObj->pt->last_level ||
           (stObj->pt->last_level == level &&
            stObj->pt->target != PIPE_TEXTURE_CUBE &&
            !st_texture_match_image(stObj->pt, &stImage->base,
@@ -803,7 +803,6 @@ st_TexSubimage(GLcontext * ctx,
                                             PIPE_TRANSFER_WRITE,
                                             xoffset, yoffset,
                                             width, height);
-      dstRowStride = stImage->transfer->stride;
    }
 
    if (!texImage->Data) {
@@ -812,6 +811,7 @@ st_TexSubimage(GLcontext * ctx,
    }
 
    src = (const GLubyte *) pixels;
+   dstRowStride = stImage->transfer->stride;
 
    for (i = 0; i++ < depth;) {
       if (!texImage->TexFormat->StoreImage(ctx, dims, texImage->_BaseFormat,
diff --git a/src/mesa/state_tracker/st_context.h b/src/mesa/state_tracker/st_context.h
index d7518ab6897..ae8c2978bf8 100644
--- a/src/mesa/state_tracker/st_context.h
+++ b/src/mesa/state_tracker/st_context.h
@@ -93,12 +93,13 @@ struct st_context
       struct pipe_constant_buffer constants[2];
       struct pipe_framebuffer_state framebuffer;
       struct pipe_texture *sampler_texture[PIPE_MAX_SAMPLERS];
-      struct pipe_poly_stipple poly_stipple;
       struct pipe_scissor_state scissor;
       struct pipe_viewport_state viewport;
 
       GLuint num_samplers;
       GLuint num_textures;
+
+      GLuint poly_stipple[32];  /**< In OpenGL's bottom-to-top order */
    } state;
 
    struct {
diff --git a/src/mesa/state_tracker/st_mesa_to_tgsi.c b/src/mesa/state_tracker/st_mesa_to_tgsi.c
index cbf3f334c08..ffa607dd87c 100644
--- a/src/mesa/state_tracker/st_mesa_to_tgsi.c
+++ b/src/mesa/state_tracker/st_mesa_to_tgsi.c
@@ -219,8 +219,9 @@ compile_instruction(
    const GLuint immediateMapping[],
    GLboolean indirectAccess,
    GLuint preamble_size,
-   GLuint processor,
-   GLboolean *insideSubroutine)
+   GLuint procType,
+   GLboolean *insideSubroutine,
+   GLint wposTemp)
 {
    GLuint i;
    struct tgsi_full_dst_register *fulldst;
@@ -247,19 +248,29 @@ compile_instruction(
       GLuint j;
 
       fullsrc = &fullinst->FullSrcRegisters[i];
-      fullsrc->SrcRegister.File = map_register_file(
-         inst->SrcReg[i].File,
-         inst->SrcReg[i].Index,
-         immediateMapping,
-         indirectAccess );
-      fullsrc->SrcRegister.Index = map_register_file_index(
-         fullsrc->SrcRegister.File,
-         inst->SrcReg[i].Index,
-         inputMapping,
-         outputMapping,
-         immediateMapping,
-         indirectAccess );
 
+      if (procType == TGSI_PROCESSOR_FRAGMENT &&
+          inst->SrcReg[i].File == PROGRAM_INPUT &&
+          inst->SrcReg[i].Index == FRAG_ATTRIB_WPOS) {
+         /* special case of INPUT[WPOS] */
+         fullsrc->SrcRegister.File = TGSI_FILE_TEMPORARY;
+         fullsrc->SrcRegister.Index = wposTemp;
+      }
+      else {
+         /* any other src register */
+         fullsrc->SrcRegister.File = map_register_file(
+            inst->SrcReg[i].File,
+            inst->SrcReg[i].Index,
+            immediateMapping,
+            indirectAccess );
+         fullsrc->SrcRegister.Index = map_register_file_index(
+            fullsrc->SrcRegister.File,
+            inst->SrcReg[i].Index,
+            inputMapping,
+            outputMapping,
+            immediateMapping,
+            indirectAccess );
+      }
 
       /* swizzle (ext swizzle also depends on negation) */
       {
@@ -733,6 +744,111 @@ find_temporaries(const struct gl_program *program,
 }
 
 
+/**
+ * Find an unused temporary in the tempsUsed array.
+ */
+static int
+find_free_temporary(GLboolean tempsUsed[MAX_PROGRAM_TEMPS])
+{
+   int i;
+   for (i = 0; i < MAX_PROGRAM_TEMPS; i++) {
+      if (!tempsUsed[i]) {
+         tempsUsed[i] = GL_TRUE;
+         return i;
+      }
+   }
+   return -1;
+}
+
+
+/** helper for building simple TGSI instruction, one src register */
+static void
+build_tgsi_instruction1(struct tgsi_full_instruction *inst,
+                        int opcode,
+                        int dstFile, int dstIndex, int writemask,
+                        int srcFile1, int srcIndex1)
+{
+   *inst = tgsi_default_full_instruction();
+
+   inst->Instruction.Opcode = opcode;
+
+   inst->Instruction.NumDstRegs = 1;
+   inst->FullDstRegisters[0].DstRegister.File = dstFile;
+   inst->FullDstRegisters[0].DstRegister.Index = dstIndex;
+   inst->FullDstRegisters[0].DstRegister.WriteMask = writemask;
+
+   inst->Instruction.NumSrcRegs = 1;
+   inst->FullSrcRegisters[0].SrcRegister.File = srcFile1;
+   inst->FullSrcRegisters[0].SrcRegister.Index = srcIndex1;
+}
+
+
+/** helper for building simple TGSI instruction, two src registers */
+static void
+build_tgsi_instruction2(struct tgsi_full_instruction *inst,
+                        int opcode,
+                        int dstFile, int dstIndex, int writemask,
+                        int srcFile1, int srcIndex1,
+                        int srcFile2, int srcIndex2)
+{
+   *inst = tgsi_default_full_instruction();
+
+   inst->Instruction.Opcode = opcode;
+
+   inst->Instruction.NumDstRegs = 1;
+   inst->FullDstRegisters[0].DstRegister.File = dstFile;
+   inst->FullDstRegisters[0].DstRegister.Index = dstIndex;
+   inst->FullDstRegisters[0].DstRegister.WriteMask = writemask;
+
+   inst->Instruction.NumSrcRegs = 2;
+   inst->FullSrcRegisters[0].SrcRegister.File = srcFile1;
+   inst->FullSrcRegisters[0].SrcRegister.Index = srcIndex1;
+   inst->FullSrcRegisters[1].SrcRegister.File = srcFile2;
+   inst->FullSrcRegisters[1].SrcRegister.Index = srcIndex2;
+}
+
+
+
+/**
+ * Emit the TGSI instructions for inverting the WPOS y coordinate.
+ */
+static int
+emit_inverted_wpos(struct tgsi_token *tokens,
+                   int wpos_temp,
+                   int winsize_const,
+                   int wpos_input,
+                   struct tgsi_header *header, int maxTokens)
+{
+   struct tgsi_full_instruction fullinst;
+   int ti = 0;
+
+   /* MOV wpos_temp.xzw, input[wpos]; */
+   build_tgsi_instruction1(&fullinst,
+                           TGSI_OPCODE_MOV,
+                           TGSI_FILE_TEMPORARY, wpos_temp, WRITEMASK_XZW,
+                           TGSI_FILE_INPUT, 0);
+
+   ti += tgsi_build_full_instruction(&fullinst,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti);
+
+   /* SUB wpos_temp.y, const[winsize_const] - input[wpos_input]; */
+   build_tgsi_instruction2(&fullinst,
+                           TGSI_OPCODE_SUB,
+                           TGSI_FILE_TEMPORARY, wpos_temp, WRITEMASK_Y,
+                           TGSI_FILE_CONSTANT, winsize_const,
+                           TGSI_FILE_INPUT, wpos_input);
+
+   ti += tgsi_build_full_instruction(&fullinst,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti);
+
+   return ti;
+}
+
+
 
 
 /**
@@ -778,16 +894,34 @@ st_translate_mesa_program(
    GLuint ti;  /* token index */
    struct tgsi_header *header;
    struct tgsi_processor *processor;
-   struct tgsi_full_instruction fullinst;
    GLuint preamble_size = 0;
    GLuint immediates[1000];
    GLuint numImmediates = 0;
    GLboolean insideSubroutine = GL_FALSE;
    GLboolean indirectAccess = GL_FALSE;
+   GLboolean tempsUsed[MAX_PROGRAM_TEMPS + 1];
+   GLint wposTemp = -1, winHeightConst = -1;
 
    assert(procType == TGSI_PROCESSOR_FRAGMENT ||
           procType == TGSI_PROCESSOR_VERTEX);
 
+   find_temporaries(program, tempsUsed);
+
+   if (procType == TGSI_PROCESSOR_FRAGMENT) {
+      if (program->InputsRead & FRAG_BIT_WPOS) {
+         /* Fragment program uses fragment position input.
+          * Need to replace instances of INPUT[WPOS] with temp T
+          * where T = INPUT[WPOS] by y is inverted.
+          */
+         static const gl_state_index winSizeState[STATE_LENGTH]
+            = { STATE_INTERNAL, STATE_FB_SIZE, 0, 0, 0 };
+         winHeightConst = _mesa_add_state_reference(program->Parameters,
+                                                    winSizeState);
+         wposTemp = find_free_temporary(tempsUsed);
+      }
+   }
+
+
    *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
 
    header = (struct tgsi_header *) &tokens[1];
@@ -884,11 +1018,9 @@ st_translate_mesa_program(
 
    /* temporary decls */
    {
-      GLboolean tempsUsed[MAX_PROGRAM_TEMPS + 1];
       GLboolean inside_range = GL_FALSE;
       GLuint start_range = 0;
 
-      find_temporaries(program, tempsUsed);
       tempsUsed[MAX_PROGRAM_TEMPS] = GL_FALSE;
       for (i = 0; i < MAX_PROGRAM_TEMPS + 1; i++) {
          if (tempsUsed[i] && !inside_range) {
@@ -1018,7 +1150,17 @@ st_translate_mesa_program(
       }
    }
 
+   /* invert WPOS fragment input */
+   if (wposTemp >= 0) {
+      ti += emit_inverted_wpos(&tokens[ti], wposTemp, winHeightConst,
+                               inputMapping[FRAG_ATTRIB_WPOS],
+                               header, maxTokens - ti);
+      preamble_size = 2; /* two instructions added */
+   }
+
    for (i = 0; i < program->NumInstructions; i++) {
+      struct tgsi_full_instruction fullinst;
+
       compile_instruction(
          &program->Instructions[i],
          &fullinst,
@@ -1028,7 +1170,8 @@ st_translate_mesa_program(
          indirectAccess,
          preamble_size,
          procType,
-         &insideSubroutine );
+         &insideSubroutine,
+         wposTemp);
 
       ti += tgsi_build_full_instruction(
          &fullinst,