Merge remote branch 'origin/master' into lp-binning

Conflicts:
	Makefile
	src/gallium/auxiliary/util/u_surface.c
	src/gallium/drivers/llvmpipe/lp_flush.c
	src/gallium/drivers/llvmpipe/lp_setup.c
	src/gallium/drivers/llvmpipe/lp_state_derived.c
	src/gallium/drivers/llvmpipe/lp_state_fs.c
	src/gallium/drivers/llvmpipe/lp_state_surface.c
	src/gallium/drivers/llvmpipe/lp_tex_cache.c
	src/gallium/drivers/llvmpipe/lp_texture.c
	src/gallium/drivers/llvmpipe/lp_tile_cache.c
	src/mesa/state_tracker/st_cb_condrender.c

13 files changed, 589 insertions, 273 deletions

diff --git a/src/gallium/auxiliary/tgsi/tgsi_build.c b/src/gallium/auxiliary/tgsi/tgsi_build.c
index de9cbc86305..e38b0be7ab5 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_build.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_build.c
@@ -103,6 +103,7 @@ tgsi_default_declaration( void )
    declaration.File = TGSI_FILE_NULL;
    declaration.UsageMask = TGSI_WRITEMASK_XYZW;
    declaration.Interpolate = TGSI_INTERPOLATE_CONSTANT;
+   declaration.Dimension = 0;
    declaration.Semantic = 0;
    declaration.Centroid = 0;
    declaration.Invariant = 0;
@@ -116,6 +117,7 @@ tgsi_build_declaration(
    unsigned file,
    unsigned usage_mask,
    unsigned interpolate,
+   unsigned dimension,
    unsigned semantic,
    unsigned centroid,
    unsigned invariant,
@@ -130,6 +132,7 @@ tgsi_build_declaration(
    declaration.File = file;
    declaration.UsageMask = usage_mask;
    declaration.Interpolate = interpolate;
+   declaration.Dimension = dimension;
    declaration.Semantic = semantic;
    declaration.Centroid = centroid;
    declaration.Invariant = invariant;
@@ -183,6 +186,7 @@ tgsi_build_full_declaration(
       full_decl->Declaration.File,
       full_decl->Declaration.UsageMask,
       full_decl->Declaration.Interpolate,
+      full_decl->Declaration.Dimension,
       full_decl->Declaration.Semantic,
       full_decl->Declaration.Centroid,
       full_decl->Declaration.Invariant,
@@ -199,6 +203,20 @@ tgsi_build_full_declaration(
       declaration,
       header );
 
+   if (full_decl->Declaration.Dimension) {
+      struct tgsi_declaration_dimension *dd;
+
+      if (maxsize <= size) {
+         return 0;
+      }
+      dd = (struct tgsi_declaration_dimension *)&tokens[size];
+      size++;
+
+      *dd = tgsi_build_declaration_dimension(full_decl->Dim.Index2D,
+                                             declaration,
+                                             header);
+   }
+
    if( full_decl->Declaration.Semantic ) {
       struct tgsi_declaration_semantic *ds;
 
@@ -249,6 +267,34 @@ tgsi_build_declaration_range(
    return declaration_range;
 }
 
+struct tgsi_declaration_dimension
+tgsi_default_declaration_dimension(void)
+{
+   struct tgsi_declaration_dimension dd;
+
+   dd.Index2D = 0;
+   dd.Padding = 0;
+
+   return dd;
+}
+
+struct tgsi_declaration_dimension
+tgsi_build_declaration_dimension(unsigned index_2d,
+                                 struct tgsi_declaration *declaration,
+                                 struct tgsi_header *header)
+{
+   struct tgsi_declaration_dimension dd;
+
+   assert(index_2d <= 0xFFFF);
+
+   dd = tgsi_default_declaration_dimension();
+   dd.Index2D = index_2d;
+
+   declaration_grow(declaration, header);
+
+   return dd;
+}
+
 struct tgsi_declaration_semantic
 tgsi_default_declaration_semantic( void )
 {
diff --git a/src/gallium/auxiliary/tgsi/tgsi_build.h b/src/gallium/auxiliary/tgsi/tgsi_build.h
index 9de2757fe40..ebee4ce5f6a 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_build.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_build.h
@@ -64,6 +64,7 @@ tgsi_build_declaration(
    unsigned file,
    unsigned usage_mask,
    unsigned interpolate,
+   unsigned dimension,
    unsigned semantic,
    unsigned centroid,
    unsigned invariant,
@@ -89,6 +90,14 @@ tgsi_build_declaration_range(
    struct tgsi_declaration *declaration,
    struct tgsi_header *header );
 
+struct tgsi_declaration_dimension
+tgsi_default_declaration_dimension(void);
+
+struct tgsi_declaration_dimension
+tgsi_build_declaration_dimension(unsigned index_2d,
+                                 struct tgsi_declaration *declaration,
+                                 struct tgsi_header *header);
+
 struct tgsi_declaration_semantic
 tgsi_default_declaration_semantic( void );
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c b/src/gallium/auxiliary/tgsi/tgsi_dump.c
index d7ff262f30a..dd365558755 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_dump.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c
@@ -159,7 +159,9 @@ static const char *property_names[] =
 {
    "GS_INPUT_PRIMITIVE",
    "GS_OUTPUT_PRIMITIVE",
-   "GS_MAX_OUTPUT_VERTICES"
+   "GS_MAX_OUTPUT_VERTICES",
+   "FS_COORD_ORIGIN",
+   "FS_COORD_PIXEL_CENTER"
 };
 
 static const char *primitive_names[] =
@@ -176,29 +178,18 @@ static const char *primitive_names[] =
    "POLYGON"
 };
 
-
-static void
-_dump_register_decl(
-   struct dump_ctx *ctx,
-   uint file,
-   int first,
-   int last )
+static const char *fs_coord_origin_names[] =
 {
-   ENM( file, file_names );
+   "UPPER_LEFT",
+   "LOWER_LEFT"
+};
 
-   /* all geometry shader inputs are two dimensional */
-   if (file == TGSI_FILE_INPUT &&
-       ctx->iter.processor.Processor == TGSI_PROCESSOR_GEOMETRY)
-      TXT("[]");
+static const char *fs_coord_pixel_center_names[] =
+{
+   "HALF_INTEGER",
+   "INTEGER"
+};
 
-   CHR( '[' );
-   SID( first );
-   if (first != last) {
-      TXT( ".." );
-      SID( last );
-   }
-   CHR( ']' );
-}
 
 static void
 _dump_register_dst(
@@ -219,8 +210,13 @@ _dump_register_src(
    struct dump_ctx *ctx,
    const struct tgsi_full_src_register *src )
 {
+   ENM(src->Register.File, file_names);
+   if (src->Register.Dimension) {
+      CHR('[');
+      SID(src->Dimension.Index);
+      CHR(']');
+   }
    if (src->Register.Indirect) {
-      ENM( src->Register.File, file_names );
       CHR( '[' );
       ENM( src->Indirect.File, file_names );
       CHR( '[' );
@@ -234,16 +230,10 @@ _dump_register_src(
       }
       CHR( ']' );
    } else {
-      ENM( src->Register.File, file_names );
       CHR( '[' );
       SID( src->Register.Index );
       CHR( ']' );
    }
-   if (src->Register.Dimension) {
-      CHR( '[' );
-      SID( src->Dimension.Index );
-      CHR( ']' );
-   }
 }
 
 static void
@@ -300,11 +290,28 @@ iter_declaration(
 
    TXT( "DCL " );
 
-   _dump_register_decl(
-      ctx,
-      decl->Declaration.File,
-      decl->Range.First,
-      decl->Range.Last );
+   ENM(decl->Declaration.File, file_names);
+
+   /* all geometry shader inputs are two dimensional */
+   if (decl->Declaration.File == TGSI_FILE_INPUT &&
+       iter->processor.Processor == TGSI_PROCESSOR_GEOMETRY) {
+      TXT("[]");
+   }
+
+   if (decl->Declaration.Dimension) {
+      CHR('[');
+      SID(decl->Dim.Index2D);
+      CHR(']');
+   }
+
+   CHR('[');
+   SID(decl->Range.First);
+   if (decl->Range.First != decl->Range.Last) {
+      TXT("..");
+      SID(decl->Range.Last);
+   }
+   CHR(']');
+
    _dump_writemask(
       ctx,
       decl->Declaration.UsageMask );
@@ -373,6 +380,12 @@ iter_property(
       case TGSI_PROPERTY_GS_OUTPUT_PRIM:
          ENM(prop->u[i].Data, primitive_names);
          break;
+      case TGSI_PROPERTY_FS_COORD_ORIGIN:
+         ENM(prop->u[i].Data, fs_coord_origin_names);
+         break;
+      case TGSI_PROPERTY_FS_COORD_PIXEL_CENTER:
+         ENM(prop->u[i].Data, fs_coord_pixel_center_names);
+         break;
       default:
          SID( prop->u[i].Data );
          break;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 118a638ab48..fbb9aa0e63a 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -953,107 +953,90 @@ micro_sub(
 }
 
 static void
-fetch_src_file_channel(
-   const struct tgsi_exec_machine *mach,
-   const uint file,
-   const uint swizzle,
-   const union tgsi_exec_channel *index,
-   union tgsi_exec_channel *chan )
-{
-   switch( swizzle ) {
-   case TGSI_SWIZZLE_X:
-   case TGSI_SWIZZLE_Y:
-   case TGSI_SWIZZLE_Z:
-   case TGSI_SWIZZLE_W:
-      switch( file ) {
-      case TGSI_FILE_CONSTANT:
-         assert(mach->Consts);
-         if (index->i[0] < 0)
-            chan->f[0] = 0.0f;
-         else
-            chan->f[0] = mach->Consts[index->i[0]][swizzle];
-         if (index->i[1] < 0)
-            chan->f[1] = 0.0f;
-         else
-            chan->f[1] = mach->Consts[index->i[1]][swizzle];
-         if (index->i[2] < 0)
-            chan->f[2] = 0.0f;
-         else
-            chan->f[2] = mach->Consts[index->i[2]][swizzle];
-         if (index->i[3] < 0)
-            chan->f[3] = 0.0f;
-         else
-            chan->f[3] = mach->Consts[index->i[3]][swizzle];
-         break;
+fetch_src_file_channel(const struct tgsi_exec_machine *mach,
+                       const uint file,
+                       const uint swizzle,
+                       const union tgsi_exec_channel *index,
+                       const union tgsi_exec_channel *index2D,
+                       union tgsi_exec_channel *chan)
+{
+   uint i;
 
-      case TGSI_FILE_INPUT:
-      case TGSI_FILE_SYSTEM_VALUE:
-         chan->u[0] = mach->Inputs[index->i[0]].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Inputs[index->i[1]].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Inputs[index->i[2]].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Inputs[index->i[3]].xyzw[swizzle].u[3];
-         break;
+   switch (file) {
+   case TGSI_FILE_CONSTANT:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index2D->i[i] >= 0 && index2D->i[i] < PIPE_MAX_CONSTANT_BUFFERS);
+         assert(mach->Consts[index2D->i[i]]);
 
-      case TGSI_FILE_TEMPORARY:
-         assert(index->i[0] < TGSI_EXEC_NUM_TEMPS);
-         chan->u[0] = mach->Temps[index->i[0]].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Temps[index->i[1]].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Temps[index->i[2]].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Temps[index->i[3]].xyzw[swizzle].u[3];
-         break;
+         if (index->i[i] < 0) {
+            chan->u[i] = 0;
+         } else {
+            const uint *p = (const uint *)mach->Consts[index2D->i[i]];
 
-      case TGSI_FILE_IMMEDIATE:
-         assert( index->i[0] < (int) mach->ImmLimit );
-         chan->f[0] = mach->Imms[index->i[0]][swizzle];
-         assert( index->i[1] < (int) mach->ImmLimit );
-         chan->f[1] = mach->Imms[index->i[1]][swizzle];
-         assert( index->i[2] < (int) mach->ImmLimit );
-         chan->f[2] = mach->Imms[index->i[2]][swizzle];
-         assert( index->i[3] < (int) mach->ImmLimit );
-         chan->f[3] = mach->Imms[index->i[3]][swizzle];
-         break;
+            chan->u[i] = p[index->i[i] * 4 + swizzle];
+         }
+      }
+      break;
 
-      case TGSI_FILE_ADDRESS:
-         chan->u[0] = mach->Addrs[index->i[0]].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Addrs[index->i[1]].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Addrs[index->i[2]].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Addrs[index->i[3]].xyzw[swizzle].u[3];
-         break;
+   case TGSI_FILE_INPUT:
+   case TGSI_FILE_SYSTEM_VALUE:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         /* XXX: 2D indexing */
+         chan->u[i] = mach->Inputs[index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i]].xyzw[swizzle].u[i];
+      }
+      break;
 
-      case TGSI_FILE_PREDICATE:
-         assert(index->i[0] < TGSI_EXEC_NUM_PREDS);
-         assert(index->i[1] < TGSI_EXEC_NUM_PREDS);
-         assert(index->i[2] < TGSI_EXEC_NUM_PREDS);
-         assert(index->i[3] < TGSI_EXEC_NUM_PREDS);
-         chan->u[0] = mach->Predicates[0].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Predicates[0].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Predicates[0].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Predicates[0].xyzw[swizzle].u[3];
-         break;
+   case TGSI_FILE_TEMPORARY:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] < TGSI_EXEC_NUM_TEMPS);
+         assert(index2D->i[i] == 0);
 
-      case TGSI_FILE_OUTPUT:
-         /* vertex/fragment output vars can be read too */
-         chan->u[0] = mach->Outputs[index->i[0]].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Outputs[index->i[1]].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Outputs[index->i[2]].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Outputs[index->i[3]].xyzw[swizzle].u[3];
-         break;
+         chan->u[i] = mach->Temps[index->i[i]].xyzw[swizzle].u[i];
+      }
+      break;
 
-      default:
-         assert( 0 );
-         chan->u[0] = 0;
-         chan->u[1] = 0;
-         chan->u[2] = 0;
-         chan->u[3] = 0;
+   case TGSI_FILE_IMMEDIATE:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] >= 0 && index->i[i] < (int)mach->ImmLimit);
+         assert(index2D->i[i] == 0);
+
+         chan->f[i] = mach->Imms[index->i[i]][swizzle];
+      }
+      break;
+
+   case TGSI_FILE_ADDRESS:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] >= 0);
+         assert(index2D->i[i] == 0);
+
+         chan->u[i] = mach->Addrs[index->i[i]].xyzw[swizzle].u[i];
+      }
+      break;
+
+   case TGSI_FILE_PREDICATE:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] >= 0 && index->i[i] < TGSI_EXEC_NUM_PREDS);
+         assert(index2D->i[i] == 0);
+
+         chan->u[i] = mach->Predicates[0].xyzw[swizzle].u[i];
+      }
+      break;
+
+   case TGSI_FILE_OUTPUT:
+      /* vertex/fragment output vars can be read too */
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] >= 0);
+         assert(index2D->i[i] == 0);
+
+         chan->u[i] = mach->Outputs[index->i[i]].xyzw[swizzle].u[i];
       }
       break;
 
    default:
-      assert( 0 );
-      chan->u[0] = 0;
-      chan->u[1] = 0;
-      chan->u[2] = 0;
-      chan->u[3] = 0;
+      assert(0);
+      for (i = 0; i < QUAD_SIZE; i++) {
+         chan->u[i] = 0;
+      }
    }
 }
 
@@ -1065,6 +1048,7 @@ fetch_source(const struct tgsi_exec_machine *mach,
              enum tgsi_exec_datatype src_datatype)
 {
    union tgsi_exec_channel index;
+   union tgsi_exec_channel index2D;
    uint swizzle;
 
    /* We start with a direct index into a register file.
@@ -1103,12 +1087,12 @@ fetch_source(const struct tgsi_exec_machine *mach,
 
       /* get current value of address register[swizzle] */
       swizzle = tgsi_util_get_src_register_swizzle( &reg->Indirect, CHAN_X );
-      fetch_src_file_channel(
-         mach,
-         reg->Indirect.File,
-         swizzle,
-         &index2,
-         &indir_index );
+      fetch_src_file_channel(mach,
+                             reg->Indirect.File,
+                             swizzle,
+                             &index2,
+                             &ZeroVec,
+                             &indir_index);
 
       /* add value of address register to the offset */
       index.i[0] += indir_index.i[0];
@@ -1129,44 +1113,22 @@ fetch_source(const struct tgsi_exec_machine *mach,
     * subscript to a register file. Effectively it means that
     * the register file is actually a 2D array of registers.
     *
-    *    file[1][3] == file[1*sizeof(file[1])+3],
+    *    file[3][1],
     *    where:
     *       [3] = Dimension.Index
     */
    if (reg->Register.Dimension) {
-      /* The size of the first-order array depends on the register file type.
-       * We need to multiply the index to the first array to get an effective,
-       * "flat" index that points to the beginning of the second-order array.
-       */
-      switch (reg->Register.File) {
-      case TGSI_FILE_INPUT:
-      case TGSI_FILE_SYSTEM_VALUE:
-         index.i[0] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
-         index.i[1] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
-         index.i[2] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
-         index.i[3] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
-         break;
-      case TGSI_FILE_CONSTANT:
-         index.i[0] *= TGSI_EXEC_MAX_CONST_BUFFER;
-         index.i[1] *= TGSI_EXEC_MAX_CONST_BUFFER;
-         index.i[2] *= TGSI_EXEC_MAX_CONST_BUFFER;
-         index.i[3] *= TGSI_EXEC_MAX_CONST_BUFFER;
-         break;
-      default:
-         assert( 0 );
-      }
-
-      index.i[0] += reg->Dimension.Index;
-      index.i[1] += reg->Dimension.Index;
-      index.i[2] += reg->Dimension.Index;
-      index.i[3] += reg->Dimension.Index;
+      index2D.i[0] =
+      index2D.i[1] =
+      index2D.i[2] =
+      index2D.i[3] = reg->Dimension.Index;
 
       /* Again, the second subscript index can be addressed indirectly
        * identically to the first one.
        * Nothing stops us from indirectly addressing the indirect register,
        * but there is no need for that, so we won't exercise it.
        *
-       *    file[1][ind[4].y+3],
+       *    file[ind[4].y+3][1],
        *    where:
        *       ind = DimIndirect.File
        *       [4] = DimIndirect.Index
@@ -1184,24 +1146,25 @@ fetch_source(const struct tgsi_exec_machine *mach,
          index2.i[3] = reg->DimIndirect.Index;
 
          swizzle = tgsi_util_get_src_register_swizzle( &reg->DimIndirect, CHAN_X );
-         fetch_src_file_channel(
-            mach,
-            reg->DimIndirect.File,
-            swizzle,
-            &index2,
-            &indir_index );
-
-         index.i[0] += indir_index.i[0];
-         index.i[1] += indir_index.i[1];
-         index.i[2] += indir_index.i[2];
-         index.i[3] += indir_index.i[3];
+         fetch_src_file_channel(mach,
+                                reg->DimIndirect.File,
+                                swizzle,
+                                &index2,
+                                &ZeroVec,
+                                &indir_index);
+
+         index2D.i[0] += indir_index.i[0];
+         index2D.i[1] += indir_index.i[1];
+         index2D.i[2] += indir_index.i[2];
+         index2D.i[3] += indir_index.i[3];
 
          /* for disabled execution channels, zero-out the index to
           * avoid using a potential garbage value.
           */
          for (i = 0; i < QUAD_SIZE; i++) {
-            if ((execmask & (1 << i)) == 0)
-               index.i[i] = 0;
+            if ((execmask & (1 << i)) == 0) {
+               index2D.i[i] = 0;
+            }
          }
       }
 
@@ -1209,15 +1172,20 @@ fetch_source(const struct tgsi_exec_machine *mach,
        * files, we would have to check whether Dimension is followed
        * by a dimension register and continue the saga.
        */
+   } else {
+      index2D.i[0] =
+      index2D.i[1] =
+      index2D.i[2] =
+      index2D.i[3] = 0;
    }
 
    swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index );
-   fetch_src_file_channel(
-      mach,
-      reg->Register.File,
-      swizzle,
-      &index,
-      chan );
+   fetch_src_file_channel(mach,
+                          reg->Register.File,
+                          swizzle,
+                          &index,
+                          &index2D,
+                          chan);
 
    if (reg->Register.Absolute) {
       if (src_datatype == TGSI_EXEC_DATA_FLOAT) {
@@ -1280,12 +1248,12 @@ store_dest(struct tgsi_exec_machine *mach,
       swizzle = tgsi_util_get_src_register_swizzle( &reg->Indirect, CHAN_X );
 
       /* fetch values from the address/indirection register */
-      fetch_src_file_channel(
-         mach,
-         reg->Indirect.File,
-         swizzle,
-         &index,
-         &indir_index );
+      fetch_src_file_channel(mach,
+                             reg->Indirect.File,
+                             swizzle,
+                             &index,
+                             &ZeroVec,
+                             &indir_index);
 
       /* save indirection offset */
       offset = indir_index.i[0];
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index 59e3b445cc3..a22873e4c2b 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -260,7 +260,7 @@ struct tgsi_exec_machine
    struct tgsi_sampler           **Samplers;
 
    unsigned                      ImmLimit;
-   const float                   (*Consts)[4];
+   const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
    const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
    unsigned                      Processor; /**< TGSI_PROCESSOR_x */
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.c b/src/gallium/auxiliary/tgsi/tgsi_parse.c
index 8c7062d850c..fd37fc3079b 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_parse.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_parse.c
@@ -109,6 +109,10 @@ tgsi_parse_token(
 
       next_token( ctx, &decl->Range );
 
+      if (decl->Declaration.Dimension) {
+         next_token(ctx, &decl->Dim);
+      }
+
       if( decl->Declaration.Semantic ) {
          next_token( ctx, &decl->Semantic );
       }
diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.h b/src/gallium/auxiliary/tgsi/tgsi_parse.h
index 439a57269b7..8150e3cd29d 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_parse.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_parse.h
@@ -58,6 +58,7 @@ struct tgsi_full_declaration
 {
    struct tgsi_declaration Declaration;
    struct tgsi_declaration_range Range;
+   struct tgsi_declaration_dimension Dim;
    struct tgsi_declaration_semantic Semantic;
 };
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.c b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
index e1e4f97967d..91e1b27da12 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sanity.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
@@ -335,10 +335,6 @@ iter_instruction(
          fill_scan_register1d(ind_reg,
                               inst->Src[i].Indirect.File,
                               inst->Src[i].Indirect.Index);
-         if (!(ind_reg->file == TGSI_FILE_ADDRESS || ind_reg->file == TGSI_FILE_LOOP) ||
-             ind_reg->indices[0] != 0) {
-            report_warning(ctx, "Indirect register neither ADDR[0] nor LOOP[0]");
-         }
          check_register_usage(
             ctx,
             ind_reg,
@@ -412,12 +408,16 @@ iter_declaration(
          uint vert;
          for (vert = 0; vert < ctx->implied_array_size; ++vert) {
             scan_register *reg = MALLOC(sizeof(scan_register));
-            fill_scan_register2d(reg, file, vert, i);
+            fill_scan_register2d(reg, file, i, vert);
             check_and_declare(ctx, reg);
          }
       } else {
          scan_register *reg = MALLOC(sizeof(scan_register));
-         fill_scan_register1d(reg, file, i);
+         if (decl->Declaration.Dimension) {
+            fill_scan_register2d(reg, file, i, decl->Dim.Index2D);
+         } else {
+            fill_scan_register1d(reg, file, i);
+         }
          check_and_declare(ctx, reg);
       }
    }
diff --git a/src/gallium/auxiliary/tgsi/tgsi_scan.c b/src/gallium/auxiliary/tgsi/tgsi_scan.c
index a6cc773003a..b9be8dc0a31 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_scan.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_scan.c
@@ -101,12 +101,10 @@ tgsi_scan_shader(const struct tgsi_token *tokens,
                       src->Register.File == TGSI_FILE_SYSTEM_VALUE) {
                      const int ind = src->Register.Index;
                      if (info->input_semantic_name[ind] == TGSI_SEMANTIC_FOG) {
-                        if (src->Register.SwizzleX == TGSI_SWIZZLE_X) {
-                           info->uses_fogcoord = TRUE;
-                        }
-                        else if (src->Register.SwizzleX == TGSI_SWIZZLE_Y) {
-                           info->uses_frontfacing = TRUE;
-                        }
+                        info->uses_fogcoord = TRUE;
+                     }
+                     else if (info->input_semantic_name[ind] == TGSI_SEMANTIC_FACE) {
+                        info->uses_frontfacing = TRUE;
                      }
                   }
                }
diff --git a/src/gallium/auxiliary/tgsi/tgsi_text.c b/src/gallium/auxiliary/tgsi/tgsi_text.c
index 9fcffeda368..96be353e264 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_text.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_text.c
@@ -553,7 +553,7 @@ parse_register_dcl_bracket(
       report_error( ctx, "Expected literal unsigned integer" );
       return FALSE;
    }
-   bracket->first = (int) uindex;
+   bracket->first = uindex;
 
    eat_opt_white( &ctx->cur );
 
@@ -617,10 +617,12 @@ parse_register_dcl(
        * input primitive. so we want to declare just
        * the index relevant to the semantics which is in
        * the second bracket */
-      if (ctx->processor == TGSI_PROCESSOR_GEOMETRY) {
+      if (ctx->processor == TGSI_PROCESSOR_GEOMETRY && *file == TGSI_FILE_INPUT) {
          brackets[0] = brackets[1];
+         *num_brackets = 1;
+      } else {
+         *num_brackets = 2;
       }
-      *num_brackets = 2;
    }
 
    return TRUE;
@@ -738,6 +740,13 @@ parse_src_operand(
       return FALSE;
 
    src->Register.File = file;
+   if (parsed_opt_brackets) {
+      src->Register.Dimension = 1;
+      src->Dimension.Indirect = 0;
+      src->Dimension.Dimension = 0;
+      src->Dimension.Index = bracket[0].index;
+      bracket[0] = bracket[1];
+   }
    src->Register.Index = bracket[0].index;
    if (bracket[0].ind_file != TGSI_FILE_NULL) {
       src->Register.Indirect = 1;
@@ -748,12 +757,6 @@ parse_src_operand(
       src->Indirect.SwizzleZ = bracket[0].ind_comp;
       src->Indirect.SwizzleW = bracket[0].ind_comp;
    }
-   if (parsed_opt_brackets) {
-      src->Register.Dimension = 1;
-      src->Dimension.Indirect = 0;
-      src->Dimension.Dimension = 0;
-      src->Dimension.Index = bracket[1].index;
-   }
 
    /* Parse optional swizzle.
     */
@@ -933,7 +936,8 @@ static const char *semantic_names[TGSI_SEMANTIC_COUNT] =
    "NORMAL",
    "FACE",
    "EDGEFLAG",
-   "PRIM_ID"
+   "PRIM_ID",
+   "INSTANCEID"
 };
 
 static const char *interpolate_names[TGSI_INTERPOLATE_COUNT] =
@@ -968,8 +972,17 @@ static boolean parse_declaration( struct translate_ctx *ctx )
    decl = tgsi_default_full_declaration();
    decl.Declaration.File = file;
    decl.Declaration.UsageMask = writemask;
-   decl.Range.First = brackets[0].first;
-   decl.Range.Last = brackets[0].last;
+
+   if (num_brackets == 1) {
+      decl.Range.First = brackets[0].first;
+      decl.Range.Last = brackets[0].last;
+   } else {
+      decl.Range.First = brackets[1].first;
+      decl.Range.Last = brackets[1].last;
+
+      decl.Declaration.Dimension = 1;
+      decl.Dim.Index2D = brackets[0].first;
+   }
 
    cur = ctx->cur;
    eat_opt_white( &cur );
@@ -1116,7 +1129,9 @@ static const char *property_names[] =
 {
    "GS_INPUT_PRIMITIVE",
    "GS_OUTPUT_PRIMITIVE",
-   "GS_MAX_OUTPUT_VERTICES"
+   "GS_MAX_OUTPUT_VERTICES",
+   "FS_COORD_ORIGIN",
+   "FS_COORD_PIXEL_CENTER"
 };
 
 static const char *primitive_names[] =
@@ -1133,6 +1148,19 @@ static const char *primitive_names[] =
    "POLYGON"
 };
 
+static const char *fs_coord_origin_names[] =
+{
+   "UPPER_LEFT",
+   "LOWER_LEFT"
+};
+
+static const char *fs_coord_pixel_center_names[] =
+{
+   "HALF_INTEGER",
+   "INTEGER"
+};
+
+
 static boolean
 parse_primitive( const char **pcur, uint *primitive )
 {
@@ -1150,6 +1178,40 @@ parse_primitive( const char **pcur, uint *primitive )
    return FALSE;
 }
 
+static boolean
+parse_fs_coord_origin( const char **pcur, uint *fs_coord_origin )
+{
+   uint i;
+
+   for (i = 0; i < sizeof(fs_coord_origin_names) / sizeof(fs_coord_origin_names[0]); i++) {
+      const char *cur = *pcur;
+
+      if (str_match_no_case( &cur, fs_coord_origin_names[i])) {
+         *fs_coord_origin = i;
+         *pcur = cur;
+         return TRUE;
+      }
+   }
+   return FALSE;
+}
+
+static boolean
+parse_fs_coord_pixel_center( const char **pcur, uint *fs_coord_pixel_center )
+{
+   uint i;
+
+   for (i = 0; i < sizeof(fs_coord_pixel_center_names) / sizeof(fs_coord_pixel_center_names[0]); i++) {
+      const char *cur = *pcur;
+
+      if (str_match_no_case( &cur, fs_coord_pixel_center_names[i])) {
+         *fs_coord_pixel_center = i;
+         *pcur = cur;
+         return TRUE;
+      }
+   }
+   return FALSE;
+}
+
 
 static boolean parse_property( struct translate_ctx *ctx )
 {
@@ -1191,6 +1253,18 @@ static boolean parse_property( struct translate_ctx *ctx )
          ctx->implied_array_size = u_vertices_per_prim(values[0]);
       }
       break;
+   case TGSI_PROPERTY_FS_COORD_ORIGIN:
+      if (!parse_fs_coord_origin(&ctx->cur, &values[0] )) {
+         report_error( ctx, "Unknown coord origin as property: must be UPPER_LEFT or LOWER_LEFT!" );
+         return FALSE;
+      }
+      break;
+   case TGSI_PROPERTY_FS_COORD_PIXEL_CENTER:
+      if (!parse_fs_coord_pixel_center(&ctx->cur, &values[0] )) {
+         report_error( ctx, "Unknown coord pixel center as property: must be HALF_INTEGER or INTEGER!" );
+         return FALSE;
+      }
+      break;
    default:
       if (!parse_uint(&ctx->cur, &values[0] )) {
          report_error( ctx, "Expected unsigned integer as property!" );
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.c b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
index 8bd6f68dcc0..0ae46785bbe 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
@@ -44,6 +44,7 @@ union tgsi_any_token {
    struct tgsi_property_data prop_data;
    struct tgsi_declaration decl;
    struct tgsi_declaration_range decl_range;
+   struct tgsi_declaration_dimension decl_dim;
    struct tgsi_declaration_semantic decl_semantic;
    struct tgsi_immediate imm;
    union  tgsi_immediate_data imm_data;
@@ -75,6 +76,14 @@ struct ureg_tokens {
 #define UREG_MAX_LOOP 1
 #define UREG_MAX_PRED 1
 
+struct const_decl {
+   struct {
+      unsigned first;
+      unsigned last;
+   } constant_range[UREG_MAX_CONSTANT_RANGE];
+   unsigned nr_constant_ranges;
+};
+
 #define DOMAIN_DECL 0
 #define DOMAIN_INSN 1
 
@@ -127,13 +136,14 @@ struct ureg_program
    unsigned temps_active[UREG_MAX_TEMP / 32];
    unsigned nr_temps;
 
-   struct {
-      unsigned first;
-      unsigned last;
-   } constant_range[UREG_MAX_CONSTANT_RANGE];
-   unsigned nr_constant_ranges;
+   struct const_decl const_decls;
+   struct const_decl const_decls2D[PIPE_MAX_CONSTANT_BUFFERS];
 
    unsigned property_gs_input_prim;
+   unsigned property_gs_output_prim;
+   unsigned property_gs_max_vertices;
+   unsigned char property_fs_coord_origin; /* = TGSI_FS_COORD_ORIGIN_* */
+   unsigned char property_fs_coord_pixel_center; /* = TGSI_FS_COORD_PIXEL_CENTER_* */
 
    unsigned nr_addrs;
    unsigned nr_preds;
@@ -235,36 +245,40 @@ ureg_dst_register( unsigned file,
    return dst;
 }
 
-static INLINE struct ureg_src 
-ureg_src_register( unsigned file,
-                   unsigned index )
+
+void
+ureg_property_gs_input_prim(struct ureg_program *ureg,
+                            unsigned input_prim)
 {
-   struct ureg_src src;
-
-   src.File     = file;
-   src.SwizzleX = TGSI_SWIZZLE_X;
-   src.SwizzleY = TGSI_SWIZZLE_Y;
-   src.SwizzleZ = TGSI_SWIZZLE_Z;
-   src.SwizzleW = TGSI_SWIZZLE_W;
-   src.Indirect = 0;
-   src.IndirectIndex = 0;
-   src.IndirectSwizzle = 0;
-   src.Absolute = 0;
-   src.Index    = index;
-   src.Negate   = 0;
-   src.Dimension = 0;
-   src.DimensionIndex = 0;
-
-   return src;
+   ureg->property_gs_input_prim = input_prim;
 }
 
+void
+ureg_property_gs_output_prim(struct ureg_program *ureg,
+                             unsigned output_prim)
+{
+   ureg->property_gs_output_prim = output_prim;
+}
 
+void
+ureg_property_gs_max_vertices(struct ureg_program *ureg,
+                              unsigned max_vertices)
+{
+   ureg->property_gs_max_vertices = max_vertices;
+}
 
 void
-ureg_property_gs_input_prim(struct ureg_program *ureg,
-                            unsigned gs_input_prim)
+ureg_property_fs_coord_origin(struct ureg_program *ureg,
+                            unsigned fs_coord_origin)
+{
+   ureg->property_fs_coord_origin = fs_coord_origin;
+}
+
+void
+ureg_property_fs_coord_pixel_center(struct ureg_program *ureg,
+                            unsigned fs_coord_pixel_center)
 {
-   ureg->property_gs_input_prim = gs_input_prim;
+   ureg->property_fs_coord_pixel_center = fs_coord_pixel_center;
 }
 
 
@@ -374,62 +388,92 @@ out:
 /* Returns a new constant register.  Keep track of which have been
  * referred to so that we can emit decls later.
  *
+ * Constant operands declared with this function must be addressed
+ * with a two-dimensional index.
+ *
  * There is nothing in this code to bind this constant to any tracked
  * value or manage any constant_buffer contents -- that's the
  * resposibility of the calling code.
  */
-struct ureg_src ureg_DECL_constant(struct ureg_program *ureg, 
-                                   unsigned index )
+void
+ureg_DECL_constant2D(struct ureg_program *ureg,
+                     unsigned first,
+                     unsigned last,
+                     unsigned index2D)
+{
+   struct const_decl *decl = &ureg->const_decls2D[index2D];
+
+   assert(index2D < PIPE_MAX_CONSTANT_BUFFERS);
+
+   if (decl->nr_constant_ranges < UREG_MAX_CONSTANT_RANGE) {
+      uint i = decl->nr_constant_ranges++;
+
+      decl->constant_range[i].first = first;
+      decl->constant_range[i].last = last;
+   }
+}
+
+
+/* A one-dimensional, depricated version of ureg_DECL_constant2D().
+ *
+ * Constant operands declared with this function must be addressed
+ * with a one-dimensional index.
+ */
+struct ureg_src
+ureg_DECL_constant(struct ureg_program *ureg,
+                   unsigned index)
 {
+   struct const_decl *decl = &ureg->const_decls;
    unsigned minconst = index, maxconst = index;
    unsigned i;
 
    /* Inside existing range?
     */
-   for (i = 0; i < ureg->nr_constant_ranges; i++) {
-      if (ureg->constant_range[i].first <= index &&
-          ureg->constant_range[i].last >= index)
+   for (i = 0; i < decl->nr_constant_ranges; i++) {
+      if (decl->constant_range[i].first <= index &&
+          decl->constant_range[i].last >= index) {
          goto out;
+      }
    }
 
    /* Extend existing range?
     */
-   for (i = 0; i < ureg->nr_constant_ranges; i++) {
-      if (ureg->constant_range[i].last == index - 1) {
-         ureg->constant_range[i].last = index;
+   for (i = 0; i < decl->nr_constant_ranges; i++) {
+      if (decl->constant_range[i].last == index - 1) {
+         decl->constant_range[i].last = index;
          goto out;
       }
 
-      if (ureg->constant_range[i].first == index + 1) {
-         ureg->constant_range[i].first = index;
+      if (decl->constant_range[i].first == index + 1) {
+         decl->constant_range[i].first = index;
          goto out;
       }
 
-      minconst = MIN2(minconst, ureg->constant_range[i].first);
-      maxconst = MAX2(maxconst, ureg->constant_range[i].last);
+      minconst = MIN2(minconst, decl->constant_range[i].first);
+      maxconst = MAX2(maxconst, decl->constant_range[i].last);
    }
 
    /* Create new range?
     */
-   if (ureg->nr_constant_ranges < UREG_MAX_CONSTANT_RANGE) {
-      i = ureg->nr_constant_ranges++;
-      ureg->constant_range[i].first = index;
-      ureg->constant_range[i].last = index;
+   if (decl->nr_constant_ranges < UREG_MAX_CONSTANT_RANGE) {
+      i = decl->nr_constant_ranges++;
+      decl->constant_range[i].first = index;
+      decl->constant_range[i].last = index;
       goto out;
    }
 
    /* Collapse all ranges down to one:
     */
    i = 0;
-   ureg->constant_range[0].first = minconst;
-   ureg->constant_range[0].last = maxconst;
-   ureg->nr_constant_ranges = 1;
+   decl->constant_range[0].first = minconst;
+   decl->constant_range[0].last = maxconst;
+   decl->nr_constant_ranges = 1;
 
 out:
-   assert(i < ureg->nr_constant_ranges);
-   assert(ureg->constant_range[i].first <= index);
-   assert(ureg->constant_range[i].last >= index);
-   return ureg_src_register( TGSI_FILE_CONSTANT, index );
+   assert(i < decl->nr_constant_ranges);
+   assert(decl->constant_range[i].first <= index);
+   assert(decl->constant_range[i].last >= index);
+   return ureg_src_register(TGSI_FILE_CONSTANT, index);
 }
 
 
@@ -578,7 +622,7 @@ decl_immediate( struct ureg_program *ureg,
                 unsigned type )
 {
    unsigned i, j;
-   unsigned swizzle;
+   unsigned swizzle = 0;
 
    /* Could do a first pass where we examine all existing immediates
     * without expanding.
@@ -656,6 +700,35 @@ ureg_DECL_immediate_uint( struct ureg_program *ureg,
 
 
 struct ureg_src
+ureg_DECL_immediate_block_uint( struct ureg_program *ureg,
+                                const unsigned *v,
+                                unsigned nr )
+{
+   uint index;
+   uint i;
+
+   if (ureg->nr_immediates + (nr + 3) / 4 > UREG_MAX_IMMEDIATE) {
+      set_bad(ureg);
+      return ureg_src_register(TGSI_FILE_IMMEDIATE, 0);
+   }
+
+   index = ureg->nr_immediates;
+   ureg->nr_immediates += (nr + 3) / 4;
+
+   for (i = index; i < ureg->nr_immediates; i++) {
+      ureg->immediate[i].type = TGSI_IMM_UINT32;
+      ureg->immediate[i].nr = nr > 4 ? 4 : nr;
+      memcpy(ureg->immediate[i].value.u,
+             &v[(i - index) * 4],
+             ureg->immediate[i].nr * sizeof(uint));
+      nr -= 4;
+   }
+
+   return ureg_src_register(TGSI_FILE_IMMEDIATE, index);
+}
+
+
+struct ureg_src
 ureg_DECL_immediate_int( struct ureg_program *ureg,
                          const int *v,
                          unsigned nr )
@@ -691,7 +764,7 @@ ureg_emit_src( struct ureg_program *ureg,
    if (src.Indirect) {
       out[0].src.Indirect = 1;
       out[n].value = 0;
-      out[n].src.File = TGSI_FILE_ADDRESS;
+      out[n].src.File = src.IndirectFile;
       out[n].src.SwizzleX = src.IndirectSwizzle;
       out[n].src.SwizzleY = src.IndirectSwizzle;
       out[n].src.SwizzleZ = src.IndirectSwizzle;
@@ -1058,6 +1131,31 @@ static void emit_decl_range( struct ureg_program *ureg,
 }
 
 static void
+emit_decl_range2D(struct ureg_program *ureg,
+                  unsigned file,
+                  unsigned first,
+                  unsigned last,
+                  unsigned index2D)
+{
+   union tgsi_any_token *out = get_tokens(ureg, DOMAIN_DECL, 3);
+
+   out[0].value = 0;
+   out[0].decl.Type = TGSI_TOKEN_TYPE_DECLARATION;
+   out[0].decl.NrTokens = 3;
+   out[0].decl.File = file;
+   out[0].decl.UsageMask = 0xf;
+   out[0].decl.Interpolate = TGSI_INTERPOLATE_CONSTANT;
+   out[0].decl.Dimension = 1;
+
+   out[1].value = 0;
+   out[1].decl_range.First = first;
+   out[1].decl_range.Last = last;
+
+   out[2].value = 0;
+   out[2].decl_dim.Index2D = index2D;
+}
+
+static void
 emit_immediate( struct ureg_program *ureg,
                 const unsigned *v,
                 unsigned type )
@@ -1104,6 +1202,38 @@ static void emit_decls( struct ureg_program *ureg )
                     ureg->property_gs_input_prim);
    }
 
+   if (ureg->property_gs_output_prim != ~0) {
+      assert(ureg->processor == TGSI_PROCESSOR_GEOMETRY);
+
+      emit_property(ureg,
+                    TGSI_PROPERTY_GS_OUTPUT_PRIM,
+                    ureg->property_gs_output_prim);
+   }
+
+   if (ureg->property_gs_max_vertices != ~0) {
+      assert(ureg->processor == TGSI_PROCESSOR_GEOMETRY);
+
+      emit_property(ureg,
+                    TGSI_PROPERTY_GS_MAX_VERTICES,
+                    ureg->property_gs_max_vertices);
+   }
+
+   if (ureg->property_fs_coord_origin) {
+      assert(ureg->processor == TGSI_PROCESSOR_FRAGMENT);
+
+      emit_property(ureg,
+                    TGSI_PROPERTY_FS_COORD_ORIGIN,
+                    ureg->property_fs_coord_origin);
+   }
+
+   if (ureg->property_fs_coord_pixel_center) {
+      assert(ureg->processor == TGSI_PROCESSOR_FRAGMENT);
+
+      emit_property(ureg,
+                    TGSI_PROPERTY_FS_COORD_PIXEL_CENTER,
+                    ureg->property_fs_coord_pixel_center);
+   }
+
    if (ureg->processor == TGSI_PROCESSOR_VERTEX) {
       for (i = 0; i < UREG_MAX_INPUT; i++) {
          if (ureg->vs_inputs[i/32] & (1 << (i%32))) {
@@ -1152,13 +1282,29 @@ static void emit_decls( struct ureg_program *ureg )
                        ureg->sampler[i].Index, 1 );
    }
 
-   if (ureg->nr_constant_ranges) {
-      for (i = 0; i < ureg->nr_constant_ranges; i++)
-         emit_decl_range( ureg,
-                          TGSI_FILE_CONSTANT,
-                          ureg->constant_range[i].first, 
-                          (ureg->constant_range[i].last + 1 -
-                           ureg->constant_range[i].first) );
+   if (ureg->const_decls.nr_constant_ranges) {
+      for (i = 0; i < ureg->const_decls.nr_constant_ranges; i++) {
+         emit_decl_range(ureg,
+                         TGSI_FILE_CONSTANT,
+                         ureg->const_decls.constant_range[i].first,
+                         ureg->const_decls.constant_range[i].last - ureg->const_decls.constant_range[i].first + 1);
+      }
+   }
+
+   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
+      struct const_decl *decl = &ureg->const_decls2D[i];
+
+      if (decl->nr_constant_ranges) {
+         uint j;
+
+         for (j = 0; j < decl->nr_constant_ranges; j++) {
+            emit_decl_range2D(ureg,
+                              TGSI_FILE_CONSTANT,
+                              decl->constant_range[j].first,
+                              decl->constant_range[j].last,
+                              i);
+         }
+      }
    }
 
    if (ureg->nr_temps) {
@@ -1314,6 +1460,8 @@ struct ureg_program *ureg_create( unsigned processor )
 
    ureg->processor = processor;
    ureg->property_gs_input_prim = ~0;
+   ureg->property_gs_output_prim = ~0;
+   ureg->property_gs_max_vertices = ~0;
    return ureg;
 }
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.h b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
index 03eaf24854c..e25f35c6dc3 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
@@ -52,9 +52,10 @@ struct ureg_src
    unsigned Absolute    : 1;  /* BOOL */
    unsigned Negate      : 1;  /* BOOL */
    int      Index       : 16; /* SINT */
+   unsigned IndirectFile    : 4;  /* TGSI_FILE_ */
    int      IndirectIndex   : 16; /* SINT */
-   int      IndirectSwizzle : 2;  /* TGSI_SWIZZLE_ */
-   int      DimensionIndex : 16;  /* SINT */
+   unsigned IndirectSwizzle : 2;  /* TGSI_SWIZZLE_ */
+   int      DimensionIndex  : 16; /* SINT */
 };
 
 /* Very similar to a tgsi_dst_register, removing unsupported fields
@@ -125,8 +126,23 @@ ureg_create_shader_and_destroy( struct ureg_program *p,
 
 void
 ureg_property_gs_input_prim(struct ureg_program *ureg,
-                            unsigned gs_input_prim);
+                            unsigned input_prim);
+
+void
+ureg_property_gs_output_prim(struct ureg_program *ureg,
+                             unsigned output_prim);
 
+void
+ureg_property_gs_max_vertices(struct ureg_program *ureg,
+                              unsigned max_vertices);
+
+void
+ureg_property_fs_coord_origin(struct ureg_program *ureg,
+                            unsigned fs_coord_origin);
+
+void
+ureg_property_fs_coord_pixel_center(struct ureg_program *ureg,
+                            unsigned fs_coord_pixel_center);
 
 /***********************************************************************
  * Build shader declarations:
@@ -168,10 +184,21 @@ ureg_DECL_immediate_uint( struct ureg_program *,
                           unsigned nr );
 
 struct ureg_src
+ureg_DECL_immediate_block_uint( struct ureg_program *,
+                                const unsigned *v,
+                                unsigned nr );
+
+struct ureg_src
 ureg_DECL_immediate_int( struct ureg_program *,
                          const int *v,
                          unsigned nr );
 
+void
+ureg_DECL_constant2D(struct ureg_program *ureg,
+                     unsigned first,
+                     unsigned last,
+                     unsigned index2D);
+
 struct ureg_src
 ureg_DECL_constant( struct ureg_program *,
                     unsigned index );
@@ -768,8 +795,9 @@ static INLINE struct ureg_src
 ureg_src_indirect( struct ureg_src reg, struct ureg_src addr )
 {
    assert(reg.File != TGSI_FILE_NULL);
-   assert(addr.File == TGSI_FILE_ADDRESS);
+   assert(addr.File == TGSI_FILE_ADDRESS || addr.File == TGSI_FILE_TEMPORARY);
    reg.Indirect = 1;
+   reg.IndirectFile = addr.File;
    reg.IndirectIndex = addr.Index;
    reg.IndirectSwizzle = addr.SwizzleX;
    return reg;
@@ -789,6 +817,8 @@ ureg_dst( struct ureg_src src )
 {
    struct ureg_dst dst;
 
+   assert(!src.Indirect || src.IndirectFile == TGSI_FILE_ADDRESS);
+
    dst.File      = src.File;
    dst.WriteMask = TGSI_WRITEMASK_XYZW;
    dst.Indirect  = src.Indirect;
@@ -807,6 +837,30 @@ ureg_dst( struct ureg_src src )
 }
 
 static INLINE struct ureg_src
+ureg_src_register(unsigned file,
+                  unsigned index)
+{
+   struct ureg_src src;
+
+   src.File = file;
+   src.SwizzleX = TGSI_SWIZZLE_X;
+   src.SwizzleY = TGSI_SWIZZLE_Y;
+   src.SwizzleZ = TGSI_SWIZZLE_Z;
+   src.SwizzleW = TGSI_SWIZZLE_W;
+   src.Indirect = 0;
+   src.IndirectFile = TGSI_FILE_NULL;
+   src.IndirectIndex = 0;
+   src.IndirectSwizzle = 0;
+   src.Absolute = 0;
+   src.Index = index;
+   src.Negate = 0;
+   src.Dimension = 0;
+   src.DimensionIndex = 0;
+
+   return src;
+}
+
+static INLINE struct ureg_src
 ureg_src( struct ureg_dst dst )
 {
    struct ureg_src src;
@@ -817,6 +871,7 @@ ureg_src( struct ureg_dst dst )
    src.SwizzleZ  = TGSI_SWIZZLE_Z;
    src.SwizzleW  = TGSI_SWIZZLE_W;
    src.Indirect  = dst.Indirect;
+   src.IndirectFile = TGSI_FILE_ADDRESS;
    src.IndirectIndex = dst.IndirectIndex;
    src.IndirectSwizzle = dst.IndirectSwizzle;
    src.Absolute  = 0;
@@ -863,6 +918,7 @@ ureg_src_undef( void )
    src.SwizzleZ  = 0;
    src.SwizzleW  = 0;
    src.Indirect  = 0;
+   src.IndirectFile = TGSI_FILE_NULL;
    src.IndirectIndex = 0;
    src.IndirectSwizzle = 0;
    src.Absolute  = 0;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_util.c b/src/gallium/auxiliary/tgsi/tgsi_util.c
index f4ca9e21ed9..0a7e4105a80 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_util.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_util.c
@@ -28,7 +28,6 @@
 #include "util/u_debug.h"
 #include "pipe/p_shader_tokens.h"
 #include "tgsi_parse.h"
-#include "tgsi_build.h"
 #include "tgsi_util.h"
 
 union pointer_hack