16 files changed, 3232 insertions, 1884 deletions
diff --git a/mesalib/src/glsl/Makefile b/mesalib/src/glsl/Makefile
index f964b7487..7660cb3dd 100644
--- a/mesalib/src/glsl/Makefile
+++ b/mesalib/src/glsl/Makefile
@@ -208,6 +208,6 @@ builtin_compiler: $(GLSL2_OBJECTS) $(OBJECTS) builtin_stubs.o
 
 builtin_function.cpp: builtins/profiles/* builtins/ir/* builtins/tools/generate_builtins.py builtins/tools/texture_builtins.py builtin_compiler
 	@echo Regenerating builtin_function.cpp...
-	$(PYTHON2) $(PYTHON_FLAGS) builtins/tools/generate_builtins.py ./builtin_compiler > builtin_function.cpp
+	$(PYTHON2) $(PYTHON_FLAGS) builtins/tools/generate_builtins.py ./builtin_compiler > builtin_function.cpp || rm -f builtin_function.cpp
 
 -include depend
diff --git a/mesalib/src/mesa/main/extensions.c b/mesalib/src/mesa/main/extensions.c
index 9f58b45e2..1b540e99a 100644
--- a/mesalib/src/mesa/main/extensions.c
+++ b/mesalib/src/mesa/main/extensions.c
@@ -428,6 +428,7 @@ _mesa_enable_sw_extensions(struct gl_context *ctx)
    ctx->Extensions.ARB_texture_mirrored_repeat = GL_TRUE;
    ctx->Extensions.ARB_texture_non_power_of_two = GL_TRUE;
    ctx->Extensions.ARB_texture_rg = GL_TRUE;
+   ctx->Extensions.ARB_texture_compression_rgtc = GL_TRUE;
    ctx->Extensions.ARB_vertex_array_object = GL_TRUE;
 #if FEATURE_ARB_vertex_program
    ctx->Extensions.ARB_vertex_program = GL_TRUE;
diff --git a/mesalib/src/mesa/main/formats.c b/mesalib/src/mesa/main/formats.c
index 5c9b3068e..1d7bdad24 100644
--- a/mesalib/src/mesa/main/formats.c
+++ b/mesalib/src/mesa/main/formats.c
@@ -890,7 +890,43 @@ static struct gl_format_info format_info[MESA_FORMAT_COUNT] =
       16, 16, 16, 16,
       0, 0, 0, 0, 0,
       1, 1, 8
-   }
+   },
+   {
+     MESA_FORMAT_RED_RGTC1,
+     "MESA_FORMAT_RED_RGTC1",
+     GL_RED,
+     GL_UNSIGNED_NORMALIZED,
+     4, 0, 0, 0,
+     0, 0, 0, 0, 0,
+     4, 4, 8                     /* 8 bytes per 4x4 block */
+   },
+   {
+     MESA_FORMAT_SIGNED_RED_RGTC1,
+     "MESA_FORMAT_SIGNED_RED_RGTC1",
+     GL_RED,
+     GL_SIGNED_NORMALIZED,
+     4, 0, 0, 0,
+     0, 0, 0, 0, 0,
+     4, 4, 8                     /* 8 bytes per 4x4 block */
+   },
+   {
+     MESA_FORMAT_RG_RGTC2,
+     "MESA_FORMAT_RG_RGTC2",
+     GL_RG,
+     GL_UNSIGNED_NORMALIZED,
+     4, 4, 0, 0,
+     0, 0, 0, 0, 0,
+     4, 4, 16                     /* 16 bytes per 4x4 block */
+   },
+   {
+     MESA_FORMAT_SIGNED_RG_RGTC2,
+     "MESA_FORMAT_SIGNED_RG_RGTC2",
+     GL_RG,
+     GL_SIGNED_NORMALIZED,
+     4, 4, 0, 0,
+     0, 0, 0, 0, 0,
+     4, 4, 16                     /* 16 bytes per 4x4 block */
+   },
 };
 
 
@@ -1530,6 +1566,10 @@ _mesa_format_to_type_and_comps(gl_format format,
    case MESA_FORMAT_SRGBA_DXT5:
 #endif
 #endif
+   case MESA_FORMAT_RED_RGTC1:
+   case MESA_FORMAT_SIGNED_RED_RGTC1:
+   case MESA_FORMAT_RG_RGTC2:
+   case MESA_FORMAT_SIGNED_RG_RGTC2:
       /* XXX generate error instead? */
       *datatype = GL_UNSIGNED_BYTE;
       *comps = 0;
diff --git a/mesalib/src/mesa/main/formats.h b/mesalib/src/mesa/main/formats.h
index cd7e0e25e..b32829145 100644
--- a/mesalib/src/mesa/main/formats.h
+++ b/mesalib/src/mesa/main/formats.h
@@ -179,6 +179,12 @@ typedef enum
    MESA_FORMAT_RGBA_16,           /* ... */
    /*@}*/
 
+   /*@{*/
+   MESA_FORMAT_RED_RGTC1,
+   MESA_FORMAT_SIGNED_RED_RGTC1,
+   MESA_FORMAT_RG_RGTC2,
+   MESA_FORMAT_SIGNED_RG_RGTC2,
+   /*@}*/
    MESA_FORMAT_COUNT
 } gl_format;
 
diff --git a/mesalib/src/mesa/main/texcompress.c b/mesalib/src/mesa/main/texcompress.c
index eaebb06d3..9891946df 100644
--- a/mesalib/src/mesa/main/texcompress.c
+++ b/mesalib/src/mesa/main/texcompress.c
@@ -64,6 +64,7 @@ _mesa_get_compressed_formats(struct gl_context *ctx, GLint *formats, GLboolean a
          n += 2;
       }
    }
+   /* don't return RGTC - ARB_texture_compression_rgtc query 19 */
    if (ctx->Extensions.EXT_texture_compression_s3tc) {
       if (formats) {
          formats[n++] = GL_COMPRESSED_RGB_S3TC_DXT1_EXT;
@@ -163,6 +164,15 @@ _mesa_glenum_to_compressed_format(GLenum format)
    case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT:
       return MESA_FORMAT_SRGBA_DXT5;
 
+   case GL_COMPRESSED_RED_RGTC1:
+      return MESA_FORMAT_RED_RGTC1;
+   case GL_COMPRESSED_SIGNED_RED_RGTC1:
+      return MESA_FORMAT_SIGNED_RED_RGTC1;
+   case GL_COMPRESSED_RG_RGTC2:
+      return MESA_FORMAT_RG_RGTC2;
+   case GL_COMPRESSED_SIGNED_RG_RGTC2:
+      return MESA_FORMAT_SIGNED_RG_RGTC2;
+
    default:
       return MESA_FORMAT_NONE;
    }
@@ -209,6 +219,16 @@ _mesa_compressed_format_to_glenum(struct gl_context *ctx, GLuint mesaFormat)
       return GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT;
 #endif
 #endif
+
+   case MESA_FORMAT_RED_RGTC1:
+      return GL_COMPRESSED_RED_RGTC1;
+   case MESA_FORMAT_SIGNED_RED_RGTC1:
+      return GL_COMPRESSED_SIGNED_RED_RGTC1;
+   case MESA_FORMAT_RG_RGTC2:
+      return GL_COMPRESSED_RG_RGTC2;
+   case MESA_FORMAT_SIGNED_RG_RGTC2:
+      return GL_COMPRESSED_SIGNED_RG_RGTC2;
+
    default:
       _mesa_problem(ctx, "Unexpected mesa texture format in"
                     " _mesa_compressed_format_to_glenum()");
diff --git a/mesalib/src/mesa/main/texcompress_rgtc.c b/mesalib/src/mesa/main/texcompress_rgtc.c
new file mode 100644
index 000000000..586715083
--- /dev/null
+++ b/mesalib/src/mesa/main/texcompress_rgtc.c
@@ -0,0 +1,1122 @@
+/*
+ * Copyright (C) 2011 Red Hat Inc.
+ * 
+ * block compression parts are:
+ * Copyright (C) 2004  Roland Scheidegger   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Author:
+ *    Dave Airlie
+ */
+
+/**
+ * \file texcompress_rgtc.c
+ * GL_EXT_texture_compression_rgtc support.
+ */
+
+
+#include "glheader.h"
+#include "imports.h"
+#include "colormac.h"
+#include "image.h"
+#include "macros.h"
+#include "mfeatures.h"
+#include "mipmap.h"
+#include "texcompress.h"
+#include "texcompress_rgtc.h"
+#include "texstore.h"
+
+#define RGTC_DEBUG 0
+
+static void encode_rgtc_chan_u(GLubyte *blkaddr, GLubyte srccolors[4][4],
+			     GLint numxpixels, GLint numypixels);
+static void encode_rgtc_chan_s(GLbyte *blkaddr, GLbyte srccolors[4][4],
+			     GLint numxpixels, GLint numypixels);
+
+static void extractsrc_u( GLubyte srcpixels[4][4], const GLchan *srcaddr,
+			  GLint srcRowStride, GLint numxpixels, GLint numypixels, GLint comps)
+{
+   GLubyte i, j;
+   const GLchan *curaddr;
+   for (j = 0; j < numypixels; j++) {
+      curaddr = srcaddr + j * srcRowStride * comps;
+      for (i = 0; i < numxpixels; i++) {
+	 srcpixels[j][i] = *curaddr / (CHAN_MAX / 255);
+	 curaddr += comps;
+      }
+   }
+}
+
+static void extractsrc_s( GLbyte srcpixels[4][4], const GLfloat *srcaddr,
+			  GLint srcRowStride, GLint numxpixels, GLint numypixels, GLint comps)
+{
+   GLubyte i, j;
+   const GLfloat *curaddr;
+   for (j = 0; j < numypixels; j++) {
+      curaddr = srcaddr + j * srcRowStride * comps;
+      for (i = 0; i < numxpixels; i++) {
+	 srcpixels[j][i] = FLOAT_TO_BYTE_TEX(*curaddr);
+	 curaddr += comps;
+      }
+   }
+}
+
+
+GLboolean
+_mesa_texstore_red_rgtc1(TEXSTORE_PARAMS)
+{
+   GLubyte *dst;
+   const GLint texWidth = dstRowStride * 4 / 8; /* a bit of a hack */
+   const GLchan *tempImage = NULL;
+   int i, j;
+   int numxpixels, numypixels;
+   const char *srcaddr;
+   GLubyte srcpixels[4][4];
+   GLubyte *blkaddr;
+   GLint dstRowDiff;
+   ASSERT(dstFormat == MESA_FORMAT_RED_RGTC1);
+   ASSERT(dstXoffset % 4 == 0);
+   ASSERT(dstYoffset % 4 == 0);
+   ASSERT(dstZoffset % 4 == 0);
+   (void) dstZoffset;
+   (void) dstImageOffsets;
+
+
+   tempImage = _mesa_make_temp_chan_image(ctx, dims,
+					  baseInternalFormat,
+					  _mesa_get_format_base_format(dstFormat),
+					  srcWidth, srcHeight, srcDepth,
+					  srcFormat, srcType, srcAddr,
+					  srcPacking);
+   if (!tempImage)
+      return GL_FALSE; /* out of memory */
+
+   dst = _mesa_compressed_image_address(dstXoffset, dstYoffset, 0,
+                                        dstFormat,
+                                        texWidth, (GLubyte *) dstAddr);
+
+   blkaddr = dst;
+   dstRowDiff = dstRowStride >= (srcWidth * 4) ? dstRowStride - (((srcWidth + 3) & ~3) * 4) : 0;
+   for (j = 0; j < srcHeight; j+=4) {
+      if (srcHeight > j + 3) numypixels = 4;
+      else numypixels = srcHeight - j;
+      srcaddr = tempImage + j * srcWidth;
+      for (i = 0; i < srcWidth; i += 4) {
+	 if (srcWidth > i + 3) numxpixels = 4;
+	 else numxpixels = srcWidth - i;
+	 extractsrc_u(srcpixels, srcaddr, srcWidth, numxpixels, numypixels, 1);
+	 encode_rgtc_chan_u(blkaddr, srcpixels, numxpixels, numypixels);
+	 srcaddr += numxpixels;
+	 blkaddr += 8;
+      }
+      blkaddr += dstRowDiff;
+   }
+   if (tempImage)
+      free((void *) tempImage);
+
+   return GL_TRUE;
+}
+
+GLboolean
+_mesa_texstore_signed_red_rgtc1(TEXSTORE_PARAMS)
+{
+   GLbyte *dst;
+   const GLint texWidth = dstRowStride * 4 / 8; /* a bit of a hack */
+   const GLfloat *tempImage = NULL;
+   int i, j;
+   int numxpixels, numypixels;
+   const GLfloat *srcaddr;
+   GLbyte srcpixels[4][4];
+   GLbyte *blkaddr;
+   GLint dstRowDiff;
+   ASSERT(dstFormat == MESA_FORMAT_SIGNED_RED_RGTC1);
+   ASSERT(dstXoffset % 4 == 0);
+   ASSERT(dstYoffset % 4 == 0);
+   ASSERT(dstZoffset % 4 == 0);
+   (void) dstZoffset;
+   (void) dstImageOffsets;
+
+   tempImage = _mesa_make_temp_float_image(ctx, dims,
+					   baseInternalFormat,
+					   _mesa_get_format_base_format(dstFormat),
+					   srcWidth, srcHeight, srcDepth,
+					   srcFormat, srcType, srcAddr,
+					   srcPacking, 0x0);
+   if (!tempImage)
+      return GL_FALSE; /* out of memory */
+
+   dst = (GLbyte *)_mesa_compressed_image_address(dstXoffset, dstYoffset, 0,
+						  dstFormat,
+						  texWidth, (GLubyte *) dstAddr);
+
+   blkaddr = dst;
+   dstRowDiff = dstRowStride >= (srcWidth * 4) ? dstRowStride - (((srcWidth + 3) & ~3) * 4) : 0;
+   for (j = 0; j < srcHeight; j+=4) {
+      if (srcHeight > j + 3) numypixels = 4;
+      else numypixels = srcHeight - j;
+      srcaddr = tempImage + j * srcWidth;
+      for (i = 0; i < srcWidth; i += 4) {
+	 if (srcWidth > i + 3) numxpixels = 4;
+	 else numxpixels = srcWidth - i;
+	 extractsrc_s(srcpixels, srcaddr, srcWidth, numxpixels, numypixels, 1);
+	 encode_rgtc_chan_s(blkaddr, srcpixels, numxpixels, numypixels);
+	 srcaddr += numxpixels;
+	 blkaddr += 8;
+      }
+      blkaddr += dstRowDiff;
+   }
+   if (tempImage)
+      free((void *) tempImage);
+
+   return GL_TRUE;
+}
+
+GLboolean
+_mesa_texstore_rg_rgtc2(TEXSTORE_PARAMS)
+{
+   GLubyte *dst;
+   const GLint texWidth = dstRowStride * 4 / 16; /* a bit of a hack */
+   const GLchan *tempImage = NULL;
+   int i, j;
+   int numxpixels, numypixels;
+   const char *srcaddr;
+   GLubyte srcpixels[4][4];
+   GLubyte *blkaddr;
+   GLint dstRowDiff;
+
+   ASSERT(dstFormat == MESA_FORMAT_RG_RGTC2);
+   ASSERT(dstXoffset % 4 == 0);
+   ASSERT(dstYoffset % 4 == 0);
+   ASSERT(dstZoffset % 4 == 0);
+   (void) dstZoffset;
+   (void) dstImageOffsets;
+
+   tempImage = _mesa_make_temp_chan_image(ctx, dims,
+					  baseInternalFormat,
+					  _mesa_get_format_base_format(dstFormat),
+					  srcWidth, srcHeight, srcDepth,
+					  srcFormat, srcType, srcAddr,
+					  srcPacking);
+   if (!tempImage)
+      return GL_FALSE; /* out of memory */
+
+   dst = _mesa_compressed_image_address(dstXoffset, dstYoffset, 0,
+                                        dstFormat,
+                                        texWidth, (GLubyte *) dstAddr);
+
+   blkaddr = dst;
+   dstRowDiff = dstRowStride >= (srcWidth * 8) ? dstRowStride - (((srcWidth + 7) & ~7) * 8) : 0;
+   for (j = 0; j < srcHeight; j+=4) {
+      if (srcHeight > j + 3) numypixels = 4;
+      else numypixels = srcHeight - j;
+      srcaddr = tempImage + j * srcWidth * 2;
+      for (i = 0; i < srcWidth; i += 4) {
+	 if (srcWidth > i + 3) numxpixels = 4;
+	 else numxpixels = srcWidth - i;
+	 extractsrc_u(srcpixels, srcaddr, srcWidth, numxpixels, numypixels, 2);
+	 encode_rgtc_chan_u(blkaddr, srcpixels, numxpixels, numypixels);
+
+	 blkaddr += 8;
+	 extractsrc_u(srcpixels, (GLchan *)srcaddr + 1, srcWidth, numxpixels, numypixels, 2);
+	 encode_rgtc_chan_u(blkaddr, srcpixels, numxpixels, numypixels);
+
+	 blkaddr += 8;
+
+	 srcaddr += numxpixels * 2;
+      }
+      blkaddr += dstRowDiff;
+   }
+   if (tempImage)
+      free((void *) tempImage);
+
+   return GL_TRUE;
+}
+
+GLboolean
+_mesa_texstore_signed_rg_rgtc2(TEXSTORE_PARAMS)
+{
+   GLbyte *dst;
+   const GLint texWidth = dstRowStride * 4 / 16; /* a bit of a hack */
+   const GLfloat *tempImage = NULL;
+   int i, j;
+   int numxpixels, numypixels;
+   const GLfloat *srcaddr;
+   GLbyte srcpixels[4][4];
+   GLbyte *blkaddr;
+   GLint dstRowDiff;
+
+   ASSERT(dstFormat == MESA_FORMAT_SIGNED_RG_RGTC2);
+   ASSERT(dstXoffset % 4 == 0);
+   ASSERT(dstYoffset % 4 == 0);
+   ASSERT(dstZoffset % 4 == 0);
+   (void) dstZoffset;
+   (void) dstImageOffsets;
+
+   tempImage = _mesa_make_temp_float_image(ctx, dims,
+					   baseInternalFormat,
+					   _mesa_get_format_base_format(dstFormat),
+					   srcWidth, srcHeight, srcDepth,
+					   srcFormat, srcType, srcAddr,
+					   srcPacking, 0x0);
+   if (!tempImage)
+      return GL_FALSE; /* out of memory */
+
+   dst = (GLbyte *)_mesa_compressed_image_address(dstXoffset, dstYoffset, 0,
+						  dstFormat,
+						  texWidth, (GLubyte *) dstAddr);
+
+   blkaddr = dst;
+   dstRowDiff = dstRowStride >= (srcWidth * 8) ? dstRowStride - (((srcWidth + 7) & ~7) * 8) : 0;
+   for (j = 0; j < srcHeight; j += 4) {
+      if (srcHeight > j + 3) numypixels = 4;
+      else numypixels = srcHeight - j;
+      srcaddr = tempImage + j * srcWidth * 2;
+      for (i = 0; i < srcWidth; i += 4) {
+	 if (srcWidth > i + 3) numxpixels = 4;
+	 else numxpixels = srcWidth - i;
+
+	 extractsrc_s(srcpixels, srcaddr, srcWidth, numxpixels, numypixels, 2);
+	 encode_rgtc_chan_s(blkaddr, srcpixels, numxpixels, numypixels);
+	 blkaddr += 8;
+
+	 extractsrc_s(srcpixels, srcaddr + 1, srcWidth, numxpixels, numypixels, 2);
+	 encode_rgtc_chan_s(blkaddr, srcpixels, numxpixels, numypixels);
+	 blkaddr += 8;
+
+	 srcaddr += numxpixels * 2;
+
+      }
+      blkaddr += dstRowDiff;
+   }
+   if (tempImage)
+      free((void *) tempImage);
+
+   return GL_TRUE;
+}
+
+static void _fetch_texel_rgtc_u(GLint srcRowStride, const GLubyte *pixdata,
+				GLint i, GLint j, GLchan *value, int comps)
+{
+   GLchan decode;
+   const GLubyte *blksrc = (pixdata + ((srcRowStride + 3) / 4 * (j / 4) + (i / 4)) * 8 * comps);
+   const GLubyte alpha0 = blksrc[0];
+   const GLubyte alpha1 = blksrc[1];
+   const GLubyte bit_pos = ((j&3) * 4 + (i&3)) * 3;
+   const GLubyte acodelow = blksrc[2 + bit_pos / 8];
+   const GLubyte acodehigh = blksrc[3 + bit_pos / 8];
+   const GLubyte code = (acodelow >> (bit_pos & 0x7) |
+      (acodehigh  << (8 - (bit_pos & 0x7)))) & 0x7;
+
+   if (code == 0)
+      decode = UBYTE_TO_CHAN( alpha0 );
+   else if (code == 1)
+      decode = UBYTE_TO_CHAN( alpha1 );
+   else if (alpha0 > alpha1)
+      decode = UBYTE_TO_CHAN( ((alpha0 * (8 - code) + (alpha1 * (code - 1))) / 7) );
+   else if (code < 6)
+      decode = UBYTE_TO_CHAN( ((alpha0 * (6 - code) + (alpha1 * (code - 1))) / 5) );
+   else if (code == 6)
+      decode = 0;
+   else
+      decode = CHAN_MAX;
+
+   *value = decode;
+}
+
+
+static void _fetch_texel_rgtc_s(GLint srcRowStride, const GLbyte *pixdata,
+				GLint i, GLint j, GLbyte *value, int comps)
+{
+   GLbyte decode;
+   const GLbyte *blksrc = (pixdata + ((srcRowStride + 3) / 4 * (j / 4) + (i / 4)) * 8 * comps);
+   const GLbyte alpha0 = blksrc[0];
+   const GLbyte alpha1 = blksrc[1];
+   const GLbyte bit_pos = ((j&3) * 4 + (i&3)) * 3;
+   const GLbyte acodelow = blksrc[2 + bit_pos / 8];
+   const GLbyte acodehigh = blksrc[3 + bit_pos / 8];
+   const GLbyte code = (acodelow >> (bit_pos & 0x7) |
+      (acodehigh  << (8 - (bit_pos & 0x7)))) & 0x7;
+
+   if (code == 0)
+      decode = alpha0;
+   else if (code == 1)
+      decode = alpha1;
+   else if (alpha0 > alpha1)
+      decode = ((alpha0 * (8 - code) + (alpha1 * (code - 1))) / 7);
+   else if (code < 6)
+      decode = ((alpha0 * (6 - code) + (alpha1 * (code - 1))) / 5);
+   else if (code == 6)
+      decode = -128;
+   else
+      decode = 127;
+
+   *value = decode;
+}
+
+void
+_mesa_fetch_texel_2d_f_red_rgtc1(const struct gl_texture_image *texImage,
+				 GLint i, GLint j, GLint k, GLfloat *texel)
+{
+   GLchan red;
+   _fetch_texel_rgtc_u(texImage->RowStride, (GLubyte *)(texImage->Data),
+		       i, j, &red, 1);
+   texel[RCOMP] = CHAN_TO_FLOAT(red);
+   texel[GCOMP] = 0.0;
+   texel[BCOMP] = 0.0;
+   texel[ACOMP] = 1.0;
+}
+
+void
+_mesa_fetch_texel_2d_f_signed_red_rgtc1(const struct gl_texture_image *texImage,
+					GLint i, GLint j, GLint k, GLfloat *texel)
+{
+   GLbyte red;
+   _fetch_texel_rgtc_s(texImage->RowStride, (GLbyte *)(texImage->Data),
+		       i, j, &red, 1);
+   texel[RCOMP] = BYTE_TO_FLOAT_TEX(red);
+   texel[GCOMP] = 0.0;
+   texel[BCOMP] = 0.0;
+   texel[ACOMP] = 1.0;
+}
+
+void
+_mesa_fetch_texel_2d_f_rg_rgtc2(const struct gl_texture_image *texImage,
+				 GLint i, GLint j, GLint k, GLfloat *texel)
+{
+   GLchan red, green;
+   _fetch_texel_rgtc_u(texImage->RowStride, (GLubyte *)(texImage->Data),
+		     i, j, &red, 2);
+   _fetch_texel_rgtc_u(texImage->RowStride, (GLubyte *)(texImage->Data) + 8,
+		     i, j, &green, 2);
+   texel[RCOMP] = CHAN_TO_FLOAT(red);
+   texel[GCOMP] = CHAN_TO_FLOAT(green);
+   texel[BCOMP] = 0.0;
+   texel[ACOMP] = 1.0;
+}
+
+void
+_mesa_fetch_texel_2d_f_signed_rg_rgtc2(const struct gl_texture_image *texImage,
+				       GLint i, GLint j, GLint k, GLfloat *texel)
+{
+   GLbyte red, green;
+   _fetch_texel_rgtc_s(texImage->RowStride, (GLbyte *)(texImage->Data),
+		     i, j, &red, 2);
+   _fetch_texel_rgtc_s(texImage->RowStride, (GLbyte *)(texImage->Data) + 8,
+		     i, j, &green, 2);
+   texel[RCOMP] = BYTE_TO_FLOAT_TEX(red);
+   texel[GCOMP] = BYTE_TO_FLOAT_TEX(green);
+   texel[BCOMP] = 0.0;
+   texel[ACOMP] = 1.0;
+}
+
+static void write_rgtc_encoded_channel(GLubyte *blkaddr,
+				       GLubyte alphabase1,
+				       GLubyte alphabase2,
+				       GLubyte alphaenc[16])
+{
+   *blkaddr++ = alphabase1;
+   *blkaddr++ = alphabase2;
+   *blkaddr++ = alphaenc[0] | (alphaenc[1] << 3) | ((alphaenc[2] & 3) << 6);
+   *blkaddr++ = (alphaenc[2] >> 2) | (alphaenc[3] << 1) | (alphaenc[4] << 4) | ((alphaenc[5] & 1) << 7);
+   *blkaddr++ = (alphaenc[5] >> 1) | (alphaenc[6] << 2) | (alphaenc[7] << 5);
+   *blkaddr++ = alphaenc[8] | (alphaenc[9] << 3) | ((alphaenc[10] & 3) << 6);
+   *blkaddr++ = (alphaenc[10] >> 2) | (alphaenc[11] << 1) | (alphaenc[12] << 4) | ((alphaenc[13] & 1) << 7);
+   *blkaddr++ = (alphaenc[13] >> 1) | (alphaenc[14] << 2) | (alphaenc[15] << 5);
+}
+
+static void encode_rgtc_chan_u(GLubyte *blkaddr, GLubyte srccolors[4][4],
+			     GLint numxpixels, GLint numypixels)
+{
+   GLubyte alphabase[2], alphause[2];
+   GLshort alphatest[2] = { 0 };
+   GLuint alphablockerror1, alphablockerror2, alphablockerror3;
+   GLubyte i, j, aindex, acutValues[7];
+   GLubyte alphaenc1[16], alphaenc2[16], alphaenc3[16];
+   GLboolean alphaabsmin = GL_FALSE;
+   GLboolean alphaabsmax = GL_FALSE;
+   GLshort alphadist;
+
+   /* find lowest and highest alpha value in block, alphabase[0] lowest, alphabase[1] highest */
+   alphabase[0] = 0xff; alphabase[1] = 0x0;
+   for (j = 0; j < numypixels; j++) {
+      for (i = 0; i < numxpixels; i++) {
+         if (srccolors[j][i] == 0)
+            alphaabsmin = GL_TRUE;
+         else if (srccolors[j][i] == 255)
+            alphaabsmax = GL_TRUE;
+         else {
+            if (srccolors[j][i] > alphabase[1])
+               alphabase[1] = srccolors[j][i];
+            if (srccolors[j][i] < alphabase[0])
+               alphabase[0] = srccolors[j][i];
+         }
+      }
+   }
+
+
+   if ((alphabase[0] > alphabase[1]) && !(alphaabsmin && alphaabsmax)) { /* one color, either max or min */
+      /* shortcut here since it is a very common case (and also avoids later problems) */
+      /* || (alphabase[0] == alphabase[1] && !alphaabsmin && !alphaabsmax) */
+      /* could also thest for alpha0 == alpha1 (and not min/max), but probably not common, so don't bother */
+
+      *blkaddr++ = srccolors[0][0];
+      blkaddr++;
+      *blkaddr++ = 0;
+      *blkaddr++ = 0;
+      *blkaddr++ = 0;
+      *blkaddr++ = 0;
+      *blkaddr++ = 0;
+      *blkaddr++ = 0;
+#if RGTC_DEBUG
+      fprintf(stderr, "enc0 used\n");
+#endif
+      return;
+   }
+
+   /* find best encoding for alpha0 > alpha1 */
+   /* it's possible this encoding is better even if both alphaabsmin and alphaabsmax are true */
+   alphablockerror1 = 0x0;
+   alphablockerror2 = 0xffffffff;
+   alphablockerror3 = 0xffffffff;
+   if (alphaabsmin) alphause[0] = 0;
+   else alphause[0] = alphabase[0];
+   if (alphaabsmax) alphause[1] = 255;
+   else alphause[1] = alphabase[1];
+   /* calculate the 7 cut values, just the middle between 2 of the computed alpha values */
+   for (aindex = 0; aindex < 7; aindex++) {
+      /* don't forget here is always rounded down */
+      acutValues[aindex] = (alphause[0] * (2*aindex + 1) + alphause[1] * (14 - (2*aindex + 1))) / 14;
+   }
+
+   for (j = 0; j < numypixels; j++) {
+      for (i = 0; i < numxpixels; i++) {
+         /* maybe it's overkill to have the most complicated calculation just for the error
+            calculation which we only need to figure out if encoding1 or encoding2 is better... */
+         if (srccolors[j][i] > acutValues[0]) {
+            alphaenc1[4*j + i] = 0;
+            alphadist = srccolors[j][i] - alphause[1];
+         }
+         else if (srccolors[j][i] > acutValues[1]) {
+            alphaenc1[4*j + i] = 2;
+            alphadist = srccolors[j][i] - (alphause[1] * 6 + alphause[0] * 1) / 7;
+         }
+         else if (srccolors[j][i] > acutValues[2]) {
+            alphaenc1[4*j + i] = 3;
+            alphadist = srccolors[j][i] - (alphause[1] * 5 + alphause[0] * 2) / 7;
+         }
+         else if (srccolors[j][i] > acutValues[3]) {
+            alphaenc1[4*j + i] = 4;
+            alphadist = srccolors[j][i] - (alphause[1] * 4 + alphause[0] * 3) / 7;
+         }
+         else if (srccolors[j][i] > acutValues[4]) {
+            alphaenc1[4*j + i] = 5;
+            alphadist = srccolors[j][i] - (alphause[1] * 3 + alphause[0] * 4) / 7;
+         }
+         else if (srccolors[j][i] > acutValues[5]) {
+            alphaenc1[4*j + i] = 6;
+            alphadist = srccolors[j][i] - (alphause[1] * 2 + alphause[0] * 5) / 7;
+         }
+         else if (srccolors[j][i] > acutValues[6]) {
+            alphaenc1[4*j + i] = 7;
+            alphadist = srccolors[j][i] - (alphause[1] * 1 + alphause[0] * 6) / 7;
+         }
+         else {
+            alphaenc1[4*j + i] = 1;
+            alphadist = srccolors[j][i] - alphause[0];
+         }
+         alphablockerror1 += alphadist * alphadist;
+      }
+   }
+
+#if RGTC_DEBUG
+   for (i = 0; i < 16; i++) {
+      fprintf(stderr, "%d ", alphaenc1[i]);
+   }
+   fprintf(stderr, "cutVals ");
+   for (i = 0; i < 8; i++) {
+      fprintf(stderr, "%d ", acutValues[i]);
+   }
+   fprintf(stderr, "srcVals ");
+   for (j = 0; j < numypixels; j++) {
+      for (i = 0; i < numxpixels; i++) {
+	 fprintf(stderr, "%d ", srccolors[j][i]);
+      }
+   }
+   fprintf(stderr, "\n");
+#endif
+
+   /* it's not very likely this encoding is better if both alphaabsmin and alphaabsmax
+      are false but try it anyway */
+   if (alphablockerror1 >= 32) {
+
+      /* don't bother if encoding is already very good, this condition should also imply
+      we have valid alphabase colors which we absolutely need (alphabase[0] <= alphabase[1]) */
+      alphablockerror2 = 0;
+      for (aindex = 0; aindex < 5; aindex++) {
+         /* don't forget here is always rounded down */
+         acutValues[aindex] = (alphabase[0] * (10 - (2*aindex + 1)) + alphabase[1] * (2*aindex + 1)) / 10;
+      }
+      for (j = 0; j < numypixels; j++) {
+         for (i = 0; i < numxpixels; i++) {
+             /* maybe it's overkill to have the most complicated calculation just for the error
+               calculation which we only need to figure out if encoding1 or encoding2 is better... */
+            if (srccolors[j][i] == 0) {
+               alphaenc2[4*j + i] = 6;
+               alphadist = 0;
+            }
+            else if (srccolors[j][i] == 255) {
+               alphaenc2[4*j + i] = 7;
+               alphadist = 0;
+            }
+            else if (srccolors[j][i] <= acutValues[0]) {
+               alphaenc2[4*j + i] = 0;
+               alphadist = srccolors[j][i] - alphabase[0];
+            }
+            else if (srccolors[j][i] <= acutValues[1]) {
+               alphaenc2[4*j + i] = 2;
+               alphadist = srccolors[j][i] - (alphabase[0] * 4 + alphabase[1] * 1) / 5;
+            }
+            else if (srccolors[j][i] <= acutValues[2]) {
+               alphaenc2[4*j + i] = 3;
+               alphadist = srccolors[j][i] - (alphabase[0] * 3 + alphabase[1] * 2) / 5;
+            }
+            else if (srccolors[j][i] <= acutValues[3]) {
+               alphaenc2[4*j + i] = 4;
+               alphadist = srccolors[j][i] - (alphabase[0] * 2 + alphabase[1] * 3) / 5;
+            }
+            else if (srccolors[j][i] <= acutValues[4]) {
+               alphaenc2[4*j + i] = 5;
+               alphadist = srccolors[j][i] - (alphabase[0] * 1 + alphabase[1] * 4) / 5;
+            }
+            else {
+               alphaenc2[4*j + i] = 1;
+               alphadist = srccolors[j][i] - alphabase[1];
+            }
+            alphablockerror2 += alphadist * alphadist;
+         }
+      }
+
+
+      /* skip this if the error is already very small
+         this encoding is MUCH better on average than #2 though, but expensive! */
+      if ((alphablockerror2 > 96) && (alphablockerror1 > 96)) {
+         GLshort blockerrlin1 = 0;
+         GLshort blockerrlin2 = 0;
+         GLubyte nralphainrangelow = 0;
+         GLubyte nralphainrangehigh = 0;
+         alphatest[0] = 0xff;
+         alphatest[1] = 0x0;
+         /* if we have large range it's likely there are values close to 0/255, try to map them to 0/255 */
+         for (j = 0; j < numypixels; j++) {
+            for (i = 0; i < numxpixels; i++) {
+               if ((srccolors[j][i] > alphatest[1]) && (srccolors[j][i] < (255 -(alphabase[1] - alphabase[0]) / 28)))
+                  alphatest[1] = srccolors[j][i];
+               if ((srccolors[j][i] < alphatest[0]) && (srccolors[j][i] > (alphabase[1] - alphabase[0]) / 28))
+                  alphatest[0] = srccolors[j][i];
+            }
+         }
+          /* shouldn't happen too often, don't really care about those degenerated cases */
+          if (alphatest[1] <= alphatest[0]) {
+             alphatest[0] = 1;
+             alphatest[1] = 254;
+         }
+         for (aindex = 0; aindex < 5; aindex++) {
+         /* don't forget here is always rounded down */
+            acutValues[aindex] = (alphatest[0] * (10 - (2*aindex + 1)) + alphatest[1] * (2*aindex + 1)) / 10;
+         }
+
+         /* find the "average" difference between the alpha values and the next encoded value.
+            This is then used to calculate new base values.
+            Should there be some weighting, i.e. those values closer to alphatest[x] have more weight,
+            since they will see more improvement, and also because the values in the middle are somewhat
+            likely to get no improvement at all (because the base values might move in different directions)?
+            OTOH it would mean the values in the middle are even less likely to get an improvement
+         */
+         for (j = 0; j < numypixels; j++) {
+            for (i = 0; i < numxpixels; i++) {
+               if (srccolors[j][i] <= alphatest[0] / 2) {
+               }
+               else if (srccolors[j][i] > ((255 + alphatest[1]) / 2)) {
+               }
+               else if (srccolors[j][i] <= acutValues[0]) {
+                  blockerrlin1 += (srccolors[j][i] - alphatest[0]);
+                  nralphainrangelow += 1;
+               }
+               else if (srccolors[j][i] <= acutValues[1]) {
+                  blockerrlin1 += (srccolors[j][i] - (alphatest[0] * 4 + alphatest[1] * 1) / 5);
+                  blockerrlin2 += (srccolors[j][i] - (alphatest[0] * 4 + alphatest[1] * 1) / 5);
+                  nralphainrangelow += 1;
+                  nralphainrangehigh += 1;
+               }
+               else if (srccolors[j][i] <= acutValues[2]) {
+                  blockerrlin1 += (srccolors[j][i] - (alphatest[0] * 3 + alphatest[1] * 2) / 5);
+                  blockerrlin2 += (srccolors[j][i] - (alphatest[0] * 3 + alphatest[1] * 2) / 5);
+                  nralphainrangelow += 1;
+                  nralphainrangehigh += 1;
+               }
+               else if (srccolors[j][i] <= acutValues[3]) {
+                  blockerrlin1 += (srccolors[j][i] - (alphatest[0] * 2 + alphatest[1] * 3) / 5);
+                  blockerrlin2 += (srccolors[j][i] - (alphatest[0] * 2 + alphatest[1] * 3) / 5);
+                  nralphainrangelow += 1;
+                  nralphainrangehigh += 1;
+               }
+               else if (srccolors[j][i] <= acutValues[4]) {
+                  blockerrlin1 += (srccolors[j][i] - (alphatest[0] * 1 + alphatest[1] * 4) / 5);
+                  blockerrlin2 += (srccolors[j][i] - (alphatest[0] * 1 + alphatest[1] * 4) / 5);
+                  nralphainrangelow += 1;
+                  nralphainrangehigh += 1;
+                  }
+               else {
+                  blockerrlin2 += (srccolors[j][i] - alphatest[1]);
+                  nralphainrangehigh += 1;
+               }
+            }
+         }
+         /* shouldn't happen often, needed to avoid div by zero */
+         if (nralphainrangelow == 0) nralphainrangelow = 1;
+         if (nralphainrangehigh == 0) nralphainrangehigh = 1;
+         alphatest[0] = alphatest[0] + (blockerrlin1 / nralphainrangelow);
+#if RGTC_DEBUG
+         fprintf(stderr, "block err lin low %d, nr %d\n", blockerrlin1, nralphainrangelow);
+         fprintf(stderr, "block err lin high %d, nr %d\n", blockerrlin2, nralphainrangehigh);
+#endif
+         /* again shouldn't really happen often... */
+         if (alphatest[0] < 0) {
+            alphatest[0] = 0;
+         }
+         alphatest[1] = alphatest[1] + (blockerrlin2 / nralphainrangehigh);
+         if (alphatest[1] > 255) {
+            alphatest[1] = 255;
+         }
+
+         alphablockerror3 = 0;
+         for (aindex = 0; aindex < 5; aindex++) {
+         /* don't forget here is always rounded down */
+            acutValues[aindex] = (alphatest[0] * (10 - (2*aindex + 1)) + alphatest[1] * (2*aindex + 1)) / 10;
+         }
+         for (j = 0; j < numypixels; j++) {
+            for (i = 0; i < numxpixels; i++) {
+                /* maybe it's overkill to have the most complicated calculation just for the error
+                  calculation which we only need to figure out if encoding1 or encoding2 is better... */
+               if (srccolors[j][i] <= alphatest[0] / 2) {
+                  alphaenc3[4*j + i] = 6;
+                  alphadist = srccolors[j][i];
+               }
+               else if (srccolors[j][i] > ((255 + alphatest[1]) / 2)) {
+                  alphaenc3[4*j + i] = 7;
+                  alphadist = 255 - srccolors[j][i];
+               }
+               else if (srccolors[j][i] <= acutValues[0]) {
+                  alphaenc3[4*j + i] = 0;
+                  alphadist = srccolors[j][i] - alphatest[0];
+               }
+               else if (srccolors[j][i] <= acutValues[1]) {
+                 alphaenc3[4*j + i] = 2;
+                 alphadist = srccolors[j][i] - (alphatest[0] * 4 + alphatest[1] * 1) / 5;
+               }
+               else if (srccolors[j][i] <= acutValues[2]) {
+                  alphaenc3[4*j + i] = 3;
+                  alphadist = srccolors[j][i] - (alphatest[0] * 3 + alphatest[1] * 2) / 5;
+               }
+               else if (srccolors[j][i] <= acutValues[3]) {
+                  alphaenc3[4*j + i] = 4;
+                  alphadist = srccolors[j][i] - (alphatest[0] * 2 + alphatest[1] * 3) / 5;
+               }
+               else if (srccolors[j][i] <= acutValues[4]) {
+                  alphaenc3[4*j + i] = 5;
+                  alphadist = srccolors[j][i] - (alphatest[0] * 1 + alphatest[1] * 4) / 5;
+               }
+               else {
+                  alphaenc3[4*j + i] = 1;
+                  alphadist = srccolors[j][i] - alphatest[1];
+               }
+               alphablockerror3 += alphadist * alphadist;
+            }
+         }
+      }
+   }
+  /* write the alpha values and encoding back. */
+   if ((alphablockerror1 <= alphablockerror2) && (alphablockerror1 <= alphablockerror3)) {
+#if RGTC_DEBUG
+      if (alphablockerror1 > 96) fprintf(stderr, "enc1 used, error %d\n", alphablockerror1);
+#endif
+      write_rgtc_encoded_channel( blkaddr, alphause[1], alphause[0], alphaenc1 );
+   }
+   else if (alphablockerror2 <= alphablockerror3) {
+#if RGTC_DEBUG
+      if (alphablockerror2 > 96) fprintf(stderr, "enc2 used, error %d\n", alphablockerror2);
+#endif
+      write_rgtc_encoded_channel( blkaddr, alphabase[0], alphabase[1], alphaenc2 );
+   }
+   else {
+#if RGTC_DEBUG
+      fprintf(stderr, "enc3 used, error %d\n", alphablockerror3);
+#endif
+      write_rgtc_encoded_channel( blkaddr, (GLubyte)alphatest[0], (GLubyte)alphatest[1], alphaenc3 );
+   }
+}
+
+
+static void write_rgtc_encoded_channel_s(GLbyte *blkaddr,
+					 GLbyte alphabase1,
+					 GLbyte alphabase2,
+					 GLbyte alphaenc[16])
+{
+   *blkaddr++ = alphabase1;
+   *blkaddr++ = alphabase2;
+   *blkaddr++ = alphaenc[0] | (alphaenc[1] << 3) | ((alphaenc[2] & 3) << 6);
+   *blkaddr++ = (alphaenc[2] >> 2) | (alphaenc[3] << 1) | (alphaenc[4] << 4) | ((alphaenc[5] & 1) << 7);
+   *blkaddr++ = (alphaenc[5] >> 1) | (alphaenc[6] << 2) | (alphaenc[7] << 5);
+   *blkaddr++ = alphaenc[8] | (alphaenc[9] << 3) | ((alphaenc[10] & 3) << 6);
+   *blkaddr++ = (alphaenc[10] >> 2) | (alphaenc[11] << 1) | (alphaenc[12] << 4) | ((alphaenc[13] & 1) << 7);
+   *blkaddr++ = (alphaenc[13] >> 1) | (alphaenc[14] << 2) | (alphaenc[15] << 5);
+}
+
+static void encode_rgtc_chan_s(GLbyte *blkaddr, GLbyte srccolors[4][4],
+			       GLint numxpixels, GLint numypixels)
+{
+   GLbyte alphabase[2], alphause[2];
+   GLshort alphatest[2] = { 0 };
+   GLuint alphablockerror1, alphablockerror2, alphablockerror3;
+   GLbyte i, j, aindex, acutValues[7];
+   GLbyte alphaenc1[16], alphaenc2[16], alphaenc3[16];
+   GLboolean alphaabsmin = GL_FALSE;
+   GLboolean alphaabsmax = GL_FALSE;
+   GLshort alphadist;
+
+   /* find lowest and highest alpha value in block, alphabase[0] lowest, alphabase[1] highest */
+   alphabase[0] = 0xff; alphabase[1] = 0x0;
+   for (j = 0; j < numypixels; j++) {
+      for (i = 0; i < numxpixels; i++) {
+         if (srccolors[j][i] == 0)
+            alphaabsmin = GL_TRUE;
+         else if (srccolors[j][i] == 255)
+            alphaabsmax = GL_TRUE;
+         else {
+            if (srccolors[j][i] > alphabase[1])
+               alphabase[1] = srccolors[j][i];
+            if (srccolors[j][i] < alphabase[0])
+               alphabase[0] = srccolors[j][i];
+         }
+      }
+   }
+
+
+   if ((alphabase[0] > alphabase[1]) && !(alphaabsmin && alphaabsmax)) { /* one color, either max or min */
+      /* shortcut here since it is a very common case (and also avoids later problems) */
+      /* || (alphabase[0] == alphabase[1] && !alphaabsmin && !alphaabsmax) */
+      /* could also thest for alpha0 == alpha1 (and not min/max), but probably not common, so don't bother */
+
+      *blkaddr++ = srccolors[0][0];
+      blkaddr++;
+      *blkaddr++ = 0;
+      *blkaddr++ = 0;
+      *blkaddr++ = 0;
+      *blkaddr++ = 0;
+      *blkaddr++ = 0;
+      *blkaddr++ = 0;
+#if RGTC_DEBUG
+      fprintf(stderr, "enc0 used\n");
+#endif
+      return;
+   }
+
+   /* find best encoding for alpha0 > alpha1 */
+   /* it's possible this encoding is better even if both alphaabsmin and alphaabsmax are true */
+   alphablockerror1 = 0x0;
+   alphablockerror2 = 0xffffffff;
+   alphablockerror3 = 0xffffffff;
+   if (alphaabsmin) alphause[0] = 0;
+   else alphause[0] = alphabase[0];
+   if (alphaabsmax) alphause[1] = 255;
+   else alphause[1] = alphabase[1];
+   /* calculate the 7 cut values, just the middle between 2 of the computed alpha values */
+   for (aindex = 0; aindex < 7; aindex++) {
+      /* don't forget here is always rounded down */
+      acutValues[aindex] = (alphause[0] * (2*aindex + 1) + alphause[1] * (14 - (2*aindex + 1))) / 14;
+   }
+
+   for (j = 0; j < numypixels; j++) {
+      for (i = 0; i < numxpixels; i++) {
+         /* maybe it's overkill to have the most complicated calculation just for the error
+            calculation which we only need to figure out if encoding1 or encoding2 is better... */
+         if (srccolors[j][i] > acutValues[0]) {
+            alphaenc1[4*j + i] = 0;
+            alphadist = srccolors[j][i] - alphause[1];
+         }
+         else if (srccolors[j][i] > acutValues[1]) {
+            alphaenc1[4*j + i] = 2;
+            alphadist = srccolors[j][i] - (alphause[1] * 6 + alphause[0] * 1) / 7;
+         }
+         else if (srccolors[j][i] > acutValues[2]) {
+            alphaenc1[4*j + i] = 3;
+            alphadist = srccolors[j][i] - (alphause[1] * 5 + alphause[0] * 2) / 7;
+         }
+         else if (srccolors[j][i] > acutValues[3]) {
+            alphaenc1[4*j + i] = 4;
+            alphadist = srccolors[j][i] - (alphause[1] * 4 + alphause[0] * 3) / 7;
+         }
+         else if (srccolors[j][i] > acutValues[4]) {
+            alphaenc1[4*j + i] = 5;
+            alphadist = srccolors[j][i] - (alphause[1] * 3 + alphause[0] * 4) / 7;
+         }
+         else if (srccolors[j][i] > acutValues[5]) {
+            alphaenc1[4*j + i] = 6;
+            alphadist = srccolors[j][i] - (alphause[1] * 2 + alphause[0] * 5) / 7;
+         }
+         else if (srccolors[j][i] > acutValues[6]) {
+            alphaenc1[4*j + i] = 7;
+            alphadist = srccolors[j][i] - (alphause[1] * 1 + alphause[0] * 6) / 7;
+         }
+         else {
+            alphaenc1[4*j + i] = 1;
+            alphadist = srccolors[j][i] - alphause[0];
+         }
+         alphablockerror1 += alphadist * alphadist;
+      }
+   }
+#if RGTC_DEBUG
+   for (i = 0; i < 16; i++) {
+      fprintf(stderr, "%d ", alphaenc1[i]);
+   }
+   fprintf(stderr, "cutVals ");
+   for (i = 0; i < 8; i++) {
+      fprintf(stderr, "%d ", acutValues[i]);
+   }
+   fprintf(stderr, "srcVals ");
+   for (j = 0; j < numypixels; j++)
+      for (i = 0; i < numxpixels; i++) {
+	 fprintf(stderr, "%d ", srccolors[j][i]);
+      }
+   
+   fprintf(stderr, "\n");
+#endif
+
+   /* it's not very likely this encoding is better if both alphaabsmin and alphaabsmax
+      are false but try it anyway */
+   if (alphablockerror1 >= 32) {
+
+      /* don't bother if encoding is already very good, this condition should also imply
+      we have valid alphabase colors which we absolutely need (alphabase[0] <= alphabase[1]) */
+      alphablockerror2 = 0;
+      for (aindex = 0; aindex < 5; aindex++) {
+         /* don't forget here is always rounded down */
+         acutValues[aindex] = (alphabase[0] * (10 - (2*aindex + 1)) + alphabase[1] * (2*aindex + 1)) / 10;
+      }
+      for (j = 0; j < numypixels; j++) {
+         for (i = 0; i < numxpixels; i++) {
+             /* maybe it's overkill to have the most complicated calculation just for the error
+               calculation which we only need to figure out if encoding1 or encoding2 is better... */
+            if (srccolors[j][i] == 0) {
+               alphaenc2[4*j + i] = 6;
+               alphadist = 0;
+            }
+            else if (srccolors[j][i] == 255) {
+               alphaenc2[4*j + i] = 7;
+               alphadist = 0;
+            }
+            else if (srccolors[j][i] <= acutValues[0]) {
+               alphaenc2[4*j + i] = 0;
+               alphadist = srccolors[j][i] - alphabase[0];
+            }
+            else if (srccolors[j][i] <= acutValues[1]) {
+               alphaenc2[4*j + i] = 2;
+               alphadist = srccolors[j][i] - (alphabase[0] * 4 + alphabase[1] * 1) / 5;
+            }
+            else if (srccolors[j][i] <= acutValues[2]) {
+               alphaenc2[4*j + i] = 3;
+               alphadist = srccolors[j][i] - (alphabase[0] * 3 + alphabase[1] * 2) / 5;
+            }
+            else if (srccolors[j][i] <= acutValues[3]) {
+               alphaenc2[4*j + i] = 4;
+               alphadist = srccolors[j][i] - (alphabase[0] * 2 + alphabase[1] * 3) / 5;
+            }
+            else if (srccolors[j][i] <= acutValues[4]) {
+               alphaenc2[4*j + i] = 5;
+               alphadist = srccolors[j][i] - (alphabase[0] * 1 + alphabase[1] * 4) / 5;
+            }
+            else {
+               alphaenc2[4*j + i] = 1;
+               alphadist = srccolors[j][i] - alphabase[1];
+            }
+            alphablockerror2 += alphadist * alphadist;
+         }
+      }
+
+
+      /* skip this if the error is already very small
+         this encoding is MUCH better on average than #2 though, but expensive! */
+      if ((alphablockerror2 > 96) && (alphablockerror1 > 96)) {
+         GLshort blockerrlin1 = 0;
+         GLshort blockerrlin2 = 0;
+         GLubyte nralphainrangelow = 0;
+         GLubyte nralphainrangehigh = 0;
+         alphatest[0] = 0xff;
+         alphatest[1] = 0x0;
+         /* if we have large range it's likely there are values close to 0/255, try to map them to 0/255 */
+         for (j = 0; j < numypixels; j++) {
+            for (i = 0; i < numxpixels; i++) {
+               if ((srccolors[j][i] > alphatest[1]) && (srccolors[j][i] < (255 -(alphabase[1] - alphabase[0]) / 28)))
+                  alphatest[1] = srccolors[j][i];
+               if ((srccolors[j][i] < alphatest[0]) && (srccolors[j][i] > (alphabase[1] - alphabase[0]) / 28))
+                  alphatest[0] = srccolors[j][i];
+            }
+         }
+          /* shouldn't happen too often, don't really care about those degenerated cases */
+          if (alphatest[1] <= alphatest[0]) {
+             alphatest[0] = 1;
+             alphatest[1] = 254;
+         }
+         for (aindex = 0; aindex < 5; aindex++) {
+         /* don't forget here is always rounded down */
+            acutValues[aindex] = (alphatest[0] * (10 - (2*aindex + 1)) + alphatest[1] * (2*aindex + 1)) / 10;
+         }
+
+         /* find the "average" difference between the alpha values and the next encoded value.
+            This is then used to calculate new base values.
+            Should there be some weighting, i.e. those values closer to alphatest[x] have more weight,
+            since they will see more improvement, and also because the values in the middle are somewhat
+            likely to get no improvement at all (because the base values might move in different directions)?
+            OTOH it would mean the values in the middle are even less likely to get an improvement
+         */
+         for (j = 0; j < numypixels; j++) {
+            for (i = 0; i < numxpixels; i++) {
+               if (srccolors[j][i] <= alphatest[0] / 2) {
+               }
+               else if (srccolors[j][i] > ((255 + alphatest[1]) / 2)) {
+               }
+               else if (srccolors[j][i] <= acutValues[0]) {
+                  blockerrlin1 += (srccolors[j][i] - alphatest[0]);
+                  nralphainrangelow += 1;
+               }
+               else if (srccolors[j][i] <= acutValues[1]) {
+                  blockerrlin1 += (srccolors[j][i] - (alphatest[0] * 4 + alphatest[1] * 1) / 5);
+                  blockerrlin2 += (srccolors[j][i] - (alphatest[0] * 4 + alphatest[1] * 1) / 5);
+                  nralphainrangelow += 1;
+                  nralphainrangehigh += 1;
+               }
+               else if (srccolors[j][i] <= acutValues[2]) {
+                  blockerrlin1 += (srccolors[j][i] - (alphatest[0] * 3 + alphatest[1] * 2) / 5);
+                  blockerrlin2 += (srccolors[j][i] - (alphatest[0] * 3 + alphatest[1] * 2) / 5);
+                  nralphainrangelow += 1;
+                  nralphainrangehigh += 1;
+               }
+               else if (srccolors[j][i] <= acutValues[3]) {
+                  blockerrlin1 += (srccolors[j][i] - (alphatest[0] * 2 + alphatest[1] * 3) / 5);
+                  blockerrlin2 += (srccolors[j][i] - (alphatest[0] * 2 + alphatest[1] * 3) / 5);
+                  nralphainrangelow += 1;
+                  nralphainrangehigh += 1;
+               }
+               else if (srccolors[j][i] <= acutValues[4]) {
+                  blockerrlin1 += (srccolors[j][i] - (alphatest[0] * 1 + alphatest[1] * 4) / 5);
+                  blockerrlin2 += (srccolors[j][i] - (alphatest[0] * 1 + alphatest[1] * 4) / 5);
+                  nralphainrangelow += 1;
+                  nralphainrangehigh += 1;
+                  }
+               else {
+                  blockerrlin2 += (srccolors[j][i] - alphatest[1]);
+                  nralphainrangehigh += 1;
+               }
+            }
+         }
+         /* shouldn't happen often, needed to avoid div by zero */
+         if (nralphainrangelow == 0) nralphainrangelow = 1;
+         if (nralphainrangehigh == 0) nralphainrangehigh = 1;
+         alphatest[0] = alphatest[0] + (blockerrlin1 / nralphainrangelow);
+#if RGTC_DEBUG
+	 fprintf(stderr, "block err lin low %d, nr %d\n", blockerrlin1, nralphainrangelow);
+         fprintf(stderr, "block err lin high %d, nr %d\n", blockerrlin2, nralphainrangehigh);
+#endif
+         /* again shouldn't really happen often... */
+         if (alphatest[0] < 0) {
+            alphatest[0] = 0;
+         }
+         alphatest[1] = alphatest[1] + (blockerrlin2 / nralphainrangehigh);
+         if (alphatest[1] > 255) {
+            alphatest[1] = 255;
+         }
+
+         alphablockerror3 = 0;
+         for (aindex = 0; aindex < 5; aindex++) {
+         /* don't forget here is always rounded down */
+            acutValues[aindex] = (alphatest[0] * (10 - (2*aindex + 1)) + alphatest[1] * (2*aindex + 1)) / 10;
+         }
+         for (j = 0; j < numypixels; j++) {
+            for (i = 0; i < numxpixels; i++) {
+                /* maybe it's overkill to have the most complicated calculation just for the error
+                  calculation which we only need to figure out if encoding1 or encoding2 is better... */
+               if (srccolors[j][i] <= alphatest[0] / 2) {
+                  alphaenc3[4*j + i] = 6;
+                  alphadist = srccolors[j][i];
+               }
+               else if (srccolors[j][i] > ((255 + alphatest[1]) / 2)) {
+                  alphaenc3[4*j + i] = 7;
+                  alphadist = 255 - srccolors[j][i];
+               }
+               else if (srccolors[j][i] <= acutValues[0]) {
+                  alphaenc3[4*j + i] = 0;
+                  alphadist = srccolors[j][i] - alphatest[0];
+               }
+               else if (srccolors[j][i] <= acutValues[1]) {
+                 alphaenc3[4*j + i] = 2;
+                 alphadist = srccolors[j][i] - (alphatest[0] * 4 + alphatest[1] * 1) / 5;
+               }
+               else if (srccolors[j][i] <= acutValues[2]) {
+                  alphaenc3[4*j + i] = 3;
+                  alphadist = srccolors[j][i] - (alphatest[0] * 3 + alphatest[1] * 2) / 5;
+               }
+               else if (srccolors[j][i] <= acutValues[3]) {
+                  alphaenc3[4*j + i] = 4;
+                  alphadist = srccolors[j][i] - (alphatest[0] * 2 + alphatest[1] * 3) / 5;
+               }
+               else if (srccolors[j][i] <= acutValues[4]) {
+                  alphaenc3[4*j + i] = 5;
+                  alphadist = srccolors[j][i] - (alphatest[0] * 1 + alphatest[1] * 4) / 5;
+               }
+               else {
+                  alphaenc3[4*j + i] = 1;
+                  alphadist = srccolors[j][i] - alphatest[1];
+               }
+               alphablockerror3 += alphadist * alphadist;
+            }
+         }
+      }
+   }
+  /* write the alpha values and encoding back. */
+   if ((alphablockerror1 <= alphablockerror2) && (alphablockerror1 <= alphablockerror3)) {
+#if RGTC_DEBUG
+      if (alphablockerror1 > 96) fprintf(stderr, "enc1 used, error %d\n", alphablockerror1);
+#endif
+      write_rgtc_encoded_channel_s( blkaddr, alphause[1], alphause[0], alphaenc1 );
+   }
+   else if (alphablockerror2 <= alphablockerror3) {
+#if RGTC_DEBUG
+      if (alphablockerror2 > 96) fprintf(stderr, "enc2 used, error %d\n", alphablockerror2);
+#endif
+      write_rgtc_encoded_channel_s( blkaddr, alphabase[0], alphabase[1], alphaenc2 );
+   }
+   else {
+#if RGTC_DEBUG
+      fprintf(stderr, "enc3 used, error %d\n", alphablockerror3);
+#endif
+      write_rgtc_encoded_channel_s( blkaddr, (GLubyte)alphatest[0], (GLubyte)alphatest[1], alphaenc3 );
+   }
+}
diff --git a/mesalib/src/mesa/main/texcompress_rgtc.h b/mesalib/src/mesa/main/texcompress_rgtc.h
new file mode 100644
index 000000000..424edc458
--- /dev/null
+++ b/mesalib/src/mesa/main/texcompress_rgtc.h
@@ -0,0 +1,60 @@
+/*
+ * Copyright (C) 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef TEXCOMPRESS_RGTC_H
+#define TEXCOMPRESS_RGTC_H
+
+#include "glheader.h"
+#include "mfeatures.h"
+#include "texstore.h"
+
+struct gl_texture_image;
+
+extern GLboolean
+_mesa_texstore_red_rgtc1(TEXSTORE_PARAMS);
+
+extern GLboolean
+_mesa_texstore_signed_red_rgtc1(TEXSTORE_PARAMS);
+
+extern GLboolean
+_mesa_texstore_rg_rgtc2(TEXSTORE_PARAMS);
+
+extern GLboolean
+_mesa_texstore_signed_rg_rgtc2(TEXSTORE_PARAMS);
+
+extern void
+_mesa_fetch_texel_2d_f_red_rgtc1(const struct gl_texture_image *texImage,
+				 GLint i, GLint j, GLint k, GLfloat *texel);
+
+extern void
+_mesa_fetch_texel_2d_f_signed_red_rgtc1(const struct gl_texture_image *texImage,
+					GLint i, GLint j, GLint k, GLfloat *texel);
+
+extern void
+_mesa_fetch_texel_2d_f_rg_rgtc2(const struct gl_texture_image *texImage,
+				 GLint i, GLint j, GLint k, GLfloat *texel);
+
+extern void
+_mesa_fetch_texel_2d_f_signed_rg_rgtc2(const struct gl_texture_image *texImage,
+				       GLint i, GLint j, GLint k, GLfloat *texel);
+#endif
diff --git a/mesalib/src/mesa/main/texfetch.c b/mesalib/src/mesa/main/texfetch.c
index 1e7ea9480..8666076ab 100644
--- a/mesalib/src/mesa/main/texfetch.c
+++ b/mesalib/src/mesa/main/texfetch.c
@@ -38,6 +38,7 @@
 #include "texcompress.h"
 #include "texcompress_fxt1.h"
 #include "texcompress_s3tc.h"
+#include "texcompress_rgtc.h"
 #include "texfetch.h"
 #include "teximage.h"
 
@@ -756,7 +757,35 @@ texfetch_funcs[MESA_FORMAT_COUNT] =
       fetch_texel_2d_rgba_16,
       fetch_texel_3d_rgba_16,
       store_texel_rgba_16
-   }
+   },
+   {
+      MESA_FORMAT_RED_RGTC1,
+      NULL,
+      _mesa_fetch_texel_2d_f_red_rgtc1,
+      NULL,
+      NULL
+   },
+   {
+      MESA_FORMAT_SIGNED_RED_RGTC1,
+      NULL,
+      _mesa_fetch_texel_2d_f_signed_red_rgtc1,
+      NULL,
+      NULL
+   },
+   {
+      MESA_FORMAT_RG_RGTC2,
+      NULL,
+      _mesa_fetch_texel_2d_f_rg_rgtc2,
+      NULL,
+      NULL
+   },
+   {
+      MESA_FORMAT_SIGNED_RG_RGTC2,
+      NULL,
+      _mesa_fetch_texel_2d_f_signed_rg_rgtc2,
+      NULL,
+      NULL
+   },
 };
 
 
diff --git a/mesalib/src/mesa/main/texformat.c b/mesalib/src/mesa/main/texformat.c
index 2542cea85..8040fac7d 100644
--- a/mesalib/src/mesa/main/texformat.c
+++ b/mesalib/src/mesa/main/texformat.c
@@ -1,608 +1,627 @@
-/*
- * Mesa 3-D graphics library
- * Version:  7.7
- *
- * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
- * Copyright (c) 2008-2009 VMware, Inc.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included
- * in all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
- * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
- * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- */
-
-
-/**
- * \file texformat.c
- * Texture formats.
- *
- * \author Gareth Hughes
- * \author Brian Paul
- */
-
-
-#include "context.h"
-#include "mfeatures.h"
-#include "mtypes.h"
-#include "texcompress.h"
-#include "texformat.h"
-
-#define RETURN_IF_SUPPORTED(f) do {		\
-   if (ctx->TextureFormatSupported[f])		\
-      return f;					\
-} while (0)
-
-/**
- * Choose an appropriate texture format given the format, type and
- * internalFormat parameters passed to glTexImage().
- *
- * \param ctx  the GL context.
- * \param internalFormat  user's prefered internal texture format.
- * \param format  incoming image pixel format.
- * \param type  incoming image data type.
- *
- * \return a pointer to a gl_texture_format object which describes the
- * choosen texture format, or NULL on failure.
- * 
- * This is called via dd_function_table::ChooseTextureFormat.  Hardware drivers
- * will typically override this function with a specialized version.
- */
-gl_format
-_mesa_choose_tex_format( struct gl_context *ctx, GLint internalFormat,
-                         GLenum format, GLenum type )
-{
-   (void) format;
-   (void) type;
-
-   switch (internalFormat) {
-      /* shallow RGBA formats */
-      case 4:
-      case GL_RGBA:
-	 if (type == GL_UNSIGNED_SHORT_4_4_4_4_REV) {
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB4444);
-	 } else if (type == GL_UNSIGNED_SHORT_1_5_5_5_REV) {
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB1555);
-	 }
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA8888);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
-	 break;
-
-      case GL_RGBA8:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA8888);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
-	 break;
-      case GL_RGB5_A1:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB1555);
-	 break;
-      case GL_RGBA2:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB4444_REV); /* just to test another format*/
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB4444);
-	 break;
-      case GL_RGBA4:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB4444);
-	 break;
-
-      /* deep RGBA formats */
-      case GL_RGB10_A2:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB2101010);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
-	 break;
-      case GL_RGBA12:
-      case GL_RGBA16:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_16);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_16);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA8888);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
-	 break;
-
-      /* shallow RGB formats */
-      case 3:
-      case GL_RGB:
-      case GL_RGB8:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB888);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_XRGB8888);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
-	 break;
-      case GL_R3_G3_B2:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB332);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB888);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_XRGB8888);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
-	 break;
-      case GL_RGB4:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB565_REV); /* just to test another format */
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB565);
-	 break;
-      case GL_RGB5:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB565);
-	 break;
-
-      /* deep RGB formats */
-      case GL_RGB10:
-      case GL_RGB12:
-      case GL_RGB16:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_16);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_XRGB8888);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
-	 break;
-
-      /* Alpha formats */
-      case GL_ALPHA:
-      case GL_ALPHA4:
-      case GL_ALPHA8:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_A8);
-	 break;
-
-      case GL_ALPHA12:
-      case GL_ALPHA16:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_A16);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_A8);
-	 break;
-
-      /* Luminance formats */
-      case 1:
-      case GL_LUMINANCE:
-      case GL_LUMINANCE4:
-      case GL_LUMINANCE8:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_L8);
-	 break;
-
-      case GL_LUMINANCE12:
-      case GL_LUMINANCE16:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_L16);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_L8);
-	 break;
-
-      /* Luminance/Alpha formats */
-      case GL_LUMINANCE4_ALPHA4:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_AL44);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_AL88);
-	 break;
-
-      case 2:
-      case GL_LUMINANCE_ALPHA:
-      case GL_LUMINANCE6_ALPHA2:
-      case GL_LUMINANCE8_ALPHA8:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_AL88);
-	 break;
-
-      case GL_LUMINANCE12_ALPHA4:
-      case GL_LUMINANCE12_ALPHA12:
-      case GL_LUMINANCE16_ALPHA16:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_AL1616);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_AL88);
-	 break;
-
-      case GL_INTENSITY:
-      case GL_INTENSITY4:
-      case GL_INTENSITY8:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_I8);
-	 break;
-
-      case GL_INTENSITY12:
-      case GL_INTENSITY16:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_I16);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_I8);
-	 break;
-
-      case GL_COLOR_INDEX:
-      case GL_COLOR_INDEX1_EXT:
-      case GL_COLOR_INDEX2_EXT:
-      case GL_COLOR_INDEX4_EXT:
-      case GL_COLOR_INDEX12_EXT:
-      case GL_COLOR_INDEX16_EXT:
-      case GL_COLOR_INDEX8_EXT:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_CI8);
-	 break;
-
-      default:
-         ; /* fallthrough */
-   }
-
-   if (ctx->Extensions.ARB_depth_texture) {
-      switch (internalFormat) {
-         case GL_DEPTH_COMPONENT:
-         case GL_DEPTH_COMPONENT24:
-         case GL_DEPTH_COMPONENT32:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_Z32);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_S8_Z24);
-	    break;
-         case GL_DEPTH_COMPONENT16:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_Z16);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_S8_Z24);
-         default:
-            ; /* fallthrough */
-      }
-   }
-
-   switch (internalFormat) {
-      case GL_COMPRESSED_ALPHA_ARB:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_A8);
-	 break;
-      case GL_COMPRESSED_LUMINANCE_ARB:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_L8);
-	 break;
-      case GL_COMPRESSED_LUMINANCE_ALPHA_ARB:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_AL88);
-	 break;
-      case GL_COMPRESSED_INTENSITY_ARB:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_I8);
-	 break;
-      case GL_COMPRESSED_RGB_ARB:
-         if (ctx->Extensions.EXT_texture_compression_s3tc ||
-             ctx->Extensions.S3_s3tc)
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_DXT1);
-         if (ctx->Extensions.TDFX_texture_compression_FXT1)
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_FXT1);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB888);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_XRGB8888);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
-	 break;
-      case GL_COMPRESSED_RGBA_ARB:
-         if (ctx->Extensions.EXT_texture_compression_s3tc ||
-             ctx->Extensions.S3_s3tc)
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_DXT3); /* Not rgba_dxt1, see spec */
-         if (ctx->Extensions.TDFX_texture_compression_FXT1)
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_FXT1);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA8888);
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
-	 break;
-      default:
-         ; /* fallthrough */
-   }
-
-   if (ctx->Extensions.MESA_ycbcr_texture) {
-      if (internalFormat == GL_YCBCR_MESA) {
-         if (type == GL_UNSIGNED_SHORT_8_8_MESA)
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_YCBCR);
-         else
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_YCBCR_REV);
-      }
-   }
-
-#if FEATURE_texture_fxt1
-   if (ctx->Extensions.TDFX_texture_compression_FXT1) {
-      switch (internalFormat) {
-         case GL_COMPRESSED_RGB_FXT1_3DFX:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_FXT1);
-	 break;
-         case GL_COMPRESSED_RGBA_FXT1_3DFX:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_FXT1);
-	 break;
-         default:
-            ; /* fallthrough */
-      }
-   }
-#endif
-
-#if FEATURE_texture_s3tc
-   if (ctx->Extensions.EXT_texture_compression_s3tc) {
-      switch (internalFormat) {
-         case GL_COMPRESSED_RGB_S3TC_DXT1_EXT:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_DXT1);
-	    break;
-         case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_DXT1);
-	    break;
-         case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_DXT3);
-	    break;
-         case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_DXT5);
-	    break;
-         default:
-            ; /* fallthrough */
-      }
-   }
-
-   if (ctx->Extensions.S3_s3tc) {
-      switch (internalFormat) {
-         case GL_RGB_S3TC:
-         case GL_RGB4_S3TC:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_DXT1);
-	    break;
-         case GL_RGBA_S3TC:
-         case GL_RGBA4_S3TC:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_DXT3);
-	    break;
-         default:
-            ; /* fallthrough */
-      }
-   }
-#endif
-
-   if (ctx->Extensions.ARB_texture_float) {
-      switch (internalFormat) {
-         case GL_ALPHA16F_ARB:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_ALPHA_FLOAT16);
-	    break;
-         case GL_ALPHA32F_ARB:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_ALPHA_FLOAT32);
-	    break;
-         case GL_LUMINANCE16F_ARB:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_LUMINANCE_FLOAT16);
-	    break;
-         case GL_LUMINANCE32F_ARB:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_LUMINANCE_FLOAT32);
-	    break;
-         case GL_LUMINANCE_ALPHA16F_ARB:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_LUMINANCE_ALPHA_FLOAT16);
-	    break;
-         case GL_LUMINANCE_ALPHA32F_ARB:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_LUMINANCE_ALPHA_FLOAT32);
-	    break;
-         case GL_INTENSITY16F_ARB:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_INTENSITY_FLOAT16);
-	    break;
-         case GL_INTENSITY32F_ARB:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_INTENSITY_FLOAT32);
-	    break;
-         case GL_RGB16F_ARB:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_FLOAT16);
-	    break;
-         case GL_RGB32F_ARB:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_FLOAT32);
-	    break;
-         case GL_RGBA16F_ARB:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_FLOAT16);
-	    break;
-         case GL_RGBA32F_ARB:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_FLOAT32);
-	    break;
-         default:
-            ; /* fallthrough */
-      }
-   }
-
-   if (ctx->Extensions.EXT_packed_depth_stencil) {
-      switch (internalFormat) {
-         case GL_DEPTH_STENCIL_EXT:
-         case GL_DEPTH24_STENCIL8_EXT:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_Z24_S8);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_S8_Z24);
-	    break;
-         default:
-            ; /* fallthrough */
-      }
-   }
-
-   if (ctx->Extensions.ATI_envmap_bumpmap) {
-      switch (internalFormat) {
-         case GL_DUDV_ATI:
-         case GL_DU8DV8_ATI:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_DUDV8);
-	    break;
-         default:
-            ; /* fallthrough */
-      }
-   }
-
-   if (ctx->Extensions.MESA_texture_signed_rgba) {
-      switch (internalFormat) {
-         case GL_RGBA_SNORM:
-         case GL_RGBA8_SNORM:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RGBA8888);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RGBA8888_REV);
-	    break;
-         default:
-            ; /* fallthrough */
-      }
-   }
-
-   if (ctx->VersionMajor * 10 + ctx->VersionMinor >= 31) {
-      switch (internalFormat) {
-         case GL_RED_SNORM:
-         case GL_R8_SNORM:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_R8);
-	    break;
-         case GL_RG_SNORM:
-         case GL_RG8_SNORM:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RG88);
-	    break;
-         case GL_RGB_SNORM:
-         case GL_RGB8_SNORM:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RGBX8888);
-	    break;
-         case GL_RGBA_SNORM:
-         case GL_RGBA8_SNORM:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RGBA8888);
-	    break;
-         case GL_R16_SNORM:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_R_16);
-	    break;
-         case GL_RG16_SNORM:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RG_16);
-	    break;
-         case GL_RGB16_SNORM:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RGB_16);
-	    break;
-         case GL_RGBA16_SNORM:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RGBA_16);
-	    break;
-         default:
-            ; /* fall-through */
-      }
-   }
-
-#if FEATURE_EXT_texture_sRGB
-   if (ctx->Extensions.EXT_texture_sRGB) {
-      switch (internalFormat) {
-         case GL_SRGB_EXT:
-         case GL_SRGB8_EXT:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SRGB8);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
-	    break;
-         case GL_SRGB_ALPHA_EXT:
-         case GL_SRGB8_ALPHA8_EXT:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SRGBA8);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
-	    break;
-         case GL_SLUMINANCE_EXT:
-         case GL_SLUMINANCE8_EXT:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SL8);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
-	    break;
-         case GL_SLUMINANCE_ALPHA_EXT:
-         case GL_SLUMINANCE8_ALPHA8_EXT:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SLA8);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
-	    break;
-         case GL_COMPRESSED_SLUMINANCE_EXT:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SL8);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
-	    break;
-         case GL_COMPRESSED_SLUMINANCE_ALPHA_EXT:
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SLA8);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
-	    break;
-         case GL_COMPRESSED_SRGB_EXT:
-#if FEATURE_texture_s3tc
-            if (ctx->Extensions.EXT_texture_compression_s3tc)
-	       RETURN_IF_SUPPORTED(MESA_FORMAT_SRGB_DXT1);
-#endif
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SRGB8);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
-	    break;
-         case GL_COMPRESSED_SRGB_ALPHA_EXT:
-#if FEATURE_texture_s3tc
-            if (ctx->Extensions.EXT_texture_compression_s3tc)
-	       RETURN_IF_SUPPORTED(MESA_FORMAT_SRGBA_DXT3); /* Not srgba_dxt1, see spec */
-#endif
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SRGBA8);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
-	    break;
-#if FEATURE_texture_s3tc
-         case GL_COMPRESSED_SRGB_S3TC_DXT1_EXT:
-            if (ctx->Extensions.EXT_texture_compression_s3tc)
-	       RETURN_IF_SUPPORTED(MESA_FORMAT_SRGB_DXT1);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
-            break;
-         case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT:
-            if (ctx->Extensions.EXT_texture_compression_s3tc)
-	       RETURN_IF_SUPPORTED(MESA_FORMAT_SRGBA_DXT1);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
-            break;
-         case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT:
-            if (ctx->Extensions.EXT_texture_compression_s3tc)
-	       RETURN_IF_SUPPORTED(MESA_FORMAT_SRGBA_DXT3);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
-            break;
-         case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT:
-            if (ctx->Extensions.EXT_texture_compression_s3tc)
-	       RETURN_IF_SUPPORTED(MESA_FORMAT_SRGBA_DXT5);
-	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
-            break;
-#endif
-         default:
-            ; /* fallthrough */
-      }
-   }
-#endif /* FEATURE_EXT_texture_sRGB */
-
-   if (ctx->Extensions.EXT_texture_integer) {
-      switch (internalFormat) {
-      case GL_RGBA32UI_EXT:
-      case GL_RGB32UI_EXT:
-      case GL_ALPHA32UI_EXT:
-      case GL_INTENSITY32UI_EXT:
-      case GL_LUMINANCE32UI_EXT:
-      case GL_LUMINANCE_ALPHA32UI_EXT:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_UINT32);
-	 break;
-      case GL_RGBA16UI_EXT:
-      case GL_RGB16UI_EXT:
-      case GL_ALPHA16UI_EXT:
-      case GL_INTENSITY16UI_EXT:
-      case GL_LUMINANCE16UI_EXT:
-      case GL_LUMINANCE_ALPHA16UI_EXT:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_UINT16);
-	 break;
-      case GL_RGBA8UI_EXT:
-      case GL_RGB8UI_EXT:
-      case GL_ALPHA8UI_EXT:
-      case GL_INTENSITY8UI_EXT:
-      case GL_LUMINANCE8UI_EXT:
-      case GL_LUMINANCE_ALPHA8UI_EXT:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_UINT8);
-	 break;
-      case GL_RGBA32I_EXT:
-      case GL_RGB32I_EXT:
-      case GL_ALPHA32I_EXT:
-      case GL_INTENSITY32I_EXT:
-      case GL_LUMINANCE32I_EXT:
-      case GL_LUMINANCE_ALPHA32I_EXT:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_INT32);
-	 break;
-      case GL_RGBA16I_EXT:
-      case GL_RGB16I_EXT:
-      case GL_ALPHA16I_EXT:
-      case GL_INTENSITY16I_EXT:
-      case GL_LUMINANCE16I_EXT:
-      case GL_LUMINANCE_ALPHA16I_EXT:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_INT16);
-	 break;
-      case GL_RGBA8I_EXT:
-      case GL_RGB8I_EXT:
-      case GL_ALPHA8I_EXT:
-      case GL_INTENSITY8I_EXT:
-      case GL_LUMINANCE8I_EXT:
-      case GL_LUMINANCE_ALPHA8I_EXT:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_INT8);
-	 break;
-      }
-   }
-
-   if (ctx->Extensions.ARB_texture_rg) {
-      switch (internalFormat) {
-      case GL_R8:
-      case GL_RED:
-      case GL_COMPRESSED_RED:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_R8);
-	 break;
-
-      case GL_R16:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_R16);
-	 break;
-
-      case GL_RG:
-      case GL_RG8:
-      case GL_COMPRESSED_RG:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RG88);
-	 break;
-
-      case GL_RG16:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_RG1616);
-	 break;
-
-      default:
-         ; /* fallthrough */
-      }
-   }
-
-   if (ctx->Extensions.EXT_texture_format_BGRA8888) {
-      switch (internalFormat) {
-      case GL_BGRA:
-	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
-	 break;
-
-      default:
-         ; /* fallthrough */
-      }
-   }
-
-   _mesa_problem(ctx, "unexpected format in _mesa_choose_tex_format()");
-   return MESA_FORMAT_NONE;
-}
-
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.7
+ *
+ * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
+ * Copyright (c) 2008-2009 VMware, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+/**
+ * \file texformat.c
+ * Texture formats.
+ *
+ * \author Gareth Hughes
+ * \author Brian Paul
+ */
+
+
+#include "context.h"
+#include "mfeatures.h"
+#include "mtypes.h"
+#include "texcompress.h"
+#include "texformat.h"
+
+#define RETURN_IF_SUPPORTED(f) do {		\
+   if (ctx->TextureFormatSupported[f])		\
+      return f;					\
+} while (0)
+
+/**
+ * Choose an appropriate texture format given the format, type and
+ * internalFormat parameters passed to glTexImage().
+ *
+ * \param ctx  the GL context.
+ * \param internalFormat  user's prefered internal texture format.
+ * \param format  incoming image pixel format.
+ * \param type  incoming image data type.
+ *
+ * \return a pointer to a gl_texture_format object which describes the
+ * choosen texture format, or NULL on failure.
+ * 
+ * This is called via dd_function_table::ChooseTextureFormat.  Hardware drivers
+ * will typically override this function with a specialized version.
+ */
+gl_format
+_mesa_choose_tex_format( struct gl_context *ctx, GLint internalFormat,
+                         GLenum format, GLenum type )
+{
+   (void) format;
+   (void) type;
+
+   switch (internalFormat) {
+      /* shallow RGBA formats */
+      case 4:
+      case GL_RGBA:
+	 if (type == GL_UNSIGNED_SHORT_4_4_4_4_REV) {
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB4444);
+	 } else if (type == GL_UNSIGNED_SHORT_1_5_5_5_REV) {
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB1555);
+	 }
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA8888);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
+	 break;
+
+      case GL_RGBA8:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA8888);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
+	 break;
+      case GL_RGB5_A1:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB1555);
+	 break;
+      case GL_RGBA2:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB4444_REV); /* just to test another format*/
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB4444);
+	 break;
+      case GL_RGBA4:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB4444);
+	 break;
+
+      /* deep RGBA formats */
+      case GL_RGB10_A2:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB2101010);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
+	 break;
+      case GL_RGBA12:
+      case GL_RGBA16:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_16);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_16);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA8888);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
+	 break;
+
+      /* shallow RGB formats */
+      case 3:
+      case GL_RGB:
+      case GL_RGB8:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB888);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_XRGB8888);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
+	 break;
+      case GL_R3_G3_B2:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB332);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB888);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_XRGB8888);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
+	 break;
+      case GL_RGB4:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB565_REV); /* just to test another format */
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB565);
+	 break;
+      case GL_RGB5:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB565);
+	 break;
+
+      /* deep RGB formats */
+      case GL_RGB10:
+      case GL_RGB12:
+      case GL_RGB16:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_16);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_XRGB8888);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
+	 break;
+
+      /* Alpha formats */
+      case GL_ALPHA:
+      case GL_ALPHA4:
+      case GL_ALPHA8:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_A8);
+	 break;
+
+      case GL_ALPHA12:
+      case GL_ALPHA16:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_A16);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_A8);
+	 break;
+
+      /* Luminance formats */
+      case 1:
+      case GL_LUMINANCE:
+      case GL_LUMINANCE4:
+      case GL_LUMINANCE8:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_L8);
+	 break;
+
+      case GL_LUMINANCE12:
+      case GL_LUMINANCE16:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_L16);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_L8);
+	 break;
+
+      /* Luminance/Alpha formats */
+      case GL_LUMINANCE4_ALPHA4:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_AL44);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_AL88);
+	 break;
+
+      case 2:
+      case GL_LUMINANCE_ALPHA:
+      case GL_LUMINANCE6_ALPHA2:
+      case GL_LUMINANCE8_ALPHA8:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_AL88);
+	 break;
+
+      case GL_LUMINANCE12_ALPHA4:
+      case GL_LUMINANCE12_ALPHA12:
+      case GL_LUMINANCE16_ALPHA16:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_AL1616);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_AL88);
+	 break;
+
+      case GL_INTENSITY:
+      case GL_INTENSITY4:
+      case GL_INTENSITY8:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_I8);
+	 break;
+
+      case GL_INTENSITY12:
+      case GL_INTENSITY16:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_I16);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_I8);
+	 break;
+
+      case GL_COLOR_INDEX:
+      case GL_COLOR_INDEX1_EXT:
+      case GL_COLOR_INDEX2_EXT:
+      case GL_COLOR_INDEX4_EXT:
+      case GL_COLOR_INDEX12_EXT:
+      case GL_COLOR_INDEX16_EXT:
+      case GL_COLOR_INDEX8_EXT:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_CI8);
+	 break;
+
+      default:
+         ; /* fallthrough */
+   }
+
+   if (ctx->Extensions.ARB_depth_texture) {
+      switch (internalFormat) {
+         case GL_DEPTH_COMPONENT:
+         case GL_DEPTH_COMPONENT24:
+         case GL_DEPTH_COMPONENT32:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_Z32);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_S8_Z24);
+	    break;
+         case GL_DEPTH_COMPONENT16:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_Z16);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_S8_Z24);
+         default:
+            ; /* fallthrough */
+      }
+   }
+
+   switch (internalFormat) {
+      case GL_COMPRESSED_ALPHA_ARB:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_A8);
+	 break;
+      case GL_COMPRESSED_LUMINANCE_ARB:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_L8);
+	 break;
+      case GL_COMPRESSED_LUMINANCE_ALPHA_ARB:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_AL88);
+	 break;
+      case GL_COMPRESSED_INTENSITY_ARB:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_I8);
+	 break;
+      case GL_COMPRESSED_RGB_ARB:
+         if (ctx->Extensions.EXT_texture_compression_s3tc ||
+             ctx->Extensions.S3_s3tc)
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_DXT1);
+         if (ctx->Extensions.TDFX_texture_compression_FXT1)
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_FXT1);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGB888);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_XRGB8888);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
+	 break;
+      case GL_COMPRESSED_RGBA_ARB:
+         if (ctx->Extensions.EXT_texture_compression_s3tc ||
+             ctx->Extensions.S3_s3tc)
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_DXT3); /* Not rgba_dxt1, see spec */
+         if (ctx->Extensions.TDFX_texture_compression_FXT1)
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_FXT1);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA8888);
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
+	 break;
+      default:
+         ; /* fallthrough */
+   }
+
+   if (ctx->Extensions.MESA_ycbcr_texture) {
+      if (internalFormat == GL_YCBCR_MESA) {
+         if (type == GL_UNSIGNED_SHORT_8_8_MESA)
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_YCBCR);
+         else
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_YCBCR_REV);
+      }
+   }
+
+#if FEATURE_texture_fxt1
+   if (ctx->Extensions.TDFX_texture_compression_FXT1) {
+      switch (internalFormat) {
+         case GL_COMPRESSED_RGB_FXT1_3DFX:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_FXT1);
+	 break;
+         case GL_COMPRESSED_RGBA_FXT1_3DFX:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_FXT1);
+	 break;
+         default:
+            ; /* fallthrough */
+      }
+   }
+#endif
+
+#if FEATURE_texture_s3tc
+   if (ctx->Extensions.EXT_texture_compression_s3tc) {
+      switch (internalFormat) {
+         case GL_COMPRESSED_RGB_S3TC_DXT1_EXT:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_DXT1);
+	    break;
+         case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_DXT1);
+	    break;
+         case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_DXT3);
+	    break;
+         case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_DXT5);
+	    break;
+         default:
+            ; /* fallthrough */
+      }
+   }
+
+   if (ctx->Extensions.S3_s3tc) {
+      switch (internalFormat) {
+         case GL_RGB_S3TC:
+         case GL_RGB4_S3TC:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_DXT1);
+	    break;
+         case GL_RGBA_S3TC:
+         case GL_RGBA4_S3TC:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_DXT3);
+	    break;
+         default:
+            ; /* fallthrough */
+      }
+   }
+#endif
+
+   if (ctx->Extensions.ARB_texture_float) {
+      switch (internalFormat) {
+         case GL_ALPHA16F_ARB:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_ALPHA_FLOAT16);
+	    break;
+         case GL_ALPHA32F_ARB:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_ALPHA_FLOAT32);
+	    break;
+         case GL_LUMINANCE16F_ARB:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_LUMINANCE_FLOAT16);
+	    break;
+         case GL_LUMINANCE32F_ARB:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_LUMINANCE_FLOAT32);
+	    break;
+         case GL_LUMINANCE_ALPHA16F_ARB:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_LUMINANCE_ALPHA_FLOAT16);
+	    break;
+         case GL_LUMINANCE_ALPHA32F_ARB:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_LUMINANCE_ALPHA_FLOAT32);
+	    break;
+         case GL_INTENSITY16F_ARB:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_INTENSITY_FLOAT16);
+	    break;
+         case GL_INTENSITY32F_ARB:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_INTENSITY_FLOAT32);
+	    break;
+         case GL_RGB16F_ARB:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_FLOAT16);
+	    break;
+         case GL_RGB32F_ARB:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGB_FLOAT32);
+	    break;
+         case GL_RGBA16F_ARB:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_FLOAT16);
+	    break;
+         case GL_RGBA32F_ARB:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_FLOAT32);
+	    break;
+         default:
+            ; /* fallthrough */
+      }
+   }
+
+   if (ctx->Extensions.EXT_packed_depth_stencil) {
+      switch (internalFormat) {
+         case GL_DEPTH_STENCIL_EXT:
+         case GL_DEPTH24_STENCIL8_EXT:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_Z24_S8);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_S8_Z24);
+	    break;
+         default:
+            ; /* fallthrough */
+      }
+   }
+
+   if (ctx->Extensions.ATI_envmap_bumpmap) {
+      switch (internalFormat) {
+         case GL_DUDV_ATI:
+         case GL_DU8DV8_ATI:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_DUDV8);
+	    break;
+         default:
+            ; /* fallthrough */
+      }
+   }
+
+   if (ctx->Extensions.MESA_texture_signed_rgba) {
+      switch (internalFormat) {
+         case GL_RGBA_SNORM:
+         case GL_RGBA8_SNORM:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RGBA8888);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RGBA8888_REV);
+	    break;
+         default:
+            ; /* fallthrough */
+      }
+   }
+
+   if (ctx->VersionMajor * 10 + ctx->VersionMinor >= 31) {
+      switch (internalFormat) {
+         case GL_RED_SNORM:
+         case GL_R8_SNORM:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_R8);
+	    break;
+         case GL_RG_SNORM:
+         case GL_RG8_SNORM:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RG88);
+	    break;
+         case GL_RGB_SNORM:
+         case GL_RGB8_SNORM:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RGBX8888);
+	    break;
+         case GL_RGBA_SNORM:
+         case GL_RGBA8_SNORM:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RGBA8888);
+	    break;
+         case GL_R16_SNORM:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_R_16);
+	    break;
+         case GL_RG16_SNORM:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RG_16);
+	    break;
+         case GL_RGB16_SNORM:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RGB_16);
+	    break;
+         case GL_RGBA16_SNORM:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RGBA_16);
+	    break;
+         default:
+            ; /* fall-through */
+      }
+   }
+
+#if FEATURE_EXT_texture_sRGB
+   if (ctx->Extensions.EXT_texture_sRGB) {
+      switch (internalFormat) {
+         case GL_SRGB_EXT:
+         case GL_SRGB8_EXT:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SRGB8);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
+	    break;
+         case GL_SRGB_ALPHA_EXT:
+         case GL_SRGB8_ALPHA8_EXT:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SRGBA8);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
+	    break;
+         case GL_SLUMINANCE_EXT:
+         case GL_SLUMINANCE8_EXT:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SL8);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
+	    break;
+         case GL_SLUMINANCE_ALPHA_EXT:
+         case GL_SLUMINANCE8_ALPHA8_EXT:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SLA8);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
+	    break;
+         case GL_COMPRESSED_SLUMINANCE_EXT:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SL8);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
+	    break;
+         case GL_COMPRESSED_SLUMINANCE_ALPHA_EXT:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SLA8);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
+	    break;
+         case GL_COMPRESSED_SRGB_EXT:
+#if FEATURE_texture_s3tc
+            if (ctx->Extensions.EXT_texture_compression_s3tc)
+	       RETURN_IF_SUPPORTED(MESA_FORMAT_SRGB_DXT1);
+#endif
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SRGB8);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
+	    break;
+         case GL_COMPRESSED_SRGB_ALPHA_EXT:
+#if FEATURE_texture_s3tc
+            if (ctx->Extensions.EXT_texture_compression_s3tc)
+	       RETURN_IF_SUPPORTED(MESA_FORMAT_SRGBA_DXT3); /* Not srgba_dxt1, see spec */
+#endif
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SRGBA8);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
+	    break;
+#if FEATURE_texture_s3tc
+         case GL_COMPRESSED_SRGB_S3TC_DXT1_EXT:
+            if (ctx->Extensions.EXT_texture_compression_s3tc)
+	       RETURN_IF_SUPPORTED(MESA_FORMAT_SRGB_DXT1);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
+            break;
+         case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT:
+            if (ctx->Extensions.EXT_texture_compression_s3tc)
+	       RETURN_IF_SUPPORTED(MESA_FORMAT_SRGBA_DXT1);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
+            break;
+         case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT:
+            if (ctx->Extensions.EXT_texture_compression_s3tc)
+	       RETURN_IF_SUPPORTED(MESA_FORMAT_SRGBA_DXT3);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
+            break;
+         case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT:
+            if (ctx->Extensions.EXT_texture_compression_s3tc)
+	       RETURN_IF_SUPPORTED(MESA_FORMAT_SRGBA_DXT5);
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SARGB8);
+            break;
+#endif
+         default:
+            ; /* fallthrough */
+      }
+   }
+#endif /* FEATURE_EXT_texture_sRGB */
+
+   if (ctx->Extensions.EXT_texture_integer) {
+      switch (internalFormat) {
+      case GL_RGBA32UI_EXT:
+      case GL_RGB32UI_EXT:
+      case GL_ALPHA32UI_EXT:
+      case GL_INTENSITY32UI_EXT:
+      case GL_LUMINANCE32UI_EXT:
+      case GL_LUMINANCE_ALPHA32UI_EXT:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_UINT32);
+	 break;
+      case GL_RGBA16UI_EXT:
+      case GL_RGB16UI_EXT:
+      case GL_ALPHA16UI_EXT:
+      case GL_INTENSITY16UI_EXT:
+      case GL_LUMINANCE16UI_EXT:
+      case GL_LUMINANCE_ALPHA16UI_EXT:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_UINT16);
+	 break;
+      case GL_RGBA8UI_EXT:
+      case GL_RGB8UI_EXT:
+      case GL_ALPHA8UI_EXT:
+      case GL_INTENSITY8UI_EXT:
+      case GL_LUMINANCE8UI_EXT:
+      case GL_LUMINANCE_ALPHA8UI_EXT:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_UINT8);
+	 break;
+      case GL_RGBA32I_EXT:
+      case GL_RGB32I_EXT:
+      case GL_ALPHA32I_EXT:
+      case GL_INTENSITY32I_EXT:
+      case GL_LUMINANCE32I_EXT:
+      case GL_LUMINANCE_ALPHA32I_EXT:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_INT32);
+	 break;
+      case GL_RGBA16I_EXT:
+      case GL_RGB16I_EXT:
+      case GL_ALPHA16I_EXT:
+      case GL_INTENSITY16I_EXT:
+      case GL_LUMINANCE16I_EXT:
+      case GL_LUMINANCE_ALPHA16I_EXT:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_INT16);
+	 break;
+      case GL_RGBA8I_EXT:
+      case GL_RGB8I_EXT:
+      case GL_ALPHA8I_EXT:
+      case GL_INTENSITY8I_EXT:
+      case GL_LUMINANCE8I_EXT:
+      case GL_LUMINANCE_ALPHA8I_EXT:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RGBA_INT8);
+	 break;
+      }
+   }
+
+   if (ctx->Extensions.ARB_texture_rg) {
+      switch (internalFormat) {
+      case GL_R8:
+      case GL_RED:
+      case GL_COMPRESSED_RED:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_R8);
+	 break;
+
+      case GL_R16:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_R16);
+	 break;
+
+      case GL_RG:
+      case GL_RG8:
+      case GL_COMPRESSED_RG:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RG88);
+	 break;
+
+      case GL_RG16:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_RG1616);
+	 break;
+
+      default:
+         ; /* fallthrough */
+      }
+   }
+
+   if (ctx->Extensions.EXT_texture_format_BGRA8888) {
+      switch (internalFormat) {
+      case GL_BGRA:
+	 RETURN_IF_SUPPORTED(MESA_FORMAT_ARGB8888);
+	 break;
+
+      default:
+         ; /* fallthrough */
+      }
+   }
+
+   if (ctx->Extensions.ARB_texture_compression_rgtc) {
+      switch (internalFormat) {
+         case GL_COMPRESSED_RED_RGTC1:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RED_RGTC1);
+	    break;
+         case GL_COMPRESSED_SIGNED_RED_RGTC1:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RED_RGTC1);
+	    break;
+         case GL_COMPRESSED_RG_RGTC2:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_RG_RGTC2);
+	    break;
+         case GL_COMPRESSED_SIGNED_RG_RGTC2:
+	    RETURN_IF_SUPPORTED(MESA_FORMAT_SIGNED_RG_RGTC2);
+	    break;
+         default:
+            ; /* fallthrough */
+      }
+   }
+
+   _mesa_problem(ctx, "unexpected format in _mesa_choose_tex_format()");
+   return MESA_FORMAT_NONE;
+}
+
diff --git a/mesalib/src/mesa/main/texstore.c b/mesalib/src/mesa/main/texstore.c
index 1947d32e6..60ded5852 100644
--- a/mesalib/src/mesa/main/texstore.c
+++ b/mesalib/src/mesa/main/texstore.c
@@ -65,6 +65,7 @@
 #include "pack.h"
 #include "texcompress.h"
 #include "texcompress_fxt1.h"
+#include "texcompress_rgtc.h"
 #include "texcompress_s3tc.h"
 #include "teximage.h"
 #include "texstore.h"
@@ -310,15 +311,15 @@ compute_component_mapping(GLenum inFormat, GLenum outFormat,
  * \param srcPacking  source image pixel packing
  * \return resulting image with format = textureBaseFormat and type = GLfloat.
  */
-static GLfloat *
-make_temp_float_image(struct gl_context *ctx, GLuint dims,
-                      GLenum logicalBaseFormat,
-                      GLenum textureBaseFormat,
-                      GLint srcWidth, GLint srcHeight, GLint srcDepth,
-                      GLenum srcFormat, GLenum srcType,
-                      const GLvoid *srcAddr,
-                      const struct gl_pixelstore_attrib *srcPacking,
-                      GLbitfield transferOps)
+GLfloat *
+_mesa_make_temp_float_image(struct gl_context *ctx, GLuint dims,
+			    GLenum logicalBaseFormat,
+			    GLenum textureBaseFormat,
+			    GLint srcWidth, GLint srcHeight, GLint srcDepth,
+			    GLenum srcFormat, GLenum srcType,
+			    const GLvoid *srcAddr,
+			    const struct gl_pixelstore_attrib *srcPacking,
+			    GLbitfield transferOps)
 {
    GLfloat *tempImage;
    const GLint components = _mesa_components_in_format(logicalBaseFormat);
@@ -2065,7 +2066,7 @@ _mesa_texstore_argb2101010(TEXSTORE_PARAMS)
    }
    else {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -2317,7 +2318,7 @@ _mesa_texstore_unorm1616(TEXSTORE_PARAMS)
    }
    else {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -2394,7 +2395,7 @@ _mesa_texstore_unorm16(TEXSTORE_PARAMS)
    }
    else {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -2452,7 +2453,7 @@ _mesa_texstore_rgba_16(TEXSTORE_PARAMS)
    }
    else {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -2519,7 +2520,7 @@ _mesa_texstore_signed_rgba_16(TEXSTORE_PARAMS)
    }
    else {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -2901,7 +2902,7 @@ _mesa_texstore_signed_r8(TEXSTORE_PARAMS)
    /* XXX look at adding optimized paths */
    {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -2946,7 +2947,7 @@ _mesa_texstore_signed_rg88(TEXSTORE_PARAMS)
    /* XXX look at adding optimized paths */
    {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -2991,7 +2992,7 @@ _mesa_texstore_signed_rgbx8888(TEXSTORE_PARAMS)
 
    {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -3104,7 +3105,7 @@ _mesa_texstore_signed_rgba8888(TEXSTORE_PARAMS)
    }
    else {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -3413,7 +3414,7 @@ _mesa_texstore_rgba_float32(TEXSTORE_PARAMS)
    }
    else {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -3483,7 +3484,7 @@ _mesa_texstore_rgba_float16(TEXSTORE_PARAMS)
    }
    else {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -3549,7 +3550,7 @@ _mesa_texstore_rgba_int8(TEXSTORE_PARAMS)
    }
    else {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -3614,7 +3615,7 @@ _mesa_texstore_rgba_int16(TEXSTORE_PARAMS)
    }
    else {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -3679,7 +3680,7 @@ _mesa_texstore_rgba_int32(TEXSTORE_PARAMS)
    }
    else {
       /* general path */
-      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,
                                                  baseInternalFormat,
                                                  baseFormat,
                                                  srcWidth, srcHeight, srcDepth,
@@ -4128,7 +4129,12 @@ texstore_funcs[MESA_FORMAT_COUNT] =
    { MESA_FORMAT_SIGNED_RG_16, _mesa_texstore_signed_rgba_16 },
    { MESA_FORMAT_SIGNED_RGB_16, _mesa_texstore_signed_rgba_16 },
    { MESA_FORMAT_SIGNED_RGBA_16, _mesa_texstore_signed_rgba_16 },
-   { MESA_FORMAT_RGBA_16, _mesa_texstore_rgba_16 }
+   { MESA_FORMAT_RGBA_16, _mesa_texstore_rgba_16 },
+
+   { MESA_FORMAT_RED_RGTC1, _mesa_texstore_red_rgtc1 },
+   { MESA_FORMAT_SIGNED_RED_RGTC1, _mesa_texstore_signed_red_rgtc1 },
+   { MESA_FORMAT_RG_RGTC2, _mesa_texstore_rg_rgtc2 },
+   { MESA_FORMAT_SIGNED_RG_RGTC2, _mesa_texstore_signed_rg_rgtc2 }
 };
 
 
diff --git a/mesalib/src/mesa/main/texstore.h b/mesalib/src/mesa/main/texstore.h
index 8640a23a7..35339da77 100644
--- a/mesalib/src/mesa/main/texstore.h
+++ b/mesalib/src/mesa/main/texstore.h
@@ -81,6 +81,15 @@ _mesa_make_temp_chan_image(struct gl_context *ctx, GLuint dims,
                            const GLvoid *srcAddr,
                            const struct gl_pixelstore_attrib *srcPacking);
 
+GLfloat *
+_mesa_make_temp_float_image(struct gl_context *ctx, GLuint dims,
+			    GLenum logicalBaseFormat,
+			    GLenum textureBaseFormat,
+			    GLint srcWidth, GLint srcHeight, GLint srcDepth,
+			    GLenum srcFormat, GLenum srcType,
+			    const GLvoid *srcAddr,
+			    const struct gl_pixelstore_attrib *srcPacking,
+			    GLbitfield transferOps);
 
 extern void
 _mesa_store_teximage1d(struct gl_context *ctx, GLenum target, GLint level,
diff --git a/mesalib/src/mesa/sources.mak b/mesalib/src/mesa/sources.mak
index b10881c55..b71194880 100644
--- a/mesalib/src/mesa/sources.mak
+++ b/mesalib/src/mesa/sources.mak
@@ -78,6 +78,7 @@ MAIN_SOURCES = \
 	main/stencil.c \
 	main/syncobj.c \
 	main/texcompress.c \
+	main/texcompress_rgtc.c \
 	main/texcompress_s3tc.c \
 	main/texcompress_fxt1.c \
 	main/texenv.c \
diff --git a/mesalib/src/mesa/state_tracker/st_draw.c b/mesalib/src/mesa/state_tracker/st_draw.c
index 16cd92c6b..28186378f 100644
--- a/mesalib/src/mesa/state_tracker/st_draw.c
+++ b/mesalib/src/mesa/state_tracker/st_draw.c
@@ -579,6 +579,7 @@ st_validate_varrays(struct gl_context *ctx,
    if (is_interleaved_arrays(vp, vpv, arrays)) {
       setup_interleaved_attribs(ctx, vp, vpv, arrays, vbuffer, velements,
                                 max_index);
+
       num_vbuffers = 1;
       num_velements = vpv->num_inputs;
       if (num_velements == 0)
@@ -645,6 +646,7 @@ st_draw_vbo(struct gl_context *ctx,
       for (i = 0; i < nr_prims; i++) {
          min_index = MIN2(min_index, prims[i].start);
          max_index = MAX2(max_index, prims[i].start + prims[i].count - 1);
+         max_index = MAX2(max_index, prims[i].num_instances);
       }
    }
 
diff --git a/mesalib/src/mesa/state_tracker/st_extensions.c b/mesalib/src/mesa/state_tracker/st_extensions.c
index b1ede1c8b..71c6c9252 100644
--- a/mesalib/src/mesa/state_tracker/st_extensions.c
+++ b/mesalib/src/mesa/state_tracker/st_extensions.c
@@ -416,6 +416,22 @@ void st_init_extensions(struct st_context *st)
       ctx->Extensions.S3_s3tc = GL_TRUE;
    }
 
+   if (screen->is_format_supported(screen, PIPE_FORMAT_RGTC1_UNORM,
+                                   PIPE_TEXTURE_2D, 0,
+                                   PIPE_BIND_SAMPLER_VIEW, 0) &&
+       screen->is_format_supported(screen, PIPE_FORMAT_RGTC1_SNORM,
+				   PIPE_TEXTURE_2D, 0,
+                                   PIPE_BIND_SAMPLER_VIEW, 0) &&
+       screen->is_format_supported(screen, PIPE_FORMAT_RGTC2_UNORM,
+                                   PIPE_TEXTURE_2D, 0,
+                                   PIPE_BIND_SAMPLER_VIEW, 0) &&
+       screen->is_format_supported(screen, PIPE_FORMAT_RGTC2_SNORM,
+                                   PIPE_TEXTURE_2D, 0,
+                                   PIPE_BIND_SAMPLER_VIEW, 0)
+       ) {
+     ctx->Extensions.ARB_texture_compression_rgtc = GL_TRUE;
+   }
+
    /* ycbcr support */
    if (screen->is_format_supported(screen, PIPE_FORMAT_UYVY, 
                                    PIPE_TEXTURE_2D, 0,
diff --git a/mesalib/src/mesa/state_tracker/st_format.c b/mesalib/src/mesa/state_tracker/st_format.c
index 3620d0cb6..f004ecf03 100644
--- a/mesalib/src/mesa/state_tracker/st_format.c
+++ b/mesalib/src/mesa/state_tracker/st_format.c
@@ -241,6 +241,14 @@ st_mesa_format_to_pipe_format(gl_format mesaFormat)
    case MESA_FORMAT_RGBA_UINT32:
       return PIPE_FORMAT_R32G32B32A32_USCALED;
 
+   case MESA_FORMAT_RED_RGTC1:
+      return PIPE_FORMAT_RGTC1_UNORM;
+   case MESA_FORMAT_SIGNED_RED_RGTC1:
+      return PIPE_FORMAT_RGTC1_SNORM;
+   case MESA_FORMAT_RG_RGTC2:
+      return PIPE_FORMAT_RGTC2_UNORM;
+   case MESA_FORMAT_SIGNED_RG_RGTC2:
+      return PIPE_FORMAT_RGTC2_SNORM;
    default:
       assert(0);
       return PIPE_FORMAT_NONE;
@@ -380,6 +388,15 @@ st_pipe_format_to_mesa_format(enum pipe_format format)
    case PIPE_FORMAT_R32G32B32A32_USCALED:
       return MESA_FORMAT_RGBA_UINT32;
 
+   case PIPE_FORMAT_RGTC1_UNORM:
+      return MESA_FORMAT_RED_RGTC1;
+   case PIPE_FORMAT_RGTC1_SNORM:
+      return MESA_FORMAT_SIGNED_RED_RGTC1;
+   case PIPE_FORMAT_RGTC2_UNORM:
+      return MESA_FORMAT_RG_RGTC2;
+   case PIPE_FORMAT_RGTC2_SNORM:
+      return MESA_FORMAT_SIGNED_RG_RGTC2;
+
    default:
       assert(0);
       return MESA_FORMAT_NONE;
diff --git a/mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c b/mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c
index 5c68fd78c..bf0bfd6df 100644
--- a/mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c
+++ b/mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c
@@ -1,1249 +1,1249 @@
-/**************************************************************************
- * 
- * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-/*
- * \author
- * Michal Krol,
- * Keith Whitwell
- */
-
-#include "pipe/p_compiler.h"
-#include "pipe/p_context.h"
-#include "pipe/p_screen.h"
-#include "pipe/p_shader_tokens.h"
-#include "pipe/p_state.h"
-#include "tgsi/tgsi_ureg.h"
-#include "st_mesa_to_tgsi.h"
-#include "st_context.h"
-#include "program/prog_instruction.h"
-#include "program/prog_parameter.h"
-#include "util/u_debug.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-
-
-#define PROGRAM_ANY_CONST ((1 << PROGRAM_LOCAL_PARAM) |  \
-                           (1 << PROGRAM_ENV_PARAM) |    \
-                           (1 << PROGRAM_STATE_VAR) |    \
-                           (1 << PROGRAM_NAMED_PARAM) |  \
-                           (1 << PROGRAM_CONSTANT) |     \
-                           (1 << PROGRAM_UNIFORM))
-
-
-struct label {
-   unsigned branch_target;
-   unsigned token;
-};
-
-
-/**
- * Intermediate state used during shader translation.
- */
-struct st_translate {
-   struct ureg_program *ureg;
-
-   struct ureg_dst temps[MAX_PROGRAM_TEMPS];
-   struct ureg_src *constants;
-   struct ureg_dst outputs[PIPE_MAX_SHADER_OUTPUTS];
-   struct ureg_src inputs[PIPE_MAX_SHADER_INPUTS];
-   struct ureg_dst address[1];
-   struct ureg_src samplers[PIPE_MAX_SAMPLERS];
-   struct ureg_src systemValues[SYSTEM_VALUE_MAX];
-
-   /* Extra info for handling point size clamping in vertex shader */
-   struct ureg_dst pointSizeResult; /**< Actual point size output register */
-   struct ureg_src pointSizeConst;  /**< Point size range constant register */
-   GLint pointSizeOutIndex;         /**< Temp point size output register */
-   GLboolean prevInstWrotePointSize;
-
-   const GLuint *inputMapping;
-   const GLuint *outputMapping;
-
-   /* For every instruction that contains a label (eg CALL), keep
-    * details so that we can go back afterwards and emit the correct
-    * tgsi instruction number for each label.
-    */
-   struct label *labels;
-   unsigned labels_size;
-   unsigned labels_count;
-
-   /* Keep a record of the tgsi instruction number that each mesa
-    * instruction starts at, will be used to fix up labels after
-    * translation.
-    */
-   unsigned *insn;
-   unsigned insn_size;
-   unsigned insn_count;
-
-   unsigned procType;  /**< TGSI_PROCESSOR_VERTEX/FRAGMENT */
-
-   boolean error;
-};
-
-
-/** Map Mesa's SYSTEM_VALUE_x to TGSI_SEMANTIC_x */
-static unsigned mesa_sysval_to_semantic[SYSTEM_VALUE_MAX] = {
-   TGSI_SEMANTIC_FACE,
-   TGSI_SEMANTIC_INSTANCEID
-};
-
-
-/**
- * Make note of a branch to a label in the TGSI code.
- * After we've emitted all instructions, we'll go over the list
- * of labels built here and patch the TGSI code with the actual
- * location of each label.
- */
-static unsigned *get_label( struct st_translate *t,
-                            unsigned branch_target )
-{
-   unsigned i;
-
-   if (t->labels_count + 1 >= t->labels_size) {
-      unsigned old_size = t->labels_size;
-      t->labels_size = 1 << (util_logbase2(t->labels_size) + 1);
-      t->labels = REALLOC( t->labels, 
-                           old_size * sizeof t->labels[0], 
-                           t->labels_size * sizeof t->labels[0] );
-      if (t->labels == NULL) {
-         static unsigned dummy;
-         t->error = TRUE;
-         return &dummy;
-      }
-   }
-
-   i = t->labels_count++;
-   t->labels[i].branch_target = branch_target;
-   return &t->labels[i].token;
-}
-
-
-/**
- * Called prior to emitting the TGSI code for each Mesa instruction.
- * Allocate additional space for instructions if needed.
- * Update the insn[] array so the next Mesa instruction points to
- * the next TGSI instruction.
- */
-static void set_insn_start( struct st_translate *t,
-                            unsigned start )
-{
-   if (t->insn_count + 1 >= t->insn_size) {
-      unsigned old_size = t->insn_size;
-      t->insn_size = 1 << (util_logbase2(t->insn_size) + 1);
-      t->insn = REALLOC( t->insn, 
-                         old_size * sizeof t->insn[0], 
-                         t->insn_size * sizeof t->insn[0] );
-      if (t->insn == NULL) {
-         t->error = TRUE;
-         return;
-      }
-   }
-
-   t->insn[t->insn_count++] = start;
-}
-
-
-/**
- * Map a Mesa dst register to a TGSI ureg_dst register.
- */
-static struct ureg_dst
-dst_register( struct st_translate *t,
-              gl_register_file file,
-              GLuint index )
-{
-   switch( file ) {
-   case PROGRAM_UNDEFINED:
-      return ureg_dst_undef();
-
-   case PROGRAM_TEMPORARY:
-      if (ureg_dst_is_undef(t->temps[index]))
-         t->temps[index] = ureg_DECL_temporary( t->ureg );
-
-      return t->temps[index];
-
-   case PROGRAM_OUTPUT:
-      if (t->procType == TGSI_PROCESSOR_VERTEX && index == VERT_RESULT_PSIZ)
-         t->prevInstWrotePointSize = GL_TRUE;
-
-      if (t->procType == TGSI_PROCESSOR_VERTEX)
-         assert(index < VERT_RESULT_MAX);
-      else if (t->procType == TGSI_PROCESSOR_FRAGMENT)
-         assert(index < FRAG_RESULT_MAX);
-      else
-         assert(index < GEOM_RESULT_MAX);
-
-      assert(t->outputMapping[index] < Elements(t->outputs));
-
-      return t->outputs[t->outputMapping[index]];
-
-   case PROGRAM_ADDRESS:
-      return t->address[index];
-
-   default:
-      debug_assert( 0 );
-      return ureg_dst_undef();
-   }
-}
-
-
-/**
- * Map a Mesa src register to a TGSI ureg_src register.
- */
-static struct ureg_src
-src_register( struct st_translate *t,
-              gl_register_file file,
-              GLint index )
-{
-   switch( file ) {
-   case PROGRAM_UNDEFINED:
-      return ureg_src_undef();
-
-   case PROGRAM_TEMPORARY:
-      assert(index >= 0);
-      if (ureg_dst_is_undef(t->temps[index]))
-         t->temps[index] = ureg_DECL_temporary( t->ureg );
-      assert(index < Elements(t->temps));
-      return ureg_src(t->temps[index]);
-
-   case PROGRAM_NAMED_PARAM:
-   case PROGRAM_ENV_PARAM:
-   case PROGRAM_LOCAL_PARAM:
-   case PROGRAM_UNIFORM:
-      assert(index >= 0);
-      return t->constants[index];
-   case PROGRAM_STATE_VAR:
-   case PROGRAM_CONSTANT:       /* ie, immediate */
-      if (index < 0)
-         return ureg_DECL_constant( t->ureg, 0 );
-      else
-         return t->constants[index];
-
-   case PROGRAM_INPUT:
-      assert(t->inputMapping[index] < Elements(t->inputs));
-      return t->inputs[t->inputMapping[index]];
-
-   case PROGRAM_OUTPUT:
-      assert(t->outputMapping[index] < Elements(t->outputs));
-      return ureg_src(t->outputs[t->outputMapping[index]]); /* not needed? */
-
-   case PROGRAM_ADDRESS:
-      return ureg_src(t->address[index]);
-
-   case PROGRAM_SYSTEM_VALUE:
-      assert(index < Elements(t->systemValues));
-      return t->systemValues[index];
-
-   default:
-      debug_assert( 0 );
-      return ureg_src_undef();
-   }
-}
-
-
-/**
- * Map mesa texture target to TGSI texture target.
- */
-static unsigned
-translate_texture_target( GLuint textarget,
-                          GLboolean shadow )
-{
-   if (shadow) {
-      switch( textarget ) {
-      case TEXTURE_1D_INDEX:   return TGSI_TEXTURE_SHADOW1D;
-      case TEXTURE_2D_INDEX:   return TGSI_TEXTURE_SHADOW2D;
-      case TEXTURE_RECT_INDEX: return TGSI_TEXTURE_SHADOWRECT;
-      default: break;
-      }
-   }
-
-   switch( textarget ) {
-   case TEXTURE_1D_INDEX:   return TGSI_TEXTURE_1D;
-   case TEXTURE_2D_INDEX:   return TGSI_TEXTURE_2D;
-   case TEXTURE_3D_INDEX:   return TGSI_TEXTURE_3D;
-   case TEXTURE_CUBE_INDEX: return TGSI_TEXTURE_CUBE;
-   case TEXTURE_RECT_INDEX: return TGSI_TEXTURE_RECT;
-   case TEXTURE_1D_ARRAY_INDEX:   return TGSI_TEXTURE_1D_ARRAY;
-   case TEXTURE_2D_ARRAY_INDEX:   return TGSI_TEXTURE_2D_ARRAY;
-   default:
-      debug_assert( 0 );
-      return TGSI_TEXTURE_1D;
-   }
-}
-
-
-/**
- * Create a TGSI ureg_dst register from a Mesa dest register.
- */
-static struct ureg_dst
-translate_dst( struct st_translate *t,
-               const struct prog_dst_register *DstReg,
-               boolean saturate )
-{
-   struct ureg_dst dst = dst_register( t, 
-                                       DstReg->File,
-                                       DstReg->Index );
-
-   dst = ureg_writemask( dst, 
-                         DstReg->WriteMask );
-   
-   if (saturate)
-      dst = ureg_saturate( dst );
-
-   if (DstReg->RelAddr)
-      dst = ureg_dst_indirect( dst, ureg_src(t->address[0]) );
-
-   return dst;
-}
-
-
-/**
- * Create a TGSI ureg_src register from a Mesa src register.
- */
-static struct ureg_src
-translate_src( struct st_translate *t,
-               const struct prog_src_register *SrcReg )
-{
-   struct ureg_src src = src_register( t, SrcReg->File, SrcReg->Index );
-
-   if (t->procType == TGSI_PROCESSOR_GEOMETRY && SrcReg->HasIndex2) {
-      src = src_register( t, SrcReg->File, SrcReg->Index2 );
-      if (SrcReg->RelAddr2)
-         src = ureg_src_dimension_indirect( src, ureg_src(t->address[0]),
-                                            SrcReg->Index);
-      else
-         src = ureg_src_dimension( src, SrcReg->Index);
-   }
-
-   src = ureg_swizzle( src,
-                       GET_SWZ( SrcReg->Swizzle, 0 ) & 0x3,
-                       GET_SWZ( SrcReg->Swizzle, 1 ) & 0x3,
-                       GET_SWZ( SrcReg->Swizzle, 2 ) & 0x3,
-                       GET_SWZ( SrcReg->Swizzle, 3 ) & 0x3);
-
-   if (SrcReg->Negate == NEGATE_XYZW)
-      src = ureg_negate(src);
-
-   if (SrcReg->Abs) 
-      src = ureg_abs(src);
-
-   if (SrcReg->RelAddr) {
-      src = ureg_src_indirect( src, ureg_src(t->address[0]));
-      if (SrcReg->File != PROGRAM_INPUT &&
-          SrcReg->File != PROGRAM_OUTPUT) {
-         /* If SrcReg->Index was negative, it was set to zero in
-          * src_register().  Reassign it now.  But don't do this
-          * for input/output regs since they get remapped while
-          * const buffers don't.
-          */
-         src.Index = SrcReg->Index;
-      }
-   }
-
-   return src;
-}
-
-
-static struct ureg_src swizzle_4v( struct ureg_src src,
-                                   const unsigned *swz )
-{
-   return ureg_swizzle( src, swz[0], swz[1], swz[2], swz[3] );
-}
-
-
-/**
- * Translate a SWZ instruction into a MOV, MUL or MAD instruction.  EG:
- *
- *   SWZ dst, src.x-y10 
- * 
- * becomes:
- *
- *   MAD dst {1,-1,0,0}, src.xyxx, {0,0,1,0}
- */
-static void emit_swz( struct st_translate *t,
-                      struct ureg_dst dst,
-                      const struct prog_src_register *SrcReg )
-{
-   struct ureg_program *ureg = t->ureg;
-   struct ureg_src src = src_register( t, SrcReg->File, SrcReg->Index );
-
-   unsigned negate_mask =  SrcReg->Negate;
-
-   unsigned one_mask = ((GET_SWZ(SrcReg->Swizzle, 0) == SWIZZLE_ONE) << 0 |
-                        (GET_SWZ(SrcReg->Swizzle, 1) == SWIZZLE_ONE) << 1 |
-                        (GET_SWZ(SrcReg->Swizzle, 2) == SWIZZLE_ONE) << 2 |
-                        (GET_SWZ(SrcReg->Swizzle, 3) == SWIZZLE_ONE) << 3);
-
-   unsigned zero_mask = ((GET_SWZ(SrcReg->Swizzle, 0) == SWIZZLE_ZERO) << 0 |
-                         (GET_SWZ(SrcReg->Swizzle, 1) == SWIZZLE_ZERO) << 1 |
-                         (GET_SWZ(SrcReg->Swizzle, 2) == SWIZZLE_ZERO) << 2 |
-                         (GET_SWZ(SrcReg->Swizzle, 3) == SWIZZLE_ZERO) << 3);
-
-   unsigned negative_one_mask = one_mask & negate_mask;
-   unsigned positive_one_mask = one_mask & ~negate_mask;
-   
-   struct ureg_src imm;
-   unsigned i;
-   unsigned mul_swizzle[4] = {0,0,0,0};
-   unsigned add_swizzle[4] = {0,0,0,0};
-   unsigned src_swizzle[4] = {0,0,0,0};
-   boolean need_add = FALSE;
-   boolean need_mul = FALSE;
-
-   if (dst.WriteMask == 0)
-      return;
-
-   /* Is this just a MOV?
-    */
-   if (zero_mask == 0 &&
-       one_mask == 0 &&
-       (negate_mask == 0 || negate_mask == TGSI_WRITEMASK_XYZW)) 
-   {
-      ureg_MOV( ureg, dst, translate_src( t, SrcReg ));
-      return;
-   }
-
-#define IMM_ZERO    0
-#define IMM_ONE     1
-#define IMM_NEG_ONE 2
-
-   imm = ureg_imm3f( ureg, 0, 1, -1 );
-
-   for (i = 0; i < 4; i++) {
-      unsigned bit = 1 << i;
-
-      if (dst.WriteMask & bit) {
-         if (positive_one_mask & bit) {
-            mul_swizzle[i] = IMM_ZERO;
-            add_swizzle[i] = IMM_ONE;
-            need_add = TRUE;
-         }
-         else if (negative_one_mask & bit) {
-            mul_swizzle[i] = IMM_ZERO;
-            add_swizzle[i] = IMM_NEG_ONE;
-            need_add = TRUE;
-         }
-         else if (zero_mask & bit) {
-            mul_swizzle[i] = IMM_ZERO;
-            add_swizzle[i] = IMM_ZERO;
-            need_add = TRUE;
-         }
-         else {
-            add_swizzle[i] = IMM_ZERO;
-            src_swizzle[i] = GET_SWZ(SrcReg->Swizzle, i);
-            need_mul = TRUE;
-            if (negate_mask & bit) {
-               mul_swizzle[i] = IMM_NEG_ONE;
-            }
-            else {
-               mul_swizzle[i] = IMM_ONE;
-            }
-         }
-      }
-   }
-
-   if (need_mul && need_add) {
-      ureg_MAD( ureg, 
-                dst,
-                swizzle_4v( src, src_swizzle ),
-                swizzle_4v( imm, mul_swizzle ),
-                swizzle_4v( imm, add_swizzle ) );
-   }
-   else if (need_mul) {
-      ureg_MUL( ureg, 
-                dst,
-                swizzle_4v( src, src_swizzle ),
-                swizzle_4v( imm, mul_swizzle ) );
-   }
-   else if (need_add) {
-      ureg_MOV( ureg, 
-                dst,
-                swizzle_4v( imm, add_swizzle ) );
-   }
-   else {
-      debug_assert(0);
-   }
-
-#undef IMM_ZERO
-#undef IMM_ONE
-#undef IMM_NEG_ONE
-}
-
-
-/**
- * Negate the value of DDY to match GL semantics where (0,0) is the
- * lower-left corner of the window.
- * Note that the GL_ARB_fragment_coord_conventions extension will
- * effect this someday.
- */
-static void emit_ddy( struct st_translate *t,
-                      struct ureg_dst dst,
-                      const struct prog_src_register *SrcReg )
-{
-   struct ureg_program *ureg = t->ureg;
-   struct ureg_src src = translate_src( t, SrcReg );
-   src = ureg_negate( src );
-   ureg_DDY( ureg, dst, src );
-}
-
-
-
-static unsigned
-translate_opcode( unsigned op )
-{
-   switch( op ) {
-   case OPCODE_ARL:
-      return TGSI_OPCODE_ARL;
-   case OPCODE_ABS:
-      return TGSI_OPCODE_ABS;
-   case OPCODE_ADD:
-      return TGSI_OPCODE_ADD;
-   case OPCODE_BGNLOOP:
-      return TGSI_OPCODE_BGNLOOP;
-   case OPCODE_BGNSUB:
-      return TGSI_OPCODE_BGNSUB;
-   case OPCODE_BRA:
-      return TGSI_OPCODE_BRA;
-   case OPCODE_BRK:
-      return TGSI_OPCODE_BRK;
-   case OPCODE_CAL:
-      return TGSI_OPCODE_CAL;
-   case OPCODE_CMP:
-      return TGSI_OPCODE_CMP;
-   case OPCODE_CONT:
-      return TGSI_OPCODE_CONT;
-   case OPCODE_COS:
-      return TGSI_OPCODE_COS;
-   case OPCODE_DDX:
-      return TGSI_OPCODE_DDX;
-   case OPCODE_DDY:
-      return TGSI_OPCODE_DDY;
-   case OPCODE_DP2:
-      return TGSI_OPCODE_DP2;
-   case OPCODE_DP2A:
-      return TGSI_OPCODE_DP2A;
-   case OPCODE_DP3:
-      return TGSI_OPCODE_DP3;
-   case OPCODE_DP4:
-      return TGSI_OPCODE_DP4;
-   case OPCODE_DPH:
-      return TGSI_OPCODE_DPH;
-   case OPCODE_DST:
-      return TGSI_OPCODE_DST;
-   case OPCODE_ELSE:
-      return TGSI_OPCODE_ELSE;
-   case OPCODE_EMIT_VERTEX:
-      return TGSI_OPCODE_EMIT;
-   case OPCODE_END_PRIMITIVE:
-      return TGSI_OPCODE_ENDPRIM;
-   case OPCODE_ENDIF:
-      return TGSI_OPCODE_ENDIF;
-   case OPCODE_ENDLOOP:
-      return TGSI_OPCODE_ENDLOOP;
-   case OPCODE_ENDSUB:
-      return TGSI_OPCODE_ENDSUB;
-   case OPCODE_EX2:
-      return TGSI_OPCODE_EX2;
-   case OPCODE_EXP:
-      return TGSI_OPCODE_EXP;
-   case OPCODE_FLR:
-      return TGSI_OPCODE_FLR;
-   case OPCODE_FRC:
-      return TGSI_OPCODE_FRC;
-   case OPCODE_IF:
-      return TGSI_OPCODE_IF;
-   case OPCODE_TRUNC:
-      return TGSI_OPCODE_TRUNC;
-   case OPCODE_KIL:
-      return TGSI_OPCODE_KIL;
-   case OPCODE_KIL_NV:
-      return TGSI_OPCODE_KILP;
-   case OPCODE_LG2:
-      return TGSI_OPCODE_LG2;
-   case OPCODE_LOG:
-      return TGSI_OPCODE_LOG;
-   case OPCODE_LIT:
-      return TGSI_OPCODE_LIT;
-   case OPCODE_LRP:
-      return TGSI_OPCODE_LRP;
-   case OPCODE_MAD:
-      return TGSI_OPCODE_MAD;
-   case OPCODE_MAX:
-      return TGSI_OPCODE_MAX;
-   case OPCODE_MIN:
-      return TGSI_OPCODE_MIN;
-   case OPCODE_MOV:
-      return TGSI_OPCODE_MOV;
-   case OPCODE_MUL:
-      return TGSI_OPCODE_MUL;
-   case OPCODE_NOP:
-      return TGSI_OPCODE_NOP;
-   case OPCODE_NRM3:
-      return TGSI_OPCODE_NRM;
-   case OPCODE_NRM4:
-      return TGSI_OPCODE_NRM4;
-   case OPCODE_POW:
-      return TGSI_OPCODE_POW;
-   case OPCODE_RCP:
-      return TGSI_OPCODE_RCP;
-   case OPCODE_RET:
-      return TGSI_OPCODE_RET;
-   case OPCODE_RSQ:
-      return TGSI_OPCODE_RSQ;
-   case OPCODE_SCS:
-      return TGSI_OPCODE_SCS;
-   case OPCODE_SEQ:
-      return TGSI_OPCODE_SEQ;
-   case OPCODE_SGE:
-      return TGSI_OPCODE_SGE;
-   case OPCODE_SGT:
-      return TGSI_OPCODE_SGT;
-   case OPCODE_SIN:
-      return TGSI_OPCODE_SIN;
-   case OPCODE_SLE:
-      return TGSI_OPCODE_SLE;
-   case OPCODE_SLT:
-      return TGSI_OPCODE_SLT;
-   case OPCODE_SNE:
-      return TGSI_OPCODE_SNE;
-   case OPCODE_SSG:
-      return TGSI_OPCODE_SSG;
-   case OPCODE_SUB:
-      return TGSI_OPCODE_SUB;
-   case OPCODE_TEX:
-      return TGSI_OPCODE_TEX;
-   case OPCODE_TXB:
-      return TGSI_OPCODE_TXB;
-   case OPCODE_TXD:
-      return TGSI_OPCODE_TXD;
-   case OPCODE_TXL:
-      return TGSI_OPCODE_TXL;
-   case OPCODE_TXP:
-      return TGSI_OPCODE_TXP;
-   case OPCODE_XPD:
-      return TGSI_OPCODE_XPD;
-   case OPCODE_END:
-      return TGSI_OPCODE_END;
-   default:
-      debug_assert( 0 );
-      return TGSI_OPCODE_NOP;
-   }
-}
-
-
-static void
-compile_instruction(
-   struct st_translate *t,
-   const struct prog_instruction *inst )
-{
-   struct ureg_program *ureg = t->ureg;
-   GLuint i;
-   struct ureg_dst dst[1];
-   struct ureg_src src[4];
-   unsigned num_dst;
-   unsigned num_src;
-
-   num_dst = _mesa_num_inst_dst_regs( inst->Opcode );
-   num_src = _mesa_num_inst_src_regs( inst->Opcode );
-
-   if (num_dst) 
-      dst[0] = translate_dst( t, 
-                              &inst->DstReg,
-                              inst->SaturateMode );
-
-   for (i = 0; i < num_src; i++) 
-      src[i] = translate_src( t, &inst->SrcReg[i] );
-
-   switch( inst->Opcode ) {
-   case OPCODE_SWZ:
-      emit_swz( t, dst[0], &inst->SrcReg[0] );
-      return;
-
-   case OPCODE_BGNLOOP:
-   case OPCODE_CAL:
-   case OPCODE_ELSE:
-   case OPCODE_ENDLOOP:
-   case OPCODE_IF:
-      debug_assert(num_dst == 0);
-      ureg_label_insn( ureg,
-                       translate_opcode( inst->Opcode ),
-                       src, num_src,
-                       get_label( t, inst->BranchTarget ));
-      return;
-
-   case OPCODE_TEX:
-   case OPCODE_TXB:
-   case OPCODE_TXD:
-   case OPCODE_TXL:
-   case OPCODE_TXP:
-      src[num_src++] = t->samplers[inst->TexSrcUnit];
-      ureg_tex_insn( ureg,
-                     translate_opcode( inst->Opcode ),
-                     dst, num_dst, 
-                     translate_texture_target( inst->TexSrcTarget,
-                                               inst->TexShadow ),
-                     src, num_src );
-      return;
-
-   case OPCODE_SCS:
-      dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XY );
-      ureg_insn( ureg, 
-                 translate_opcode( inst->Opcode ), 
-                 dst, num_dst, 
-                 src, num_src );
-      break;
-
-   case OPCODE_XPD:
-      dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XYZ );
-      ureg_insn( ureg, 
-                 translate_opcode( inst->Opcode ), 
-                 dst, num_dst, 
-                 src, num_src );
-      break;
-
-   case OPCODE_NOISE1:
-   case OPCODE_NOISE2:
-   case OPCODE_NOISE3:
-   case OPCODE_NOISE4:
-      /* At some point, a motivated person could add a better
-       * implementation of noise.  Currently not even the nvidia
-       * binary drivers do anything more than this.  In any case, the
-       * place to do this is in the GL state tracker, not the poor
-       * driver.
-       */
-      ureg_MOV( ureg, dst[0], ureg_imm1f(ureg, 0.5) );
-      break;
-		 
-   case OPCODE_DDY:
-      emit_ddy( t, dst[0], &inst->SrcReg[0] );
-      break;
-
-   default:
-      ureg_insn( ureg, 
-                 translate_opcode( inst->Opcode ), 
-                 dst, num_dst, 
-                 src, num_src );
-      break;
-   }
-}
-
-
-/**
- * Emit the TGSI instructions to adjust the WPOS pixel center convention
- * Basically, add (adjX, adjY) to the fragment position.
- */
-static void
-emit_adjusted_wpos( struct st_translate *t,
-                    const struct gl_program *program,
-                    GLfloat adjX, GLfloat adjY)
-{
-   struct ureg_program *ureg = t->ureg;
-   struct ureg_dst wpos_temp = ureg_DECL_temporary(ureg);
-   struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]];
-
-   /* Note that we bias X and Y and pass Z and W through unchanged.
-    * The shader might also use gl_FragCoord.w and .z.
-    */
-   ureg_ADD(ureg, wpos_temp, wpos_input,
-            ureg_imm4f(ureg, adjX, adjY, 0.0f, 0.0f));
-
-   t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]] = ureg_src(wpos_temp);
-}
-
-
-/**
- * Emit the TGSI instructions for inverting the WPOS y coordinate.
- * This code is unavoidable because it also depends on whether
- * a FBO is bound (STATE_FB_WPOS_Y_TRANSFORM).
- */
-static void
-emit_wpos_inversion( struct st_translate *t,
-                     const struct gl_program *program,
-                     boolean invert)
-{
-   struct ureg_program *ureg = t->ureg;
-
-   /* 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 wposTransformState[STATE_LENGTH]
-      = { STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM, 0, 0, 0 };
-   
-   /* XXX: note we are modifying the incoming shader here!  Need to
-    * do this before emitting the constant decls below, or this
-    * will be missed:
-    */
-   unsigned wposTransConst = _mesa_add_state_reference(program->Parameters,
-                                                       wposTransformState);
-
-   struct ureg_src wpostrans = ureg_DECL_constant( ureg, wposTransConst );
-   struct ureg_dst wpos_temp;
-   struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]];
-
-   /* MOV wpos_temp, input[wpos]
-    */
-   if (wpos_input.File == TGSI_FILE_TEMPORARY)
-      wpos_temp = ureg_dst(wpos_input);
-   else {
-      wpos_temp = ureg_DECL_temporary( ureg );
-      ureg_MOV( ureg, wpos_temp, wpos_input );
-   }
-
-   if (invert) {
-      /* MAD wpos_temp.y, wpos_input, wpostrans.xxxx, wpostrans.yyyy
-       */
-      ureg_MAD( ureg,
-                ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ),
-                wpos_input,
-                ureg_scalar(wpostrans, 0),
-                ureg_scalar(wpostrans, 1));
-   } else {
-      /* MAD wpos_temp.y, wpos_input, wpostrans.zzzz, wpostrans.wwww
-       */
-      ureg_MAD( ureg,
-                ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ),
-                wpos_input,
-                ureg_scalar(wpostrans, 2),
-                ureg_scalar(wpostrans, 3));
-   }
-
-   /* Use wpos_temp as position input from here on:
-    */
-   t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]] = ureg_src(wpos_temp);
-}
-
-
-/**
- * Emit fragment position/ooordinate code.
- */
-static void
-emit_wpos(struct st_context *st,
-          struct st_translate *t,
-          const struct gl_program *program,
-          struct ureg_program *ureg)
-{
-   const struct gl_fragment_program *fp =
-      (const struct gl_fragment_program *) program;
-   struct pipe_screen *pscreen = st->pipe->screen;
-   boolean invert = FALSE;
-
-   if (fp->OriginUpperLeft) {
-      /* Fragment shader wants origin in upper-left */
-      if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT)) {
-         /* the driver supports upper-left origin */
-      }
-      else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT)) {
-         /* the driver supports lower-left origin, need to invert Y */
-         ureg_property_fs_coord_origin(ureg, TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
-         invert = TRUE;
-      }
-      else
-         assert(0);
-   }
-   else {
-      /* Fragment shader wants origin in lower-left */
-      if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT))
-         /* the driver supports lower-left origin */
-         ureg_property_fs_coord_origin(ureg, TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
-      else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT))
-         /* the driver supports upper-left origin, need to invert Y */
-         invert = TRUE;
-      else
-         assert(0);
-   }
-   
-   if (fp->PixelCenterInteger) {
-      /* Fragment shader wants pixel center integer */
-      if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER))
-         /* the driver supports pixel center integer */
-         ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
-      else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER))
-         /* the driver supports pixel center half integer, need to bias X,Y */
-         emit_adjusted_wpos(t, program, 0.5f, invert ? 0.5f : -0.5f);
-      else
-         assert(0);
-   }
-   else {
-      /* Fragment shader wants pixel center half integer */
-      if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER)) {
-         /* the driver supports pixel center half integer */
-      }
-      else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) {
-         /* the driver supports pixel center integer, need to bias X,Y */
-         ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
-         emit_adjusted_wpos(t, program, 0.5f, invert ? -0.5f : 0.5f);
-      }
-      else
-         assert(0);
-   }
-
-   /* we invert after adjustment so that we avoid the MOV to temporary,
-    * and reuse the adjustment ADD instead */
-   emit_wpos_inversion(t, program, invert);
-}
-
-
-/**
- * OpenGL's fragment gl_FrontFace input is 1 for front-facing, 0 for back.
- * TGSI uses +1 for front, -1 for back.
- * This function converts the TGSI value to the GL value.  Simply clamping/
- * saturating the value to [0,1] does the job.
- */
-static void
-emit_face_var( struct st_translate *t,
-               const struct gl_program *program )
-{
-   struct ureg_program *ureg = t->ureg;
-   struct ureg_dst face_temp = ureg_DECL_temporary( ureg );
-   struct ureg_src face_input = t->inputs[t->inputMapping[FRAG_ATTRIB_FACE]];
-
-   /* MOV_SAT face_temp, input[face]
-    */
-   face_temp = ureg_saturate( face_temp );
-   ureg_MOV( ureg, face_temp, face_input );
-
-   /* Use face_temp as face input from here on:
-    */
-   t->inputs[t->inputMapping[FRAG_ATTRIB_FACE]] = ureg_src(face_temp);
-}
-
-
-static void
-emit_edgeflags( struct st_translate *t,
-                 const struct gl_program *program )
-{
-   struct ureg_program *ureg = t->ureg;
-   struct ureg_dst edge_dst = t->outputs[t->outputMapping[VERT_RESULT_EDGE]];
-   struct ureg_src edge_src = t->inputs[t->inputMapping[VERT_ATTRIB_EDGEFLAG]];
-
-   ureg_MOV( ureg, edge_dst, edge_src );
-}
-
-
-/**
- * Translate Mesa program to TGSI format.
- * \param program  the program to translate
- * \param numInputs  number of input registers used
- * \param inputMapping  maps Mesa fragment program inputs to TGSI generic
- *                      input indexes
- * \param inputSemanticName  the TGSI_SEMANTIC flag for each input
- * \param inputSemanticIndex  the semantic index (ex: which texcoord) for
- *                            each input
- * \param interpMode  the TGSI_INTERPOLATE_LINEAR/PERSP mode for each input
- * \param numOutputs  number of output registers used
- * \param outputMapping  maps Mesa fragment program outputs to TGSI
- *                       generic outputs
- * \param outputSemanticName  the TGSI_SEMANTIC flag for each output
- * \param outputSemanticIndex  the semantic index (ex: which texcoord) for
- *                             each output
- *
- * \return  PIPE_OK or PIPE_ERROR_OUT_OF_MEMORY
- */
-enum pipe_error
-st_translate_mesa_program(
-   struct gl_context *ctx,
-   uint procType,
-   struct ureg_program *ureg,
-   const struct gl_program *program,
-   GLuint numInputs,
-   const GLuint inputMapping[],
-   const ubyte inputSemanticName[],
-   const ubyte inputSemanticIndex[],
-   const GLuint interpMode[],
-   GLuint numOutputs,
-   const GLuint outputMapping[],
-   const ubyte outputSemanticName[],
-   const ubyte outputSemanticIndex[],
-   boolean passthrough_edgeflags )
-{
-   struct st_translate translate, *t;
-   unsigned i;
-   enum pipe_error ret = PIPE_OK;
-
-   assert(numInputs <= Elements(t->inputs));
-   assert(numOutputs <= Elements(t->outputs));
-
-   t = &translate;
-   memset(t, 0, sizeof *t);
-
-   t->procType = procType;
-   t->inputMapping = inputMapping;
-   t->outputMapping = outputMapping;
-   t->ureg = ureg;
-   t->pointSizeOutIndex = -1;
-   t->prevInstWrotePointSize = GL_FALSE;
-
-   /*_mesa_print_program(program);*/
-
-   /*
-    * Declare input attributes.
-    */
-   if (procType == TGSI_PROCESSOR_FRAGMENT) {
-      for (i = 0; i < numInputs; i++) {
-         if (program->InputFlags[0] & PROG_PARAM_BIT_CYL_WRAP) {
-            t->inputs[i] = ureg_DECL_fs_input_cyl(ureg,
-                                                  inputSemanticName[i],
-                                                  inputSemanticIndex[i],
-                                                  interpMode[i],
-                                                  TGSI_CYLINDRICAL_WRAP_X);
-         }
-         else {
-            t->inputs[i] = ureg_DECL_fs_input(ureg,
-                                              inputSemanticName[i],
-                                              inputSemanticIndex[i],
-                                              interpMode[i]);
-         }
-      }
-
-      if (program->InputsRead & FRAG_BIT_WPOS) {
-         /* Must do this after setting up t->inputs, and before
-          * emitting constant references, below:
-          */
-         emit_wpos(st_context(ctx), t, program, ureg);
-      }
-
-      if (program->InputsRead & FRAG_BIT_FACE) {
-         emit_face_var( t, program );
-      }
-
-      /*
-       * Declare output attributes.
-       */
-      for (i = 0; i < numOutputs; i++) {
-         switch (outputSemanticName[i]) {
-         case TGSI_SEMANTIC_POSITION:
-            t->outputs[i] = ureg_DECL_output( ureg,
-                                              TGSI_SEMANTIC_POSITION, /* Z / Depth */
-                                              outputSemanticIndex[i] );
-
-            t->outputs[i] = ureg_writemask( t->outputs[i],
-                                            TGSI_WRITEMASK_Z );
-            break;
-         case TGSI_SEMANTIC_STENCIL:
-            t->outputs[i] = ureg_DECL_output( ureg,
-                                              TGSI_SEMANTIC_STENCIL, /* Stencil */
-                                              outputSemanticIndex[i] );
-            t->outputs[i] = ureg_writemask( t->outputs[i],
-                                            TGSI_WRITEMASK_Y );
-            break;
-         case TGSI_SEMANTIC_COLOR:
-            t->outputs[i] = ureg_DECL_output( ureg,
-                                              TGSI_SEMANTIC_COLOR,
-                                              outputSemanticIndex[i] );
-            break;
-         default:
-            debug_assert(0);
-            return 0;
-         }
-      }
-   }
-   else if (procType == TGSI_PROCESSOR_GEOMETRY) {
-      for (i = 0; i < numInputs; i++) {
-         t->inputs[i] = ureg_DECL_gs_input(ureg,
-                                           i,
-                                           inputSemanticName[i],
-                                           inputSemanticIndex[i]);
-      }
-
-      for (i = 0; i < numOutputs; i++) {
-         t->outputs[i] = ureg_DECL_output( ureg,
-                                           outputSemanticName[i],
-                                           outputSemanticIndex[i] );
-      }
-   }
-   else {
-      assert(procType == TGSI_PROCESSOR_VERTEX);
-
-      for (i = 0; i < numInputs; i++) {
-         t->inputs[i] = ureg_DECL_vs_input(ureg, i);
-      }
-
-      for (i = 0; i < numOutputs; i++) {
-         t->outputs[i] = ureg_DECL_output( ureg,
-                                           outputSemanticName[i],
-                                           outputSemanticIndex[i] );
-         if ((outputSemanticName[i] == TGSI_SEMANTIC_PSIZE) && program->Id) {
-            /* Writing to the point size result register requires special
-             * handling to implement clamping.
-             */
-            static const gl_state_index pointSizeClampState[STATE_LENGTH]
-               = { STATE_INTERNAL, STATE_POINT_SIZE_IMPL_CLAMP, 0, 0, 0 };
-               /* XXX: note we are modifying the incoming shader here!  Need to
-               * do this before emitting the constant decls below, or this
-               * will be missed:
-               */
-            unsigned pointSizeClampConst =
-               _mesa_add_state_reference(program->Parameters,
-                                         pointSizeClampState);
-            struct ureg_dst psizregtemp = ureg_DECL_temporary( ureg );
-            t->pointSizeConst = ureg_DECL_constant( ureg, pointSizeClampConst );
-            t->pointSizeResult = t->outputs[i];
-            t->pointSizeOutIndex = i;
-            t->outputs[i] = psizregtemp;
-         }
-      }
-      if (passthrough_edgeflags)
-         emit_edgeflags( t, program );
-   }
-
-   /* Declare address register.
-    */
-   if (program->NumAddressRegs > 0) {
-      debug_assert( program->NumAddressRegs == 1 );
-      t->address[0] = ureg_DECL_address( ureg );
-   }
-
-   /* Declare misc input registers
-    */
-   {
-      GLbitfield sysInputs = program->SystemValuesRead;
-      unsigned numSys = 0;
-      for (i = 0; sysInputs; i++) {
-         if (sysInputs & (1 << i)) {
-            unsigned semName = mesa_sysval_to_semantic[i];
-            t->systemValues[i] = ureg_DECL_system_value(ureg, numSys, semName, 0);
-            numSys++;
-            sysInputs &= ~(1 << i);
-         }
-      }
-   }
-
-   if (program->IndirectRegisterFiles & (1 << PROGRAM_TEMPORARY)) {
-      /* If temps are accessed with indirect addressing, declare temporaries
-       * in sequential order.  Else, we declare them on demand elsewhere.
-       */
-      for (i = 0; i < program->NumTemporaries; i++) {
-         /* XXX use TGSI_FILE_TEMPORARY_ARRAY when it's supported by ureg */
-         t->temps[i] = ureg_DECL_temporary( t->ureg );
-      }
-   }
-
-   /* Emit constants and immediates.  Mesa uses a single index space
-    * for these, so we put all the translated regs in t->constants.
-    */
-   if (program->Parameters) {
-      t->constants = CALLOC( program->Parameters->NumParameters,
-                             sizeof t->constants[0] );
-      if (t->constants == NULL) {
-         ret = PIPE_ERROR_OUT_OF_MEMORY;
-         goto out;
-      }
-
-      for (i = 0; i < program->Parameters->NumParameters; i++) {
-         switch (program->Parameters->Parameters[i].Type) {
-         case PROGRAM_ENV_PARAM:
-         case PROGRAM_LOCAL_PARAM:
-         case PROGRAM_STATE_VAR:
-         case PROGRAM_NAMED_PARAM:
-         case PROGRAM_UNIFORM:
-            t->constants[i] = ureg_DECL_constant( ureg, i );
-            break;
-
-            /* Emit immediates only when there's no indirect addressing of
-             * the const buffer.
-             * FIXME: Be smarter and recognize param arrays:
-             * indirect addressing is only valid within the referenced
-             * array.
-             */
-         case PROGRAM_CONSTANT:
-            if (program->IndirectRegisterFiles & PROGRAM_ANY_CONST)
-               t->constants[i] = ureg_DECL_constant( ureg, i );
-            else
-               t->constants[i] = 
-                  ureg_DECL_immediate( ureg,
-                                       program->Parameters->ParameterValues[i],
-                                       4 );
-            break;
-         default:
-            break;
-         }
-      }
-   }
-
-   /* texture samplers */
-   for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) {
-      if (program->SamplersUsed & (1 << i)) {
-         t->samplers[i] = ureg_DECL_sampler( ureg, i );
-      }
-   }
-
-   /* Emit each instruction in turn:
-    */
-   for (i = 0; i < program->NumInstructions; i++) {
-      set_insn_start( t, ureg_get_instruction_number( ureg ));
-      compile_instruction( t, &program->Instructions[i] );
-
-      if (t->prevInstWrotePointSize && program->Id) {
-         /* The previous instruction wrote to the (fake) vertex point size
-          * result register.  Now we need to clamp that value to the min/max
-          * point size range, putting the result into the real point size
-          * register.
-          * Note that we can't do this easily at the end of program due to
-          * possible early return.
-          */
-         set_insn_start( t, ureg_get_instruction_number( ureg ));
-         ureg_MAX( t->ureg,
-                   ureg_writemask(t->outputs[t->pointSizeOutIndex], WRITEMASK_X),
-                   ureg_src(t->outputs[t->pointSizeOutIndex]),
-                   ureg_swizzle(t->pointSizeConst, 1,1,1,1));
-         ureg_MIN( t->ureg, ureg_writemask(t->pointSizeResult, WRITEMASK_X),
-                   ureg_src(t->outputs[t->pointSizeOutIndex]),
-                   ureg_swizzle(t->pointSizeConst, 2,2,2,2));
-      }
-      t->prevInstWrotePointSize = GL_FALSE;
-   }
-
-   /* Fix up all emitted labels:
-    */
-   for (i = 0; i < t->labels_count; i++) {
-      ureg_fixup_label( ureg,
-                        t->labels[i].token,
-                        t->insn[t->labels[i].branch_target] );
-   }
-
-out:
-   FREE(t->insn);
-   FREE(t->labels);
-   FREE(t->constants);
-
-   if (t->error) {
-      debug_printf("%s: translate error flag set\n", __FUNCTION__);
-   }
-
-   return ret;
-}
-
-
-/**
- * Tokens cannot be free with free otherwise the builtin gallium
- * malloc debugging will get confused.
- */
-void
-st_free_tokens(const struct tgsi_token *tokens)
-{
-   FREE((void *)tokens);
-}
+/**************************************************************************
+ * 
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/*
+ * \author
+ * Michal Krol,
+ * Keith Whitwell
+ */
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_context.h"
+#include "pipe/p_screen.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/p_state.h"
+#include "tgsi/tgsi_ureg.h"
+#include "st_mesa_to_tgsi.h"
+#include "st_context.h"
+#include "program/prog_instruction.h"
+#include "program/prog_parameter.h"
+#include "util/u_debug.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+
+#define PROGRAM_ANY_CONST ((1 << PROGRAM_LOCAL_PARAM) |  \
+                           (1 << PROGRAM_ENV_PARAM) |    \
+                           (1 << PROGRAM_STATE_VAR) |    \
+                           (1 << PROGRAM_NAMED_PARAM) |  \
+                           (1 << PROGRAM_CONSTANT) |     \
+                           (1 << PROGRAM_UNIFORM))
+
+
+struct label {
+   unsigned branch_target;
+   unsigned token;
+};
+
+
+/**
+ * Intermediate state used during shader translation.
+ */
+struct st_translate {
+   struct ureg_program *ureg;
+
+   struct ureg_dst temps[MAX_PROGRAM_TEMPS];
+   struct ureg_src *constants;
+   struct ureg_dst outputs[PIPE_MAX_SHADER_OUTPUTS];
+   struct ureg_src inputs[PIPE_MAX_SHADER_INPUTS];
+   struct ureg_dst address[1];
+   struct ureg_src samplers[PIPE_MAX_SAMPLERS];
+   struct ureg_src systemValues[SYSTEM_VALUE_MAX];
+
+   /* Extra info for handling point size clamping in vertex shader */
+   struct ureg_dst pointSizeResult; /**< Actual point size output register */
+   struct ureg_src pointSizeConst;  /**< Point size range constant register */
+   GLint pointSizeOutIndex;         /**< Temp point size output register */
+   GLboolean prevInstWrotePointSize;
+
+   const GLuint *inputMapping;
+   const GLuint *outputMapping;
+
+   /* For every instruction that contains a label (eg CALL), keep
+    * details so that we can go back afterwards and emit the correct
+    * tgsi instruction number for each label.
+    */
+   struct label *labels;
+   unsigned labels_size;
+   unsigned labels_count;
+
+   /* Keep a record of the tgsi instruction number that each mesa
+    * instruction starts at, will be used to fix up labels after
+    * translation.
+    */
+   unsigned *insn;
+   unsigned insn_size;
+   unsigned insn_count;
+
+   unsigned procType;  /**< TGSI_PROCESSOR_VERTEX/FRAGMENT */
+
+   boolean error;
+};
+
+
+/** Map Mesa's SYSTEM_VALUE_x to TGSI_SEMANTIC_x */
+static unsigned mesa_sysval_to_semantic[SYSTEM_VALUE_MAX] = {
+   TGSI_SEMANTIC_FACE,
+   TGSI_SEMANTIC_INSTANCEID
+};
+
+
+/**
+ * Make note of a branch to a label in the TGSI code.
+ * After we've emitted all instructions, we'll go over the list
+ * of labels built here and patch the TGSI code with the actual
+ * location of each label.
+ */
+static unsigned *get_label( struct st_translate *t,
+                            unsigned branch_target )
+{
+   unsigned i;
+
+   if (t->labels_count + 1 >= t->labels_size) {
+      unsigned old_size = t->labels_size;
+      t->labels_size = 1 << (util_logbase2(t->labels_size) + 1);
+      t->labels = REALLOC( t->labels, 
+                           old_size * sizeof t->labels[0], 
+                           t->labels_size * sizeof t->labels[0] );
+      if (t->labels == NULL) {
+         static unsigned dummy;
+         t->error = TRUE;
+         return &dummy;
+      }
+   }
+
+   i = t->labels_count++;
+   t->labels[i].branch_target = branch_target;
+   return &t->labels[i].token;
+}
+
+
+/**
+ * Called prior to emitting the TGSI code for each Mesa instruction.
+ * Allocate additional space for instructions if needed.
+ * Update the insn[] array so the next Mesa instruction points to
+ * the next TGSI instruction.
+ */
+static void set_insn_start( struct st_translate *t,
+                            unsigned start )
+{
+   if (t->insn_count + 1 >= t->insn_size) {
+      unsigned old_size = t->insn_size;
+      t->insn_size = 1 << (util_logbase2(t->insn_size) + 1);
+      t->insn = REALLOC( t->insn, 
+                         old_size * sizeof t->insn[0], 
+                         t->insn_size * sizeof t->insn[0] );
+      if (t->insn == NULL) {
+         t->error = TRUE;
+         return;
+      }
+   }
+
+   t->insn[t->insn_count++] = start;
+}
+
+
+/**
+ * Map a Mesa dst register to a TGSI ureg_dst register.
+ */
+static struct ureg_dst
+dst_register( struct st_translate *t,
+              gl_register_file file,
+              GLuint index )
+{
+   switch( file ) {
+   case PROGRAM_UNDEFINED:
+      return ureg_dst_undef();
+
+   case PROGRAM_TEMPORARY:
+      if (ureg_dst_is_undef(t->temps[index]))
+         t->temps[index] = ureg_DECL_temporary( t->ureg );
+
+      return t->temps[index];
+
+   case PROGRAM_OUTPUT:
+      if (t->procType == TGSI_PROCESSOR_VERTEX && index == VERT_RESULT_PSIZ)
+         t->prevInstWrotePointSize = GL_TRUE;
+
+      if (t->procType == TGSI_PROCESSOR_VERTEX)
+         assert(index < VERT_RESULT_MAX);
+      else if (t->procType == TGSI_PROCESSOR_FRAGMENT)
+         assert(index < FRAG_RESULT_MAX);
+      else
+         assert(index < GEOM_RESULT_MAX);
+
+      assert(t->outputMapping[index] < Elements(t->outputs));
+
+      return t->outputs[t->outputMapping[index]];
+
+   case PROGRAM_ADDRESS:
+      return t->address[index];
+
+   default:
+      debug_assert( 0 );
+      return ureg_dst_undef();
+   }
+}
+
+
+/**
+ * Map a Mesa src register to a TGSI ureg_src register.
+ */
+static struct ureg_src
+src_register( struct st_translate *t,
+              gl_register_file file,
+              GLint index )
+{
+   switch( file ) {
+   case PROGRAM_UNDEFINED:
+      return ureg_src_undef();
+
+   case PROGRAM_TEMPORARY:
+      assert(index >= 0);
+      assert(index < Elements(t->temps));
+      if (ureg_dst_is_undef(t->temps[index]))
+         t->temps[index] = ureg_DECL_temporary( t->ureg );
+      return ureg_src(t->temps[index]);
+
+   case PROGRAM_NAMED_PARAM:
+   case PROGRAM_ENV_PARAM:
+   case PROGRAM_LOCAL_PARAM:
+   case PROGRAM_UNIFORM:
+      assert(index >= 0);
+      return t->constants[index];
+   case PROGRAM_STATE_VAR:
+   case PROGRAM_CONSTANT:       /* ie, immediate */
+      if (index < 0)
+         return ureg_DECL_constant( t->ureg, 0 );
+      else
+         return t->constants[index];
+
+   case PROGRAM_INPUT:
+      assert(t->inputMapping[index] < Elements(t->inputs));
+      return t->inputs[t->inputMapping[index]];
+
+   case PROGRAM_OUTPUT:
+      assert(t->outputMapping[index] < Elements(t->outputs));
+      return ureg_src(t->outputs[t->outputMapping[index]]); /* not needed? */
+
+   case PROGRAM_ADDRESS:
+      return ureg_src(t->address[index]);
+
+   case PROGRAM_SYSTEM_VALUE:
+      assert(index < Elements(t->systemValues));
+      return t->systemValues[index];
+
+   default:
+      debug_assert( 0 );
+      return ureg_src_undef();
+   }
+}
+
+
+/**
+ * Map mesa texture target to TGSI texture target.
+ */
+static unsigned
+translate_texture_target( GLuint textarget,
+                          GLboolean shadow )
+{
+   if (shadow) {
+      switch( textarget ) {
+      case TEXTURE_1D_INDEX:   return TGSI_TEXTURE_SHADOW1D;
+      case TEXTURE_2D_INDEX:   return TGSI_TEXTURE_SHADOW2D;
+      case TEXTURE_RECT_INDEX: return TGSI_TEXTURE_SHADOWRECT;
+      default: break;
+      }
+   }
+
+   switch( textarget ) {
+   case TEXTURE_1D_INDEX:   return TGSI_TEXTURE_1D;
+   case TEXTURE_2D_INDEX:   return TGSI_TEXTURE_2D;
+   case TEXTURE_3D_INDEX:   return TGSI_TEXTURE_3D;
+   case TEXTURE_CUBE_INDEX: return TGSI_TEXTURE_CUBE;
+   case TEXTURE_RECT_INDEX: return TGSI_TEXTURE_RECT;
+   case TEXTURE_1D_ARRAY_INDEX:   return TGSI_TEXTURE_1D_ARRAY;
+   case TEXTURE_2D_ARRAY_INDEX:   return TGSI_TEXTURE_2D_ARRAY;
+   default:
+      debug_assert( 0 );
+      return TGSI_TEXTURE_1D;
+   }
+}
+
+
+/**
+ * Create a TGSI ureg_dst register from a Mesa dest register.
+ */
+static struct ureg_dst
+translate_dst( struct st_translate *t,
+               const struct prog_dst_register *DstReg,
+               boolean saturate )
+{
+   struct ureg_dst dst = dst_register( t, 
+                                       DstReg->File,
+                                       DstReg->Index );
+
+   dst = ureg_writemask( dst, 
+                         DstReg->WriteMask );
+   
+   if (saturate)
+      dst = ureg_saturate( dst );
+
+   if (DstReg->RelAddr)
+      dst = ureg_dst_indirect( dst, ureg_src(t->address[0]) );
+
+   return dst;
+}
+
+
+/**
+ * Create a TGSI ureg_src register from a Mesa src register.
+ */
+static struct ureg_src
+translate_src( struct st_translate *t,
+               const struct prog_src_register *SrcReg )
+{
+   struct ureg_src src = src_register( t, SrcReg->File, SrcReg->Index );
+
+   if (t->procType == TGSI_PROCESSOR_GEOMETRY && SrcReg->HasIndex2) {
+      src = src_register( t, SrcReg->File, SrcReg->Index2 );
+      if (SrcReg->RelAddr2)
+         src = ureg_src_dimension_indirect( src, ureg_src(t->address[0]),
+                                            SrcReg->Index);
+      else
+         src = ureg_src_dimension( src, SrcReg->Index);
+   }
+
+   src = ureg_swizzle( src,
+                       GET_SWZ( SrcReg->Swizzle, 0 ) & 0x3,
+                       GET_SWZ( SrcReg->Swizzle, 1 ) & 0x3,
+                       GET_SWZ( SrcReg->Swizzle, 2 ) & 0x3,
+                       GET_SWZ( SrcReg->Swizzle, 3 ) & 0x3);
+
+   if (SrcReg->Negate == NEGATE_XYZW)
+      src = ureg_negate(src);
+
+   if (SrcReg->Abs) 
+      src = ureg_abs(src);
+
+   if (SrcReg->RelAddr) {
+      src = ureg_src_indirect( src, ureg_src(t->address[0]));
+      if (SrcReg->File != PROGRAM_INPUT &&
+          SrcReg->File != PROGRAM_OUTPUT) {
+         /* If SrcReg->Index was negative, it was set to zero in
+          * src_register().  Reassign it now.  But don't do this
+          * for input/output regs since they get remapped while
+          * const buffers don't.
+          */
+         src.Index = SrcReg->Index;
+      }
+   }
+
+   return src;
+}
+
+
+static struct ureg_src swizzle_4v( struct ureg_src src,
+                                   const unsigned *swz )
+{
+   return ureg_swizzle( src, swz[0], swz[1], swz[2], swz[3] );
+}
+
+
+/**
+ * Translate a SWZ instruction into a MOV, MUL or MAD instruction.  EG:
+ *
+ *   SWZ dst, src.x-y10 
+ * 
+ * becomes:
+ *
+ *   MAD dst {1,-1,0,0}, src.xyxx, {0,0,1,0}
+ */
+static void emit_swz( struct st_translate *t,
+                      struct ureg_dst dst,
+                      const struct prog_src_register *SrcReg )
+{
+   struct ureg_program *ureg = t->ureg;
+   struct ureg_src src = src_register( t, SrcReg->File, SrcReg->Index );
+
+   unsigned negate_mask =  SrcReg->Negate;
+
+   unsigned one_mask = ((GET_SWZ(SrcReg->Swizzle, 0) == SWIZZLE_ONE) << 0 |
+                        (GET_SWZ(SrcReg->Swizzle, 1) == SWIZZLE_ONE) << 1 |
+                        (GET_SWZ(SrcReg->Swizzle, 2) == SWIZZLE_ONE) << 2 |
+                        (GET_SWZ(SrcReg->Swizzle, 3) == SWIZZLE_ONE) << 3);
+
+   unsigned zero_mask = ((GET_SWZ(SrcReg->Swizzle, 0) == SWIZZLE_ZERO) << 0 |
+                         (GET_SWZ(SrcReg->Swizzle, 1) == SWIZZLE_ZERO) << 1 |
+                         (GET_SWZ(SrcReg->Swizzle, 2) == SWIZZLE_ZERO) << 2 |
+                         (GET_SWZ(SrcReg->Swizzle, 3) == SWIZZLE_ZERO) << 3);
+
+   unsigned negative_one_mask = one_mask & negate_mask;
+   unsigned positive_one_mask = one_mask & ~negate_mask;
+   
+   struct ureg_src imm;
+   unsigned i;
+   unsigned mul_swizzle[4] = {0,0,0,0};
+   unsigned add_swizzle[4] = {0,0,0,0};
+   unsigned src_swizzle[4] = {0,0,0,0};
+   boolean need_add = FALSE;
+   boolean need_mul = FALSE;
+
+   if (dst.WriteMask == 0)
+      return;
+
+   /* Is this just a MOV?
+    */
+   if (zero_mask == 0 &&
+       one_mask == 0 &&
+       (negate_mask == 0 || negate_mask == TGSI_WRITEMASK_XYZW)) 
+   {
+      ureg_MOV( ureg, dst, translate_src( t, SrcReg ));
+      return;
+   }
+
+#define IMM_ZERO    0
+#define IMM_ONE     1
+#define IMM_NEG_ONE 2
+
+   imm = ureg_imm3f( ureg, 0, 1, -1 );
+
+   for (i = 0; i < 4; i++) {
+      unsigned bit = 1 << i;
+
+      if (dst.WriteMask & bit) {
+         if (positive_one_mask & bit) {
+            mul_swizzle[i] = IMM_ZERO;
+            add_swizzle[i] = IMM_ONE;
+            need_add = TRUE;
+         }
+         else if (negative_one_mask & bit) {
+            mul_swizzle[i] = IMM_ZERO;
+            add_swizzle[i] = IMM_NEG_ONE;
+            need_add = TRUE;
+         }
+         else if (zero_mask & bit) {
+            mul_swizzle[i] = IMM_ZERO;
+            add_swizzle[i] = IMM_ZERO;
+            need_add = TRUE;
+         }
+         else {
+            add_swizzle[i] = IMM_ZERO;
+            src_swizzle[i] = GET_SWZ(SrcReg->Swizzle, i);
+            need_mul = TRUE;
+            if (negate_mask & bit) {
+               mul_swizzle[i] = IMM_NEG_ONE;
+            }
+            else {
+               mul_swizzle[i] = IMM_ONE;
+            }
+         }
+      }
+   }
+
+   if (need_mul && need_add) {
+      ureg_MAD( ureg, 
+                dst,
+                swizzle_4v( src, src_swizzle ),
+                swizzle_4v( imm, mul_swizzle ),
+                swizzle_4v( imm, add_swizzle ) );
+   }
+   else if (need_mul) {
+      ureg_MUL( ureg, 
+                dst,
+                swizzle_4v( src, src_swizzle ),
+                swizzle_4v( imm, mul_swizzle ) );
+   }
+   else if (need_add) {
+      ureg_MOV( ureg, 
+                dst,
+                swizzle_4v( imm, add_swizzle ) );
+   }
+   else {
+      debug_assert(0);
+   }
+
+#undef IMM_ZERO
+#undef IMM_ONE
+#undef IMM_NEG_ONE
+}
+
+
+/**
+ * Negate the value of DDY to match GL semantics where (0,0) is the
+ * lower-left corner of the window.
+ * Note that the GL_ARB_fragment_coord_conventions extension will
+ * effect this someday.
+ */
+static void emit_ddy( struct st_translate *t,
+                      struct ureg_dst dst,
+                      const struct prog_src_register *SrcReg )
+{
+   struct ureg_program *ureg = t->ureg;
+   struct ureg_src src = translate_src( t, SrcReg );
+   src = ureg_negate( src );
+   ureg_DDY( ureg, dst, src );
+}
+
+
+
+static unsigned
+translate_opcode( unsigned op )
+{
+   switch( op ) {
+   case OPCODE_ARL:
+      return TGSI_OPCODE_ARL;
+   case OPCODE_ABS:
+      return TGSI_OPCODE_ABS;
+   case OPCODE_ADD:
+      return TGSI_OPCODE_ADD;
+   case OPCODE_BGNLOOP:
+      return TGSI_OPCODE_BGNLOOP;
+   case OPCODE_BGNSUB:
+      return TGSI_OPCODE_BGNSUB;
+   case OPCODE_BRA:
+      return TGSI_OPCODE_BRA;
+   case OPCODE_BRK:
+      return TGSI_OPCODE_BRK;
+   case OPCODE_CAL:
+      return TGSI_OPCODE_CAL;
+   case OPCODE_CMP:
+      return TGSI_OPCODE_CMP;
+   case OPCODE_CONT:
+      return TGSI_OPCODE_CONT;
+   case OPCODE_COS:
+      return TGSI_OPCODE_COS;
+   case OPCODE_DDX:
+      return TGSI_OPCODE_DDX;
+   case OPCODE_DDY:
+      return TGSI_OPCODE_DDY;
+   case OPCODE_DP2:
+      return TGSI_OPCODE_DP2;
+   case OPCODE_DP2A:
+      return TGSI_OPCODE_DP2A;
+   case OPCODE_DP3:
+      return TGSI_OPCODE_DP3;
+   case OPCODE_DP4:
+      return TGSI_OPCODE_DP4;
+   case OPCODE_DPH:
+      return TGSI_OPCODE_DPH;
+   case OPCODE_DST:
+      return TGSI_OPCODE_DST;
+   case OPCODE_ELSE:
+      return TGSI_OPCODE_ELSE;
+   case OPCODE_EMIT_VERTEX:
+      return TGSI_OPCODE_EMIT;
+   case OPCODE_END_PRIMITIVE:
+      return TGSI_OPCODE_ENDPRIM;
+   case OPCODE_ENDIF:
+      return TGSI_OPCODE_ENDIF;
+   case OPCODE_ENDLOOP:
+      return TGSI_OPCODE_ENDLOOP;
+   case OPCODE_ENDSUB:
+      return TGSI_OPCODE_ENDSUB;
+   case OPCODE_EX2:
+      return TGSI_OPCODE_EX2;
+   case OPCODE_EXP:
+      return TGSI_OPCODE_EXP;
+   case OPCODE_FLR:
+      return TGSI_OPCODE_FLR;
+   case OPCODE_FRC:
+      return TGSI_OPCODE_FRC;
+   case OPCODE_IF:
+      return TGSI_OPCODE_IF;
+   case OPCODE_TRUNC:
+      return TGSI_OPCODE_TRUNC;
+   case OPCODE_KIL:
+      return TGSI_OPCODE_KIL;
+   case OPCODE_KIL_NV:
+      return TGSI_OPCODE_KILP;
+   case OPCODE_LG2:
+      return TGSI_OPCODE_LG2;
+   case OPCODE_LOG:
+      return TGSI_OPCODE_LOG;
+   case OPCODE_LIT:
+      return TGSI_OPCODE_LIT;
+   case OPCODE_LRP:
+      return TGSI_OPCODE_LRP;
+   case OPCODE_MAD:
+      return TGSI_OPCODE_MAD;
+   case OPCODE_MAX:
+      return TGSI_OPCODE_MAX;
+   case OPCODE_MIN:
+      return TGSI_OPCODE_MIN;
+   case OPCODE_MOV:
+      return TGSI_OPCODE_MOV;
+   case OPCODE_MUL:
+      return TGSI_OPCODE_MUL;
+   case OPCODE_NOP:
+      return TGSI_OPCODE_NOP;
+   case OPCODE_NRM3:
+      return TGSI_OPCODE_NRM;
+   case OPCODE_NRM4:
+      return TGSI_OPCODE_NRM4;
+   case OPCODE_POW:
+      return TGSI_OPCODE_POW;
+   case OPCODE_RCP:
+      return TGSI_OPCODE_RCP;
+   case OPCODE_RET:
+      return TGSI_OPCODE_RET;
+   case OPCODE_RSQ:
+      return TGSI_OPCODE_RSQ;
+   case OPCODE_SCS:
+      return TGSI_OPCODE_SCS;
+   case OPCODE_SEQ:
+      return TGSI_OPCODE_SEQ;
+   case OPCODE_SGE:
+      return TGSI_OPCODE_SGE;
+   case OPCODE_SGT:
+      return TGSI_OPCODE_SGT;
+   case OPCODE_SIN:
+      return TGSI_OPCODE_SIN;
+   case OPCODE_SLE:
+      return TGSI_OPCODE_SLE;
+   case OPCODE_SLT:
+      return TGSI_OPCODE_SLT;
+   case OPCODE_SNE:
+      return TGSI_OPCODE_SNE;
+   case OPCODE_SSG:
+      return TGSI_OPCODE_SSG;
+   case OPCODE_SUB:
+      return TGSI_OPCODE_SUB;
+   case OPCODE_TEX:
+      return TGSI_OPCODE_TEX;
+   case OPCODE_TXB:
+      return TGSI_OPCODE_TXB;
+   case OPCODE_TXD:
+      return TGSI_OPCODE_TXD;
+   case OPCODE_TXL:
+      return TGSI_OPCODE_TXL;
+   case OPCODE_TXP:
+      return TGSI_OPCODE_TXP;
+   case OPCODE_XPD:
+      return TGSI_OPCODE_XPD;
+   case OPCODE_END:
+      return TGSI_OPCODE_END;
+   default:
+      debug_assert( 0 );
+      return TGSI_OPCODE_NOP;
+   }
+}
+
+
+static void
+compile_instruction(
+   struct st_translate *t,
+   const struct prog_instruction *inst )
+{
+   struct ureg_program *ureg = t->ureg;
+   GLuint i;
+   struct ureg_dst dst[1];
+   struct ureg_src src[4];
+   unsigned num_dst;
+   unsigned num_src;
+
+   num_dst = _mesa_num_inst_dst_regs( inst->Opcode );
+   num_src = _mesa_num_inst_src_regs( inst->Opcode );
+
+   if (num_dst) 
+      dst[0] = translate_dst( t, 
+                              &inst->DstReg,
+                              inst->SaturateMode );
+
+   for (i = 0; i < num_src; i++) 
+      src[i] = translate_src( t, &inst->SrcReg[i] );
+
+   switch( inst->Opcode ) {
+   case OPCODE_SWZ:
+      emit_swz( t, dst[0], &inst->SrcReg[0] );
+      return;
+
+   case OPCODE_BGNLOOP:
+   case OPCODE_CAL:
+   case OPCODE_ELSE:
+   case OPCODE_ENDLOOP:
+   case OPCODE_IF:
+      debug_assert(num_dst == 0);
+      ureg_label_insn( ureg,
+                       translate_opcode( inst->Opcode ),
+                       src, num_src,
+                       get_label( t, inst->BranchTarget ));
+      return;
+
+   case OPCODE_TEX:
+   case OPCODE_TXB:
+   case OPCODE_TXD:
+   case OPCODE_TXL:
+   case OPCODE_TXP:
+      src[num_src++] = t->samplers[inst->TexSrcUnit];
+      ureg_tex_insn( ureg,
+                     translate_opcode( inst->Opcode ),
+                     dst, num_dst, 
+                     translate_texture_target( inst->TexSrcTarget,
+                                               inst->TexShadow ),
+                     src, num_src );
+      return;
+
+   case OPCODE_SCS:
+      dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XY );
+      ureg_insn( ureg, 
+                 translate_opcode( inst->Opcode ), 
+                 dst, num_dst, 
+                 src, num_src );
+      break;
+
+   case OPCODE_XPD:
+      dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XYZ );
+      ureg_insn( ureg, 
+                 translate_opcode( inst->Opcode ), 
+                 dst, num_dst, 
+                 src, num_src );
+      break;
+
+   case OPCODE_NOISE1:
+   case OPCODE_NOISE2:
+   case OPCODE_NOISE3:
+   case OPCODE_NOISE4:
+      /* At some point, a motivated person could add a better
+       * implementation of noise.  Currently not even the nvidia
+       * binary drivers do anything more than this.  In any case, the
+       * place to do this is in the GL state tracker, not the poor
+       * driver.
+       */
+      ureg_MOV( ureg, dst[0], ureg_imm1f(ureg, 0.5) );
+      break;
+		 
+   case OPCODE_DDY:
+      emit_ddy( t, dst[0], &inst->SrcReg[0] );
+      break;
+
+   default:
+      ureg_insn( ureg, 
+                 translate_opcode( inst->Opcode ), 
+                 dst, num_dst, 
+                 src, num_src );
+      break;
+   }
+}
+
+
+/**
+ * Emit the TGSI instructions to adjust the WPOS pixel center convention
+ * Basically, add (adjX, adjY) to the fragment position.
+ */
+static void
+emit_adjusted_wpos( struct st_translate *t,
+                    const struct gl_program *program,
+                    GLfloat adjX, GLfloat adjY)
+{
+   struct ureg_program *ureg = t->ureg;
+   struct ureg_dst wpos_temp = ureg_DECL_temporary(ureg);
+   struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]];
+
+   /* Note that we bias X and Y and pass Z and W through unchanged.
+    * The shader might also use gl_FragCoord.w and .z.
+    */
+   ureg_ADD(ureg, wpos_temp, wpos_input,
+            ureg_imm4f(ureg, adjX, adjY, 0.0f, 0.0f));
+
+   t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]] = ureg_src(wpos_temp);
+}
+
+
+/**
+ * Emit the TGSI instructions for inverting the WPOS y coordinate.
+ * This code is unavoidable because it also depends on whether
+ * a FBO is bound (STATE_FB_WPOS_Y_TRANSFORM).
+ */
+static void
+emit_wpos_inversion( struct st_translate *t,
+                     const struct gl_program *program,
+                     boolean invert)
+{
+   struct ureg_program *ureg = t->ureg;
+
+   /* 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 wposTransformState[STATE_LENGTH]
+      = { STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM, 0, 0, 0 };
+   
+   /* XXX: note we are modifying the incoming shader here!  Need to
+    * do this before emitting the constant decls below, or this
+    * will be missed:
+    */
+   unsigned wposTransConst = _mesa_add_state_reference(program->Parameters,
+                                                       wposTransformState);
+
+   struct ureg_src wpostrans = ureg_DECL_constant( ureg, wposTransConst );
+   struct ureg_dst wpos_temp;
+   struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]];
+
+   /* MOV wpos_temp, input[wpos]
+    */
+   if (wpos_input.File == TGSI_FILE_TEMPORARY)
+      wpos_temp = ureg_dst(wpos_input);
+   else {
+      wpos_temp = ureg_DECL_temporary( ureg );
+      ureg_MOV( ureg, wpos_temp, wpos_input );
+   }
+
+   if (invert) {
+      /* MAD wpos_temp.y, wpos_input, wpostrans.xxxx, wpostrans.yyyy
+       */
+      ureg_MAD( ureg,
+                ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ),
+                wpos_input,
+                ureg_scalar(wpostrans, 0),
+                ureg_scalar(wpostrans, 1));
+   } else {
+      /* MAD wpos_temp.y, wpos_input, wpostrans.zzzz, wpostrans.wwww
+       */
+      ureg_MAD( ureg,
+                ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ),
+                wpos_input,
+                ureg_scalar(wpostrans, 2),
+                ureg_scalar(wpostrans, 3));
+   }
+
+   /* Use wpos_temp as position input from here on:
+    */
+   t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]] = ureg_src(wpos_temp);
+}
+
+
+/**
+ * Emit fragment position/ooordinate code.
+ */
+static void
+emit_wpos(struct st_context *st,
+          struct st_translate *t,
+          const struct gl_program *program,
+          struct ureg_program *ureg)
+{
+   const struct gl_fragment_program *fp =
+      (const struct gl_fragment_program *) program;
+   struct pipe_screen *pscreen = st->pipe->screen;
+   boolean invert = FALSE;
+
+   if (fp->OriginUpperLeft) {
+      /* Fragment shader wants origin in upper-left */
+      if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT)) {
+         /* the driver supports upper-left origin */
+      }
+      else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT)) {
+         /* the driver supports lower-left origin, need to invert Y */
+         ureg_property_fs_coord_origin(ureg, TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
+         invert = TRUE;
+      }
+      else
+         assert(0);
+   }
+   else {
+      /* Fragment shader wants origin in lower-left */
+      if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT))
+         /* the driver supports lower-left origin */
+         ureg_property_fs_coord_origin(ureg, TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
+      else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT))
+         /* the driver supports upper-left origin, need to invert Y */
+         invert = TRUE;
+      else
+         assert(0);
+   }
+   
+   if (fp->PixelCenterInteger) {
+      /* Fragment shader wants pixel center integer */
+      if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER))
+         /* the driver supports pixel center integer */
+         ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
+      else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER))
+         /* the driver supports pixel center half integer, need to bias X,Y */
+         emit_adjusted_wpos(t, program, 0.5f, invert ? 0.5f : -0.5f);
+      else
+         assert(0);
+   }
+   else {
+      /* Fragment shader wants pixel center half integer */
+      if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER)) {
+         /* the driver supports pixel center half integer */
+      }
+      else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) {
+         /* the driver supports pixel center integer, need to bias X,Y */
+         ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
+         emit_adjusted_wpos(t, program, 0.5f, invert ? -0.5f : 0.5f);
+      }
+      else
+         assert(0);
+   }
+
+   /* we invert after adjustment so that we avoid the MOV to temporary,
+    * and reuse the adjustment ADD instead */
+   emit_wpos_inversion(t, program, invert);
+}
+
+
+/**
+ * OpenGL's fragment gl_FrontFace input is 1 for front-facing, 0 for back.
+ * TGSI uses +1 for front, -1 for back.
+ * This function converts the TGSI value to the GL value.  Simply clamping/
+ * saturating the value to [0,1] does the job.
+ */
+static void
+emit_face_var( struct st_translate *t,
+               const struct gl_program *program )
+{
+   struct ureg_program *ureg = t->ureg;
+   struct ureg_dst face_temp = ureg_DECL_temporary( ureg );
+   struct ureg_src face_input = t->inputs[t->inputMapping[FRAG_ATTRIB_FACE]];
+
+   /* MOV_SAT face_temp, input[face]
+    */
+   face_temp = ureg_saturate( face_temp );
+   ureg_MOV( ureg, face_temp, face_input );
+
+   /* Use face_temp as face input from here on:
+    */
+   t->inputs[t->inputMapping[FRAG_ATTRIB_FACE]] = ureg_src(face_temp);
+}
+
+
+static void
+emit_edgeflags( struct st_translate *t,
+                 const struct gl_program *program )
+{
+   struct ureg_program *ureg = t->ureg;
+   struct ureg_dst edge_dst = t->outputs[t->outputMapping[VERT_RESULT_EDGE]];
+   struct ureg_src edge_src = t->inputs[t->inputMapping[VERT_ATTRIB_EDGEFLAG]];
+
+   ureg_MOV( ureg, edge_dst, edge_src );
+}
+
+
+/**
+ * Translate Mesa program to TGSI format.
+ * \param program  the program to translate
+ * \param numInputs  number of input registers used
+ * \param inputMapping  maps Mesa fragment program inputs to TGSI generic
+ *                      input indexes
+ * \param inputSemanticName  the TGSI_SEMANTIC flag for each input
+ * \param inputSemanticIndex  the semantic index (ex: which texcoord) for
+ *                            each input
+ * \param interpMode  the TGSI_INTERPOLATE_LINEAR/PERSP mode for each input
+ * \param numOutputs  number of output registers used
+ * \param outputMapping  maps Mesa fragment program outputs to TGSI
+ *                       generic outputs
+ * \param outputSemanticName  the TGSI_SEMANTIC flag for each output
+ * \param outputSemanticIndex  the semantic index (ex: which texcoord) for
+ *                             each output
+ *
+ * \return  PIPE_OK or PIPE_ERROR_OUT_OF_MEMORY
+ */
+enum pipe_error
+st_translate_mesa_program(
+   struct gl_context *ctx,
+   uint procType,
+   struct ureg_program *ureg,
+   const struct gl_program *program,
+   GLuint numInputs,
+   const GLuint inputMapping[],
+   const ubyte inputSemanticName[],
+   const ubyte inputSemanticIndex[],
+   const GLuint interpMode[],
+   GLuint numOutputs,
+   const GLuint outputMapping[],
+   const ubyte outputSemanticName[],
+   const ubyte outputSemanticIndex[],
+   boolean passthrough_edgeflags )
+{
+   struct st_translate translate, *t;
+   unsigned i;
+   enum pipe_error ret = PIPE_OK;
+
+   assert(numInputs <= Elements(t->inputs));
+   assert(numOutputs <= Elements(t->outputs));
+
+   t = &translate;
+   memset(t, 0, sizeof *t);
+
+   t->procType = procType;
+   t->inputMapping = inputMapping;
+   t->outputMapping = outputMapping;
+   t->ureg = ureg;
+   t->pointSizeOutIndex = -1;
+   t->prevInstWrotePointSize = GL_FALSE;
+
+   /*_mesa_print_program(program);*/
+
+   /*
+    * Declare input attributes.
+    */
+   if (procType == TGSI_PROCESSOR_FRAGMENT) {
+      for (i = 0; i < numInputs; i++) {
+         if (program->InputFlags[0] & PROG_PARAM_BIT_CYL_WRAP) {
+            t->inputs[i] = ureg_DECL_fs_input_cyl(ureg,
+                                                  inputSemanticName[i],
+                                                  inputSemanticIndex[i],
+                                                  interpMode[i],
+                                                  TGSI_CYLINDRICAL_WRAP_X);
+         }
+         else {
+            t->inputs[i] = ureg_DECL_fs_input(ureg,
+                                              inputSemanticName[i],
+                                              inputSemanticIndex[i],
+                                              interpMode[i]);
+         }
+      }
+
+      if (program->InputsRead & FRAG_BIT_WPOS) {
+         /* Must do this after setting up t->inputs, and before
+          * emitting constant references, below:
+          */
+         emit_wpos(st_context(ctx), t, program, ureg);
+      }
+
+      if (program->InputsRead & FRAG_BIT_FACE) {
+         emit_face_var( t, program );
+      }
+
+      /*
+       * Declare output attributes.
+       */
+      for (i = 0; i < numOutputs; i++) {
+         switch (outputSemanticName[i]) {
+         case TGSI_SEMANTIC_POSITION:
+            t->outputs[i] = ureg_DECL_output( ureg,
+                                              TGSI_SEMANTIC_POSITION, /* Z / Depth */
+                                              outputSemanticIndex[i] );
+
+            t->outputs[i] = ureg_writemask( t->outputs[i],
+                                            TGSI_WRITEMASK_Z );
+            break;
+         case TGSI_SEMANTIC_STENCIL:
+            t->outputs[i] = ureg_DECL_output( ureg,
+                                              TGSI_SEMANTIC_STENCIL, /* Stencil */
+                                              outputSemanticIndex[i] );
+            t->outputs[i] = ureg_writemask( t->outputs[i],
+                                            TGSI_WRITEMASK_Y );
+            break;
+         case TGSI_SEMANTIC_COLOR:
+            t->outputs[i] = ureg_DECL_output( ureg,
+                                              TGSI_SEMANTIC_COLOR,
+                                              outputSemanticIndex[i] );
+            break;
+         default:
+            debug_assert(0);
+            return 0;
+         }
+      }
+   }
+   else if (procType == TGSI_PROCESSOR_GEOMETRY) {
+      for (i = 0; i < numInputs; i++) {
+         t->inputs[i] = ureg_DECL_gs_input(ureg,
+                                           i,
+                                           inputSemanticName[i],
+                                           inputSemanticIndex[i]);
+      }
+
+      for (i = 0; i < numOutputs; i++) {
+         t->outputs[i] = ureg_DECL_output( ureg,
+                                           outputSemanticName[i],
+                                           outputSemanticIndex[i] );
+      }
+   }
+   else {
+      assert(procType == TGSI_PROCESSOR_VERTEX);
+
+      for (i = 0; i < numInputs; i++) {
+         t->inputs[i] = ureg_DECL_vs_input(ureg, i);
+      }
+
+      for (i = 0; i < numOutputs; i++) {
+         t->outputs[i] = ureg_DECL_output( ureg,
+                                           outputSemanticName[i],
+                                           outputSemanticIndex[i] );
+         if ((outputSemanticName[i] == TGSI_SEMANTIC_PSIZE) && program->Id) {
+            /* Writing to the point size result register requires special
+             * handling to implement clamping.
+             */
+            static const gl_state_index pointSizeClampState[STATE_LENGTH]
+               = { STATE_INTERNAL, STATE_POINT_SIZE_IMPL_CLAMP, 0, 0, 0 };
+               /* XXX: note we are modifying the incoming shader here!  Need to
+               * do this before emitting the constant decls below, or this
+               * will be missed:
+               */
+            unsigned pointSizeClampConst =
+               _mesa_add_state_reference(program->Parameters,
+                                         pointSizeClampState);
+            struct ureg_dst psizregtemp = ureg_DECL_temporary( ureg );
+            t->pointSizeConst = ureg_DECL_constant( ureg, pointSizeClampConst );
+            t->pointSizeResult = t->outputs[i];
+            t->pointSizeOutIndex = i;
+            t->outputs[i] = psizregtemp;
+         }
+      }
+      if (passthrough_edgeflags)
+         emit_edgeflags( t, program );
+   }
+
+   /* Declare address register.
+    */
+   if (program->NumAddressRegs > 0) {
+      debug_assert( program->NumAddressRegs == 1 );
+      t->address[0] = ureg_DECL_address( ureg );
+   }
+
+   /* Declare misc input registers
+    */
+   {
+      GLbitfield sysInputs = program->SystemValuesRead;
+      unsigned numSys = 0;
+      for (i = 0; sysInputs; i++) {
+         if (sysInputs & (1 << i)) {
+            unsigned semName = mesa_sysval_to_semantic[i];
+            t->systemValues[i] = ureg_DECL_system_value(ureg, numSys, semName, 0);
+            numSys++;
+            sysInputs &= ~(1 << i);
+         }
+      }
+   }
+
+   if (program->IndirectRegisterFiles & (1 << PROGRAM_TEMPORARY)) {
+      /* If temps are accessed with indirect addressing, declare temporaries
+       * in sequential order.  Else, we declare them on demand elsewhere.
+       */
+      for (i = 0; i < program->NumTemporaries; i++) {
+         /* XXX use TGSI_FILE_TEMPORARY_ARRAY when it's supported by ureg */
+         t->temps[i] = ureg_DECL_temporary( t->ureg );
+      }
+   }
+
+   /* Emit constants and immediates.  Mesa uses a single index space
+    * for these, so we put all the translated regs in t->constants.
+    */
+   if (program->Parameters) {
+      t->constants = CALLOC( program->Parameters->NumParameters,
+                             sizeof t->constants[0] );
+      if (t->constants == NULL) {
+         ret = PIPE_ERROR_OUT_OF_MEMORY;
+         goto out;
+      }
+
+      for (i = 0; i < program->Parameters->NumParameters; i++) {
+         switch (program->Parameters->Parameters[i].Type) {
+         case PROGRAM_ENV_PARAM:
+         case PROGRAM_LOCAL_PARAM:
+         case PROGRAM_STATE_VAR:
+         case PROGRAM_NAMED_PARAM:
+         case PROGRAM_UNIFORM:
+            t->constants[i] = ureg_DECL_constant( ureg, i );
+            break;
+
+            /* Emit immediates only when there's no indirect addressing of
+             * the const buffer.
+             * FIXME: Be smarter and recognize param arrays:
+             * indirect addressing is only valid within the referenced
+             * array.
+             */
+         case PROGRAM_CONSTANT:
+            if (program->IndirectRegisterFiles & PROGRAM_ANY_CONST)
+               t->constants[i] = ureg_DECL_constant( ureg, i );
+            else
+               t->constants[i] = 
+                  ureg_DECL_immediate( ureg,
+                                       program->Parameters->ParameterValues[i],
+                                       4 );
+            break;
+         default:
+            break;
+         }
+      }
+   }
+
+   /* texture samplers */
+   for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) {
+      if (program->SamplersUsed & (1 << i)) {
+         t->samplers[i] = ureg_DECL_sampler( ureg, i );
+      }
+   }
+
+   /* Emit each instruction in turn:
+    */
+   for (i = 0; i < program->NumInstructions; i++) {
+      set_insn_start( t, ureg_get_instruction_number( ureg ));
+      compile_instruction( t, &program->Instructions[i] );
+
+      if (t->prevInstWrotePointSize && program->Id) {
+         /* The previous instruction wrote to the (fake) vertex point size
+          * result register.  Now we need to clamp that value to the min/max
+          * point size range, putting the result into the real point size
+          * register.
+          * Note that we can't do this easily at the end of program due to
+          * possible early return.
+          */
+         set_insn_start( t, ureg_get_instruction_number( ureg ));
+         ureg_MAX( t->ureg,
+                   ureg_writemask(t->outputs[t->pointSizeOutIndex], WRITEMASK_X),
+                   ureg_src(t->outputs[t->pointSizeOutIndex]),
+                   ureg_swizzle(t->pointSizeConst, 1,1,1,1));
+         ureg_MIN( t->ureg, ureg_writemask(t->pointSizeResult, WRITEMASK_X),
+                   ureg_src(t->outputs[t->pointSizeOutIndex]),
+                   ureg_swizzle(t->pointSizeConst, 2,2,2,2));
+      }
+      t->prevInstWrotePointSize = GL_FALSE;
+   }
+
+   /* Fix up all emitted labels:
+    */
+   for (i = 0; i < t->labels_count; i++) {
+      ureg_fixup_label( ureg,
+                        t->labels[i].token,
+                        t->insn[t->labels[i].branch_target] );
+   }
+
+out:
+   FREE(t->insn);
+   FREE(t->labels);
+   FREE(t->constants);
+
+   if (t->error) {
+      debug_printf("%s: translate error flag set\n", __FUNCTION__);
+   }
+
+   return ret;
+}
+
+
+/**
+ * Tokens cannot be free with free otherwise the builtin gallium
+ * malloc debugging will get confused.
+ */
+void
+st_free_tokens(const struct tgsi_token *tokens)
+{
+   FREE((void *)tokens);
+}