- Merge from trunk into branchsarea-1-0-0-20010307

121 files changed, 9005 insertions, 8920 deletions

diff --git a/xc/config/cf/host.def b/xc/config/cf/host.def
index 99628a34a..bdf4e48e4 100644
--- a/xc/config/cf/host.def
+++ b/xc/config/cf/host.def
@@ -7,7 +7,7 @@
 #define LibraryCDebugFlags -O2
 #endif
 #define BuildServersOnly YES
-#define XF86CardDrivers vga i810 mga ati glint
+#define XF86CardDrivers vga tdfx i810 mga ati glint
 #define LinuxDistribution LinuxRedHat
 #define GccWarningOptions -Wall -Wpointer-arith -Wstrict-prototypes \
                           -Wmissing-prototypes -Wmissing-declarations \
@@ -16,7 +16,7 @@
 #define NormalLibGlx NO
 
 #define BuildXF86DRI YES
-#define HasGlide3 NO
+#define HasGlide3 YES
 
 #ifdef i386Architecture
 #define MesaUse3DNow YES
@@ -45,11 +45,11 @@
 /* #define DoLoadableServer NO */
 
 /* Optionally turn this on to change the place where you install the build */
-#define ProjectRoot /mnt/src/work/binary/
+/* #define ProjectRoot /usr/X11R6-DRI */
 
 /* Optionally turn this on to force the kernel modules to build */
+/* #define BuildXF86DRM YES */
 
-#define BuildXF86DRM NO
 #define XnestServer NO
 #define XVirtualFramebufferServer NO
 #define XprtServer NO
diff --git a/xc/extras/Mesa/src/image.c b/xc/extras/Mesa/src/image.c
index c0f463c39..5ffb84dce 100644
--- a/xc/extras/Mesa/src/image.c
+++ b/xc/extras/Mesa/src/image.c
@@ -2,19 +2,19 @@
 /*
  * Mesa 3-D graphics library
  * Version:  3.4.1
- * 
+ *
  * Copyright (C) 1999-2000  Brian Paul   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 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
@@ -1545,7 +1545,6 @@ _mesa_pack_rgba_span( GLcontext *ctx,
       _mesa_pack_float_rgba_span(ctx, n, (const GLfloat (*)[4]) rgba,
                                  format, type, destination,
                                  packing, applyTransferOps);
-
    }
 }
 
@@ -1811,7 +1810,7 @@ extract_float_rgba(GLuint n, GLfloat rgba[][4],
          alphaIndex = 0;
          stride = 1;
          break;
-      case GL_LUMINANCE: 
+      case GL_LUMINANCE:
          redIndex = greenIndex = blueIndex = 0;
          alphaIndex = -1;
          stride = 1;
@@ -2159,7 +2158,7 @@ extract_float_rgba(GLuint n, GLfloat rgba[][4],
             GLuint i;
             for (i = 0; i < n; i ++) {
                GLuint p = uisrc[i];
-               rgba[i][rComp] = UBYTE_COLOR_TO_FLOAT_COLOR((p >> 24)       ); 
+               rgba[i][rComp] = UBYTE_COLOR_TO_FLOAT_COLOR((p >> 24)       );
                rgba[i][gComp] = UBYTE_COLOR_TO_FLOAT_COLOR((p >> 16) & 0xff);
                rgba[i][bComp] = UBYTE_COLOR_TO_FLOAT_COLOR((p >>  8) & 0xff);
                rgba[i][aComp] = UBYTE_COLOR_TO_FLOAT_COLOR((p      ) & 0xff);
@@ -2173,7 +2172,7 @@ extract_float_rgba(GLuint n, GLfloat rgba[][4],
                rgba[i][rComp] = UBYTE_COLOR_TO_FLOAT_COLOR((p      ) & 0xff);
                rgba[i][gComp] = UBYTE_COLOR_TO_FLOAT_COLOR((p >>  8) & 0xff);
                rgba[i][bComp] = UBYTE_COLOR_TO_FLOAT_COLOR((p >> 16) & 0xff);
-               rgba[i][aComp] = UBYTE_COLOR_TO_FLOAT_COLOR((p >> 24)       ); 
+               rgba[i][aComp] = UBYTE_COLOR_TO_FLOAT_COLOR((p >> 24)       );
             }
          }
          break;
@@ -2262,7 +2261,7 @@ _mesa_unpack_ubyte_color_span( GLcontext *ctx,
                                GLboolean applyTransferOps )
 {
    ASSERT(dstFormat == GL_ALPHA ||
-          dstFormat == GL_LUMINANCE || 
+          dstFormat == GL_LUMINANCE ||
           dstFormat == GL_LUMINANCE_ALPHA ||
           dstFormat == GL_INTENSITY ||
           dstFormat == GL_RGB ||
@@ -2474,7 +2473,7 @@ _mesa_unpack_ubyte_color_span( GLcontext *ctx,
             dstRedIndex = dstGreenIndex = dstBlueIndex = -1;
             dstLuminanceIndex = dstIntensityIndex = -1;
             break;
-         case GL_LUMINANCE: 
+         case GL_LUMINANCE:
             dstLuminanceIndex = 0;
             dstRedIndex = dstGreenIndex = dstBlueIndex = dstAlphaIndex = -1;
             dstIntensityIndex = -1;
@@ -2581,7 +2580,7 @@ _mesa_unpack_float_color_span( GLcontext *ctx,
                                GLboolean applyTransferOps, GLboolean clamp )
 {
    ASSERT(dstFormat == GL_ALPHA ||
-          dstFormat == GL_LUMINANCE || 
+          dstFormat == GL_LUMINANCE ||
           dstFormat == GL_LUMINANCE_ALPHA ||
           dstFormat == GL_INTENSITY ||
           dstFormat == GL_RGB ||
@@ -2743,7 +2742,7 @@ _mesa_unpack_float_color_span( GLcontext *ctx,
             dstRedIndex = dstGreenIndex = dstBlueIndex = -1;
             dstLuminanceIndex = dstIntensityIndex = -1;
             break;
-         case GL_LUMINANCE: 
+         case GL_LUMINANCE:
             dstLuminanceIndex = 0;
             dstRedIndex = dstGreenIndex = dstBlueIndex = dstAlphaIndex = -1;
             dstIntensityIndex = -1;
diff --git a/xc/extras/Mesa/src/macros.h b/xc/extras/Mesa/src/macros.h
index 16038d3b1..8010e28bd 100644
--- a/xc/extras/Mesa/src/macros.h
+++ b/xc/extras/Mesa/src/macros.h
@@ -2,19 +2,19 @@
 /*
  * Mesa 3-D graphics library
  * Version:  3.3
- * 
+ *
  * Copyright (C) 1999  Brian Paul   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 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
@@ -89,7 +89,7 @@
 #define TEST_BITS(WORD, BITS)   ((WORD) & (BITS))
 
 
-/* Stepping a GLfloat pointer by a byte stride 
+/* Stepping a GLfloat pointer by a byte stride
  */
 #define STRIDE_F(p, i)  (p = (GLfloat *)((GLubyte *)p + i))
 #define STRIDE_UI(p, i)  (p = (GLuint *)((GLubyte *)p + i))
@@ -160,7 +160,7 @@ do {						\
    case 2: (DST)[1] = (SRC)[1];			\
    case 1: (DST)[0] = (SRC)[0];			\
    }  						\
-} while(0)			   
+} while(0)
 
 #define SUB_4V( DST, SRCA, SRCB )		\
 do {						\
@@ -496,4 +496,31 @@ do {						\
 #define FLOAT_TO_INT(X)		( (GLint) (2147483647.0 * (X)) )
 
 
+
+/* Generic color packing macros
+ */
+
+#define PACK_COLOR_8888( a, b, c, d )					\
+   (((a) << 24) | ((b) << 16) | ((c) << 8) | (d))
+
+#define PACK_COLOR_888( a, b, c )					\
+   (((a) << 16) | ((b) << 8) | (c))
+
+#define PACK_COLOR_565( a, b, c )					\
+   ((((a) & 0xf8) << 8) | (((b) & 0xfc) << 3) | (((c) & 0xf8) >> 3))
+
+#define PACK_COLOR_1555( a, b, c, d )					\
+   ((((b) & 0xf8) << 7) | (((c) & 0xf8) << 2) | (((d) & 0xf8) >> 3) |	\
+    ((a) ? 0x8000 : 0))
+
+#define PACK_COLOR_4444( a, b, c, d )					\
+   ((((a) & 0xf0) << 8) | (((b) & 0xf0) << 4) | ((c) & 0xf0) | ((d) >> 4))
+
+#define PACK_COLOR_88( a, b )						\
+   (((a) << 8) | (b))
+
+#define PACK_COLOR_332( a, b, c )					\
+   (((a) & 0xe0) | (((b) & 0xe0) >> 3) | (((c) & 0xc0) >> 6))
+
+
 #endif /*MACROS_H*/
diff --git a/xc/extras/Mesa/src/stencil.c b/xc/extras/Mesa/src/stencil.c
index 74e03af38..77755c313 100644
--- a/xc/extras/Mesa/src/stencil.c
+++ b/xc/extras/Mesa/src/stencil.c
@@ -1,9 +1,9 @@
 
 /*
  * Mesa 3-D graphics library
- * Version:  3.4
+ * Version:  3.4.2
  * 
- * Copyright (C) 1999-2000  Brian Paul   All Rights Reserved.
+ * Copyright (C) 1999-2001  Brian Paul   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"),
@@ -1038,47 +1038,45 @@ _mesa_stencil_and_ztest_pixels( GLcontext *ctx,
    if (ctx->Driver.WriteStencilPixels) {
       /*** Hardware stencil buffer ***/
       GLstencil stencil[PB_SIZE];
-      GLubyte mask[PB_SIZE];
+      GLubyte origMask[PB_SIZE];
 
       ASSERT(ctx->Driver.ReadStencilPixels);
       (*ctx->Driver.ReadStencilPixels)(ctx, n, x, y, stencil);
 
+      MEMCPY(origMask, mask, n * sizeof(GLubyte));
 
-      if (do_stencil_test( ctx, n, stencil, mask ) == GL_FALSE) {
-         /* all fragments failed the stencil test, we're done. */
-         return GL_FALSE;
-      }
+      (void) do_stencil_test( ctx, n, stencil, mask );
 
       if (ctx->Depth.Test == GL_FALSE) {
          apply_stencil_op( ctx, ctx->Stencil.ZPassFunc, n, stencil, mask );
       }
       else {
-         GLubyte passmask[PB_SIZE], failmask[PB_SIZE], oldmask[PB_SIZE];
-         GLuint i;
-
-         MEMCPY(oldmask, mask, n * sizeof(GLubyte));
-
          _mesa_depth_test_pixels(ctx, n, x, y, z, mask);
 
-         for (i=0;i<n;i++) {
-            ASSERT(mask[i] == 0 || mask[i] == 1);
-            passmask[i] = oldmask[i] & mask[i];
-            failmask[i] = oldmask[i] & (mask[i] ^ 1);
-         }
-
          if (ctx->Stencil.ZFailFunc != GL_KEEP) {
-            apply_stencil_op( ctx, ctx->Stencil.ZFailFunc, n, stencil, failmask );
+            GLubyte failmask[PB_SIZE];
+            GLuint i;
+            for (i = 0; i < n; i++) {
+               ASSERT(mask[i] == 0 || mask[i] == 1);
+               failmask[i] = origMask[i] & (mask[i] ^ 1);
+            }
+            apply_stencil_op(ctx, ctx->Stencil.ZFailFunc, n, stencil, failmask);
          }
          if (ctx->Stencil.ZPassFunc != GL_KEEP) {
-            apply_stencil_op( ctx, ctx->Stencil.ZPassFunc, n, stencil, passmask );
+            GLubyte passmask[PB_SIZE];
+            GLuint i;
+            for (i = 0; i < n; i++) {
+               ASSERT(mask[i] == 0 || mask[i] == 1);
+               passmask[i] = origMask[i] & mask[i];
+            }
+            apply_stencil_op(ctx, ctx->Stencil.ZPassFunc, n, stencil, passmask);
          }
       }
 
       /* Write updated stencil values into hardware stencil buffer */
-      (ctx->Driver.WriteStencilPixels)(ctx, n, x, y, stencil, mask );
+      (ctx->Driver.WriteStencilPixels)(ctx, n, x, y, stencil, origMask );
 
       return GL_TRUE;
-
    }
    else {
       /*** Software stencil buffer ***/
@@ -1088,7 +1086,6 @@ _mesa_stencil_and_ztest_pixels( GLcontext *ctx,
          return GL_FALSE;
       }
 
-
       if (ctx->Depth.Test==GL_FALSE) {
          apply_stencil_op_to_pixels( ctx, n, x, y, ctx->Stencil.ZPassFunc, mask );
       }
diff --git a/xc/extras/Mesa/src/texformat.c b/xc/extras/Mesa/src/texformat.c
new file mode 100644
index 000000000..f85725843
--- /dev/null
+++ b/xc/extras/Mesa/src/texformat.c
@@ -0,0 +1,351 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.4
+ *
+ * Copyright (C) 1999-2000  Brian Paul   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 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.
+ *
+ * Author:
+ *    Gareth Hughes <gareth@valinux.com>
+ */
+
+#ifdef PC_HEADER
+#include "all.h"
+#else
+#include "glheader.h"
+#include "context.h"
+#include "image.h"
+#include "mem.h"
+#include "mmath.h"
+#include "span.h"
+#include "texformat.h"
+#include "teximage.h"
+#include "texstate.h"
+#include "types.h"
+#endif
+
+
+/* ================================================================
+ * Texture internal formats:
+ */
+
+const struct gl_texture_format _mesa_texformat_rgba8888 = {
+   MESA_FORMAT_RGBA8888,	/* IntFormat */
+   8,				/* RedBits */
+   8,				/* GreenBits */
+   8,				/* BlueBits */
+   8,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   4,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_abgr8888 = {
+   MESA_FORMAT_ABGR8888,	/* IntFormat */
+   8,				/* RedBits */
+   8,				/* GreenBits */
+   8,				/* BlueBits */
+   8,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   4,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_argb8888 = {
+   MESA_FORMAT_ARGB8888,	/* IntFormat */
+   8,				/* RedBits */
+   8,				/* GreenBits */
+   8,				/* BlueBits */
+   8,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   4,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_rgb888 = {
+   MESA_FORMAT_RGB888,		/* IntFormat */
+   8,				/* RedBits */
+   8,				/* GreenBits */
+   8,				/* BlueBits */
+   0,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   3,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_bgr888 = {
+   MESA_FORMAT_BGR888,		/* IntFormat */
+   8,				/* RedBits */
+   8,				/* GreenBits */
+   8,				/* BlueBits */
+   0,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   3,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_rgb565 = {
+   MESA_FORMAT_RGB565,		/* IntFormat */
+   5,				/* RedBits */
+   6,				/* GreenBits */
+   5,				/* BlueBits */
+   0,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   2,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_argb4444 = {
+   MESA_FORMAT_ARGB4444,	/* IntFormat */
+   4,				/* RedBits */
+   4,				/* GreenBits */
+   4,				/* BlueBits */
+   4,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   2,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_argb1555 = {
+   MESA_FORMAT_ARGB1555,	/* IntFormat */
+   5,				/* RedBits */
+   5,				/* GreenBits */
+   5,				/* BlueBits */
+   1,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   2,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_al88 = {
+   MESA_FORMAT_AL88,		/* IntFormat */
+   0,				/* RedBits */
+   0,				/* GreenBits */
+   0,				/* BlueBits */
+   8,				/* AlphaBits */
+   8,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   2,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_rgb332 = {
+   MESA_FORMAT_RGB332,		/* IntFormat */
+   3,				/* RedBits */
+   3,				/* GreenBits */
+   2,				/* BlueBits */
+   0,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   1,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_a8 = {
+   MESA_FORMAT_A8,		/* IntFormat */
+   0,				/* RedBits */
+   0,				/* GreenBits */
+   0,				/* BlueBits */
+   8,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   1,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_l8 = {
+   MESA_FORMAT_L8,		/* IntFormat */
+   0,				/* RedBits */
+   0,				/* GreenBits */
+   0,				/* BlueBits */
+   0,				/* AlphaBits */
+   8,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   1,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_i8 = {
+   MESA_FORMAT_I8,		/* IntFormat */
+   0,				/* RedBits */
+   0,				/* GreenBits */
+   0,				/* BlueBits */
+   0,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   8,				/* IntensityBits */
+   0,				/* IndexBits */
+   1,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_texformat_ci8 = {
+   MESA_FORMAT_CI8,		/* IntFormat */
+   0,				/* RedBits */
+   0,				/* GreenBits */
+   0,				/* BlueBits */
+   0,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   8,				/* IndexBits */
+   1,				/* TexelBytes */
+};
+
+const struct gl_texture_format _mesa_null_texformat = {
+   -1,				/* IntFormat */
+   0,				/* RedBits */
+   0,				/* GreenBits */
+   0,				/* BlueBits */
+   0,				/* AlphaBits */
+   0,				/* LuminanceBits */
+   0,				/* IntensityBits */
+   0,				/* IndexBits */
+   0,				/* TexelBytes */
+};
+
+
+
+/*
+ * Given an internal texture format enum or 1, 2, 3, 4 return the
+ * corresponding _base_ internal format:  GL_ALPHA, GL_LUMINANCE,
+ * GL_LUMANCE_ALPHA, GL_INTENSITY, GL_RGB, or GL_RGBA.
+ * Return -1 if invalid enum.
+ */
+void _mesa_init_texture_format( GLcontext *ctx,
+				struct gl_texture_image *texImage,
+				GLenum internalFormat )
+{
+   texImage->IntFormat = internalFormat;
+
+   /* Ask the driver for the base format, if it doesn't know, it will
+    * return -1;
+    */
+   if ( ctx->Driver.BaseCompressedTexFormat ) {
+      GLint format = 0;			/* Silence compiler warning */
+      format = (*ctx->Driver.BaseCompressedTexFormat)( ctx, format );
+      if ( format >= 0 ) {
+	 internalFormat = format;
+      }
+   }
+
+   switch ( internalFormat ) {
+      /* GH: Bias towards GL_RGB, GL_RGBA texture formats.  This has
+       * got to be better than sticking them way down the end of this
+       * huge list.
+       */
+   case 4:				/* Quake3 uses this... */
+   case GL_RGBA:
+      texImage->Format = GL_RGBA;
+      texImage->TexFormat = &_mesa_texformat_abgr8888;
+      break;
+
+   case 3:				/* ... and this. */
+   case GL_RGB:
+      texImage->Format = GL_RGB;
+      texImage->TexFormat = &_mesa_texformat_bgr888;
+      break;
+
+      /* GH: Okay, keep checking as normal.  Still test for GL_RGB,
+       * GL_RGBA formats first.
+       */
+   case GL_RGBA2:
+   case GL_RGBA4:
+   case GL_RGB5_A1:
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      texImage->Format = GL_RGBA;
+      texImage->TexFormat = &_mesa_texformat_abgr8888;
+      break;
+
+   case GL_R3_G3_B2:
+   case GL_RGB4:
+   case GL_RGB5:
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+      texImage->Format = GL_RGB;
+      texImage->TexFormat = &_mesa_texformat_bgr888;
+      break;
+
+   case GL_ALPHA:
+   case GL_ALPHA4:
+   case GL_ALPHA8:
+   case GL_ALPHA12:
+   case GL_ALPHA16:
+      texImage->Format = GL_ALPHA;
+      texImage->TexFormat = &_mesa_texformat_a8;
+      break;
+
+   case 1:
+   case GL_LUMINANCE:
+   case GL_LUMINANCE4:
+   case GL_LUMINANCE8:
+   case GL_LUMINANCE12:
+   case GL_LUMINANCE16:
+      texImage->Format = GL_LUMINANCE;
+      texImage->TexFormat = &_mesa_texformat_l8;
+      break;
+
+   case 2:
+   case GL_LUMINANCE_ALPHA:
+   case GL_LUMINANCE4_ALPHA4:
+   case GL_LUMINANCE6_ALPHA2:
+   case GL_LUMINANCE8_ALPHA8:
+   case GL_LUMINANCE12_ALPHA4:
+   case GL_LUMINANCE12_ALPHA12:
+   case GL_LUMINANCE16_ALPHA16:
+      texImage->Format = GL_LUMINANCE_ALPHA;
+      texImage->TexFormat = &_mesa_texformat_al88;
+      break;
+
+   case GL_INTENSITY:
+   case GL_INTENSITY4:
+   case GL_INTENSITY8:
+   case GL_INTENSITY12:
+   case GL_INTENSITY16:
+      texImage->Format = GL_INTENSITY;
+      texImage->TexFormat = &_mesa_texformat_i8;
+      break;
+
+   case GL_COLOR_INDEX:
+   case GL_COLOR_INDEX1_EXT:
+   case GL_COLOR_INDEX2_EXT:
+   case GL_COLOR_INDEX4_EXT:
+   case GL_COLOR_INDEX8_EXT:
+   case GL_COLOR_INDEX12_EXT:
+   case GL_COLOR_INDEX16_EXT:
+      texImage->Format = GL_COLOR_INDEX;
+      texImage->TexFormat = &_mesa_texformat_ci8;
+      break;
+
+   default:
+      gl_problem( ctx, "unexpected format in _mesa_init_texture_format" );
+      return;
+   }
+}
diff --git a/xc/extras/Mesa/src/texformat.h b/xc/extras/Mesa/src/texformat.h
new file mode 100644
index 000000000..8f5e7de75
--- /dev/null
+++ b/xc/extras/Mesa/src/texformat.h
@@ -0,0 +1,80 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.4
+ *
+ * Copyright (C) 1999-2000  Brian Paul   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 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.
+ *
+ * Author:
+ *    Gareth Hughes <gareth@valinux.com>
+ */
+
+#ifndef TEXFORMAT_H
+#define TEXFORMAT_H
+
+#include "types.h"
+
+
+/* The Mesa internal texture image types.  These will be set to their
+ * default value, but may be changed by drivers as required.
+ */
+				/* msb <------ TEXEL BITS -----------> lsb */
+enum _format {			/* ---- ---- ---- ---- ---- ---- ---- ---- */
+   MESA_FORMAT_RGBA8888,	/* RRRR RRRR GGGG GGGG BBBB BBBB AAAA AAAA */
+   MESA_FORMAT_ABGR8888,	/* AAAA AAAA BBBB BBBB GGGG GGGG RRRR RRRR */
+   MESA_FORMAT_ARGB8888,	/* AAAA AAAA RRRR RRRR GGGG GGGG BBBB BBBB */
+   MESA_FORMAT_RGB888,		/*           RRRR RRRR GGGG GGGG BBBB BBBB */
+   MESA_FORMAT_BGR888,		/*           BBBB BBBB GGGG GGGG RRRR RRRR */
+   MESA_FORMAT_RGB565,		/*                     RRRR RGGG GGGB BBBB */
+   MESA_FORMAT_ARGB4444,	/*                     AAAA RRRR GGGG BBBB */
+   MESA_FORMAT_ARGB1555,	/*                     ARRR RRGG GGGB BBBB */
+   MESA_FORMAT_AL88,		/*                     AAAA AAAA LLLL LLLL */
+   MESA_FORMAT_RGB332,		/*                               RRRG GGBB */
+   MESA_FORMAT_A8,		/*                               AAAA AAAA */
+   MESA_FORMAT_L8,		/*                               LLLL LLLL */
+   MESA_FORMAT_I8,		/*                               IIII IIII */
+   MESA_FORMAT_CI8		/*                               CCCC CCCC */
+};
+
+
+extern void
+_mesa_init_texture_format( GLcontext *ctx,
+			   struct gl_texture_image *texImage,
+			   GLenum internalFormat );
+
+
+extern const struct gl_texture_format _mesa_texformat_rgba8888;
+extern const struct gl_texture_format _mesa_texformat_abgr8888;
+extern const struct gl_texture_format _mesa_texformat_argb8888;
+extern const struct gl_texture_format _mesa_texformat_rgb888;
+extern const struct gl_texture_format _mesa_texformat_bgr888;
+extern const struct gl_texture_format _mesa_texformat_rgb565;
+extern const struct gl_texture_format _mesa_texformat_argb4444;
+extern const struct gl_texture_format _mesa_texformat_argb1555;
+extern const struct gl_texture_format _mesa_texformat_al88;
+extern const struct gl_texture_format _mesa_texformat_rgb332;
+extern const struct gl_texture_format _mesa_texformat_a8;
+extern const struct gl_texture_format _mesa_texformat_l8;
+extern const struct gl_texture_format _mesa_texformat_i8;
+extern const struct gl_texture_format _mesa_texformat_ci8;
+
+extern const struct gl_texture_format _mesa_null_texformat;
+
+#endif
diff --git a/xc/extras/Mesa/src/teximage.c b/xc/extras/Mesa/src/teximage.c
index 4b71d5178..e1d676dbb 100644
--- a/xc/extras/Mesa/src/teximage.c
+++ b/xc/extras/Mesa/src/teximage.c
@@ -2,19 +2,19 @@
 /*
  * Mesa 3-D graphics library
  * Version:  3.4
- * 
+ *
  * Copyright (C) 1999-2000  Brian Paul   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 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
@@ -33,6 +33,7 @@
 #include "mem.h"
 #include "mmath.h"
 #include "span.h"
+#include "texformat.h"
 #include "teximage.h"
 #include "texstate.h"
 #include "types.h"
@@ -300,115 +301,6 @@ is_compressed_format(GLcontext *ctx, GLenum internalFormat)
 
 
 
-/*
- * Examine the texImage->Format field and set the Red, Green, Blue, etc
- * texel component sizes to default values.
- * These fields are set only here by core Mesa but device drivers may
- * overwritting these fields to indicate true texel resolution.
- */
-static void
-set_teximage_component_sizes( struct gl_texture_image *texImage )
-{
-   switch (texImage->Format) {
-      case GL_ALPHA:
-         texImage->RedBits = 0;
-         texImage->GreenBits = 0;
-         texImage->BlueBits = 0;
-         texImage->AlphaBits = 8;
-         texImage->IntensityBits = 0;
-         texImage->LuminanceBits = 0;
-         texImage->IndexBits = 0;
-         break;
-      case GL_LUMINANCE:
-         texImage->RedBits = 0;
-         texImage->GreenBits = 0;
-         texImage->BlueBits = 0;
-         texImage->AlphaBits = 0;
-         texImage->IntensityBits = 0;
-         texImage->LuminanceBits = 8;
-         texImage->IndexBits = 0;
-         break;
-      case GL_LUMINANCE_ALPHA:
-         texImage->RedBits = 0;
-         texImage->GreenBits = 0;
-         texImage->BlueBits = 0;
-         texImage->AlphaBits = 8;
-         texImage->IntensityBits = 0;
-         texImage->LuminanceBits = 8;
-         texImage->IndexBits = 0;
-         break;
-      case GL_INTENSITY:
-         texImage->RedBits = 0;
-         texImage->GreenBits = 0;
-         texImage->BlueBits = 0;
-         texImage->AlphaBits = 0;
-         texImage->IntensityBits = 8;
-         texImage->LuminanceBits = 0;
-         texImage->IndexBits = 0;
-         break;
-      case GL_RED:
-         texImage->RedBits = 8;
-         texImage->GreenBits = 0;
-         texImage->BlueBits = 0;
-         texImage->AlphaBits = 0;
-         texImage->IntensityBits = 0;
-         texImage->LuminanceBits = 0;
-         texImage->IndexBits = 0;
-         break;
-      case GL_GREEN:
-         texImage->RedBits = 0;
-         texImage->GreenBits = 8;
-         texImage->BlueBits = 0;
-         texImage->AlphaBits = 0;
-         texImage->IntensityBits = 0;
-         texImage->LuminanceBits = 0;
-         texImage->IndexBits = 0;
-         break;
-      case GL_BLUE:
-         texImage->RedBits = 0;
-         texImage->GreenBits = 0;
-         texImage->BlueBits = 8;
-         texImage->AlphaBits = 0;
-         texImage->IntensityBits = 0;
-         texImage->LuminanceBits = 0;
-         texImage->IndexBits = 0;
-         break;
-      case GL_RGB:
-      case GL_BGR:
-         texImage->RedBits = 8;
-         texImage->GreenBits = 8;
-         texImage->BlueBits = 8;
-         texImage->AlphaBits = 0;
-         texImage->IntensityBits = 0;
-         texImage->LuminanceBits = 0;
-         texImage->IndexBits = 0;
-         break;
-      case GL_RGBA:
-      case GL_BGRA:
-      case GL_ABGR_EXT:
-         texImage->RedBits = 8;
-         texImage->GreenBits = 8;
-         texImage->BlueBits = 8;
-         texImage->AlphaBits = 8;
-         texImage->IntensityBits = 0;
-         texImage->LuminanceBits = 0;
-         texImage->IndexBits = 0;
-         break;
-      case GL_COLOR_INDEX:
-         texImage->RedBits = 0;
-         texImage->GreenBits = 0;
-         texImage->BlueBits = 0;
-         texImage->AlphaBits = 0;
-         texImage->IntensityBits = 0;
-         texImage->LuminanceBits = 0;
-         texImage->IndexBits = 8;
-         break;
-      default:
-         gl_problem(NULL, "unexpected format in set_teximage_component_sizes");
-   }
-}
-
-
 static void
 set_tex_image(struct gl_texture_object *tObj,
               GLenum target, GLint level,
@@ -467,9 +359,7 @@ init_texture_image( GLcontext *ctx,
 {
    ASSERT(img);
    ASSERT(!img->Data);
-   img->Format = (GLenum) _mesa_base_tex_format(ctx, internalFormat);
-   set_teximage_component_sizes( img );
-   img->IntFormat = (GLenum) internalFormat;
+   _mesa_init_texture_format( ctx, img, internalFormat );
    img->Border = border;
    img->Width = width;
    img->Height = height;
@@ -745,7 +635,7 @@ make_texture_image( GLcontext *ctx,
          }
          return;  /* all done */
       }
-   }      
+   }
 
 
    /*
@@ -851,13 +741,6 @@ clear_proxy_teximage(struct gl_texture_image *img)
    ASSERT(img);
    img->Format = 0;
    img->IntFormat = 0;
-   img->RedBits = 0;
-   img->GreenBits = 0;
-   img->BlueBits = 0;
-   img->AlphaBits = 0;
-   img->IntensityBits = 0;
-   img->LuminanceBits = 0;
-   img->IndexBits = 0;
    img->Border = 0;
    img->Width = 0;
    img->Height = 0;
@@ -871,6 +754,7 @@ clear_proxy_teximage(struct gl_texture_image *img)
    img->Data = NULL;
    img->IsCompressed = 0;
    img->CompressedSize = 0;
+   img->TexFormat = &_mesa_null_texformat;
 }
 
 
@@ -1994,11 +1878,10 @@ void
 _mesa_GetTexImage( GLenum target, GLint level, GLenum format,
                    GLenum type, GLvoid *pixels )
 {
-   GET_CURRENT_CONTEXT(ctx);
    const struct gl_texture_object *texObj;
    struct gl_texture_image *texImage;
    GLboolean discardImage;
-
+   GET_CURRENT_CONTEXT(ctx);
    ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx, "glGetTexImage");
 
    if (level < 0 || level >= ctx->Const.MaxTextureLevels) {
@@ -2174,11 +2057,12 @@ _mesa_TexSubImage1D( GLenum target, GLint level,
                      GLenum format, GLenum type,
                      const GLvoid *pixels )
 {
-   GET_CURRENT_CONTEXT(ctx);
    struct gl_texture_unit *texUnit;
    struct gl_texture_object *texObj;
    struct gl_texture_image *texImage;
    GLboolean success = GL_FALSE;
+   GET_CURRENT_CONTEXT(ctx);
+   ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx, "glTexSubImage1D");
 
    if (subtexture_error_check(ctx, 1, target, level, xoffset, 0, 0,
                               width, 1, 1, format, type)) {
@@ -2257,11 +2141,12 @@ _mesa_TexSubImage2D( GLenum target, GLint level,
                      GLenum format, GLenum type,
                      const GLvoid *pixels )
 {
-   GET_CURRENT_CONTEXT(ctx);
    struct gl_texture_unit *texUnit;
    struct gl_texture_object *texObj;
    struct gl_texture_image *texImage;
    GLboolean success = GL_FALSE;
+   GET_CURRENT_CONTEXT(ctx);
+   ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx, "glTexSubImage2D");
 
    if (subtexture_error_check(ctx, 2, target, level, xoffset, yoffset, 0,
                               width, height, 1, format, type)) {
@@ -2370,11 +2255,12 @@ _mesa_TexSubImage3D( GLenum target, GLint level,
                      GLenum format, GLenum type,
                      const GLvoid *pixels )
 {
-   GET_CURRENT_CONTEXT(ctx);
    struct gl_texture_unit *texUnit;
    struct gl_texture_object *texObj;
    struct gl_texture_image *texImage;
    GLboolean success = GL_FALSE;
+   GET_CURRENT_CONTEXT(ctx);
+   ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx, "glTexSubImage3D");
 
    if (subtexture_error_check(ctx, 3, target, level, xoffset, yoffset, zoffset,
                               width, height, depth, format, type)) {
@@ -2521,7 +2407,7 @@ _mesa_CopyTexImage1D( GLenum target, GLint level,
       return;
 
    if (ctx->Pixel.MapColorFlag || ctx->Pixel.ScaleOrBiasRGBA
-       || !ctx->Driver.CopyTexImage1D 
+       || !ctx->Driver.CopyTexImage1D
        || !(*ctx->Driver.CopyTexImage1D)(ctx, target, level,
                          internalFormat, x, y, width, border))
    {
@@ -2616,7 +2502,7 @@ _mesa_CopyTexSubImage1D( GLenum target, GLint level,
          gl_error( ctx, GL_OUT_OF_MEMORY, "glCopyTexSubImage2D" );
          return;
       }
-      
+
       /* now call glTexSubImage1D to do the real work */
       unpackSave = ctx->Unpack;
       ctx->Unpack = _mesa_native_packing;
@@ -2668,7 +2554,7 @@ _mesa_CopyTexSubImage2D( GLenum target, GLint level,
       _mesa_TexSubImage2D(target, level, xoffset, yoffset, width, height,
                           GL_RGBA, GL_UNSIGNED_BYTE, image);
       ctx->Unpack = unpackSave;
-      
+
       FREE(image);
    }
 }
@@ -2713,7 +2599,7 @@ _mesa_CopyTexSubImage3D( GLenum target, GLint level,
       _mesa_TexSubImage3D(target, level, xoffset, yoffset, zoffset,
                           width, height, 1, GL_RGBA, GL_UNSIGNED_BYTE, image);
       ctx->Unpack = unpackSave;
-      
+
       FREE(image);
    }
 }
@@ -3116,11 +3002,12 @@ _mesa_CompressedTexSubImage1DARB(GLenum target, GLint level, GLint xoffset,
                                  GLsizei width, GLenum format,
                                  GLsizei imageSize, const GLvoid *data)
 {
-   GET_CURRENT_CONTEXT(ctx);
    struct gl_texture_unit *texUnit;
    struct gl_texture_object *texObj;
    struct gl_texture_image *texImage;
    GLboolean success = GL_FALSE;
+   GET_CURRENT_CONTEXT(ctx);
+   ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx, "glCompressedTexSubImage1DARB");
 
    if (subtexture_error_check(ctx, 1, target, level, xoffset, 0, 0,
                               width, 1, 1, format, GL_NONE)) {
@@ -3153,11 +3040,12 @@ _mesa_CompressedTexSubImage2DARB(GLenum target, GLint level, GLint xoffset,
                                  GLenum format, GLsizei imageSize,
                                  const GLvoid *data)
 {
-   GET_CURRENT_CONTEXT(ctx);
    struct gl_texture_unit *texUnit;
    struct gl_texture_object *texObj;
    struct gl_texture_image *texImage;
    GLboolean success = GL_FALSE;
+   GET_CURRENT_CONTEXT(ctx);
+   ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx, "glCompressedTexSubImage2DARB");
 
    if (subtexture_error_check(ctx, 2, target, level, xoffset, yoffset, 0,
                               width, height, 1, format, GL_NONE)) {
@@ -3191,11 +3079,12 @@ _mesa_CompressedTexSubImage3DARB(GLenum target, GLint level, GLint xoffset,
                                  GLsizei height, GLsizei depth, GLenum format,
                                  GLsizei imageSize, const GLvoid *data)
 {
-   GET_CURRENT_CONTEXT(ctx);
    struct gl_texture_unit *texUnit;
    struct gl_texture_object *texObj;
    struct gl_texture_image *texImage;
    GLboolean success = GL_FALSE;
+   GET_CURRENT_CONTEXT(ctx);
+   ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx, "glCompressedTexSubImage3DARB");
 
    if (subtexture_error_check(ctx, 3, target, level, xoffset, yoffset, zoffset,
                               width, height, depth, format, GL_NONE)) {
@@ -3226,10 +3115,9 @@ _mesa_CompressedTexSubImage3DARB(GLenum target, GLint level, GLint xoffset,
 void
 _mesa_GetCompressedTexImageARB(GLenum target, GLint level, GLvoid *img)
 {
-   GET_CURRENT_CONTEXT(ctx);
    const struct gl_texture_object *texObj;
    struct gl_texture_image *texImage;
-
+   GET_CURRENT_CONTEXT(ctx);
    ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx, "glGetCompressedTexImageARB");
 
    if (level < 0 || level >= ctx->Const.MaxTextureLevels) {
diff --git a/xc/extras/Mesa/src/teximage.h b/xc/extras/Mesa/src/teximage.h
index ebc45e690..45ce79a1b 100644
--- a/xc/extras/Mesa/src/teximage.h
+++ b/xc/extras/Mesa/src/teximage.h
@@ -2,19 +2,19 @@
 /*
  * Mesa 3-D graphics library
  * Version:  3.4
- * 
+ *
  * Copyright (C) 1999-2000  Brian Paul   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 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
@@ -142,7 +142,7 @@ _mesa_CopyTexImage2D( GLenum target, GLint level,
 
 
 extern void
-_mesa_CopyTexSubImage1D( GLenum target, GLint level, GLint xoffset, 
+_mesa_CopyTexSubImage1D( GLenum target, GLint level, GLint xoffset,
                          GLint x, GLint y, GLsizei width );
 
 
@@ -152,7 +152,7 @@ _mesa_CopyTexSubImage2D( GLenum target, GLint level,
                          GLint x, GLint y, GLsizei width, GLsizei height );
 
 
-extern void 
+extern void
 _mesa_CopyTexSubImage3D( GLenum target, GLint level,
                          GLint xoffset, GLint yoffset, GLint zoffset,
                          GLint x, GLint y, GLsizei width, GLsizei height );
@@ -204,4 +204,3 @@ _mesa_GetCompressedTexImageARB(GLenum target, GLint lod, GLvoid *img);
 
 
 #endif
-
diff --git a/xc/extras/Mesa/src/texstate.c b/xc/extras/Mesa/src/texstate.c
index 08204bd88..d74feb578 100644
--- a/xc/extras/Mesa/src/texstate.c
+++ b/xc/extras/Mesa/src/texstate.c
@@ -890,25 +890,25 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level,
          *params = img->Border;
          return;
       case GL_TEXTURE_RED_SIZE:
-         *params = img->RedBits;
+         *params = img->TexFormat->RedBits;
          return;
       case GL_TEXTURE_GREEN_SIZE:
-         *params = img->GreenBits;
+         *params = img->TexFormat->GreenBits;
          return;
       case GL_TEXTURE_BLUE_SIZE:
-         *params = img->BlueBits;
+         *params = img->TexFormat->BlueBits;
          return;
       case GL_TEXTURE_ALPHA_SIZE:
-         *params = img->AlphaBits;
+         *params = img->TexFormat->AlphaBits;
          return;
       case GL_TEXTURE_INTENSITY_SIZE:
-         *params = img->IntensityBits;
+         *params = img->TexFormat->IntensityBits;
          return;
       case GL_TEXTURE_LUMINANCE_SIZE:
-         *params = img->LuminanceBits;
+         *params = img->TexFormat->LuminanceBits;
          return;
       case GL_TEXTURE_INDEX_SIZE_EXT:
-         *params = img->IndexBits;
+         *params = img->TexFormat->IndexBits;
          return;
 
       /* GL_ARB_texture_compression */
diff --git a/xc/extras/Mesa/src/texutil.c b/xc/extras/Mesa/src/texutil.c
index 313ef1455..4143830c1 100644
--- a/xc/extras/Mesa/src/texutil.c
+++ b/xc/extras/Mesa/src/texutil.c
@@ -2,1687 +2,1304 @@
 /*
  * Mesa 3-D graphics library
  * Version:  3.4
- * 
+ *
  * Copyright (C) 1999-2000  Brian Paul   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 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.
+ *
+ * Authors:
+ *    Gareth Hughes <gareth@valinux.com>
  */
 
-
 #ifdef PC_HEADER
 #include "all.h"
 #else
 #include "glheader.h"
 #include "context.h"
+#include "enums.h"
 #include "image.h"
 #include "mem.h"
+#include "texformat.h"
 #include "texutil.h"
 #include "types.h"
 #endif
 
+#define DBG 0
 
-/*
- * Texture utilities which may be useful to device drivers.
- * See the texutil.h for the list of supported internal formats.
- * It's expected that new formats will be added for new hardware.
- */
 
+struct gl_texture_convert {
+   GLint xoffset, yoffset, zoffset;	/* Subimage offset */
+   GLint width, height, depth;		/* Subimage region */
 
-/*
- * If the system is little endian and can do 4-byte word stores on
- * non 4-byte-aligned addresses then we can use this optimization.
- */
-#if defined(__i386__)
-#define DO_32BIT_STORES000
+   GLint imageWidth, imageHeight;	/* Full image dimensions */
+   GLenum format, type;
+
+   const struct gl_pixelstore_attrib *packing;
+
+   const GLvoid *srcImage;
+   GLvoid *dstImage;
+
+   GLint index;
+};
+
+typedef GLboolean (*convert_func)( struct gl_texture_convert *convert );
+typedef void (*unconvert_func)( struct gl_texture_convert *convert );
+
+#define CONVERT_STRIDE_BIT	0x1
+#define CONVERT_PACKING_BIT	0x2
+
+#ifdef __GNUC__
+#define ERROR_STRING	__FUNCTION__
+#else
+#define ERROR_STRING	__FILE__ ":" __LINE__
 #endif
+#define UNCONVERT_ERROR( format )					\
+do {									\
+   static char buffer[128];						\
+   sprintf( buffer, "illegal format %s in " ERROR_STRING "\n",		\
+	    gl_lookup_enum_by_nr( format ) );				\
+   gl_problem( NULL, buffer );						\
+} while (0)
 
 
 
-/*
- * Convert texture image data into one a specific internal format.
- * Input:
- *   dstFormat - the destination internal format
- *   dstWidth, dstHeight - the destination image size
- *   dstImage - pointer to destination image buffer
- *   dstRowStride - bytes to jump between image rows
- *   srcWidth, srcHeight - size of texture image
- *   srcFormat, srcType - format and datatype of source image
- *   srcImage - pointer to user's texture image
- *   packing - describes how user's texture image is packed.
- * Return: pointer to resulting image data or NULL if error or out of memory
- *   or NULL if we can't process the given image format/type/internalFormat
- *   combination.
- *
- * Supported type conversions:
- *   srcFormat           srcType                          dstFormat
- *   GL_INTENSITY        GL_UNSIGNED_BYTE                 MESA_I8
- *   GL_LUMINANCE        GL_UNSIGNED_BYTE                 MESA_L8
- *   GL_ALPHA            GL_UNSIGNED_BYTE                 MESA_A8
- *   GL_COLOR_INDEX      GL_UNSIGNED_BYTE                 MESA_C8
- *   GL_LUMINANCE_ALPHA  GL_UNSIGNED_BYTE                 MESA_A8_L8
- *   GL_RGB              GL_UNSIGNED_BYTE                 MESA_R5_G6_B5
- *   GL_RGB              GL_UNSIGNED_SHORT_5_6_5          MESA_R5_G6_B5
- *   GL_RGBA             GL_UNSIGNED_BYTE                 MESA_A4_R4_G4_B4
- *   GL_BGRA             GL_UNSIGNED_SHORT_4_4_4_4_REV    MESA_A4_R4_G4_B4
- *   GL_BGRA             GL_UNSIGHED_SHORT_1_5_5_5_REV    MESA_A1_R5_G5_B5
- *   GL_RGBA             GL_UNSIGNED_BYTE                 MESA_A1_R5_G5_B5
- *   GL_BGRA             GL_UNSIGNED_INT_8_8_8_8_REV      MESA_A8_R8_G8_B8
- *   GL_RGBA             GL_UNSIGNED_BYTE                 MESA_A8_R8_G8_B8
- *   GL_RGB              GL_UNSIGNED_BYTE                 MESA_A8_R8_G8_B8
- *   more to be added for new drivers...
- *
- * Notes:
- *   Some hardware only supports texture images of specific aspect ratios.
- *   This code will do power-of-two image up-scaling if that's needed.
- *
- *
+/* ================================================================
+ * RGBA8888 textures:
  */
-GLboolean
-_mesa_convert_teximage(MesaIntTexFormat dstFormat,
-                       GLint dstWidth, GLint dstHeight, GLvoid *dstImage,
-                       GLint dstRowStride,
-                       GLint srcWidth, GLint srcHeight,
-                       GLenum srcFormat, GLenum srcType,
-                       const GLvoid *srcImage,
-                       const struct gl_pixelstore_attrib *packing)
+
+#define DST_TYPE		GLuint
+#define DST_TEXELS_PER_DWORD	1
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_8888( src[3], src[2], src[1], src[0] )
+
+#define CONVERT_DIRECT
+
+#define SRC_TEXEL_BYTES		4
+
+#define TAG(x) x##_rgba8888_direct
+#define PRESERVE_DST_TYPE
+#include "texutil_tmp.h"
+
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_8888( src[0], src[1], src[2], src[3] )
+
+#define CONVERT_TEXEL_DWORD( src )	CONVERT_TEXEL( src )
+
+#define SRC_TEXEL_BYTES		4
+
+#define TAG(x) x##_abgr8888_to_rgba8888
+#define PRESERVE_DST_TYPE
+#include "texutil_tmp.h"
+
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_8888( src[0], src[1], src[2], 0xff )
+
+#define CONVERT_TEXEL_DWORD( src )	CONVERT_TEXEL( src )
+
+#define SRC_TEXEL_BYTES		3
+
+#define TAG(x) x##_bgr888_to_rgba8888
+#include "texutil_tmp.h"
+
+
+#define CONVERT_RGBA8888( name )					\
+static GLboolean							\
+convert_##name##_rgba8888( struct gl_texture_convert *convert )		\
+{									\
+   convert_func *tab;							\
+   GLint index = convert->index;					\
+									\
+   if ( convert->format == GL_ABGR_EXT &&				\
+	convert->type == GL_UNSIGNED_INT_8_8_8_8_REV )			\
+   {									\
+      tab = name##_tab_rgba8888_direct;					\
+   }									\
+   else if ( convert->format == GL_RGBA &&				\
+	     ( convert->type == GL_UNSIGNED_BYTE ||			\
+	       convert->type == GL_UNSIGNED_INT_8_8_8_8 ) )		\
+   {									\
+      tab = name##_tab_abgr8888_to_rgba8888;				\
+   }									\
+   else if ( convert->format == GL_RGB &&				\
+	     convert->type == GL_UNSIGNED_BYTE )			\
+   {									\
+      tab = name##_tab_bgr888_to_rgba8888;				\
+   }									\
+   else									\
+   {									\
+      /* Can't handle this source format/type combination */		\
+      return GL_FALSE;							\
+   }									\
+									\
+   return tab[index]( convert );					\
+}
+
+CONVERT_RGBA8888( texsubimage2d )
+CONVERT_RGBA8888( texsubimage3d )
+
+
+static void
+unconvert_teximage_rgba8888( struct gl_texture_convert *convert )
 {
-   const GLint wScale = dstWidth / srcWidth;   /* must be power of two */
-   const GLint hScale = dstHeight / srcHeight; /* must be power of two */
-   ASSERT(dstWidth >= srcWidth);
-   ASSERT(dstHeight >= srcHeight);
-   ASSERT(dstImage);
-   ASSERT(srcImage);
-   ASSERT(packing);
-
-   switch (dstFormat) {
-      case MESA_I8:
-      case MESA_L8:
-      case MESA_A8:
-      case MESA_C8:
-         if (srcType != GL_UNSIGNED_BYTE ||
-             ((srcFormat != GL_INTENSITY) &&
-              (srcFormat != GL_LUMINANCE) &&
-              (srcFormat != GL_ALPHA) &&
-              (srcFormat != GL_COLOR_INDEX))) {
-            /* bad internal format / srcFormat combination */
-            return GL_FALSE;
-         }
-         else {
-            /* store as 8-bit texels */
-            if (wScale == 1 && hScale == 1) {
-               /* no scaling needed - fast case */
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-               GLubyte *dst = (GLubyte *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  MEMCPY(dst, src, dstWidth * sizeof(GLubyte));
-                  dst += dstRowStride;
-                  src += srcStride;
-               }
-            }
-            else {
-               /* must rescale image */
-               GLubyte *dst = (GLubyte *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                             srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < dstWidth; col++) {
-                     dst[col] = src[col / wScale];
-                  }
-                  dst += dstRowStride;
-               }
-            }
-         }
-         break;
-
-      case MESA_A8_L8:
-         if (srcType != GL_UNSIGNED_BYTE || srcFormat != GL_LUMINANCE_ALPHA) {
-            return GL_FALSE;
-         }
-         else {
-            /* store as 16-bit texels */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row, col;
-               for (row = 0; row < dstHeight; row++) {
-                  for (col = 0; col < dstWidth; col++) {
-                     GLubyte luminance = src[col * 2 + 0];
-                     GLubyte alpha = src[col * 2 + 1];
-                     dst[col] = ((GLushort) alpha << 8) | luminance;
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-                  src += srcStride;
-               }
-            }
-            else {
-               /* must rescale */
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row, col;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-                  for (col = 0; col < dstWidth; col++) {
-                     GLint srcCol = col / wScale;
-                     GLubyte luminance = src[srcCol * 2 + 0];
-                     GLubyte alpha = src[srcCol * 2 + 1];
-                     dst[col] = ((GLushort) alpha << 8) | luminance;
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-                  src += srcStride;
-               }
-            }
-         }
-         break;
-
-      case MESA_R5_G6_B5:
-         if (srcFormat == GL_RGB && srcType == GL_UNSIGNED_SHORT_5_6_5) {
-            /* special, optimized case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  MEMCPY(dst, src, dstWidth * sizeof(GLushort));
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLushort *src = (const GLushort *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < dstWidth; col++) {
-                     dst[col] = src[col / wScale];
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else if (srcFormat == GL_RGB && srcType == GL_UNSIGNED_BYTE) {
-            /* general case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-#ifdef DO_32BIT_STORES
-               GLuint *dst = (GLuint *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint col, col3;
-                  GLint halfDstWidth = dstWidth >> 1;
-                  for (col = col3 = 0; col < halfDstWidth; col++, col3 += 6) {
-                     GLubyte r0 = src[col3 + 0];
-                     GLubyte g0 = src[col3 + 1];
-                     GLubyte b0 = src[col3 + 2];
-                     GLubyte r1 = src[col3 + 3];
-                     GLubyte g1 = src[col3 + 4];
-                     GLubyte b1 = src[col3 + 5];
-                     GLuint d0 = ((r0 & 0xf8) << 8)
-                               | ((g0 & 0xfc) << 3)
-                               | ((b0 & 0xf8) >> 3);
-                     GLuint d1 = ((r1 & 0xf8) << 8)
-                               | ((g1 & 0xfc) << 3)
-                               | ((b1 & 0xf8) >> 3);
-                     dst[col] = (d1 << 16) | d0;
-                  }
-                  src += srcStride;
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-#else /* 16-bit stores */
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint col, col3;
-                  for (col = col3 = 0; col < dstWidth; col++, col3 += 3) {
-                     GLubyte r = src[col3 + 0];
-                     GLubyte g = src[col3 + 1];
-                     GLubyte b = src[col3 + 2];
-                     dst[col] = ((r & 0xf8) << 8)
-                              | ((g & 0xfc) << 3)
-                              | ((b & 0xf8) >> 3);
-                  }
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-#endif
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < dstWidth; col++) {
-                     GLint col3 = (col / wScale) * 3;
-                     GLubyte r = src[col3 + 0];
-                     GLubyte g = src[col3 + 1];
-                     GLubyte b = src[col3 + 2];
-                     dst[col] = ((r & 0xf8) << 8)
-                              | ((g & 0xfc) << 3)
-                              | ((b & 0xf8) >> 3);
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else if (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE) {
-            /* general case (used by Quake3) */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-#ifdef DO_32BIT_STORES
-               GLuint *dst = dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint col, col4;
-                  GLint halfDstWidth = dstWidth >> 1;
-                  for (col = col4 = 0; col < halfDstWidth; col++, col4 += 8) {
-                     GLubyte r0 = src[col4 + 0];
-                     GLubyte g0 = src[col4 + 1];
-                     GLubyte b0 = src[col4 + 2];
-                     GLubyte r1 = src[col4 + 4];
-                     GLubyte g1 = src[col4 + 5];
-                     GLubyte b1 = src[col4 + 6];
-                     GLuint d0 = ((r0 & 0xf8) << 8)
-                               | ((g0 & 0xfc) << 3)
-                               | ((b0 & 0xf8) >> 3);
-                     GLuint d1 = ((r1 & 0xf8) << 8)
-                               | ((g1 & 0xfc) << 3)
-                               | ((b1 & 0xf8) >> 3);
-                     dst[col] = (d1 << 16) | d0;
-                  }
-                  src += srcStride;
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-#else /* 16-bit stores */
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint col, col4;
-                  for (col = col4 = 0; col < dstWidth; col++, col4 += 4) {
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     dst[col] = ((r & 0xf8) << 8)
-                              | ((g & 0xfc) << 3)
-                              | ((b & 0xf8) >> 3);
-                  }
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-#endif
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < dstWidth; col++) {
-                     GLint col4 = (col / wScale) * 4;
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     dst[col] = ((r & 0xf8) << 8)
-                              | ((g & 0xfc) << 3)
-                              | ((b & 0xf8) >> 3);
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else {
-            /* can't handle this srcFormat/srcType combination */
-            return GL_FALSE;
-         }
-         break;
-
-      case MESA_A4_R4_G4_B4:
-         /* store as 16-bit texels (GR_TEXFMT_ARGB_4444) */
-         if (srcFormat == GL_BGRA && srcType == GL_UNSIGNED_SHORT_4_4_4_4_REV){
-            /* special, optimized case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  MEMCPY(dst, src, dstWidth * sizeof(GLushort));
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLushort *src = (const GLushort *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < dstWidth; col++) {
-                     dst[col] = src[col / wScale];
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else if (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE) {
-            /* general case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-#ifdef DO_32BIT_STORES
-               GLuint *dst = dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint col, col4;
-                  GLint halfDstWidth = dstWidth >> 1;
-                  for (col = col4 = 0; col < halfDstWidth; col++, col4 += 8) {
-                     GLubyte r0 = src[col4 + 0];
-                     GLubyte g0 = src[col4 + 1];
-                     GLubyte b0 = src[col4 + 2];
-                     GLubyte a0 = src[col4 + 3];
-                     GLubyte r1 = src[col4 + 4];
-                     GLubyte g1 = src[col4 + 5];
-                     GLubyte b1 = src[col4 + 6];
-                     GLubyte a1 = src[col4 + 7];
-                     GLuint d0 = ((a0 & 0xf0) << 8)
-                               | ((r0 & 0xf0) << 4)
-                               | ((g0 & 0xf0)     )
-                               | ((b0 & 0xf0) >> 4);
-                     GLuint d1 = ((a1 & 0xf0) << 8)
-                               | ((r1 & 0xf0) << 4)
-                               | ((g1 & 0xf0)     )
-                               | ((b1 & 0xf0) >> 4);
-                     dst[col] = (d1 << 16) | d0;
-                  }
-                  src += srcStride;
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-#else /* 16-bit stores */
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint col, col4;
-                  for (col = col4 = 0; col < dstWidth; col++, col4 += 4) {
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     GLubyte a = src[col4 + 3];
-                     dst[col] = ((a & 0xf0) << 8)
-                              | ((r & 0xf0) << 4)
-                              | ((g & 0xf0)     )
-                              | ((b & 0xf0) >> 4);
-                  }
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-#endif
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < dstWidth; col++) {
-                     GLint col4 = (col / wScale) * 4;
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     GLubyte a = src[col4 + 3];
-                     dst[col] = ((a & 0xf0) << 8)
-                              | ((r & 0xf0) << 4)
-                              | ((g & 0xf0)     )
-                              | ((b & 0xf0) >> 4);
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else {
-            /* can't handle this format/srcType combination */
-            return GL_FALSE;
-         }
-         break;
-
-      case MESA_A1_R5_G5_B5:
-         /* store as 16-bit texels (GR_TEXFMT_ARGB_1555) */
-         if (srcFormat == GL_BGRA && srcType == GL_UNSIGNED_SHORT_1_5_5_5_REV){
-            /* special, optimized case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  MEMCPY(dst, src, dstWidth * sizeof(GLushort));
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLushort *src = (const GLushort *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < dstWidth; col++) {
-                     dst[col] = src[col / wScale];
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else if (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE) {
-            /* general case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint col, col4;
-                  for (col = col4 = 0; col < dstWidth; col++, col4 += 4) {
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     GLubyte a = src[col4 + 3];
-                     dst[col] = ((a & 0x80) << 8)
-                              | ((r & 0xf8) << 7)
-                              | ((g & 0xf8) << 2)
-                              | ((b & 0xf8) >> 3);
-                  }
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < dstWidth; col++) {
-                     GLint col4 = (col / wScale) * 4;
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     GLubyte a = src[col4 + 3];
-                     dst[col] = ((a & 0x80) << 8)
-                              | ((r & 0xf8) << 7)
-                              | ((g & 0xf8) << 2)
-                              | ((b & 0xf8) >> 3);
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else {
-            /* can't handle this source format/type combination */
-            return GL_FALSE;
-         }
-         break;
-
-      case MESA_A8_R8_G8_B8:
-      case MESA_FF_R8_G8_B8:
-         /* 32-bit texels */
-         if (srcFormat == GL_BGRA && srcType == GL_UNSIGNED_INT_8_8_8_8_REV){
-            /* special, optimized case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-               GLuint *dst = (GLuint *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  MEMCPY(dst, src, dstWidth * sizeof(GLuint));
-                  src += srcStride;
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLuint *dst = (GLuint *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLuint *src = (const GLuint *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < dstWidth; col++) {
-                     dst[col] = src[col / wScale];
-                  }
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else if (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE) {
-            /* general case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-               GLuint *dst = (GLuint *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint col, col4;
-                  for (col = col4 = 0; col < dstWidth; col++, col4 += 4) {
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     GLubyte a = src[col4 + 3];
-                     dst[col] = (a << 24) | (r << 16) | (g << 8) | b;
-                  }
-                  src += srcStride;
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLuint *dst = (GLuint *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < dstWidth; col++) {
-                     GLint col4 = (col / wScale) * 4;
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     GLubyte a = src[col4 + 3];
-                     dst[col] = (a << 24) | (r << 16) | (g << 8) | b;
-                  }
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else if (srcFormat == GL_RGB && srcType == GL_UNSIGNED_BYTE) {
-            /* general case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 srcWidth, srcFormat, srcType);
-               GLuint *dst = (GLuint *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint col, col3;
-                  for (col = col3 = 0; col < dstWidth; col++, col3 += 3) {
-                     GLubyte r = src[col3 + 0];
-                     GLubyte g = src[col3 + 1];
-                     GLubyte b = src[col3 + 2];
-                     GLubyte a = 255;
-                     dst[col] = (a << 24) | (r << 16) | (g << 8) | b;
-                  }
-                  src += srcStride;
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLuint *dst = (GLuint *) dstImage;
-               GLint row;
-               for (row = 0; row < dstHeight; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < dstWidth; col++) {
-                     GLint col3 = (col / wScale) * 3;
-                     GLubyte r = src[col3 + 0];
-                     GLubyte g = src[col3 + 1];
-                     GLubyte b = src[col3 + 2];
-                     GLubyte a = 255;
-                     dst[col] = (a << 24) | (r << 16) | (g << 8) | b;
-                  }
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else {
-            /* can't handle this source format/type combination */
-            return GL_FALSE;
-         }
-         if (dstFormat == MESA_FF_R8_G8_B8) {
-            /* set alpha bytes to 0xff */
-            GLint i;
-            GLubyte *dst = (GLubyte *) dstImage;
-            for (i = 0; i < dstWidth * dstHeight; i++) {
-               dst[i * 4 + 3] = 0xff;
-            }
-         }
-         break;
-
-      default:
-         /* unexpected internal format! */
-         return GL_FALSE;
+   const GLubyte *src = (const GLubyte *)convert->srcImage;
+   GLint texels, i;
+
+   texels = convert->width * convert->height * convert->depth;
+
+   switch ( convert->format ) {
+   case GL_RGBA: {
+      GLuint *dst = (GLuint *)convert->dstImage;
+      for ( i = 0 ; i < texels ; i++ ) {
+	 *dst++ = PACK_COLOR_8888( src[0], src[1], src[2], src[3] );
+	 src += 4;
+      }
+      break;
+   }
+   case GL_RGB: {
+      GLubyte *dst = (GLubyte *)convert->dstImage;
+      for ( i = 0 ; i < texels ; i++ ) {
+	 *dst++ = src[3];
+	 *dst++ = src[2];
+	 *dst++ = src[1];
+	 src += 4;
+      }
+      break;
+   }
+   default:
+      UNCONVERT_ERROR( convert->format );
+      break;
    }
-   return GL_TRUE;
 }
 
 
 
-/*
- * Replace a subregion of a texture image with new data.
- * Input:
- *   dstFormat - destination image format
- *   dstXoffset, dstYoffset - destination for new subregion (in texels)
- *   dstWidth, dstHeight - total size of dest image (in texels)
- *   dstImage - pointer to dest image
- *   dstRowStride - bytes to jump between image rows (in bytes)
- *   width, height - size of region to copy/replace (in texels)
- *   srcWidth, srcHeight - size of the corresponding gl_texture_image
- *   srcFormat, srcType - source image format and datatype
- *   srcImage - source image
- *   packing - source image packing information.
- * Return:  GL_TRUE or GL_FALSE for success, failure
- *
- * Notes:
- *   Like _mesa_convert_teximage(), we can do power-of-two image scaling
- *   to accomodate hardware with texture image aspect ratio constraints.
- *   dstWidth / srcWidth is used to compute the horizontal scaling factor and
- *   dstHeight / srcHeight is used to compute the vertical scaling factor.
+/* ================================================================
+ * ABGR8888 textures:
  */
-GLboolean
-_mesa_convert_texsubimage(MesaIntTexFormat dstFormat,
-                          GLint dstXoffset, GLint dstYoffset,
-                          GLint dstWidth, GLint dstHeight, GLvoid *dstImage,
-                          GLint dstRowStride,
-                          GLint width, GLint height,
-                          GLint srcWidth, GLint srcHeight,
-                          GLenum srcFormat, GLenum srcType,
-                          const GLvoid *srcImage,
-                          const struct gl_pixelstore_attrib *packing)
+
+#define DST_TYPE		GLuint
+#define DST_TEXELS_PER_DWORD	1
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_8888( src[3], src[2], src[1], src[0] )
+
+#define CONVERT_DIRECT
+
+#define SRC_TEXEL_BYTES		4
+
+#define TAG(x) x##_abgr8888_direct
+#define PRESERVE_DST_TYPE
+#include "texutil_tmp.h"
+
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_8888( 0xff, src[2], src[1], src[0] )
+
+#define CONVERT_TEXEL_DWORD( src )	CONVERT_TEXEL( src )
+
+#define SRC_TEXEL_BYTES		3
+
+#define TAG(x) x##_bgr888_to_abgr8888
+#include "texutil_tmp.h"
+
+
+#define CONVERT_ABGR8888( name )					\
+static GLboolean							\
+convert_##name##_abgr8888( struct gl_texture_convert *convert )		\
+{									\
+   convert_func *tab;							\
+   GLint index = convert->index;					\
+									\
+   if ( convert->format == GL_RGBA &&					\
+	convert->type == GL_UNSIGNED_BYTE )				\
+   {									\
+      tab = name##_tab_abgr8888_direct;					\
+   }									\
+   else if ( convert->format == GL_RGB &&				\
+	     convert->type == GL_UNSIGNED_BYTE )			\
+   {									\
+      tab = name##_tab_bgr888_to_abgr8888;				\
+   }									\
+   else									\
+   {									\
+      /* Can't handle this source format/type combination */		\
+      return GL_FALSE;							\
+   }									\
+									\
+   return tab[index]( convert );					\
+}
+
+CONVERT_ABGR8888( texsubimage2d )
+CONVERT_ABGR8888( texsubimage3d )
+
+
+static void
+unconvert_teximage_abgr8888( struct gl_texture_convert *convert )
 {
-   const GLint wScale = dstWidth / srcWidth;   /* must be power of two */
-   const GLint hScale = dstHeight / srcHeight; /* must be power of two */
-   ASSERT(dstWidth >= srcWidth);
-   ASSERT(dstHeight >= srcHeight);
-   ASSERT(dstImage);
-   ASSERT(srcImage);
-   ASSERT(packing);
-
-   width *= wScale;
-   height *= hScale;
-   dstXoffset *= wScale;
-   dstYoffset *= hScale;
-
-   /* XXX hscale != 1 and wscale != 1 not tested!!!! */
-   
-   switch (dstFormat) {
-      case MESA_I8:
-      case MESA_L8:
-      case MESA_A8:
-      case MESA_C8:
-         if (srcType != GL_UNSIGNED_BYTE ||
-             ((srcFormat != GL_INTENSITY) &&
-              (srcFormat != GL_LUMINANCE) &&
-              (srcFormat != GL_ALPHA) &&
-              (srcFormat != GL_COLOR_INDEX))) {
-            /* bad internal format / srcFormat combination */
-            return GL_FALSE;
-         }
-         else {
-            /* store as 8-bit texels */
-            if (wScale == 1 && hScale == 1) {
-               /* no scaling needed - fast case */
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcRowStride = _mesa_image_row_stride(packing,
-                                                  width, srcFormat, srcType);
-               GLubyte *dst = (GLubyte *) dstImage
-                            + dstYoffset * dstRowStride + dstXoffset;
-               const GLint rowSize = width * sizeof(GLubyte);
-               GLint row;
-               if (dstRowStride == srcRowStride && dstRowStride == rowSize) {
-                  MEMCPY(dst, src, rowSize * height);
-               }
-               else {
-                  for (row = 0; row < height; row++) {
-                     MEMCPY(dst, src, rowSize);
-                     dst += dstRowStride;
-                     src += srcRowStride;
-                  }
-               }
-            }
-            else {
-               /* must rescale image */
-               GLubyte *dst = (GLubyte *) dstImage
-                            + dstYoffset * dstRowStride + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < width; col++) {
-                     dst[col] = src[col / wScale];
-                  }
-                  dst += dstRowStride;
-               }
-            }
-         }
-         break;
-
-      case MESA_A8_L8:
-         if (srcType != GL_UNSIGNED_BYTE || srcFormat != GL_LUMINANCE_ALPHA) {
-            return GL_FALSE;
-         }
-         else {
-            /* store as 16-bit texels */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 width, srcFormat, srcType);
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row, col;
-               for (row = 0; row < height; row++) {
-                  for (col = 0; col < width; col++) {
-                     GLubyte luminance = src[col * 2 + 0];
-                     GLubyte alpha = src[col * 2 + 1];
-                     dst[col] = ((GLushort) alpha << 8) | luminance;
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-                  src += srcStride;
-               }
-            }
-            else {
-               /* must rescale */
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row, col;
-               for (row = 0; row < height; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 width, srcFormat, srcType);
-                  for (col = 0; col < width; col++) {
-                     GLint srcCol = col / wScale;
-                     GLubyte luminance = src[srcCol * 2 + 0];
-                     GLubyte alpha = src[srcCol * 2 + 1];
-                     dst[col] = ((GLushort) alpha << 8) | luminance;
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-                  src += srcStride;
-               }
-            }
-         }
-         break;
-
-      case MESA_R5_G6_B5:
-         if (srcFormat == GL_RGB && srcType == GL_UNSIGNED_SHORT_5_6_5) {
-            /* special, optimized case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 width, srcFormat, srcType);
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  MEMCPY(dst, src, width * sizeof(GLushort));
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLushort *src = (const GLushort *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < width; col++) {
-                     dst[col] = src[col / wScale];
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else if (srcFormat == GL_RGB && srcType == GL_UNSIGNED_BYTE) {
-            /* general case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 width, srcFormat, srcType);
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint col, col3;
-                  for (col = col3 = 0; col < width; col++, col3 += 3) {
-                     GLubyte r = src[col3 + 0];
-                     GLubyte g = src[col3 + 1];
-                     GLubyte b = src[col3 + 2];
-                     dst[col] = ((r & 0xf8) << 8)
-                              | ((g & 0xfc) << 3)
-                              | ((b & 0xf8) >> 3);
-                  }
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < width; col++) {
-                     GLint col3 = (col / wScale) * 3;
-                     GLubyte r = src[col3 + 0];
-                     GLubyte g = src[col3 + 1];
-                     GLubyte b = src[col3 + 2];
-                     dst[col] = ((r & 0xf8) << 8)
-                              | ((g & 0xfc) << 3)
-                              | ((b & 0xf8) >> 3);
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else if (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE) {
-            /* general case (used by Quake3) */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 width, srcFormat, srcType);
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint col, col4;
-                  for (col = col4 = 0; col < width; col++, col4 += 4) {
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     dst[col] = ((r & 0xf8) << 8)
-                              | ((g & 0xfc) << 3)
-                              | ((b & 0xf8) >> 3);
-                  }
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < width; col++) {
-                     GLint col4 = (col / wScale) * 4;
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     dst[col] = ((r & 0xf8) << 8)
-                              | ((g & 0xfc) << 3)
-                              | ((b & 0xf8) >> 3);
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else {
-            /* can't handle this srcFormat/srcType combination */
-            return GL_FALSE;
-         }
-         break;
-
-      case MESA_A4_R4_G4_B4:
-         /* store as 16-bit texels (GR_TEXFMT_ARGB_4444) */
-         if (srcFormat == GL_BGRA && srcType == GL_UNSIGNED_SHORT_4_4_4_4_REV){
-            /* special, optimized case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 width, srcFormat, srcType);
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  MEMCPY(dst, src, width * sizeof(GLushort));
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLushort *src = (const GLushort *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < width; col++) {
-                     dst[col] = src[col / wScale];
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else if (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE) {
-            /* general case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 width, srcFormat, srcType);
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint col, col4;
-                  for (col = col4 = 0; col < width; col++, col4 += 4) {
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     GLubyte a = src[col4 + 3];
-                     dst[col] = ((a & 0xf0) << 8)
-                              | ((r & 0xf0) << 4)
-                              | ((g & 0xf0)     )
-                              | ((b & 0xf0) >> 4);
-                  }
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < width; col++) {
-                     GLint col4 = (col / wScale) * 4;
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     GLubyte a = src[col4 + 3];
-                     dst[col] = ((a & 0xf0) << 8)
-                              | ((r & 0xf0) << 4)
-                              | ((g & 0xf0)     )
-                              | ((b & 0xf0) >> 4);
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else {
-            /* can't handle this format/srcType combination */
-            return GL_FALSE;
-         }
-         break;
-
-      case MESA_A1_R5_G5_B5:
-         /* store as 16-bit texels (GR_TEXFMT_ARGB_1555) */
-         if (srcFormat == GL_BGRA && srcType == GL_UNSIGNED_SHORT_1_5_5_5_REV){
-            /* special, optimized case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 width, srcFormat, srcType);
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  MEMCPY(dst, src, width * sizeof(GLushort));
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLushort *src = (const GLushort *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < width; col++) {
-                     dst[col] = src[col / wScale];
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else if (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE) {
-            /* general case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 width, srcFormat, srcType);
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint col, col4;
-                  for (col = col4 = 0; col < width; col++, col4 += 4) {
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     GLubyte a = src[col4 + 3];
-                     dst[col] = ((a & 0x80) << 8)
-                              | ((r & 0xf8) << 7)
-                              | ((g & 0xf8) << 2)
-                              | ((b & 0xf8) >> 3);
-                  }
-                  src += srcStride;
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLushort *dst = (GLushort *) ((GLubyte *) dstImage
-                             + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < width; col++) {
-                     GLint col4 = (col / wScale) * 4;
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     GLubyte a = src[col4 + 3];
-                     dst[col] = ((a & 0x80) << 8)
-                              | ((r & 0xf8) << 7)
-                              | ((g & 0xf8) << 2)
-                              | ((b & 0xf8) >> 3);
-                  }
-                  dst = (GLushort *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else {
-            /* can't handle this source format/type combination */
-            return GL_FALSE;
-         }
-         break;
-
-      case MESA_A8_R8_G8_B8:
-      case MESA_FF_R8_G8_B8:
-         /* 32-bit texels */
-         if (srcFormat == GL_BGRA && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) {
-            /* special, optimized case */
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 width, srcFormat, srcType);
-               GLuint *dst = (GLuint *) ((GLubyte *) dstImage
-                           + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  MEMCPY(dst, src, width * sizeof(GLuint));
-                  src += srcStride;
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLuint *dst = (GLuint *) ((GLubyte *) dstImage
-                           + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLuint *src = (const GLuint *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < width; col++) {
-                     dst[col] = src[col / wScale];
-                  }
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            if (dstFormat == MESA_FF_R8_G8_B8) {
-               /* set alpha bytes to 0xff */
-               GLint row, col;
-               GLubyte *dst = (GLubyte *) dstImage
-                              + dstYoffset * dstRowStride + dstXoffset * 4;
-               assert(wScale == 1 && hScale == 1); /* XXX not done */
-               for (row = 0; row < height; row++) {
-                  for (col = 0; col < width; col++) {
-                     dst[col * 4 + 3] = 0xff;
-                  }
-                  dst = dst + dstRowStride;
-               }
-            }
-         }
-         else if (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE) {
-            /* general case */
-            const GLubyte aMask = (dstFormat==MESA_FF_R8_G8_B8) ? 0xff : 0x00;
-            if (wScale == 1 && hScale == 1) {
-               const GLubyte *src = (const GLubyte *)
-                  _mesa_image_address(packing, srcImage, srcWidth, srcHeight,
-                                      srcFormat, srcType, 0, 0, 0);
-               const GLint srcStride = _mesa_image_row_stride(packing,
-                                                 width, srcFormat, srcType);
-               GLuint *dst = (GLuint *) ((GLubyte *) dstImage
-                           + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint col, col4;
-                  for (col = col4 = 0; col < width; col++, col4 += 4) {
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     GLubyte a = src[col4 + 3] | aMask;
-                     dst[col] = (a << 24) | (r << 16) | (g << 8) | b;
-                  }
-                  src += srcStride;
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-            else {
-               /* must rescale image */
-               GLuint *dst = (GLuint *) ((GLubyte *) dstImage
-                           + dstYoffset * dstRowStride) + dstXoffset;
-               GLint row;
-               for (row = 0; row < height; row++) {
-                  GLint srcRow = row / hScale;
-                  const GLubyte *src = (const GLubyte *)
-                     _mesa_image_address(packing, srcImage, srcWidth,
-                                  srcHeight, srcFormat, srcType, 0, srcRow, 0);
-                  GLint col;
-                  for (col = 0; col < width; col++) {
-                     GLint col4 = (col / wScale) * 4;
-                     GLubyte r = src[col4 + 0];
-                     GLubyte g = src[col4 + 1];
-                     GLubyte b = src[col4 + 2];
-                     GLubyte a = src[col4 + 3] | aMask;
-                     dst[col] = (a << 24) | (r << 16) | (g << 8) | b;
-                  }
-                  dst = (GLuint *) ((GLubyte *) dst + dstRowStride);
-               }
-            }
-         }
-         else {
-            /* can't handle this source format/type combination */
-            return GL_FALSE;
-         }
-         break;
-
-
-      default:
-         /* unexpected internal format! */
-         return GL_FALSE;
+   const GLubyte *src = (const GLubyte *)convert->srcImage;
+   GLint texels, i;
+
+   texels = convert->width * convert->height * convert->depth;
+
+   switch ( convert->format ) {
+   case GL_RGBA:
+      MEMCPY( convert->dstImage, src, texels * 4 );
+      break;
+   case GL_RGB: {
+      GLubyte *dst = (GLubyte *)convert->dstImage;
+      for ( i = 0 ; i < texels ; i++ ) {
+	 *dst++ = src[0];
+	 *dst++ = src[1];
+	 *dst++ = src[2];
+	 src += 4;
+      }
+      break;
+   }
+   default:
+      UNCONVERT_ERROR( convert->format );
+      break;
    }
-   return GL_TRUE;
 }
 
 
 
+/* ================================================================
+ * ARGB8888 textures:
+ */
 
-/*
- * Used to convert 16-bit texels into GLubyte color components.
+#define DST_TYPE		GLuint
+#define DST_TEXELS_PER_DWORD	1
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_8888( src[3], src[2], src[1], src[0] )
+
+#define CONVERT_DIRECT
+
+#define SRC_TEXEL_BYTES		4
+
+#define TAG(x) x##_argb8888_direct
+#define PRESERVE_DST_TYPE
+#include "texutil_tmp.h"
+
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_8888( src[3], src[0], src[1], src[2] )
+
+#define CONVERT_TEXEL_DWORD( src )	CONVERT_TEXEL( src )
+
+#define SRC_TEXEL_BYTES		4
+
+#define TAG(x) x##_abgr8888_to_argb8888
+#define PRESERVE_DST_TYPE
+#include "texutil_tmp.h"
+
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_8888( 0xff, src[0], src[1], src[2] )
+
+#define CONVERT_TEXEL_DWORD( src )	CONVERT_TEXEL( src )
+
+#define SRC_TEXEL_BYTES		3
+
+#define TAG(x) x##_bgr888_to_argb8888
+#include "texutil_tmp.h"
+
+
+#define CONVERT_ARGB8888( name )					\
+static GLboolean							\
+convert_##name##_argb8888( struct gl_texture_convert *convert )		\
+{									\
+   convert_func *tab;							\
+   GLint index = convert->index;					\
+									\
+   if ( convert->format == GL_BGRA &&					\
+	convert->type == GL_UNSIGNED_INT_8_8_8_8_REV )			\
+   {									\
+      tab = name##_tab_argb8888_direct;					\
+   }									\
+   else if ( convert->format == GL_RGBA &&				\
+	     convert->type == GL_UNSIGNED_BYTE )			\
+   {									\
+      tab = name##_tab_abgr8888_to_argb8888;				\
+   }									\
+   else if ( convert->format == GL_RGB &&				\
+	     convert->type == GL_UNSIGNED_BYTE )			\
+   {									\
+      tab = name##_tab_bgr888_to_argb8888;				\
+   }									\
+   else									\
+   {									\
+      /* Can't handle this source format/type combination */		\
+      return GL_FALSE;							\
+   }									\
+									\
+   return tab[index]( convert );					\
+}
+
+CONVERT_ARGB8888( texsubimage2d )
+CONVERT_ARGB8888( texsubimage3d )
+
+
+static void
+unconvert_teximage_argb8888( struct gl_texture_convert *convert )
+{
+   const GLubyte *src = (const GLubyte *)convert->srcImage;
+   GLint texels, i;
+
+   texels = convert->width * convert->height * convert->depth;
+
+   switch ( convert->format ) {
+   case GL_RGBA: {
+      GLuint *dst = (GLuint *)convert->dstImage;
+      for ( i = 0 ; i < texels ; i++ ) {
+	 *dst++ = PACK_COLOR_8888( src[3], src[0], src[1], src[2] );
+	 src += 4;
+      }
+      break;
+   }
+   case GL_RGB: {
+      GLubyte *dst = (GLubyte *)convert->dstImage;
+      for ( i = 0 ; i < texels ; i++ ) {
+	 *dst++ = src[2];
+	 *dst++ = src[1];
+	 *dst++ = src[0];
+	 src += 4;
+      }
+      break;
+   }
+   default:
+      UNCONVERT_ERROR( convert->format );
+      break;
+   }
+}
+
+
+
+/* ================================================================
+ * RGB888 textures:
+ */
+
+static GLboolean
+convert_texsubimage2d_rgb888( struct gl_texture_convert *convert )
+{
+   /* This is a placeholder for now...
+    */
+   return GL_FALSE;
+}
+
+static GLboolean
+convert_texsubimage3d_rgb888( struct gl_texture_convert *convert )
+{
+   /* This is a placeholder for now...
+    */
+   return GL_FALSE;
+}
+
+
+static void
+unconvert_teximage_rgb888( struct gl_texture_convert *convert )
+{
+   UNCONVERT_ERROR( convert->format );
+}
+
+
+
+/* ================================================================
+ * BGR888 textures:
  */
-static GLubyte R5G6B5toRed[0xffff];
-static GLubyte R5G6B5toGreen[0xffff];
-static GLubyte R5G6B5toBlue[0xffff];
 
-static GLubyte A4R4G4B4toRed[0xffff];
-static GLubyte A4R4G4B4toGreen[0xffff];
-static GLubyte A4R4G4B4toBlue[0xffff];
-static GLubyte A4R4G4B4toAlpha[0xffff];
+static GLboolean
+convert_texsubimage2d_bgr888( struct gl_texture_convert *convert )
+{
+   /* This is a placeholder for now...
+    */
+   return GL_FALSE;
+}
+
+static GLboolean
+convert_texsubimage3d_bgr888( struct gl_texture_convert *convert )
+{
+   /* This is a placeholder for now...
+    */
+   return GL_FALSE;
+}
+
+
+static void
+unconvert_teximage_bgr888( struct gl_texture_convert *convert )
+{
+   UNCONVERT_ERROR( convert->format );
+}
+
+
+
+/* ================================================================
+ * RGB565 textures:
+ */
+
+#define DST_TYPE		GLushort
+#define DST_TEXELS_PER_DWORD	2
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_565( src[0], src[1], src[2] )
+
+#define CONVERT_DIRECT
+
+#define SRC_TEXEL_BYTES		2
+
+#define TAG(x) x##_rgb565_direct
+#define PRESERVE_DST_TYPE
+#include "texutil_tmp.h"
+
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_565( src[0], src[1], src[2] )
+
+#define CONVERT_TEXEL_DWORD( src )					\
+		((PACK_COLOR_565( src[0], src[1], src[2] )) |		\
+		 (PACK_COLOR_565( src[3], src[4], src[5] ) << 16))
+
+#define SRC_TEXEL_BYTES		3
+
+#define TAG(x) x##_bgr888_to_rgb565
+#define PRESERVE_DST_TYPE
+#include "texutil_tmp.h"
+
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_565( src[0], src[1], src[2] )
+
+#define CONVERT_TEXEL_DWORD( src )					\
+		((PACK_COLOR_565( src[0], src[1], src[2] )) |		\
+		 (PACK_COLOR_565( src[4], src[5], src[6] ) << 16))
+
+#define SRC_TEXEL_BYTES		4
+
+#define TAG(x) x##_abgr8888_to_rgb565
+#include "texutil_tmp.h"
+
+
+#define CONVERT_RGB565( name )						\
+static GLboolean							\
+convert_##name##_rgb565( struct gl_texture_convert *convert )		\
+{									\
+   convert_func *tab;							\
+   GLint index = convert->index;					\
+									\
+   if ( convert->format == GL_RGB &&					\
+	convert->type == GL_UNSIGNED_SHORT_5_6_5 )			\
+   {									\
+      tab = name##_tab_rgb565_direct;					\
+   }									\
+   else if ( convert->format == GL_RGB &&				\
+	     convert->type == GL_UNSIGNED_BYTE )			\
+   {									\
+      tab = name##_tab_bgr888_to_rgb565;				\
+   }									\
+   else if ( convert->format == GL_RGBA &&				\
+	     convert->type == GL_UNSIGNED_BYTE )			\
+   {									\
+      tab = name##_tab_abgr8888_to_rgb565;				\
+   }									\
+   else									\
+   {									\
+      /* Can't handle this source format/type combination */		\
+      return GL_FALSE;							\
+   }									\
+									\
+   return tab[index]( convert );					\
+}
+
+CONVERT_RGB565( texsubimage2d )
+CONVERT_RGB565( texsubimage3d )
+
+
+static void
+unconvert_teximage_rgb565( struct gl_texture_convert *convert )
+{
+   const GLushort *src = (const GLushort *)convert->srcImage;
+   GLubyte *dst = (GLubyte *)convert->dstImage;
+   GLint texels, i;
+
+   texels = convert->width * convert->height * convert->depth;
+
+   switch ( convert->format ) {
+   case GL_RGBA:
+      for ( i = 0 ; i < texels ; i++ ) {
+	 GLushort s = *src++;
+	 *dst++ = ((s >> 8) & 0xf8) * 255 / 0xf8;
+	 *dst++ = ((s >> 3) & 0xfc) * 255 / 0xfc;
+	 *dst++ = ((s << 3) & 0xf8) * 255 / 0xf8;
+	 *dst++ = 0xff;
+      }
+      break;
+   case GL_RGB:
+      for ( i = 0 ; i < texels ; i++ ) {
+	 GLushort s = *src++;
+	 *dst++ = ((s >> 8) & 0xf8) * 255 / 0xf8;
+	 *dst++ = ((s >> 3) & 0xfc) * 255 / 0xfc;
+	 *dst++ = ((s << 3) & 0xf8) * 255 / 0xf8;
+      }
+      break;
+   default:
+      UNCONVERT_ERROR( convert->format );
+      break;
+   }
+}
+
+
+
+/* ================================================================
+ * ARGB4444 textures:
+ */
+
+#define DST_TYPE		GLushort
+#define DST_TEXELS_PER_DWORD	2
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_4444( src[3], src[0], src[1], src[2] )
+
+#define CONVERT_DIRECT
+
+#define SRC_TEXEL_BYTES		2
+
+#define TAG(x) x##_argb4444_direct
+#define PRESERVE_DST_TYPE
+#include "texutil_tmp.h"
+
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_4444( src[3], src[0], src[1], src[2] )
+
+#define CONVERT_TEXEL_DWORD( src )					\
+		((PACK_COLOR_4444( src[3], src[0], src[1], src[2] )) |	\
+		 (PACK_COLOR_4444( src[7], src[4], src[5], src[6] ) << 16))
+
+#define SRC_TEXEL_BYTES		4
+
+#define TAG(x) x##_rgba8888_to_argb4444
+#include "texutil_tmp.h"
+
+
+#define CONVERT_ARGB4444( name )					\
+static GLboolean							\
+convert_##name##_argb4444( struct gl_texture_convert *convert )		\
+{									\
+   convert_func *tab;							\
+   GLint index = convert->index;					\
+									\
+   if ( convert->format == GL_BGRA &&					\
+	convert->type == GL_UNSIGNED_SHORT_4_4_4_4_REV )		\
+   {									\
+      tab = name##_tab_argb4444_direct;					\
+   }									\
+   else if ( convert->format == GL_RGBA &&				\
+	     convert->type == GL_UNSIGNED_BYTE )			\
+   {									\
+      tab = name##_tab_rgba8888_to_argb4444;				\
+   }									\
+   else									\
+   {									\
+      /* Can't handle this source format/type combination */		\
+      return GL_FALSE;							\
+   }									\
+									\
+   return tab[index]( convert );					\
+}
+
+CONVERT_ARGB4444( texsubimage2d )
+CONVERT_ARGB4444( texsubimage3d )
 
-static GLubyte A1R5G5B5toRed[0xffff];
-static GLubyte A1R5G5B5toGreen[0xffff];
-static GLubyte A1R5G5B5toBlue[0xffff];
-static GLubyte A1R5G5B5toAlpha[0xffff];
 
 static void
-generate_lookup_tables(void)
+unconvert_teximage_argb4444( struct gl_texture_convert *convert )
 {
-   GLint i;
-   for (i = 0; i <= 0xffff; i++) {
-      GLint r = (i >> 8) & 0xf8;
-      GLint g = (i >> 3) & 0xfc;
-      GLint b = (i << 3) & 0xf8;
-      r = r * 255 / 0xf8;
-      g = g * 255 / 0xfc;
-      b = b * 255 / 0xf8;
-      R5G6B5toRed[i]   = r;
-      R5G6B5toGreen[i] = g;
-      R5G6B5toBlue[i]  = b;
+   const GLushort *src = (const GLushort *)convert->srcImage;
+   GLubyte *dst = (GLubyte *)convert->dstImage;
+   GLint texels, i;
+
+   texels = convert->width * convert->height * convert->depth;
+
+   switch ( convert->format ) {
+   case GL_RGBA:
+      for ( i = 0 ; i < texels ; i++ ) {
+	 GLushort s = *src++;
+	 *dst++ = ((s >>  8) & 0xf) * 255 / 0xf;
+	 *dst++ = ((s >>  4) & 0xf) * 255 / 0xf;
+	 *dst++ = ((s      ) & 0xf) * 255 / 0xf;
+	 *dst++ = ((s >> 12) & 0xf) * 255 / 0xf;
+      }
+      break;
+   default:
+      UNCONVERT_ERROR( convert->format );
+      break;
    }
+}
+
+
+
+/* ================================================================
+ * ARGB1555 textures:
+ */
+
+#define DST_TYPE		GLushort
+#define DST_TEXELS_PER_DWORD	2
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_1555( src[3], src[0], src[1], src[2] )
+
+#define CONVERT_DIRECT
+
+#define SRC_TEXEL_BYTES		2
 
-   for (i = 0; i <= 0xffff; i++) {
-      GLint r = (i >>  8) & 0xf;
-      GLint g = (i >>  4) & 0xf;
-      GLint b = (i      ) & 0xf;
-      GLint a = (i >> 12) & 0xf;
-      r = r * 255 / 0xf;
-      g = g * 255 / 0xf;
-      b = b * 255 / 0xf;
-      a = a * 255 / 0xf;
-      A4R4G4B4toRed[i]   = r;
-      A4R4G4B4toGreen[i] = g;
-      A4R4G4B4toBlue[i]  = b;
-      A4R4G4B4toAlpha[i] = a;
+#define TAG(x) x##_argb1555_direct
+#define PRESERVE_DST_TYPE
+#include "texutil_tmp.h"
+
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_1555( src[3], src[0], src[1], src[2] )
+
+#define CONVERT_TEXEL_DWORD( src )					\
+		((PACK_COLOR_1555( src[3], src[0], src[1], src[2] )) |	\
+		 (PACK_COLOR_1555( src[7], src[4], src[5], src[6] ) << 16))
+
+#define SRC_TEXEL_BYTES		4
+
+#define TAG(x) x##_rgba8888_to_argb1555
+#include "texutil_tmp.h"
+
+
+#define CONVERT_ARGB1555( name )					\
+static GLboolean							\
+convert_##name##_argb1555( struct gl_texture_convert *convert )		\
+{									\
+   convert_func *tab;							\
+   GLint index = convert->index;					\
+									\
+   if ( convert->format == GL_BGRA &&					\
+	convert->type == GL_UNSIGNED_SHORT_1_5_5_5_REV )		\
+   {									\
+      tab = name##_tab_argb1555_direct;					\
+   }									\
+   else if ( convert->format == GL_RGBA &&				\
+	     convert->type == GL_UNSIGNED_BYTE )			\
+   {									\
+      tab = name##_tab_rgba8888_to_argb1555;				\
+   }									\
+   else									\
+   {									\
+      /* Can't handle this source format/type combination */		\
+      return GL_FALSE;							\
+   }									\
+									\
+   return tab[index]( convert );					\
+}
+
+CONVERT_ARGB1555( texsubimage2d )
+CONVERT_ARGB1555( texsubimage3d )
+
+
+static void
+unconvert_teximage_argb1555( struct gl_texture_convert *convert )
+{
+   const GLushort *src = (const GLushort *)convert->srcImage;
+   GLubyte *dst = (GLubyte *)convert->dstImage;
+   GLint texels, i;
+
+   texels = convert->width * convert->height * convert->depth;
+
+   switch ( convert->format ) {
+   case GL_RGBA:
+      for ( i = 0 ; i < texels ; i++ ) {
+	 GLushort s = *src++;
+	 *dst++ = ((s >> 10) & 0xf8) * 255 / 0xf8;
+	 *dst++ = ((s >>  5) & 0xf8) * 255 / 0xf8;
+	 *dst++ = ((s      ) & 0xf8) * 255 / 0xf8;
+	 *dst++ = ((s >> 15) & 0x01) * 255;
+      }
+      break;
+   default:
+      UNCONVERT_ERROR( convert->format );
+      break;
    }
+}
+
+
+
+/* ================================================================
+ * AL88 textures:
+ */
+
+#define DST_TYPE		GLushort
+#define DST_TEXELS_PER_DWORD	2
 
-   for (i = 0; i <= 0xffff; i++) {
-      GLint r = (i >> 10) & 0xf8;
-      GLint g = (i >>  5) & 0xf8;
-      GLint b = (i      ) & 0xf8;
-      GLint a = (i >> 15) & 0x1;
-      r = r * 255 / 0xf8;
-      g = g * 255 / 0xf8;
-      b = b * 255 / 0xf8;
-      a = a * 255;
-      A1R5G5B5toRed[i]   = r;
-      A1R5G5B5toGreen[i] = g;
-      A1R5G5B5toBlue[i]  = b;
-      A1R5G5B5toAlpha[i] = a;
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_88( src[0], src[1] )
+
+#define CONVERT_DIRECT
+
+#define SRC_TEXEL_BYTES		2
+
+#define TAG(x) x##_al88_direct
+#define PRESERVE_DST_TYPE
+#include "texutil_tmp.h"
+
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_88( src[0], 0x00 )
+
+#define CONVERT_TEXEL_DWORD( src )					\
+		((PACK_COLOR_88( src[0], 0x00 )) |			\
+		 (PACK_COLOR_88( src[1], 0x00 ) << 16))
+
+#define SRC_TEXEL_BYTES		1
+
+#define TAG(x) x##_a8_to_al88
+#define PRESERVE_DST_TYPE
+#include "texutil_tmp.h"
+
+
+#define CONVERT_TEXEL( src )						\
+		PACK_COLOR_88( 0xff, src[0] )
+
+#define CONVERT_TEXEL_DWORD( src )					\
+		((PACK_COLOR_88( 0xff, src[0] )) |			\
+		 (PACK_COLOR_88( 0xff, src[1] ) << 16))
+
+#define SRC_TEXEL_BYTES		1
+
+#define TAG(x) x##_l8_to_al88
+#include "texutil_tmp.h"
+
+
+#define CONVERT_AL88( name )						\
+static GLboolean							\
+convert_##name##_al88( struct gl_texture_convert *convert )		\
+{									\
+   convert_func *tab;							\
+   GLint index = convert->index;					\
+									\
+   if ( convert->format == GL_LUMINANCE_ALPHA &&			\
+	convert->type == GL_UNSIGNED_BYTE )				\
+   {									\
+      tab = name##_tab_al88_direct;					\
+   }									\
+   else if ( convert->format == GL_ALPHA &&				\
+	     convert->type == GL_UNSIGNED_BYTE )			\
+   {									\
+      tab = name##_tab_a8_to_al88;					\
+   }									\
+   else if ( convert->format == GL_LUMINANCE &&				\
+	     convert->type == GL_UNSIGNED_BYTE )			\
+   {									\
+      tab = name##_tab_l8_to_al88;					\
+   }									\
+   else									\
+   {									\
+      /* Can't handle this source format/type combination */		\
+      return GL_FALSE;							\
+   }									\
+									\
+   return tab[index]( convert );					\
+}
+
+CONVERT_AL88( texsubimage2d )
+CONVERT_AL88( texsubimage3d )
+
+
+static void
+unconvert_teximage_al88( struct gl_texture_convert *convert )
+{
+   const GLubyte *src = (const GLubyte *)convert->srcImage;
+   GLint texels, i;
+
+   texels = convert->width * convert->height * convert->depth;
+
+   switch ( convert->format ) {
+   case GL_LUMINANCE_ALPHA:
+      MEMCPY( convert->dstImage, src, texels * 2 );
+      break;
+   case GL_ALPHA: {
+      GLubyte *dst = (GLubyte *)convert->dstImage;
+      for ( i = 0 ; i < texels ; i++ ) {
+	 *dst++ = src[1];
+	 src += 2;
+      }
+      break;
+   }
+   case GL_LUMINANCE: {
+      GLubyte *dst = (GLubyte *)convert->dstImage;
+      for ( i = 0 ; i < texels ; i++ ) {
+	 *dst++ = src[0];
+	 src += 2;
+      }
+      break;
+   }
+   default:
+      UNCONVERT_ERROR( convert->format );
+      break;
    }
 }
 
 
 
-/*
- * Convert a texture image from an internal format to one of Mesa's
- * core internal formats.  This is likely to be used by glGetTexImage
- * and for fetching texture images when falling back to software rendering.
- *
- * Input:
- *   srcFormat - source image format
- *   srcWidth, srcHeight - source image size
- *   srcImage - source image pointer
- *   srcRowStride - bytes to jump between image rows
- *   dstWidth, dstHeight - size of dest image
- *   dstFormat - format of dest image (must be one of Mesa's IntFormat values)
- *   dstImage - pointer to dest image
- * Notes:
- *   This function will do power of two image down-scaling to accomodate
- *   drivers with limited texture image aspect ratios.
- *   The implicit dest data type is GL_UNSIGNED_BYTE.
+/* ================================================================
+ * RGB332 textures:
+ */
+
+static GLboolean
+convert_texsubimage2d_rgb332( struct gl_texture_convert *convert )
+{
+   /* This is a placeholder for now...
+    */
+   return GL_FALSE;
+}
+
+static GLboolean
+convert_texsubimage3d_rgb332( struct gl_texture_convert *convert )
+{
+   /* This is a placeholder for now...
+    */
+   return GL_FALSE;
+}
+
+
+static void
+unconvert_teximage_rgb332( struct gl_texture_convert *convert )
+{
+   UNCONVERT_ERROR( convert->format );
+}
+
+
+
+/* ================================================================
+ * CI8 (and all other single-byte texel) textures:
  */
-void
-_mesa_unconvert_teximage(MesaIntTexFormat srcFormat,
-                         GLint srcWidth, GLint srcHeight,
-                         const GLvoid *srcImage, GLint srcRowStride,
-                         GLint dstWidth, GLint dstHeight,
-                         GLenum dstFormat, GLubyte *dstImage)
+
+#define DST_TYPE		GLubyte
+#define DST_TEXELS_PER_DWORD	4
+
+#define CONVERT_TEXEL( src )	src[0]
+
+#define CONVERT_DIRECT
+
+#define SRC_TEXEL_BYTES		1
+
+#define TAG(x) x##_ci8_direct
+#include "texutil_tmp.h"
+
+
+#define CONVERT_CI8( name )						\
+static GLboolean							\
+convert_##name##_ci8( struct gl_texture_convert *convert )		\
+{									\
+   convert_func *tab;							\
+   GLint index = convert->index;					\
+									\
+   if ( ( convert->format == GL_ALPHA ||				\
+	  convert->format == GL_LUMINANCE ||				\
+	  convert->format == GL_INTENSITY ||				\
+	  convert->format == GL_COLOR_INDEX ) &&			\
+	convert->type == GL_UNSIGNED_BYTE )				\
+   {									\
+      tab = name##_tab_ci8_direct;					\
+   }									\
+   else									\
+   {									\
+      /* Can't handle this source format/type combination */		\
+      return GL_FALSE;							\
+   }									\
+									\
+   return tab[index]( convert );					\
+}
+
+CONVERT_CI8( texsubimage2d )
+CONVERT_CI8( texsubimage3d )
+
+
+static void
+unconvert_teximage_ci8( struct gl_texture_convert *convert )
 {
-   static GLboolean firstCall = GL_TRUE;
-   const GLint wScale = srcWidth / dstWidth;   /* must be power of two */
-   const GLint hScale = srcHeight / dstHeight; /* must be power of two */
-   ASSERT(srcWidth >= dstWidth);
-   ASSERT(srcHeight >= dstHeight);
-   ASSERT(dstImage);
-   ASSERT(srcImage);
-
-   if (firstCall) {
-      generate_lookup_tables();
-      firstCall = GL_FALSE;
+   const GLubyte *src = (const GLubyte *)convert->srcImage;
+   GLint texels;
+
+   texels = convert->width * convert->height * convert->depth;
+
+   switch ( convert->format ) {
+   case GL_ALPHA:
+   case GL_LUMINANCE:
+   case GL_INTENSITY:
+   case GL_COLOR_INDEX:
+      MEMCPY( convert->dstImage, src, texels );
+      break;
+   default:
+      UNCONVERT_ERROR( convert->format );
+      break;
    }
+}
 
-   switch (srcFormat) {
-      case MESA_I8:
-      case MESA_L8:
-      case MESA_A8:
-      case MESA_C8:
-#ifdef DEBUG
-         if (srcFormat == MESA_I8) {
-            ASSERT(dstFormat == GL_INTENSITY);
-         }
-         else if (srcFormat == MESA_L8) {
-            ASSERT(dstFormat == GL_LUMINANCE);
-         }
-         else if (srcFormat == MESA_A8) {
-            ASSERT(dstFormat == GL_ALPHA);
-         }
-         else if (srcFormat == MESA_C8) {
-            ASSERT(dstFormat == GL_COLOR_INDEX);
-         }
-#endif
-         if (wScale == 1 && hScale == 1) {
-            /* easy! */
-            MEMCPY(dstImage, srcImage, dstWidth * dstHeight * sizeof(GLubyte));
-         }
-         else {
-            /* rescale */
-            const GLubyte *src8 = (const GLubyte *) srcImage;
-            GLint row, col;
-            for (row = 0; row < dstHeight; row++) {
-               GLint srcRow = row * hScale;
-               for (col = 0; col < dstWidth; col++) {
-                  GLint srcCol = col * wScale;
-                  *dstImage++ = src8[srcRow * srcWidth + srcCol];
-               }
-            }
-         }
-         break;
-      case MESA_A8_L8:
-         ASSERT(dstFormat == GL_LUMINANCE_ALPHA);
-         if (wScale == 1 && hScale == 1) {
-            GLint i, n = dstWidth * dstHeight;
-            const GLushort *texel = (const GLushort *) srcImage;
-            for (i = 0; i < n; i++) {
-               const GLushort tex = *texel++;
-               *dstImage++ = (tex & 0xff); /* luminance */
-               *dstImage++ = (tex >> 8);   /* alpha */
-            }
-         }
-         else {
-            /* rescale */
-            const GLushort *src16 = (const GLushort *) srcImage;
-            GLint row, col;
-            for (row = 0; row < dstHeight; row++) {
-               GLint srcRow = row * hScale;
-               for (col = 0; col < dstWidth; col++) {
-                  GLint srcCol = col * wScale;
-                  const GLushort tex = src16[srcRow * srcWidth + srcCol];
-                  *dstImage++ = (tex & 0xff); /* luminance */
-                  *dstImage++ = (tex >> 8);   /* alpha */
-               }
-            }
-         }
-         break;
-      case MESA_R5_G6_B5:
-         ASSERT(dstFormat == GL_RGB);
-         if (wScale == 1 && hScale == 1) {
-            GLint i, n = dstWidth * dstHeight;
-            const GLushort *texel = (const GLushort *) srcImage;
-            for (i = 0; i < n; i++) {
-               const GLushort tex = *texel++;
-               *dstImage++ = R5G6B5toRed[tex];
-               *dstImage++ = R5G6B5toGreen[tex];
-               *dstImage++ = R5G6B5toBlue[tex];
-            }
-         }
-         else {
-            /* rescale */
-            const GLushort *src16 = (const GLushort *) srcImage;
-            GLint row, col;
-            for (row = 0; row < dstHeight; row++) {
-               GLint srcRow = row * hScale;
-               for (col = 0; col < dstWidth; col++) {
-                  GLint srcCol = col * wScale;
-                  const GLushort tex = src16[srcRow * srcWidth + srcCol];
-                  *dstImage++ = R5G6B5toRed[tex];
-                  *dstImage++ = R5G6B5toGreen[tex];
-                  *dstImage++ = R5G6B5toBlue[tex];
-               }
-            }
-         }
-         break;
-      case MESA_A4_R4_G4_B4:
-         ASSERT(dstFormat == GL_RGBA);
-         if (wScale == 1 && hScale == 1) {
-            GLint i, n = dstWidth * dstHeight;
-            const GLushort *texel = (const GLushort *) srcImage;
-            for (i = 0; i < n; i++) {
-               const GLushort tex = *texel++;
-               *dstImage++ = A4R4G4B4toRed[tex];
-               *dstImage++ = A4R4G4B4toGreen[tex];
-               *dstImage++ = A4R4G4B4toBlue[tex];
-               *dstImage++ = A4R4G4B4toAlpha[tex];
-            }
-         }
-         else {
-            /* rescale */
-            const GLushort *src16 = (const GLushort *) srcImage;
-            GLint row, col;
-            for (row = 0; row < dstHeight; row++) {
-               GLint srcRow = row * hScale;
-               for (col = 0; col < dstWidth; col++) {
-                  GLint srcCol = col * wScale;
-                  const GLushort tex = src16[srcRow * srcWidth + srcCol];
-                  *dstImage++ = A4R4G4B4toRed[tex];
-                  *dstImage++ = A4R4G4B4toGreen[tex];
-                  *dstImage++ = A4R4G4B4toBlue[tex];
-                  *dstImage++ = A4R4G4B4toAlpha[tex];
-               }
-            }
-         }
-         break;
-      case MESA_A1_R5_G5_B5:
-         ASSERT(dstFormat == GL_RGBA);
-         if (wScale == 1 && hScale == 1) {
-            GLint i, n = dstWidth * dstHeight;
-            const GLushort *texel = (const GLushort *) srcImage;
-            for (i = 0; i < n; i++) {
-               const GLushort tex = *texel++;
-               *dstImage++ = A1R5G5B5toRed[tex];
-               *dstImage++ = A1R5G5B5toGreen[tex];
-               *dstImage++ = A1R5G5B5toBlue[tex];
-               *dstImage++ = A1R5G5B5toAlpha[tex];
-            }
-         }
-         else {
-            /* rescale */
-            const GLushort *src16 = (const GLushort *) srcImage;
-            GLint row, col;
-            for (row = 0; row < dstHeight; row++) {
-               GLint srcRow = row * hScale;
-               for (col = 0; col < dstWidth; col++) {
-                  GLint srcCol = col * wScale;
-                  const GLushort tex = src16[srcRow * srcWidth + srcCol];
-                  *dstImage++ = A1R5G5B5toRed[tex];
-                  *dstImage++ = A1R5G5B5toGreen[tex];
-                  *dstImage++ = A1R5G5B5toBlue[tex];
-                  *dstImage++ = A1R5G5B5toAlpha[tex];
-               }
-            }
-         }
-         break;
-      case MESA_A8_R8_G8_B8:
-      case MESA_FF_R8_G8_B8:
-         ASSERT(dstFormat == GL_RGBA);
-         if (wScale == 1 && hScale == 1) {
-            GLint i, n = dstWidth * dstHeight;
-            const GLuint *texel = (const GLuint *) srcImage;
-            for (i = 0; i < n; i++) {
-               const GLuint tex = *texel++;
-               *dstImage++ = (tex >> 16) & 0xff; /* R */
-               *dstImage++ = (tex >>  8) & 0xff; /* G */
-               *dstImage++ = (tex      ) & 0xff; /* B */
-               *dstImage++ = (tex >> 24) & 0xff; /* A */
-            }
-         }
-         else {
-            /* rescale */
-            const GLuint *src = (const GLuint *) srcImage;
-            GLint row, col;
-            for (row = 0; row < dstHeight; row++) {
-               GLint srcRow = row * hScale;
-               for (col = 0; col < dstWidth; col++) {
-                  GLint srcCol = col * wScale;
-                  const GLuint tex = src[srcRow * srcWidth + srcCol];
-                  *dstImage++ = (tex >> 16) & 0xff; /* R */
-                  *dstImage++ = (tex >>  8) & 0xff; /* G */
-                  *dstImage++ = (tex      ) & 0xff; /* B */
-                  *dstImage++ = (tex >> 24) & 0xff; /* A */
-               }
-            }
-         }
-         break;
-      default:
-         gl_problem(NULL, "bad srcFormat in _mesa_uncovert_teximage()");
+
+
+/* ================================================================
+ * Global entry points
+ */
+
+static convert_func gl_convert_texsubimage2d_tab[] = {
+   convert_texsubimage2d_rgba8888,
+   convert_texsubimage2d_abgr8888,
+   convert_texsubimage2d_argb8888,
+   convert_texsubimage2d_rgb888,
+   convert_texsubimage2d_bgr888,
+   convert_texsubimage2d_rgb565,
+   convert_texsubimage2d_argb4444,
+   convert_texsubimage2d_argb1555,
+   convert_texsubimage2d_al88,
+   convert_texsubimage2d_rgb332,
+   convert_texsubimage2d_ci8,		/* These are all the same... */
+   convert_texsubimage2d_ci8,
+   convert_texsubimage2d_ci8,
+   convert_texsubimage2d_ci8,
+};
+
+static convert_func gl_convert_texsubimage3d_tab[] = {
+   convert_texsubimage3d_rgba8888,
+   convert_texsubimage3d_abgr8888,
+   convert_texsubimage3d_argb8888,
+   convert_texsubimage3d_rgb888,
+   convert_texsubimage3d_bgr888,
+   convert_texsubimage3d_rgb565,
+   convert_texsubimage3d_argb4444,
+   convert_texsubimage3d_argb1555,
+   convert_texsubimage3d_al88,
+   convert_texsubimage3d_rgb332,
+   convert_texsubimage3d_ci8,		/* These are all the same... */
+   convert_texsubimage3d_ci8,
+   convert_texsubimage3d_ci8,
+   convert_texsubimage3d_ci8,
+};
+
+static unconvert_func gl_unconvert_teximage_tab[] = {
+   unconvert_teximage_rgba8888,
+   unconvert_teximage_abgr8888,
+   unconvert_teximage_argb8888,
+   unconvert_teximage_rgb888,
+   unconvert_teximage_bgr888,
+   unconvert_teximage_rgb565,
+   unconvert_teximage_argb4444,
+   unconvert_teximage_argb1555,
+   unconvert_teximage_al88,
+   unconvert_teximage_rgb332,
+   unconvert_teximage_ci8,		/* These are all the same... */
+   unconvert_teximage_ci8,
+   unconvert_teximage_ci8,
+   unconvert_teximage_ci8,
+};
+
+
+/* See if we need to care about the pixel store attributes when we're
+ * converting the texture image.  This should be stored as
+ * packing->_SomeBoolean and updated when the values change, to avoid
+ * testing every time...
+ */
+static INLINE GLboolean
+convert_needs_packing( const struct gl_pixelstore_attrib *packing,
+		       GLenum format, GLenum type )
+{
+   if ( ( packing->Alignment == 1 ||
+	  ( packing->Alignment == 4 &&	/* Pick up the common Q3A case... */
+	    format == GL_RGBA && type == GL_UNSIGNED_BYTE ) ) &&
+	packing->RowLength == 0 &&
+	packing->SkipPixels == 0 &&
+	packing->SkipRows == 0 &&
+	packing->ImageHeight == 0 &&
+	packing->SkipImages == 0 &&
+	packing->SwapBytes == GL_FALSE &&
+	packing->LsbFirst == GL_FALSE ) {
+      return GL_FALSE;
+   } else {
+      return GL_TRUE;
    }
 }
 
 
+GLboolean
+_mesa_convert_texsubimage1d( GLint mesaFormat,
+			     GLint xoffset,
+			     GLint width,
+			     GLenum format, GLenum type,
+			     const struct gl_pixelstore_attrib *packing,
+			     const GLvoid *srcImage, GLvoid *dstImage )
+{
+   struct gl_texture_convert convert;
 
-/*
- * Given an internal Mesa driver texture format, fill in the component
- * bit sizes in the given texture image struct.
+   ASSERT( packing );
+   ASSERT( srcImage );
+   ASSERT( dstImage );
+
+   ASSERT( mesaFormat >= MESA_FORMAT_RGBA8888 );
+   ASSERT( mesaFormat <= MESA_FORMAT_CI8 );
+
+   /* Make it easier to pass all the parameters around.
+    */
+   convert.xoffset = xoffset;
+   convert.yoffset = 0;
+   convert.width = width;
+   convert.height = 1;
+   convert.format = format;
+   convert.type = type;
+   convert.packing = packing;
+   convert.srcImage = srcImage;
+   convert.dstImage = dstImage;
+
+   convert.index = 0;
+
+   if ( convert_needs_packing( packing, format, type ) )
+      convert.index |= CONVERT_PACKING_BIT;
+
+   return gl_convert_texsubimage2d_tab[mesaFormat]( &convert );
+}
+
+GLboolean
+_mesa_convert_texsubimage2d( GLint mesaFormat,
+			     GLint xoffset, GLint yoffset,
+			     GLint width, GLint height,
+			     GLint imageWidth,
+			     GLenum format, GLenum type,
+			     const struct gl_pixelstore_attrib *packing,
+			     const GLvoid *srcImage, GLvoid *dstImage )
+{
+   struct gl_texture_convert convert;
+
+   ASSERT( packing );
+   ASSERT( srcImage );
+   ASSERT( dstImage );
+
+   ASSERT( mesaFormat >= MESA_FORMAT_RGBA8888 );
+   ASSERT( mesaFormat <= MESA_FORMAT_CI8 );
+
+   /* Make it easier to pass all the parameters around.
+    */
+   convert.xoffset = xoffset;
+   convert.yoffset = yoffset;
+   convert.width = width;
+   convert.height = height;
+   convert.imageWidth = imageWidth;
+   convert.format = format;
+   convert.type = type;
+   convert.packing = packing;
+   convert.srcImage = srcImage;
+   convert.dstImage = dstImage;
+
+   convert.index = 0;
+
+   if ( convert_needs_packing( packing, format, type ) )
+      convert.index |= CONVERT_PACKING_BIT;
+
+   if ( width != imageWidth )
+      convert.index |= CONVERT_STRIDE_BIT;
+
+   return gl_convert_texsubimage2d_tab[mesaFormat]( &convert );
+}
+
+GLboolean
+_mesa_convert_texsubimage3d( GLint mesaFormat,
+			     GLint xoffset, GLint yoffset, GLint zoffset,
+			     GLint width, GLint height, GLint depth,
+			     GLint imageWidth, GLint imageHeight,
+			     GLenum format, GLenum type,
+			     const struct gl_pixelstore_attrib *packing,
+			     const GLvoid *srcImage, GLvoid *dstImage )
+{
+   struct gl_texture_convert convert;
+
+   ASSERT( packing );
+   ASSERT( srcImage );
+   ASSERT( dstImage );
+
+   ASSERT( mesaFormat >= MESA_FORMAT_RGBA8888 );
+   ASSERT( mesaFormat <= MESA_FORMAT_CI8 );
+
+   /* Make it easier to pass all the parameters around.
+    */
+   convert.xoffset = xoffset;
+   convert.yoffset = yoffset;
+   convert.zoffset = zoffset;
+   convert.width = width;
+   convert.height = height;
+   convert.depth = depth;
+   convert.imageWidth = imageWidth;
+   convert.imageHeight = imageHeight;
+   convert.format = format;
+   convert.type = type;
+   convert.packing = packing;
+   convert.srcImage = srcImage;
+   convert.dstImage = dstImage;
+
+   convert.index = 0;
+
+   if ( convert_needs_packing( packing, format, type ) )
+      convert.index |= CONVERT_PACKING_BIT;
+
+   if ( width != imageWidth || height != imageHeight )
+      convert.index |= CONVERT_STRIDE_BIT;
+
+   return gl_convert_texsubimage3d_tab[mesaFormat]( &convert );
+}
+
+
+
+void _mesa_unconvert_teximage1d( GLint mesaFormat, GLenum format, GLint width,
+				 const GLvoid *srcImage, GLvoid *dstImage )
+{
+   struct gl_texture_convert convert;
+
+   ASSERT( srcImage );
+   ASSERT( dstImage );
+
+   ASSERT( mesaFormat >= MESA_FORMAT_RGBA8888 );
+   ASSERT( mesaFormat <= MESA_FORMAT_CI8 );
+
+   /* Make it easier to pass all the parameters around.
+    */
+   convert.width = width;
+   convert.height = 1;
+   convert.depth = 1;
+   convert.format = format;
+   convert.srcImage = srcImage;
+   convert.dstImage = dstImage;
+
+   gl_unconvert_teximage_tab[mesaFormat]( &convert );
+}
+
+void _mesa_unconvert_teximage2d( GLint mesaFormat, GLenum format,
+				 GLint width, GLint height,
+				 const GLvoid *srcImage, GLvoid *dstImage )
+{
+   struct gl_texture_convert convert;
+
+   ASSERT( srcImage );
+   ASSERT( dstImage );
+
+   ASSERT( mesaFormat >= MESA_FORMAT_RGBA8888 );
+   ASSERT( mesaFormat <= MESA_FORMAT_CI8 );
+
+   /* Make it easier to pass all the parameters around.
+    */
+   convert.width = width;
+   convert.height = height;
+   convert.depth = 1;
+   convert.format = format;
+   convert.srcImage = srcImage;
+   convert.dstImage = dstImage;
+
+   gl_unconvert_teximage_tab[mesaFormat]( &convert );
+}
+
+void _mesa_unconvert_teximage3d( GLint mesaFormat, GLenum format,
+				 GLint width, GLint height, GLint depth,
+				 const GLvoid *srcImage, GLvoid *dstImage )
+{
+   struct gl_texture_convert convert;
+
+   ASSERT( srcImage );
+   ASSERT( dstImage );
+
+   ASSERT( mesaFormat >= MESA_FORMAT_RGBA8888 );
+   ASSERT( mesaFormat <= MESA_FORMAT_CI8 );
+
+   /* Make it easier to pass all the parameters around.
+    */
+   convert.width = width;
+   convert.height = height;
+   convert.depth = depth;
+   convert.format = format;
+   convert.srcImage = srcImage;
+   convert.dstImage = dstImage;
+
+   gl_unconvert_teximage_tab[mesaFormat]( &convert );
+}
+
+
+
+/* Nearest filtering only (for broken hardware that can't support
+ * all aspect ratios).  This can be made a lot faster, but I don't
+ * really care enough...
  */
-void
-_mesa_set_teximage_component_sizes(MesaIntTexFormat mesaFormat,
-                                   struct gl_texture_image *texImage)
+void _mesa_rescale_teximage2d( const struct gl_texture_format *texFormat,
+			       GLint srcWidth, GLint srcHeight,
+			       GLint dstWidth, GLint dstHeight,
+			       const GLvoid *srcImage, GLvoid *dstImage )
 {
-   static const GLint bitSizes [][8] = {
-      /* format            R  G  B  A  I  L  C */
-      { MESA_I8,           0, 0, 0, 0, 8, 0, 0 },
-      { MESA_L8,           0, 0, 0, 0, 0, 8, 0 },
-      { MESA_A8,           0, 0, 0, 8, 0, 0, 0 },
-      { MESA_C8,           0, 0, 0, 0, 0, 0, 8 },
-      { MESA_A8_L8,        0, 0, 0, 8, 0, 8, 0 },
-      { MESA_R5_G6_B5,     5, 6, 5, 0, 0, 0, 0 },
-      { MESA_A4_R4_G4_B4,  4, 4, 4, 4, 0, 0, 0 },
-      { MESA_A1_R5_G5_B5,  5, 5, 5, 1, 0, 0, 0 },
-      { MESA_A8_R8_G8_B8,  8, 8, 8, 8, 0, 0, 0 },
-      { MESA_FF_R8_G8_B8,  8, 8, 8, 8, 0, 0, 0 },
-      { -1,                0, 0, 0, 0, 0, 0, 0 }
-   };
-   GLint i;
-   for (i = 0; i < bitSizes[i][0] >= 0; i++) {
-      if (bitSizes[i][0] == mesaFormat) {
-         texImage->RedBits       = bitSizes[i][1];
-         texImage->GreenBits     = bitSizes[i][2];
-         texImage->BlueBits      = bitSizes[i][3];
-         texImage->AlphaBits     = bitSizes[i][4];
-         texImage->IntensityBits = bitSizes[i][5];
-         texImage->LuminanceBits = bitSizes[i][6];
-         texImage->IndexBits     = bitSizes[i][7];
-         return;
-      }
+   GLint row, col;
+
+#define INNER_LOOP( HOP, WOP )						\
+   for ( row = 0 ; row < dstHeight ; row++ ) {				\
+      GLint srcRow = row HOP hScale;					\
+      for ( col = 0 ; col < dstWidth ; col++ ) {			\
+	 GLint srcCol = col WOP wScale;					\
+	 *dst++ = src[srcRow * srcWidth + srcCol];			\
+      }									\
+   }									\
+
+#define RESCALE_IMAGE( TYPE )						\
+do {									\
+   const TYPE *src = (const TYPE *)srcImage;				\
+   TYPE *dst = (TYPE *)dstImage;					\
+									\
+   if ( srcHeight <= dstHeight ) {					\
+      const GLint hScale = dstHeight / srcHeight;			\
+      if ( srcWidth <= dstWidth ) {					\
+	 const GLint wScale = dstWidth / srcWidth;			\
+	 INNER_LOOP( /, / );						\
+      }									\
+      else {								\
+	 const GLint wScale = srcWidth / dstWidth;			\
+	 INNER_LOOP( /, * );						\
+      }									\
+   }									\
+   else {								\
+      const GLint hScale = srcHeight / dstHeight;			\
+      if ( srcWidth <= dstWidth ) {					\
+	 const GLint wScale = dstWidth / srcWidth;			\
+	 INNER_LOOP( *, / );						\
+      }									\
+      else {								\
+	 const GLint wScale = srcWidth / dstWidth;			\
+	 INNER_LOOP( *, * );						\
+      }									\
+   }									\
+} while (0)
+
+   switch ( texFormat->TexelBytes ) {
+   case 4:
+      RESCALE_IMAGE( GLuint );
+      break;
+
+   case 2:
+      RESCALE_IMAGE( GLushort );
+      break;
+
+   case 1:
+      RESCALE_IMAGE( GLubyte );
+      break;
    }
-   gl_problem(NULL, "bad format in _mesa_set_teximage_component_sizes");
 }
diff --git a/xc/extras/Mesa/src/texutil.h b/xc/extras/Mesa/src/texutil.h
index 22e916d50..a3e842667 100644
--- a/xc/extras/Mesa/src/texutil.h
+++ b/xc/extras/Mesa/src/texutil.h
@@ -2,91 +2,86 @@
 /*
  * Mesa 3-D graphics library
  * Version:  3.4
- * 
+ *
  * Copyright (C) 1999-2000  Brian Paul   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 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.
+ *
+ * Authors:
+ *    Gareth Hughes <gareth@valinux.com>
  */
 
 
 #ifndef TEXUTIL_H
 #define TEXUTIL_H
 
-
 #include "types.h"
-
-
-
-/*
- * NOTE: "FF" means fill with byte value 0xff
- */
-                           /* msb <------ TEXEL BITS -----------> lsb */
-typedef enum {             /* ---- ---- ---- ---- ---- ---- ---- ---- */
-   MESA_I8,                /*                               IIII IIII */
-   MESA_L8,                /*                               LLLL LLLL */
-   MESA_A8,                /*                               AAAA AAAA */
-   MESA_C8,                /*                               CCCC CCCC */
-   MESA_A8_L8,             /*                     AAAA AAAA LLLL LLLL */
-   MESA_R5_G6_B5,          /*                     RRRR RGGG GGGB BBBB */
-   MESA_A4_R4_G4_B4,       /*                     AAAA RRRR GGGG BBBB */
-   MESA_A1_R5_G5_B5,       /*                     ARRR RRGG GGGB BBBB */
-   MESA_A8_R8_G8_B8,       /* AAAA AAAA RRRR RRRR GGGG GGGG BBBB BBBB */
-   MESA_FF_R8_G8_B8        /* FFFF FFFF RRRR RRRR GGGG GGGG BBBB BBBB */
-} MesaIntTexFormat;
-
-
-
+#include "texformat.h"
 
 extern GLboolean
-_mesa_convert_teximage(MesaIntTexFormat dstFormat,
-                       GLint dstWidth, GLint dstHeight, GLvoid *dstImage,
-                       GLint dstRowStride,
-                       GLint srcWidth, GLint srcHeight,
-                       GLenum srcFormat, GLenum srcType,
-                       const GLvoid *srcImage,
-                       const struct gl_pixelstore_attrib *packing);
-
-
+_mesa_convert_texsubimage1d( GLint mesaFormat,
+			     GLint xoffset,
+			     GLint width,
+			     GLenum format, GLenum type,
+			     const struct gl_pixelstore_attrib *packing,
+			     const GLvoid *srcImage, GLvoid *dstImage );
 
 extern GLboolean
-_mesa_convert_texsubimage(MesaIntTexFormat dstFormat,
-                          GLint dstXoffset, GLint dstYoffset,
-                          GLint dstWidth, GLint dstHeight, GLvoid *dstImage,
-                          GLint dstRowStride,
-                          GLint width, GLint height,
-                          GLint srcWidth, GLint srcHeight,
-                          GLenum srcFormat, GLenum srcType,
-                          const GLvoid *srcImage,
-                          const struct gl_pixelstore_attrib *packing);
-
+_mesa_convert_texsubimage2d( GLint mesaFormat,
+			     GLint xoffset, GLint yoffset,
+			     GLint width, GLint height,
+			     GLint imageWidth,
+			     GLenum format, GLenum type,
+			     const struct gl_pixelstore_attrib *packing,
+			     const GLvoid *srcImage, GLvoid *dstImage );
 
+extern GLboolean
+_mesa_convert_texsubimage3d( GLint mesaFormat,
+			     GLint xoffset, GLint yoffset, GLint zoffset,
+			     GLint width, GLint height, GLint depth,
+			     GLint imageWidth, GLint imageHeight,
+			     GLenum format, GLenum type,
+			     const struct gl_pixelstore_attrib *packing,
+			     const GLvoid *srcImage, GLvoid *dstImage );
+
+/* Deprecated in 3.5:
+ */
 extern void
-_mesa_unconvert_teximage(MesaIntTexFormat srcFormat,
-                         GLint srcWidth, GLint srcHeight,
-                         const GLvoid *srcImage, GLint srcRowStride,
-                         GLint dstWidth, GLint dstHeight,
-                         GLenum dstFormat, GLubyte *dstImage);
+_mesa_unconvert_teximage1d( GLint mesaFormat, GLenum format, GLint width,
+			    const GLvoid *srcImage, GLvoid *dstImage );
 
+extern void
+_mesa_unconvert_teximage2d( GLint mesaFormat, GLenum format,
+			    GLint width, GLint height,
+			    const GLvoid *srcImage, GLvoid *dstImage );
 
 extern void
-_mesa_set_teximage_component_sizes(MesaIntTexFormat mesaFormat,
-                                   struct gl_texture_image *texImage);
+_mesa_unconvert_teximage3d( GLint mesaFormat, GLenum format,
+			    GLint width, GLint height, GLint depth,
+			    const GLvoid *srcImage, GLvoid *dstImage );
 
+/* Nearest filtering only (for broken hardware that can't support
+ * all aspect ratios).  FIXME: Make this a subimage update as well...
+ */
+extern void
+_mesa_rescale_teximage2d( const struct gl_texture_format *texFormat,
+			  GLint srcWidth, GLint srcHeight,
+			  GLint dstWidth, GLint dstHeight,
+			  const GLvoid *srcImage, GLvoid *dstImage );
 
 #endif
-
diff --git a/xc/extras/Mesa/src/texutil_tmp.h b/xc/extras/Mesa/src/texutil_tmp.h
new file mode 100644
index 000000000..ea6c4ee52
--- /dev/null
+++ b/xc/extras/Mesa/src/texutil_tmp.h
@@ -0,0 +1,375 @@
+/*
+ * NOTE: All 3D code is untested and most definitely broken...
+ */
+
+#define DST_TEXEL_BYTES		(4 / DST_TEXELS_PER_DWORD)
+#define DST_ROW_WIDTH		(convert->width * DST_TEXEL_BYTES)
+#define DST_ROW_STRIDE		(convert->imageWidth * DST_TEXEL_BYTES)
+#define DST_IMG_STRIDE		(convert->imageWidth *			\
+				 convert->imageHeight * DST_TEXEL_BYTES)
+
+
+/* ================================================================
+ * PRE: No pixelstore attribs, width == imageWidth.
+ */
+static GLboolean
+TAG(texsubimage2d)( struct gl_texture_convert *convert )
+{
+   const GLubyte *src = (const GLubyte *)convert->srcImage;
+   GLuint *dst = (GLuint *)((GLubyte *)convert->dstImage +
+			    (convert->yoffset * convert->imageWidth +
+			     convert->xoffset) * DST_TEXEL_BYTES);
+   GLint dwords, i;
+   (void) dwords; (void) i;
+
+   if ( DBG )
+      fprintf( stderr, __FUNCTION__ "\n" );
+
+#ifdef CONVERT_DIRECT
+   MEMCPY( dst, src, convert->height * DST_ROW_WIDTH );
+#else
+   dwords = (convert->width * convert->height +
+	     DST_TEXELS_PER_DWORD - 1) / DST_TEXELS_PER_DWORD;
+
+   for ( i = 0 ; i < dwords ; i++ ) {
+      *dst++ = CONVERT_TEXEL_DWORD( src );
+      src += SRC_TEXEL_BYTES * DST_TEXELS_PER_DWORD;
+   }
+#endif
+
+   return GL_TRUE;
+}
+
+/* PRE: As above, height == imageHeight also.
+ */
+static GLboolean
+TAG(texsubimage3d)( struct gl_texture_convert *convert )
+{
+   const GLubyte *src = (const GLubyte *)convert->srcImage;
+   GLuint *dst = (GLuint *)((GLubyte *)convert->dstImage +
+			    ((convert->zoffset * convert->height +
+			      convert->yoffset) * convert->width +
+			     convert->xoffset) * DST_TEXEL_BYTES);
+   GLint dwords, i;
+   (void) dwords; (void) i;
+
+   if ( DBG )
+      fprintf( stderr, __FUNCTION__ "\n" );
+
+#ifdef CONVERT_DIRECT
+   MEMCPY( dst, src, convert->depth * convert->height * DST_ROW_WIDTH );
+#else
+   dwords = (convert->width * convert->height * convert->depth +
+	     DST_TEXELS_PER_DWORD - 1) / DST_TEXELS_PER_DWORD;
+
+   for ( i = 0 ; i < dwords ; i++ ) {
+      *dst++ = CONVERT_TEXEL_DWORD( src );
+      src += SRC_TEXEL_BYTES * DST_TEXELS_PER_DWORD;
+   }
+#endif
+
+   return GL_TRUE;
+}
+
+
+
+/* ================================================================
+ * PRE: No pixelstore attribs, width != imageWidth.
+ */
+static GLboolean
+TAG(texsubimage2d_stride)( struct gl_texture_convert *convert )
+{
+   const GLubyte *src = (const GLubyte *)convert->srcImage;
+   DST_TYPE *dst = (DST_TYPE *)((GLubyte *)convert->dstImage +
+				(convert->yoffset * convert->imageWidth +
+				 convert->xoffset) * DST_TEXEL_BYTES);
+   GLint adjust;
+   GLint row, col;
+
+   adjust = convert->imageWidth - convert->width;
+
+   if ( DBG ) {
+      fprintf( stderr, __FUNCTION__ ":\n" );
+      fprintf( stderr, "   x=%d y=%d w=%d h=%d s=%d\n",
+	       convert->xoffset, convert->yoffset, convert->width,
+	       convert->height, convert->imageWidth );
+      fprintf( stderr, "   adjust=%d\n", adjust );
+   }
+
+   for ( row = 0 ; row < convert->height ; row++ ) {
+      for ( col = 0 ; col < convert->width ; col++ ) {
+	 *dst++ = CONVERT_TEXEL( src );
+	 src += SRC_TEXEL_BYTES;
+      }
+      dst += adjust;
+   }
+
+   return GL_TRUE;
+}
+
+/* PRE: As above, or height != imageHeight also.
+ */
+static GLboolean
+TAG(texsubimage3d_stride)( struct gl_texture_convert *convert )
+{
+   const GLubyte *src = (const GLubyte *)convert->srcImage;
+   DST_TYPE *dst = (DST_TYPE *)((GLubyte *)convert->dstImage +
+				((convert->zoffset * convert->imageHeight +
+				  convert->yoffset) * convert->imageWidth +
+				 convert->xoffset) * DST_TEXEL_BYTES);
+   GLint adjust;
+   GLint row, col, img;
+
+   adjust = convert->imageWidth - convert->width;
+
+   if ( DBG ) {
+      fprintf( stderr, __FUNCTION__ ":\n" );
+      fprintf( stderr, "   x=%d y=%d w=%d h=%d s=%d\n",
+	       convert->xoffset, convert->yoffset, convert->width,
+	       convert->height, convert->imageWidth );
+      fprintf( stderr, "   adjust=%d\n", adjust );
+   }
+
+   for ( img = 0 ; img < convert->depth ; img++ ) {
+      for ( row = 0 ; row < convert->height ; row++ ) {
+	 for ( col = 0 ; col < convert->width ; col++ ) {
+	    *dst++ = CONVERT_TEXEL( src );
+	    src += SRC_TEXEL_BYTES;
+	 }
+	 dst += adjust;
+      }
+      /* FIXME: ... */
+   }
+
+   return GL_TRUE;
+}
+
+
+
+/* ================================================================
+ * PRE: Require pixelstore attribs, width == imageWidth.
+ */
+static GLboolean
+TAG(texsubimage2d_pack)( struct gl_texture_convert *convert )
+{
+   const GLubyte *src = (const GLubyte *)
+      _mesa_image_address( convert->packing, convert->srcImage,
+			   convert->width, convert->height,
+			   convert->format, convert->type, 0, 0, 0 );
+   const GLint srcRowStride =
+      _mesa_image_row_stride( convert->packing, convert->width,
+			      convert->format, convert->type );
+   GLuint *dst = (GLuint *)((GLubyte *)convert->dstImage +
+			    (convert->yoffset * convert->width +
+			     convert->xoffset) * DST_TEXEL_BYTES);
+   GLint width;
+   GLint row, col;
+   (void) col;
+
+   if ( DBG )
+      fprintf( stderr, __FUNCTION__ "\n" );
+
+   width = ((convert->width + DST_TEXELS_PER_DWORD - 1)
+	    & ~(DST_TEXELS_PER_DWORD - 1));
+
+   for ( row = 0 ; row < convert->height ; row++ ) {
+#ifdef CONVERT_DIRECT
+      MEMCPY( dst, src, DST_ROW_STRIDE );
+      src += srcRowStride;
+      dst = (GLuint *)((GLubyte *)dst + DST_ROW_STRIDE);
+#else
+      const GLubyte *srcRow = src;
+      for ( col = width / DST_TEXELS_PER_DWORD ; col ; col-- ) {
+	 *dst++ = CONVERT_TEXEL_DWORD( src );
+	 src += SRC_TEXEL_BYTES * DST_TEXELS_PER_DWORD;
+      }
+      src = srcRow + srcRowStride;
+#endif
+   }
+
+   return GL_TRUE;
+}
+
+/* PRE: as above, height == imageHeight also.
+ */
+static GLboolean
+TAG(texsubimage3d_pack)( struct gl_texture_convert *convert )
+{
+   const GLubyte *src = (const GLubyte *)
+      _mesa_image_address( convert->packing, convert->srcImage,
+			   convert->width, convert->height,
+			   convert->format, convert->type, 0, 0, 0 );
+   const GLint srcRowStride =
+      _mesa_image_row_stride( convert->packing, convert->width,
+			      convert->format, convert->type );
+   GLuint *dst = (GLuint *)((GLubyte *)convert->dstImage +
+			    ((convert->zoffset * convert->height +
+			      convert->yoffset) * convert->width +
+			     convert->xoffset) * DST_TEXEL_BYTES);
+   GLint width;
+   GLint row, col, img;
+   (void) col;
+
+   if ( DBG )
+      fprintf( stderr, __FUNCTION__ "\n" );
+
+   width = ((convert->width + DST_TEXELS_PER_DWORD - 1)
+	    & ~(DST_TEXELS_PER_DWORD - 1));
+
+   for ( img = 0 ; img < convert->depth ; img++ ) {
+      for ( row = 0 ; row < convert->height ; row++ ) {
+#ifdef CONVERT_DIRECT
+	 MEMCPY( dst, src, DST_ROW_STRIDE );
+	 src += srcRowStride;
+	 dst = (GLuint *)((GLubyte *)dst + DST_ROW_STRIDE);
+#else
+	 const GLubyte *srcRow = src;
+	 for ( col = width / DST_TEXELS_PER_DWORD ; col ; col-- ) {
+	    *dst++ = CONVERT_TEXEL_DWORD( src );
+	    src += SRC_TEXEL_BYTES * DST_TEXELS_PER_DWORD;
+	 }
+	 src = srcRow + srcRowStride;
+#endif
+      }
+   }
+
+   return GL_TRUE;
+}
+
+
+
+/* ================================================================
+ * PRE: Require pixelstore attribs, width != imageWidth.
+ */
+static GLboolean
+TAG(texsubimage2d_stride_pack)( struct gl_texture_convert *convert )
+{
+   const GLubyte *src = (const GLubyte *)
+      _mesa_image_address( convert->packing, convert->srcImage,
+			   convert->width, convert->height,
+			   convert->format, convert->type, 0, 0, 0 );
+   const GLint srcRowStride =
+      _mesa_image_row_stride( convert->packing, convert->width,
+			      convert->format, convert->type );
+   DST_TYPE *dst = (DST_TYPE *)((GLubyte *)convert->dstImage +
+				(convert->yoffset * convert->imageWidth +
+				 convert->xoffset) * DST_TEXEL_BYTES);
+   GLint adjust;
+   GLint row, col;
+   (void) col;
+
+   adjust = convert->imageWidth - convert->width;
+
+   if ( DBG ) {
+      fprintf( stderr, __FUNCTION__ ":\n" );
+      fprintf( stderr, "   x=%d y=%d w=%d h=%d s=%d\n",
+	       convert->xoffset, convert->yoffset, convert->width,
+	       convert->height, convert->imageWidth );
+      fprintf( stderr, "   adjust=%d\n", adjust );
+   }
+
+   for ( row = 0 ; row < convert->height ; row++ ) {
+#ifdef CONVERT_DIRECT
+      MEMCPY( dst, src, DST_ROW_WIDTH );
+      src += srcRowStride;
+      dst += convert->imageWidth;
+#else
+      const GLubyte *srcRow = src;
+      for ( col = 0 ; col < convert->width ; col++ ) {
+	 *dst++ = CONVERT_TEXEL( src );
+	 src += SRC_TEXEL_BYTES;
+      }
+      src = srcRow + srcRowStride;
+      dst += adjust;
+#endif
+   }
+
+   return GL_TRUE;
+}
+
+/* PRE: As above, or height != imageHeight also.
+ */
+static GLboolean
+TAG(texsubimage3d_stride_pack)( struct gl_texture_convert *convert )
+{
+   const GLubyte *src = (const GLubyte *)
+      _mesa_image_address( convert->packing, convert->srcImage,
+			   convert->width, convert->height,
+			   convert->format, convert->type, 0, 0, 0 );
+   const GLint srcRowStride =
+      _mesa_image_row_stride( convert->packing, convert->width,
+			      convert->format, convert->type );
+   DST_TYPE *dst = (DST_TYPE *)((GLubyte *)convert->dstImage +
+				((convert->zoffset * convert->imageHeight +
+				  convert->yoffset) * convert->imageWidth +
+				 convert->xoffset) * DST_TEXEL_BYTES);
+   GLint adjust;
+   GLint row, col, img;
+   (void) col;
+
+   adjust = convert->imageWidth - convert->width;
+
+   if ( DBG ) {
+      fprintf( stderr, __FUNCTION__ ":\n" );
+      fprintf( stderr, "   x=%d y=%d w=%d h=%d s=%d\n",
+	       convert->xoffset, convert->yoffset, convert->width,
+	       convert->height, convert->imageWidth );
+      fprintf( stderr, "   adjust=%d\n", adjust );
+   }
+
+   for ( img = 0 ; img < convert->depth ; img++ ) {
+      for ( row = 0 ; row < convert->height ; row++ ) {
+#ifdef CONVERT_DIRECT
+	 MEMCPY( dst, src, DST_ROW_WIDTH );
+	 src += srcRowStride;
+	 dst += convert->imageWidth;
+#else
+	 const GLubyte *srcRow = src;
+	 for ( col = 0 ; col < convert->width ; col++ ) {
+	    *dst++ = CONVERT_TEXEL( src );
+	    src += SRC_TEXEL_BYTES;
+	 }
+	 src = srcRow + srcRowStride;
+	 dst += adjust;
+#endif
+      }
+      /* FIXME: ... */
+   }
+
+   return GL_TRUE;
+}
+
+
+
+static convert_func TAG(texsubimage2d_tab)[] = {
+   TAG(texsubimage2d),
+   TAG(texsubimage2d_stride),
+   TAG(texsubimage2d_pack),
+   TAG(texsubimage2d_stride_pack),
+};
+
+static convert_func TAG(texsubimage3d_tab)[] = {
+   TAG(texsubimage3d),
+   TAG(texsubimage3d_stride),
+   TAG(texsubimage3d_pack),
+   TAG(texsubimage3d_stride_pack),
+};
+
+
+#ifndef PRESERVE_DST_TYPE
+#undef DST_TYPE
+#undef DST_TEXELS_PER_DWORD
+#endif
+
+#undef SRC_TEXEL_BYTES
+#undef DST_TEXEL_BYTES
+#undef DST_ROW_WIDTH
+#undef DST_ROW_STRIDE
+
+#undef CONVERT_TEXEL
+#undef CONVERT_TEXEL_DWORD
+#undef CONVERT_DIRECT
+
+#undef TAG
+
+#undef PRESERVE_DST_TYPE
diff --git a/xc/extras/Mesa/src/tritemp.h b/xc/extras/Mesa/src/tritemp.h
index 03a38852c..71e62a31a 100644
--- a/xc/extras/Mesa/src/tritemp.h
+++ b/xc/extras/Mesa/src/tritemp.h
@@ -1,21 +1,20 @@
-/* $Id: tritemp.h,v 1.15 2001/02/12 20:42:42 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
  * Version:  3.4.1
- * 
+ *
  * Copyright (C) 1999-2001  Brian Paul   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 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
@@ -57,7 +56,7 @@
  *
  * Optionally, one may provide one-time setup code per triangle:
  *    SETUP_CODE    - code which is to be executed once per triangle
- * 
+ *
  * The following macro MUST be defined:
  *    INNER_LOOP(LEFT,RIGHT,Y) - code to write a span of pixels.
  *        Something like:
@@ -227,7 +226,7 @@
     * By stepping rasterization parameters along the major edge,
     * we can avoid recomputing them at the discontinuity where
     * the top and bottom edges meet.  However, this forces us to
-    * be able to scan both left-to-right and right-to-left. 
+    * be able to scan both left-to-right and right-to-left.
     * Also, we must determine whether the major edge is at the
     * left or right side of the triangle.  We do this by
     * computing the magnitude of the cross-product of the major
@@ -445,7 +444,7 @@
 	    dtdy = oneOverArea * (eMaj.dx * eBot_dt - eMaj_dt * eBot.dx);
 	 } else {
 	    dtdx = 0;
-	    dtdy = 0; 
+	    dtdy = 0;
 	 }
 
 	 if (VB->TexCoordPtr[0]->size > 2)
@@ -553,7 +552,7 @@
        * inside the triangle.
        *
        * Next we creep down the major edge until we reach that y,
-       * and compute the corresponding x coordinate on the edge. 
+       * and compute the corresponding x coordinate on the edge.
        * Then we find the half-integral x that lies on or just
        * inside the edge.  This is the first pixel that might lie in
        * the interior of the triangle.  (We won't know for sure
@@ -784,7 +783,7 @@
 		     t0 = VB->TexCoordPtr[0]->data[vLower][1] * T_SCALE;
 		     ft = (GLfixed)(t0 * FIXED_SCALE + dtdx * adjx + dtdy * adjy) + FIXED_HALF;
 		     fdtOuter = SignedFloatToFixed(dtdy + dxOuter * dtdx);
-		  } 
+		  }
 		  else
 		  {
 		     t0 = 0;
@@ -801,7 +800,7 @@
                   sLeft = s0 + (dsdx * adjx + dsdy * adjy) * (1.0F/FIXED_SCALE);
                   dsOuter = dsdy + dxOuter * dsdx;
 		  if (VB->TexCoordPtr[0]->size > 1)
-		  {		  
+		  {
 		     t0 = VB->TexCoordPtr[0]->data[vLower][1] * invW;
 		     tLeft = t0 + (dtdx * adjx + dtdy * adjy) * (1.0F/FIXED_SCALE);
 		     dtOuter = dtdy + dxOuter * dtdx;
@@ -809,7 +808,7 @@
 		     tLeft = dtOuter = 0;
 		  }
 		  if (VB->TexCoordPtr[0]->size > 2)
-		  {		  
+		  {
 		     u0 = VB->TexCoordPtr[0]->data[vLower][2] * invW;
 		     uLeft = u0 + (dudx * adjx + dudy * adjy) * (1.0F/FIXED_SCALE);
 		     duOuter = dudy + dxOuter * dudx;
@@ -817,7 +816,7 @@
 		     uLeft = duOuter = 0;
 		  }
 		  if (VB->TexCoordPtr[0]->size > 3)
-		  {		  
+		  {
 		     v0 = VB->TexCoordPtr[0]->data[vLower][3] * invW;
 		  } else {
 		     v0 = invW;
@@ -834,7 +833,7 @@
                   s1Left = s0 + (ds1dx * adjx + ds1dy * adjy) * (1.0F/FIXED_SCALE);
                   ds1Outer = ds1dy + dxOuter * ds1dx;
 		  if (VB->TexCoordPtr[0]->size > 1)
-		  {		  
+		  {
 		     t0 = VB->TexCoordPtr[1]->data[vLower][1] * invW;
 		     t1Left = t0 + (dt1dx * adjx + dt1dy * adjy) * (1.0F/FIXED_SCALE);
 		     dt1Outer = dt1dy + dxOuter * dt1dx;
@@ -842,7 +841,7 @@
 		     t1Left = dt1Outer = 0;
 		  }
 		  if (VB->TexCoordPtr[0]->size > 2)
-		  {		  
+		  {
 		     u0 = VB->TexCoordPtr[1]->data[vLower][2] * invW;
 		     u1Left = u0 + (du1dx * adjx + du1dy * adjy) * (1.0F/FIXED_SCALE);
 		     du1Outer = du1dy + dxOuter * du1dx;
@@ -850,7 +849,7 @@
 		     u1Left = du1Outer = 0;
 		  }
 		  if (VB->TexCoordPtr[0]->size > 3)
-		  {		  
+		  {
 		     v0 = VB->TexCoordPtr[1]->data[vLower][3] * invW;
 		  } else {
 		     v0 =  invW;
diff --git a/xc/extras/Mesa/src/types.h b/xc/extras/Mesa/src/types.h
index 804de925e..5c297eb2a 100644
--- a/xc/extras/Mesa/src/types.h
+++ b/xc/extras/Mesa/src/types.h
@@ -1,4 +1,4 @@
-/* -*- mode: C; tab-width:8; c-basic-offset:8 -*- */
+/* -*- mode: C; tab-width:8; c-basic-offset:3 -*- */
 
 /*
  * Mesa 3-D graphics library
@@ -171,7 +171,21 @@ typedef void (*TextureSampleFunc)( const struct gl_texture_object *tObj,
                                    const GLfloat u[], const GLfloat lambda[],
                                    GLubyte rgba[][4] );
 
+/* Texture format record */
+/* GH: This is an interim structure until 3.5 */
+struct gl_texture_format {
+   GLint IntFormat;		/* One of the MESA_FORMAT_* values */
 
+   GLubyte RedBits;		/* Bits per texel component */
+   GLubyte GreenBits;
+   GLubyte BlueBits;
+   GLubyte AlphaBits;
+   GLubyte LuminanceBits;
+   GLubyte IntensityBits;
+   GLubyte IndexBits;
+
+   GLint TexelBytes;
+};
 
 /* Texture image record */
 struct gl_texture_image {
@@ -180,13 +194,6 @@ struct gl_texture_image {
 				 * GL_COLOR_INDEX only
 				 */
    GLenum IntFormat;		/* Internal format as given by the user */
-   GLubyte RedBits;		/* Bits per texel component              */
-   GLubyte GreenBits;		/*   These are initialized by Mesa but   */
-   GLubyte BlueBits;		/*   may be reassigned by the device     */
-   GLubyte AlphaBits;		/*   driver to indicate the true texture */
-   GLubyte IntensityBits;	/*   color resolution.                   */
-   GLubyte LuminanceBits;
-   GLubyte IndexBits;
    GLuint Border;		/* 0 or 1 */
    GLuint Width;		/* = 2^WidthLog2 + 2*Border */
    GLuint Height;		/* = 2^HeightLog2 + 2*Border */
@@ -200,6 +207,8 @@ struct gl_texture_image {
    GLuint MaxLog2;		/* = MAX(WidthLog2, HeightLog2) */
    GLubyte *Data;		/* Image data as unsigned bytes */
 
+   const struct gl_texture_format *TexFormat;
+
    GLboolean IsCompressed;	/* GL_ARB_texture_compression */
    GLuint CompressedSize;	/* GL_ARB_texture_compression */
 
diff --git a/xc/lib/GL/mesa/src/Imakefile b/xc/lib/GL/mesa/src/Imakefile
index cc88803f4..7ccc16fd7 100644
--- a/xc/lib/GL/mesa/src/Imakefile
+++ b/xc/lib/GL/mesa/src/Imakefile
@@ -152,6 +152,8 @@ LinkSourceFile(state.c, $(MESASRCDIR)/src)
 LinkSourceFile(state.h, $(MESASRCDIR)/src)
 LinkSourceFile(stencil.c, $(MESASRCDIR)/src)
 LinkSourceFile(stencil.h, $(MESASRCDIR)/src)
+LinkSourceFile(texformat.c, $(MESASRCDIR)/src)
+LinkSourceFile(texformat.h, $(MESASRCDIR)/src)
 LinkSourceFile(texgen_tmp.h, $(MESASRCDIR)/src)
 LinkSourceFile(teximage.c, $(MESASRCDIR)/src)
 LinkSourceFile(teximage.h, $(MESASRCDIR)/src)
@@ -163,6 +165,7 @@ LinkSourceFile(texture.c, $(MESASRCDIR)/src)
 LinkSourceFile(texture.h, $(MESASRCDIR)/src)
 LinkSourceFile(texutil.c, $(MESASRCDIR)/src)
 LinkSourceFile(texutil.h, $(MESASRCDIR)/src)
+LinkSourceFile(texutil_tmp.h, $(MESASRCDIR)/src)
 LinkSourceFile(trans_tmp.h, $(MESASRCDIR)/src)
 LinkSourceFile(translate.c, $(MESASRCDIR)/src)
 LinkSourceFile(translate.h, $(MESASRCDIR)/src)
@@ -252,6 +255,7 @@ LinkSourceFile(zoom.h, $(MESASRCDIR)/src)
 		stages.c \
 		state.c \
 		stencil.c \
+		texformat.c \
 		teximage.c \
 		texobj.c \
 		texstate.c \
@@ -325,6 +329,7 @@ LinkSourceFile(zoom.h, $(MESASRCDIR)/src)
 		stages.o \
 		state.o \
 		stencil.o \
+		texformat.o \
 		teximage.o \
 		texobj.o \
 		texstate.o \
@@ -346,8 +351,8 @@ LinkSourceFile(zoom.h, $(MESASRCDIR)/src)
 		zoom.o
 
 #ifdef i386Architecture
-     ASM_SRCS = 
-     ASM_OBJS = 
+     ASM_SRCS =
+     ASM_OBJS =
 #if MesaUseMMX
      MMX_DEFS = -DUSE_MMX_ASM
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810_3d_reg.h b/xc/lib/GL/mesa/src/drv/i810/i810_3d_reg.h
index a7200f1b8..cad45626d 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810_3d_reg.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810_3d_reg.h
@@ -113,8 +113,8 @@
 #define LCS_LINEWIDTH_2_0       (0x4<<12)
 #define LCS_LINEWIDTH_3_0       (0x6<<12)
 #define LCS_UPDATE_ALPHA_INTERP (0x1<<11)
-#define LCS_ALPHA_FLAT          (0x0<<10)
-#define LCS_ALPHA_INTERP        (0x1<<10)
+#define LCS_ALPHA_FLAT          (0x1<<10)
+#define LCS_ALPHA_INTERP        (0x0<<10)
 #define LCS_UPDATE_FOG_INTERP   (0x1<<9)
 #define LCS_FOG_INTERP          (0x0<<8)
 #define LCS_FOG_FLAT            (0x1<<8)
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810ioctl.c b/xc/lib/GL/mesa/src/drv/i810/i810ioctl.c
index 2f16e33d7..146b6c1b9 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810ioctl.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810ioctl.c
@@ -200,7 +200,6 @@ void i810DmaFinish( i810ContextPtr imesa  )
    FLUSH_BATCH( imesa );
 
    if (imesa->sarea->last_quiescent != imesa->sarea->last_enqueue) {
-     
       if (I810_DEBUG&DEBUG_VERBOSE_IOCTL) 
 	 fprintf(stderr, "i810DmaFinish\n");
 
@@ -214,14 +213,22 @@ void i810DmaFinish( i810ContextPtr imesa  )
 
 void i810RegetLockQuiescent( i810ContextPtr imesa  ) 
 {
+   /* XXX I disabled this conditional.  Doing so fixes all the readpixels
+    * problems.  The problem was that we'd sometimes read from the frame
+    * buffer (via the span functions) before rendering was completed.
+    * Taking out this conditional solves that problem.  (BrianP)
+    *
    if (imesa->sarea->last_quiescent != imesa->sarea->last_enqueue) {
+   */
       if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
 	 fprintf(stderr, "i810RegetLockQuiescent\n");
 
       drmUnlock(imesa->driFd, imesa->hHWContext);
       i810GetLock( imesa, DRM_LOCK_QUIESCENT ); 
       imesa->sarea->last_quiescent = imesa->sarea->last_enqueue;
+   /*
    }
+   */
 }
 
 void i810WaitAgeLocked( i810ContextPtr imesa, int age  ) 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810span.c b/xc/lib/GL/mesa/src/drv/i810/i810span.c
index 9e86e9b68..4bf02ad13 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810span.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810span.c
@@ -30,7 +30,7 @@
 			dPriv->x * 2 +			\
 			dPriv->y * pitch)
 
-#define INIT_MONO_PIXEL(p) 
+#define INIT_MONO_PIXEL(p)
 
 #define CLIPPIXEL(_x,_y) (_x >= minx && _x < maxx && \
 			  _y >= miny && _y < maxy)
@@ -50,6 +50,8 @@
 
 #define HW_LOCK()				\
    i810ContextPtr imesa = I810_CONTEXT(ctx);	\
+   FLUSH_BATCH(imesa);				\
+   i810DmaFinish(imesa);			\
    LOCK_HARDWARE_QUIESCENT(imesa);
 
 #define HW_CLIPLOOP()						\
@@ -85,9 +87,9 @@
 #define READ_RGBA( rgba, _x, _y )				\
 do {								\
    GLushort p = *(GLushort *)(read_buf + _x*2 + _y*pitch);	\
-   rgba[0] = (p >> 8) & 0xf8;					\
-   rgba[1] = (p >> 3) & 0xfc;					\
-   rgba[2] = (p << 3) & 0xf8;					\
+   rgba[0] = (((p >> 11) & 0x1f) * 255) / 31;			\
+   rgba[1] = (((p >>  5) & 0x3f) * 255) / 63;			\
+   rgba[2] = (((p >>  0) & 0x1f) * 255) / 31;			\
    rgba[3] = 255;						\
 } while(0)
 
@@ -128,7 +130,7 @@ do {								\
    *(GLushort *)(buf + _x*2 + _y*pitch)  = d;
 
 #define READ_DEPTH( d, _x, _y )	\
-   d = *(GLushort *)(buf + _x*2 + _y*pitch);	 
+   d = *(GLushort *)(buf + _x*2 + _y*pitch);
 
 /*     d = 0xffff; */
 
@@ -144,7 +146,7 @@ void i810DDInitSpanFuncs( GLcontext *ctx )
       ctx->Driver.WriteRGBSpan = i810WriteRGBSpan_565;
       ctx->Driver.WriteMonoRGBASpan = i810WriteMonoRGBASpan_565;
       ctx->Driver.WriteRGBAPixels = i810WriteRGBAPixels_565;
-      ctx->Driver.WriteMonoRGBAPixels = i810WriteMonoRGBAPixels_565; 
+      ctx->Driver.WriteMonoRGBAPixels = i810WriteMonoRGBAPixels_565;
       ctx->Driver.ReadRGBASpan = i810ReadRGBASpan_565;
       ctx->Driver.ReadRGBAPixels = i810ReadRGBAPixels_565;
    } else {
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810state.c b/xc/lib/GL/mesa/src/drv/i810/i810state.c
index cdccbd5ec..50a916fb7 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810state.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810state.c
@@ -182,6 +182,7 @@ static void i810DDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
    GLubyte p[4];
    int i,j,k;
    int active = (ctx->Polygon.StippleFlag && ctx->PB->primitive == GL_POLYGON);
+   GLuint newMask;
 
    FLUSH_BATCH(imesa);
    ctx->Driver.TriangleCaps |= DD_TRI_STIPPLE;
@@ -191,10 +192,10 @@ static void i810DDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
       imesa->Setup[I810_CTXREG_ST1] &= ~ST1_ENABLE;
    }
 
-   p[0] = mask[0] & 0xf; p[0] |= p[0] << 4;
-   p[1] = mask[4] & 0xf; p[1] |= p[1] << 4;
-   p[2] = mask[8] & 0xf; p[2] |= p[2] << 4;
-   p[3] = mask[12] & 0xf; p[3] |= p[3] << 4;
+   p[0] = mask[12] & 0xf; p[0] |= p[0] << 4;
+   p[1] = mask[8] & 0xf; p[1] |= p[1] << 4;
+   p[2] = mask[4] & 0xf; p[2] |= p[2] << 4;
+   p[3] = mask[0] & 0xf; p[3] |= p[3] << 4;
 
    for (k = 0 ; k < 8 ; k++)
       for (j = 0 ; j < 4; j++) 
@@ -203,12 +204,19 @@ static void i810DDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
 	       ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
 	       return;
 	    }
-   
+
+   newMask = ((p[0] & 0xf) << 0) |
+             ((p[1] & 0xf) << 4) |
+             ((p[2] & 0xf) << 8) |
+             ((p[3] & 0xf) << 12);
+   if (newMask == 0xffff) {
+      /* do opaque stipple in software for conformance */
+      ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
+      return;
+   }
+
    imesa->Setup[I810_CTXREG_ST1] &= ~0xffff;
-   imesa->Setup[I810_CTXREG_ST1] |= ( ((p[0] & 0xf) << 0) |
-				      ((p[1] & 0xf) << 4) |
-				      ((p[2] & 0xf) << 8) |
-				      ((p[3] & 0xf) << 12) );   
+   imesa->Setup[I810_CTXREG_ST1] |= newMask;
 
    if (active)
       imesa->Setup[I810_CTXREG_ST1] |= ST1_ENABLE;
@@ -439,26 +447,23 @@ static GLboolean i810DDColorMask(GLcontext *ctx,
 				 GLboolean b, GLboolean a )
 {
    i810ContextPtr imesa = I810_CONTEXT( ctx );
-   GLuint tmp = 0;
-   GLuint rv = 1;
-
-   imesa->Fallback &= ~I810_FALLBACK_COLORMASK;
+   GLuint tmp;
 
-   if (r && g && b) {
-      tmp = imesa->Setup[I810_CTXREG_B2] | B2_FB_WRITE_ENABLE;
-   } else if (!r && !g && !b) {
-      tmp = imesa->Setup[I810_CTXREG_B2] & ~B2_FB_WRITE_ENABLE;
-   } else {
-      rv = 0;
+   if (r && g && b)
+      imesa->Fallback &= ~I810_FALLBACK_COLORMASK;
+   else
       imesa->Fallback |= I810_FALLBACK_COLORMASK;
-   }
-      
+
+   tmp = imesa->Setup[I810_CTXREG_B2] |
+      (B2_FB_WRITE_ENABLE | B2_UPDATE_FB_WRITE_ENABLE);
+
    if (tmp != imesa->Setup[I810_CTXREG_B2]) {
       FLUSH_BATCH(imesa);
       imesa->Setup[I810_CTXREG_B2] = tmp;
+      imesa->dirty |= I810_UPLOAD_CTX;
    }
 
-   return rv;
+   return GL_FALSE;  /* makes s/w path always do s/w masking */
 }
 
 /* Seperate specular not fully implemented in hardware...  Needs
@@ -625,16 +630,6 @@ static void i810DDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
    case GL_TEXTURE_2D:      
       FLUSH_BATCH(imesa);
       imesa->new_state |= I810_NEW_TEXTURE;
-      imesa->dirty |= I810_UPLOAD_CTX;
-      if (ctx->Texture.CurrentUnit == 0) {
-	 imesa->Setup[I810_CTXREG_MT] &= ~MT_TEXEL0_ENABLE;
-	 if (state)
-	    imesa->Setup[I810_CTXREG_MT] |= MT_TEXEL0_ENABLE;
-      } else {
-	 imesa->Setup[I810_CTXREG_MT] &= ~MT_TEXEL1_ENABLE;
-	 if (state)
-	    imesa->Setup[I810_CTXREG_MT] |= MT_TEXEL1_ENABLE;
-      }
       break;
    case GL_COLOR_LOGIC_OP:
       FLUSH_BATCH( imesa );
@@ -995,8 +990,8 @@ void i810DDInitState( i810ContextPtr imesa )
 
 #define INTERESTED (~(NEW_MODELVIEW|NEW_PROJECTION|\
                       NEW_TEXTURE_MATRIX|\
-                      NEW_USER_CLIP|NEW_CLIENT_STATE|\
-                      NEW_TEXTURE_ENABLE))
+                      NEW_USER_CLIP|NEW_CLIENT_STATE))
+
 
 void i810DDUpdateState( GLcontext *ctx )
 {
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tex.c b/xc/lib/GL/mesa/src/drv/i810/i810tex.c
index 5c1615128..ff537aece 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tex.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tex.c
@@ -705,44 +705,50 @@ static void i810UpdateTex0State( GLcontext *ctx )
    struct gl_texture_object	*tObj;
    i810TextureObjectPtr t;
    int ma_modulate_op;
+   int format;
 
-
-   tObj = ctx->Texture.Unit[0].Current;
-
-   if ( tObj != ctx->Texture.Unit[0].CurrentD[2] ) 
-      tObj = 0;
-
-
+   /* disable */
+   imesa->Setup[I810_CTXREG_MT] &= ~MT_TEXEL0_ENABLE;
    imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
-				     MC_STAGE_0 |
-				     MC_UPDATE_DEST |
-				     MC_DEST_CURRENT |
-				     MC_UPDATE_ARG1 |
-				     MC_ARG1_ITERATED_COLOR | 
-				     MC_UPDATE_ARG2 |
-				     MC_ARG2_ONE |
-				     MC_UPDATE_OP |
-				     MC_OP_ARG1 );
-
+                                     MC_STAGE_0 |
+                                     MC_UPDATE_DEST |
+                                     MC_DEST_CURRENT |
+                                     MC_UPDATE_ARG1 |
+                                     MC_ARG1_ITERATED_COLOR | 
+                                     MC_UPDATE_ARG2 |
+                                     MC_ARG2_ONE |
+                                     MC_UPDATE_OP |
+                                     MC_OP_ARG1 );
    imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
-				     MA_STAGE_0 |
-				     MA_UPDATE_ARG1 |
-				     MA_ARG1_ITERATED_ALPHA |
-				     MA_UPDATE_ARG2 |
-				     MA_ARG2_TEX0_ALPHA |
-				     MA_UPDATE_OP |
-				     MA_OP_ARG1 );
-    
+                                     MA_STAGE_0 |
+                                     MA_UPDATE_ARG1 |
+                                     MA_ARG1_ITERATED_ALPHA |
+                                     MA_UPDATE_ARG2 |
+                                     MA_ARG2_TEX0_ALPHA |
+                                     MA_UPDATE_OP |
+                                     MA_OP_ARG1 );
+
+   if (ctx->Texture.Unit[0].ReallyEnabled == 0) {
+      return;
+   }
 
-   if (!(ctx->Texture.ReallyEnabled & 0xf) || !tObj || !tObj->Complete) {
+   tObj = ctx->Texture.Unit[0].Current;
+   if (ctx->Texture.Unit[0].ReallyEnabled != TEXTURE0_2D ||
+       tObj->Image[tObj->BaseLevel]->Border > 0) {
+      /* 1D or 3D texturing enabled, or texture border - fallback */
+      imesa->Fallback |= I810_FALLBACK_TEXTURE;
       return;
    }
 
+   /* Do 2D texture setup */
+
+   imesa->Setup[I810_CTXREG_MT] |= MT_TEXEL0_ENABLE;
+
    t = tObj->DriverData;
-  
    if (!t) {
       t = i810CreateTexObj( imesa, tObj );
-      if (!t) return;
+      if (!t)
+         return;
    }
 
    if (t->current_unit != 0)
@@ -757,9 +763,11 @@ static void i810UpdateTex0State( GLcontext *ctx )
    if (t->MemBlock)
       i810UpdateTexLRU( imesa, t );
   
+   format = t->image[0].internalFormat;
+
    switch (ctx->Texture.Unit[0].EnvMode) {
    case GL_REPLACE:
-      if (t->image[0].internalFormat == GL_ALPHA) 
+      if (format == GL_ALPHA) 
 	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
 					   MC_STAGE_0 |
 					   MC_UPDATE_DEST |
@@ -782,7 +790,7 @@ static void i810UpdateTex0State( GLcontext *ctx )
 					   MC_UPDATE_OP |
 					   MC_OP_ARG1 );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 ma_modulate_op = MA_OP_ARG1;
       } else {
 	 ma_modulate_op = MA_OP_ARG2;
@@ -809,7 +817,7 @@ static void i810UpdateTex0State( GLcontext *ctx )
 					MC_UPDATE_OP |
 					MC_OP_MODULATE );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 ma_modulate_op = MA_OP_ARG1;
       } else {
 	 ma_modulate_op = MA_OP_MODULATE;
@@ -826,30 +834,74 @@ static void i810UpdateTex0State( GLcontext *ctx )
       break;
 
    case GL_ADD:
-      imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
-					MC_STAGE_0 |
-					MC_UPDATE_DEST |
-					MC_DEST_CURRENT |
-					MC_UPDATE_ARG1 |
-					MC_ARG1_TEX0_COLOR | 
-					MC_UPDATE_ARG2 |
-					MC_ARG2_ITERATED_COLOR |
-					MC_UPDATE_OP |
-					MC_OP_ADD );
+      if (format == GL_ALPHA) {
+         /* Cv = Cf */
+	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
+					   MC_STAGE_0 |
+					   MC_UPDATE_DEST |
+					   MC_DEST_CURRENT |
+					   MC_UPDATE_ARG1 |
+					   MC_ARG1_TEX0_COLOR | 
+					   MC_UPDATE_ARG2 |
+					   MC_ARG2_ITERATED_COLOR |
+					   MC_UPDATE_OP |
+					   MC_OP_ARG2 );
+      }
+      else {
+         /* Cv = Cf + Ct */
+         imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
+                                           MC_STAGE_0 |
+                                           MC_UPDATE_DEST |
+                                           MC_DEST_CURRENT |
+                                           MC_UPDATE_ARG1 |
+                                           MC_ARG1_TEX0_COLOR | 
+                                           MC_UPDATE_ARG2 |
+                                           MC_ARG2_ITERATED_COLOR |
+                                           MC_UPDATE_OP |
+                                           MC_OP_ADD );
+      }
 
-      imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
-					MA_STAGE_0 |
-					MA_UPDATE_ARG1 |
-					MA_ARG1_ITERATED_ALPHA |
-					MA_UPDATE_ARG2 |
-					MA_ARG2_TEX0_ALPHA |
-					MA_UPDATE_OP |
-					MA_OP_ADD );
+      /* alpha */
+      if (format == GL_ALPHA ||
+          format == GL_LUMINANCE_ALPHA ||
+          format == GL_RGBA) {
+         /* Av = Af * At */
+         imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
+                                           MA_STAGE_0 |
+                                           MA_UPDATE_ARG1 |
+                                           MA_ARG1_ITERATED_ALPHA |
+                                           MA_UPDATE_ARG2 |
+                                           MA_ARG2_TEX0_ALPHA |
+                                           MA_UPDATE_OP |
+                                           MA_OP_MODULATE );
+      }
+      else if (format == GL_LUMINANCE || format == GL_RGB) {
+         /* Av = Af */
+         imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
+                                           MA_STAGE_0 |
+                                           MA_UPDATE_ARG1 |
+                                           MA_ARG1_ITERATED_ALPHA |
+                                           MA_UPDATE_ARG2 |
+                                           MA_ARG2_ITERATED_ALPHA |
+                                           MA_UPDATE_OP |
+                                           MA_OP_ARG1 );
+      }
+      else {
+         /* Av = Af + At */
+         imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
+                                           MA_STAGE_0 |
+                                           MA_UPDATE_ARG1 |
+                                           MA_ARG1_ITERATED_ALPHA |
+                                           MA_UPDATE_ARG2 |
+                                           MA_ARG2_TEX0_ALPHA |
+                                           MA_UPDATE_OP |
+                                           MA_OP_ADD );
+      }
       break;
 
    case GL_DECAL:
-
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
+         /* C = Ct */
 	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
 					   MC_STAGE_0 |
 					   MC_UPDATE_DEST |
@@ -862,29 +914,32 @@ static void i810UpdateTex0State( GLcontext *ctx )
 					   MC_OP_ARG2 );
 
       } else {
+         /* RGBA or undefined result */
+         /* C = Cf*(1-At)+Ct*At */
 	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
 					   MC_STAGE_0 |
 					   MC_UPDATE_DEST |
 					   MC_DEST_CURRENT |
 					   MC_UPDATE_ARG1 |
-					   MC_ARG1_COLOR_FACTOR | 
+					   MC_ARG1_TEX0_COLOR |
 					   MC_UPDATE_ARG2 |
-					   MC_ARG2_TEX0_COLOR |
+					   MC_ARG2_ITERATED_COLOR | 
 					   MC_UPDATE_OP |
 					   MC_OP_LIN_BLEND_TEX0_ALPHA );
       }
 
+      /* Av = Af */
       imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
 					MA_STAGE_0 |
 					MA_UPDATE_ARG1 |
-					MA_ARG1_ALPHA_FACTOR |
+					MA_ARG1_ITERATED_ALPHA |
 					MA_UPDATE_ARG2 |
-					MA_ARG2_ALPHA_FACTOR |
+					MA_ARG2_ITERATED_ALPHA |
 					MA_UPDATE_OP |
 					MA_OP_ARG1 );
       break;
    case GL_BLEND:
-      if (t->image[0].internalFormat == GL_ALPHA) 
+      if (format == GL_ALPHA) 
 	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
 					   MC_STAGE_0 |
 					   MC_UPDATE_DEST |
@@ -907,7 +962,20 @@ static void i810UpdateTex0State( GLcontext *ctx )
 					   MC_UPDATE_OP |
 					   MC_OP_LIN_BLEND_TEX0_COLOR );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      /* alpha */
+      if (format == GL_LUMINANCE || format == GL_RGB) {
+         /* Av = Af */
+         imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
+                                           MA_STAGE_0 |
+                                           MA_UPDATE_ARG1 |
+                                           MA_ARG1_ITERATED_ALPHA |
+                                           MA_UPDATE_ARG2 |
+                                           MA_ARG2_ITERATED_ALPHA |
+                                           MA_UPDATE_OP |
+                                           MA_OP_ARG1 );
+      }
+      else if (format == GL_INTENSITY) {
+         /* Av = Af(1-It)+AcIt */
 	 imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
 					   MA_STAGE_0 |
 					   MA_UPDATE_ARG1 |
@@ -915,16 +983,17 @@ static void i810UpdateTex0State( GLcontext *ctx )
 					   MA_UPDATE_ARG2 |
 					   MA_ARG2_ITERATED_ALPHA |
 					   MA_UPDATE_OP |
-					   MA_OP_ARG1 );
+					   MA_OP_LIN_BLEND_TEX0_ALPHA );
       } else {
+         /* Av = AfAt */
 	 imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
 					   MA_STAGE_0 |
 					   MA_UPDATE_ARG1 |
-					   MA_ARG1_ALPHA_FACTOR |
+					   MA_ARG1_TEX0_ALPHA |
 					   MA_UPDATE_ARG2 |
 					   MA_ARG2_ITERATED_ALPHA |
 					   MA_UPDATE_OP |
-					   MA_OP_LIN_BLEND_TEX0_ALPHA );
+					   MA_OP_MODULATE );
       }
       break;
 
@@ -942,50 +1011,56 @@ static void i810UpdateTex1State( GLcontext *ctx )
    i810ContextPtr imesa = I810_CONTEXT(ctx);
    struct gl_texture_object	*tObj;
    i810TextureObjectPtr t;
-   int ma_modulate_op;
-
-
-   tObj = ctx->Texture.Unit[1].Current;
-
-   if ( tObj != ctx->Texture.Unit[1].CurrentD[2] ) 
-      tObj = 0;
-
+   int ma_modulate_op, format;
 
+   /* disable */
+   imesa->Setup[I810_CTXREG_MT] &= ~MT_TEXEL1_ENABLE;
    imesa->Setup[I810_CTXREG_MC1] = ( GFX_OP_MAP_COLOR_STAGES |
-				     MC_STAGE_1 |
-				     MC_UPDATE_DEST |
-				     MC_DEST_CURRENT |
-				     MC_UPDATE_ARG1 |
-				     MC_ARG1_ONE | 
-				     MC_ARG1_DONT_REPLICATE_ALPHA |
-				     MC_ARG1_DONT_INVERT |
-				     MC_UPDATE_ARG2 |
-				     MC_ARG2_ONE |
-				     MC_ARG2_DONT_REPLICATE_ALPHA |
-				     MC_ARG2_DONT_INVERT |
-				     MC_UPDATE_OP |
-				     MC_OP_DISABLE );
-
+                                     MC_STAGE_1 |
+                                     MC_UPDATE_DEST |
+                                     MC_DEST_CURRENT |
+                                     MC_UPDATE_ARG1 |
+                                     MC_ARG1_ONE | 
+                                     MC_ARG1_DONT_REPLICATE_ALPHA |
+                                     MC_ARG1_DONT_INVERT |
+                                     MC_UPDATE_ARG2 |
+                                     MC_ARG2_ONE |
+                                     MC_ARG2_DONT_REPLICATE_ALPHA |
+                                     MC_ARG2_DONT_INVERT |
+                                     MC_UPDATE_OP |
+                                     MC_OP_DISABLE );
    imesa->Setup[I810_CTXREG_MA1] = ( GFX_OP_MAP_ALPHA_STAGES |
-				     MA_STAGE_1 |
-				     MA_UPDATE_ARG1 |
-				     MA_ARG1_CURRENT_ALPHA |
-				     MA_ARG1_DONT_INVERT |
-				     MA_UPDATE_ARG2 |
-				     MA_ARG2_CURRENT_ALPHA |
-				     MA_ARG2_DONT_INVERT |
-				     MA_UPDATE_OP |
-				     MA_OP_ARG1 );
-
-   if (!(ctx->Texture.ReallyEnabled & 0xf0) || !tObj || !tObj->Complete) {
+                                     MA_STAGE_1 |
+                                     MA_UPDATE_ARG1 |
+                                     MA_ARG1_CURRENT_ALPHA |
+                                     MA_ARG1_DONT_INVERT |
+                                     MA_UPDATE_ARG2 |
+                                     MA_ARG2_CURRENT_ALPHA |
+                                     MA_ARG2_DONT_INVERT |
+                                     MA_UPDATE_OP |
+                                     MA_OP_ARG1 );
+
+   if (ctx->Texture.Unit[1].ReallyEnabled == 0) {
+      return;
+   }
+
+   tObj = ctx->Texture.Unit[1].Current;
+   if (ctx->Texture.Unit[1].ReallyEnabled != TEXTURE0_2D ||
+       tObj->Image[tObj->BaseLevel]->Border > 0) {
+      /* 1D or 3D texturing enabled, or texture border - fallback */
+      imesa->Fallback |= I810_FALLBACK_TEXTURE;
       return;
    }
 
+   /* Do 2D texture setup */
+
+   imesa->Setup[I810_CTXREG_MT] |= MT_TEXEL1_ENABLE;
+
    t = tObj->DriverData;
-  
    if (!t) {
       t = i810CreateTexObj( imesa, tObj );
-      if (!t) return;
+      if (!t)
+         return;
    }
     
    if (t->current_unit != 1)
@@ -1000,6 +1075,8 @@ static void i810UpdateTex1State( GLcontext *ctx )
    if (t->MemBlock)
       i810UpdateTexLRU( imesa, t );
 
+   format = t->image[0].internalFormat;
+
    switch (ctx->Texture.Unit[1].EnvMode) {
    case GL_REPLACE:
       imesa->Setup[I810_CTXREG_MC1] = ( GFX_OP_MAP_COLOR_STAGES |
@@ -1013,7 +1090,7 @@ static void i810UpdateTex1State( GLcontext *ctx )
 					MC_UPDATE_OP |
 					MC_OP_ARG1 );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 ma_modulate_op = MA_OP_ARG1;
       } else {
 	 ma_modulate_op = MA_OP_ARG2;
@@ -1040,7 +1117,7 @@ static void i810UpdateTex1State( GLcontext *ctx )
 					MC_UPDATE_OP |
 					MC_OP_MODULATE );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 ma_modulate_op = MA_OP_ARG1;
       } else {
 	 ma_modulate_op = MA_OP_MODULATE;
@@ -1068,7 +1145,7 @@ static void i810UpdateTex1State( GLcontext *ctx )
 					MC_UPDATE_OP |
 					MC_OP_ADD );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 ma_modulate_op = MA_OP_ARG1;
       } else {
 	 ma_modulate_op = MA_OP_ADD;
@@ -1086,7 +1163,7 @@ static void i810UpdateTex1State( GLcontext *ctx )
 
 
    case GL_DECAL:
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 imesa->Setup[I810_CTXREG_MC1] = ( GFX_OP_MAP_COLOR_STAGES |
 					   MC_STAGE_1 |
 					   MC_UPDATE_DEST |
@@ -1133,7 +1210,7 @@ static void i810UpdateTex1State( GLcontext *ctx )
 					MC_UPDATE_OP |
 					MC_OP_LIN_BLEND_TEX1_COLOR );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 imesa->Setup[I810_CTXREG_MA1] = ( GFX_OP_MAP_ALPHA_STAGES |
 					   MA_STAGE_1 |
 					   MA_UPDATE_ARG1 |
@@ -1169,11 +1246,12 @@ void i810UpdateTextureState( GLcontext *ctx )
    if (imesa->CurrentTexObj[1]) imesa->CurrentTexObj[1]->bound = 0;
    imesa->CurrentTexObj[0] = 0;
    imesa->CurrentTexObj[1] = 0;   
+   imesa->Fallback &= ~I810_FALLBACK_TEXTURE;
    i810UpdateTex0State( ctx );
    i810UpdateTex1State( ctx );
    I810_CONTEXT( ctx )->dirty |= (I810_UPLOAD_CTX |
-				  I810_UPLOAD_TEX0 | 
-				  I810_UPLOAD_TEX1);
+                                  I810_UPLOAD_TEX0 | 
+                                  I810_UPLOAD_TEX1);
 }
 
 
@@ -1196,13 +1274,17 @@ static void i810TexEnv( GLcontext *ctx, GLenum target,
 
       struct gl_texture_unit *texUnit = 
 	 &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
-      GLfloat *fc = texUnit->EnvColor;
-      GLuint col;
-
-      col = ((((GLubyte)fc[3])<<24) | 
-	     (((GLubyte)fc[0])<<16) | 
-	     (((GLubyte)fc[1])<<8) | 
-	     (((GLubyte)fc[2])<<0));
+      const GLfloat *fc = texUnit->EnvColor;
+      GLuint r, g, b, a, col;
+      FLOAT_COLOR_TO_UBYTE_COLOR(r, fc[0]);
+      FLOAT_COLOR_TO_UBYTE_COLOR(g, fc[1]);
+      FLOAT_COLOR_TO_UBYTE_COLOR(b, fc[2]);
+      FLOAT_COLOR_TO_UBYTE_COLOR(a, fc[3]);
+
+      col = ((a << 24) | 
+	     (r << 16) | 
+	     (g <<  8) | 
+	     (b <<  0));
     
       if (imesa->Setup[I810_CTXREG_CF1] != col) {
 	 FLUSH_BATCH(imesa);	
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tris.h b/xc/lib/GL/mesa/src/drv/i810/i810tris.h
index a83adad95..a416f82dc 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tris.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tris.h
@@ -87,40 +87,42 @@ static __inline__ void i810_draw_point( i810ContextPtr imesa,
 {
    int vertsize = imesa->vertsize;
    GLuint *vb = i810AllocDwordsInline( imesa, 6 * vertsize );
+   const GLfloat x = tmp->v.x + 0.125;
+   const GLfloat y = tmp->v.y - 0.5F;
    int j;
 
-   *(float *)&vb[0] = tmp->v.x - sz;
-   *(float *)&vb[1] = tmp->v.y - sz;
+   *(float *)&vb[0] = x - sz;
+   *(float *)&vb[1] = y - sz;
    for (j = 2 ; j < vertsize ; j++)
       vb[j] = tmp->ui[j];
    vb += vertsize;
 
-   *(float *)&vb[0] = tmp->v.x + sz;
-   *(float *)&vb[1] = tmp->v.y - sz;
+   *(float *)&vb[0] = x + sz;
+   *(float *)&vb[1] = y - sz;
    for (j = 2 ; j < vertsize ; j++)
       vb[j] = tmp->ui[j];
    vb += vertsize;
 
-   *(float *)&vb[0] = tmp->v.x + sz;
-   *(float *)&vb[1] = tmp->v.y + sz;
+   *(float *)&vb[0] = x + sz;
+   *(float *)&vb[1] = y + sz;
    for (j = 2 ; j < vertsize ; j++)
       vb[j] = tmp->ui[j];
    vb += vertsize;
 
-   *(float *)&vb[0] = tmp->v.x + sz;
-   *(float *)&vb[1] = tmp->v.y + sz;
+   *(float *)&vb[0] = x + sz;
+   *(float *)&vb[1] = y + sz;
    for (j = 2 ; j < vertsize ; j++)
       vb[j] = tmp->ui[j];
    vb += vertsize;
 
-   *(float *)&vb[0] = tmp->v.x - sz;
-   *(float *)&vb[1] = tmp->v.y + sz;
+   *(float *)&vb[0] = x - sz;
+   *(float *)&vb[1] = y + sz;
    for (j = 2 ; j < vertsize ; j++)
       vb[j] = tmp->ui[j];
    vb += vertsize;
 
-   *(float *)&vb[0] = tmp->v.x - sz;
-   *(float *)&vb[1] = tmp->v.y - sz;
+   *(float *)&vb[0] = x - sz;
+   *(float *)&vb[1] = y - sz;
    for (j = 2 ; j < vertsize ; j++)
       vb[j] = tmp->ui[j];
 }
@@ -143,13 +145,33 @@ static __inline__ void i810_draw_line( i810ContextPtr imesa,
 			 : "=%c" (j)
 			 : "0" (vertsize), "S" ((long)v1)
 			 : "memory" );
+#elif 0
+   for (j = 0 ; j < vertsize ; j++)
+      vb[j] = v0->ui[j];
+
+   vb += vertsize;
+   for (j = 0 ; j < vertsize ; j++)
+      vb[j] = v1->ui[j];
 #else
+   const GLfloat dx = -0.5;
+   const GLfloat dy = -0.5;
+
+   v0->v.x += dx;
+   v0->v.y += dy;
+   v1->v.x += dx;
+   v1->v.y += dy;
+
    for (j = 0 ; j < vertsize ; j++)
       vb[j] = v0->ui[j];
 
    vb += vertsize;
    for (j = 0 ; j < vertsize ; j++)
       vb[j] = v1->ui[j];
+
+   v0->v.x -= dx;
+   v0->v.y -= dy;
+   v1->v.x -= dx;
+   v1->v.y -= dy;
 #endif
 }
 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810vb.c b/xc/lib/GL/mesa/src/drv/i810/i810vb.c
index a6ad760a6..5ecc3e2af 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810vb.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810vb.c
@@ -87,15 +87,18 @@
 
 #define COORD					\
       GLfloat *win = VB->Win.data[i];		\
-      v->v.x =                 win[0];	\
-      v->v.y =    i810height - win[1];	\
-      v->v.z = (1.0/0x10000) * win[2];	\
+      v->v.x =       xoffset + win[0];		\
+      v->v.y =       yoffset - win[1];		\
+      v->v.z = (1.0/0x10000) * win[2];		\
       v->v.oow =               win[3]; 
 
 
 
 #define NOP
 
+#define SUBPIXEL_X -0.5
+#define SUBPIXEL_Y -0.375
+
 
 #define SETUPFUNC(name,win,col,tex0,tex1,tex0_4,spec,fog)		\
 static void name(struct vertex_buffer *VB, GLuint start, GLuint end)	\
@@ -105,9 +108,12 @@ static void name(struct vertex_buffer *VB, GLuint start, GLuint end)	\
    i810VertexPtr v;							\
    GLfloat (*tc0)[4];							\
    GLfloat (*tc1)[4];							\
-   GLfloat i810height = dPriv->h;					\
+   const GLfloat xoffset = SUBPIXEL_X;					\
+   const GLfloat yoffset = dPriv->h + SUBPIXEL_Y;			\
    int i;								\
-   (void) i810height; (void) imesa;					\
+   (void) xoffset;							\
+   (void) yoffset;							\
+   (void) imesa;							\
 									\
 									\
    gl_import_client_data( VB, VB->ctx->RenderFlags,			\
@@ -269,7 +275,7 @@ void i810ChooseRasterSetupFunc(GLcontext *ctx)
      funcindex |= I810_TEX0_BIT;
   
   if (ctx->Texture.ReallyEnabled & 0xf0) {
-     funcindex |= I810_TEX1_BIT;
+     funcindex |= (I810_TEX0_BIT | I810_TEX1_BIT);
      imesa->vertsize = 10;
      imesa->Setup[I810_CTXREG_VF] = I810_VFMT_T0T1;
   }
diff --git a/xc/lib/GL/mesa/src/drv/mga/Imakefile b/xc/lib/GL/mesa/src/drv/mga/Imakefile
index 2630d8159..e4c73ddc0 100644
--- a/xc/lib/GL/mesa/src/drv/mga/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/mga/Imakefile
@@ -151,6 +151,7 @@ MESA_INCLUDES = -I. -I.. -I../../include \
 		../../stages.c \
 		../../state.c \
 		../../stencil.c \
+		../../texformat.c \
 		../../teximage.c \
 		../../texobj.c \
 		../../texstate.c \
@@ -224,6 +225,7 @@ MESA_INCLUDES = -I. -I.. -I../../include \
 		../../stages.o \
 		../../state.o \
 		../../stencil.o \
+		../../texformat.o \
 		../../teximage.o \
 		../../texobj.o \
 		../../texstate.o \
diff --git a/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c b/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c
index f428614f0..8e7bb0f24 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c
@@ -314,6 +314,7 @@ GLboolean XMesaCreateContext( Display *dpy, GLvisual *mesaVis,
    mmesa->hHWContext = driContextPriv->hHWContext;
    mmesa->driFd = sPriv->fd;
    mmesa->driHwLock = &sPriv->pSAREA->lock;
+
    mmesa->mgaScreen = mgaScreen;
    mmesa->driScreen = sPriv;
    mmesa->sarea = saPriv;
@@ -374,7 +375,7 @@ GLboolean XMesaCreateContext( Display *dpy, GLvisual *mesaVis,
       break;
    };
 
-
+   mmesa->canDoStipple = GL_FALSE;
    mmesa->renderindex = -1;		/* impossible value */
    mmesa->new_state = ~0;
    mmesa->dirty = ~0;
@@ -444,14 +445,16 @@ GLframebuffer *XMesaCreateWindowBuffer( Display *dpy,
                                         __DRIdrawablePrivate *driDrawPriv,
                                         GLvisual *mesaVis)
 {
+   GLboolean swStencil = mesaVis->StencilBits > 0 && mesaVis->DepthBits != 24;
+
    if (MGA_DEBUG&DEBUG_VERBOSE_DRI)
       fprintf(stderr, "XMesaCreateWindowBuffer\n");
 
    return gl_create_framebuffer(mesaVis,
                                 GL_FALSE,  /* software depth buffer? */
-                                mesaVis->StencilBits > 0,
+                                swStencil,
                                 mesaVis->AccumRedBits > 0,
-                                mesaVis->AlphaBits > 0
+                                GL_FALSE   /* software alpha buffer/ */
                                 );
 }
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgacontext.h b/xc/lib/GL/mesa/src/drv/mga/mgacontext.h
index 1ae30deec..e1aacb193 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgacontext.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgacontext.h
@@ -57,11 +57,13 @@
  *    - texture env GL_BLEND -- can be fixed
  *    - 1D and 3D textures
  *    - incomplete textures
+ *    - GL_DEPTH_FUNC == GL_NEVER not in h/w
  */
 #define MGA_FALLBACK_TEXTURE   0x1
 #define MGA_FALLBACK_BUFFER    0x2
 #define MGA_FALLBACK_LOGICOP   0x4
 #define MGA_FALLBACK_STENCIL   0x8
+#define MGA_FALLBACK_DEPTH     0x10
 
 
 /* For mgaCtx->new_state.
@@ -191,6 +193,7 @@ struct mga_context_t {
    GLuint        depth_clear_mask;
    GLuint        stencil_clear_mask;
    GLuint        hw_stencil;
+   GLboolean     canDoStipple;
 
    /* Dma buffers
     */
@@ -220,7 +223,6 @@ struct mga_context_t {
     */
    unsigned int texAge[MGA_NR_TEX_HEAPS];/* texture LRU age  */
    unsigned int dirtyAge;		/* buffer age for synchronization */
-   unsigned int lastSwap;		/* throttling runaway apps */
    GLuint *status;
    GLuint status_offset;
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgadd.c b/xc/lib/GL/mesa/src/drv/mga/mgadd.c
index dced4235b..6f055c5be 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgadd.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgadd.c
@@ -51,7 +51,7 @@
 #include "X86/common_x86_asm.h"
 #endif
 
-#define MGA_DATE	"20010215"
+#define MGA_DATE	"20010306"
 
 
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
index c803ac5b5..c31d23eca 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
@@ -28,7 +28,7 @@ static void mga_iload_dma_ioctl(mgaContextPtr mmesa,
 
    if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
       fprintf(stderr, "DRM_IOCTL_MGA_ILOAD idx %d dst %x length %d\n",
-	      buf->idx, dest, length);
+	      buf->idx, (int) dest, length);
 
    ret = drmMGATextureLoad( mmesa->driFd, buf->idx, dest, length );
    if ( ret < 0 ) {
@@ -156,8 +156,9 @@ drmBufPtr mga_get_buffer_ioctl( mgaContextPtr mmesa )
 
 
 
-GLbitfield mgaDDClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
-		       GLint cx, GLint cy, GLint cw, GLint ch )
+static GLbitfield
+mgaDDClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
+            GLint cx, GLint cy, GLint cw, GLint ch )
 {
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
    __DRIdrawablePrivate *dPriv = mmesa->driDrawable;
@@ -290,7 +291,6 @@ void mgaSwapBuffers( mgaContextPtr mmesa )
    int nbox;
    int ret;
    int i;
-   int tmp;
    GLuint last_frame, last_wrap;
 
    FLUSH_BATCH( mmesa );
@@ -300,6 +300,16 @@ void mgaSwapBuffers( mgaContextPtr mmesa )
    last_frame = mmesa->sarea->last_frame.head;
    last_wrap = mmesa->sarea->last_frame.wrap;
 
+   /* FIXME: Add a timeout to this loop...
+    */
+   while ( 1 ) {
+      if ( last_wrap < mmesa->sarea->last_wrap ||
+	   ( last_wrap == mmesa->sarea->last_wrap &&
+	     last_frame <= *mmesa->status - mmesa->status_offset ) ) {
+	 break;
+      }
+   }
+
    /* Use the frontbuffer cliprects
     */
    if (mmesa->dirty_cliprects & MGA_FRONT)
@@ -329,32 +339,8 @@ void mgaSwapBuffers( mgaContextPtr mmesa )
       }
    }
 
-   if ( 0 ) {
-      fprintf( stderr, "waiting...\n" );
-   }
-
-   while ( 1 ) {
-      if ( last_wrap < mmesa->sarea->last_wrap ||
-	   ( last_wrap == mmesa->sarea->last_wrap &&
-	      last_frame <= *mmesa->status - mmesa->status_offset ) ) {
-	 break;
-      }
-      if ( 0 ) {
-	 fprintf( stderr, "   head=0x%06x wrap=%d\n",
-		  *mmesa->status - mmesa->status_offset,
-		  mmesa->sarea->last_wrap );
-	 fprintf( stderr, "  frame=0x%06x wrap=%d\n\n",
-		  last_frame, last_wrap );
-      }
-   }
-
-   if ( 0 )
-      fprintf( stderr, "waiting... done.\n" );
-
    UNLOCK_HARDWARE( mmesa );
 
-
-   mmesa->lastSwap = tmp;
    mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
 }
 
@@ -408,8 +394,8 @@ static int intersect_rect( XF86DRIClipRectPtr out,
    if (b->y1 > out->y1) out->y1 = b->y1;
    if (b->x2 < out->x2) out->x2 = b->x2;
    if (b->y2 < out->y2) out->y2 = b->y2;
-   if (out->x1 >= out->x2) return 0;
-   if (out->y1 >= out->y2) return 0;
+   if (out->x1 > out->x2) return 0;
+   if (out->y1 > out->y2) return 0;
    return 1;
 }
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgapixel.c b/xc/lib/GL/mesa/src/drv/mga/mgapixel.c
index d970e49b4..72903c209 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgapixel.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgapixel.c
@@ -545,9 +545,9 @@ mgaDDDrawPixels( GLcontext *ctx,
 	  * negligble speedup when the buffers/images don't exactly
 	  * match:
 	  */
-#if 1
+#if 0
 	 if (cpp == 2) {
-	    if (!_mesa_convert_teximage( MESA_R5_G6_B5,
+	    if (!_mesa_convert_teximage( MESA_FORMAT_RGB565,
 					 width, rows,
 					 address, bufferpitch,
 					 width, rows,
@@ -558,7 +558,7 @@ mgaDDDrawPixels( GLcontext *ctx,
 	       return GL_FALSE;
 	    }
 	 } else {
-	    if (!_mesa_convert_teximage( MESA_A8_R8_G8_B8,
+	    if (!_mesa_convert_teximage( MESA_FORMAT_ARGB8888,
 					 width, rows,
 					 address, bufferpitch,
 					 width, rows,
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaspan.c b/xc/lib/GL/mesa/src/drv/mga/mgaspan.c
index b707bc406..83a78f2dc 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgaspan.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaspan.c
@@ -25,7 +25,6 @@
 			dPriv->y * pitch);		\
    GLuint p = MGA_CONTEXT( ctx )->MonoColor;		\
    (void) read_buf; (void) buf; (void) p
-   
 
 
 #define LOCAL_DEPTH_VARS				\
@@ -59,8 +58,10 @@
 
 #define HW_LOCK()				\
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);	\
+   FLUSH_BATCH(mmesa);				\
    LOCK_HARDWARE_QUIESCENT(mmesa);
 
+
 #define HW_CLIPLOOP()						\
   do {								\
     int _nc = mmesa->numClipRects;				\
@@ -97,9 +98,9 @@
 #define READ_RGBA( rgba, _x, _y )				\
 do {								\
    GLushort p = *(GLushort *)(read_buf + _x*2 + _y*pitch);	\
-   rgba[0] = (p >> 8) & 0xf8;					\
-   rgba[1] = (p >> 3) & 0xfc;					\
-   rgba[2] = (p << 3) & 0xf8;					\
+   rgba[0] = (((p >> 11) & 0x1f) * 255) / 31;			\
+   rgba[1] = (((p >>  5) & 0x3f) * 255) / 63;			\
+   rgba[2] = (((p >>  0) & 0x1f) * 255) / 31;			\
    rgba[3] = 255;						\
 } while(0)
 
@@ -127,7 +128,7 @@ do {								\
 	rgba[0] = (p >> 16) & 0xff;				\
 	rgba[1] = (p >> 8)  & 0xff;				\
 	rgba[2] = (p >> 0)  & 0xff;				\
-	rgba[3] = (p >> 24) & 0xff;				\
+	rgba[3] = 0xff;						\
     } while (0)
 
 #define TAG(x) mga##x##_8888
@@ -168,13 +169,13 @@ do {								\
 #define WRITE_DEPTH( _x, _y, d ) {			\
    GLuint tmp = *(GLuint *)(buf + _x*4 + _y*pitch);	\
    tmp &= 0xff;						\
-   tmp |= (d) & 0xffffff00;				\
+   tmp |= (d) << 8;					\
    *(GLuint *)(buf + _x*4 + _y*pitch) = tmp;		\
 }
 
-#define READ_DEPTH( d, _x, _y )		\
-   d = *(GLuint *)(buf + _x*4 + _y*pitch) & ~0xff;	
-
+#define READ_DEPTH( d, _x, _y )	{				\
+   d = (*(GLuint *)(buf + _x*4 + _y*pitch) & ~0xff) >> 8;	\
+}
 
 #define TAG(x) mga##x##_24_8
 #include "depthtmp.h"
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgastate.c b/xc/lib/GL/mesa/src/drv/mga/mgastate.c
index eb45233e0..54cc4de29 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgastate.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgastate.c
@@ -190,7 +190,9 @@ static void mgaUpdateZMode(const GLcontext *ctx)
    if (ctx->Depth.Test) {
       switch(ctx->Depth.Func)  {
       case GL_NEVER:
-	 zmode = DC_zmode_nozcmp; break;
+         /* can't do this in h/w, we'll use a s/w fallback */
+	 zmode = DC_zmode_nozcmp;
+         break;
       case GL_ALWAYS:
 	 zmode = DC_zmode_nozcmp; break;
       case GL_LESS:
@@ -208,15 +210,17 @@ static void mgaUpdateZMode(const GLcontext *ctx)
       default:
 	 break;
       }
-   } else {
+
+      if (ctx->Depth.Mask)
+         zmode |= DC_atype_zi;
+      else
+         zmode |= DC_atype_i;
+   }
+   else {
       zmode |= DC_zmode_nozcmp;
+      zmode |= DC_atype_i;  /* don't write to zbuffer */
    }
 
-   if (ctx->Depth.Mask)
-      zmode |= DC_atype_zi;
-   else
-      zmode |= DC_atype_i;
-
 #if defined(ACCEL_ROP)
    mmesa->setup.dwgctl &= DC_bop_MASK;
    if (ctx->Color.ColorLogicOpEnabled)
@@ -287,6 +291,10 @@ static void mgaDDDepthFunc(GLcontext *ctx, GLenum func)
 {
    FLUSH_BATCH( MGA_CONTEXT(ctx) );
    MGA_CONTEXT(ctx)->new_state |= MGA_NEW_DEPTH;
+   if (func == GL_NEVER && ctx->Depth.Test)
+      MGA_CONTEXT(ctx)->Fallback |= MGA_FALLBACK_DEPTH;
+   else
+      MGA_CONTEXT(ctx)->Fallback &= ~MGA_FALLBACK_DEPTH;
 }
 
 static void mgaDDDepthMask(GLcontext *ctx, GLboolean flag)
@@ -439,7 +447,10 @@ static void mgaUpdateAlphaMode(GLcontext *ctx)
 	    a |= AC_src_zero;
 	 break;
       case GL_SRC_ALPHA_SATURATE:
-	 a |= AC_src_src_alpha_sat;
+         if (ctx->Visual->AlphaBits > 0)
+            a |= AC_src_src_alpha_sat;
+         else
+            a |= AC_src_zero;
 	 break;
       default:		/* never happens */
 	 break;
@@ -499,10 +510,10 @@ void mgaUpdateClipping(const GLcontext *ctx)
    if (mmesa->driDrawable)
    {
       int x1 = mmesa->driDrawable->x + ctx->Scissor.X;
-      int y1 = mmesa->driDrawable->y + mmesa->driDrawable->h - (ctx->Scissor.Y+
-								ctx->Scissor.Height);
-      int x2 = mmesa->driDrawable->x + ctx->Scissor.X+ctx->Scissor.Width;
-      int y2 = mmesa->driDrawable->y + mmesa->driDrawable->h - ctx->Scissor.Y;
+      int y1 = mmesa->driDrawable->y + mmesa->driDrawable->h
+               - (ctx->Scissor.Y + ctx->Scissor.Height);
+      int x2 = x1 + ctx->Scissor.Width - 1;
+      int y2 = y1 + ctx->Scissor.Height - 1;
 
       if (x1 < 0) x1 = 0;
       if (y1 < 0) y1 = 0;
@@ -636,7 +647,7 @@ static GLboolean mgaDDColorMask(GLcontext *ctx,
       mmesa->dirty |= MGA_UPLOAD_CONTEXT;
    }
 
-   return 1;
+   return 0;  /* Kind of a hack */
 }
 
 /* =============================================================
@@ -644,9 +655,12 @@ static GLboolean mgaDDColorMask(GLcontext *ctx,
  *
  * The mga supports a subset of possible 4x4 stipples natively, GL
  * wants 32x32.  Fortunately stipple is usually a repeating pattern.
+ *
+ * Note: the fully opaque pattern (0xffff) has been disabled in order
+ * to work around a conformance issue.
  */
 static int mgaStipples[16] = {
-   0xffff,
+   0xffff1,  /* See above note */
    0xa5a5,
    0x5a5a,
    0xa0a0,
@@ -674,7 +688,12 @@ static void mgaDDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
    GLuint stipple;
 
    FLUSH_BATCH(mmesa);
-   ctx->Driver.TriangleCaps |= DD_TRI_STIPPLE;
+
+   /* Turn off flags.  We'll turn them on below if this stipple pattern
+    * works in h/w.
+    */
+   ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
+   mmesa->canDoStipple = GL_FALSE;
 
    if (active) {
       mmesa->dirty |= MGA_UPLOAD_CONTEXT;
@@ -690,7 +709,6 @@ static void mgaDDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
       for (j = 0 ; j < 4; j++)
 	 for (i = 0 ; i < 4 ; i++)
 	    if (*m++ != p[j]) {
-	       ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
 	       return;
 	    }
 
@@ -706,13 +724,15 @@ static void mgaDDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
       }
 
    if (i == 16) {
-      ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
       return;
    }
 
+   mmesa->canDoStipple = GL_TRUE;
+
    if (active) {
       mmesa->setup.dwgctl &= ~(0xf<<20);
       mmesa->setup.dwgctl |= mmesa->poly_stipple;
+      ctx->Driver.TriangleCaps |= DD_TRI_STIPPLE;
    }
 }
 
@@ -778,6 +798,16 @@ void mgaEmitHwStateLocked( mgaContextPtr mmesa )
    mmesa->sarea->dirty |= mmesa->dirty;
 
    mmesa->dirty &= (MGA_UPLOAD_CLIPRECTS|MGA_WAIT_AGE);
+
+   /* This is a bit of a hack but seems to be the best place to ensure
+    * that separate specular is disabled when not needed.
+    */
+   if (mmesa->glCtx->Texture.ReallyEnabled == 0 ||
+       !mmesa->glCtx->Light.Enabled ||
+       mmesa->glCtx->Light.Model.ColorControl == GL_SINGLE_COLOR) {
+      sarea->TexState[0].texctl2 &= ~TMC_specen_enable;
+      sarea->TexState[1].texctl2 &= ~TMC_specen_enable;
+   }
 }
 
 
@@ -808,6 +838,10 @@ static void mgaDDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
    case GL_DEPTH_TEST:
       FLUSH_BATCH( mmesa );
       mmesa->new_state |= MGA_NEW_DEPTH;
+      if (ctx->Depth.Func == GL_NEVER && ctx->Depth.Test)
+         mmesa->Fallback |= MGA_FALLBACK_DEPTH;
+      else
+         mmesa->Fallback &= ~MGA_FALLBACK_DEPTH;
       break;
    case GL_SCISSOR_TEST:
       FLUSH_BATCH( mmesa );
@@ -829,14 +863,20 @@ static void mgaDDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
       mmesa->new_state |= (MGA_NEW_TEXTURE|MGA_NEW_ALPHA);
       break;
    case GL_POLYGON_STIPPLE:
-      if ((ctx->Driver.TriangleCaps & DD_TRI_STIPPLE) &&
-	  ctx->PB->primitive == GL_POLYGON)
-      {
-	 FLUSH_BATCH(mmesa);
-	 mmesa->dirty |= MGA_UPLOAD_CONTEXT;
-	 mmesa->setup.dwgctl &= ~(0xf<<20);
-	 if (state)
+      FLUSH_BATCH(mmesa);
+      mmesa->dirty |= MGA_UPLOAD_CONTEXT;
+      mmesa->setup.dwgctl &= ~(0xf<<20);
+      if (state) {
+         if (mmesa->canDoStipple && ctx->PB->primitive == GL_POLYGON) {
 	    mmesa->setup.dwgctl |= mmesa->poly_stipple;
+            ctx->Driver.TriangleCaps |= DD_TRI_STIPPLE;
+         }
+         else {
+            ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
+         }
+      }
+      else {
+         ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
       }
       break;
    case GL_COLOR_LOGIC_OP:
@@ -948,11 +988,6 @@ void mgaDDUpdateHwState( GLcontext *ctx )
 }
 
 
-
-
-
-
-
 void mgaDDReducedPrimitiveChange( GLcontext *ctx, GLenum prim )
 {
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
@@ -960,20 +995,24 @@ void mgaDDReducedPrimitiveChange( GLcontext *ctx, GLenum prim )
    FLUSH_BATCH( mmesa );
    mgaUpdateCull(ctx);
 
-   if (ctx->Polygon.StippleFlag && (ctx->Driver.TriangleCaps & DD_TRI_STIPPLE))
-   {
+   if (ctx->Polygon.StippleFlag) {
       mmesa->dirty |= MGA_UPLOAD_CONTEXT;
       mmesa->setup.dwgctl &= ~(0xf<<20);
-      if (ctx->PB->primitive == GL_POLYGON)
+      ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
+      if (ctx->PB->primitive == GL_POLYGON && mmesa->canDoStipple) {
 	 mmesa->setup.dwgctl |= mmesa->poly_stipple;
+         ctx->Driver.TriangleCaps |= DD_TRI_STIPPLE;
+      }
+   }
+   else {
+      ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
    }
 }
 
 
 #define INTERESTED (~(NEW_MODELVIEW|NEW_PROJECTION|\
                       NEW_TEXTURE_MATRIX|\
-                      NEW_USER_CLIP|NEW_CLIENT_STATE|\
-                      NEW_TEXTURE_ENABLE))
+                      NEW_USER_CLIP|NEW_CLIENT_STATE))
 
 void mgaDDUpdateState( GLcontext *ctx )
 {
@@ -999,6 +1038,9 @@ void mgaDDUpdateState( GLcontext *ctx )
       ctx->Driver.TriangleFunc=mmesa->TriangleFunc;
       ctx->Driver.QuadFunc=mmesa->QuadFunc;
    }
+   else {
+      ctx->IndirectTriangles |= mmesa->IndirectTriangles;
+   }
 }
 
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatex.c b/xc/lib/GL/mesa/src/drv/mga/mgatex.c
index 03f1e39f4..35945224a 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatex.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatex.c
@@ -37,9 +37,11 @@
 #include "mgaregs.h"
 #include "mgaioctl.h"
 
+#include "context.h"
 #include "enums.h"
 #include "simple_list.h"
 #include "mem.h"
+#include "texutil.h"
 
 #define TEX_0 1
 #define TEX_1 2
@@ -144,11 +146,166 @@ static void mgaSetTexBorderColor(mgaTextureObjectPtr t, GLubyte color[4])
 }
 
 
+static GLint mgaChooseTexFormat( mgaContextPtr mmesa,
+				 struct gl_texture_image *texImage,
+				 GLenum format, GLenum type )
+{
+   const GLboolean do32bpt = mmesa->default32BitTextures;
+   const struct gl_texture_format *texFormat;
+   GLint ret;
+
+   if ( 0 )
+      fprintf( stderr, "internal=%s format=%s type=%s\n",
+	       texImage->IntFormat == 3 ? "GL_RGB (3)" :
+	       texImage->IntFormat == 4 ? "GL_RGBA (4)" :
+	       gl_lookup_enum_by_nr( texImage->IntFormat ),
+	       gl_lookup_enum_by_nr( format ),
+	       gl_lookup_enum_by_nr( type ) );
+
+#define SET_FORMAT( r, gl )						\
+   do {									\
+      ret = (r);							\
+      texFormat = &(gl);						\
+   } while (0)
+
+#define SET_FORMAT_32BPT( r32, gl32, r16, gl16 )			\
+   do {									\
+      if ( do32bpt ) {							\
+	 ret = (r32);							\
+	 texFormat = &(gl32);						\
+      } else {								\
+	 ret = (r16);							\
+	 texFormat = &(gl16);						\
+      }									\
+   } while (0)
+
+   switch ( texImage->IntFormat ) {
+      /* GH: Bias towards GL_RGB, GL_RGBA texture formats.  This has
+       * got to be better than sticking them way down the end of this
+       * huge list.
+       */
+   case GL_RGBA:
+   case 4:
+      if ( format == GL_BGRA ) {
+	 if ( type == GL_UNSIGNED_INT_8_8_8_8_REV ) {
+	    SET_FORMAT( TMC_tformat_tw32, _mesa_texformat_argb8888 );
+	    break;
+	 } else if ( type == GL_UNSIGNED_SHORT_4_4_4_4_REV ) {
+	    SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
+	    break;
+	 } else if ( type == GL_UNSIGNED_SHORT_1_5_5_5_REV ) {
+	    SET_FORMAT( TMC_tformat_tw15, _mesa_texformat_argb1555 );
+	    break;
+	 }
+      }
+      SET_FORMAT_32BPT( TMC_tformat_tw32, _mesa_texformat_argb8888,
+			TMC_tformat_tw12, _mesa_texformat_argb4444 );
+      break;
+
+   case GL_RGB:
+   case 3:
+      if ( format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5 ) {
+	 SET_FORMAT( TMC_tformat_tw16, _mesa_texformat_rgb565 );
+	 break;
+      }
+      SET_FORMAT_32BPT( TMC_tformat_tw32, _mesa_texformat_argb8888,
+			TMC_tformat_tw16, _mesa_texformat_rgb565 );
+      break;
+
+      /* GH: Okay, keep checking as normal.  Still test for GL_RGB,
+       * GL_RGBA formats first.
+       */
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      SET_FORMAT_32BPT( TMC_tformat_tw32, _mesa_texformat_argb8888,
+			TMC_tformat_tw12, _mesa_texformat_argb4444 );
+      break;
+
+   case GL_RGBA4:
+   case GL_RGBA2:
+      SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
+      break;
+
+   case GL_RGB5_A1:
+      SET_FORMAT( TMC_tformat_tw15, _mesa_texformat_argb1555 );
+      break;
+
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+      SET_FORMAT_32BPT( TMC_tformat_tw32, _mesa_texformat_argb8888,
+			TMC_tformat_tw16, _mesa_texformat_rgb565 );
+      break;
+
+   case GL_RGB5:
+   case GL_RGB4:
+   case GL_R3_G3_B2:
+      SET_FORMAT( TMC_tformat_tw16, _mesa_texformat_rgb565 );
+      break;
+
+   case GL_ALPHA:
+   case GL_ALPHA4:
+   case GL_ALPHA8:
+   case GL_ALPHA12:
+   case GL_ALPHA16:
+      /* FIXME: This will report incorrect component sizes... */
+      SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
+      break;
+
+   case 1:
+   case GL_LUMINANCE:
+   case GL_LUMINANCE4:
+   case GL_LUMINANCE8:
+   case GL_LUMINANCE12:
+   case GL_LUMINANCE16:
+      /* FIXME: This will report incorrect component sizes... */
+      SET_FORMAT( TMC_tformat_tw16, _mesa_texformat_rgb565 );
+      break;
 
+   case 2:
+   case GL_LUMINANCE_ALPHA:
+   case GL_LUMINANCE4_ALPHA4:
+   case GL_LUMINANCE6_ALPHA2:
+   case GL_LUMINANCE8_ALPHA8:
+   case GL_LUMINANCE12_ALPHA4:
+   case GL_LUMINANCE12_ALPHA12:
+   case GL_LUMINANCE16_ALPHA16:
+      /* FIXME: This will report incorrect component sizes... */
+      SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
+      break;
 
+   case GL_INTENSITY:
+   case GL_INTENSITY4:
+   case GL_INTENSITY8:
+   case GL_INTENSITY12:
+   case GL_INTENSITY16:
+      /* FIXME: This will report incorrect component sizes... */
+      SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
+      break;
 
+   case GL_COLOR_INDEX:
+   case GL_COLOR_INDEX1_EXT:
+   case GL_COLOR_INDEX2_EXT:
+   case GL_COLOR_INDEX4_EXT:
+   case GL_COLOR_INDEX8_EXT:
+   case GL_COLOR_INDEX12_EXT:
+   case GL_COLOR_INDEX16_EXT:
+      SET_FORMAT( TMC_tformat_tw8, _mesa_texformat_ci8 );
+      break;
+
+   default:
+      fprintf( stderr, "bad texture format in mgaChooseTexFormat() %d",
+	       texImage->IntFormat );
+      return -1;
+   }
 
+   texImage->TexFormat = texFormat;
 
+   return ret;
+}
 
 
 /*
@@ -160,60 +317,27 @@ static void mgaSetTexBorderColor(mgaTextureObjectPtr t, GLubyte color[4])
 static void mgaCreateTexObj(mgaContextPtr mmesa,
 			    struct gl_texture_object *tObj)
 {
-   const struct gl_texture_image *image = tObj->Image[ 0 ];
+   const GLint baseLevel = tObj->BaseLevel;
+   struct gl_texture_image *image = tObj->Image[baseLevel];
    mgaTextureObjectPtr t;
    int i, ofs;
    int LastLevel;
    int s, s2;
-   int textureFormat;
-
+   int tformat;
 
    if (!image) return;
 
-
    tObj->DriverData = t = CALLOC( sizeof( *t ) );
    if (!t) {
       fprintf(stderr, "mgaCreateTexObj: Failed to malloc mgaTextureObject\n" );
       return;
    }
 
-   switch( image->Format ) {
-   case GL_RGB:
-   case GL_LUMINANCE:
-      if ( image->IntFormat != GL_RGB5 && ( image->IntFormat == GL_RGB8 ||
-					    mmesa->default32BitTextures ) ) {
-	 t->texelBytes = 4;
-	 textureFormat = TMC_tformat_tw32;
-      } else {
-	 t->texelBytes = 2;
-	 textureFormat = TMC_tformat_tw16;
-      }
-      break;
-   case GL_ALPHA:
-   case GL_LUMINANCE_ALPHA:
-   case GL_INTENSITY:
-   case GL_RGBA:
-      if ( image->IntFormat != GL_RGBA4 && ( image->IntFormat == GL_RGBA8 ||
-					     mmesa->default32BitTextures ) ) {
-	 t->texelBytes = 4;
-	 textureFormat = TMC_tformat_tw32;
-      } else {
-	 t->texelBytes = 2;
-	 textureFormat = TMC_tformat_tw12;
-      }
-      break;
-   case GL_COLOR_INDEX:
-      textureFormat = TMC_tformat_tw8;
-      t->texelBytes = 1;
-      break;
-   default:
-      fprintf(stderr, "mgaCreateTexObj: bad image->Format %x/%s\n",
-	      image->Format,
-	      gl_lookup_enum_by_nr(image->Format));
-      FREE( t );
-      tObj->DriverData = 0;
-      return;
-   }
+   /* FIXME: Use the real DD interface...
+    */
+   tformat = mgaChooseTexFormat( mmesa, image, image->Format,
+				 GL_UNSIGNED_BYTE );
+   t->texelBytes = image->TexFormat->TexelBytes;
 
    /* We are going to upload all levels that are present, even if
     * later levels wouldn't be used by the current filtering mode.  This
@@ -236,7 +360,6 @@ static void mgaCreateTexObj(mgaContextPtr mmesa,
 	       MAX2( tObj->Image[i]->Height, 8 ) *
 	       t->texelBytes) + 31) & ~31;
    }
-
    t->totalSize = ofs;
    t->lastLevel = LastLevel;
    t->tObj = tObj;
@@ -249,9 +372,7 @@ static void mgaCreateTexObj(mgaContextPtr mmesa,
 
 
    /* setup hardware register values */
-   t->setup.texctl = (TMC_takey_1 |
-		      TMC_tamask_0 |
-		      textureFormat );
+   t->setup.texctl = TMC_takey_1 | TMC_tamask_0 | tformat;
 
    if (image->WidthLog2 >= 3)
       t->setup.texctl |= ((image->WidthLog2 - 3) << TMC_tpitch_SHIFT);
@@ -331,6 +452,7 @@ static void mgaUpdateTextureEnvG400( GLcontext *ctx, int unit )
    GLuint *reg = ((GLuint *)&mmesa->setup.tdualstage0 + unit);
    GLuint source = mmesa->tmu_source[unit];
    struct gl_texture_object *tObj = ctx->Texture.Unit[source].Current;
+   GLenum format;
 
    if ( tObj != ctx->Texture.Unit[source].CurrentD[2] ||
 	!tObj ||
@@ -338,95 +460,183 @@ static void mgaUpdateTextureEnvG400( GLcontext *ctx, int unit )
 	((ctx->Enabled>>(source*4))&TEXTURE0_ANY) != TEXTURE0_2D )
       return;
 
+   format = tObj->Image[tObj->BaseLevel]->Format;
 
    switch (ctx->Texture.Unit[source].EnvMode) {
    case GL_REPLACE:
-      *reg = (TD0_color_sel_arg1 |
-	      TD0_alpha_sel_arg1 );
+      if (format == GL_RGB || format == GL_LUMINANCE) {
+	 *reg = (TD0_color_sel_arg1 |
+                 TD0_alpha_arg2_diffuse |
+		 TD0_alpha_sel_arg2 );
+      }
+      else if (format == GL_ALPHA) {
+         *reg = (TD0_color_sel_arg2 |
+                 TD0_color_arg2_diffuse |
+                 TD0_alpha_sel_arg1 );
+      }
+      else {
+         *reg = (TD0_color_sel_arg1 |
+                 TD0_alpha_sel_arg1 );
+      }
       break;
 
    case GL_MODULATE:
-      if (unit == 0)
+      if (unit == 0) {
 	 *reg = ( TD0_color_arg2_diffuse |
 		  TD0_color_sel_mul |
 		  TD0_alpha_arg2_diffuse |
 		  TD0_alpha_sel_mul);
-      else
+      }
+      else {
 	 *reg = ( TD0_color_arg2_prevstage |
 		  TD0_color_alpha_prevstage |
 		  TD0_color_sel_mul |
 		  TD0_alpha_arg2_prevstage |
 		  TD0_alpha_sel_mul);
+      }
       break;
    case GL_DECAL:
-      if (tObj->Image[0]->Format == GL_RGB)
-	 *reg = (TD0_color_sel_arg1 |
-		 TD0_alpha_sel_arg1 );
-      else if (unit == 0)
-	 *reg = (TD0_color_arg2_diffuse |
-		 TD0_color_alpha_currtex |
-		 TD0_color_alpha2inv_enable |
-		 TD0_color_arg2mul_alpha2 |
-		 TD0_color_arg1mul_alpha1 |
-		 TD0_color_add_add |
-		 TD0_color_sel_add |
-		 TD0_alpha_arg2_diffuse |
-		 TD0_alpha_sel_arg2 );
-      else
-	 *reg = (TD0_color_arg2_prevstage |
-		 TD0_color_alpha_currtex |
-		 TD0_color_alpha2inv_enable |
-		 TD0_color_arg2mul_alpha2 |
-		 TD0_color_arg1mul_alpha1 |
-		 TD0_color_add_add |
-		 TD0_color_sel_add |
-		 TD0_alpha_arg2_prevstage |
-		 TD0_alpha_sel_arg2 );
-
+      if (format == GL_RGB) {
+         if (unit == 0) {
+            *reg = (TD0_color_sel_arg1 |
+                    TD0_alpha_arg2_diffuse |
+                    TD0_alpha_sel_arg2 );
+         }
+         else {
+            *reg = (TD0_color_sel_arg1 |
+                    TD0_alpha_arg2_prevstage |
+                    TD0_alpha_sel_arg2 );
+         }
+      }
+      else if ( format == GL_RGBA ) {
+#if 0
+         if (unit == 0) {
+            /* this doesn't work */
+            *reg = (TD0_color_arg2_diffuse |
+                    TD0_color_alpha_currtex |
+                    TD0_color_alpha2inv_enable |
+                    TD0_color_arg2mul_alpha2 |
+                    TD0_color_arg1mul_alpha1 |
+                    TD0_color_blend_enable |
+                    TD0_color_arg1add_mulout |
+                    TD0_color_arg2add_mulout |
+                    TD0_color_add_add |
+                    TD0_color_sel_mul |
+                    TD0_alpha_arg2_diffuse |
+                    TD0_alpha_sel_arg2 );
+         }
+         else {
+            *reg = (TD0_color_arg2_prevstage |
+                    TD0_color_alpha_currtex |
+                    TD0_color_alpha2inv_enable |
+                    TD0_color_arg2mul_alpha2 |
+                    TD0_color_arg1mul_alpha1 |
+                    TD0_color_add_add |
+                    TD0_color_sel_add |
+                    TD0_alpha_arg2_prevstage |
+                    TD0_alpha_sel_arg2 );
+         }
+#else
+         /* s/w fallback, pretty sure we can't do in h/w */
+	 mmesa->Fallback |= MGA_FALLBACK_TEXTURE;
+#endif
+      }
+      else {
+	 if (unit == 0) {
+	    *reg = ( TD0_color_arg2_diffuse |
+		     TD0_color_sel_arg2 |
+		     TD0_alpha_arg2_diffuse |
+		     TD0_alpha_sel_arg2);
+	 }
+	 else {
+	    *reg = ( TD0_color_arg2_prevstage |
+		     TD0_color_sel_arg2 |
+		     TD0_alpha_arg2_prevstage |
+		     TD0_alpha_sel_arg2);
+	 }
+      }
       break;
 
    case GL_ADD:
-      if (unit == 0)
-	 *reg = ( TD0_color_arg2_diffuse |
-		  TD0_color_add_add |
-		  TD0_color_sel_add |
-		  TD0_alpha_arg2_diffuse |
-		  TD0_alpha_sel_add);
-      else
-	 *reg = ( TD0_color_arg2_prevstage |
-		  TD0_color_alpha_prevstage |
-		  TD0_color_add_add |
-		  TD0_color_sel_add |
-		  TD0_alpha_arg2_prevstage |
-		  TD0_alpha_sel_add);
+      if (unit == 0) {
+         if (format == GL_INTENSITY)
+            *reg = ( TD0_color_arg2_diffuse |
+                     TD0_color_add_add |
+                     TD0_color_sel_add |
+                     TD0_alpha_arg2_diffuse |
+                     TD0_alpha_add_enable |
+                     TD0_alpha_sel_add);
+         else if (format == GL_ALPHA)
+            *reg = ( TD0_color_arg2_diffuse |
+                     TD0_color_sel_mul |
+                     TD0_alpha_arg2_diffuse |
+                     TD0_alpha_sel_mul);
+         else
+            *reg = ( TD0_color_arg2_diffuse |
+                     TD0_color_add_add |
+                     TD0_color_sel_add |
+                     TD0_alpha_arg2_diffuse |
+                     TD0_alpha_sel_mul);
+      }
+      else {
+         if (format == GL_INTENSITY) {
+            *reg = ( TD0_color_arg2_prevstage |
+                     TD0_color_add_add |
+                     TD0_color_sel_add |
+                     TD0_alpha_arg2_prevstage |
+                     TD0_alpha_add_enable |
+                     TD0_alpha_sel_add);
+         }
+         else if (format == GL_ALPHA) {
+            *reg = ( TD0_color_arg2_prevstage |
+                     TD0_color_sel_mul |
+                     TD0_alpha_arg2_prevstage |
+                     TD0_alpha_sel_mul);
+         }
+         else {
+            *reg = ( TD0_color_arg2_prevstage |
+                     TD0_color_alpha_prevstage |
+                     TD0_color_add_add |
+                     TD0_color_sel_add |
+                     TD0_alpha_arg2_prevstage |
+                     TD0_alpha_sel_mul);
+         }
+      }
       break;
 
    case GL_BLEND:
-      if (mmesa->blend_flags)
-	 mmesa->Fallback |= MGA_FALLBACK_TEXTURE;
-
-      /* Do singletexture GL_BLEND with 'all ones' env-color
-       * by using both texture units.  Multitexture gl_blend
-       * is a fallback.
-       */
-      if (unit == 0) {
-	 /* Part 1: R1 = Rf ( 1 - Rt )
-	  *         A1 = Af At
-	  */
+      if (format == GL_ALPHA) {
 	 *reg = ( TD0_color_arg2_diffuse |
-		  TD0_color_arg1_inv_enable |
 		  TD0_color_sel_mul |
 		  TD0_alpha_arg2_diffuse |
-		  TD0_alpha_sel_arg1);
-      } else {
-	 /* Part 2: R2 = R1 + Rt
-	  *         A2 = A1
-	  */
-	 *reg = ( TD0_color_arg2_prevstage |
-		  TD0_color_add_add |
-		  TD0_color_sel_add |
-		  TD0_alpha_arg2_prevstage |
-		  TD0_alpha_sel_arg2);
+		  TD0_alpha_sel_mul);
+      }
+      else {
+	 mmesa->Fallback |= MGA_FALLBACK_TEXTURE;
+
+         /* Do singletexture GL_BLEND with 'all ones' env-color
+          * by using both texture units.  Multitexture gl_blend
+          * is a fallback.
+          */
+         if (unit == 0) {
+            /* Part 1: R1 = Rf ( 1 - Rt )
+             *         A1 = Af At
+             */
+            *reg = ( TD0_color_arg2_diffuse |
+                     TD0_color_arg1_inv_enable |
+                     TD0_color_sel_mul |
+                     TD0_alpha_arg2_diffuse |
+                     TD0_alpha_sel_arg1);
+         } else {
+            /* Part 2: R2 = R1 + Rt
+             *         A2 = A1
+             */
+            *reg = ( TD0_color_arg2_prevstage |
+                     TD0_color_add_add |
+                     TD0_color_sel_add |
+                     TD0_alpha_arg2_prevstage |
+                     TD0_alpha_sel_arg2);
+         }
       }
       break;
    default:
@@ -459,6 +669,11 @@ static void mgaUpdateTextureObject( GLcontext *ctx, int unit )
       return;
    }
 
+   if (tObj->Image[tObj->BaseLevel]->Border > 0) {
+      mmesa->Fallback |= MGA_FALLBACK_TEXTURE;
+      return;
+   }
+
 /*     if (!tObj) tObj = ctx->Texture.Unit[0].Current; */
 /*     if (!tObj) return; */
 
@@ -597,6 +812,9 @@ static void mgaDDTexImage( GLcontext *ctx, GLenum target,
 {
    mgaContextPtr mmesa = MGA_CONTEXT( ctx );
    mgaTextureObjectPtr t;
+   GLint tformat;
+   /* hack: cast-away const */
+   struct gl_texture_image *img = (struct gl_texture_image *) image;
 
    /* just free the mga texture if it exists, it will be recreated at
       mgaUpdateTextureState time. */
@@ -608,6 +826,9 @@ static void mgaDDTexImage( GLcontext *ctx, GLenum target,
       mmesa->new_state |= MGA_NEW_TEXTURE;
    }
 
+   tformat = mgaChooseTexFormat( mmesa, img, img->Format,
+				 GL_UNSIGNED_BYTE );
+
    if (0)
       fprintf(stderr, "mgaDDTexImage tObj %p, level %d, image %p\n",
 	      tObj, level, image);
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatris.c b/xc/lib/GL/mesa/src/drv/mga/mgatris.c
index cd051bf26..4caa1bfa4 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatris.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatris.c
@@ -38,25 +38,6 @@
 #include "mgavb.h"
 
 
-static void mga_null_quad( GLcontext *ctx, GLuint v0,
-			   GLuint v1, GLuint v2, GLuint v3, GLuint pv ) 
-{
-}     
-
-static void mga_null_triangle( GLcontext *ctx, GLuint v0,
-			       GLuint v1, GLuint v2, GLuint pv ) 
-{
-}     
-
-static void mga_null_line( GLcontext *ctx, GLuint v1, GLuint v2, GLuint pv ) 
-{
-}
-
-static void mga_null_points( GLcontext *ctx, GLuint first, GLuint last ) 
-{
-}
-
-
 #define MGA_COLOR(to, from) {			\
   (to)[0] = (from)[2];				\
   (to)[1] = (from)[1];				\
@@ -64,6 +45,12 @@ static void mga_null_points( GLcontext *ctx, GLuint first, GLuint last )
   (to)[3] = (from)[3];				\
 }
 
+#define MGA_COLOR3(to, from) {			\
+  (to)[0] = (from)[2];				\
+  (to)[1] = (from)[1];				\
+  (to)[2] = (from)[0];				\
+}
+
 
 
 static triangle_func tri_tab[0x10];   
@@ -106,9 +93,6 @@ static points_func   points_tab[0x10];
 
 void mgaDDTrifuncInit()
 {
-   int i;
-
-
    init();
    init_flat();
    init_offset();
@@ -117,14 +101,6 @@ void mgaDDTrifuncInit()
    init_twoside_flat();
    init_twoside_offset();
    init_twoside_offset_flat();
-
-   for (i = 0 ; i < 0x20 ; i++) 
-      if (i & MGA_NODRAW_BIT) {
-	 quad_tab[i] = mga_null_quad; 
-	 tri_tab[i] = mga_null_triangle; 
-	 line_tab[i] = mga_null_line;
-	 points_tab[i] = mga_null_points;
-      }
 }
 
 
@@ -134,7 +110,7 @@ void mgaDDTrifuncInit()
 #define LINE_FALLBACK (ALL_FALLBACK | DD_LINE_SMOOTH | DD_LINE_STIPPLE)
 #define TRI_FALLBACK (ALL_FALLBACK | DD_TRI_SMOOTH | DD_TRI_UNFILLED)
 #define ANY_FALLBACK (POINT_FALLBACK|LINE_FALLBACK|TRI_FALLBACK|DD_TRI_STIPPLE)
-#define ANY_RASTER_FLAGS (DD_FLATSHADE|DD_TRI_LIGHT_TWOSIDE|DD_TRI_OFFSET|DD_Z_NEVER)
+#define ANY_RASTER_FLAGS (DD_FLATSHADE|DD_TRI_LIGHT_TWOSIDE|DD_TRI_OFFSET)
 
 /* Setup the Point, Line, Triangle and Quad functions based on the
    current rendering state.  Wherever possible, use the hardware to
@@ -147,6 +123,9 @@ void mgaDDChooseRenderState(GLcontext *ctx)
 
     if (mmesa->Fallback) {
 	mmesa->renderindex = MGA_FALLBACK_BIT;
+        if (flags & DD_TRI_LIGHT_TWOSIDE) {
+           mmesa->IndirectTriangles = DD_TRI_LIGHT_TWOSIDE;
+        }
 	return;
     }
 
@@ -154,7 +133,6 @@ void mgaDDChooseRenderState(GLcontext *ctx)
 	if (flags & DD_FLATSHADE)               index |= MGA_FLAT_BIT;
 	if (flags & DD_TRI_LIGHT_TWOSIDE)       index |= MGA_TWOSIDE_BIT;
 	if (flags & DD_TRI_OFFSET)              index |= MGA_OFFSET_BIT; 
-	if (flags & DD_Z_NEVER)                 index |= MGA_NODRAW_BIT; 
     }
 	
     mmesa->PointsFunc = points_tab[index];
@@ -194,6 +172,9 @@ void mgaDDChooseRenderState(GLcontext *ctx)
 	    mmesa->QuadFunc = 0;
 	    mmesa->IndirectTriangles |= (DD_TRI_SW_RASTERIZE |
 					 DD_QUAD_SW_RASTERIZE);
+            if (flags & DD_TRI_LIGHT_TWOSIDE) {
+               mmesa->IndirectTriangles |= DD_TRI_LIGHT_TWOSIDE;
+            }
 	}
     }
 }
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatris.h b/xc/lib/GL/mesa/src/drv/mga/mgatris.h
index f50c83974..9880bbdf1 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatris.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatris.h
@@ -38,8 +38,7 @@ extern void mgaDDTrifuncInit( void );
 #define MGA_FLAT_BIT	    0x1
 #define MGA_OFFSET_BIT	    0x2	
 #define MGA_TWOSIDE_BIT	    0x4	
-#define MGA_NODRAW_BIT	    0x8
-#define MGA_FALLBACK_BIT    0x10
+#define MGA_FALLBACK_BIT    0x8
 
 static __inline void mga_draw_triangle( mgaContextPtr mmesa,
 				      mgaVertex *v0, 
@@ -90,40 +89,42 @@ static __inline void mga_draw_point( mgaContextPtr mmesa,
 {
    GLuint vertsize = mmesa->vertsize;
    GLuint *wv = mgaAllocVertexDwords( mmesa, 6*vertsize);
-   int j;
+   GLuint j;
+   const GLfloat x = tmp->v.x + 0.125;
+   const GLfloat y = tmp->v.y - 0.125;
 
-   *(float *)&wv[0] = tmp->v.x - sz;
-   *(float *)&wv[1] = tmp->v.y - sz;
+   *(float *)&wv[0] = x - sz;
+   *(float *)&wv[1] = y - sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->v.x + sz;
-   *(float *)&wv[1] = tmp->v.y - sz;
+   *(float *)&wv[0] = x + sz;
+   *(float *)&wv[1] = y - sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->v.x + sz;
-   *(float *)&wv[1] = tmp->v.y + sz;
+   *(float *)&wv[0] = x + sz;
+   *(float *)&wv[1] = y + sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->v.x + sz;
-   *(float *)&wv[1] = tmp->v.y + sz;
+   *(float *)&wv[0] = x + sz;
+   *(float *)&wv[1] = y + sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->v.x - sz;
-   *(float *)&wv[1] = tmp->v.y + sz;
+   *(float *)&wv[0] = x - sz;
+   *(float *)&wv[1] = y + sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->v.x - sz;
-   *(float *)&wv[1] = tmp->v.y - sz;
+   *(float *)&wv[0] = x - sz;
+   *(float *)&wv[1] = y - sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
 }
@@ -134,55 +135,79 @@ static __inline void mga_draw_line( mgaContextPtr mmesa,
 				  const mgaVertex *tmp1,
 				  float width )
 {
-   GLuint vertsize = mmesa->vertsize;
+   const GLuint vertsize = mmesa->vertsize;
    GLuint *wv = mgaAllocVertexDwords( mmesa, 6 * vertsize );
-   float dx, dy, ix, iy;
-   int j;
+   GLuint j;
+   GLfloat x0 = tmp0->v.x;
+   GLfloat y0 = tmp0->v.y;
+   GLfloat x1 = tmp1->v.x;
+   GLfloat y1 = tmp1->v.y;
+   GLfloat dx, dy, ix, iy;
+   GLfloat hw;
+
+   hw = 0.5F * width;
+   if (hw > 0.1F && hw < 0.5F) {
+      hw = 0.5F;
+   }
 
+   /* adjust vertices depending on line direction */
    dx = tmp0->v.x - tmp1->v.x;
    dy = tmp0->v.y - tmp1->v.y;
-
-   ix = width * .5; iy = 0;
-  
-   if ((ix<.5) && (ix>0.1)) ix = .5; /* I want to see lines with width
-                                        0.5 also */
-  
    if (dx * dx > dy * dy) {
-      iy = ix; ix = 0;
+      /* X-major line */
+      ix = 0.0F;
+      iy = hw;
+      if (x1 < x0) {
+         x0 += 0.5F;
+         x1 += 0.5F;
+      }
+      y0 -= 0.5F;
+      y1 -= 0.5F;
+   }
+   else {
+      /* Y-major line */
+      ix = hw;
+      iy = 0.0F;
+      if (y1 > y0) {
+         y0 -= 0.5F;
+         y1 -= 0.5F;
+      }
+      x0 += 0.5F;
+      x1 += 0.5F;
    }
-  
-   *(float *)&wv[0] = tmp0->v.x - ix;
-   *(float *)&wv[1] = tmp0->v.y - iy;
+
+   *(float *)&wv[0] = x0 - ix;
+   *(float *)&wv[1] = y0 - iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp0->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp1->v.x + ix;
-   *(float *)&wv[1] = tmp1->v.y + iy;
+   *(float *)&wv[0] = x1 + ix;
+   *(float *)&wv[1] = y1 + iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp1->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp0->v.x + ix;
-   *(float *)&wv[1] = tmp0->v.y + iy;
+   *(float *)&wv[0] = x0 + ix;
+   *(float *)&wv[1] = y0 + iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp0->ui[j];
    wv += vertsize;
 	 
-   *(float *)&wv[0] = tmp0->v.x - ix;
-   *(float *)&wv[1] = tmp0->v.y - iy;
+   *(float *)&wv[0] = x0 - ix;
+   *(float *)&wv[1] = y0 - iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp0->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp1->v.x - ix;
-   *(float *)&wv[1] = tmp1->v.y - iy;
+   *(float *)&wv[0] = x1 - ix;
+   *(float *)&wv[1] = y1 - iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp1->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp1->v.x + ix;
-   *(float *)&wv[1] = tmp1->v.y + iy;
+   *(float *)&wv[0] = x1 + ix;
+   *(float *)&wv[1] = y1 + iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp1->ui[j];
    wv += vertsize;
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h b/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
index f578871ce..a041d887d 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
@@ -16,6 +16,7 @@ static __inline void TAG(triangle)(GLcontext *ctx,
 
 #if (IND & (MGA_TWOSIDE_BIT | MGA_FLAT_BIT))
    GLuint c[3];
+   GLuint s[3];
 #endif
 
    v[0] = &mgaverts[e0];
@@ -26,6 +27,9 @@ static __inline void TAG(triangle)(GLcontext *ctx,
    c[0] = v[0]->ui[4];
    c[1] = v[1]->ui[4];
    c[2] = v[2]->ui[4];
+   s[0] = v[0]->ui[5];
+   s[1] = v[1]->ui[5];
+   s[2] = v[2]->ui[5];
 #endif
 
 
@@ -41,13 +45,19 @@ static __inline void TAG(triangle)(GLcontext *ctx,
       {
 	 GLuint  facing        = (cc > 0.0) ^ ctx->Polygon.FrontBit;
 	 GLubyte (*vbcolor)[4] = VB->Color[facing]->data;
+	 GLubyte (*vbspec)[4] = VB->Spec[facing];
 	 if (IND & MGA_FLAT_BIT) {
 	    MGA_COLOR((char *)&v[0]->ui[4], vbcolor[pv]);
 	    v[2]->ui[4] = v[1]->ui[4] = v[0]->ui[4];
+            MGA_COLOR3((char *)&v[0]->ui[5], vbspec[pv]);
+	    v[2]->ui[5] = v[1]->ui[5] = v[0]->ui[5];
 	 } else {
 	    MGA_COLOR((char *)&v[0]->ui[4], vbcolor[e0]);
 	    MGA_COLOR((char *)&v[1]->ui[4], vbcolor[e1]);
 	    MGA_COLOR((char *)&v[2]->ui[4], vbcolor[e2]);
+	    MGA_COLOR3((char *)&v[0]->ui[5], vbspec[e0]);
+	    MGA_COLOR3((char *)&v[1]->ui[5], vbspec[e1]);
+	    MGA_COLOR3((char *)&v[2]->ui[5], vbspec[e2]);
 	 }
       }
 #endif
@@ -79,9 +89,13 @@ static __inline void TAG(triangle)(GLcontext *ctx,
 #elif (IND & MGA_FLAT_BIT)
    {
       GLuint color = mgaverts[pv].ui[4];
+      GLuint spec = mgaverts[pv].ui[5];
       v[0]->ui[4] = color;
       v[1]->ui[4] = color;
       v[2]->ui[4] = color;
+      v[0]->ui[5] = spec;
+      v[1]->ui[5] = spec;
+      v[2]->ui[5] = spec;
    }
 #endif
 
@@ -97,6 +111,9 @@ static __inline void TAG(triangle)(GLcontext *ctx,
    v[0]->ui[4] = c[0];
    v[1]->ui[4] = c[1];
    v[2]->ui[4] = c[2];    
+   v[0]->ui[5] = s[0];
+   v[1]->ui[5] = s[1];
+   v[2]->ui[5] = s[2];    
 #endif
 
 }
@@ -120,25 +137,34 @@ static void TAG(line)( GLcontext *ctx, GLuint v0, GLuint v1, GLuint pv )
    float width = ctx->Line.Width;
    GLfloat z0, z1;
    GLuint c0, c1;
+   GLuint s0, s1;
    mgaVertex *vert0 = &mgaVB[v0];
    mgaVertex *vert1 = &mgaVB[v1];
 
-
    if (IND & MGA_TWOSIDE_BIT) {
       GLubyte (*vbcolor)[4] = ctx->VB->ColorPtr->data;
+      GLubyte (*vbspec)[4] = ctx->VB->Specular;
 
       if (IND & MGA_FLAT_BIT) {
 	 MGA_COLOR((char *)&vert0->v.color,vbcolor[pv]);
 	 *(int *)&vert1->v.color = *(int *)&vert0->v.color;
+         MGA_COLOR3((char *)&vert0->v.specular, vbspec[pv]);
+         *(int *)&vert1->v.specular = *(int *)&vert0->v.specular;
       } else {
 	 MGA_COLOR((char *)&vert0->v.color,vbcolor[v0]);
 	 MGA_COLOR((char *)&vert1->v.color,vbcolor[v1]);
+         MGA_COLOR3((char *)&vert0->v.specular, vbspec[v0]);
+         MGA_COLOR3((char *)&vert1->v.specular, vbspec[v1]);
       }
    } else if (IND & MGA_FLAT_BIT) {
       c0 = *(GLuint *) &(vert0->v.color);
       c1 = *(GLuint *) &(vert1->v.color);
       *(int *)&vert0->v.color = 
 	 *(int *)&vert1->v.color = *(int *)&mgaVB[pv].v.color;
+      s0 = *(GLuint *) &(vert0->v.specular);
+      s1 = *(GLuint *) &(vert1->v.specular);
+      *(int *)&vert0->v.specular = 
+         *(int *)&vert1->v.specular = *(int *)&mgaVB[pv].v.specular;
    }
 
    if (IND & MGA_OFFSET_BIT) {
@@ -159,6 +185,8 @@ static void TAG(line)( GLcontext *ctx, GLuint v0, GLuint v1, GLuint pv )
    if ((IND & MGA_FLAT_BIT) && !(IND & MGA_TWOSIDE_BIT)) {
       *(GLuint *) &(vert0->v.color) = c0;
       *(GLuint *) &(vert1->v.color) = c1;
+      *(GLuint *) &(vert0->v.specular) = s0;
+      *(GLuint *) &(vert1->v.specular) = s1;
    }
 }
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgavb.c b/xc/lib/GL/mesa/src/drv/mga/mgavb.c
index ae6316fc8..1dfc89d2b 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgavb.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgavb.c
@@ -35,34 +35,34 @@
 #include <stdio.h>
 #include <stdlib.h>
 
-#define TEX0 {					\
+#define TEX0 {				\
   v->v.tu0 = tc0[i][0];			\
   v->v.tv0 = tc0[i][1];			\
 }
 
-#define TEX1 {					\
+#define TEX1 {				\
   v->v.tu1 = tc1[i][0];			\
   v->v.tv1 = tc1[i][1];			\
 }
 
-#define SPC {					\
-  GLubyte *spec = &(VB->Spec[0][i][0]);		\
+#define SPC {				\
+  GLubyte *spec = &(VB->Spec[0][i][0]);	\
   v->v.specular.red = spec[0];		\
-  v->v.specular.green = spec[1];		\
+  v->v.specular.green = spec[1];	\
   v->v.specular.blue = spec[2];		\
 }
 
-#define FOG {					\
-  GLubyte *spec = &(VB->Spec[0][i][0]);		\
-  v->v.specular.alpha = spec[3];		\
+#define FOG {				\
+  GLubyte *spec = &(VB->Spec[0][i][0]);	\
+  v->v.specular.alpha = spec[3];	\
 }
 
 #define COL {					\
   GLubyte *col = &(VB->Color[0]->data[i][0]);	\
-  v->v.color.blue  = col[2];		\
-  v->v.color.green = col[1];		\
-  v->v.color.red   = col[0];		\
-  v->v.color.alpha = col[3];		\
+  v->v.color.blue  = col[2];			\
+  v->v.color.green = col[1];			\
+  v->v.color.red   = col[0];			\
+  v->v.color.alpha = col[3];			\
 }
 
 /* The v code we have doesn't seem to support projective texturing
@@ -70,14 +70,13 @@
  * second set of texcoords).  This may be a problem for the g400.  
  */
 #define TEX0_4						\
-  if (VB->TexCoordPtr[0]->size == 4)			\
-  {							\
+  if (VB->TexCoordPtr[0]->size == 4) {			\
      GLfloat (*tc)[4] = VB->TexCoordPtr[0]->data;	\
      v = &(MGA_DRIVER_DATA(VB)->verts[start]);		\
      mmesa->setupdone &= ~MGA_WIN_BIT;			\
-     for (i=start; i < end; i++, v++)	{		\
-        float oow = 1.0 / tc[i][3];			\
-	v->v.rhw *= tc[i][3];			\
+     for (i = start; i < end; i++, v++) {		\
+        GLfloat oow = 1.0 / tc[i][3];			\
+	v->v.rhw *= tc[i][3];				\
 	v->v.tu0 *= oow;				\
 	v->v.tv0 *= oow;				\
      }							\
@@ -138,7 +137,7 @@ static void name(struct vertex_buffer *VB, GLuint start, GLuint end)	\
 	    spec;							\
 	    fog;							\
 	 }								\
-	    col;							\
+         col;								\
       }									\
    tex0_4;								\
 }
diff --git a/xc/lib/GL/mesa/src/drv/radeon/Imakefile b/xc/lib/GL/mesa/src/drv/radeon/Imakefile
index d862565d0..2dc0b3083 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/radeon/Imakefile
@@ -59,6 +59,8 @@ MESA_INCLUDES = -I. -I.. -I../../include \
 		radeon_span.c \
 		radeon_state.c \
 		radeon_tex.c \
+		radeon_texmem.c \
+		radeon_texstate.c \
 		radeon_tris.c \
 		radeon_vb.c \
 		radeon_xmesa.c
@@ -74,6 +76,8 @@ MESA_INCLUDES = -I. -I.. -I../../include \
 		radeon_span.o \
 		radeon_state.o \
 		radeon_tex.o \
+		radeon_texmem.o \
+		radeon_texstate.o \
 		radeon_tris.o \
 		radeon_vb.o \
 		radeon_xmesa.o
@@ -152,6 +156,7 @@ MESA_INCLUDES = -I. -I.. -I../../include \
 		../../stages.c \
 		../../state.c \
 		../../stencil.c \
+		../../texformat.c \
 		../../teximage.c \
 		../../texobj.c \
 		../../texstate.c \
@@ -225,6 +230,7 @@ MESA_INCLUDES = -I. -I.. -I../../include \
 		../../stages.o \
 		../../state.o \
 		../../stencil.o \
+		../../texformat.o \
 		../../teximage.o \
 		../../texobj.o \
 		../../texstate.o \
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_context.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_context.c
index 9b2c5b2f1..172e0f62a 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_context.c
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_context.c
@@ -163,16 +163,8 @@ GLboolean radeonCreateContext( Display *dpy, GLvisual *glVisual,
       ctx->Const.MaxTextureSize = (1 << 10);
    }
 
-   /* FIXME: Support all available texture units... */
    ctx->Const.MaxTextureUnits = 2;
 
-#if ENABLE_PERF_BOXES
-   if (getenv("LIBGL_PERFORMANCE_BOXES"))
-      rmesa->boxes = 1;
-   else
-      rmesa->boxes = 0;
-#endif
-
    ctx->DriverCtx = (void *)rmesa;
 
    radeonDDInitExtensions( ctx );
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_dd.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_dd.c
index b86506a8d..4ef0ebd30 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_dd.c
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_dd.c
@@ -46,7 +46,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "X86/common_x86_asm.h"
 #endif
 
-#define RADEON_DATE	"20010105"
+#define RADEON_DATE	"20010305"
 
 
 /* Return the width and height of the current color buffer.
@@ -100,6 +100,11 @@ static const GLubyte *radeonDDGetString( GLcontext *ctx, GLenum name )
 	 strncat( buffer, " x86", 4 );
       }
 #endif
+#ifdef USE_MMX_ASM
+      if ( cpu_has_mmx ) {
+	 strncat( buffer, "/MMX", 4 );
+      }
+#endif
 #ifdef USE_3DNOW_ASM
       if ( cpu_has_3dnow ) {
 	 strncat( buffer, "/3DNow!", 7 );
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_ioctl.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_ioctl.c
index a02794e49..c2cd6a203 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_ioctl.c
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_ioctl.c
@@ -321,6 +321,7 @@ void radeonFireBlitLocked( radeonContextPtr rmesa, drmBufPtr buffer,
 			   GLint offset, GLint pitch, GLint format,
 			   GLint x, GLint y, GLint width, GLint height )
 {
+#if 0
    GLint ret;
 
    ret = drmRadeonTextureBlit( rmesa->driFd, buffer->idx,
@@ -332,6 +333,7 @@ void radeonFireBlitLocked( radeonContextPtr rmesa, drmBufPtr buffer,
       fprintf( stderr, "drmRadeonTextureBlit: return = %d\n", ret );
       exit( 1 );
    }
+#endif
 }
 
 
@@ -616,12 +618,6 @@ static GLbitfield radeonDDClear( GLcontext *ctx, GLbitfield mask,
    }
 #endif
 
-#if 0
-   if ( rmesa->dirty & ~RADEON_UPLOAD_CLIPRECTS ) {
-      radeonEmitHwStateLocked( rmesa );
-   }
-#endif
-
    for ( i = 0 ; i < rmesa->numClipRects ; ) {
       GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS, rmesa->numClipRects );
       XF86DRIClipRectPtr box = rmesa->pClipRects;
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_state.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_state.c
index d82013338..d80004dd6 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_state.c
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_state.c
@@ -511,7 +511,8 @@ static void radeonDDLightModelfv( GLcontext *ctx, GLenum pname,
 
       FLUSH_BATCH( rmesa );
 
-      if ( ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR ) {
+      if ( ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR &&
+           ctx->Light.Enabled && ctx->Texture.ReallyEnabled) {
 	 p |=  RADEON_SPECULAR_ENABLE;
       } else {
 	 p &= ~RADEON_SPECULAR_ENABLE;
@@ -805,6 +806,22 @@ static void radeonDDEnable( GLcontext *ctx, GLenum cap, GLboolean state )
       }
       break;
 
+   case GL_LIGHTING:
+      {
+         GLuint p = rmesa->setup.pp_cntl;
+         if ( ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR &&
+              ctx->Light.Enabled && ctx->Texture.ReallyEnabled) {
+            p |=  RADEON_SPECULAR_ENABLE;
+         } else {
+            p &= ~RADEON_SPECULAR_ENABLE;
+         }
+         if ( rmesa->setup.pp_cntl != p ) {
+            rmesa->setup.pp_cntl = p;
+            rmesa->dirty |= RADEON_UPLOAD_CONTEXT;
+         }
+         break;
+      }
+
    case GL_SCISSOR_TEST:
       FLUSH_BATCH( rmesa );
       rmesa->scissor = state;
@@ -919,29 +936,29 @@ void radeonEmitHwStateLocked( radeonContextPtr rmesa )
    if ( (rmesa->dirty & RADEON_UPLOAD_TEX0) && t0 ) {
       radeon_texture_regs_t *tex = &sarea->TexState[0];
 
-      tex->pp_txfilter = t0->setup.pp_txfilter | rmesa->lod_bias[0] << 8;
-      tex->pp_txformat = t0->setup.pp_txformat | RADEON_TXF_ST_ROUTE_STQ0;
-      tex->pp_txoffset = t0->setup.pp_txoffset;
+      tex->pp_txfilter = t0->pp_txfilter | rmesa->lod_bias[0] << 8;
+      tex->pp_txformat = t0->pp_txformat | RADEON_TXFORMAT_ST_ROUTE_STQ0;
+      tex->pp_txoffset = t0->pp_txoffset;
 
       tex->pp_txcblend = rmesa->color_combine[0];
       tex->pp_txablend = rmesa->alpha_combine[0];
       tex->pp_tfactor = rmesa->env_color[0];
 
-      tex->pp_border_color = t0->setup.pp_border_color;
+      tex->pp_border_color = t0->pp_border_color;
    }
 
    if ( (rmesa->dirty & RADEON_UPLOAD_TEX1) && t1 ) {
       radeon_texture_regs_t *tex = &sarea->TexState[1];
 
-      tex->pp_txfilter = t1->setup.pp_txfilter | rmesa->lod_bias[1] << 8;
-      tex->pp_txformat = t1->setup.pp_txformat | RADEON_TXF_ST_ROUTE_STQ1;
-      tex->pp_txoffset = t1->setup.pp_txoffset;
+      tex->pp_txfilter = t1->pp_txfilter | rmesa->lod_bias[1] << 8;
+      tex->pp_txformat = t1->pp_txformat | RADEON_TXFORMAT_ST_ROUTE_STQ1;
+      tex->pp_txoffset = t1->pp_txoffset;
 
       tex->pp_txcblend = rmesa->color_combine[1];
       tex->pp_txablend = rmesa->alpha_combine[1];
       tex->pp_tfactor = rmesa->env_color[1];
 
-      tex->pp_border_color = t1->setup.pp_border_color;
+      tex->pp_border_color = t1->pp_border_color;
    }
 
    if ( rmesa->dirty & RADEON_UPLOAD_TEX2 ) {
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.c
index dd9c4557b..a77a7f154 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.c
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.c
@@ -44,1963 +44,701 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "simple_list.h"
 #include "enums.h"
 #include "mem.h"
+#include "texutil.h"
 
 
 static void radeonSetTexWrap( radeonTexObjPtr t, GLenum swrap, GLenum twrap )
 {
-   t->setup.pp_txfilter &= ~(RADEON_CLAMP_S_MASK | RADEON_CLAMP_T_MASK);
+   t->pp_txfilter &= ~(RADEON_CLAMP_S_MASK | RADEON_CLAMP_T_MASK);
 
    switch ( swrap ) {
    case GL_REPEAT:
-      t->setup.pp_txfilter |= RADEON_CLAMP_S_WRAP;
+      t->pp_txfilter |= RADEON_CLAMP_S_WRAP;
       break;
    case GL_CLAMP:
-      t->setup.pp_txfilter |= RADEON_CLAMP_S_CLAMP_LAST;
+      t->pp_txfilter |= RADEON_CLAMP_S_CLAMP_LAST;
       break;
    case GL_CLAMP_TO_EDGE:
-      t->setup.pp_txfilter |= RADEON_CLAMP_S_CLAMP_LAST;
+      t->pp_txfilter |= RADEON_CLAMP_S_CLAMP_LAST;
       break;
    }
 
    switch ( twrap ) {
    case GL_REPEAT:
-      t->setup.pp_txfilter |= RADEON_CLAMP_T_WRAP;
+      t->pp_txfilter |= RADEON_CLAMP_T_WRAP;
       break;
    case GL_CLAMP:
-      t->setup.pp_txfilter |= RADEON_CLAMP_T_CLAMP_LAST;
+      t->pp_txfilter |= RADEON_CLAMP_T_CLAMP_LAST;
       break;
    case GL_CLAMP_TO_EDGE:
-      t->setup.pp_txfilter |= RADEON_CLAMP_T_CLAMP_LAST;
+      t->pp_txfilter |= RADEON_CLAMP_T_CLAMP_LAST;
       break;
    }
 }
 
 static void radeonSetTexFilter( radeonTexObjPtr t, GLenum minf, GLenum magf )
 {
-   t->setup.pp_txfilter &= ~(RADEON_MIN_FILTER_MASK | RADEON_MAG_FILTER_MASK);
+   t->pp_txfilter &= ~(RADEON_MIN_FILTER_MASK | RADEON_MAG_FILTER_MASK);
 
    switch ( minf ) {
    case GL_NEAREST:
-      t->setup.pp_txfilter |= RADEON_MIN_FILTER_NEAREST;
+      t->pp_txfilter |= RADEON_MIN_FILTER_NEAREST;
       break;
    case GL_LINEAR:
-      t->setup.pp_txfilter |= RADEON_MIN_FILTER_LINEAR;
+      t->pp_txfilter |= RADEON_MIN_FILTER_LINEAR;
       break;
    case GL_NEAREST_MIPMAP_NEAREST:
-      t->setup.pp_txfilter |= RADEON_MIN_FILTER_NEAREST_MIP_NEAREST;
+      t->pp_txfilter |= RADEON_MIN_FILTER_NEAREST_MIP_NEAREST;
       break;
    case GL_LINEAR_MIPMAP_NEAREST:
-      t->setup.pp_txfilter |= RADEON_MIN_FILTER_LINEAR_MIP_NEAREST;
+      t->pp_txfilter |= RADEON_MIN_FILTER_LINEAR_MIP_NEAREST;
       break;
    case GL_NEAREST_MIPMAP_LINEAR:
-      t->setup.pp_txfilter |= RADEON_MIN_FILTER_NEAREST_MIP_LINEAR;
+      t->pp_txfilter |= RADEON_MIN_FILTER_NEAREST_MIP_LINEAR;
       break;
    case GL_LINEAR_MIPMAP_LINEAR:
-      t->setup.pp_txfilter |= RADEON_MIN_FILTER_LINEAR_MIP_LINEAR;
+      t->pp_txfilter |= RADEON_MIN_FILTER_LINEAR_MIP_LINEAR;
       break;
    }
 
    switch ( magf ) {
    case GL_NEAREST:
-      t->setup.pp_txfilter |= RADEON_MAG_FILTER_NEAREST;
+      t->pp_txfilter |= RADEON_MAG_FILTER_NEAREST;
       break;
    case GL_LINEAR:
-      t->setup.pp_txfilter |= RADEON_MAG_FILTER_LINEAR;
+      t->pp_txfilter |= RADEON_MAG_FILTER_LINEAR;
       break;
    }
 }
 
 static void radeonSetTexBorderColor( radeonTexObjPtr t, GLubyte c[4] )
 {
-   t->setup.pp_border_color = radeonPackColor( 4, c[0], c[1], c[2], c[3] );
+   t->pp_border_color = radeonPackColor( 4, c[0], c[1], c[2], c[3] );
 }
 
-
-/* Allocate and initialize hardware state associated with texture `t'.
- */
-static radeonTexObjPtr radeonCreateTexObj( radeonContextPtr rmesa,
-					   struct gl_texture_object *tObj )
+static radeonTexObjPtr radeonAllocTexObj( struct gl_texture_object *tObj )
 {
    radeonTexObjPtr t;
-   struct gl_texture_image *image;
-   GLint log2Width, log2Height, log2Size, log2MinSize;
-   GLint totalSize;
-   GLint texelsPerDword = 0, blitWidth = 0, blitPitch = 0;
-   GLint x, y, width, height;
-   GLint i;
-   GLuint txformat, txalpha;
-
-   image = tObj->Image[0];
-   if ( !image )
-      return NULL;
-
-   t = (radeonTexObjPtr) CALLOC( sizeof(*t) );
-   if ( !t )
-      return NULL;
-
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API )
-      fprintf( stderr, "%s( %p )\n", __FUNCTION__, tObj );
-
-   switch ( image->Format ) {
-   case GL_RGBA:
-      if ( image->IntFormat != GL_RGBA4 &&
-	   ( image->IntFormat == GL_RGBA8 ||
-	     rmesa->radeonScreen->cpp == 4 ) ) {
-	 t->texelBytes = 4;
-	 txformat = RADEON_TXF_32BPP_ARGB8888;
-      } else {
-	 t->texelBytes = 2;
-	 txformat = RADEON_TXF_16BPP_ARGB4444;
-      }
-      txalpha = RADEON_TXF_ALPHA_IN_MAP;
-      break;
-
-   case GL_RGB:
-      if ( image->IntFormat != GL_RGB5 &&
-	   ( image->IntFormat == GL_RGB8 ||
-	     rmesa->radeonScreen->cpp == 4 ) ) {
-	 t->texelBytes = 4;
-	 txformat = RADEON_TXF_32BPP_ARGB8888;
-      } else {
-	 t->texelBytes = 2;
-	 txformat = RADEON_TXF_16BPP_RGB565;
-      }
-      txalpha = 0;
-      break;
-
-   case GL_ALPHA:
-   case GL_LUMINANCE_ALPHA:
-      t->texelBytes = 2;
-      txformat = RADEON_TXF_16BPP_AI88;
-      txalpha = RADEON_TXF_ALPHA_IN_MAP;
-      break;
-
-   case GL_LUMINANCE:
-      t->texelBytes = 2;
-      txformat = RADEON_TXF_16BPP_AI88;
-      txalpha = 0;
-      break;
-
-   case GL_INTENSITY:
-      t->texelBytes = 1;
-      txformat = RADEON_TXF_8BPP_I;
-      txalpha = 0;
-      break;
-
-   case GL_COLOR_INDEX:
-   default:
-      fprintf( stderr, "%s: bad image->Format\n", __FUNCTION__ );
-      FREE( t );
-      return NULL;
-   }
-
-   /* Calculate dimensions in log domain.
-    */
-   for ( i = 1, log2Height = 0 ; i < image->Height ; i *= 2 ) {
-      log2Height++;
-   }
-   for ( i = 1, log2Width = 0 ;  i < image->Width ;  i *= 2 ) {
-      log2Width++;
-   }
-   if ( image->Width > image->Height ) {
-      log2Size = log2Width;
-   } else {
-      log2Size = log2Height;
-   }
-
-   t->dirty_images = 0;
-
-   /* The Radeon has a 64-byte minimum pitch for all blits.  We
-    * calculate the equivalent number of texels to simplify the
-    * calculation of the texture image area.
-    */
-   switch ( t->texelBytes ) {
-   case 1:
-      texelsPerDword = 4;
-      blitPitch = 64;
-      break;
-   case 2:
-      texelsPerDword = 2;
-      blitPitch = 32;
-      break;
-   case 4:
-      texelsPerDword = 1;
-      blitPitch = 16;
-      break;
-   }
-
-   /* Select the larger of the two widths for our global texture image
-    * coordinate space.  As the Radeon has very strict offset rules, we
-    * can't upload mipmaps directly and have to reference their location
-    * from the aligned start of the whole image.
-    */
-   blitWidth = MAX2( image->Width, blitPitch );
-
-   /* Calculate mipmap offsets and dimensions.
-    */
-   totalSize = 0;
-   x = 0;
-   y = 0;
-
-   for ( i = 0 ; i <= log2Size ; i++ ) {
-      GLuint size;
 
-      image = tObj->Image[i];
-      if ( !image )
-	 break;
-
-      width = image->Width;
-      height = image->Height;
-
-      /* Texture images have a minimum pitch of 32 bytes (half of the
-       * 64-byte minimum pitch for blits).  For images that have a
-       * width smaller than this, we must pad each texture image
-       * scanline out to this amount.
-       */
-      if ( width < blitPitch / 2 ) {
-	 width = blitPitch / 2;
-      }
-
-      size = width * height * t->texelBytes;
-      totalSize += size;
-
-      t->dirty_images |= (1 << i);
-
-      while ( width < blitWidth && height > 1 ) {
-	 width *= 2;
-	 height /= 2;
-      }
-
-      t->image[i].x = x;
-      t->image[i].y = y;
-
-      t->image[i].width  = width;
-      t->image[i].height = height;
-
-      t->image[i].dwords = size / sizeof(CARD32);
-
-      /* While blits must have a pitch of at least 64 bytes, mipmaps
-       * must be aligned on a 32-byte boundary (just like each texture
-       * image scanline).
-       */
-      if ( width >= blitWidth ) {
-	 y += height;
-      } else {
-	 x += width;
-	 if ( x >= blitWidth ) {
-	    x = 0;
-	    y++;
-	 }
-      }
+   t = CALLOC_STRUCT( radeon_tex_obj );
 
-      if ( 0 )
-	 fprintf( stderr, "level=%d p=%d   %dx%d -> %dx%d at (%d,%d)  %d dwords\n",
-		  i, blitWidth, image->Width, image->Height,
-		  t->image[i].width, t->image[i].height,
-		  t->image[i].x, t->image[i].y,
-		  t->image[i].dwords );
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p )\n", __FUNCTION__, t );
    }
 
-   log2MinSize = log2Size - i + 1;
-
-   /* Align the total size of texture memory block.
+   /* Initialize non-image-dependent parts of the state:
     */
-   totalSize = (totalSize + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
-
-   t->totalSize = totalSize;
-
-   t->bound = 0;
-   t->heap = 0;
    t->tObj = tObj;
+   t->dirty_images = ~0;
+   t->pp_txfilter = RADEON_BORDER_MODE_OGL;
+   t->pp_txformat = (RADEON_TXFORMAT_ENDIAN_NO_SWAP |
+		     RADEON_TXFORMAT_PERSPECTIVE_ENABLE);
 
-   t->memBlock = NULL;
-   t->bufAddr = 0;
-
-   /* Hardware state:
-    */
-   t->setup.pp_txfilter = ((log2Size << RADEON_MAX_MIP_LEVEL_SHIFT) |
-			   RADEON_BORDER_MODE_OGL);
-
-   t->setup.pp_txformat = (txformat | txalpha |
-			   (log2Width << RADEON_TXF_WIDTH_SHIFT) |
-			   (log2Height << RADEON_TXF_HEIGHT_SHIFT) |
-			   RADEON_TXF_ENDIAN_NO_SWAP |
-			   RADEON_TXF_PERSPECTIVE_ENABLE);
-
-   t->setup.pp_txoffset		= 0x00000000;
-   t->setup.pp_txcblend		= 0x00000000;
-   t->setup.pp_txablend		= 0x00000000;
-   t->setup.pp_tfactor		= 0x00000000;
-   t->setup.pp_border_color	= 0x00000000;
+   make_empty_list( t );
 
    radeonSetTexWrap( t, tObj->WrapS, tObj->WrapT );
    radeonSetTexFilter( t, tObj->MinFilter, tObj->MagFilter );
    radeonSetTexBorderColor( t, tObj->BorderColor );
 
-   tObj->DriverData = t;
-
-   make_empty_list( t );
-
    return t;
 }
 
-/* Destroy hardware state associated with texture `t'.
- */
-void radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t )
-{
-#if ENABLE_PERF_BOXES
-   /* Bump the performace counter */
-   rmesa->c_textureSwaps++;
-#endif
-   if ( !t ) return;
-
-   if ( t->memBlock ) {
-      mmFreeMem( t->memBlock );
-      t->memBlock = NULL;
-   }
-
-   if ( t->tObj )
-      t->tObj->DriverData = NULL;
-   if ( t->bound )
-      rmesa->CurrentTexObj[t->bound-1] = NULL;
-
-   remove_from_list( t );
-   FREE( t );
-}
-
-/* Keep track of swapped out texture objects.
- */
-static void radeonSwapOutTexObj( radeonContextPtr rmesa, radeonTexObjPtr t )
-{
-#if ENABLE_PERF_BOXES
-   /* Bump the performace counter */
-   rmesa->c_textureSwaps++;
-#endif
-   if ( t->memBlock ) {
-      mmFreeMem( t->memBlock );
-      t->memBlock = NULL;
-   }
-
-   t->dirty_images = ~0;
-   move_to_tail( &rmesa->SwappedOut, t );
-}
-
-/* Print out debugging information about texture LRU.
- */
-void radeonPrintLocalLRU( radeonContextPtr rmesa, int heap )
-{
-   radeonTexObjPtr t;
-   int sz = 1 << (rmesa->radeonScreen->logTexGranularity[heap]);
-
-   fprintf( stderr, "\nLocal LRU, heap %d:\n", heap );
-
-   foreach ( t, &rmesa->TexObjList[heap] ) {
-      if (!t->tObj) {
-	 fprintf( stderr, "Placeholder %d at 0x%x sz 0x%x\n",
-		  t->memBlock->ofs / sz,
-		  t->memBlock->ofs,
-		  t->memBlock->size );
-      } else {
-	 fprintf( stderr, "Texture (bound %d) at 0x%x sz 0x%x\n",
-		  t->bound,
-		  t->memBlock->ofs,
-		  t->memBlock->size );
-      }
-   }
-
-   fprintf( stderr, "\n" );
-}
-
-void radeonPrintGlobalLRU( radeonContextPtr rmesa, int heap )
-{
-   radeon_tex_region_t *list = rmesa->sarea->texList[heap];
-   int i, j;
-
-   fprintf( stderr, "\nGlobal LRU, heap %d list %p:\n", heap, list );
-
-   for ( i = 0, j = RADEON_NR_TEX_REGIONS ; i < RADEON_NR_TEX_REGIONS ; i++ ) {
-      fprintf( stderr, "list[%d] age %d next %d prev %d\n",
-	       j, list[j].age, list[j].next, list[j].prev );
-      j = list[j].next;
-      if ( j == RADEON_NR_TEX_REGIONS ) break;
-   }
-
-   if ( j != RADEON_NR_TEX_REGIONS ) {
-      fprintf( stderr, "Loop detected in global LRU\n" );
-      for ( i = 0 ; i < RADEON_NR_TEX_REGIONS ; i++ ) {
-	 fprintf( stderr, "list[%d] age %d next %d prev %d\n",
-		  i, list[i].age, list[i].next, list[i].prev );
-      }
-   }
-
-   fprintf( stderr, "\n" );
-}
-
-/* Reset the global texture LRU.
- */
-static void radeonResetGlobalLRU( radeonContextPtr rmesa, int heap )
-{
-   radeon_tex_region_t *list = rmesa->sarea->texList[heap];
-   int sz = 1 << rmesa->radeonScreen->logTexGranularity[heap];
-   int i;
-
-   /*
-    * (Re)initialize the global circular LRU list.  The last element in
-    * the array (RADEON_NR_TEX_REGIONS) is the sentinal.  Keeping it at
-    * the end of the array allows it to be addressed rationally when
-    * looking up objects at a particular location in texture memory.
-    */
-   for ( i = 0 ; (i+1) * sz <= rmesa->radeonScreen->texSize[heap] ; i++ ) {
-      list[i].prev = i-1;
-      list[i].next = i+1;
-      list[i].age = 0;
-   }
-
-   i--;
-   list[0].prev = RADEON_NR_TEX_REGIONS;
-   list[i].prev = i-1;
-   list[i].next = RADEON_NR_TEX_REGIONS;
-   list[RADEON_NR_TEX_REGIONS].prev = i;
-   list[RADEON_NR_TEX_REGIONS].next = 0;
-   rmesa->sarea->texAge[heap] = 0;
-}
-
-/* Update the local and glock texture LRUs.
- */
-static void radeonUpdateTexLRU(radeonContextPtr rmesa, radeonTexObjPtr t )
-{
-   int heap = t->heap;
-   radeon_tex_region_t *list = rmesa->sarea->texList[heap];
-   int sz = rmesa->radeonScreen->logTexGranularity[heap];
-   int start = t->memBlock->ofs >> sz;
-   int end = (t->memBlock->ofs + t->memBlock->size-1) >> sz;
-   int i;
-
-   rmesa->lastTexAge[heap] = ++rmesa->sarea->texAge[heap];
-
-   if ( !t->memBlock ) {
-      fprintf( stderr, "no memblock\n\n" );
-      return;
-   }
-
-   /* Update our local LRU */
-   move_to_head( &rmesa->TexObjList[heap], t );
-
-   /* Update the global LRU */
-   for ( i = start ; i <= end ; i++ ) {
-      list[i].in_use = 1;
-      list[i].age = rmesa->lastTexAge[heap];
-
-      /* remove_from_list(i) */
-      list[(CARD32)list[i].next].prev = list[i].prev;
-      list[(CARD32)list[i].prev].next = list[i].next;
-
-      /* insert_at_head(list, i) */
-      list[i].prev = RADEON_NR_TEX_REGIONS;
-      list[i].next = list[RADEON_NR_TEX_REGIONS].next;
-      list[(CARD32)list[RADEON_NR_TEX_REGIONS].next].prev = i;
-      list[RADEON_NR_TEX_REGIONS].next = i;
-   }
-
-   if ( 0 ) {
-      radeonPrintGlobalLRU( rmesa, t->heap );
-      radeonPrintLocalLRU( rmesa, t->heap );
-   }
-}
-
-/* Update our notion of what textures have been changed since we last
- * held the lock.  This pertains to both our local textures and the
- * textures belonging to other clients.  Keep track of other client's
- * textures by pushing a placeholder texture onto the LRU list -- these
- * are denoted by (tObj == NULL).
- */
-static void radeonTexturesGone( radeonContextPtr rmesa, int heap,
-				int offset, int size, int in_use )
-{
-   radeonTexObjPtr t, tmp;
-
-   foreach_s ( t, tmp, &rmesa->TexObjList[heap] ) {
-      if ( t->memBlock->ofs >= offset + size ||
-	   t->memBlock->ofs + t->memBlock->size <= offset )
-	 continue;
-
-      /* It overlaps - kick it out.  Need to hold onto the currently
-       * bound objects, however.
-       */
-      if ( t->bound ) {
-	 radeonSwapOutTexObj( rmesa, t );
-      } else {
-	 radeonDestroyTexObj( rmesa, t );
-      }
-   }
-
-   if ( in_use ) {
-      t = (radeonTexObjPtr) CALLOC( sizeof(*t) );
-      if ( !t ) return;
 
-      t->memBlock = mmAllocMem( rmesa->texHeap[heap], size, 0, offset );
-      if ( !t->memBlock ) {
-	 fprintf( stderr, "Couldn't alloc placeholder sz %x ofs %x\n",
-		  (int)size, (int)offset );
-	 mmDumpMemInfo( rmesa->texHeap[heap] );
-	 return;
-      }
-      insert_at_head( &rmesa->TexObjList[heap], t );
-   }
-}
-
-/* Update our client's shared texture state.  If another client has
- * modified a region in which we have textures, then we need to figure
- * out which of our textures has been removed, and update our global
- * LRU.
- */
-void radeonAgeTextures( radeonContextPtr rmesa, int heap )
+static GLint radeonChooseTexFormat( radeonContextPtr rmesa,
+				    struct gl_texture_image *texImage,
+				    GLenum format, GLenum type )
 {
-   RADEONSAREAPrivPtr sarea = rmesa->sarea;
-
-   if ( sarea->texAge[heap] != rmesa->lastTexAge[heap] ) {
-      int sz = 1 << rmesa->radeonScreen->logTexGranularity[heap];
-      int nr = 0;
-      int idx;
-
-      for ( idx = sarea->texList[heap][RADEON_NR_TEX_REGIONS].prev ;
-	    idx != RADEON_NR_TEX_REGIONS && nr < RADEON_NR_TEX_REGIONS ;
-	    idx = sarea->texList[heap][idx].prev, nr++ )
-      {
-	 /* If switching texturing schemes, then the SAREA might not
-	  * have been properly cleared, so we need to reset the
-	  * global texture LRU.
-	  */
-	 if ( idx * sz > rmesa->radeonScreen->texSize[heap] ) {
-	    nr = RADEON_NR_TEX_REGIONS;
-	    break;
-	 }
-
-	 if ( sarea->texList[heap][idx].age > rmesa->lastTexAge[heap] ) {
-	    radeonTexturesGone( rmesa, heap, idx * sz, sz,
-				sarea->texList[heap][idx].in_use );
-	 }
-      }
-
-      if ( nr == RADEON_NR_TEX_REGIONS ) {
-	 radeonTexturesGone( rmesa, heap, 0,
-			     rmesa->radeonScreen->texSize[heap], 0 );
-	 radeonResetGlobalLRU( rmesa, heap );
-      }
-
-      rmesa->dirty |= (RADEON_UPLOAD_CONTEXT |
-		       RADEON_UPLOAD_TEX0IMAGES |
-		       RADEON_UPLOAD_TEX1IMAGES);
-      rmesa->lastTexAge[heap] = sarea->texAge[heap];
-   }
-}
+   const GLboolean do32bpt = ( rmesa->radeonScreen->cpp == 4 );
+   const struct gl_texture_format *texFormat;
+   GLint ret;
 
+   if ( 0 )
+      fprintf( stderr, "internal=%s format=%s type=%s\n",
+	       texImage->IntFormat == 3 ? "GL_RGB (3)" :
+	       texImage->IntFormat == 4 ? "GL_RGBA (4)" :
+	       gl_lookup_enum_by_nr( texImage->IntFormat ),
+	       gl_lookup_enum_by_nr( format ),
+	       gl_lookup_enum_by_nr( type ) );
 
-/* ================================================================
- * Texture image conversions
- */
-
-/* Convert a block of Mesa-formatted texture to an 8bpp hardware format.
- */
-static void radeonConvertTexture8bpp( CARD32 *dst,
-				      struct gl_texture_image *image,
-				      int x, int y, int width, int height,
-				      int pitch )
-{
-   CARD8 *src;
-   int i, j;
-
-   if ( width < 4 ) {
-      width = 4;
-   }
-
-#define ALIGN_DST							\
+#define SET_FORMAT( r, gl )						\
    do {									\
-      if ( width < 32 ) {						\
-	 dst += ((32 - width) / 4);					\
-      }									\
+      ret = (r);							\
+      texFormat = &(gl);						\
    } while (0)
 
-   switch ( image->Format ) {
-   case GL_INTENSITY:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x);
-	 for ( j = width >> 2 ; j ; j-- ) {
-	    *dst++ = src[0] | (src[1] << 8) | (src[2] << 16) | (src[3] << 24);
-	    src += 4;
-	 }
-	 ALIGN_DST;
-      }
-      break;
-
-   case GL_COLOR_INDEX:
-   default:
-      fprintf( stderr, "%s: unsupported format 0x%x\n",
-	       __FUNCTION__, image->Format );
-      break;
-   }
-}
-#undef ALIGN_DST
-
-/* Convert a block of Mesa-formatted texture to a 16bpp hardware format.
- */
-static void radeonConvertTexture16bpp( CARD32 *dst,
-				       struct gl_texture_image *image,
-				       int x, int y, int width, int height,
-				       int pitch )
-{
-   CARD8 *src;
-   int i, j;
-
-   if ( width < 2 ) {
-      width = 2;
-   }
-
-#define ALIGN_DST							\
+#define SET_FORMAT_32BPT( r32, gl32, r16, gl16 )			\
    do {									\
-      if ( width < 16 ) {						\
-	 dst += ((16 - width) / 2);					\
+      if ( do32bpt ) {							\
+	 ret = (r32);							\
+	 texFormat = &(gl32);						\
+      } else {								\
+	 ret = (r16);							\
+	 texFormat = &(gl16);						\
       }									\
    } while (0)
 
-   switch ( image->Format ) {
+   switch ( texImage->IntFormat ) {
+      /* GH: Bias towards GL_RGB, GL_RGBA texture formats.  This has
+       * got to be better than sticking them way down the end of this
+       * huge list.
+       */
    case GL_RGBA:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 4;
-	 for ( j = width >> 1 ; j ; j-- ) {
-	    *dst++ = ((RADEONPACKCOLOR4444( src[0], src[1], src[2], src[3] )) |
-		      (RADEONPACKCOLOR4444( src[4], src[5], src[6], src[7] ) << 16));
-	    src += 8;
+   case 4:
+      if ( format == GL_BGRA ) {
+	 if ( type == GL_UNSIGNED_INT_8_8_8_8_REV ) {
+	    SET_FORMAT( RADEON_TXFORMAT_ARGB8888, _mesa_texformat_argb8888 );
+	    break;
+	 } else if ( type == GL_UNSIGNED_SHORT_4_4_4_4_REV ) {
+	    SET_FORMAT( RADEON_TXFORMAT_ARGB4444, _mesa_texformat_argb4444 );
+	    break;
+	 } else if ( type == GL_UNSIGNED_SHORT_1_5_5_5_REV ) {
+	    SET_FORMAT( RADEON_TXFORMAT_ARGB1555, _mesa_texformat_argb1555 );
+	    break;
 	 }
-	 ALIGN_DST;
       }
+      SET_FORMAT_32BPT( RADEON_TXFORMAT_RGBA8888, _mesa_texformat_rgba8888,
+			RADEON_TXFORMAT_ARGB4444, _mesa_texformat_argb4444 );
       break;
 
    case GL_RGB:
-   {
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 3;
-	 for ( j = width >> 1 ; j ; j-- ) {
-	    *dst++ = ((RADEONPACKCOLOR565( src[0], src[1], src[2] )) |
-		      (RADEONPACKCOLOR565( src[3], src[4], src[5] ) << 16));
-	    src += 6;
-	 }
-	 ALIGN_DST;
+   case 3:
+      if ( format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5 ) {
+	 SET_FORMAT( RADEON_TXFORMAT_RGB565, _mesa_texformat_rgb565 );
+	 break;
       }
-   }
+      SET_FORMAT_32BPT( RADEON_TXFORMAT_RGBA8888, _mesa_texformat_rgba8888,
+			RADEON_TXFORMAT_RGB565,   _mesa_texformat_rgb565 );
       break;
 
-   case GL_ALPHA:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x);
-	 for ( j = width >> 1 ; j ; j-- ) {
-	    *dst++ = ((RADEONPACKCOLOR88( 0x00, src[0] )) |
-		      (RADEONPACKCOLOR88( 0x00, src[1] ) << 16));
-	    src += 2;
-	 }
-	 ALIGN_DST;
-      }
+      /* GH: Okay, keep checking as normal.  Still test for GL_RGB,
+       * GL_RGBA formats first.
+       */
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      SET_FORMAT_32BPT( RADEON_TXFORMAT_RGBA8888, _mesa_texformat_rgba8888,
+			RADEON_TXFORMAT_ARGB4444, _mesa_texformat_argb4444 );
       break;
 
-   case GL_LUMINANCE:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x);
-	 for ( j = width >> 1 ; j ; j-- ) {
-	    *dst++ = ((RADEONPACKCOLOR88( src[0], 0xff )) |
-		      (RADEONPACKCOLOR88( src[1], 0xff ) << 16));
-	    src += 2;
-	 }
-	 ALIGN_DST;
-      }
+   case GL_RGBA4:
+   case GL_RGBA2:
+      SET_FORMAT( RADEON_TXFORMAT_ARGB4444, _mesa_texformat_argb4444 );
       break;
 
-   case GL_LUMINANCE_ALPHA:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 2;
-	 for ( j = width >> 1 ; j ; j-- ) {
-	    *dst++ = ((RADEONPACKCOLOR88( src[0], src[1] )) |
-		      (RADEONPACKCOLOR88( src[2], src[3] ) << 16));
-	    src += 4;
-	 }
-	 ALIGN_DST;
-      }
+   case GL_RGB5_A1:
+      SET_FORMAT( RADEON_TXFORMAT_ARGB1555, _mesa_texformat_argb1555 );
       break;
 
-   default:
-      fprintf( stderr, "%s: unsupported format 0x%x\n",
-	       __FUNCTION__, image->Format );
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+      SET_FORMAT_32BPT( RADEON_TXFORMAT_RGBA8888, _mesa_texformat_rgba8888,
+			RADEON_TXFORMAT_RGB565,   _mesa_texformat_rgb565 );
       break;
-   }
-}
-#undef ALIGN_DST
 
-/* Convert a block of Mesa-formatted texture to a 32bpp hardware format.
- */
-static void radeonConvertTexture32bpp( CARD32 *dst,
-				       struct gl_texture_image *image,
-				       int x, int y, int width, int height,
-				       int pitch )
-{
-   CARD8 *src;
-   int i, j;
+   case GL_RGB5:
+   case GL_RGB4:
+   case GL_R3_G3_B2:
+      SET_FORMAT( RADEON_TXFORMAT_RGB565, _mesa_texformat_rgb565 );
+      break;
 
-#define ALIGN_DST							\
-   do {									\
-      if ( width < 8 ) {						\
-	 dst += (8 - width);						\
-      }									\
-   } while (0)
+   case GL_ALPHA:
+   case GL_ALPHA4:
+   case GL_ALPHA8:
+   case GL_ALPHA12:
+   case GL_ALPHA16:
+      SET_FORMAT( RADEON_TXFORMAT_AI88, _mesa_texformat_al88 );
+      break;
 
-   switch ( image->Format ) {
-   case GL_RGBA:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 4;
-	 for ( j = width ; j ; j-- ) {
-	    *dst++ = RADEONPACKCOLOR8888( src[0], src[1], src[2], src[3] );
-	    src += 4;
-	 }
-	 ALIGN_DST;
-      }
+   case 1:
+   case GL_LUMINANCE:
+   case GL_LUMINANCE4:
+   case GL_LUMINANCE8:
+   case GL_LUMINANCE12:
+   case GL_LUMINANCE16:
+      SET_FORMAT( RADEON_TXFORMAT_AI88, _mesa_texformat_al88 );
       break;
 
-   case GL_RGB:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 3;
-	 for ( j = width ; j ; j-- ) {
-	    *dst++ = RADEONPACKCOLOR8888( src[0], src[1], src[2], 0xff );
-	    src += 3;
-	 }
-	 ALIGN_DST;
-      }
+   case 2:
+   case GL_LUMINANCE_ALPHA:
+   case GL_LUMINANCE4_ALPHA4:
+   case GL_LUMINANCE6_ALPHA2:
+   case GL_LUMINANCE8_ALPHA8:
+   case GL_LUMINANCE12_ALPHA4:
+   case GL_LUMINANCE12_ALPHA12:
+   case GL_LUMINANCE16_ALPHA16:
+      SET_FORMAT( RADEON_TXFORMAT_AI88, _mesa_texformat_al88 );
       break;
 
-   default:
-      fprintf( stderr, "%s: unsupported format 0x%x\n",
-	       __FUNCTION__, image->Format );
+   case GL_INTENSITY:
+   case GL_INTENSITY4:
+   case GL_INTENSITY8:
+   case GL_INTENSITY12:
+   case GL_INTENSITY16:
+      SET_FORMAT( RADEON_TXFORMAT_I8, _mesa_texformat_i8 );
       break;
+
+   case GL_COLOR_INDEX:
+   case GL_COLOR_INDEX1_EXT:
+   case GL_COLOR_INDEX2_EXT:
+   case GL_COLOR_INDEX4_EXT:
+   case GL_COLOR_INDEX8_EXT:
+   case GL_COLOR_INDEX12_EXT:
+   case GL_COLOR_INDEX16_EXT:
+   default:
+      fprintf( stderr, "bad texture format in radeonChooseTexFormat() %d",
+	       texImage->IntFormat );
+      return -1;
    }
+
+   texImage->TexFormat = texFormat;
+
+   return ret;
 }
 
-/* Upload the texture image associated with texture `t' at level `level'
- * at the address relative to `start'.
+
+/* ================================================================
+ * Texture image callbacks
  */
-static void radeonUploadSubImage( radeonContextPtr rmesa,
-				  radeonTexObjPtr t, GLint level,
-				  GLint x, GLint y, GLint width, GLint height )
-{
-   struct gl_texture_image *image;
-   GLint texelsPerDword = 0;
-   GLint imageX, imageY, imageWidth, imageHeight;
-   GLint blitX, blitY, blitWidth, blitHeight;
-   GLint imageRows, blitRows;
-   GLint remaining ;
-   GLint format, dwords;
-   CARD32 pitch, offset;
-   drmBufPtr buffer;
-   CARD32 *dst;
-
-   /* Ensure we have a valid texture to upload */
-   if ( ( level < 0 ) || ( level >= RADEON_MAX_TEXTURE_LEVELS ) )
-      return;
 
-   image = t->tObj->Image[level];
-   if ( !image )
-      return;
+static GLboolean
+radeonDDTexImage1D( GLcontext *ctx, GLenum target, GLint level,
+		    GLenum format, GLenum type, const GLvoid *pixels,
+		    const struct gl_pixelstore_attrib *packing,
+		    struct gl_texture_object *texObj,
+		    struct gl_texture_image *texImage,
+		    GLboolean *retainInternalCopy )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   const struct gl_texture_format *texFormat;
+   GLuint unit = ctx->Texture.CurrentUnit;
+   GLuint texSize;
+   GLint txformat;
+   GLubyte *data;
 
-   switch ( t->texelBytes ) {
-   case 1:
-      texelsPerDword = 4;
-      break;
-   case 2:
-      texelsPerDword = 2;
-      break;
-   case 4:
-      texelsPerDword = 1;
-      break;
-   }
+   if ( target != GL_TEXTURE_1D )
+      return GL_FALSE;
 
-   format = t->setup.pp_txformat & RADEON_TXF_FORMAT_MASK;
-
-   imageX = 0;
-   imageY = 0;
-   imageWidth = image->Width;
-   imageHeight = image->Height;
-
-   blitX = t->image[level].x;
-   blitY = t->image[level].y;
-   blitWidth = t->image[level].width;
-   blitHeight = t->image[level].height;
-
-   dwords = t->image[level].dwords;
-   offset = t->bufAddr;
-   pitch = (t->image[0].width * t->texelBytes) / 64;
-
-#if ENABLE_PERF_BOXES
-   /* Bump the performace counter */
-   rmesa->c_textureBytes += (dwords << 2);
-#endif
-
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_MSG ) {
-      fprintf( stderr, "   upload image: %d,%d at %d,%d\n",
-	       imageWidth, imageHeight, imageX, imageY );
-      fprintf( stderr, "   upload  blit: %d,%d at %d,%d\n",
-	       blitWidth, blitHeight, blitX, blitY );
-      fprintf( stderr, "       blit ofs: 0x%07x pitch: 0x%x dwords: %d "
-	       "level: %d format: %x\n",
-	       (GLuint)offset, (GLuint)pitch, dwords, level, format );
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p, %d )\n", __FUNCTION__, texObj, level );
    }
 
-   /* Subdivide the texture if required */
-   if ( dwords <= RADEON_BUFFER_MAX_DWORDS / 2 ) {
-      imageRows = imageHeight;
-      blitRows = blitHeight;
+   if ( t ) {
+      if ( t->bound ) FLUSH_BATCH( rmesa );
+      radeonSwapOutTexObj( rmesa, t );
    } else {
-      imageRows = (RADEON_BUFFER_MAX_DWORDS * texelsPerDword)/(2 * imageWidth);
-      blitRows = (RADEON_BUFFER_MAX_DWORDS * texelsPerDword) / (2 * blitWidth);
+      t = radeonAllocTexObj( texObj );
+      texObj->DriverData = t;
    }
 
-   for ( remaining = imageHeight ;
-	 remaining > 0 ;
-	 remaining -= imageRows, imageY += imageRows, blitY += blitRows )
-   {
-      if ( remaining >= imageRows ) {
-	 imageHeight = imageRows;
-	 blitHeight = blitRows;
-      } else {
-	 imageHeight = remaining;
-	 blitHeight = blitRows;
-      }
-      dwords = blitWidth * blitHeight / texelsPerDword;
+   txformat = radeonChooseTexFormat( rmesa, texImage, format, type );
+   if ( txformat < 0 )
+      return GL_FALSE;
 
-      if ( RADEON_DEBUG & DEBUG_VERBOSE_MSG ) {
-	 fprintf( stderr, "       blitting: %d,%d at %d,%d\n",
-		  imageWidth, imageHeight, imageX, imageY );
-	 fprintf( stderr, "                 %d,%d at %d,%d - %d dwords\n",
-		  blitWidth, blitHeight, blitX, blitY, dwords );
-      }
+   texFormat = texImage->TexFormat;
+   texSize = texImage->Width * texFormat->TexelBytes;
 
-      /* Grab the indirect buffer for the texture blit */
-      buffer = radeonGetBufferLocked( rmesa );
+   /* We really shouldn't have to keep the texture image, it should be
+    * hung from the main texImage structure.
+    */
+   if ( t->image[level].data ) {
+      FREE( t->image[level].data );
+      t->image[level].data = NULL;
+   }
 
-      dst = (CARD32 *)((char *)buffer->address + RADEON_HOSTDATA_BLIT_OFFSET);
+   data = (GLubyte *) MALLOC( texSize );
+   if ( !data )
+      return GL_FALSE;
 
-      /* Actually do the texture conversion */
-      switch ( t->texelBytes ) {
-      case 1:
-	 radeonConvertTexture8bpp( dst, image,
-				   imageX, imageY, imageWidth, imageHeight,
-				   imageWidth );
-	 break;
-      case 2:
-	 radeonConvertTexture16bpp( dst, image,
-				    imageX, imageY, imageWidth, imageHeight,
-				    imageWidth );
-	 break;
-      case 4:
-	 radeonConvertTexture32bpp( dst, image,
-				    imageX, imageY, imageWidth, imageHeight,
-				    imageWidth );
-	 break;
-      }
-
-      radeonFireBlitLocked( rmesa, buffer,
-			    offset, pitch, format,
-			    blitX, blitY, blitWidth, blitHeight );
+   if ( !_mesa_convert_texsubimage1d( texFormat->IntFormat,
+				      0, texImage->Width,
+				      format, type, packing,
+				      pixels, data ) ) {
+      /*fprintf( stderr, "   *** convert failed!\n" );*/
+      FREE( data );
+      return GL_FALSE;
    }
 
-   rmesa->new_state |= RADEON_NEW_CONTEXT;
-   rmesa->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_MASKS;
-}
+   t->image[level].data = data;
+   t->dirty_images |= (1 << level);
 
-/* Upload the texture images associated with texture `t'.  This might
- * require removing our own and/or other client's texture objects to
- * make room for these images.
- */
-/* NOTE: This function is only called while holding the hardware lock */
-int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t )
-{
-   int i;
-   int heap;
+   /* Format-specific hardware state:
+    */
+   t->pp_txformat &= ~(RADEON_TXFORMAT_FORMAT_MASK |
+		       RADEON_TXFORMAT_ALPHA_IN_MAP);
+   t->pp_txformat |= txformat;
 
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
-      fprintf( stderr, "%s( %p, %p )\n",
-	       __FUNCTION__, rmesa->glCtx, t );
+   if ( txformat == RADEON_TXFORMAT_RGBA8888 ||
+	txformat == RADEON_TXFORMAT_ARGB4444 ||
+	txformat == RADEON_TXFORMAT_ARGB1555 ||
+	txformat == RADEON_TXFORMAT_AI88 ) {
+      t->pp_txformat |= RADEON_TXFORMAT_ALPHA_IN_MAP;
    }
 
-   if ( !t )
-      return 0;
-
-   /* Choose the heap appropriately */
-   heap = t->heap = RADEON_CARD_HEAP;
-#if 0
-   if ( !rmesa->radeonScreen->IsPCI &&
-	t->totalSize > rmesa->radeonScreen->texSize[heap] ) {
-      heap = t->heap = RADEON_AGP_HEAP;
+   /* If we don't have a texture bound, bind this one.
+    */
+   if ( !rmesa->CurrentTexObj[unit] ) {
+      rmesa->CurrentTexObj[unit] = t;
+      t->bound |= unit + 1;
    }
-#endif
-
-   /* Do we need to eject LRU texture objects? */
-   if ( !t->memBlock ) {
-      /* Allocate a memory block on a 4k boundary (1<<12 == 4096) */
-      t->memBlock = mmAllocMem( rmesa->texHeap[heap],
-				t->totalSize, 12, 0 );
-
-#if 0
-      /* Try AGP before kicking anything out of local mem */
-      if ( !t->memBlock && heap == RADEON_CARD_HEAP ) {
-	 t->memBlock = mmAllocMem( rmesa->texHeap[RADEON_AGP_HEAP],
-				   t->totalSize, 12, 0 );
-
-	 if ( t->memBlock )
-	    heap = t->heap = RADEON_AGP_HEAP;
-      }
-#endif
-
-      /* Kick out textures until the requested texture fits */
-      while ( !t->memBlock ) {
-	 if ( rmesa->TexObjList[heap].prev->bound ) {
-	    fprintf( stderr,
-		     "radeonUploadTexImages: ran into bound texture\n" );
-	    return -1;
-	 }
-	 if ( rmesa->TexObjList[heap].prev ==
-	      &rmesa->TexObjList[heap] ) {
-	    if ( rmesa->radeonScreen->IsPCI ) {
-	       fprintf( stderr, "radeonUploadTexImages: upload texture "
-			"failure on local texture heaps, sz=%d\n",
-			t->totalSize );
-	       return -1;
-#if 0
-	    } else if ( heap == RADEON_CARD_HEAP ) {
-	       heap = t->heap = RADEON_AGP_HEAP;
-	       continue;
-#endif
-	    } else {
-	       fprintf( stderr, "radeonUploadTexImages: upload texture "
-			"failure on both local and AGP texture heaps, "
-			"sz=%d\n",
-			t->totalSize );
-	       return -1;
-	    }
-	 }
-
-	 radeonDestroyTexObj( rmesa, rmesa->TexObjList[heap].prev );
 
-	 t->memBlock = mmAllocMem( rmesa->texHeap[heap],
-				   t->totalSize, 12, 0 );
-      }
-
-      /* Set the base offset of the texture image */
-      t->bufAddr = rmesa->radeonScreen->texOffset[heap] + t->memBlock->ofs;
+   rmesa->new_state |= RADEON_NEW_TEXTURE;
 
-      t->setup.pp_txoffset = t->bufAddr;
-#if 0
-      /* Fix AGP texture offsets */
-      if ( heap == RADEON_AGP_HEAP ) {
-	  t->setup.pp_tx_offset += RADEON_AGP_TEX_OFFSET +
-	      rmesa->radeonScreen->agpTexOffset;
-      }
-#endif
+   *retainInternalCopy = GL_FALSE;
+   return GL_TRUE;
+}
 
-      /* Force loading the new state into the hardware */
-      switch ( t->bound ) {
-      case 1:
-	 rmesa->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_TEX0;
-	 break;
+static GLboolean
+radeonDDTexImage2D( GLcontext *ctx, GLenum target, GLint level,
+		    GLenum format, GLenum type, const GLvoid *pixels,
+		    const struct gl_pixelstore_attrib *packing,
+		    struct gl_texture_object *texObj,
+		    struct gl_texture_image *texImage,
+		    GLboolean *retainInternalCopy )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   const struct gl_texture_format *texFormat;
+   GLuint unit = ctx->Texture.CurrentUnit;
+   GLuint texSize;
+   GLint txformat;
+   GLubyte *data;
 
-      case 2:
-	 rmesa->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_TEX1;
-	 break;
+   if ( target != GL_TEXTURE_2D )
+      return GL_FALSE;
 
-      default:
-	 return -1;
-      }
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p, %d )\n", __FUNCTION__, texObj, level );
    }
 
-   /* Let the world know we've used this memory recently */
-   radeonUpdateTexLRU( rmesa, t );
+   if ( t ) {
+      if ( t->bound ) FLUSH_BATCH( rmesa );
+      radeonSwapOutTexObj( rmesa, t );
+   } else {
+      t = radeonAllocTexObj( texObj );
+      texObj->DriverData = t;
+   }
 
-   /* Upload any images that are new */
-   if ( t->dirty_images ) {
-      int num_levels = ((t->setup.pp_txfilter & RADEON_MAX_MIP_LEVEL_MASK) >>
-			RADEON_MAX_MIP_LEVEL_SHIFT);
+   txformat = radeonChooseTexFormat( rmesa, texImage, format, type );
+   if ( txformat < 0 )
+      return GL_FALSE;
 
-      for ( i = 0 ; i <= num_levels ; i++ ) {
-	 if ( t->dirty_images & (1 << i) ) {
-	    radeonUploadSubImage( rmesa, t, i, 0, 0,
-				  t->image[i].width, t->image[i].height );
-	 }
-      }
+   texFormat = texImage->TexFormat;
+   texSize = texImage->Width * texImage->Height * texFormat->TexelBytes;
 
-      rmesa->dirty |= RADEON_UPLOAD_CONTEXT;
+   /* We really shouldn't have to keep the texture image, it should be
+    * hung from the main texImage structure.
+    */
+   if ( t->image[level].data ) {
+      FREE( t->image[level].data );
+      t->image[level].data = NULL;
    }
 
-   t->dirty_images = 0;
-   return 0;
-}
-
+   data = (GLubyte *) MALLOC( texSize );
+   if ( !data )
+      return GL_FALSE;
 
-/* ================================================================
- * Texture combine functions
- */
+   if ( !_mesa_convert_texsubimage2d( texFormat->IntFormat,
+				      0, 0, texImage->Width, texImage->Height,
+				      texImage->Width, format, type, packing,
+				      pixels, data ) ) {
+      if ( 0 )
+	 fprintf( stderr, "   *** convert failed!  %s/%s-> %s\n",
+		  gl_lookup_enum_by_nr( format ),
+		  gl_lookup_enum_by_nr( type ),
+		  gl_lookup_enum_by_nr( texImage->IntFormat ) );
+      FREE( data );
+      return GL_FALSE;
+   }
 
-#define RADEON_DISABLE		0
-#define RADEON_REPLACE		1
-#define RADEON_MODULATE		2
-#define RADEON_DECAL		3
-#define RADEON_BLEND		4
-#define RADEON_ADD		5
-#define RADEON_MAX_COMBFUNC	6
+   t->image[level].data = data;
+   t->dirty_images |= (1 << level);
 
-static GLuint radeon_color_combine[][RADEON_MAX_COMBFUNC] =
-{
-   /* Unit 0:
-    */
-   {
-      /* Disable combiner stage
-       */
-      (RADEON_COLOR_ARG_A_ZERO |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_CURRENT_COLOR |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_REPLACE = 0x00802800
-       */
-      (RADEON_COLOR_ARG_A_ZERO |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_T0_COLOR |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_MODULATE = 0x00800142
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_T0_COLOR |
-       RADEON_COLOR_ARG_C_ZERO |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_DECAL = 0x008c2d42
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_T0_COLOR |
-       RADEON_COLOR_ARG_C_T0_ALPHA |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_BLEND = 0x008c2902
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
-       RADEON_COLOR_ARG_C_T0_COLOR |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_ADD = 0x00812802
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_T0_COLOR |
-       RADEON_COMP_ARG_B |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-   },
-
-   /* Unit 1:
-    */
-   {
-      /* Disable combiner stage
-       */
-      (RADEON_COLOR_ARG_A_ZERO |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_CURRENT_COLOR |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_REPLACE = 0x00803000
-       */
-      (RADEON_COLOR_ARG_A_ZERO |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_T1_COLOR |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_MODULATE = 0x00800182
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_T1_COLOR |
-       RADEON_COLOR_ARG_C_ZERO |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_DECAL = 0x008c3582
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_T1_COLOR |
-       RADEON_COLOR_ARG_C_T1_ALPHA |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_BLEND = 0x008c3102
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
-       RADEON_COLOR_ARG_C_T1_COLOR |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_ADD = 0x00813002
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_T1_COLOR |
-       RADEON_COMP_ARG_B |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-   },
-
-   /* Unit 2:
+   /* Format-specific hardware state:
     */
-   {
-      /* Disable combiner stage
-       */
-      (RADEON_COLOR_ARG_A_ZERO |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_CURRENT_COLOR |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_REPLACE = 0x00803800
-       */
-      (RADEON_COLOR_ARG_A_ZERO |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_T2_COLOR |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_MODULATE = 0x008001c2
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_T2_COLOR |
-       RADEON_COLOR_ARG_C_ZERO |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_DECAL = 0x008c3dc2
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_T2_COLOR |
-       RADEON_COLOR_ARG_C_T2_ALPHA |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_BLEND = 0x008c3902
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
-       RADEON_COLOR_ARG_C_T2_COLOR |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_ADD = 0x00813802
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_T2_COLOR |
-       RADEON_COMP_ARG_B |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
+   t->pp_txformat &= ~(RADEON_TXFORMAT_FORMAT_MASK |
+		       RADEON_TXFORMAT_ALPHA_IN_MAP);
+   t->pp_txformat |= txformat;
+
+   if ( txformat == RADEON_TXFORMAT_RGBA8888 ||
+	txformat == RADEON_TXFORMAT_ARGB4444 ||
+	txformat == RADEON_TXFORMAT_ARGB1555 ||
+	txformat == RADEON_TXFORMAT_AI88 ) {
+      t->pp_txformat |= RADEON_TXFORMAT_ALPHA_IN_MAP;
    }
-};
 
-static GLuint radeon_alpha_combine[][RADEON_MAX_COMBFUNC] =
-{
-   /* Unit 0:
-    */
-   {
-      /* Disable combiner stage
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_REPLACE = 0x00800500
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_T0_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_MODULATE = 0x00800051
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_T0_ALPHA |
-       RADEON_ALPHA_ARG_C_ZERO |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_DECAL = 0x00800100
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_BLEND = 0x00800051
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
-       RADEON_ALPHA_ARG_C_T0_ALPHA |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_ADD = 0x00800051
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_T0_ALPHA |
-       RADEON_COMP_ARG_B |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-   },
-
-   /* Unit 1:
+   /* If we don't have a texture bound, bind this one.
     */
-   {
-      /* Disable combiner stage
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_REPLACE = 0x00800600
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_T1_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_MODULATE = 0x00800061
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_T1_ALPHA |
-       RADEON_ALPHA_ARG_C_ZERO |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_DECAL = 0x00800100
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_BLEND = 0x00800061
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
-       RADEON_ALPHA_ARG_C_T1_ALPHA |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_ADD = 0x00800061
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_T1_ALPHA |
-       RADEON_COMP_ARG_B |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-   },
-
-   /* Unit 2:
-    */
-   {
-      /* Disable combiner stage
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_REPLACE = 0x00800700
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_T2_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_MODULATE = 0x00800071
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_T2_ALPHA |
-       RADEON_ALPHA_ARG_C_ZERO |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_DECAL = 0x00800100
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_BLEND = 0x00800071
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
-       RADEON_ALPHA_ARG_C_T2_ALPHA |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_ADD = 0x00800021
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_T2_ALPHA |
-       RADEON_COMP_ARG_B |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
+   if ( !rmesa->CurrentTexObj[unit] ) {
+      rmesa->CurrentTexObj[unit] = t;
+      t->bound |= unit + 1;
    }
-};
 
+   rmesa->new_state |= RADEON_NEW_TEXTURE;
 
-/* GL_EXT_texture_env_combine support
- */
+   *retainInternalCopy = GL_FALSE;
+   return GL_TRUE;
+}
 
-/* The color tables have combine functions for GL_SRC_COLOR,
- * GL_ONE_MINUS_SRC_COLOR, GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
+/* GH: This is undoubtedly broken...
  */
-static GLuint radeon_texture_color[][RADEON_MAX_TEXTURE_UNITS] =
-{
-   {
-      RADEON_COLOR_ARG_A_T0_COLOR,
-      RADEON_COLOR_ARG_A_T1_COLOR,
-      RADEON_COLOR_ARG_A_T2_COLOR
-   },
-   {
-      RADEON_COLOR_ARG_A_T0_COLOR | RADEON_COMP_ARG_A,
-      RADEON_COLOR_ARG_A_T1_COLOR | RADEON_COMP_ARG_A,
-      RADEON_COLOR_ARG_A_T2_COLOR | RADEON_COMP_ARG_A
-   },
-   {
-      RADEON_COLOR_ARG_A_T0_ALPHA,
-      RADEON_COLOR_ARG_A_T1_ALPHA,
-      RADEON_COLOR_ARG_A_T2_ALPHA
-   },
-   {
-      RADEON_COLOR_ARG_A_T0_ALPHA | RADEON_COMP_ARG_A,
-      RADEON_COLOR_ARG_A_T1_ALPHA | RADEON_COMP_ARG_A,
-      RADEON_COLOR_ARG_A_T2_ALPHA | RADEON_COMP_ARG_A
-   },
-};
-
-static GLuint radeon_tfactor_color[] =
-{
-   RADEON_COLOR_ARG_A_TFACTOR_COLOR,
-   RADEON_COLOR_ARG_A_TFACTOR_COLOR | RADEON_COMP_ARG_A,
-   RADEON_COLOR_ARG_A_TFACTOR_ALPHA,
-   RADEON_COLOR_ARG_A_TFACTOR_ALPHA | RADEON_COMP_ARG_A
-};
-
-static GLuint radeon_primary_color[] =
-{
-   RADEON_COLOR_ARG_A_DIFFUSE_COLOR,
-   RADEON_COLOR_ARG_A_DIFFUSE_COLOR | RADEON_COMP_ARG_A,
-   RADEON_COLOR_ARG_A_DIFFUSE_ALPHA,
-   RADEON_COLOR_ARG_A_DIFFUSE_ALPHA | RADEON_COMP_ARG_A
-};
-
-static GLuint radeon_previous_color[] =
+static GLboolean
+radeonDDTexImage3D( GLcontext *ctx, GLenum target, GLint level,
+		    GLenum format, GLenum type, const GLvoid *pixels,
+		    const struct gl_pixelstore_attrib *packing,
+		    struct gl_texture_object *texObj,
+		    struct gl_texture_image *texImage,
+		    GLboolean *retainInternalCopy )
 {
-   RADEON_COLOR_ARG_A_CURRENT_COLOR,
-   RADEON_COLOR_ARG_A_CURRENT_COLOR | RADEON_COMP_ARG_A,
-   RADEON_COLOR_ARG_A_CURRENT_ALPHA,
-   RADEON_COLOR_ARG_A_CURRENT_ALPHA | RADEON_COMP_ARG_A
-};
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   const struct gl_texture_format *texFormat;
+   GLuint unit = ctx->Texture.CurrentUnit;
+   GLuint texSize;
+   GLint txformat;
+   GLubyte *data;
 
-/* The alpha tables only have GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
- */
-static GLuint radeon_texture_alpha[][RADEON_MAX_TEXTURE_UNITS] =
-{
-   {
-      RADEON_ALPHA_ARG_A_T0_ALPHA,
-      RADEON_ALPHA_ARG_A_T1_ALPHA,
-      RADEON_ALPHA_ARG_A_T2_ALPHA
-   },
-   {
-      RADEON_ALPHA_ARG_A_T0_ALPHA | RADEON_COMP_ARG_A,
-      RADEON_ALPHA_ARG_A_T1_ALPHA | RADEON_COMP_ARG_A,
-      RADEON_ALPHA_ARG_A_T2_ALPHA | RADEON_COMP_ARG_A
-   },
-};
-
-static GLuint radeon_tfactor_alpha[] =
-{
-   RADEON_ALPHA_ARG_A_TFACTOR_ALPHA,
-   RADEON_ALPHA_ARG_A_TFACTOR_ALPHA | RADEON_COMP_ARG_A
-};
+   if ( target != GL_TEXTURE_3D )
+      return GL_FALSE;
 
-static GLuint radeon_primary_alpha[] =
-{
-   RADEON_ALPHA_ARG_A_DIFFUSE_ALPHA,
-   RADEON_ALPHA_ARG_A_DIFFUSE_ALPHA | RADEON_COMP_ARG_A
-};
+   if ( t ) {
+      if ( t->bound ) FLUSH_BATCH( rmesa );
+      radeonSwapOutTexObj( rmesa, t );
+   } else {
+      t = radeonAllocTexObj( texObj );
+      texObj->DriverData = t;
+   }
 
-static GLuint radeon_previous_alpha[] =
-{
-   RADEON_ALPHA_ARG_A_CURRENT_ALPHA,
-   RADEON_ALPHA_ARG_A_CURRENT_ALPHA | RADEON_COMP_ARG_A
-};
+   txformat = radeonChooseTexFormat( rmesa, texImage, format, type );
+   if ( txformat < 0 )
+      return GL_FALSE;
 
+   texFormat = texImage->TexFormat;
+   texSize = (texImage->Width * texImage->Height *
+	      texImage->Depth * texFormat->TexelBytes);
 
-/* Extract the arg from slot A, shift it into the correct argument slot
- * and set the corresponding complement bit.
- */
-#define RADEON_COLOR_ARG( n, arg )					\
-do {									\
-   color_combine |=							\
-      ((color_arg[n] & RADEON_COLOR_ARG_MASK)				\
-       << RADEON_COLOR_ARG_##arg##_SHIFT);				\
-   color_combine |=							\
-      ((color_arg[n] >> RADEON_COMP_ARG_SHIFT)				\
-       << RADEON_COMP_ARG_##arg##_SHIFT);				\
-} while (0)
-
-#define RADEON_ALPHA_ARG( n, arg )					\
-do {									\
-   alpha_combine |=							\
-      ((alpha_arg[n] & RADEON_ALPHA_ARG_MASK)				\
-       << RADEON_ALPHA_ARG_##arg##_SHIFT);				\
-   alpha_combine |=							\
-      ((alpha_arg[n] >> RADEON_COMP_ARG_SHIFT)				\
-       << RADEON_COMP_ARG_##arg##_SHIFT);				\
-} while (0)
+   /* We really shouldn't have to keep the texture image, it should be
+    * hung from the main texImage structure.
+    */
+   if ( t->image[level].data ) {
+      FREE( t->image[level].data );
+      t->image[level].data = NULL;
+   }
 
+   data = (GLubyte *) MALLOC( texSize );
+   if ( !data )
+      return GL_FALSE;
 
-/* ================================================================
- * Texture unit state management
- */
+   if ( !_mesa_convert_texsubimage3d( texFormat->IntFormat,
+				      0, 0, 0, texImage->Width,
+				      texImage->Height, texImage->Depth,
+				      texImage->Width, texImage->Height,
+				      format, type, packing,
+				      pixels, data ) ) {
+      FREE( data );
+      return GL_FALSE;
+   }
 
-static void radeonUpdateTextureEnv( GLcontext *ctx, int unit )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   int source = rmesa->tmu_source[unit];
-   struct gl_texture_object *tObj;
-   struct gl_texture_unit *texUnit;
-   GLuint enabled;
-   GLuint color_combine, alpha_combine;
-   GLuint color_arg[3], alpha_arg[3];
-   GLuint i, numColorArgs = 0, numAlphaArgs = 0;
-   GLuint op;
+   t->image[level].data = data;
+   t->dirty_images |= (1 << level);
 
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
-      fprintf( stderr, "%s( %p, %d )\n",
-	       __FUNCTION__, ctx, unit );
+   /* If we don't have a texture bound, bind this one.
+    */
+   if ( !rmesa->CurrentTexObj[unit] ) {
+      rmesa->CurrentTexObj[unit] = t;
+      t->bound |= unit + 1;
    }
 
-   enabled = (ctx->Texture.ReallyEnabled >> (source * 4)) & TEXTURE0_ANY;
-   if ( enabled != TEXTURE0_2D && enabled != TEXTURE0_1D )
-      return;
+   rmesa->new_state |= RADEON_NEW_TEXTURE;
 
-   /* Only update the hardware texture state if the texture is current,
-    * complete and enabled.
-    */
-   texUnit = &ctx->Texture.Unit[source];
-   tObj = texUnit->Current;
-   if ( !tObj || !tObj->Complete )
-      return;
+   *retainInternalCopy = GL_FALSE;
+   return GL_TRUE;
+}
 
-   if ( ( tObj != texUnit->CurrentD[2] ) &&
-	( tObj != texUnit->CurrentD[1] ) )
-      return;
 
-   /* Set the texture environment state.  Isn't this nice and clean?
-    * The Radeon will automagically set the texture alpha to 0xff when
-    * the texture format does not include an alpha component.  This
-    * reduces the amount of special-casing we have to do, alpha-only
-    * textures being a notable exception.
-    */
-   switch ( texUnit->EnvMode ) {
-   case GL_REPLACE:
-      switch ( tObj->Image[0]->Format ) {
-      case GL_RGBA:
-      case GL_RGB:
-      case GL_LUMINANCE:
-      case GL_LUMINANCE_ALPHA:
-      case GL_INTENSITY:
-	 color_combine = radeon_color_combine[unit][RADEON_REPLACE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_REPLACE];
-	 break;
-      case GL_ALPHA:
-	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_REPLACE];
-	 break;
-      case GL_COLOR_INDEX:
-      default:
-	 return;
-      }
-      break;
+/* ================================================================
+ * Texture subimage callbacks
+ */
 
-   case GL_MODULATE:
-      switch ( tObj->Image[0]->Format ) {
-      case GL_RGBA:
-      case GL_RGB:
-      case GL_LUMINANCE:
-      case GL_LUMINANCE_ALPHA:
-      case GL_INTENSITY:
-	 color_combine = radeon_color_combine[unit][RADEON_MODULATE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
-	 break;
-      case GL_ALPHA:
-	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
-	 break;
-      case GL_COLOR_INDEX:
-      default:
-	 return;
-      }
-      break;
+static GLboolean
+radeonDDTexSubImage1D( GLcontext *ctx, GLenum target, GLint level,
+		       GLint xoffset, GLsizei width,
+		       GLenum format, GLenum type,
+		       const GLvoid *pixels,
+		       const struct gl_pixelstore_attrib *packing,
+		       struct gl_texture_object *texObj,
+		       struct gl_texture_image *texImage )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   const struct gl_texture_format *texFormat;
 
-   case GL_DECAL:
-      switch ( tObj->Image[0]->Format ) {
-      case GL_RGBA:
-      case GL_RGB:
-	 color_combine = radeon_color_combine[unit][RADEON_DECAL];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
-	 break;
-      case GL_ALPHA:
-      case GL_LUMINANCE:
-      case GL_LUMINANCE_ALPHA:
-      case GL_INTENSITY:
-	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
-	 break;
-      case GL_COLOR_INDEX:
-      default:
-	 return;
-      }
-      break;
+   if ( target != GL_TEXTURE_1D )
+      return GL_FALSE;
 
-   case GL_BLEND:
-      switch ( tObj->Image[0]->Format ) {
-      case GL_RGBA:
-      case GL_RGB:
-      case GL_LUMINANCE:
-      case GL_LUMINANCE_ALPHA:
-	 color_combine = radeon_color_combine[unit][RADEON_BLEND];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
-	 break;
-      case GL_ALPHA:
-	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
-	 break;
-      case GL_INTENSITY:
-	 color_combine = radeon_color_combine[unit][RADEON_BLEND];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_BLEND];
-	 break;
-      case GL_COLOR_INDEX:
-      default:
-	 return;
-      }
-      break;
+   /* FIXME: Can this ever be NULL???
+    */
+   ASSERT( t );
 
-   case GL_ADD:
-      switch ( tObj->Image[0]->Format ) {
-      case GL_RGBA:
-      case GL_RGB:
-      case GL_LUMINANCE:
-      case GL_LUMINANCE_ALPHA:
-	 color_combine = radeon_color_combine[unit][RADEON_ADD];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
-	 break;
-      case GL_ALPHA:
-	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
-	 break;
-      case GL_INTENSITY:
-	 color_combine = radeon_color_combine[unit][RADEON_ADD];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_ADD];
-	 break;
-      case GL_COLOR_INDEX:
-      default:
-	 return;
-      }
-      break;
+   if ( t->bound )
+      FLUSH_BATCH( rmesa );
 
-   case GL_COMBINE_EXT:
-      /* Step 0:
-       * Calculate how many arguments we need to process.
-       */
-      switch ( texUnit->CombineModeRGB ) {
-      case GL_REPLACE:
-	 numColorArgs = 1;
-	 break;
-      case GL_MODULATE:
-      case GL_ADD:
-      case GL_ADD_SIGNED_EXT:
-      case GL_DOT3_RGB_EXT:
-      case GL_DOT3_RGBA_EXT:
-	 numColorArgs = 2;
-	 break;
-      case GL_INTERPOLATE_EXT:
-	 numColorArgs = 3;
-	 break;
-      default:
-	 return;
-      }
+   texFormat = texImage->TexFormat;
 
-      switch ( texUnit->CombineModeA ) {
-      case GL_REPLACE:
-	 numAlphaArgs = 1;
-	 break;
-      case GL_MODULATE:
-      case GL_ADD:
-      case GL_ADD_SIGNED_EXT:
-	 numAlphaArgs = 2;
-	 break;
-      case GL_INTERPOLATE_EXT:
-	 numAlphaArgs = 3;
-	 break;
-      default:
-	 return;
-      }
+   if ( !_mesa_convert_texsubimage1d( texFormat->IntFormat,
+				      xoffset, width, format, type, packing,
+				      pixels, t->image[level].data ) ) {
+      /*fprintf( stderr, "   *** convert failed!\n" );*/
+      return GL_FALSE;
+   }
 
-      /* Step 1:
-       * Extract the color and alpha combine function arguments.
-       */
-      for ( i = 0 ; i < numColorArgs ; i++ ) {
-	 op = texUnit->CombineOperandRGB[i] - GL_SRC_COLOR;
-	 switch ( texUnit->CombineSourceRGB[i] ) {
-	 case GL_TEXTURE:
-	    color_arg[i] = radeon_texture_color[op][unit];
-	    break;
-	 case GL_CONSTANT_EXT:
-	    color_arg[i] = radeon_tfactor_color[op];
-	    break;
-	 case GL_PRIMARY_COLOR_EXT:
-	    color_arg[i] = radeon_primary_color[op];
-	    break;
-	 case GL_PREVIOUS_EXT:
-	    color_arg[i] = radeon_previous_color[op];
-	    break;
-	 default:
-	    return;
-	 }
-      }
+   t->dirty_images |= (1 << level);
+   rmesa->new_state |= RADEON_NEW_TEXTURE;
 
-      for ( i = 0 ; i < numAlphaArgs ; i++ ) {
-	 op = texUnit->CombineOperandA[i] - GL_SRC_ALPHA;
-	 switch ( texUnit->CombineSourceA[i] ) {
-	 case GL_TEXTURE:
-	    alpha_arg[i] = radeon_texture_alpha[op][unit];
-	    break;
-	 case GL_CONSTANT_EXT:
-	    alpha_arg[i] = radeon_tfactor_alpha[op];
-	    break;
-	 case GL_PRIMARY_COLOR_EXT:
-	    alpha_arg[i] = radeon_primary_alpha[op];
-	    break;
-	 case GL_PREVIOUS_EXT:
-	    alpha_arg[i] = radeon_previous_alpha[op];
-	    break;
-	 default:
-	    return;
-	 }
-      }
+   return GL_TRUE;
+}
 
-      /* Step 2:
-       * Build up the color and alpha combine functions.
-       */
-      switch ( texUnit->CombineModeRGB ) {
-      case GL_REPLACE:
-	 color_combine = (RADEON_COLOR_ARG_A_ZERO |
-			  RADEON_COLOR_ARG_B_ZERO |
-			  RADEON_BLEND_CTL_ADD |
-			  RADEON_CLAMP_TX);
-	 RADEON_COLOR_ARG( 0, C );
-	 break;
-      case GL_MODULATE:
-	 color_combine = (RADEON_COLOR_ARG_C_ZERO |
-			  RADEON_BLEND_CTL_ADD |
-			  RADEON_CLAMP_TX);
-	 RADEON_COLOR_ARG( 0, A );
-	 RADEON_COLOR_ARG( 1, B );
-	 break;
-      case GL_ADD:
-	 color_combine = (RADEON_COLOR_ARG_B_ZERO |
-			  RADEON_COMP_ARG_B |
-			  RADEON_BLEND_CTL_ADD |
-			  RADEON_CLAMP_TX);
-	 RADEON_COLOR_ARG( 0, A );
-	 RADEON_COLOR_ARG( 1, C );
-	 break;
-      case GL_ADD_SIGNED_EXT:
-	 color_combine = (RADEON_COLOR_ARG_B_ZERO |
-			  RADEON_COMP_ARG_B |
-			  RADEON_BLEND_CTL_ADDSIGNED |
-			  RADEON_CLAMP_TX);
-	 RADEON_COLOR_ARG( 0, A );
-	 RADEON_COLOR_ARG( 1, C );
-	 break;
-      case GL_INTERPOLATE_EXT:
-	 color_combine = (RADEON_BLEND_CTL_BLEND |
-			  RADEON_CLAMP_TX);
-	 RADEON_COLOR_ARG( 0, B );
-	 RADEON_COLOR_ARG( 1, A );
-	 RADEON_COLOR_ARG( 2, C );
-	 break;
-      case GL_DOT3_RGB_EXT:
-      case GL_DOT3_RGBA_EXT:
-	 color_combine = (RADEON_COLOR_ARG_C_ZERO |
-			  RADEON_BLEND_CTL_DOT3 |
-			  RADEON_CLAMP_TX);
-	 RADEON_COLOR_ARG( 0, A );
-	 RADEON_COLOR_ARG( 1, B );
-	 break;
-      default:
-	 return;
-      }
+static GLboolean
+radeonDDTexSubImage2D( GLcontext *ctx, GLenum target, GLint level,
+		       GLint xoffset, GLint yoffset,
+		       GLsizei width, GLsizei height,
+		       GLenum format, GLenum type,
+		       const GLvoid *pixels,
+		       const struct gl_pixelstore_attrib *packing,
+		       struct gl_texture_object *texObj,
+		       struct gl_texture_image *texImage )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   const struct gl_texture_format *texFormat;
 
-      switch ( texUnit->CombineModeA ) {
-      case GL_REPLACE:
-	 alpha_combine = (RADEON_ALPHA_ARG_A_ZERO |
-			  RADEON_ALPHA_ARG_B_ZERO |
-			  RADEON_BLEND_CTL_ADD |
-			  RADEON_CLAMP_TX);
-	 RADEON_ALPHA_ARG( 0, C );
-	 break;
-      case GL_MODULATE:
-	 alpha_combine = (RADEON_ALPHA_ARG_C_ZERO |
-			  RADEON_BLEND_CTL_ADD |
-			  RADEON_CLAMP_TX);
-	 RADEON_ALPHA_ARG( 0, A );
-	 RADEON_ALPHA_ARG( 1, B );
-	 break;
-      case GL_ADD:
-	 alpha_combine = (RADEON_ALPHA_ARG_B_ZERO |
-			  RADEON_COMP_ARG_B |
-			  RADEON_BLEND_CTL_ADD |
-			  RADEON_CLAMP_TX);
-	 RADEON_ALPHA_ARG( 0, A );
-	 RADEON_ALPHA_ARG( 1, C );
-	 break;
-      case GL_ADD_SIGNED_EXT:
-	 alpha_combine = (RADEON_ALPHA_ARG_B_ZERO |
-			  RADEON_COMP_ARG_B |
-			  RADEON_BLEND_CTL_ADDSIGNED |
-			  RADEON_CLAMP_TX);
-	 RADEON_ALPHA_ARG( 0, A );
-	 RADEON_ALPHA_ARG( 1, C );
-	 break;
-      case GL_INTERPOLATE_EXT:
-	 alpha_combine = (RADEON_BLEND_CTL_BLEND |
-			  RADEON_CLAMP_TX);
-	 RADEON_ALPHA_ARG( 0, B );
-	 RADEON_ALPHA_ARG( 1, A );
-	 RADEON_ALPHA_ARG( 2, C );
-	 break;
-      default:
-	 return;
-      }
+   if ( target != GL_TEXTURE_2D )
+      return GL_FALSE;
 
-      if ( texUnit->CombineModeRGB == GL_DOT3_RGB_EXT ) {
-	 alpha_combine |= RADEON_DOT_ALPHA_DONT_REPLICATE;
-      }
+   /* FIXME: Can this ever be NULL???
+    */
+   ASSERT( t );
 
-      /* Step 3:
-       * Apply the scale factor.  The EXT extension has a somewhat
-       * unnecessary restriction that the scale must be 4x.  The ARB
-       * extension will likely drop this and we can just apply the
-       * scale factors regardless.
-       */
-      if ( texUnit->CombineModeRGB != GL_DOT3_RGB_EXT &&
-	   texUnit->CombineModeRGB != GL_DOT3_RGBA_EXT ) {
-	 color_combine |= (texUnit->CombineScaleShiftRGB << 21);
-	 alpha_combine |= (texUnit->CombineScaleShiftA << 21);
-      } else {
-	 color_combine |= RADEON_SCALE_4X;
-	 alpha_combine |= RADEON_SCALE_4X;
-      }
+   if ( t->bound )
+      FLUSH_BATCH( rmesa );
 
-      /* All done!
-       */
-      break;
+   texFormat = texImage->TexFormat;
 
-   default:
-      return;
+   if ( !_mesa_convert_texsubimage2d( texFormat->IntFormat,
+				      xoffset, yoffset, width, height,
+				      texImage->Width, format, type, packing,
+				      pixels, t->image[level].data ) ) {
+      /*fprintf( stderr, "   *** convert failed!\n" );*/
+      return GL_FALSE;
    }
 
-   rmesa->color_combine[source] = color_combine;
-   rmesa->alpha_combine[source] = alpha_combine;
+   t->dirty_images |= (1 << level);
+   rmesa->new_state |= RADEON_NEW_TEXTURE;
+
+   return GL_TRUE;
 }
 
-static void radeonUpdateTextureObject( GLcontext *ctx, int unit )
+/* GH: This is undoubtedly broken...
+ */
+static GLboolean
+radeonDDTexSubImage3D( GLcontext *ctx, GLenum target, GLint level,
+		       GLint xoffset, GLint yoffset, GLint zoffset,
+		       GLsizei width, GLsizei height, GLint depth,
+		       GLenum format, GLenum type,
+		       const GLvoid *pixels,
+		       const struct gl_pixelstore_attrib *packing,
+		       struct gl_texture_object *texObj,
+		       struct gl_texture_image *texImage )
 {
    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   int source = rmesa->tmu_source[unit];
-   struct gl_texture_object *tObj;
-   radeonTexObjPtr t;
-   GLuint enabled;
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   const struct gl_texture_format *texFormat;
 
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
-      fprintf( stderr, "%s( %p, %d )\n",
-	       __FUNCTION__, ctx, unit );
-   }
-
-   enabled = (ctx->Texture.ReallyEnabled >> (source * 4)) & TEXTURE0_ANY;
-   if ( enabled != TEXTURE0_2D && enabled != TEXTURE0_1D ) {
-      if ( enabled )
-	 rmesa->Fallback |= RADEON_FALLBACK_TEXTURE;
-      return;
-   }
+   if ( target != GL_TEXTURE_3D )
+      return GL_FALSE;
 
-   /* Only update the hardware texture state if the texture is current,
-    * complete and enabled.
+   /* FIXME: Can this ever be NULL???
     */
-   tObj = ctx->Texture.Unit[source].Current;
-   if ( !tObj || !tObj->Complete )
-      return;
-
-   if ( ( tObj != ctx->Texture.Unit[source].CurrentD[2] ) &&
-	( tObj != ctx->Texture.Unit[source].CurrentD[1] ) )
-      return;
+   ASSERT( t );
 
-   if ( !tObj->DriverData ) {
-      /* If this is the first time the texture has been used, then create
-       * a new texture object for it.
-       */
-      radeonCreateTexObj( rmesa, tObj );
-
-      if ( !tObj->DriverData ) {
-	 /* Can't create a texture object... */
-	 fprintf( stderr, "%s: texture object creation failed!\n",
-		  __FUNCTION__ );
-	 rmesa->Fallback |= RADEON_FALLBACK_TEXTURE;
-	 return;
-      }
-   }
-
-   /* We definately have a valid texture now */
-   t = tObj->DriverData;
-
-   /* Force the texture unit state to be loaded into the hardware */
-   rmesa->dirty |= RADEON_UPLOAD_CONTEXT | (RADEON_UPLOAD_TEX0 << unit);
+   if ( t->bound )
+      FLUSH_BATCH( rmesa );
 
-   /* Force any texture images to be loaded into the hardware */
-   if ( t->dirty_images )
-      rmesa->dirty |= (RADEON_UPLOAD_TEX0IMAGES << unit);
+   texFormat = texImage->TexFormat;
 
-   /* Bind to the given texture unit */
-   rmesa->CurrentTexObj[unit] = t;
-   t->bound = unit + 1;
+   if ( !_mesa_convert_texsubimage3d( texFormat->IntFormat,
+				      xoffset, yoffset, zoffset,
+				      width, height, depth,
+				      texImage->Width, texImage->Height,
+				      format, type, packing,
+				      pixels, t->image[level].data ) ) {
+      /*fprintf( stderr, "   *** convert failed!\n" );*/
+      return GL_FALSE;
+   }
 
-   if ( t->memBlock )
-      radeonUpdateTexLRU( rmesa, t );
+   t->dirty_images |= (1 << level);
+   rmesa->new_state |= RADEON_NEW_TEXTURE;
 
-   switch ( unit ) {
-   case 0:
-      rmesa->setup.pp_cntl |= (RADEON_TEX_0_ENABLE |
-			       RADEON_TEX_BLEND_0_ENABLE);
-      break;
-   case 1:
-      rmesa->setup.pp_cntl |= (RADEON_TEX_1_ENABLE |
-			       RADEON_TEX_BLEND_1_ENABLE);
-      break;
-   }
+   return GL_TRUE;
 }
 
-void radeonUpdateTextureState( GLcontext *ctx )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
-      fprintf( stderr, "%s( %p ) en=0x%x\n",
-	       __FUNCTION__, ctx, ctx->Texture.ReallyEnabled );
-   }
-
-   /* Clear any texturing fallbacks */
-   rmesa->Fallback &= ~RADEON_FALLBACK_TEXTURE;
+/* ================================================================
+ * DEPRECATED...
+ */
 
-   /* Unbind any currently bound textures */
-   if ( rmesa->CurrentTexObj[0] ) rmesa->CurrentTexObj[0]->bound = 0;
-   if ( rmesa->CurrentTexObj[1] ) rmesa->CurrentTexObj[1]->bound = 0;
-   rmesa->CurrentTexObj[0] = NULL;
-   rmesa->CurrentTexObj[1] = NULL;
+static GLvoid *radeonDDGetTexImage( GLcontext *ctx, GLenum target, GLint level,
+				    const struct gl_texture_object *texObj,
+				    GLenum *formatOut, GLenum *typeOut,
+				    GLboolean *freeImageOut )
+{
+   const struct gl_texture_image *texImage = texObj->Image[level];
+   const struct gl_texture_format *texFormat = texImage->TexFormat;
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   GLubyte *data;
 
-   if ( ctx->Enabled & (TEXTURE0_3D|TEXTURE1_3D) )
-      rmesa->Fallback |= RADEON_FALLBACK_TEXTURE;
+   if ( !t || !t->image[level].data )
+      return NULL;
 
-   /* Disable all texturing until it is known to be good */
-   rmesa->setup.pp_cntl &= ~(RADEON_TEX_ENABLE_MASK |
-			     RADEON_TEX_BLEND_ENABLE_MASK);
+   data = (GLubyte *) MALLOC( texImage->Width * texImage->Height * 4 );
+   if ( !data )
+      return NULL;
 
-   radeonUpdateTextureObject( ctx, 0 );
-   radeonUpdateTextureEnv( ctx, 0 );
+   if ( 0 )
+      fprintf( stderr, "   in=%d out=%s\n",
+	       texFormat->IntFormat,
+	       gl_lookup_enum_by_nr( texImage->Format ) );
 
-   if ( rmesa->multitex ) {
-      radeonUpdateTextureObject( ctx, 1 );
-      radeonUpdateTextureEnv( ctx, 1 );
+   switch ( target ) {
+   case GL_TEXTURE_1D:
+      _mesa_unconvert_teximage1d( texFormat->IntFormat, texImage->Format,
+				  texImage->Width,
+				  t->image[level].data, data );
+      break;
+   case GL_TEXTURE_2D:
+      _mesa_unconvert_teximage2d( texFormat->IntFormat, texImage->Format,
+				  texImage->Width, texImage->Height,
+				  t->image[level].data, data );
+      break;
+   default:
+      return NULL;
    }
 
-   rmesa->dirty |= RADEON_UPLOAD_CONTEXT;
+   *formatOut = texImage->Format;
+   *typeOut = GL_UNSIGNED_BYTE;
+   *freeImageOut = GL_TRUE;
+
+   return data;
 }
 
 
 /* ================================================================
- * DD interface texturing functions
- *
- * FIXME: Many of these are deprecated -- we should move to the new
- * single-copy texture interface.
+ * Texture state callbacks
  */
+
 #define SCALED_FLOAT_TO_BYTE( x, scale ) \
 		((((GLint)((256.0F / scale) * (x))) - 1) / 2)
 
@@ -2067,79 +805,6 @@ static void radeonDDTexEnv( GLcontext *ctx, GLenum target,
    }
 }
 
-static void radeonDDTexImage( GLcontext *ctx, GLenum target,
-			      struct gl_texture_object *tObj, GLint level,
-			      GLint internalFormat,
-			      const struct gl_texture_image *image )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   radeonTexObjPtr t;
-
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API )
-      fprintf( stderr, "%s( %p, level %d )\n", __FUNCTION__, tObj, level );
-
-   if ( ( target != GL_TEXTURE_2D ) &&
-	( target != GL_TEXTURE_1D ) )
-      return;
-
-   if ( level >= RADEON_MAX_TEXTURE_LEVELS )
-      return;
-
-   t = (radeonTexObjPtr)tObj->DriverData;
-   if ( t ) {
-      if ( t->bound ) FLUSH_BATCH( rmesa );
-
-      /* Destroy the old texture, and upload a new one.  The actual
-       * uploading of the texture image occurs in the UploadSubImage
-       * function.
-       */
-      radeonDestroyTexObj( rmesa, t );
-      rmesa->new_state |= RADEON_NEW_TEXTURE;
-   }
-}
-
-static void radeonDDTexSubImage( GLcontext *ctx, GLenum target,
-				 struct gl_texture_object *tObj, GLint level,
-				 GLint xoffset, GLint yoffset,
-				 GLsizei width, GLsizei height,
-				 GLint internalFormat,
-				 const struct gl_texture_image *image )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   radeonTexObjPtr t;
-
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
-      fprintf( stderr, "%s( %p, level %d ) size: %d,%d of %d,%d\n",
-	       __FUNCTION__, tObj, level, width, height,
-	       image->Width, image->Height );
-   }
-
-   if ( ( target != GL_TEXTURE_2D ) &&
-	( target != GL_TEXTURE_1D ) )
-      return;
-
-   if ( level >= RADEON_MAX_TEXTURE_LEVELS )
-      return;
-
-   t = (radeonTexObjPtr)tObj->DriverData;
-   if ( t ) {
-      if ( t->bound ) FLUSH_BATCH( rmesa );
-
-#if 0
-      /* FIXME: Only upload textures if we already have space in the heap.
-       */
-      LOCK_HARDWARE( rmesa );
-      radeonUploadSubImage( rmesa, t, level,
-			    xoffset, yoffset, width, height );
-      UNLOCK_HARDWARE( rmesa );
-#else
-      radeonDestroyTexObj( rmesa, t );
-#endif
-      /* Update the context state */
-      rmesa->new_state |= RADEON_NEW_TEXTURE;
-   }
-}
-
 static void radeonDDTexParameter( GLcontext *ctx, GLenum target,
 				  struct gl_texture_object *tObj,
 				  GLenum pname, const GLfloat *params )
@@ -2156,7 +821,7 @@ static void radeonDDTexParameter( GLcontext *ctx, GLenum target,
     * created with current state before it is used, so we don't have
     * to do anything now.
     */
-   if ( !t || !t->bound )
+   if ( !t )
       return;
 
    if ( ( target != GL_TEXTURE_2D ) &&
@@ -2192,34 +857,51 @@ static void radeonDDBindTexture( GLcontext *ctx, GLenum target,
 				 struct gl_texture_object *tObj )
 {
    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   GLint unit = ctx->Texture.CurrentUnit;
+   radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData;
+   GLuint unit = ctx->Texture.CurrentUnit;
 
    if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
       fprintf( stderr, "%s( %p ) unit=%d\n",
-	       __FUNCTION__, tObj, ctx->Texture.CurrentUnit );
+	       __FUNCTION__, tObj, unit );
    }
 
    FLUSH_BATCH( rmesa );
 
+   if ( !t ) {
+      t = radeonAllocTexObj( tObj );
+      tObj->DriverData = t;
+   }
+
+   /* Unbind a currently bound texture.
+    */
    if ( rmesa->CurrentTexObj[unit] ) {
-      rmesa->CurrentTexObj[unit]->bound = 0;
+      rmesa->CurrentTexObj[unit]->bound &= ~(unit + 1);
       rmesa->CurrentTexObj[unit] = NULL;
    }
 
+   /* Bind to the given texture unit.
+    */
+   rmesa->CurrentTexObj[unit] = t;
+   t->bound |= unit + 1;
+
    rmesa->new_state |= RADEON_NEW_TEXTURE;
 }
 
 static void radeonDDDeleteTexture( GLcontext *ctx,
-				 struct gl_texture_object *tObj )
+				   struct gl_texture_object *tObj )
 {
    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
    radeonTexObjPtr t = (radeonTexObjPtr)tObj->DriverData;
 
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, __FUNCTION__ "( %p )\n", tObj );
+   }
+
    if ( t ) {
       if ( t->bound ) {
 	 FLUSH_BATCH( rmesa );
-
-	 rmesa->CurrentTexObj[t->bound-1] = 0;
+	 if ( t->bound & TEX_0 ) rmesa->CurrentTexObj[0] = NULL;
+	 if ( t->bound & TEX_1 ) rmesa->CurrentTexObj[1] = NULL;
 	 rmesa->new_state |= RADEON_NEW_TEXTURE;
       }
 
@@ -2240,14 +922,19 @@ static GLboolean radeonDDIsTextureResident( GLcontext *ctx,
 
 void radeonDDInitTextureFuncs( GLcontext *ctx )
 {
+   ctx->Driver.TexImage1D		= radeonDDTexImage1D;
+   ctx->Driver.TexImage2D		= radeonDDTexImage2D;
+   ctx->Driver.TexImage3D		= NULL /*radeonDDTexImage3D*/;
+   ctx->Driver.TexSubImage1D		= radeonDDTexSubImage1D;
+   ctx->Driver.TexSubImage2D		= radeonDDTexSubImage2D;
+   ctx->Driver.TexSubImage3D		= NULL /*radeonDDTexSubImage3D*/;
+   ctx->Driver.GetTexImage		= radeonDDGetTexImage;
    ctx->Driver.TexEnv			= radeonDDTexEnv;
-   ctx->Driver.TexImage			= radeonDDTexImage;
-   ctx->Driver.TexSubImage		= radeonDDTexSubImage;
    ctx->Driver.TexParameter		= radeonDDTexParameter;
    ctx->Driver.BindTexture		= radeonDDBindTexture;
    ctx->Driver.DeleteTexture		= radeonDDDeleteTexture;
-   ctx->Driver.UpdateTexturePalette	= NULL;
-   ctx->Driver.ActiveTexture		= NULL;
    ctx->Driver.IsTextureResident	= radeonDDIsTextureResident;
    ctx->Driver.PrioritizeTexture	= NULL;
+   ctx->Driver.ActiveTexture		= NULL;
+   ctx->Driver.UpdateTexturePalette	= NULL;
 }
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.h b/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.h
index 6bf967761..967048dd1 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.h
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.h
@@ -45,9 +45,11 @@ extern int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t );
 
 extern void radeonAgeTextures( radeonContextPtr rmesa, int heap );
 extern void radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t );
+extern void radeonSwapOutTexObj( radeonContextPtr rmesa, radeonTexObjPtr t );
 
 extern void radeonPrintLocalLRU( radeonContextPtr rmesa, int heap );
 extern void radeonPrintGlobalLRU( radeonContextPtr rmesa, int heap );
+extern void radeonUpdateTexLRU(radeonContextPtr rmesa, radeonTexObjPtr t );
 
 extern void radeonDDInitTextureFuncs( GLcontext *ctx );
 
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_texmem.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_texmem.c
new file mode 100644
index 000000000..13f6d4b59
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_texmem.c
@@ -0,0 +1,508 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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
+on 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
+ATI, VA LINUX SYSTEMS AND/OR THEIR 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.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <martin@valinux.com>
+ *   Gareth Hughes <gareth@valinux.com>
+ *
+ */
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+#include "radeon_vb.h"
+#include "radeon_tex.h"
+
+#include "mmath.h"
+#include "simple_list.h"
+#include "enums.h"
+#include "mem.h"
+
+
+/* Destroy hardware state associated with texture `t'.
+ */
+void radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t )
+{
+#if ENABLE_PERF_BOXES
+   /* Bump the performace counter */
+   rmesa->c_textureSwaps++;
+#endif
+   if ( !t ) return;
+
+   if ( t->memBlock ) {
+      mmFreeMem( t->memBlock );
+      t->memBlock = NULL;
+   }
+
+   if ( t->tObj )
+      t->tObj->DriverData = NULL;
+
+   if ( t->bound & TEX_0 ) rmesa->CurrentTexObj[0] = NULL;
+   if ( t->bound & TEX_1 ) rmesa->CurrentTexObj[1] = NULL;
+
+   remove_from_list( t );
+   FREE( t );
+}
+
+/* Keep track of swapped out texture objects.
+ */
+void radeonSwapOutTexObj( radeonContextPtr rmesa, radeonTexObjPtr t )
+{
+#if ENABLE_PERF_BOXES
+   /* Bump the performace counter */
+   rmesa->c_textureSwaps++;
+#endif
+   if ( t->memBlock ) {
+      mmFreeMem( t->memBlock );
+      t->memBlock = NULL;
+   }
+
+   t->dirty_images = ~0;
+   move_to_tail( &rmesa->SwappedOut, t );
+}
+
+/* Print out debugging information about texture LRU.
+ */
+void radeonPrintLocalLRU( radeonContextPtr rmesa, int heap )
+{
+   radeonTexObjPtr t;
+   int sz = 1 << (rmesa->radeonScreen->logTexGranularity[heap]);
+
+   fprintf( stderr, "\nLocal LRU, heap %d:\n", heap );
+
+   foreach ( t, &rmesa->TexObjList[heap] ) {
+      if (!t->tObj) {
+	 fprintf( stderr, "Placeholder %d at 0x%x sz 0x%x\n",
+		  t->memBlock->ofs / sz,
+		  t->memBlock->ofs,
+		  t->memBlock->size );
+      } else {
+	 fprintf( stderr, "Texture (bound %d) at 0x%x sz 0x%x\n",
+		  t->bound,
+		  t->memBlock->ofs,
+		  t->memBlock->size );
+      }
+   }
+
+   fprintf( stderr, "\n" );
+}
+
+void radeonPrintGlobalLRU( radeonContextPtr rmesa, int heap )
+{
+   radeon_tex_region_t *list = rmesa->sarea->texList[heap];
+   int i, j;
+
+   fprintf( stderr, "\nGlobal LRU, heap %d list %p:\n", heap, list );
+
+   for ( i = 0, j = RADEON_NR_TEX_REGIONS ; i < RADEON_NR_TEX_REGIONS ; i++ ) {
+      fprintf( stderr, "list[%d] age %d next %d prev %d\n",
+	       j, list[j].age, list[j].next, list[j].prev );
+      j = list[j].next;
+      if ( j == RADEON_NR_TEX_REGIONS ) break;
+   }
+
+   if ( j != RADEON_NR_TEX_REGIONS ) {
+      fprintf( stderr, "Loop detected in global LRU\n" );
+      for ( i = 0 ; i < RADEON_NR_TEX_REGIONS ; i++ ) {
+	 fprintf( stderr, "list[%d] age %d next %d prev %d\n",
+		  i, list[i].age, list[i].next, list[i].prev );
+      }
+   }
+
+   fprintf( stderr, "\n" );
+}
+
+/* Reset the global texture LRU.
+ */
+static void radeonResetGlobalLRU( radeonContextPtr rmesa, int heap )
+{
+   radeon_tex_region_t *list = rmesa->sarea->texList[heap];
+   int sz = 1 << rmesa->radeonScreen->logTexGranularity[heap];
+   int i;
+
+   /*
+    * (Re)initialize the global circular LRU list.  The last element in
+    * the array (RADEON_NR_TEX_REGIONS) is the sentinal.  Keeping it at
+    * the end of the array allows it to be addressed rationally when
+    * looking up objects at a particular location in texture memory.
+    */
+   for ( i = 0 ; (i+1) * sz <= rmesa->radeonScreen->texSize[heap] ; i++ ) {
+      list[i].prev = i-1;
+      list[i].next = i+1;
+      list[i].age = 0;
+   }
+
+   i--;
+   list[0].prev = RADEON_NR_TEX_REGIONS;
+   list[i].prev = i-1;
+   list[i].next = RADEON_NR_TEX_REGIONS;
+   list[RADEON_NR_TEX_REGIONS].prev = i;
+   list[RADEON_NR_TEX_REGIONS].next = 0;
+   rmesa->sarea->texAge[heap] = 0;
+}
+
+/* Update the local and glock texture LRUs.
+ */
+void radeonUpdateTexLRU(radeonContextPtr rmesa, radeonTexObjPtr t )
+{
+   int heap = t->heap;
+   radeon_tex_region_t *list = rmesa->sarea->texList[heap];
+   int sz = rmesa->radeonScreen->logTexGranularity[heap];
+   int start = t->memBlock->ofs >> sz;
+   int end = (t->memBlock->ofs + t->memBlock->size-1) >> sz;
+   int i;
+
+   rmesa->lastTexAge[heap] = ++rmesa->sarea->texAge[heap];
+
+   if ( !t->memBlock ) {
+      fprintf( stderr, "no memblock\n\n" );
+      return;
+   }
+
+   /* Update our local LRU */
+   move_to_head( &rmesa->TexObjList[heap], t );
+
+   /* Update the global LRU */
+   for ( i = start ; i <= end ; i++ ) {
+      list[i].in_use = 1;
+      list[i].age = rmesa->lastTexAge[heap];
+
+      /* remove_from_list(i) */
+      list[(CARD32)list[i].next].prev = list[i].prev;
+      list[(CARD32)list[i].prev].next = list[i].next;
+
+      /* insert_at_head(list, i) */
+      list[i].prev = RADEON_NR_TEX_REGIONS;
+      list[i].next = list[RADEON_NR_TEX_REGIONS].next;
+      list[(CARD32)list[RADEON_NR_TEX_REGIONS].next].prev = i;
+      list[RADEON_NR_TEX_REGIONS].next = i;
+   }
+
+   if ( 0 ) {
+      radeonPrintGlobalLRU( rmesa, t->heap );
+      radeonPrintLocalLRU( rmesa, t->heap );
+   }
+}
+
+/* Update our notion of what textures have been changed since we last
+ * held the lock.  This pertains to both our local textures and the
+ * textures belonging to other clients.  Keep track of other client's
+ * textures by pushing a placeholder texture onto the LRU list -- these
+ * are denoted by (tObj == NULL).
+ */
+static void radeonTexturesGone( radeonContextPtr rmesa, int heap,
+				int offset, int size, int in_use )
+{
+   radeonTexObjPtr t, tmp;
+
+   foreach_s ( t, tmp, &rmesa->TexObjList[heap] ) {
+      if ( t->memBlock->ofs >= offset + size ||
+	   t->memBlock->ofs + t->memBlock->size <= offset )
+	 continue;
+
+      /* It overlaps - kick it out.  Need to hold onto the currently
+       * bound objects, however.
+       */
+      if ( t->bound ) {
+	 radeonSwapOutTexObj( rmesa, t );
+      } else {
+	 radeonDestroyTexObj( rmesa, t );
+      }
+   }
+
+   if ( in_use ) {
+      t = (radeonTexObjPtr) CALLOC( sizeof(*t) );
+      if ( !t ) return;
+
+      t->memBlock = mmAllocMem( rmesa->texHeap[heap], size, 0, offset );
+      if ( !t->memBlock ) {
+	 fprintf( stderr, "Couldn't alloc placeholder sz %x ofs %x\n",
+		  (int)size, (int)offset );
+	 mmDumpMemInfo( rmesa->texHeap[heap] );
+	 return;
+      }
+      insert_at_head( &rmesa->TexObjList[heap], t );
+   }
+}
+
+/* Update our client's shared texture state.  If another client has
+ * modified a region in which we have textures, then we need to figure
+ * out which of our textures has been removed, and update our global
+ * LRU.
+ */
+void radeonAgeTextures( radeonContextPtr rmesa, int heap )
+{
+   RADEONSAREAPrivPtr sarea = rmesa->sarea;
+
+   if ( sarea->texAge[heap] != rmesa->lastTexAge[heap] ) {
+      int sz = 1 << rmesa->radeonScreen->logTexGranularity[heap];
+      int nr = 0;
+      int idx;
+
+      for ( idx = sarea->texList[heap][RADEON_NR_TEX_REGIONS].prev ;
+	    idx != RADEON_NR_TEX_REGIONS && nr < RADEON_NR_TEX_REGIONS ;
+	    idx = sarea->texList[heap][idx].prev, nr++ )
+      {
+	 /* If switching texturing schemes, then the SAREA might not
+	  * have been properly cleared, so we need to reset the
+	  * global texture LRU.
+	  */
+	 if ( idx * sz > rmesa->radeonScreen->texSize[heap] ) {
+	    nr = RADEON_NR_TEX_REGIONS;
+	    break;
+	 }
+
+	 if ( sarea->texList[heap][idx].age > rmesa->lastTexAge[heap] ) {
+	    radeonTexturesGone( rmesa, heap, idx * sz, sz,
+				sarea->texList[heap][idx].in_use );
+	 }
+      }
+
+      if ( nr == RADEON_NR_TEX_REGIONS ) {
+	 radeonTexturesGone( rmesa, heap, 0,
+			     rmesa->radeonScreen->texSize[heap], 0 );
+	 radeonResetGlobalLRU( rmesa, heap );
+      }
+
+      rmesa->dirty |= (RADEON_UPLOAD_CONTEXT |
+		       RADEON_UPLOAD_TEX0IMAGES |
+		       RADEON_UPLOAD_TEX1IMAGES);
+      rmesa->lastTexAge[heap] = sarea->texAge[heap];
+   }
+}
+
+
+/* ================================================================
+ * Texture image uploads
+ */
+
+/* Upload the texture image associated with texture `t' at level `level'
+ * at the address relative to `start'.
+ */
+static void radeonUploadSubImage( radeonContextPtr rmesa,
+				  radeonTexObjPtr t, GLint level,
+				  GLint x, GLint y, GLint width, GLint height )
+{
+   struct gl_texture_image *texImage;
+   const struct gl_texture_format *texFormat;
+   GLint texelsPerDword = 0;
+   GLint imageX, imageY, imageWidth, imageHeight;
+   GLint blitX, blitY, blitWidth, blitHeight;
+   GLint imageRows, blitRows;
+   GLint remaining;
+   GLint format, dwords;
+   CARD32 pitch, offset;
+   drmBufPtr buffer;
+   CARD32 *dst;
+   GLint ret;
+
+   /* Ensure we have a valid texture to upload */
+   texImage = t->tObj->Image[level];
+   if ( !texImage )
+      return;
+
+   texFormat = texImage->TexFormat;
+
+   switch ( texFormat->TexelBytes ) {
+   case 1:
+      texelsPerDword = 4;
+      break;
+   case 2:
+      texelsPerDword = 2;
+      break;
+   case 4:
+      texelsPerDword = 1;
+      break;
+   }
+
+   format = t->pp_txformat & RADEON_TXFORMAT_FORMAT_MASK;
+
+   imageX = 0;
+   imageY = 0;
+   imageWidth = texImage->Width;
+   imageHeight = texImage->Height;
+
+   blitX = t->image[level].x;
+   blitY = t->image[level].y;
+   blitWidth = t->image[level].width;
+   blitHeight = t->image[level].height;
+
+   /* dwords = t->image[level].dwords; */
+   offset = t->bufAddr;
+   pitch = (t->image[0].width * texFormat->TexelBytes) / 64;
+
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_MSG ) {
+      fprintf( stderr, "   upload image: %d,%d at %d,%d\n",
+	       imageWidth, imageHeight, imageX, imageY );
+      fprintf( stderr, "   upload  blit: %d,%d at %d,%d\n",
+	       blitWidth, blitHeight, blitX, blitY );
+      fprintf( stderr, "       blit ofs: 0x%07x pitch: 0x%x "
+	       "level: %d format: %x\n",
+	       (GLuint)offset, (GLuint)pitch, level, format );
+   }
+
+   ret = drmRadeonLoadTexture( rmesa->driFd, offset, pitch, format,
+			       imageWidth, imageHeight, &t->image[level] );
+
+   if ( ret ) {
+      UNLOCK_HARDWARE( rmesa );
+      fprintf( stderr, "drmRadeonTextureBlit: return = %d\n", ret );
+      exit( 1 );
+   }
+
+   rmesa->new_state |= RADEON_NEW_CONTEXT;
+   rmesa->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_MASKS;
+}
+
+/* Upload the texture images associated with texture `t'.  This might
+ * require removing our own and/or other client's texture objects to
+ * make room for these images.
+ */
+int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t )
+{
+   int i;
+   int heap;
+
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p, %p )\n",
+	       __FUNCTION__, rmesa->glCtx, t->tObj );
+   }
+
+   if ( !t )
+      return 0;
+
+   /* Choose the heap appropriately */
+   heap = t->heap = RADEON_CARD_HEAP;
+#if 0
+   if ( !rmesa->radeonScreen->IsPCI &&
+	t->totalSize > rmesa->radeonScreen->texSize[heap] ) {
+      heap = t->heap = RADEON_AGP_HEAP;
+   }
+#endif
+
+   /* Do we need to eject LRU texture objects? */
+   if ( !t->memBlock ) {
+      /* Allocate a memory block on a 4k boundary (1<<12 == 4096) */
+      t->memBlock = mmAllocMem( rmesa->texHeap[heap],
+				t->totalSize, 12, 0 );
+
+#if 0
+      /* Try AGP before kicking anything out of local mem */
+      if ( !t->memBlock && heap == RADEON_CARD_HEAP ) {
+	 t->memBlock = mmAllocMem( rmesa->texHeap[RADEON_AGP_HEAP],
+				   t->totalSize, 12, 0 );
+
+	 if ( t->memBlock )
+	    heap = t->heap = RADEON_AGP_HEAP;
+      }
+#endif
+
+      /* Kick out textures until the requested texture fits */
+      while ( !t->memBlock ) {
+	 if ( rmesa->TexObjList[heap].prev->bound ) {
+	    fprintf( stderr,
+		     "radeonUploadTexImages: ran into bound texture\n" );
+	    return -1;
+	 }
+	 if ( rmesa->TexObjList[heap].prev ==
+	      &rmesa->TexObjList[heap] ) {
+	    if ( rmesa->radeonScreen->IsPCI ) {
+	       fprintf( stderr, "radeonUploadTexImages: upload texture "
+			"failure on local texture heaps, sz=%d\n",
+			t->totalSize );
+	       return -1;
+#if 0
+	    } else if ( heap == RADEON_CARD_HEAP ) {
+	       heap = t->heap = RADEON_AGP_HEAP;
+	       continue;
+#endif
+	    } else {
+	       fprintf( stderr, "radeonUploadTexImages: upload texture "
+			"failure on both local and AGP texture heaps, "
+			"sz=%d\n",
+			t->totalSize );
+	       return -1;
+	    }
+	 }
+
+	 radeonDestroyTexObj( rmesa, rmesa->TexObjList[heap].prev );
+
+	 t->memBlock = mmAllocMem( rmesa->texHeap[heap],
+				   t->totalSize, 12, 0 );
+      }
+
+      /* Set the base offset of the texture image */
+      t->bufAddr = rmesa->radeonScreen->texOffset[heap] + t->memBlock->ofs;
+
+      t->pp_txoffset = t->bufAddr;
+#if 0
+      /* Fix AGP texture offsets */
+      if ( heap == RADEON_AGP_HEAP ) {
+	  t->setup.pp_tx_offset += RADEON_AGP_TEX_OFFSET +
+	      rmesa->radeonScreen->agpTexOffset;
+      }
+#endif
+
+      /* Force loading the new state into the hardware */
+      switch ( t->bound ) {
+      case 1:
+	 rmesa->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_TEX0;
+	 break;
+
+      case 2:
+	 rmesa->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_TEX1;
+	 break;
+
+      default:
+	 return -1;
+      }
+   }
+
+   /* Let the world know we've used this memory recently */
+   radeonUpdateTexLRU( rmesa, t );
+
+   /* Upload any images that are new */
+   if ( t->dirty_images ) {
+      int levels = ((t->pp_txfilter & RADEON_MAX_MIP_LEVEL_MASK) >>
+		    RADEON_MAX_MIP_LEVEL_SHIFT);
+
+      for ( i = 0 ; i <= levels ; i++ ) {
+	 if ( t->dirty_images & (1 << i) ) {
+	    radeonUploadSubImage( rmesa, t, i, 0, 0,
+				  t->image[i].width, t->image[i].height );
+	 }
+      }
+
+      rmesa->dirty |= RADEON_UPLOAD_CONTEXT;
+   }
+
+   t->dirty_images = 0;
+   return 0;
+}
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_texobj.h b/xc/lib/GL/mesa/src/drv/radeon/radeon_texobj.h
index b63d5c6a7..7cec84e9a 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_texobj.h
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_texobj.h
@@ -40,17 +40,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_sarea.h"
 #include "mm.h"
 
-/* Handle the Radeon's tightly packed mipmaps and strict offset,
- * pitch rules for blits by assigning each mipmap a set of
- * coordinates that can be used for a hostdata blit.
- */
-typedef struct {
-   GLuint x;				/* Blit coordinates */
-   GLuint y;
-   GLuint width;			/* Blit dimensions */
-   GLuint height;
-   GLuint dwords;			/* Size of image level */
-} radeonTexImage;
+#define TEX_0 1
+#define TEX_1 2
 
 typedef struct radeon_tex_obj radeonTexObj, *radeonTexObjPtr;
 
@@ -72,16 +63,15 @@ struct radeon_tex_obj {
    GLint bound;				/* Texture unit currently bound to */
    GLint heap;				/* Texture heap currently stored in */
 
-   radeonTexImage image[RADEON_MAX_TEXTURE_LEVELS]; /* Image data for all
-						      mipmap levels */
+   drmRadeonTexImage image[RADEON_MAX_TEXTURE_LEVELS];
 
    GLint totalSize;			/* Total size of the texture
 					   including all mipmap levels */
-   GLint texelBytes;			/* Number of bytes per texel */
-
-   GLboolean hasAlpha;
 
-   radeon_texture_regs_t setup;		/* Setup regs for texture */
+   GLuint pp_txfilter;		        /* Hardware register values */
+   GLuint pp_txformat;
+   GLuint pp_txoffset;
+   GLuint pp_border_color;
 };
 
 #endif /* __RADEON_TEXOBJ_H__ */
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_texstate.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_texstate.c
new file mode 100644
index 000000000..2a076717e
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_texstate.c
@@ -0,0 +1,1103 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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
+on 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
+ATI, VA LINUX SYSTEMS AND/OR THEIR 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.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <martin@valinux.com>
+ *   Gareth Hughes <gareth@valinux.com>
+ *
+ */
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+#include "radeon_vb.h"
+#include "radeon_tex.h"
+
+#include "mmath.h"
+#include "simple_list.h"
+#include "enums.h"
+#include "mem.h"
+
+static void radeonSetTexImages( radeonContextPtr rmesa,
+				struct gl_texture_object *tObj )
+{
+   radeonTexObjPtr t = (radeonTexObjPtr)tObj->DriverData;
+   struct gl_texture_image *texImage = tObj->Image[0];
+   const struct gl_texture_format *texFormat = texImage->TexFormat;
+   GLint log2Width, log2Height, log2Size;
+   GLint totalSize;
+   GLint texelsPerDword = 0, blitWidth = 0, blitPitch = 0;
+   GLint x, y, width, height;
+   GLint i;
+
+   /* Calculate dimensions in log domain.
+    */
+   for ( i = 1, log2Height = 0 ; i < texImage->Height ; i *= 2 ) {
+      log2Height++;
+   }
+   for ( i = 1, log2Width = 0 ;  i < texImage->Width ;  i *= 2 ) {
+      log2Width++;
+   }
+   log2Size = MAX2( log2Width, log2Height );
+
+   /* The Radeon has a 64-byte minimum pitch for all blits.  We
+    * calculate the equivalent number of texels to simplify the
+    * calculation of the texture image area.
+    */
+   switch ( texFormat->TexelBytes ) {
+   case 4:
+      texelsPerDword = 1;
+      blitPitch = 16;
+      break;
+   case 2:
+      texelsPerDword = 2;
+      blitPitch = 32;
+      break;
+   case 1:
+      texelsPerDword = 4;
+      blitPitch = 64;
+      break;
+   }
+
+   /* Select the larger of the two widths for our global texture image
+    * coordinate space.  As the Radeon has very strict offset rules, we
+    * can't upload mipmaps directly and have to reference their location
+    * from the aligned start of the whole image.
+    */
+   blitWidth = MAX2( texImage->Width, blitPitch );
+
+   /* Calculate mipmap offsets and dimensions.
+    */
+   totalSize = 0;
+   x = 0;
+   y = 0;
+
+   for ( i = 0 ; i <= log2Size ; i++ ) {
+      GLuint size;
+
+      texImage = tObj->Image[i];
+      if ( !texImage )
+	 break;
+
+      width = texImage->Width;
+      height = texImage->Height;
+
+      /* Texture images have a minimum pitch of 32 bytes (half of the
+       * 64-byte minimum pitch for blits).  For images that have a
+       * width smaller than this, we must pad each texture image
+       * scanline out to this amount.
+       */
+      if ( width < blitPitch / 2 ) {
+	 width = blitPitch / 2;
+      }
+
+      size = width * height * texFormat->TexelBytes;
+      totalSize += size;
+      ASSERT( (totalSize & 31) == 0 );
+
+      while ( width < blitWidth && height > 1 ) {
+	 width *= 2;
+	 height /= 2;
+      }
+
+      t->image[i].x = x;
+      t->image[i].y = y;
+
+      t->image[i].width  = width;
+      t->image[i].height = height;
+
+      /* While blits must have a pitch of at least 64 bytes, mipmaps
+       * must be aligned on a 32-byte boundary (just like each texture
+       * image scanline).
+       */
+      if ( width >= blitWidth ) {
+	 y += height;
+      } else {
+	 x += width;
+	 if ( x >= blitWidth ) {
+	    x = 0;
+	    y++;
+	 }
+      }
+
+      if ( 0 ) {
+	 fprintf( stderr, "level=%d p=%d   %dx%d -> %dx%d at (%d,%d)\n",
+		  i, blitWidth, texImage->Width, texImage->Height,
+		  t->image[i].width, t->image[i].height,
+		  t->image[i].x, t->image[i].y );
+      }
+   }
+
+   /* Align the total size of texture memory block.
+    */
+   t->totalSize = (totalSize + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
+
+   /* Hardware state:
+    */
+   t->pp_txfilter &= ~RADEON_MAX_MIP_LEVEL_MASK;
+   t->pp_txfilter |= i << RADEON_MAX_MIP_LEVEL_SHIFT;
+
+   t->pp_txformat &= ~(RADEON_TXFORMAT_WIDTH_MASK |
+		       RADEON_TXFORMAT_HEIGHT_MASK);
+   t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_WIDTH_SHIFT) |
+		      (log2Height << RADEON_TXFORMAT_HEIGHT_SHIFT));
+}
+
+
+/* ================================================================
+ * Texture combine functions
+ */
+
+#define RADEON_DISABLE		0
+#define RADEON_REPLACE		1
+#define RADEON_MODULATE		2
+#define RADEON_DECAL		3
+#define RADEON_BLEND		4
+#define RADEON_ADD		5
+#define RADEON_MAX_COMBFUNC	6
+
+static GLuint radeon_color_combine[][RADEON_MAX_COMBFUNC] =
+{
+   /* Unit 0:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_CURRENT_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00802800
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T0_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800142
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T0_COLOR |
+       RADEON_COLOR_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x008c2d42
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T0_COLOR |
+       RADEON_COLOR_ARG_C_T0_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x008c2902
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
+       RADEON_COLOR_ARG_C_T0_COLOR |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00812802
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T0_COLOR |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+   /* Unit 1:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_CURRENT_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00803000
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T1_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800182
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T1_COLOR |
+       RADEON_COLOR_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x008c3582
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T1_COLOR |
+       RADEON_COLOR_ARG_C_T1_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x008c3102
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
+       RADEON_COLOR_ARG_C_T1_COLOR |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00813002
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T1_COLOR |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+   /* Unit 2:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_CURRENT_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00803800
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T2_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x008001c2
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T2_COLOR |
+       RADEON_COLOR_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x008c3dc2
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T2_COLOR |
+       RADEON_COLOR_ARG_C_T2_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x008c3902
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
+       RADEON_COLOR_ARG_C_T2_COLOR |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00813802
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T2_COLOR |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   }
+};
+
+static GLuint radeon_alpha_combine[][RADEON_MAX_COMBFUNC] =
+{
+   /* Unit 0:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00800500
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T0_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800051
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_T0_ALPHA |
+       RADEON_ALPHA_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x00800100
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x00800051
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
+       RADEON_ALPHA_ARG_C_T0_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00800051
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T0_ALPHA |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+   /* Unit 1:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00800600
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T1_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800061
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_T1_ALPHA |
+       RADEON_ALPHA_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x00800100
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x00800061
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
+       RADEON_ALPHA_ARG_C_T1_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00800061
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T1_ALPHA |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+   /* Unit 2:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00800700
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T2_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800071
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_T2_ALPHA |
+       RADEON_ALPHA_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x00800100
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x00800071
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
+       RADEON_ALPHA_ARG_C_T2_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00800021
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T2_ALPHA |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   }
+};
+
+
+/* GL_EXT_texture_env_combine support
+ */
+
+/* The color tables have combine functions for GL_SRC_COLOR,
+ * GL_ONE_MINUS_SRC_COLOR, GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
+ */
+static GLuint radeon_texture_color[][RADEON_MAX_TEXTURE_UNITS] =
+{
+   {
+      RADEON_COLOR_ARG_A_T0_COLOR,
+      RADEON_COLOR_ARG_A_T1_COLOR,
+      RADEON_COLOR_ARG_A_T2_COLOR
+   },
+   {
+      RADEON_COLOR_ARG_A_T0_COLOR | RADEON_COMP_ARG_A,
+      RADEON_COLOR_ARG_A_T1_COLOR | RADEON_COMP_ARG_A,
+      RADEON_COLOR_ARG_A_T2_COLOR | RADEON_COMP_ARG_A
+   },
+   {
+      RADEON_COLOR_ARG_A_T0_ALPHA,
+      RADEON_COLOR_ARG_A_T1_ALPHA,
+      RADEON_COLOR_ARG_A_T2_ALPHA
+   },
+   {
+      RADEON_COLOR_ARG_A_T0_ALPHA | RADEON_COMP_ARG_A,
+      RADEON_COLOR_ARG_A_T1_ALPHA | RADEON_COMP_ARG_A,
+      RADEON_COLOR_ARG_A_T2_ALPHA | RADEON_COMP_ARG_A
+   },
+};
+
+static GLuint radeon_tfactor_color[] =
+{
+   RADEON_COLOR_ARG_A_TFACTOR_COLOR,
+   RADEON_COLOR_ARG_A_TFACTOR_COLOR | RADEON_COMP_ARG_A,
+   RADEON_COLOR_ARG_A_TFACTOR_ALPHA,
+   RADEON_COLOR_ARG_A_TFACTOR_ALPHA | RADEON_COMP_ARG_A
+};
+
+static GLuint radeon_primary_color[] =
+{
+   RADEON_COLOR_ARG_A_DIFFUSE_COLOR,
+   RADEON_COLOR_ARG_A_DIFFUSE_COLOR | RADEON_COMP_ARG_A,
+   RADEON_COLOR_ARG_A_DIFFUSE_ALPHA,
+   RADEON_COLOR_ARG_A_DIFFUSE_ALPHA | RADEON_COMP_ARG_A
+};
+
+static GLuint radeon_previous_color[] =
+{
+   RADEON_COLOR_ARG_A_CURRENT_COLOR,
+   RADEON_COLOR_ARG_A_CURRENT_COLOR | RADEON_COMP_ARG_A,
+   RADEON_COLOR_ARG_A_CURRENT_ALPHA,
+   RADEON_COLOR_ARG_A_CURRENT_ALPHA | RADEON_COMP_ARG_A
+};
+
+/* The alpha tables only have GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
+ */
+static GLuint radeon_texture_alpha[][RADEON_MAX_TEXTURE_UNITS] =
+{
+   {
+      RADEON_ALPHA_ARG_A_T0_ALPHA,
+      RADEON_ALPHA_ARG_A_T1_ALPHA,
+      RADEON_ALPHA_ARG_A_T2_ALPHA
+   },
+   {
+      RADEON_ALPHA_ARG_A_T0_ALPHA | RADEON_COMP_ARG_A,
+      RADEON_ALPHA_ARG_A_T1_ALPHA | RADEON_COMP_ARG_A,
+      RADEON_ALPHA_ARG_A_T2_ALPHA | RADEON_COMP_ARG_A
+   },
+};
+
+static GLuint radeon_tfactor_alpha[] =
+{
+   RADEON_ALPHA_ARG_A_TFACTOR_ALPHA,
+   RADEON_ALPHA_ARG_A_TFACTOR_ALPHA | RADEON_COMP_ARG_A
+};
+
+static GLuint radeon_primary_alpha[] =
+{
+   RADEON_ALPHA_ARG_A_DIFFUSE_ALPHA,
+   RADEON_ALPHA_ARG_A_DIFFUSE_ALPHA | RADEON_COMP_ARG_A
+};
+
+static GLuint radeon_previous_alpha[] =
+{
+   RADEON_ALPHA_ARG_A_CURRENT_ALPHA,
+   RADEON_ALPHA_ARG_A_CURRENT_ALPHA | RADEON_COMP_ARG_A
+};
+
+
+/* Extract the arg from slot A, shift it into the correct argument slot
+ * and set the corresponding complement bit.
+ */
+#define RADEON_COLOR_ARG( n, arg )					\
+do {									\
+   color_combine |=							\
+      ((color_arg[n] & RADEON_COLOR_ARG_MASK)				\
+       << RADEON_COLOR_ARG_##arg##_SHIFT);				\
+   color_combine |=							\
+      ((color_arg[n] >> RADEON_COMP_ARG_SHIFT)				\
+       << RADEON_COMP_ARG_##arg##_SHIFT);				\
+} while (0)
+
+#define RADEON_ALPHA_ARG( n, arg )					\
+do {									\
+   alpha_combine |=							\
+      ((alpha_arg[n] & RADEON_ALPHA_ARG_MASK)				\
+       << RADEON_ALPHA_ARG_##arg##_SHIFT);				\
+   alpha_combine |=							\
+      ((alpha_arg[n] >> RADEON_COMP_ARG_SHIFT)				\
+       << RADEON_COMP_ARG_##arg##_SHIFT);				\
+} while (0)
+
+
+/* ================================================================
+ * Texture unit state management
+ */
+
+static void radeonUpdateTextureEnv( GLcontext *ctx, int unit )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   int source = rmesa->tmu_source[unit];
+   struct gl_texture_object *tObj;
+   struct gl_texture_unit *texUnit;
+   GLuint enabled;
+   GLuint color_combine, alpha_combine;
+   GLuint color_arg[3], alpha_arg[3];
+   GLuint i, numColorArgs = 0, numAlphaArgs = 0;
+   GLuint op;
+
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_MSG ) {
+      fprintf( stderr, "%s( %p, %d )\n",
+	       __FUNCTION__, ctx, unit );
+   }
+
+   enabled = (ctx->Texture.ReallyEnabled >> (source * 4)) & TEXTURE0_ANY;
+   if ( enabled != TEXTURE0_2D && enabled != TEXTURE0_1D )
+      return;
+
+   /* Only update the hardware texture state if the texture is current,
+    * complete and enabled.
+    */
+   texUnit = &ctx->Texture.Unit[source];
+   tObj = texUnit->Current;
+   if ( !tObj || !tObj->Complete )
+      return;
+
+   if ( ( tObj != texUnit->CurrentD[2] ) &&
+	( tObj != texUnit->CurrentD[1] ) )
+      return;
+
+   /* Set the texture environment state.  Isn't this nice and clean?
+    * The Radeon will automagically set the texture alpha to 0xff when
+    * the texture format does not include an alpha component.  This
+    * reduces the amount of special-casing we have to do, alpha-only
+    * textures being a notable exception.
+    */
+   switch ( texUnit->EnvMode ) {
+   case GL_REPLACE:
+      switch ( tObj->Image[0]->Format ) {
+      case GL_RGBA:
+      case GL_RGB:
+      case GL_LUMINANCE:
+      case GL_LUMINANCE_ALPHA:
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_REPLACE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_REPLACE];
+	 break;
+      case GL_ALPHA:
+	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_REPLACE];
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return;
+      }
+      break;
+
+   case GL_MODULATE:
+      switch ( tObj->Image[0]->Format ) {
+      case GL_RGBA:
+      case GL_RGB:
+      case GL_LUMINANCE:
+      case GL_LUMINANCE_ALPHA:
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_MODULATE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_ALPHA:
+	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return;
+      }
+      break;
+
+   case GL_DECAL:
+      switch ( tObj->Image[0]->Format ) {
+      case GL_RGBA:
+      case GL_RGB:
+	 color_combine = radeon_color_combine[unit][RADEON_DECAL];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+	 break;
+      case GL_ALPHA:
+      case GL_LUMINANCE:
+      case GL_LUMINANCE_ALPHA:
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return;
+      }
+      break;
+
+   case GL_BLEND:
+      switch ( tObj->Image[0]->Format ) {
+      case GL_RGBA:
+      case GL_RGB:
+      case GL_LUMINANCE:
+      case GL_LUMINANCE_ALPHA:
+	 color_combine = radeon_color_combine[unit][RADEON_BLEND];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_ALPHA:
+	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_BLEND];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_BLEND];
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return;
+      }
+      break;
+
+   case GL_ADD:
+      switch ( tObj->Image[0]->Format ) {
+      case GL_RGBA:
+      case GL_RGB:
+      case GL_LUMINANCE:
+      case GL_LUMINANCE_ALPHA:
+	 color_combine = radeon_color_combine[unit][RADEON_ADD];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_ALPHA:
+	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_ADD];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_ADD];
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return;
+      }
+      break;
+
+   case GL_COMBINE_EXT:
+      /* Step 0:
+       * Calculate how many arguments we need to process.
+       */
+      switch ( texUnit->CombineModeRGB ) {
+      case GL_REPLACE:
+	 numColorArgs = 1;
+	 break;
+      case GL_MODULATE:
+      case GL_ADD:
+      case GL_ADD_SIGNED_EXT:
+      case GL_DOT3_RGB_EXT:
+      case GL_DOT3_RGBA_EXT:
+	 numColorArgs = 2;
+	 break;
+      case GL_INTERPOLATE_EXT:
+	 numColorArgs = 3;
+	 break;
+      default:
+	 return;
+      }
+
+      switch ( texUnit->CombineModeA ) {
+      case GL_REPLACE:
+	 numAlphaArgs = 1;
+	 break;
+      case GL_MODULATE:
+      case GL_ADD:
+      case GL_ADD_SIGNED_EXT:
+	 numAlphaArgs = 2;
+	 break;
+      case GL_INTERPOLATE_EXT:
+	 numAlphaArgs = 3;
+	 break;
+      default:
+	 return;
+      }
+
+      /* Step 1:
+       * Extract the color and alpha combine function arguments.
+       */
+      for ( i = 0 ; i < numColorArgs ; i++ ) {
+	 op = texUnit->CombineOperandRGB[i] - GL_SRC_COLOR;
+	 switch ( texUnit->CombineSourceRGB[i] ) {
+	 case GL_TEXTURE:
+	    color_arg[i] = radeon_texture_color[op][unit];
+	    break;
+	 case GL_CONSTANT_EXT:
+	    color_arg[i] = radeon_tfactor_color[op];
+	    break;
+	 case GL_PRIMARY_COLOR_EXT:
+	    color_arg[i] = radeon_primary_color[op];
+	    break;
+	 case GL_PREVIOUS_EXT:
+	    color_arg[i] = radeon_previous_color[op];
+	    break;
+	 default:
+	    return;
+	 }
+      }
+
+      for ( i = 0 ; i < numAlphaArgs ; i++ ) {
+	 op = texUnit->CombineOperandA[i] - GL_SRC_ALPHA;
+	 switch ( texUnit->CombineSourceA[i] ) {
+	 case GL_TEXTURE:
+	    alpha_arg[i] = radeon_texture_alpha[op][unit];
+	    break;
+	 case GL_CONSTANT_EXT:
+	    alpha_arg[i] = radeon_tfactor_alpha[op];
+	    break;
+	 case GL_PRIMARY_COLOR_EXT:
+	    alpha_arg[i] = radeon_primary_alpha[op];
+	    break;
+	 case GL_PREVIOUS_EXT:
+	    alpha_arg[i] = radeon_previous_alpha[op];
+	    break;
+	 default:
+	    return;
+	 }
+      }
+
+      /* Step 2:
+       * Build up the color and alpha combine functions.
+       */
+      switch ( texUnit->CombineModeRGB ) {
+      case GL_REPLACE:
+	 color_combine = (RADEON_COLOR_ARG_A_ZERO |
+			  RADEON_COLOR_ARG_B_ZERO |
+			  RADEON_BLEND_CTL_ADD |
+			  RADEON_CLAMP_TX);
+	 RADEON_COLOR_ARG( 0, C );
+	 break;
+      case GL_MODULATE:
+	 color_combine = (RADEON_COLOR_ARG_C_ZERO |
+			  RADEON_BLEND_CTL_ADD |
+			  RADEON_CLAMP_TX);
+	 RADEON_COLOR_ARG( 0, A );
+	 RADEON_COLOR_ARG( 1, B );
+	 break;
+      case GL_ADD:
+	 color_combine = (RADEON_COLOR_ARG_B_ZERO |
+			  RADEON_COMP_ARG_B |
+			  RADEON_BLEND_CTL_ADD |
+			  RADEON_CLAMP_TX);
+	 RADEON_COLOR_ARG( 0, A );
+	 RADEON_COLOR_ARG( 1, C );
+	 break;
+      case GL_ADD_SIGNED_EXT:
+	 color_combine = (RADEON_COLOR_ARG_B_ZERO |
+			  RADEON_COMP_ARG_B |
+			  RADEON_BLEND_CTL_ADDSIGNED |
+			  RADEON_CLAMP_TX);
+	 RADEON_COLOR_ARG( 0, A );
+	 RADEON_COLOR_ARG( 1, C );
+	 break;
+      case GL_INTERPOLATE_EXT:
+	 color_combine = (RADEON_BLEND_CTL_BLEND |
+			  RADEON_CLAMP_TX);
+	 RADEON_COLOR_ARG( 0, B );
+	 RADEON_COLOR_ARG( 1, A );
+	 RADEON_COLOR_ARG( 2, C );
+	 break;
+      case GL_DOT3_RGB_EXT:
+      case GL_DOT3_RGBA_EXT:
+	 color_combine = (RADEON_COLOR_ARG_C_ZERO |
+			  RADEON_BLEND_CTL_DOT3 |
+			  RADEON_CLAMP_TX);
+	 RADEON_COLOR_ARG( 0, A );
+	 RADEON_COLOR_ARG( 1, B );
+	 break;
+      default:
+	 return;
+      }
+
+      switch ( texUnit->CombineModeA ) {
+      case GL_REPLACE:
+	 alpha_combine = (RADEON_ALPHA_ARG_A_ZERO |
+			  RADEON_ALPHA_ARG_B_ZERO |
+			  RADEON_BLEND_CTL_ADD |
+			  RADEON_CLAMP_TX);
+	 RADEON_ALPHA_ARG( 0, C );
+	 break;
+      case GL_MODULATE:
+	 alpha_combine = (RADEON_ALPHA_ARG_C_ZERO |
+			  RADEON_BLEND_CTL_ADD |
+			  RADEON_CLAMP_TX);
+	 RADEON_ALPHA_ARG( 0, A );
+	 RADEON_ALPHA_ARG( 1, B );
+	 break;
+      case GL_ADD:
+	 alpha_combine = (RADEON_ALPHA_ARG_B_ZERO |
+			  RADEON_COMP_ARG_B |
+			  RADEON_BLEND_CTL_ADD |
+			  RADEON_CLAMP_TX);
+	 RADEON_ALPHA_ARG( 0, A );
+	 RADEON_ALPHA_ARG( 1, C );
+	 break;
+      case GL_ADD_SIGNED_EXT:
+	 alpha_combine = (RADEON_ALPHA_ARG_B_ZERO |
+			  RADEON_COMP_ARG_B |
+			  RADEON_BLEND_CTL_ADDSIGNED |
+			  RADEON_CLAMP_TX);
+	 RADEON_ALPHA_ARG( 0, A );
+	 RADEON_ALPHA_ARG( 1, C );
+	 break;
+      case GL_INTERPOLATE_EXT:
+	 alpha_combine = (RADEON_BLEND_CTL_BLEND |
+			  RADEON_CLAMP_TX);
+	 RADEON_ALPHA_ARG( 0, B );
+	 RADEON_ALPHA_ARG( 1, A );
+	 RADEON_ALPHA_ARG( 2, C );
+	 break;
+      default:
+	 return;
+      }
+
+      if ( texUnit->CombineModeRGB == GL_DOT3_RGB_EXT ) {
+	 alpha_combine |= RADEON_DOT_ALPHA_DONT_REPLICATE;
+      }
+
+      /* Step 3:
+       * Apply the scale factor.  The EXT extension has a somewhat
+       * unnecessary restriction that the scale must be 4x.  The ARB
+       * extension will likely drop this and we can just apply the
+       * scale factors regardless.
+       */
+      if ( texUnit->CombineModeRGB != GL_DOT3_RGB_EXT &&
+	   texUnit->CombineModeRGB != GL_DOT3_RGBA_EXT ) {
+	 color_combine |= (texUnit->CombineScaleShiftRGB << 21);
+	 alpha_combine |= (texUnit->CombineScaleShiftA << 21);
+      } else {
+	 color_combine |= RADEON_SCALE_4X;
+	 alpha_combine |= RADEON_SCALE_4X;
+      }
+
+      /* All done!
+       */
+      break;
+
+   default:
+      return;
+   }
+
+   rmesa->color_combine[source] = color_combine;
+   rmesa->alpha_combine[source] = alpha_combine;
+}
+
+static void radeonUpdateTextureObject( GLcontext *ctx, int unit )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   int source = rmesa->tmu_source[unit];
+   struct gl_texture_object *tObj;
+   radeonTexObjPtr t;
+   GLuint enabled;
+
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_MSG ) {
+      fprintf( stderr, "%s( %p, %d )\n",
+	       __FUNCTION__, ctx, unit );
+   }
+
+   enabled = (ctx->Texture.ReallyEnabled >> (source * 4)) & TEXTURE0_ANY;
+   if ( enabled != TEXTURE0_2D && enabled != TEXTURE0_1D ) {
+      if ( enabled )
+	 rmesa->Fallback |= RADEON_FALLBACK_TEXTURE;
+      return;
+   }
+
+   /* Only update the hardware texture state if the texture is current,
+    * complete and enabled.
+    */
+   tObj = ctx->Texture.Unit[source].Current;
+   if ( !tObj || !tObj->Complete )
+      return;
+
+   if ( ( tObj != ctx->Texture.Unit[source].CurrentD[2] ) &&
+	( tObj != ctx->Texture.Unit[source].CurrentD[1] ) )
+      return;
+
+   /* We definately have a valid texture now */
+   t = tObj->DriverData;
+
+   /* Force the texture unit state to be loaded into the hardware */
+   rmesa->dirty |= RADEON_UPLOAD_CONTEXT | (RADEON_UPLOAD_TEX0 << unit);
+
+   /* Force any texture images to be loaded into the hardware */
+   if ( t->dirty_images ) {
+      if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+	 fprintf( stderr, "   t->dirty_images = 0x%x\n", t->dirty_images );
+      }
+      radeonSetTexImages( rmesa, tObj );
+      rmesa->dirty |= (RADEON_UPLOAD_TEX0IMAGES << unit);
+   }
+
+   if ( t->memBlock )
+      radeonUpdateTexLRU( rmesa, t );
+
+   switch ( unit ) {
+   case 0:
+      rmesa->setup.pp_cntl |= (RADEON_TEX_0_ENABLE |
+			       RADEON_TEX_BLEND_0_ENABLE);
+      break;
+   case 1:
+      rmesa->setup.pp_cntl |= (RADEON_TEX_1_ENABLE |
+			       RADEON_TEX_BLEND_1_ENABLE);
+      break;
+   }
+}
+
+void radeonUpdateTextureState( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p ) en=0x%x\n",
+	       __FUNCTION__, ctx, ctx->Texture.ReallyEnabled );
+   }
+
+   /* Clear any texturing fallbacks */
+   rmesa->Fallback &= ~RADEON_FALLBACK_TEXTURE;
+
+   /* Disable all texturing until it is known to be good */
+   rmesa->setup.pp_cntl &= ~(RADEON_TEX_ENABLE_MASK |
+			     RADEON_TEX_BLEND_ENABLE_MASK);
+
+   radeonUpdateTextureObject( ctx, 0 );
+   radeonUpdateTextureEnv( ctx, 0 );
+
+   if ( rmesa->multitex ) {
+      radeonUpdateTextureObject( ctx, 1 );
+      radeonUpdateTextureEnv( ctx, 1 );
+   }
+
+   rmesa->dirty |= RADEON_UPLOAD_CONTEXT;
+}
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_tris.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_tris.c
index 71ed6910b..e4b6394ee 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_tris.c
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_tris.c
@@ -55,6 +55,13 @@ do {									\
    *(GLuint *)(to) = *(GLuint *)(from);					\
 } while (0)
 
+#define RADEON_COLOR3( to, from )		\
+do {						\
+  (to)[0] = (from)[2];				\
+  (to)[1] = (from)[1];				\
+  (to)[2] = (from)[0];				\
+} while (0)
+
 
 static void radeon_null_quad( GLcontext *ctx, GLuint v0,
 			      GLuint v1, GLuint v2, GLuint v3, GLuint pv )
@@ -162,6 +169,10 @@ void radeonDDChooseRenderState( GLcontext *ctx )
 
    if ( rmesa->Fallback ) {
       rmesa->RenderIndex = RADEON_FALLBACK_BIT;
+      /* fixes vorder.c failure: */
+      if (flags & DD_TRI_LIGHT_TWOSIDE) {
+         rmesa->IndirectTriangles = DD_TRI_LIGHT_TWOSIDE;
+      }
       return;
    }
 
@@ -200,6 +211,11 @@ void radeonDDChooseRenderState( GLcontext *ctx )
 	 rmesa->IndirectTriangles |= (DD_TRI_SW_RASTERIZE |
 				      DD_QUAD_SW_RASTERIZE);
       }
+
+      /* fixes vorder.c failure: */
+      if (flags & DD_TRI_LIGHT_TWOSIDE) {
+         rmesa->IndirectTriangles |= DD_TRI_LIGHT_TWOSIDE;
+      }
    }
 
    if ( 0 ) {
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_tritmp.h b/xc/lib/GL/mesa/src/drv/radeon/radeon_tritmp.h
index ed5006ea2..a53512a30 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_tritmp.h
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_tritmp.h
@@ -247,7 +247,8 @@ static void TAG(line)( GLcontext *ctx,
    GLfloat z[2];
 #endif
 #if (IND & RADEON_TWOSIDE_BIT)
-   int c[2];
+   GLuint c[2];
+   GLuint s[2];
 #endif
 
    v[0] = &verts[e0];
@@ -256,17 +257,24 @@ static void TAG(line)( GLcontext *ctx,
 #if (IND & RADEON_TWOSIDE_BIT)
    c[0] = v[0]->ui[4];
    c[1] = v[1]->ui[4];
+   s[0] = v[0]->ui[5];
+   s[1] = v[1]->ui[5];
 #endif
 
 #if (IND & RADEON_TWOSIDE_BIT)
    {
       GLubyte (*vbcolor)[4] = ctx->VB->ColorPtr->data;
+      GLubyte (*vbspec)[4] = ctx->VB->Specular;
       if ( IND & RADEON_FLAT_BIT ) {
 	 RADEON_COLOR( (char *)&v[0]->ui[4], vbcolor[pv] );
 	 v[1]->ui[4] = v[0]->ui[4];
+	 RADEON_COLOR3( (char *)&v[0]->ui[5], vbspec[pv] );
+	 v[1]->ui[5] = v[0]->ui[5];
       } else {
 	 RADEON_COLOR( (char *)&v[0]->ui[4], vbcolor[e0] );
 	 RADEON_COLOR( (char *)&v[1]->ui[4], vbcolor[e1] );
+	 RADEON_COLOR3( (char *)&v[0]->ui[5], vbspec[e0] );
+	 RADEON_COLOR3( (char *)&v[1]->ui[5], vbspec[e1] );
       }
    }
 #endif
@@ -289,6 +297,8 @@ static void TAG(line)( GLcontext *ctx,
 #if (IND & RADEON_TWOSIDE_BIT)
    v[0]->ui[4] = c[0];
    v[1]->ui[4] = c[1];
+   v[0]->ui[5] = s[0];
+   v[1]->ui[5] = s[1];
 #endif
 }
 
@@ -309,7 +319,10 @@ static void TAG(points)( GLcontext *ctx,
 
 	    if ( IND & RADEON_TWOSIDE_BIT ) {
 	       GLubyte (*vbcolor)[4] = VB->ColorPtr->data;
+               GLubyte (*vbspec)[4] = VB->Specular;
 	       RADEON_COLOR( (char *)&tmp0.v.color, vbcolor[i] );
+               if (vbspec)
+                  RADEON_COLOR3( (char *)&tmp0.v.specular, vbspec[i] );
 	    }
 	    if ( IND & RADEON_OFFSET_BIT ) {
 	       GLfloat offset = ctx->PointZoffset * rmesa->depth_scale;
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/Imakefile b/xc/lib/GL/mesa/src/drv/tdfx/Imakefile
index ebb353844..9f45df60d 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/tdfx/Imakefile
@@ -150,6 +150,7 @@ MESA_INCLUDES = -I. -I.. -I../../include
 		../../stages.c \
 		../../state.c \
 		../../stencil.c \
+		../../texformat.c \
 		../../teximage.c \
 		../../texobj.c \
 		../../texstate.c \
@@ -223,6 +224,7 @@ MESA_INCLUDES = -I. -I.. -I../../include
 		../../stages.o \
 		../../state.o \
 		../../stencil.o \
+		../../texformat.o \
 		../../teximage.o \
 		../../texobj.o \
 		../../texstate.o \
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.c
index 4cb7efaba..293c42fd7 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.c
@@ -43,9 +43,22 @@
 #include "tdfx_render.h"
 #include "tdfx_pipeline.h"
 #include "tdfx_span.h"
+#include "tdfx_tex.h"
 #include "tdfx_texman.h"
 #include "extensions.h"
 
+#ifndef TDFX_DEBUG
+int TDFX_DEBUG = (0
+/*  		  | DEBUG_ALWAYS_SYNC */
+/*		  | DEBUG_VERBOSE_API */
+/*		  | DEBUG_VERBOSE_MSG */
+/*		  | DEBUG_VERBOSE_LRU */
+/*  		  | DEBUG_VERBOSE_DRI */
+/*  		  | DEBUG_VERBOSE_IOCTL */
+/*   		  | DEBUG_VERBOSE_2D */
+/*   		  | DEBUG_VERBOSE_TEXTURE */
+   );
+#endif
 
 
 #if 0
@@ -75,15 +88,13 @@ static void tdfxDDInitExtensions( GLcontext *ctx )
    gl_extensions_disable( ctx, "GL_INGR_blend_func_separate" );
    gl_extensions_enable( ctx, "GL_HP_occlusion_test" );
 
-   if ( !fxMesa->haveTwoTMUs ) {
+   if ( fxMesa->numTMUs == 1 ) {
       gl_extensions_disable( ctx, "GL_EXT_texture_env_add" );
       gl_extensions_disable( ctx, "GL_ARB_multitexture" );
    }
 
    if ( TDFX_IS_NAPALM( fxMesa ) ) {
-      gl_extensions_enable( ctx, "GL_ARB_texture_compression" );
       gl_extensions_enable( ctx, "GL_EXT_texture_env_combine" );
-      gl_extensions_enable( ctx, "GL_3DFX_texture_compression_FXT1" );
    }
 
    if (fxMesa->haveHwStencil) {
@@ -178,16 +189,10 @@ GLboolean tdfxCreateContext( Display *dpy, GLvisual *mesaVis,
    fxMesa->Glide.Initialized = GL_FALSE;
    fxMesa->Glide.Board = 0;
 
-
-   if (getenv("FX_EMULATE_SINGLE_TMU")) {
-      fxMesa->haveTwoTMUs = GL_FALSE;
-   }
-   else {
-      if ( TDFX_IS_BANSHEE( fxMesa ) ) {
-         fxMesa->haveTwoTMUs = GL_FALSE;
-      } else {
-         fxMesa->haveTwoTMUs = GL_TRUE;
-      }
+   if ( getenv( "FX_EMULATE_SINGLE_TMU" ) || TDFX_IS_BANSHEE( fxMesa ) ) {
+      fxMesa->numTMUs = 1;
+   } else {
+      fxMesa->numTMUs = 2;
    }
 
    fxMesa->stats.swapBuffer = 0;
@@ -218,6 +223,7 @@ GLboolean tdfxCreateContext( Display *dpy, GLvisual *mesaVis,
    tdfxDDInitStateFuncs( ctx );
    tdfxDDInitRenderFuncs( ctx );
    tdfxDDInitSpanFuncs( ctx );
+   tdfxDDInitTextureFuncs( ctx );
 
    ctx->Driver.TriangleCaps = (DD_TRI_CULL |
 			       DD_TRI_LIGHT_TWOSIDE |
@@ -448,19 +454,21 @@ void tdfxDestroyContext( tdfxContextPtr fxMesa )
    }
 
    if ( fxMesa ) {
-      if (fxMesa->glCtx->Shared->RefCount == 1) {
+      GLcontext *ctx = fxMesa->glCtx;
+      struct gl_texture_object *tObj;
+
+      if ( ctx->Shared->RefCount == 1 ) {
          /* This share group is about to go away, free our private
           * texture object data.
           */
-         struct gl_texture_object *tObj;
-         tObj = fxMesa->glCtx->Shared->TexObjectList;
-         while (tObj) {
-            tdfxTMFreeTexture(fxMesa, tObj);
-            tObj = tObj->Next;
+	 LOCK_HARDWARE( fxMesa );
+         for ( tObj = ctx->Shared->TexObjectList ; tObj ; tObj = tObj->Next ) {
+            tdfxTMFreeTextureLocked( fxMesa, tObj );
          }
+	 UNLOCK_HARDWARE( fxMesa );
       }
 
-      tdfxTMClose(fxMesa);  /* free texture memory */
+      tdfxTMClose( fxMesa );  /* free texture memory */
       XFree( fxMesa );
    }
 
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.h b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.h
index d630184bd..a73347435 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.h
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.h
@@ -280,9 +280,7 @@ typedef struct {
    volatile int fifoOwner;
    volatile int ctxOwner;
    volatile int texOwner;
-}
-TDFXSAREAPriv;
-
+} TDFXSAREAPriv;
 
 typedef struct {
    GLuint swapBuffer;
@@ -292,73 +290,72 @@ typedef struct {
    GLuint texSwaps;
 } tdfxStats;
 
-
-
 /*
  *  Memory range from startAddr to endAddr-1
  */
 typedef struct mem_range {
    struct mem_range *next;
    FxU32 startAddr, endAddr;
-}
-tdfxMemRange;
+} tdfxMemRange;
 
+typedef struct {
+   GLvoid *data;
+   GLsizei width, height;
+   FxU32 size;
+} tdfxTexRawData;
 
 typedef struct {
-   GLsizei		width, height;	/* image size */
-   GLint		texelSize;	/* How many bytes to a texel */
-   GrTextureFormat_t	glideFormat;	/* Glide image format */
-   void			*data;		/* Glide-formated texture image */
-   FxU32		dataSize;	/* image size in bytes */
-}
-tdfxMipMapLevel;
+   tdfxTexRawData original;		/* Mesa-formatted texture image */
+   tdfxTexRawData rescaled;		/* Only needed if aspect ratio > 8:1 */
+
+   GLvoid *data;			/* Final version of texture image */
+   FxU32 size;				/* image size in bytes */
+
+   GrTextureFormat_t glideFormat;	/* Glide image format */
+   GLint wScale, hScale;		/* Broken hardware... */
+} tdfxTexImage, *tdfxTexImagePtr;
 
 
 #define TDFX_NUM_TMU		2
 
 
-typedef struct tdfxTexInfo_t
-{
+typedef struct {
    GLboolean isInTM;
    GLboolean reloadImages;  /* if true, resend images to Glide */
    GLuint lastTimeUsed;
    FxU32 whichTMU;
 
    GrTexInfo info;
-   GrAspectRatio_t aspectRatio;
-   tdfxMipMapLevel mipmapLevel[MAX_TEXTURE_LEVELS];
-   tdfxMemRange *tm[TDFX_NUM_TMU];
+   tdfxTexImage image[MAX_TEXTURE_LEVELS];
+   tdfxMemRange *range[TDFX_NUM_TMU];
 
    GLint minLevel, maxLevel;
+   GrMipMapMode_t mmMode;
+   GrAspectRatio_t aspectRatio;
+   FxBool LODblend;
    GrTextureFilterMode_t minFilt;
    GrTextureFilterMode_t magFilt;
    GrTextureClampMode_t sClamp;
    GrTextureClampMode_t tClamp;
-   FxBool LODblend;
-   GrMipMapMode_t mmMode;
 
-   GLfloat sScale, tScale;  /* texcoord scale factor */
+   GLfloat sScale, tScale;		/* texcoord scale factor */
 
    GuTexPalette palette;
-}
-tdfxTexInfo;
+} tdfxTexObj, *tdfxTexObjPtr;
 
+#define TDFX_TEXTURE_DATA(tObj)		((tdfxTexObjPtr)((tObj)->DriverData))
 
-#define TDFX_TEXTURE_DATA(mesaObj) ((tdfxTexInfo *)((mesaObj)->DriverData))
 
-
-
-/*
- * This is state which may be shared by several tdfx contexts.
+/* This is state which may be shared by several tdfx contexts.
  * It hangs off of Mesa's gl_shared_state object (ctx->Shared->DriverData).
  */
-struct tdfxSharedState {
+typedef struct tdfx_shared_state {
    GLboolean umaTexMemory;
-   GLuint totalTexMem[TDFX_NUM_TMU]; /* constant */
-   GLuint freeTexMem[TDFX_NUM_TMU]; /* changes as we go */
-   tdfxMemRange *tmPool;
-   tdfxMemRange *tmFree[TDFX_NUM_TMU];
-};
+   GLuint totalTexMem[TDFX_NUM_TMU];	/* constant */
+   GLuint freeTexMem[TDFX_NUM_TMU];	/* changes as we go */
+   tdfxMemRange *rangePool;
+   tdfxMemRange *freeRanges[TDFX_NUM_TMU];
+} tdfxSharedState, *tdfxSharedStatePtr;
 
 
 
@@ -516,6 +513,11 @@ struct tdfx_depth {
    FxBool Mask;				/* Write enable flag */
 };
 
+#ifndef GR_STIPPLE_PATTERN
+#error "You MUST upgrade your Glide3 libraries and headers."
+#error "Get the latest from http://dri.sourceforge.net/resources.html"
+#endif
+
 struct tdfx_stipple {
    GrStippleMode_t Mode;		/* Stipple enable/disable */
    FxU32 Pattern;			/* 8x4 Stipple Pattern */
@@ -630,6 +632,7 @@ struct tdfx_context {
 
    GLuint tmu_source[TDFX_NUM_TMU];
    GLuint tex_dest[MAX_TEXTURE_UNITS];
+   GLuint numTMUs;
 
    GLuint SetupIndex;
    GLuint SetupDone;
@@ -641,10 +644,6 @@ struct tdfx_context {
    GLfloat sScale0, tScale0;
    GLfloat sScale1, tScale1;
 
-#if 0
-   GLuint last_tri_caps;
-   GLuint stw_hint_state;      /* for grHints */
-#endif
    GLuint using_fast_path, passes, multipass;
    GLuint texBindNumber;
    GLint tmuSrc;
@@ -652,13 +651,6 @@ struct tdfx_context {
    int screen_width;
    int screen_height;
 
-#if 0
-   void *state;                 /* Glide state buffer */
-
-   GLint textureAlign;
-   GLboolean verbose;
-#endif
-   GLboolean haveTwoTMUs;      /* True if we have 2 tmu's  */
    GLboolean haveHwStencil;
 
    GLint maxPendingSwapBuffers;
@@ -768,6 +760,7 @@ extern int TDFX_DEBUG;
 #define DEBUG_VERBOSE_DRI	0x10
 #define DEBUG_VERBOSE_IOCTL	0x20
 #define DEBUG_VERBOSE_2D	0x40
+#define DEBUG_VERBOSE_TEXTURE	0x80
 
 #endif /* GLX_DIRECT_RENDERING */
 
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_dd.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_dd.c
index ec5b25871..3f76fb88d 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_dd.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_dd.c
@@ -43,9 +43,11 @@
 
 #include "enums.h"
 #include "pb.h"
+#if defined(USE_X86_ASM) || defined(USE_3DNOW_ASM) || defined(USE_KATMAI_ASM)
+#include "X86/common_x86_asm.h"
+#endif
 
-
-#define TDFX_DATE	"20010104"
+#define TDFX_DATE	"20010305"
 
 
 /* These are used in calls to FX_grColorMaskv() */
@@ -63,10 +65,12 @@ static const GLubyte *tdfxDDGetString( GLcontext *ctx, GLenum name )
    tdfxContextPtr fxMesa = (tdfxContextPtr) ctx->DriverCtx;
 
    switch ( name ) {
-   case GL_RENDERER:
-   {
-      static char buffer[100];
-      char hardware[100];
+   case GL_VENDOR:
+      return "VA Linux Systems, Inc.";
+
+   case GL_RENDERER: {
+      static char buffer[128];
+      char hardware[128];
 
       strcpy( hardware, FX_grGetString( fxMesa, GR_HARDWARE ) );
 
@@ -83,19 +87,41 @@ static const GLubyte *tdfxDDGetString( GLcontext *ctx, GLenum name )
 	 strcpy( hardware, "Voodoo5" );
       }
       else {
-	 /* unexpected result: replace spaces with hyphens */
+	 /* Unexpected result: replace spaces with hyphens */
 	 int i;
 	 for ( i = 0 ; hardware[i] ; i++ ) {
 	    if ( hardware[i] == ' ' || hardware[i] == '\t' )
 	       hardware[i] = '-';
 	 }
       }
-      /* now make the GL_RENDERER string */
+      /* Now make the GL_RENDERER string */
       sprintf( buffer, "Mesa DRI %s " TDFX_DATE, hardware );
+
+      /* Append any CPU-specific information.
+       */
+#ifdef USE_X86_ASM
+      if ( gl_x86_cpu_features ) {
+	 strncat( buffer, " x86", 4 );
+      }
+#endif
+#ifdef USE_MMX_ASM
+      if ( cpu_has_mmx ) {
+	 strncat( buffer, "/MMX", 4 );
+      }
+#endif
+#ifdef USE_3DNOW_ASM
+      if ( cpu_has_3dnow ) {
+	 strncat( buffer, "/3DNow!", 7 );
+      }
+#endif
+#ifdef USE_KATMAI_ASM
+      if ( cpu_has_xmm ) {
+	 strncat( buffer, "/SSE", 4 );
+      }
+#endif
       return buffer;
    }
-   case GL_VENDOR:
-      return "VA Linux Systems, Inc.";
+
    default:
       return NULL;
    }
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_render.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_render.c
index 97931a351..da60bee16 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_render.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_render.c
@@ -594,16 +594,13 @@ static void uploadTextureImages( tdfxContextPtr fxMesa )
 {
    GLcontext *ctx = fxMesa->glCtx;
    int unit;
-   for (unit = 0; unit < TDFX_NUM_TMU; unit++) {
-      if (ctx->Texture.Unit[unit].ReallyEnabled == TEXTURE0_2D) {
+   for ( unit = 0 ; unit < TDFX_NUM_TMU ; unit++ ) {
+      if ( ctx->Texture.Unit[unit].ReallyEnabled == TEXTURE0_2D ) {
          struct gl_texture_object *tObj = ctx->Texture.Unit[unit].CurrentD[2];
-         tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-         if (ti && ti->reloadImages) {
-            /*
-            printf("download texture image on unit %d\n", unit);
-            */
-            tdfxTMDownloadTexture(fxMesa, tObj);
-            ti->reloadImages = GL_FALSE;
+         tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+         if ( t && t->reloadImages ) {
+            tdfxTMDownloadTextureLocked( fxMesa, tObj );
+            t->reloadImages = GL_FALSE;
          }
       }
    }
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_state.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_state.c
index eb9a6a98b..d630fbd8c 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_state.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_state.c
@@ -45,7 +45,6 @@
 #include "tdfx_vb.h"
 #include "tdfx_tex.h"
 #include "tdfx_texman.h"
-#include "tdfx_texstate.h"
 #include "tdfx_tris.h"
 #include "tdfx_render.h"
 
@@ -1535,15 +1534,6 @@ void tdfxDDInitStateFuncs( GLcontext *ctx )
    ctx->Driver.Scissor			= tdfxDDScissor;
    ctx->Driver.ShadeModel		= tdfxDDShadeModel;
 
-   ctx->Driver.BindTexture		= tdfxDDBindTexture;
-   ctx->Driver.DeleteTexture		= tdfxDDDeleteTexture;
-   ctx->Driver.TexEnv			= tdfxDDTexEnv;
-   ctx->Driver.TexParameter		= tdfxDDTexParameter;
-   ctx->Driver.TexImage2D		= tdfxDDTexImage2D;
-   ctx->Driver.TexSubImage2D		= tdfxDDTexSubImage2D;
-   ctx->Driver.GetTexImage		= tdfxDDGetTexImage;
-   ctx->Driver.UpdateTexturePalette	= tdfxDDTexturePalette;
-
    if ( fxMesa->haveHwStencil ) {
       ctx->Driver.StencilFunc		= tdfxDDStencilFunc;
       ctx->Driver.StencilMask		= tdfxDDStencilMask;
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.c
index c572e22a8..e0f2d2eb0 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.c
@@ -35,39 +35,37 @@
  *
  */
 
-#include "image.h"
-#include "texutil.h"
 #include "tdfx_context.h"
 #include "tdfx_tex.h"
 #include "tdfx_texman.h"
 
+#include "enums.h"
+#include "image.h"
+#include "texutil.h"
+
+#define TX_DITHER_NONE		0x00000000
 
 
-static int
-logbase2(int n)
+static int logbase2( int n )
 {
-    GLint i = 1;
-    GLint log2 = 0;
-
-    if (n < 0) {
-        return -1;
-    }
-
-    while (n > i) {
-        i *= 2;
-        log2++;
-    }
-    if (i != n) {
-        return -1;
-    }
-    else {
-        return log2;
-    }
+   GLint i = 1;
+   GLint log2 = 0;
+
+   if ( n < 0 )
+      return -1;
+
+   while ( n > i ) {
+      i *= 2;
+      log2++;
+   }
+   if ( i != n ) {
+      return -1;
+   } else {
+      return log2;
+   }
 }
 
-
-/*
- * Compute various texture image parameters.
+/* Compute various texture image parameters.
  * Input:  w, h - source texture width and height
  * Output:  lodlevel - Glide lod level token for the larger texture dimension
  *          aspectratio - Glide aspect ratio token
@@ -84,1520 +82,977 @@ logbase2(int n)
  *      32   64  GR_LOD_LOG2_64 (=6)   GR_ASPECT_LOG2_1x2 (=-1)
  *      32   32  GR_LOD_LOG2_32 (=5)   GR_ASPECT_LOG2_1x1 (=0)
  */
-static void
-tdfxTexGetInfo(const GLcontext *ctx, int w, int h,
-               GrLOD_t *lodlevel, GrAspectRatio_t *aspectratio,
-               float *sscale, float *tscale,
-               int *wscale, int *hscale)
+static void tdfxTexGetInfo( const GLcontext *ctx, int w, int h,
+			    GrLOD_t *lodlevel, GrAspectRatio_t *aspectratio,
+			    float *sscale, float *tscale,
+			    int *wscale, int *hscale )
 {
-    int logw, logh, ar, lod, ws, hs;
-    float s, t;
-
-    ASSERT(w >= 1);
-    ASSERT(h >= 1);
-
-    logw = logbase2(w);
-    logh = logbase2(h);
-    ar = logw - logh;  /* aspect ratio = difference in log dimensions */
-
-    /* Hardware only allows a maximum aspect ratio of 8x1, so handle
-       |ar| > 3 by scaling the image and using an 8x1 aspect ratio */
-    if (ar >= 0) {
-        ASSERT(width >= height);
-        lod = logw;
-        s = 256.0;
-        ws = 1;
-        if (ar <= GR_ASPECT_LOG2_8x1) {
-            t = 256 >> ar;
-            hs = 1;
-        }
-        else {
-            /* have to stretch image height */
-            t = 32.0;
-            hs = 1 << (ar - 3);
-        }
-    }
-    else {
-        ASSERT(width < height);
-        lod = logh;
-        t = 256.0;
-        hs = 1;
-        if (ar >= GR_ASPECT_LOG2_1x8) {
-            s = 256 >> -ar;
-            ws = 1;
-        }
-        else {
-            /* have to stretch image width */
-            s = 32.0;
-            ws = 1 << (-ar - 3);
-        }
-    }
-
-    if (ar < GR_ASPECT_LOG2_1x8)
-        ar = GR_ASPECT_LOG2_1x8;
-    else if (ar > GR_ASPECT_LOG2_8x1)
-        ar = GR_ASPECT_LOG2_8x1;
-
-    if (lodlevel)
-        *lodlevel = (GrLOD_t) lod;
-    if (aspectratio)
-        *aspectratio = (GrAspectRatio_t) ar;
-    if (sscale)
-        *sscale = s;
-    if (tscale)
-        *tscale = t;
-    if (wscale)
-        *wscale = ws;
-    if (hscale)
-        *hscale = hs;
+   int logw, logh, ar, lod, ws, hs;
+   float s, t;
+
+   ASSERT( w >= 1 );
+   ASSERT( h >= 1 );
+
+   logw = logbase2( w );
+   logh = logbase2( h );
+   ar = logw - logh;  /* aspect ratio = difference in log dimensions */
+
+   /* Hardware only allows a maximum aspect ratio of 8x1, so handle
+    * |ar| > 3 by scaling the image and using an 8x1 aspect ratio.
+    */
+   if ( ar >= 0 ) {
+      ASSERT( width >= height );
+      lod = logw;
+      s = 256.0;
+      ws = 1;
+      if ( ar <= GR_ASPECT_LOG2_8x1 ) {
+	 t = 256 >> ar;
+	 hs = 1;
+      } else {
+	 /* have to stretch image height */
+	 t = 32.0;
+	 hs = 1 << (ar - 3);
+      }
+   } else {
+      ASSERT( width < height );
+      lod = logh;
+      t = 256.0;
+      hs = 1;
+      if ( ar >= GR_ASPECT_LOG2_1x8 ) {
+	 s = 256 >> -ar;
+	 ws = 1;
+      } else {
+	 /* have to stretch image width */
+	 s = 32.0;
+	 ws = 1 << (-ar - 3);
+      }
+   }
+
+   if ( ar < GR_ASPECT_LOG2_1x8 ) {
+      ar = GR_ASPECT_LOG2_1x8;
+   } else if ( ar > GR_ASPECT_LOG2_8x1 ) {
+      ar = GR_ASPECT_LOG2_8x1;
+   }
+
+   if ( lodlevel )
+      *lodlevel = (GrLOD_t)lod;
+   if ( aspectratio )
+      *aspectratio = (GrAspectRatio_t)ar;
+   if ( sscale )
+      *sscale = s;
+   if ( tscale )
+      *tscale = t;
+   if ( wscale )
+      *wscale = ws;
+   if ( hscale )
+      *hscale = hs;
 }
 
 
-/*
- * We need to call this when a texture object's minification filter
+/* We need to call this when a texture object's minification filter
  * or texture image sizes change.
  */
-static void RevalidateTexture(GLcontext *ctx, struct gl_texture_object *tObj)
+static void tdfxRevalidateTexture( GLcontext *ctx,
+				   struct gl_texture_object *tObj )
 {
-    tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-    GLint minl, maxl;
-
-    if (!ti)
-       return;
-
-    minl = maxl = tObj->BaseLevel;
-
-    if (tObj->Image[minl]) {
-       maxl = MIN2(tObj->MaxLevel, tObj->Image[minl]->MaxLog2);
-
-       /* compute largeLodLog2, aspect ratio and texcoord scale factors */
-       tdfxTexGetInfo(ctx, tObj->Image[minl]->Width, tObj->Image[minl]->Height,
-                      &ti->info.largeLodLog2,
-                      &ti->info.aspectRatioLog2,
-                      &(ti->sScale), &(ti->tScale), NULL, NULL);
-    }
-
-    if (tObj->Image[maxl] && (tObj->MinFilter != GL_NEAREST) && (tObj->MinFilter != GL_LINEAR)) {
-        /* mipmapping: need to compute smallLodLog2 */
-        tdfxTexGetInfo(ctx, tObj->Image[maxl]->Width,
-                       tObj->Image[maxl]->Height,
-                       &ti->info.smallLodLog2, NULL,
-                       NULL, NULL, NULL, NULL);
-    }
-    else {
-        /* not mipmapping: smallLodLog2 = largeLodLog2 */
-        ti->info.smallLodLog2 = ti->info.largeLodLog2;
-    }
-
-    ti->minLevel = minl;
-    ti->maxLevel = maxl;
-    ti->info.data = NULL;
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+   GLint minl, maxl;
+
+   if ( !t )
+      return;
+
+   minl = maxl = tObj->BaseLevel;
+
+   if ( tObj->Image[minl] ) {
+      maxl = MIN2( tObj->MaxLevel, tObj->Image[minl]->MaxLog2 );
+
+      /* Compute largeLodLog2, aspect ratio and texcoord scale factors.
+       */
+      tdfxTexGetInfo( ctx,
+		      tObj->Image[minl]->Width, tObj->Image[minl]->Height,
+		      &t->info.largeLodLog2, &t->info.aspectRatioLog2,
+		      &t->sScale, &t->tScale, NULL, NULL );
+   }
+
+   if ( tObj->Image[maxl] &&
+	tObj->MinFilter != GL_NEAREST  &&
+	tObj->MinFilter != GL_LINEAR ) {
+      /* Mipmapping: need to compute smallLodLog2 */
+      tdfxTexGetInfo( ctx,
+		      tObj->Image[maxl]->Width, tObj->Image[maxl]->Height,
+		      &t->info.smallLodLog2,
+		      NULL, NULL, NULL, NULL, NULL );
+   } else {
+      /* Not mipmapping: smallLodLog2 = largeLodLog2 */
+      t->info.smallLodLog2 = t->info.largeLodLog2;
+   }
+
+   t->minLevel = minl;
+   t->maxLevel = maxl;
+   t->info.data = NULL;
 }
 
 
-static tdfxTexInfo *
-fxAllocTexObjData(tdfxContextPtr fxMesa)
+static tdfxTexObjPtr tdfxAllocTexObj( tdfxContextPtr fxMesa )
 {
-    tdfxTexInfo *ti;
-    int i;
+   tdfxTexObjPtr t;
+   int i;
 
-    if (!(ti = CALLOC(sizeof(tdfxTexInfo)))) {
-        gl_problem(NULL, "tdfx driver: out of memory");
-        return NULL;
-    }
+   t = CALLOC( sizeof(tdfxTexObj) );
+   if ( !t ) {
+      gl_problem( NULL, "tdfx driver: out of memory" );
+      return NULL;
+   }
 
-    ti->isInTM = GL_FALSE;
+   t->isInTM = GL_FALSE;
 
-    ti->whichTMU = TDFX_TMU_NONE;
+   t->whichTMU = TDFX_TMU_NONE;
 
-    ti->tm[TDFX_TMU0] = NULL;
-    ti->tm[TDFX_TMU1] = NULL;
+   t->range[TDFX_TMU0] = NULL;
+   t->range[TDFX_TMU1] = NULL;
 
-    ti->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
-    ti->magFilt = GR_TEXTUREFILTER_BILINEAR;
+   t->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
+   t->magFilt = GR_TEXTUREFILTER_BILINEAR;
 
-    ti->sClamp = GR_TEXTURECLAMP_WRAP;
-    ti->tClamp = GR_TEXTURECLAMP_WRAP;
+   t->sClamp = GR_TEXTURECLAMP_WRAP;
+   t->tClamp = GR_TEXTURECLAMP_WRAP;
 
-    ti->mmMode = GR_MIPMAP_NEAREST;
-    ti->LODblend = FXFALSE;
+   t->mmMode = GR_MIPMAP_NEAREST;
+   t->LODblend = FXFALSE;
 
-    for (i = 0; i < MAX_TEXTURE_LEVELS; i++) {
-        ti->mipmapLevel[i].data = NULL;
-    }
+   for ( i = 0 ; i < MAX_TEXTURE_LEVELS ; i++ ) {
+      t->image[i].original.data = NULL;
+      t->image[i].rescaled.data = NULL;
+   }
 
-    return ti;
+   return t;
 }
 
 
-/*
- * Called via glBindTexture.
+/* Given an OpenGL internal texture format, return the corresponding
+ * Glide internal texture format and MesaIntTexFormat.
+ * If allow32bpp is true, we'll return 32-bit texel formats when
+ * appropriate.
  */
-
-void
-tdfxDDBindTexture(GLcontext * ctx, GLenum target,
-                  struct gl_texture_object *tObj)
+static GrTextureFormat_t
+tdfxTexGetFormat( tdfxContextPtr fxMesa, struct gl_texture_image *texImage,
+		  GLenum format, GLenum type )
 {
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    tdfxTexInfo *ti;
+   const GLboolean allow32bpp = TDFX_IS_NAPALM(fxMesa);
+   const GLboolean is32bpp = ( fxMesa->fxScreen->cpp == 4 );
+   const struct gl_texture_format *texFormat;
+   GrTextureFormat_t ret;
+
+   if ( 0 )
+      fprintf( stderr, "internal=%s format=%s type=%s\n",
+	       texImage->IntFormat == 3 ? "GL_RGB (3)" :
+	       texImage->IntFormat == 4 ? "GL_RGBA (4)" :
+	       gl_lookup_enum_by_nr( texImage->IntFormat ),
+	       gl_lookup_enum_by_nr( format ),
+	       gl_lookup_enum_by_nr( type ) );
+
+#define SET_FORMAT( gr, gl )						\
+   do {									\
+      ret = (gr);							\
+      texFormat = &(gl);						\
+   } while (0)
+
+#define SET_FORMAT_32BPP( gr32, gl32, gr16, gl16 )			\
+   do {									\
+      if ( allow32bpp ) {						\
+	 ret = (gr32);							\
+	 texFormat = &(gl32);						\
+      } else {								\
+	 ret = (gr16);							\
+	 texFormat = &(gl16);						\
+      }									\
+   } while (0)
+
+   switch ( texImage->IntFormat ) {
+      /* GH: Bias towards GL_RGB, GL_RGBA texture formats.  This has
+       * got to be better than sticking them way down the end of this
+       * huge list.
+       */
+   case GL_RGBA:
+   case 4:
+      if ( format == GL_BGRA ) {
+	 if ( type == GL_UNSIGNED_INT_8_8_8_8_REV && allow32bpp ) {
+	    SET_FORMAT( GR_TEXFMT_ARGB_8888, _mesa_texformat_argb8888 );
+	    break;
+	 } else if ( type == GL_UNSIGNED_SHORT_4_4_4_4_REV ) {
+	    SET_FORMAT( GR_TEXFMT_ARGB_4444, _mesa_texformat_argb4444 );
+	    break;
+	 } else if ( type == GL_UNSIGNED_SHORT_1_5_5_5_REV ) {
+	    SET_FORMAT( GR_TEXFMT_ARGB_1555, _mesa_texformat_argb1555 );
+	    break;
+	 }
+      }
+      if ( allow32bpp && is32bpp ) {
+	 SET_FORMAT( GR_TEXFMT_ARGB_8888, _mesa_texformat_rgba8888 );
+      } else {
+	 SET_FORMAT( GR_TEXFMT_ARGB_4444, _mesa_texformat_argb4444 );
+      }
+      break;
+
+   case GL_RGB:
+   case 3:
+      if ( format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5 ) {
+	 SET_FORMAT( GR_TEXFMT_RGB_565, _mesa_texformat_rgb565 );
+	 break;
+      }
+      if ( allow32bpp && is32bpp ) {
+	 SET_FORMAT( GR_TEXFMT_ARGB_8888, _mesa_texformat_rgba8888 );
+      } else {
+	 SET_FORMAT( GR_TEXFMT_RGB_565, _mesa_texformat_rgb565 );
+      }
+      break;
+
+      /* GH: Okay, keep checking as normal.  Still test for GL_RGB,
+       * GL_RGBA formats first.
+       */
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      SET_FORMAT_32BPP( GR_TEXFMT_ARGB_8888, _mesa_texformat_argb8888,
+			GR_TEXFMT_ARGB_4444, _mesa_texformat_argb4444 );
+      break;
+
+   case GL_RGBA4:
+   case GL_RGBA2:
+      SET_FORMAT( GR_TEXFMT_ARGB_4444, _mesa_texformat_argb4444 );
+      break;
+
+   case GL_RGB5_A1:
+      SET_FORMAT( GR_TEXFMT_ARGB_1555, _mesa_texformat_argb1555 );
+      break;
+
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+      SET_FORMAT_32BPP( GR_TEXFMT_ARGB_8888, _mesa_texformat_argb8888,
+			GR_TEXFMT_RGB_565,   _mesa_texformat_rgb565 );
+      break;
+
+   case GL_RGB5:
+   case GL_RGB4:
+   case GL_R3_G3_B2:
+      SET_FORMAT( GR_TEXFMT_RGB_565, _mesa_texformat_rgb565 );
+      break;
+
+   case GL_ALPHA:
+   case GL_ALPHA4:
+   case GL_ALPHA8:
+   case GL_ALPHA12:
+   case GL_ALPHA16:
+      SET_FORMAT( GR_TEXFMT_ALPHA_8, _mesa_texformat_a8 );
+      break;
+
+   case 1:
+   case GL_LUMINANCE:
+   case GL_LUMINANCE4:
+   case GL_LUMINANCE8:
+   case GL_LUMINANCE12:
+   case GL_LUMINANCE16:
+      SET_FORMAT( GR_TEXFMT_INTENSITY_8, _mesa_texformat_l8 );
+      break;
+
+   case 2:
+   case GL_LUMINANCE_ALPHA:
+   case GL_LUMINANCE4_ALPHA4:
+   case GL_LUMINANCE6_ALPHA2:
+   case GL_LUMINANCE8_ALPHA8:
+   case GL_LUMINANCE12_ALPHA4:
+   case GL_LUMINANCE12_ALPHA12:
+   case GL_LUMINANCE16_ALPHA16:
+      SET_FORMAT( GR_TEXFMT_ALPHA_INTENSITY_88, _mesa_texformat_al88 );
+      break;
+
+   case GL_INTENSITY:
+   case GL_INTENSITY4:
+   case GL_INTENSITY8:
+   case GL_INTENSITY12:
+   case GL_INTENSITY16:
+      SET_FORMAT( GR_TEXFMT_ALPHA_8, _mesa_texformat_i8 );
+      break;
+
+   case GL_COLOR_INDEX:
+   case GL_COLOR_INDEX1_EXT:
+   case GL_COLOR_INDEX2_EXT:
+   case GL_COLOR_INDEX4_EXT:
+   case GL_COLOR_INDEX8_EXT:
+   case GL_COLOR_INDEX12_EXT:
+   case GL_COLOR_INDEX16_EXT:
+      SET_FORMAT( GR_TEXFMT_P_8, _mesa_texformat_ci8 );
+      break;
+
+   default:
+      fprintf( stderr, "bad texture format in fxTexGetFormat() %d",
+	       texImage->IntFormat );
+      return -1;
+   }
+
+   texImage->TexFormat = texFormat;
+
+   return ret;
+}
 
-    if (MESA_VERBOSE & VERBOSE_DRIVER) {
-        fprintf(stderr, "fxmesa: fxDDTexBind(%d,%p)\n", tObj->Name,
-                tObj->DriverData);
-    }
 
-    if (target != GL_TEXTURE_2D)
-        return;
+static GLboolean
+tdfxDDTexImage2D( GLcontext *ctx, GLenum target, GLint level,
+		  GLenum format, GLenum type, const GLvoid *pixels,
+		  const struct gl_pixelstore_attrib *packing,
+		  struct gl_texture_object *texObj,
+		  struct gl_texture_image *texImage,
+		  GLboolean *retainInternalCopy )
+{
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   const struct gl_texture_format *texFormat;
+   GrTextureFormat_t glideFormat;
+   tdfxTexObjPtr t;
+   tdfxTexImagePtr image;
+   GLint dstWidth, dstHeight, wScale, hScale;
+   GLint size;
+   void *data;
+
+   if ( 0 ) {
+      printf("TexImage id=%d int 0x%x  format 0x%x  type 0x%x  %dx%d\n",
+	     texObj->Name, texImage->IntFormat, format, type,
+	     texImage->Width, texImage->Height);
+   }
+
+   if ( target != GL_TEXTURE_2D || texImage->Border > 0 )
+      return GL_FALSE;
+
+   if ( !texObj->DriverData )
+      texObj->DriverData = tdfxAllocTexObj( fxMesa );
+
+   t = TDFX_TEXTURE_DATA(texObj);
+   image = &t->image[level];
+
+   /* Determine the appropriate GL internal texel format, Mesa internal
+    * texel format, and texelSize (bytes) given the user's internal
+    * texture format hint.
+    */
+   glideFormat = tdfxTexGetFormat( fxMesa, texImage, format, type );
 
-    if (!tObj->DriverData) {
-        tObj->DriverData = fxAllocTexObjData(fxMesa);
-    }
+   /* Get the destination internal format.
+    */
+   texFormat = texImage->TexFormat;
 
-    ti = TDFX_TEXTURE_DATA(tObj);
-    ti->lastTimeUsed = fxMesa->texBindNumber++;
+   /* Determine width and height scale factors for texture.  Remember,
+    * Glide is limited to 8:1 aspect ratios.
+    */
+   tdfxTexGetInfo( ctx,
+		   texImage->Width, texImage->Height,
+		   NULL, NULL, NULL, NULL,
+		   &wScale, &hScale );
+   dstWidth = texImage->Width * wScale;
+   dstHeight = texImage->Height * hScale;
+
+   /* Allocate new storage for texture image, if needed.  This
+    * conditional wants to set uncompressedImage to point to the
+    * uncompressed image, and mml->data to the texture data.  If the
+    * image is uncompressed, these are identical.  If the image is not
+    * compressed, these are different.
+    */
+   if ( !image->original.data || image->glideFormat != glideFormat ||
+	image->original.width != texImage->Width ||
+	image->original.height != texImage->Height )
+   {
+      if ( image->original.data ) {
+	 FREE( image->original.data );
+	 image->original.data = NULL;
+      }
+      if ( image->rescaled.data ) {
+	 FREE( image->rescaled.data );
+	 image->rescaled.data = NULL;
+      }
+
+      size = texImage->Width * texImage->Height * texFormat->TexelBytes;
+      image->original.data = (void *) MALLOC( size );
+      if ( !image->original.data )
+	 return GL_FALSE;
+
+      image->original.width = texImage->Width;
+      image->original.height = texImage->Height;
+      image->original.size = size;
+
+      image->glideFormat = glideFormat;
+      image->wScale = wScale;
+      image->hScale = hScale;
+
+      t->info.format = glideFormat;
+      tdfxTMMoveOutTM( fxMesa, texObj );
+   }
+
+   /* Store the texture image into the 'original' space.
+    */
+   if ( !_mesa_convert_texsubimage2d( texFormat->IntFormat,
+				      0, 0, texImage->Width,
+				      texImage->Height, texImage->Width,
+				      format, type, packing, pixels,
+				      image->original.data ) ) {
+      return GL_FALSE;
+   }
+
+   data = image->original.data;
+   size = image->original.size;
+
+   /* GH: Sigh...
+    */
+   if ( wScale > 1 || hScale > 1 ) {
+      if ( image->rescaled.data ) {
+	 FREE( image->rescaled.data );
+	 image->rescaled.data = NULL;
+      }
 
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;
-}
+      size = dstWidth * dstHeight * texFormat->TexelBytes;
+      image->rescaled.data = (void *) MALLOC( size );
+      if ( !image->rescaled.data )
+	 return GL_FALSE;
 
+      image->rescaled.width = dstWidth;
+      image->rescaled.height = dstHeight;
+      image->rescaled.size = size;
 
-/*
- * Called via glTexEnv.
- */
-void
-tdfxDDTexEnv(GLcontext * ctx, GLenum target, GLenum pname,
-             const GLfloat * param)
-{
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-
-    if ( TDFX_DEBUG & DEBUG_VERBOSE_API ) {
-        if (param)
-            fprintf(stderr, "fxmesa: texenv(%x,%x)\n", pname,
-                    (GLint) (*param));
-        else
-            fprintf(stderr, "fxmesa: texenv(%x)\n", pname);
-    }
+      _mesa_rescale_teximage2d( texFormat,
+				texImage->Width, texImage->Height,
+				dstWidth, dstHeight,
+				image->original.data, image->rescaled.data );
 
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;
-}
+      data = image->rescaled.data;
+   }
 
+   image->data = data;
+   image->size = size;
 
-/*
- * Called via glTexParameter.
- */
-void
-tdfxDDTexParameter(GLcontext * ctx, GLenum target,
-                   struct gl_texture_object *tObj,
-                   GLenum pname, const GLfloat * params)
-{
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    GLenum param = (GLenum) (GLint) params[0];
-    tdfxTexInfo *ti;
-
-    if (MESA_VERBOSE & VERBOSE_DRIVER) {
-        fprintf(stderr, "fxmesa: fxDDTexParam(%d,%p,%x,%x)\n", tObj->Name,
-                tObj->DriverData, pname, param);
-    }
-
-    if (target != GL_TEXTURE_2D)
-        return;
-
-    if (!tObj->DriverData)
-        tObj->DriverData = fxAllocTexObjData(fxMesa);
-
-    ti = TDFX_TEXTURE_DATA(tObj);
-
-    switch (pname) {
-    case GL_TEXTURE_MIN_FILTER:
-        switch (param) {
-        case GL_NEAREST:
-            ti->mmMode = GR_MIPMAP_DISABLE;
-            ti->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
-            ti->LODblend = FXFALSE;
-            break;
-        case GL_LINEAR:
-            ti->mmMode = GR_MIPMAP_DISABLE;
-            ti->minFilt = GR_TEXTUREFILTER_BILINEAR;
-            ti->LODblend = FXFALSE;
-            break;
-        case GL_NEAREST_MIPMAP_LINEAR:
-            if (TDFX_IS_NAPALM(fxMesa)) {
-                 if (fxMesa->haveTwoTMUs) {
-                     ti->mmMode = GR_MIPMAP_NEAREST;
-                     ti->LODblend = FXTRUE;
-                 }
-                 else {
-                     ti->mmMode = GR_MIPMAP_NEAREST_DITHER;
-                     ti->LODblend = FXFALSE;
-                 }
-                 ti->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
-                 break;
-            }
-            /* XXX Voodoo3/Banshee mipmap blending seems to produce
-             * incorrectly filtered colors for the smallest mipmap levels.
-             * To work-around we fall-through here and use a different filter.
-             */
-        case GL_NEAREST_MIPMAP_NEAREST:
-            ti->mmMode = GR_MIPMAP_NEAREST;
-            ti->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
-            ti->LODblend = FXFALSE;
-            break;
-        case GL_LINEAR_MIPMAP_LINEAR:
-            if (TDFX_IS_NAPALM(fxMesa)) {
-                if (fxMesa->haveTwoTMUs) {
-                    ti->mmMode = GR_MIPMAP_NEAREST;
-                    ti->LODblend = FXTRUE;
-                }
-                else {
-                    ti->mmMode = GR_MIPMAP_NEAREST_DITHER;
-                    ti->LODblend = FXFALSE;
-                }
-                ti->minFilt = GR_TEXTUREFILTER_BILINEAR;
-                break;
-            }
-            /* XXX Voodoo3/Banshee mipmap blending seems to produce
-             * incorrectly filtered colors for the smallest mipmap levels.
-             * To work-around we fall-through here and use a different filter.
-             */
-        case GL_LINEAR_MIPMAP_NEAREST:
-            ti->mmMode = GR_MIPMAP_NEAREST;
-            ti->minFilt = GR_TEXTUREFILTER_BILINEAR;
-            ti->LODblend = FXFALSE;
-            break;
-        default:
-            break;
-        }
-        RevalidateTexture(ctx, tObj);
-        fxMesa->new_state |= TDFX_NEW_TEXTURE;
-        break;
-
-    case GL_TEXTURE_MAG_FILTER:
-        switch (param) {
-        case GL_NEAREST:
-            ti->magFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
-            break;
-        case GL_LINEAR:
-            ti->magFilt = GR_TEXTUREFILTER_BILINEAR;
-            break;
-        default:
-            break;
-        }
-        fxMesa->new_state |= TDFX_NEW_TEXTURE;
-        break;
-
-    case GL_TEXTURE_WRAP_S:
-        switch (param) {
-        case GL_CLAMP:
-            ti->sClamp = GR_TEXTURECLAMP_CLAMP;
-            break;
-        case GL_REPEAT:
-            ti->sClamp = GR_TEXTURECLAMP_WRAP;
-            break;
-        default:
-            break;
-        }
-        fxMesa->new_state |= TDFX_NEW_TEXTURE;
-        break;
-
-    case GL_TEXTURE_WRAP_T:
-        switch (param) {
-        case GL_CLAMP:
-            ti->tClamp = GR_TEXTURECLAMP_CLAMP;
-            break;
-        case GL_REPEAT:
-            ti->tClamp = GR_TEXTURECLAMP_WRAP;
-            break;
-        default:
-            break;
-        }
-        fxMesa->new_state |= TDFX_NEW_TEXTURE;
-        break;
-
-    case GL_TEXTURE_BORDER_COLOR:
-        /* TO DO */
-        break;
-    case GL_TEXTURE_MIN_LOD:
-        /* TO DO */
-        break;
-    case GL_TEXTURE_MAX_LOD:
-        /* TO DO */
-        break;
-    case GL_TEXTURE_BASE_LEVEL:
-        RevalidateTexture(ctx, tObj);
-        break;
-    case GL_TEXTURE_MAX_LEVEL:
-        RevalidateTexture(ctx, tObj);
-        break;
-
-    default:
-        break;
-    }
-}
+   tdfxRevalidateTexture( ctx, texObj );
 
+   t->reloadImages = GL_TRUE;
+   fxMesa->new_state |= TDFX_NEW_TEXTURE;
 
-/*
- * Called via glDeleteTextures to delete a texture object.
- * Here, we delete the Glide data associated with the texture.
- */
-void
-tdfxDDDeleteTexture(GLcontext * ctx, struct gl_texture_object *tObj)
-{
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    tdfxTMFreeTexture(fxMesa, tObj);
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;
+   *retainInternalCopy = GL_FALSE;
+   return GL_TRUE;
 }
 
 
-/*
- * Return true if texture is resident, false otherwise.
- */
-GLboolean
-tdfxDDIsTextureResident(GLcontext *ctx, struct gl_texture_object *tObj)
+static GLboolean
+tdfxDDTexSubImage2D( GLcontext *ctx, GLenum target, GLint level,
+		     GLint xoffset, GLint yoffset,
+		     GLsizei width, GLsizei height,
+		     GLenum format, GLenum type, const GLvoid *pixels,
+		     const struct gl_pixelstore_attrib *packing,
+		     struct gl_texture_object *texObj,
+		     struct gl_texture_image *texImage )
 {
-    tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-    return (GLboolean) (ti && ti->isInTM);
-}
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(texObj);
+   tdfxTexImagePtr image;
 
+   if ( target != GL_TEXTURE_2D )
+      return GL_FALSE;
 
+   if ( !t )
+      return GL_FALSE;
 
-/*
- * Convert a gl_color_table texture palette to Glide's format.
- */
-static void
-convertPalette(FxU32 data[256], const struct gl_color_table *table)
-{
-    const GLubyte *tableUB = (const GLubyte *) table->Table;
-    GLint width = table->Size;
-    FxU32 r, g, b, a;
-    GLint i;
-
-    ASSERT(table->TableType == GL_UNSIGNED_BYTE);
-
-    switch (table->Format) {
-    case GL_INTENSITY:
-        for (i = 0; i < width; i++) {
-            r = tableUB[i];
-            g = tableUB[i];
-            b = tableUB[i];
-            a = tableUB[i];
-            data[i] = (a << 24) | (r << 16) | (g << 8) | b;
-        }
-        break;
-    case GL_LUMINANCE:
-        for (i = 0; i < width; i++) {
-            r = tableUB[i];
-            g = tableUB[i];
-            b = tableUB[i];
-            a = 255;
-            data[i] = (a << 24) | (r << 16) | (g << 8) | b;
-        }
-        break;
-    case GL_ALPHA:
-        for (i = 0; i < width; i++) {
-            r = g = b = 255;
-            a = tableUB[i];
-            data[i] = (a << 24) | (r << 16) | (g << 8) | b;
-        }
-        break;
-    case GL_LUMINANCE_ALPHA:
-        for (i = 0; i < width; i++) {
-            r = g = b = tableUB[i * 2 + 0];
-            a = tableUB[i * 2 + 1];
-            data[i] = (a << 24) | (r << 16) | (g << 8) | b;
-        }
-        break;
-    case GL_RGB:
-        for (i = 0; i < width; i++) {
-            r = tableUB[i * 3 + 0];
-            g = tableUB[i * 3 + 1];
-            b = tableUB[i * 3 + 2];
-            a = 255;
-            data[i] = (a << 24) | (r << 16) | (g << 8) | b;
-        }
-        break;
-    case GL_RGBA:
-        for (i = 0; i < width; i++) {
-            r = tableUB[i * 4 + 0];
-            g = tableUB[i * 4 + 1];
-            b = tableUB[i * 4 + 2];
-            a = tableUB[i * 4 + 3];
-            data[i] = (a << 24) | (r << 16) | (g << 8) | b;
-        }
-        break;
-    }
-}
+   if ( 0 ) {
+      fprintf( stderr, "TexSubImage id=%d lvl=%d int=0x%x format=0x%x type=0x%x  x=%d y=%d w=%d h=%d fullW=%d fullH=%d\n",
+	       texObj->Name, level, texImage->IntFormat, format, type,
+	       xoffset, yoffset, width, height,
+	       texImage->Width, texImage->Height );
+   }
 
+   image = &t->image[level];
 
+   /* Must have an existing texture image!
+    */
+   assert( image->original.data );
 
-void
-tdfxDDTexturePalette(GLcontext * ctx, struct gl_texture_object *tObj)
-{
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-
-    if (tObj) {
-        /* per-texture palette */
-        tdfxTexInfo *ti;
-        if (!tObj->DriverData)
-            tObj->DriverData = fxAllocTexObjData(fxMesa);
-        ti = TDFX_TEXTURE_DATA(tObj);
-        convertPalette(ti->palette.data, &tObj->Palette);
-        /*tdfxTexInvalidate(ctx, tObj);*/
-    }
-    else {
-        /* global texture palette */
-        convertPalette(fxMesa->glbPalette.data, &ctx->Texture.Palette);
-    }
-    fxMesa->new_state |= TDFX_NEW_TEXTURE; /* XXX too heavy-handed */
+   if ( !_mesa_convert_texsubimage2d( texImage->TexFormat->IntFormat,
+				      xoffset, yoffset, width, height,
+				      texImage->Width,
+				      format, type, packing,
+				      pixels, image->original.data ) ) {
+      return GL_FALSE;
+   }
+
+   /* Rescale the original image again if we have to.
+    */
+   if ( image->wScale > 1 || image->hScale > 1 ) {
+      assert( image->rescaled.data );
+      _mesa_rescale_teximage2d( texImage->TexFormat,
+				image->original.width, image->original.height,
+				image->rescaled.width, image->rescaled.height,
+				image->original.data, image->rescaled.data );
+   }
+
+   t->reloadImages = GL_TRUE; /* signal the image needs to be reloaded */
+   fxMesa->new_state |= TDFX_NEW_TEXTURE;  /* XXX this might be a bit much */
+
+   return GL_TRUE;
 }
 
 
-/*
- * Given an OpenGL internal texture format, return the corresponding
- * Glide internal texture format and MesaIntTexFormat.
- * If allow32bpp is true, we'll return 32-bit texel formats when
- * appropriate.
- */
-static void
-tdfxTexGetFormat(GLenum intFormatHint, GLboolean allow32bpp,
-                 GrTextureFormat_t *glideFormat,
-                 MesaIntTexFormat *mesaFormat,
-                 GLint *texelSize)
-{
-    switch (intFormatHint) {
-    case 1:
-    case GL_LUMINANCE:
-    case GL_LUMINANCE4:
-    case GL_LUMINANCE8:
-    case GL_LUMINANCE12:
-    case GL_LUMINANCE16:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_INTENSITY_8;
-        if (mesaFormat)
-            *mesaFormat = MESA_L8;
-        if (texelSize)
-            *texelSize = 1;
-        break;
-    case 2:
-    case GL_LUMINANCE_ALPHA:
-    case GL_LUMINANCE4_ALPHA4:
-    case GL_LUMINANCE6_ALPHA2:
-    case GL_LUMINANCE8_ALPHA8:
-    case GL_LUMINANCE12_ALPHA4:
-    case GL_LUMINANCE12_ALPHA12:
-    case GL_LUMINANCE16_ALPHA16:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ALPHA_INTENSITY_88;
-        if (mesaFormat)
-            *mesaFormat = MESA_A8_L8;
-        if (texelSize)
-            *texelSize = 2;
-        break;
-    case GL_INTENSITY:
-    case GL_INTENSITY4:
-    case GL_INTENSITY8:
-    case GL_INTENSITY12:
-    case GL_INTENSITY16:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ALPHA_8;
-        if (mesaFormat)
-            *mesaFormat = MESA_I8;
-        if (texelSize)
-            *texelSize = 1;
-       break;
-    case GL_ALPHA:
-    case GL_ALPHA4:
-    case GL_ALPHA8:
-    case GL_ALPHA12:
-    case GL_ALPHA16:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ALPHA_8;
-        if (mesaFormat)
-            *mesaFormat = MESA_A8;
-        if (texelSize)
-            *texelSize = 1;
-        break;
-    case GL_R3_G3_B2:
-    case GL_RGB4:
-    case GL_RGB5:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_RGB_565;
-        if (mesaFormat)
-            *mesaFormat = MESA_R5_G6_B5;
-        if (texelSize)
-            *texelSize = 2;
-        break;
-    case 3:
-    case GL_RGB:
-    case GL_RGB8:
-    case GL_RGB10:
-    case GL_RGB12:
-    case GL_RGB16:
-        if (allow32bpp) {
-            if (glideFormat)
-                *glideFormat = GR_TEXFMT_ARGB_8888;
-            if (mesaFormat)
-                *mesaFormat = MESA_FF_R8_G8_B8;
-            if (texelSize)
-                *texelSize = 4;
-        }
-        else {
-            if (glideFormat)
-                *glideFormat = GR_TEXFMT_RGB_565;
-            if (mesaFormat)
-                *mesaFormat = MESA_R5_G6_B5;
-            if (texelSize)
-                *texelSize = 2;
-        }
-        break;
-    case GL_RGBA2:
-    case GL_RGBA4:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ARGB_4444;
-        if (mesaFormat)
-            *mesaFormat = MESA_A4_R4_G4_B4;
-        if (texelSize)
-            *texelSize = 2;
-        break;
-    case 4:
-    case GL_RGBA:
-    case GL_RGBA8:
-    case GL_RGB10_A2:
-    case GL_RGBA12:
-    case GL_RGBA16:
-        if (allow32bpp) {
-            if (glideFormat)
-                *glideFormat = GR_TEXFMT_ARGB_8888;
-            if (mesaFormat)
-                *mesaFormat = MESA_A8_R8_G8_B8;
-            if (texelSize)
-                *texelSize = 4;
-        }
-        else {
-            if (glideFormat)
-                *glideFormat = GR_TEXFMT_ARGB_4444;
-            if (mesaFormat)
-                *mesaFormat = MESA_A4_R4_G4_B4;
-            if (texelSize)
-                *texelSize = 2;
-        }
-        break;
-    case GL_RGB5_A1:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ARGB_1555;
-        if (mesaFormat)
-            *mesaFormat = MESA_A1_R5_G5_B5;
-        if (texelSize)
-           *texelSize = 2;
-        break;
-    case GL_COLOR_INDEX:
-    case GL_COLOR_INDEX1_EXT:
-    case GL_COLOR_INDEX2_EXT:
-    case GL_COLOR_INDEX4_EXT:
-    case GL_COLOR_INDEX8_EXT:
-    case GL_COLOR_INDEX12_EXT:
-    case GL_COLOR_INDEX16_EXT:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_P_8;
-        if (mesaFormat)
-            *mesaFormat = MESA_C8;
-        if (texelSize)
-            *texelSize = 1;
-        break;
-    case GL_COMPRESSED_RGB_FXT1_3DFX:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ARGB_CMP_FXT1;
-        if (mesaFormat)
-            *mesaFormat = MESA_A8_R8_G8_B8;
-        if (texelSize)
-            *texelSize = 4;
-        break;
-    case GL_COMPRESSED_RGBA_FXT1_3DFX:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ARGB_CMP_FXT1;
-        if (mesaFormat)
-            *mesaFormat = MESA_A8_R8_G8_B8;
-        if (texelSize)
-            *texelSize = 4;
-        break;
-    default:
-        fprintf(stderr, "bad texture format in fxTexGetFormat() %d", intFormatHint);
-        break;
-    }
-}
 
+/* ================================================================
+ *
+ */
 
-/**********************************************************************/
-/**** NEW TEXTURE IMAGE FUNCTIONS                                  ****/
-/**********************************************************************/
+static void tdfxDDTexEnv( GLcontext *ctx, GLenum target,
+			  GLenum pname, const GLfloat *param )
+{
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
 
-static FxBool TexusFatalError = FXFALSE;
-static FxBool TexusError = FXFALSE;
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_API ) {
+      if ( param ) {
+	 fprintf( stderr, __FUNCTION__"( %x, %x )\n", pname, (GLint)(*param) );
+      } else {
+	 fprintf( stderr, __FUNCTION__"( %x )\n", pname );
+      }
+   }
 
-#define TX_DITHER_NONE                                  0x00000000
+   fxMesa->new_state |= TDFX_NEW_TEXTURE;
+}
 
-static void
-fxTexusError(const char *string, FxBool fatal)
+static void tdfxDDTexParameter( GLcontext *ctx, GLenum target,
+				struct gl_texture_object *tObj,
+				GLenum pname, const GLfloat *params )
 {
-    gl_problem(NULL, string);
-   /*
-    * Just propagate the fatal value up.
-    */
-    TexusError = FXTRUE;
-    TexusFatalError = fatal;
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   GLenum param = (GLenum) (GLint) params[0];
+   tdfxTexObjPtr t;
+
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, __FUNCTION__ "( %d, %p, %x, %x )\n",
+	       tObj->Name, tObj->DriverData, pname, param );
+   }
+
+   if ( target != GL_TEXTURE_2D )
+      return;
+
+   if ( !tObj->DriverData )
+      tObj->DriverData = tdfxAllocTexObj( fxMesa );
+
+   t = TDFX_TEXTURE_DATA(tObj);
+
+   switch ( pname ) {
+   case GL_TEXTURE_MIN_FILTER:
+      switch ( param ) {
+      case GL_NEAREST:
+	 t->mmMode = GR_MIPMAP_DISABLE;
+	 t->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
+	 t->LODblend = FXFALSE;
+	 break;
+
+      case GL_LINEAR:
+	 t->mmMode = GR_MIPMAP_DISABLE;
+	 t->minFilt = GR_TEXTUREFILTER_BILINEAR;
+	 t->LODblend = FXFALSE;
+	 break;
+
+      case GL_NEAREST_MIPMAP_LINEAR:
+	 if ( TDFX_IS_NAPALM(fxMesa) ) {
+	    if ( fxMesa->numTMUs > 1 ) {
+	       t->mmMode = GR_MIPMAP_NEAREST;
+	       t->LODblend = FXTRUE;
+	    } else {
+	       t->mmMode = GR_MIPMAP_NEAREST_DITHER;
+	       t->LODblend = FXFALSE;
+	    }
+	    t->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
+	    break;
+	 }
+	 /* XXX Voodoo3/Banshee mipmap blending seems to produce
+	  * incorrectly filtered colors for the smallest mipmap levels.
+	  * To work-around we fall-through here and use a different filter.
+	  */
+      case GL_NEAREST_MIPMAP_NEAREST:
+	 t->mmMode = GR_MIPMAP_NEAREST;
+	 t->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
+	 t->LODblend = FXFALSE;
+	 break;
+
+      case GL_LINEAR_MIPMAP_LINEAR:
+	 if ( TDFX_IS_NAPALM(fxMesa) ) {
+	    if ( fxMesa->numTMUs > 1 ) {
+	       t->mmMode = GR_MIPMAP_NEAREST;
+	       t->LODblend = FXTRUE;
+	    } else {
+	       t->mmMode = GR_MIPMAP_NEAREST_DITHER;
+	       t->LODblend = FXFALSE;
+	    }
+	    t->minFilt = GR_TEXTUREFILTER_BILINEAR;
+	    break;
+	 }
+	 /* XXX Voodoo3/Banshee mipmap blending seems to produce
+	  * incorrectly filtered colors for the smallest mipmap levels.
+	  * To work-around we fall-through here and use a different filter.
+	  */
+      case GL_LINEAR_MIPMAP_NEAREST:
+	 t->mmMode = GR_MIPMAP_NEAREST;
+	 t->minFilt = GR_TEXTUREFILTER_BILINEAR;
+	 t->LODblend = FXFALSE;
+	 break;
+      default:
+	 break;
+      }
+      tdfxRevalidateTexture( ctx, tObj );
+      fxMesa->new_state |= TDFX_NEW_TEXTURE;
+      break;
+
+   case GL_TEXTURE_MAG_FILTER:
+      switch ( param ) {
+      case GL_NEAREST:
+	 t->magFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
+	 break;
+      case GL_LINEAR:
+	 t->magFilt = GR_TEXTUREFILTER_BILINEAR;
+	 break;
+      default:
+	 break;
+      }
+      fxMesa->new_state |= TDFX_NEW_TEXTURE;
+      break;
+
+   case GL_TEXTURE_WRAP_S:
+      switch ( param ) {
+      case GL_CLAMP:
+	 t->sClamp = GR_TEXTURECLAMP_CLAMP;
+	 break;
+      case GL_REPEAT:
+	 t->sClamp = GR_TEXTURECLAMP_WRAP;
+	 break;
+      default:
+	 break;
+      }
+      fxMesa->new_state |= TDFX_NEW_TEXTURE;
+      break;
+
+   case GL_TEXTURE_WRAP_T:
+      switch ( param ) {
+      case GL_CLAMP:
+	 t->tClamp = GR_TEXTURECLAMP_CLAMP;
+	 break;
+      case GL_REPEAT:
+	 t->tClamp = GR_TEXTURECLAMP_WRAP;
+	 break;
+      default:
+	 break;
+      }
+      fxMesa->new_state |= TDFX_NEW_TEXTURE;
+      break;
+
+   case GL_TEXTURE_BASE_LEVEL:
+      tdfxRevalidateTexture( ctx, tObj );
+      break;
+
+   case GL_TEXTURE_MAX_LEVEL:
+      tdfxRevalidateTexture( ctx, tObj );
+      break;
+
+   case GL_TEXTURE_BORDER_COLOR:
+      /* TO DO */
+      break;
+   case GL_TEXTURE_MIN_LOD:
+      /* TO DO */
+      break;
+   case GL_TEXTURE_MAX_LOD:
+      /* TO DO */
+      break;
+
+   default:
+      break;
+   }
 }
 
-GLboolean
-tdfxDDTexImage2D(GLcontext * ctx, GLenum target, GLint level,
-                 GLenum format, GLenum type, const GLvoid * pixels,
-                 const struct gl_pixelstore_attrib * packing,
-                 struct gl_texture_object * texObj,
-                 struct gl_texture_image * texImage,
-                 GLboolean * retainInternalCopy)
+static void tdfxDDBindTexture( GLcontext *ctx, GLenum target,
+			       struct gl_texture_object *tObj )
 {
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    const GLboolean allow32bpt = TDFX_IS_NAPALM(fxMesa);
-    GrTextureFormat_t gldformat;
-    tdfxTexInfo *ti;
-    tdfxMipMapLevel *mml;
-    GLint dstWidth, dstHeight, wScale, hScale, texelSize, dstStride;
-    MesaIntTexFormat intFormat;
-    GLboolean isCompressedFormat;
-    GLint texsize;
-    void *uncompressedImage;
-
-    /*
-    printf("TexImage id=%d int 0x%x  format 0x%x  type 0x%x  %dx%d\n",
-           texObj->Name, texImage->IntFormat, format, type,
-           texImage->Width, texImage->Height);
-    */
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   tdfxTexObjPtr t;
 
-    isCompressedFormat = texImage->IsCompressed;
-    if (target != GL_TEXTURE_2D || texImage->Border > 0)
-        return GL_FALSE;
-
-    if (!texObj->DriverData)
-        texObj->DriverData = fxAllocTexObjData(fxMesa);
-
-    ti = TDFX_TEXTURE_DATA(texObj);
-    mml = &ti->mipmapLevel[level];
-
-    /* Determine the appropriate GL internal texel format, Mesa internal
-     * texel format, and texelSize (bytes) given the user's internal
-     * texture format hint.
-     */
-    tdfxTexGetFormat(texImage->IntFormat, allow32bpt,
-                     &gldformat, &intFormat, &texelSize);
-
-    /* Determine width and height scale factors for texture.
-     * Remember, Glide is limited to 8:1 aspect ratios.
-     */
-    tdfxTexGetInfo(ctx,
-                   texImage->Width, texImage->Height,
-                   NULL,       /* lod level          */
-                   NULL,       /* aspect ratio       */
-                   NULL, NULL, /* sscale, tscale     */
-                   &wScale, &hScale);
-    dstWidth = texImage->Width * wScale;
-    dstHeight = texImage->Height * hScale;
-    if (isCompressedFormat) {
-        texsize = tdfxDDCompressedImageSize(ctx,
-                                            texImage->IntFormat,
-                                            2,
-                                            texImage->Width,
-                                            texImage->Height,
-                                            1);
-    }
-    else {
-        texsize = dstWidth * dstHeight * texelSize;
-    }
-
-   /*
-    * If the image is not compressed, this doesn't
-    * matter, but it might as well have a sensible
-    * value, and it might save a failure later on.
-    */
-    texImage->CompressedSize = texsize;
-    /* housekeeping */
-    _mesa_set_teximage_component_sizes(intFormat, texImage);
-
-    /*
-     * allocate new storage for texture image, if needed.
-     * This conditional wants to set uncompressedImage to
-     * point to the uncompressed image, and mml->data to
-     * the texture data.  If the image is uncompressed,
-     * these are identical.  If the image is not compressed,
-     * these are different.
-     */
-    if (!mml->data || mml->glideFormat != gldformat ||
-        mml->width != dstWidth || mml->height != dstHeight ||
-        texsize != mml->dataSize ) {
-        if (mml->data) {
-            FREE(mml->data);
-        }
-        uncompressedImage
-            = (void *)MALLOC(dstWidth * dstHeight * texelSize);
-        if (!uncompressedImage) {
-            return GL_FALSE;
-        }
-        if (isCompressedFormat) {
-            mml->data = MALLOC(texsize);
-            if (!mml->data) {
-                FREE(uncompressedImage);
-                return GL_FALSE;
-            }
-        } else {
-            mml->data = uncompressedImage;
-        }
-        mml->texelSize = texelSize;
-        mml->glideFormat = gldformat;
-        mml->width = dstWidth;
-        mml->height = dstHeight;
-        mml->dataSize = texsize;
-        ti->info.format = gldformat;
-        tdfxTMMoveOutTM(fxMesa, texObj);
-        /*tdfxTexInvalidate(ctx, texObj);*/
-    }
-    else {
-       /*
-        * Here we don't have to allocate anything, but we
-        * do have to point uncompressedImage to the uncompressed
-        * data.
-        */
-        if (isCompressedFormat) {
-            uncompressedImage
-                = (void *)MALLOC(dstWidth * dstHeight * texelSize);
-            if (!uncompressedImage) {
-                return GL_FALSE;
-            }
-        } else {
-            uncompressedImage = mml->data;
-        }
-    }
-
-    dstStride = dstWidth * texelSize;
-
-    /* store the texture image into uncompressedImage */
-    if (!_mesa_convert_teximage(intFormat,
-                                dstWidth, dstHeight,
-                                uncompressedImage,
-                                dstStride,
-                                texImage->Width, texImage->Height,
-                                format, type, pixels, packing)) {
-        /*printf("convert failed\n");*/
-        return GL_FALSE;  /* not necessarily an error */
-    }
-
-    /*
-     * Now compress it if necessary.
-     */
-    if (isCompressedFormat) {
-        TxErrorCallbackFnc_t oldErrorCallback;
-        (*txErrorSetCallbackProc)(fxTexusError, &oldErrorCallback);
-        (*txImgQuantizeProc)((char *)mml->data,
-                             (char *)uncompressedImage,
-                             texImage->Width,
-                             texImage->Height,
-                             gldformat,
-                             TX_DITHER_NONE);
-        (*txErrorSetCallbackProc)(oldErrorCallback, NULL);
-        if (uncompressedImage != mml->data) {
-            /*
-             * We do not need this any more, errors or no.
-             */
-            FREE(uncompressedImage);
-        }
-        TexusError = FXFALSE;
-        if (TexusFatalError) {
-            FREE(mml->data);
-            mml->data = NULL;
-            TexusFatalError = FXFALSE;
-            return GL_FALSE;
-        }
-    }
-
-
-    RevalidateTexture(ctx, texObj);
-
-    ti->reloadImages = GL_TRUE;
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;
-
-    *retainInternalCopy = GL_FALSE;
-    return GL_TRUE;
-}
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, __FUNCTION__ "( %d, %p )\n",
+	       tObj->Name, tObj->DriverData );
+   }
 
+   if ( target != GL_TEXTURE_2D )
+      return;
 
-GLboolean
-tdfxDDTexSubImage2D(GLcontext * ctx, GLenum target, GLint level,
-                    GLint xoffset, GLint yoffset,
-                    GLsizei width, GLsizei height,
-                    GLenum format, GLenum type, const GLvoid * pixels,
-                    const struct gl_pixelstore_attrib * packing,
-                    struct gl_texture_object * texObj,
-                    struct gl_texture_image * texImage)
-{
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    tdfxTexInfo *ti;
-    GLint wscale, hscale, dstStride = 0;
-    tdfxMipMapLevel *mml;
-    GLboolean result;
-    void *uncompressedImage = (void *)0;
-    FxU32 uncompressedSize;
-    TxErrorCallbackFnc_t oldErrorCallback;
-
-    if (target != GL_TEXTURE_2D)
-        return GL_FALSE;
-
-    if (!texObj->DriverData)
-        return GL_FALSE;
-
-    /*
-    printf("TexSubImage id=%d lvl=%d int=0x%x format=0x%x type=0x%x  x=%d y=%d w=%d h=%d fullW=%d fullH=%d\n",
-           texObj->Name, level,
-           texImage->IntFormat, format, type, xoffset, yoffset, width, height,
-           texImage->Width, texImage->Height);
-    */
+   if ( !tObj->DriverData )
+      tObj->DriverData = tdfxAllocTexObj( fxMesa );
 
-    ti = TDFX_TEXTURE_DATA(texObj);
-    mml = &ti->mipmapLevel[level];
-
-    tdfxTexGetInfo(ctx, texImage->Width, texImage->Height, NULL, NULL,
-                   NULL, NULL, &wscale, &hscale);
-
-    /*
-     * Must have an existing texture image!
-     */
-    assert(mml->data);
-
-    switch (mml->glideFormat) {
-    case GR_TEXFMT_INTENSITY_8:
-        dstStride = mml->width;
-        result = _mesa_convert_texsubimage(MESA_I8, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    case GR_TEXFMT_ALPHA_8:
-        dstStride = mml->width;
-        result = _mesa_convert_texsubimage(MESA_A8, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    case GR_TEXFMT_P_8:
-        dstStride = mml->width;
-        result = _mesa_convert_texsubimage(MESA_C8, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    case GR_TEXFMT_ALPHA_INTENSITY_88:
-        dstStride = mml->width * 2;
-        result = _mesa_convert_texsubimage(MESA_A8_L8, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    case GR_TEXFMT_RGB_565:
-        dstStride = mml->width * 2;
-        result = _mesa_convert_texsubimage(MESA_R5_G6_B5, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    case GR_TEXFMT_ARGB_4444:
-        dstStride = mml->width * 2;
-        result = _mesa_convert_texsubimage(MESA_A4_R4_G4_B4, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    case GR_TEXFMT_ARGB_CMP_FXT1:
-        /*
-         * There are some special legality constraints for compressed
-         * textures.
-         */
-        if ((xoffset != texImage->Border)
-            || (yoffset != texImage->Border)) {
-            gl_error( ctx,
-                      GL_INVALID_OPERATION,
-                      "glTexSubImage2D(offset)" );
-            return GL_FALSE;
-        }
-        if ((width != texImage->Width)
-            || (height != texImage->Height)) {
-            gl_error( ctx,
-                      GL_INVALID_VALUE,
-                      "glTexSubImage2D(image size)" );
-            return GL_FALSE;
-        }
-        /*
-         * The width and height have to be multiples of
-         * 8 and 4 respectively.
-         */
-        width  = (mml->width  + 0x7) &~ 0x7;
-        height = (mml->height + 0x3) &~ 0x3;
-        /*
-         * A texel is 8888 for this format.
-         */
-        uncompressedSize = mml->width * mml->height * 4;
-        uncompressedImage = (void *)MALLOC(uncompressedSize);
-        /*
-         * Convert the data.
-         */
-        dstStride = mml->width * 4;
-        result = _mesa_convert_texsubimage(MESA_A8_R8_G8_B8, xoffset, yoffset,
-                                           mml->width, mml->height, uncompressedImage,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        if (!result) {
-            FREE(uncompressedImage);
-            printf("TexSubImage convert failed\n");
-            return GL_FALSE;
-        }
-        /*
-         * Now that we have converted the data, then compress it.
-         */
-        (*txErrorSetCallbackProc)(fxTexusError, &oldErrorCallback);
-        (*txImgQuantizeProc)((char *)mml->data,
-                            (char *)uncompressedImage,
-                            mml->width,
-                            mml->height,
-                            mml->glideFormat,
-                            TX_DITHER_NONE);
-        (*txErrorSetCallbackProc)(oldErrorCallback, NULL);
-        result = TexusFatalError;
-        TexusFatalError = TexusError = FXFALSE;
-        /*
-         * We don't need this any more.
-         */
-        FREE(uncompressedImage);
-        break;
-    case GR_TEXFMT_ARGB_8888:
-        {
-            MesaIntTexFormat intFormat;
-            if (texImage->Format == GL_RGB) {
-                /* An RGB image padded out to 4 bytes/texel */
-                intFormat = MESA_FF_R8_G8_B8;
-            }
-            else {
-                intFormat = MESA_A8_R8_G8_B8;
-            }
-            dstStride = mml->width * 4;
-            result = _mesa_convert_texsubimage(intFormat, xoffset, yoffset,
-                                            mml->width, mml->height, mml->data,
-                                            dstStride, width, height,
-                                            texImage->Width, texImage->Height,
-                                            format, type, pixels, packing);
-        }
-        break;
-    case GR_TEXFMT_ARGB_1555:
-        dstStride = mml->width * 2;
-        result = _mesa_convert_texsubimage(MESA_A1_R5_G5_B5, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    default:
-        gl_problem(NULL, "tdfx driver: fxTexBuildSubImageMap() bad format");
-        result = GL_FALSE;
-    }
-
-    if (!result) {
-        return GL_FALSE;
-    }
-
-    ti->reloadImages = GL_TRUE; /* signal the image needs to be reloaded */
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;  /* XXX this might be a bit much */
-
-    return GL_TRUE;
+   t = TDFX_TEXTURE_DATA(tObj);
+   t->lastTimeUsed = fxMesa->texBindNumber++;
+
+   fxMesa->new_state |= TDFX_NEW_TEXTURE;
 }
 
+static void tdfxDDDeleteTexture( GLcontext *ctx,
+				 struct gl_texture_object *tObj )
+{
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
 
-/**********************************************************************/
-/**** COMPRESSED TEXTURE IMAGE FUNCTIONS                           ****/
-/**********************************************************************/
+   LOCK_HARDWARE( fxMesa );
+   tdfxTMFreeTextureLocked( fxMesa, tObj );
+   UNLOCK_HARDWARE( fxMesa );
 
-GLboolean
-tdfxDDCompressedTexImage2D( GLcontext *ctx, GLenum target,
-                            GLint level, GLsizei imageSize,
-                            const GLvoid *data,
-                            struct gl_texture_object *texObj,
-                            struct gl_texture_image *texImage,
-                            GLboolean *retainInternalCopy)
-{
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    const GLboolean allow32bpt = TDFX_IS_NAPALM(fxMesa);
-    GrTextureFormat_t gldformat;
-    tdfxTexInfo *ti;
-    tdfxMipMapLevel *mml;
-    GLint dstWidth, dstHeight, wScale, hScale, texelSize;
-    MesaIntTexFormat intFormat;
-    GLboolean isCompressedFormat;
-    GLsizei texsize;
-
-    if (target != GL_TEXTURE_2D || texImage->Border > 0)
-        return GL_FALSE;
-
-    if (!texObj->DriverData)
-        texObj->DriverData = fxAllocTexObjData(fxMesa);
-
-    ti = TDFX_TEXTURE_DATA(texObj);
-    mml = &ti->mipmapLevel[level];
-
-    isCompressedFormat = tdfxDDIsCompressedGlideFormatMacro(texImage->IntFormat);
-    if (!isCompressedFormat) {
-        gl_error( ctx, GL_INVALID_ENUM, "glCompressedTexImage2D(format)" );
-        return GL_FALSE;
-    }
-    /* Determine the apporpriate GL internal texel format, Mesa internal
-     * texel format, and texelSize (bytes) given the user's internal
-     * texture format hint.
-     */
-    tdfxTexGetFormat(texImage->IntFormat, allow32bpt,
-                     &gldformat, &intFormat, &texelSize);
-
-    /* Determine width and height scale factors for texture.
-     * Remember, Glide is limited to 8:1 aspect ratios.
-     */
-    tdfxTexGetInfo(ctx,
-                   texImage->Width, texImage->Height,
-                   NULL,       /* lod level          */
-                   NULL,       /* aspect ratio       */
-                   NULL, NULL, /* sscale, tscale     */
-                   &wScale, &hScale);
-    dstWidth = texImage->Width * wScale;
-    dstHeight = texImage->Height * hScale;
-    /* housekeeping */
-    _mesa_set_teximage_component_sizes(intFormat, texImage);
-
-    texsize = tdfxDDCompressedImageSize(ctx,
-                                        texImage->IntFormat,
-                                        2,
-                                        texImage->Width,
-                                        texImage->Height,
-                                        1);
-    if (texsize != imageSize) {
-        gl_error(ctx,
-                 GL_INVALID_VALUE,
-                 "glCompressedTexImage2D(texsize)");
-        return GL_FALSE;
-    }
-
-    /* allocate new storage for texture image, if needed */
-    if (!mml->data || mml->glideFormat != gldformat ||
-        mml->width != dstWidth || mml->height != dstHeight ||
-        texsize != mml->dataSize) {
-        if (mml->data) {
-            FREE(mml->data);
-        }
-        mml->data = MALLOC(texsize);
-        if (!mml->data) {
-            return GL_FALSE;
-        }
-        mml->texelSize = texelSize;
-        mml->glideFormat = gldformat;
-        mml->width = dstWidth;
-        mml->height = dstHeight;
-        tdfxTMMoveOutTM(fxMesa, texObj);
-        /*tdfxTexInvalidate(ctx, texObj);*/
-    }
-
-    /* save the texture data */
-    MEMCPY(mml->data, data, imageSize);
-
-    RevalidateTexture(ctx, texObj);
-
-    ti->reloadImages = GL_TRUE;
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;
-
-    *retainInternalCopy = GL_FALSE;
-    return GL_TRUE;
+   fxMesa->new_state |= TDFX_NEW_TEXTURE;
 }
 
-GLboolean
-tdfxDDCompressedTexSubImage2D( GLcontext *ctx, GLenum target,
-                               GLint level, GLint xoffset,
-                               GLint yoffset, GLsizei width,
-                               GLint height, GLenum format,
-                               GLsizei imageSize, const GLvoid *data,
-                               struct gl_texture_object *texObj,
-                               struct gl_texture_image *texImage )
+static GLboolean tdfxDDIsTextureResident( GLcontext *ctx,
+					  struct gl_texture_object *tObj )
 {
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    tdfxTexInfo *ti;
-    tdfxMipMapLevel *mml;
-
-    /*
-     * We punt if we are not replacing the entire image.  This
-     * is allowed by the spec.
-     */
-    if ((xoffset != 0) && (yoffset != 0)
-        && (width != texImage->Width)
-        && (height != texImage->Height)) {
-        return GL_FALSE;
-    }
-
-    ti = TDFX_TEXTURE_DATA(texObj);
-    mml = &ti->mipmapLevel[level];
-    if (imageSize != mml->dataSize) {
-        return GL_FALSE;
-    }
-    MEMCPY(data, mml->data, imageSize);
-
-    ti->reloadImages = GL_TRUE;
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;
-
-    return GL_TRUE;
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+
+   return ( t && t->isInTM );
 }
 
 
-#if	0
+/* Convert a gl_color_table texture palette to Glide's format.
+ */
 static void
-PrintTexture(int w, int h, int c, const GLubyte * data)
+tdfxConvertPalette( FxU32 data[256], const struct gl_color_table *table )
 {
-    int i, j;
-    for (i = 0; i < h; i++) {
-        for (j = 0; j < w; j++) {
-            if (c == 2)
-                printf("%02x %02x  ", data[0], data[1]);
-            else if (c == 3)
-                printf("%02x %02x %02x  ", data[0], data[1], data[2]);
-            data += c;
-        }
-        printf("\n");
-    }
+   const GLubyte *tableUB = (const GLubyte *) table->Table;
+   GLint width = table->Size;
+   FxU32 r, g, b, a;
+   GLint i;
+
+   ASSERT( table->TableType == GL_UNSIGNED_BYTE );
+
+   switch ( table->Format ) {
+   case GL_RGBA:
+      for ( i = 0 ; i < width ; i++ ) {
+	 r = tableUB[i * 4 + 0];
+	 g = tableUB[i * 4 + 1];
+	 b = tableUB[i * 4 + 2];
+	 a = tableUB[i * 4 + 3];
+	 data[i] = PACK_COLOR_8888( a, r, g, b );
+      }
+      break;
+   case GL_RGB:
+      for ( i = 0 ; i < width ; i++ ) {
+	 r = tableUB[i * 3 + 0];
+	 g = tableUB[i * 3 + 1];
+	 b = tableUB[i * 3 + 2];
+	 a = 255;
+	 data[i] = PACK_COLOR_8888( a, r, g, b );
+      }
+      break;
+   case GL_LUMINANCE:
+      for ( i = 0 ; i < width ; i++ ) {
+	 r = tableUB[i];
+	 g = tableUB[i];
+	 b = tableUB[i];
+	 a = 255;
+	 data[i] = PACK_COLOR_8888( a, r, g, b );
+      }
+      break;
+   case GL_ALPHA:
+      for ( i = 0 ; i < width ; i++ ) {
+	 r = g = b = 255;
+	 a = tableUB[i];
+	 data[i] = PACK_COLOR_8888( a, r, g, b );
+      }
+      break;
+   case GL_LUMINANCE_ALPHA:
+      for ( i = 0 ; i < width ; i++ ) {
+	 r = g = b = tableUB[i * 2 + 0];
+	 a = tableUB[i * 2 + 1];
+	 data[i] = PACK_COLOR_8888( a, r, g, b );
+      }
+      break;
+   case GL_INTENSITY:
+      for ( i = 0 ; i < width ; i++ ) {
+	 r = tableUB[i];
+	 g = tableUB[i];
+	 b = tableUB[i];
+	 a = tableUB[i];
+	 data[i] = PACK_COLOR_8888( a, r, g, b );
+      }
+      break;
+   }
 }
-#endif
 
-
-GLboolean
-tdfxDDTestProxyTexImage(GLcontext *ctx, GLenum target,
-                        GLint level, GLint internalFormat,
-                        GLenum format, GLenum type,
-                        GLint width, GLint height,
-                        GLint depth, GLint border)
+static void
+tdfxDDTexturePalette( GLcontext *ctx, struct gl_texture_object *tObj )
 {
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-
-    switch (target) {
-    case GL_PROXY_TEXTURE_1D:
-        return GL_TRUE;  /* software rendering */
-    case GL_PROXY_TEXTURE_2D:
-        {
-            struct gl_texture_object *tObj;
-            tdfxTexInfo *ti;
-            int memNeeded;
-
-            tObj = ctx->Texture.Proxy2D;
-            if (!tObj->DriverData)
-                tObj->DriverData = fxAllocTexObjData(fxMesa);
-            ti = TDFX_TEXTURE_DATA(tObj);
-
-            /* assign the parameters to test against */
-            tObj->Image[level]->Width = width;
-            tObj->Image[level]->Height = height;
-            tObj->Image[level]->Border = border;
-            tObj->Image[level]->IntFormat = internalFormat;
-            if (level == 0) {
-               /* don't use mipmap levels > 0 */
-               tObj->MinFilter = tObj->MagFilter = GL_NEAREST;
-            }
-            else {
-               /* test with all mipmap levels */
-               tObj->MinFilter = GL_LINEAR_MIPMAP_LINEAR;
-               tObj->MagFilter = GL_NEAREST;
-            }
-            RevalidateTexture(ctx, tObj);
-
-            /*
-            printf("small lodlog2 0x%x\n", ti->info.smallLodLog2);
-            printf("large lodlog2 0x%x\n", ti->info.largeLodLog2);
-            printf("aspect ratio 0x%x\n", ti->info.aspectRatioLog2);
-            printf("glide format 0x%x\n", ti->info.format);
-            printf("data %p\n", ti->info.data);
-            printf("lodblend %d\n", (int) ti->LODblend);
-            */
-
-            /* determine where texture will reside */
-            if (ti->LODblend && !shared->umaTexMemory) {
-                /* XXX GR_MIPMAPLEVELMASK_BOTH might not be right, but works */
-                memNeeded = FX_grTexTextureMemRequired_NoLock(
-                                        GR_MIPMAPLEVELMASK_BOTH, &(ti->info));
-            }
-            else {
-                /* XXX GR_MIPMAPLEVELMASK_BOTH might not be right, but works */
-                memNeeded = FX_grTexTextureMemRequired_NoLock(
-                                        GR_MIPMAPLEVELMASK_BOTH, &(ti->info));
-            }
-            /*
-            printf("Proxy test %d > %d\n", memNeeded, shared->totalTexMem[0]);
-            */
-            if (memNeeded > shared->totalTexMem[0])
-                return GL_FALSE;
-            else
-                return GL_TRUE;
-        }
-    case GL_PROXY_TEXTURE_3D:
-        return GL_TRUE;  /* software rendering */
-    default:
-        return GL_TRUE;  /* never happens, silence compiler */
-    }
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   tdfxTexObjPtr t;
+
+   if ( tObj ) {
+      /* Per-texture palette */
+      if ( !tObj->DriverData )
+	 tObj->DriverData = tdfxAllocTexObj(fxMesa);
+
+      t = TDFX_TEXTURE_DATA(tObj);
+      tdfxConvertPalette( t->palette.data, &tObj->Palette );
+      /*tdfxTexInvalidate( ctx, tObj );*/
+   } else {
+      /* Global texture palette */
+      tdfxConvertPalette( fxMesa->glbPalette.data, &ctx->Texture.Palette );
+   }
+
+   fxMesa->new_state |= TDFX_NEW_TEXTURE; /* XXX too heavy-handed */
 }
 
 
-/*
- * Return a texture image to Mesa.  This is either to satisfy
- * a glGetTexImage() call or to prepare for software texturing.
- */
-GLvoid *
-tdfxDDGetTexImage(GLcontext * ctx, GLenum target, GLint level,
-                  const struct gl_texture_object *texObj,
-                  GLenum * formatOut, GLenum * typeOut,
-                  GLboolean * freeImageOut)
-{
-    tdfxTexInfo *ti;
-    tdfxMipMapLevel *mml;
-
-    if (target != GL_TEXTURE_2D)
-        return NULL;
-
-    if (!texObj->DriverData)
-        return NULL;
-
-    ti = TDFX_TEXTURE_DATA(texObj);
-    mml = &ti->mipmapLevel[level];
-    if (mml->data) {
-        MesaIntTexFormat mesaFormat;
-        GLenum glFormat;
-        struct gl_texture_image *texImage = texObj->Image[level];
-        GLint srcStride;
-        void *uncompressedImage = NULL;
-
-        GLubyte *data =
-            (GLubyte *) MALLOC(texImage->Width * texImage->Height * 4);
-        if (!data)
-            return NULL;
-
-        uncompressedImage = (void *)mml->data;
-        switch (mml->glideFormat) {
-        case GR_TEXFMT_INTENSITY_8:
-            mesaFormat = MESA_I8;
-            glFormat = GL_INTENSITY;
-            srcStride = mml->width;
-            break;
-        case GR_TEXFMT_ALPHA_INTENSITY_88:
-            mesaFormat = MESA_A8_L8;
-            glFormat = GL_LUMINANCE_ALPHA;
-            srcStride = mml->width;
-            break;
-        case GR_TEXFMT_ALPHA_8:
-            if (texImage->Format == GL_INTENSITY) {
-                mesaFormat = MESA_I8;
-                glFormat = GL_INTENSITY;
-            }
-            else {
-                mesaFormat = MESA_A8;
-                glFormat = GL_ALPHA;
-            }
-            srcStride = mml->width;
-            break;
-        case GR_TEXFMT_RGB_565:
-            mesaFormat = MESA_R5_G6_B5;
-            glFormat = GL_RGB;
-            srcStride = mml->width * 2;
-            break;
-        case GR_TEXFMT_ARGB_8888:
-            mesaFormat = MESA_A8_R8_G8_B8;
-            glFormat = GL_RGBA;
-            srcStride = mml->width * 4;
-            break;
-        case GR_TEXFMT_ARGB_4444:
-            mesaFormat = MESA_A4_R4_G4_B4;
-            glFormat = GL_RGBA;
-            srcStride = mml->width * 2;
-            break;
-        case GR_TEXFMT_ARGB_1555:
-            mesaFormat = MESA_A1_R5_G5_B5;
-            glFormat = GL_RGBA;
-            srcStride = mml->width * 2;
-            break;
-        case GR_TEXFMT_P_8:
-            mesaFormat = MESA_C8;
-            glFormat = GL_COLOR_INDEX;
-            srcStride = mml->width;
-            break;
-        case GR_TEXFMT_ARGB_CMP_FXT1:
-            mesaFormat = MESA_A8_R8_G8_B8;
-            glFormat = GL_RGBA;
-            srcStride = mml->width * 4;
-           /*
-            * Allocate data for the uncompressed image,
-            * decompress the image.  The data will be deallocated
-            * after it is converted to the mesa format.
-            */
-            uncompressedImage = MALLOC(mml->width * mml->height * 4);
-            if (!uncompressedImage) {
-                gl_problem(NULL, "can't get memory in tdfxDDGetTexImage");
-                return NULL;
-            }
-            (*txImgDequantizeFXT1Proc)((FxU32 *)uncompressedImage,
-                                       (FxU32 *)mml->data,
-                                       mml->width,
-                                       mml->height);
-            break;
-        default:
-            gl_problem(NULL, "Bad glideFormat in tdfxDDGetTexImage");
-            return NULL;
-        }
-        _mesa_unconvert_teximage(mesaFormat, mml->width, mml->height,
-                                 uncompressedImage, srcStride, texImage->Width,
-                                 texImage->Height, glFormat, data);
-        if (uncompressedImage != mml->data) {
-            FREE(uncompressedImage);
-        }
-        *formatOut = glFormat;
-        *typeOut = GL_UNSIGNED_BYTE;
-        *freeImageOut = GL_TRUE;
-        return data;
-    }
-    else {
-        return NULL;
-    }
-}
 
-/*
- * This is called from _mesa_GetCompressedTexImage.  We just
- * copy out the compressed data.
- */
-void
-tdfxDDGetCompressedTexImage( GLcontext *ctx, GLenum target,
-                             GLint lod, void *image,
-                             const struct gl_texture_object *texObj,
-                             struct gl_texture_image *texImage )
-{
-    tdfxTexInfo *ti;
-    tdfxMipMapLevel *mml;
 
-    if (target != GL_TEXTURE_2D)
-        return;
+/**********************************************************************/
+/**** NEW TEXTURE IMAGE FUNCTIONS                                  ****/
+/**********************************************************************/
 
-    if (!texObj->DriverData)
-        return;
 
-    ti = TDFX_TEXTURE_DATA(texObj);
-    mml = &ti->mipmapLevel[lod];
-    if (mml->data) {
-        MEMCPY(image, mml->data, mml->dataSize);
-    }
-}
 
-/*
- * Calculate a specific texture format given a generic
- * texture format.
- */
-GLint
-tdfxDDSpecificCompressedTexFormat(GLcontext *ctx,
-                                  GLint internalFormat,
-                                  GLint numDimensions)
+#if 0
+static void
+PrintTexture(int w, int h, int c, const GLubyte * data)
 {
-    if (numDimensions != 2) {
-        return internalFormat;
-    }
-   /*
-    * If we don't have pointers to the functions, then
-    * we drop back to uncompressed format.  The logic
-    * in Mesa proper handles this for us.
-    *
-    * This is just to ease the transition to a Glide with
-    * the texus2 library.
-    */
-    if (!txImgQuantizeProc || !txImgDequantizeFXT1Proc) {
-        return internalFormat;
-    }
-    switch (internalFormat) {
-    case GL_COMPRESSED_RGB_ARB:
-        return GL_COMPRESSED_RGB_FXT1_3DFX;
-    case GL_COMPRESSED_RGBA_ARB:
-        return GL_COMPRESSED_RGBA_FXT1_3DFX;
-    }
-    return internalFormat;
+   int i, j;
+   for (i = 0; i < h; i++) {
+      for (j = 0; j < w; j++) {
+	 if (c == 2)
+	    printf("%02x %02x  ", data[0], data[1]);
+	 else if (c == 3)
+	    printf("%02x %02x %02x  ", data[0], data[1], data[2]);
+	 data += c;
+      }
+      printf("\n");
+   }
 }
+#endif
 
-/*
- * Calculate a specific texture format given a generic
- * texture format.
- */
-GLint
-tdfxDDBaseCompressedTexFormat(GLcontext *ctx,
-                              GLint internalFormat)
+
+static GLboolean
+tdfxDDTestProxyTexImage( GLcontext *ctx, GLenum target,
+			 GLint level, GLint internalFormat,
+			 GLenum format, GLenum type,
+			 GLint width, GLint height,
+			 GLint depth, GLint border )
 {
-    switch (internalFormat) {
-    case GL_COMPRESSED_RGB_FXT1_3DFX:
-        return GL_RGB;
-    case GL_COMPRESSED_RGBA_FXT1_3DFX:
-        return GL_RGBA;
-    }
-    return -1;
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+
+   switch (target) {
+   case GL_PROXY_TEXTURE_1D:
+      return GL_TRUE;  /* software rendering */
+   case GL_PROXY_TEXTURE_2D:
+   {
+      struct gl_texture_object *tObj;
+      tdfxTexObjPtr t;
+      int memNeeded;
+
+      tObj = ctx->Texture.Proxy2D;
+      if (!tObj->DriverData)
+	 tObj->DriverData = tdfxAllocTexObj(fxMesa);
+      t = TDFX_TEXTURE_DATA(tObj);
+
+      /* assign the parameters to test against */
+      tObj->Image[level]->Width = width;
+      tObj->Image[level]->Height = height;
+      tObj->Image[level]->Border = border;
+      tObj->Image[level]->IntFormat = internalFormat;
+      if (level == 0) {
+	 /* don't use mipmap levels > 0 */
+	 tObj->MinFilter = tObj->MagFilter = GL_NEAREST;
+      }
+      else {
+	 /* test with all mipmap levels */
+	 tObj->MinFilter = GL_LINEAR_MIPMAP_LINEAR;
+	 tObj->MagFilter = GL_NEAREST;
+      }
+      tdfxRevalidateTexture(ctx, tObj);
+
+      /*
+	printf("small lodlog2 0x%x\n", t->info.smallLodLog2);
+	printf("large lodlog2 0x%x\n", t->info.largeLodLog2);
+	printf("aspect ratio 0x%x\n", t->info.aspectRatioLog2);
+	printf("glide format 0x%x\n", t->info.format);
+	printf("data %p\n", t->info.data);
+	printf("lodblend %d\n", (int) t->LODblend);
+      */
+
+      /* determine where texture will reside */
+      if (t->LODblend && !tss->umaTexMemory) {
+	 /* XXX GR_MIPMAPLEVELMASK_BOTH might not be right, but works */
+	 memNeeded = FX_grTexTextureMemRequired_NoLock(
+	    GR_MIPMAPLEVELMASK_BOTH, &(t->info));
+      }
+      else {
+	 /* XXX GR_MIPMAPLEVELMASK_BOTH might not be right, but works */
+	 memNeeded = FX_grTexTextureMemRequired_NoLock(
+	    GR_MIPMAPLEVELMASK_BOTH, &(t->info));
+      }
+      /*
+	printf("Proxy test %d > %d\n", memNeeded, tss->totalTexMem[0]);
+      */
+      if (memNeeded > tss->totalTexMem[0])
+	 return GL_FALSE;
+      else
+	 return GL_TRUE;
+   }
+   case GL_PROXY_TEXTURE_3D:
+      return GL_TRUE;  /* software rendering */
+   default:
+      return GL_TRUE;  /* never happens, silence compiler */
+   }
 }
 
-/*
- * Tell us if an image is compressed.  The real work is done
- * in a macro, but we need to have a function to create a
- * function pointer.
+
+/* Return a texture image to Mesa.  This is either to satisfy
+ * a glGetTexImage() call or to prepare for software texturing.
  */
-GLboolean
-tdfxDDIsCompressedFormat(GLcontext *ctx, GLint internalFormat)
+static GLvoid *
+tdfxDDGetTexImage( GLcontext *ctx, GLenum target, GLint level,
+		   const struct gl_texture_object *texObj,
+		   GLenum *formatOut, GLenum *typeOut,
+		   GLboolean *freeImageOut )
 {
-    return tdfxDDIsCompressedFormatMacro(internalFormat);
+   const struct gl_texture_image *texImage = texObj->Image[level];
+   const struct gl_texture_format *texFormat = texImage->TexFormat;
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(texObj);
+   tdfxTexImagePtr image;
+   GLubyte *data;
+
+   if ( target != GL_TEXTURE_2D )
+      return NULL;
+   if ( !t )
+      return NULL;
+
+   image = &t->image[level];
+   if ( !image->original.data )
+      return NULL;
+
+   data = (GLubyte *) MALLOC( texImage->Width * texImage->Height * 4 );
+   if ( !data )
+      return NULL;
+
+   _mesa_unconvert_teximage2d( texFormat->IntFormat, texImage->Format,
+			       texImage->Width, texImage->Height,
+			       image->original.data, data );
+
+   *formatOut = texImage->Format;
+   *typeOut = GL_UNSIGNED_BYTE;
+   *freeImageOut = GL_TRUE;
+
+   return data;
 }
 
 
-/*
- * Calculate the image size of a compressed texture.
- *
- * The current compressed format, the FXT1 family, all
- * map 8x32 texel blocks into 128 bits.
- *
- * We return 0 if we can't calculate the size.
- *
- * Glide would report this out to us, but we don't have
- * exactly the right parameters.
- */
-GLsizei
-tdfxDDCompressedImageSize(GLcontext *ctx,
-                          GLenum intFormat,
-                          GLuint numDimensions,
-                          GLuint width,
-                          GLuint height,
-                          GLuint depth)
+void tdfxDDInitTextureFuncs( GLcontext *ctx )
 {
-    if (numDimensions != 2) {
-        return 0;
-    }
-    switch (intFormat) {
-    case GL_COMPRESSED_RGB_FXT1_3DFX:
-    case GL_COMPRESSED_RGBA_FXT1_3DFX:
-       /*
-        * Round height and width to multiples of 4 and 8,
-        * divide the resulting product by 32 to get the number
-        * of blocks, and multiply by 32 = 128/8 to get the.
-        * number of bytes required.  That is to say, just
-        * return the product.  Remember that we are returning
-        * bytes, not texels, so we have shrunk the texture
-        * by a factor of the texel size.
-        */
-        width = (width   + 0x7) &~ 0x7;
-        height = (height + 0x3) &~ 0x3;
-        return width * height;
-    }
-    return 0;
+   ctx->Driver.TexImage2D		= tdfxDDTexImage2D;
+   ctx->Driver.TexSubImage2D		= tdfxDDTexSubImage2D;
+   ctx->Driver.GetTexImage		= tdfxDDGetTexImage;
+   ctx->Driver.TexEnv			= tdfxDDTexEnv;
+   ctx->Driver.TexParameter		= tdfxDDTexParameter;
+   ctx->Driver.BindTexture		= tdfxDDBindTexture;
+   ctx->Driver.DeleteTexture		= tdfxDDDeleteTexture;
+   ctx->Driver.IsTextureResident	= tdfxDDIsTextureResident;
+   ctx->Driver.UpdateTexturePalette	= tdfxDDTexturePalette;
 }
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.h b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.h
index 95ef3b018..1132f3f35 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.h
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.h
@@ -35,123 +35,12 @@
  *
  */
 
-#ifndef _TDFX_TEX_H_
-#define _TDFX_TEX_H_
+#ifndef __TDFX_TEX_H__
+#define __TDFX_TEX_H__
 
+extern void tdfxUpdateTextureState( GLcontext *ctx );
+extern void tdfxUpdateTextureBinding( GLcontext *ctx );
 
-#include "texutil.h"
-
-
-#define tdfxDDIsCompressedFormatMacro(internalFormat) \
-    (((internalFormat) == GL_COMPRESSED_RGB_FXT1_3DFX) || \
-     ((internalFormat) == GL_COMPRESSED_RGBA_FXT1_3DFX))
-#define tdfxDDIsCompressedGlideFormatMacro(internalFormat) \
-    ((internalFormat) == GR_TEXFMT_ARGB_CMP_FXT1)
-
-
-
-extern void
-tdfxTexValidate(GLcontext * ctx, struct gl_texture_object *tObj);
-
-extern void
-tdfxDDBindTexture(GLcontext * ctx, GLenum target,
-                  struct gl_texture_object *tObj);
-
-extern void
-tdfxDDDeleteTexture(GLcontext * ctx, struct gl_texture_object *tObj);
-
-extern GLboolean
-tdfxDDIsTextureResident(GLcontext *ctx, struct gl_texture_object *tObj);
-
-extern void
-tdfxDDTexturePalette(GLcontext * ctx, struct gl_texture_object *tObj);
-
-#if 000 /* DEAD? */
-extern void
-fxDDTexUseGlobalPalette(GLcontext * ctx, GLboolean state);
-#endif
-
-extern void
-tdfxDDTexEnv(GLcontext * ctx, GLenum target, GLenum pname,
-             const GLfloat * param);
-
-extern void
-tdfxDDTexParameter(GLcontext * ctx, GLenum target,
-                   struct gl_texture_object *tObj,
-                   GLenum pname, const GLfloat * params);
-
-extern GLboolean
-tdfxDDTexImage2D(GLcontext * ctx, GLenum target, GLint level,
-                 GLenum format, GLenum type, const GLvoid * pixels,
-                 const struct gl_pixelstore_attrib * packing,
-                 struct gl_texture_object * texObj,
-                 struct gl_texture_image * texImage,
-                 GLboolean * retainInternalCopy);
-
-extern GLboolean
-tdfxDDTexSubImage2D(GLcontext * ctx, GLenum target, GLint level,
-                    GLint xoffset, GLint yoffset,
-                    GLsizei width, GLsizei height,
-                    GLenum format, GLenum type, const GLvoid * pixels,
-                    const struct gl_pixelstore_attrib * packing,
-                    struct gl_texture_object * texObj,
-                    struct gl_texture_image * texImage);
-
-extern GLboolean
-tdfxDDCompressedTexImage2D( GLcontext *ctx, GLenum target,
-                            GLint level, GLsizei imageSize,
-                            const GLvoid *data,
-                            struct gl_texture_object *texObj,
-                            struct gl_texture_image *texImage,
-                            GLboolean *retainInternalCopy);
-
-extern GLboolean
-tdfxDDCompressedTexSubImage2D( GLcontext *ctx, GLenum target,
-                               GLint level, GLint xoffset,
-                               GLint yoffset, GLsizei width,
-                               GLint height, GLenum format,
-                               GLsizei imageSize, const GLvoid *data,
-                               struct gl_texture_object *texObj,
-                               struct gl_texture_image *texImage );
-
-extern GLboolean
-tdfxDDTestProxyTexImage(GLcontext *ctx, GLenum target,
-                        GLint level, GLint internalFormat,
-                        GLenum format, GLenum type,
-                        GLint width, GLint height,
-                        GLint depth, GLint border);
-
-extern GLvoid *
-tdfxDDGetTexImage(GLcontext * ctx, GLenum target, GLint level,
-                  const struct gl_texture_object *texObj,
-                  GLenum * formatOut, GLenum * typeOut,
-                  GLboolean * freeImageOut);
-
-extern void
-tdfxDDGetCompressedTexImage( GLcontext *ctx, GLenum target,
-                             GLint lod, void *image,
-                             const struct gl_texture_object *texObj,
-                             struct gl_texture_image *texImage );
-
-extern GLint
-tdfxDDSpecificCompressedTexFormat(GLcontext *ctx,
-                                  GLint internalFormat,
-                                  GLint numDimensions);
-
-extern GLint
-tdfxDDBaseCompressedTexFormat(GLcontext *ctx,
-                              GLint internalFormat);
-
-extern GLboolean
-tdfxDDIsCompressedFormat(GLcontext *ctx, GLint internalFormat);
-
-extern GLsizei
-tdfxDDCompressedImageSize(GLcontext *ctx,
-                          GLenum intFormat,
-                          GLuint numDimensions,
-                          GLuint width,
-                          GLuint height,
-                          GLuint depth);
-
+extern void tdfxDDInitTextureFuncs( GLcontext *ctx );
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.c
index 97fe9d4c6..8ca93f91a 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.c
@@ -39,930 +39,882 @@
 #include "tdfx_tex.h"
 #include "tdfx_texman.h"
 
+#define BAD_ADDRESS	((FxU32) -1)
 
-#define BAD_ADDRESS ((FxU32) -1)
-
-
-#if 0 /* DEBUG use */
-/*
- * Verify the consistancy of the texture memory manager.
+/* Verify the consistancy of the texture memory manager.
  * This involves:
  *    Traversing all texture objects and computing total memory used.
  *    Traverse the free block list and computing total memory free.
  *    Compare the total free and total used amounts to the total memory size.
  *    Make various assertions about the results.
  */
-static void
-VerifyFreeList(tdfxContextPtr fxMesa, FxU32 tmu)
+static void tdfxTMVerifyFreeList( tdfxContextPtr fxMesa, FxU32 unit )
 {
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    tdfxMemRange *block;
-    int prevStart = -1, prevEnd = -1;
-    int totalFree = 0;
-    int numObj = 0, numRes = 0;
-    int totalUsed = 0;
-
-    for (block = shared->tmFree[tmu]; block; block = block->next) {
-       assert( block->endAddr > 0 );
-       assert( block->startAddr <= shared->totalTexMem[tmu] );
-       assert( block->endAddr <= shared->totalTexMem[tmu] );
-       assert( (int) block->startAddr > prevStart );
-       assert( (int) block->startAddr >= prevEnd );
-       prevStart = (int) block->startAddr;
-       prevEnd = (int) block->endAddr;
-       totalFree += (block->endAddr - block->startAddr);
-    }
-    assert(totalFree == shared->freeTexMem[tmu]);
-
-    {
-       struct gl_texture_object *obj;
-       for (obj = mesaShared->TexObjectList; obj; obj = obj->Next) {
-          tdfxTexInfo *ti = TDFX_TEXTURE_DATA(obj);
-          numObj++;
-          if (ti) {
-             if (ti->isInTM) {
-                numRes++;
-                assert(ti->tm[0]);
-                if (ti->tm[tmu])
-                   totalUsed += (ti->tm[tmu]->endAddr - ti->tm[tmu]->startAddr);
-             }
-             else {
-                assert(!ti->tm[0]);
-             }
-          }
-       }
-    }
-
-    printf("totalFree: %d  totalUsed: %d  totalMem: %d #objs=%d  #res=%d\n",
-           shared->freeTexMem[tmu], totalUsed, shared->totalTexMem[tmu],
-           numObj, numRes);
-
-    assert(totalUsed + totalFree == shared->totalTexMem[tmu]);
-}
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   struct gl_texture_object *texObj;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+   tdfxMemRange *block;
+   int prevStart = -1, prevEnd = -1;
+   int totalFree = 0;
+   int numObj = 0, numRes = 0;
+   int totalUsed = 0;
+
+   for ( block = tss->freeRanges[unit] ; block ; block = block->next ) {
+      assert( block->endAddr > 0 );
+      assert( block->startAddr <= tss->totalTexMem[unit] );
+      assert( block->endAddr <= tss->totalTexMem[unit] );
+      assert( (int) block->startAddr > prevStart );
+      assert( (int) block->startAddr >= prevEnd );
+      prevStart = (int) block->startAddr;
+      prevEnd = (int) block->endAddr;
+      totalFree += (block->endAddr - block->startAddr);
+   }
+   assert( totalFree == tss->freeTexMem[unit] );
+
+   for ( texObj = ss->TexObjectList ; texObj ; texObj = texObj->Next ) {
+      tdfxTexObjPtr t = TDFX_TEXTURE_DATA(texObj);
+      numObj++;
+      if ( t ) {
+	 if ( t->isInTM ) {
+	    numRes++;
+	    assert( t->range[0] );
+	    if ( t->range[unit] )
+	       totalUsed += (t->range[unit]->endAddr - t->range[unit]->startAddr);
+	 } else {
+	    assert(!t->range[0]);
+	 }
+      }
+   }
 
+   fprintf( stderr,
+	    "totalFree: %d  totalUsed: %d  totalMem: %d #objs=%d  #res=%d\n",
+	    tss->freeTexMem[unit], totalUsed, tss->totalTexMem[unit],
+	    numObj, numRes );
+
+   assert( totalUsed + totalFree == tss->totalTexMem[unit] );
+}
 
-static void
-dump_texmem(tdfxContextPtr fxMesa)
+static void tdfxTMDumpTexMem( tdfxContextPtr fxMesa )
 {
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    struct gl_texture_object *oldestObj, *obj, *lowestPriorityObj;
-    tdfxMemRange *r;
-    FxU32 prev;
-
-    printf("DUMP Objects:\n");
-    for (obj = mesaShared->TexObjectList; obj; obj = obj->Next) {
-        tdfxTexInfo *info = TDFX_TEXTURE_DATA(obj);
-
-        if (info && info->isInTM) {
-        printf("Obj %8p: %4d  info = %p\n", obj, obj->Name, info);
-
-           printf("  isInTM=%d  whichTMU=%d  lastTimeUsed=%d\n",
-                  info->isInTM, info->whichTMU, info->lastTimeUsed);
-           printf("    tm[0] = %p", info->tm[0]);
-           assert(info->tm[0]);
-           if (info->tm[0]) {
-              printf("  tm startAddr = %d  endAddr = %d",
-                     info->tm[0]->startAddr,
-                     info->tm[0]->endAddr);
-           }
-           printf("\n");
-           printf("    tm[1] = %p", info->tm[1]);
-           if (info->tm[1]) {
-              printf("  tm startAddr = %d  endAddr = %d",
-                     info->tm[1]->startAddr,
-                     info->tm[1]->endAddr);
-           }
-           printf("\n");
-        }
-    }
-
-    VerifyFreeList(fxMesa, 0);
-    VerifyFreeList(fxMesa, 1);
-
-    printf("Free memory unit 0:  %d bytes\n", shared->freeTexMem[0]);
-    prev = 0;
-    for (r = shared->tmFree[0]; r; r = r->next) {
-       printf("%8p:  start %8d  end %8d  size %8d  gap %8d\n", r, r->startAddr, r->endAddr, r->endAddr - r->startAddr, r->startAddr - prev);
-       prev = r->endAddr;
-    }
-
-    printf("Free memory unit 1:  %d bytes\n", shared->freeTexMem[1]);
-    prev = 0;
-    for (r = shared->tmFree[1]; r; r = r->next) {
-       printf("%8p:  start %8d  end %8d  size %8d  gap %8d\n", r, r->startAddr, r->endAddr, r->endAddr - r->startAddr, r->startAddr - prev);
-       prev = r->endAddr;
-    }
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+   struct gl_texture_object *texObj;
+   tdfxMemRange *r;
+   FxU32 prev;
+
+   printf( "DUMP Objects:\n" );
+   for ( texObj = ss->TexObjectList ; texObj ; texObj = texObj->Next ) {
+      tdfxTexObjPtr t = TDFX_TEXTURE_DATA(texObj);
+
+      if ( t && t->isInTM ) {
+	 printf( "Obj %8p: %4d  info = %p\n", texObj, texObj->Name, t );
+
+	 printf( "  isInTM=%d  whichTMU=%ld  lastTimeUsed=%d\n",
+		 t->isInTM, t->whichTMU, t->lastTimeUsed );
+	 printf( "    tm[0] = %p", t->range[0] );
+	 assert( t->range[0] );
+	 if ( t->range[0] ) {
+	    printf( "  tm startAddr = %ld  endAddr = %ld",
+		    t->range[0]->startAddr,
+		    t->range[0]->endAddr );
+	 }
+	 printf( "\n" );
+	 printf( "    tm[1] = %p", t->range[1] );
+	 if ( t->range[1] ) {
+	    printf( "  tm startAddr = %ld  endAddr = %ld",
+		    t->range[1]->startAddr,
+		    t->range[1]->endAddr );
+	 }
+	 printf( "\n" );
+      }
+   }
 
-}
-#endif
+   tdfxTMVerifyFreeList( fxMesa, 0 );
+   tdfxTMVerifyFreeList( fxMesa, 1 );
+
+   printf( "Free memory unit 0:  %d bytes\n", tss->freeTexMem[0] );
+   prev = 0;
+   for ( r = tss->freeRanges[0] ; r ; r = r->next ) {
+      printf( "%8p:  start %8ld  end %8ld  size %8ld  gap %8ld\n",
+	      r, r->startAddr, r->endAddr, r->endAddr - r->startAddr,
+	      r->startAddr - prev );
+      prev = r->endAddr;
+   }
 
+   printf( "Free memory unit 1:  %d bytes\n", tss->freeTexMem[1] );
+   prev = 0;
+   for ( r = tss->freeRanges[1] ; r ; r = r->next ) {
+      printf( "%8p:  start %8ld  end %8ld  size %8ld  gap %8ld\n",
+	      r, r->startAddr, r->endAddr, r->endAddr - r->startAddr,
+	      r->startAddr - prev );
+      prev = r->endAddr;
+   }
+}
 
 
 #ifdef TEXSANITY
-static void
-fubar(void)
+static void fubar( void )
 {
+   /* GH: What am I meant to do??? */
 }
 
-/*
- * Sanity Check
+/* Sanity Check
  */
-static void
-sanity(tdfxContextPtr fxMesa)
+static void sanity( tdfxContextPtr fxMesa )
 {
-    tdfxMemRange *tmp, *prev, *pos;
-
-    prev = 0;
-    tmp = fxMesa->tmFree[0];
-    while (tmp) {
-        if (!tmp->startAddr && !tmp->endAddr) {
-            fprintf(stderr, "Textures fubar\n");
-            fubar();
-        }
-        if (tmp->startAddr >= tmp->endAddr) {
-            fprintf(stderr, "Node fubar\n");
-            fubar();
-        }
-        if (prev && (prev->startAddr >= tmp->startAddr ||
-                     prev->endAddr > tmp->startAddr)) {
-            fprintf(stderr, "Sorting fubar\n");
-            fubar();
-        }
-        prev = tmp;
-        tmp = tmp->next;
-    }
-    prev = 0;
-    tmp = fxMesa->tmFree[1];
-    while (tmp) {
-        if (!tmp->startAddr && !tmp->endAddr) {
-            fprintf(stderr, "Textures fubar\n");
-            fubar();
-        }
-        if (tmp->startAddr >= tmp->endAddr) {
-            fprintf(stderr, "Node fubar\n");
-            fubar();
-        }
-        if (prev && (prev->startAddr >= tmp->startAddr ||
-                     prev->endAddr > tmp->startAddr)) {
-            fprintf(stderr, "Sorting fubar\n");
-            fubar();
-        }
-        prev = tmp;
-        tmp = tmp->next;
-    }
+   tdfxMemRange *tmp, *prev, *pos;
+
+   prev = 0;
+   tmp = fxMesa->freeRanges[0];
+   while ( tmp ) {
+      if ( !tmp->startAddr && !tmp->endAddr ) {
+	 fprintf( stderr, "Textures fubar\n" );
+	 fubar();
+      }
+      if ( tmp->startAddr >= tmp->endAddr ) {
+	 fprintf( stderr, "Node fubar\n" );
+	 fubar();
+      }
+      if ( prev && ( prev->startAddr >= tmp->startAddr ||
+		     prev->endAddr > tmp->startAddr ) ) {
+	 fprintf( stderr, "Sorting fubar\n" );
+	 fubar();
+      }
+      prev = tmp;
+      tmp = tmp->next;
+   }
+
+   prev = 0;
+   tmp = fxMesa->freeRanges[1];
+   while ( tmp ) {
+      if ( !tmp->startAddr && !tmp->endAddr ) {
+	 fprintf( stderr, "Textures fubar\n" );
+	 fubar();
+      }
+      if ( tmp->startAddr >= tmp->endAddr ) {
+	 fprintf( stderr, "Node fubar\n" );
+	 fubar();
+      }
+      if ( prev && ( prev->startAddr >= tmp->startAddr ||
+		     prev->endAddr > tmp->startAddr ) ) {
+	 fprintf( stderr, "Sorting fubar\n" );
+	 fubar();
+      }
+      prev = tmp;
+      tmp = tmp->next;
+   }
 }
 #endif
 
 
-
-
-
-/*
- * Allocate and initialize a new MemRange struct.
- * Try to allocate it from the pool of free MemRange nodes rather than malloc.
+/* Allocate and initialize a new MemRange struct.  Try to allocate it
+ * from the pool of free MemRange nodes rather than malloc.
  */
 static tdfxMemRange *
-NewRangeNode(tdfxContextPtr fxMesa, FxU32 start, FxU32 end)
+tdfxTMNewRangeNode( tdfxContextPtr fxMesa, FxU32 start, FxU32 end )
 {
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    tdfxMemRange *result;
-
-    _glthread_LOCK_MUTEX(mesaShared->Mutex);
-    if (shared && shared->tmPool) {
-        result = shared->tmPool;
-        shared->tmPool = shared->tmPool->next;
-    }
-    else {
-        result = MALLOC(sizeof(tdfxMemRange));
-
-    }
-    _glthread_UNLOCK_MUTEX(mesaShared->Mutex);
-
-    if (!result) {
-        /*fprintf(stderr, "fxDriver: out of memory!\n");*/
-        return NULL;
-    }
-
-    result->startAddr = start;
-    result->endAddr = end;
-    result->next = NULL;
-
-    return result;
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+   tdfxMemRange *range;
+
+   _glthread_LOCK_MUTEX( ss->Mutex );
+   if ( tss && tss->rangePool ) {
+      range = tss->rangePool;
+      tss->rangePool = tss->rangePool->next;
+   } else {
+      range = MALLOC( sizeof(tdfxMemRange) );
+   }
+   _glthread_UNLOCK_MUTEX( ss->Mutex );
+
+   if ( !range ) {
+      if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	 fprintf( stderr, __FUNCTION__ ": out of memory!\n" );
+      return NULL;
+   }
+
+   range->startAddr = start;
+   range->endAddr = end;
+   range->next = NULL;
+
+   return range;
 }
 
 
-/*
- * Initialize texture memory.
- * We take care of one or both TMU's here.
+/* Initialize texture memory.  We take care of one or both TMU's here.
  */
-void
-tdfxTMInit(tdfxContextPtr fxMesa)
+void tdfxTMInit( tdfxContextPtr fxMesa )
 {
-    if (!fxMesa->glCtx->Shared->DriverData) {
-        const char *extensions;
-        struct tdfxSharedState *shared = CALLOC_STRUCT(tdfxSharedState);
-        if (!shared)
-           return;
-
-        extensions = FX_grGetString(fxMesa, GR_EXTENSION);
-        if (strstr(extensions, "TEXUMA")) {
-            FxU32 start, end;
-            shared->umaTexMemory = GL_TRUE;
-            FX_grEnable(fxMesa, GR_TEXTURE_UMA_EXT);
-            start = FX_grTexMinAddress(fxMesa, 0);
-            end = FX_grTexMaxAddress(fxMesa, 0);
-            shared->totalTexMem[0] = end - start;
-            shared->totalTexMem[1] = 0;
-            shared->freeTexMem[0] = end - start;
-            shared->freeTexMem[1] = 0;
-            shared->tmFree[0] = NewRangeNode(fxMesa, start, end);
-            shared->tmFree[1] = NULL;
-            /*printf("UMA tex memory: %d\n", (int) (end - start));*/
-        }
-        else {
-            const int numTMUs = fxMesa->haveTwoTMUs ? 2 : 1;
-            int tmu;
-            shared->umaTexMemory = GL_FALSE;
-            for (tmu = 0; tmu < numTMUs; tmu++) {
-                FxU32 start = FX_grTexMinAddress(fxMesa, tmu);
-                FxU32 end = FX_grTexMaxAddress(fxMesa, tmu);
-                shared->totalTexMem[tmu] = end - start;
-                shared->freeTexMem[tmu] = end - start;
-                shared->tmFree[tmu] = NewRangeNode(fxMesa, start, end);
-                /*printf("Split tex memory: %d\n", (int) (end - start));*/
-            }
-        }
-
-        shared->tmPool = NULL;
-        fxMesa->glCtx->Shared->DriverData = shared;
-        /*printf("Texture memory init UMA: %d\n", shared->umaTexMemory);*/
-    }
+   GLcontext *ctx = fxMesa->glCtx;
+
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+      fprintf( stderr, __FUNCTION__ "\n" );
+
+   if ( !ctx->Shared->DriverData ) {
+      const char *extensions;
+      tdfxSharedStatePtr tss = CALLOC_STRUCT( tdfx_shared_state );
+
+      if ( !tss )
+	 return;
+
+      LOCK_HARDWARE( fxMesa );
+
+      extensions = grGetString( GR_EXTENSION );
+
+      if ( strstr( extensions, " TEXUMA " ) ) {
+	 FxU32 start, end;
+
+	 tss->umaTexMemory = GL_TRUE;
+
+	 grEnable( GR_TEXTURE_UMA_EXT );
+
+	 start = grTexMinAddress( 0 );
+	 end = grTexMaxAddress( 0 );
+
+	 if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	    fprintf( stderr, "   UMA tex memory: %d\n", (int)(end - start) );
+
+	 tss->totalTexMem[0] = end - start;
+	 tss->totalTexMem[1] = 0;
+	 tss->freeTexMem[0] = end - start;
+	 tss->freeTexMem[1] = 0;
+	 tss->freeRanges[0] = tdfxTMNewRangeNode( fxMesa, start, end );
+	 tss->freeRanges[1] = NULL;
+      } else {
+	 int unit;
+
+	 tss->umaTexMemory = GL_FALSE;
+
+	 for ( unit = 0 ; unit < fxMesa->numTMUs ; unit++ ) {
+	    FxU32 start, end;
+
+	    start = grTexMinAddress( unit );
+	    end = grTexMaxAddress( unit );
+
+	    tss->totalTexMem[unit] = end - start;
+	    tss->freeTexMem[unit] = end - start;
+	    tss->freeRanges[unit] = tdfxTMNewRangeNode( fxMesa, start, end );
+
+	    if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	       fprintf( stderr, "   Split tex memory: %d\n",
+			(int)(end - start) );
+	 }
+      }
+
+      UNLOCK_HARDWARE( fxMesa );
+
+      tss->rangePool = NULL;
+      ctx->Shared->DriverData = tss;
+
+      if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	 fprintf( stderr, "  init UMA: %d\n", tss->umaTexMemory );
+   }
 }
 
 
-/*
- * Clean-up texture memory before destroying context.
+/* Clean-up texture memory before destroying context.
  */
-void
-tdfxTMClose(tdfxContextPtr fxMesa)
+void tdfxTMClose( tdfxContextPtr fxMesa )
 {
-    if (fxMesa->glCtx->Shared->RefCount == 1) {
-        /* refcount will soon go to zero, free our 3dfx stuff */
-        struct tdfxSharedState *shared = (struct tdfxSharedState *) fxMesa->glCtx->Shared->DriverData;
-
-        const int numTMUs = fxMesa->haveTwoTMUs ? 2 : 1;
-        int tmu;
-        tdfxMemRange *tmp, *next;
-
-        /* Deallocate the pool of free tdfxMemRange nodes */
-        tmp = shared->tmPool;
-        while (tmp) {
-            next = tmp->next;
-            FREE(tmp);
-            tmp = next;
-        }
-
-        /* Delete the texture memory block tdfxMemRange nodes */
-        for (tmu = 0; tmu < numTMUs; tmu++) {
-            tmp = shared->tmFree[tmu];
-            while (tmp) {
-                next = tmp->next;
-                FREE(tmp);
-                tmp = next;
-            }
-        }
-
-        FREE(shared);
-        fxMesa->glCtx->Shared->DriverData = NULL;
-    }
-}
+   GLcontext *ctx = fxMesa->glCtx;
 
+   if ( ctx->Shared->RefCount == 1 ) {
+      /* RefCount will soon go to zero, free our 3dfx stuff */
+      tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ctx->Shared->DriverData;
+      int unit;
+      tdfxMemRange *tmp, *next;
+
+      /* Deallocate the pool of free tdfxMemRange nodes */
+      tmp = tss->rangePool;
+      while ( tmp ) {
+	 next = tmp->next;
+	 FREE( tmp );
+	 tmp = next;
+      }
 
+      /* Delete the texture memory block tdfxMemRange nodes */
+      for ( unit = 0 ; unit < fxMesa->numTMUs ; unit++ ) {
+	 tmp = tss->freeRanges[unit];
+	 while ( tmp ) {
+	    next = tmp->next;
+	    FREE( tmp );
+	    tmp = next;
+	 }
+      }
 
-/*
- * Delete a tdfxMemRange struct.
- * We keep a linked list of free/available tdfxMemRange structs to
- * avoid extra malloc/free calls.
- */
-#if 0
-static void
-DeleteRangeNode_NoLock(struct TdfxSharedState *shared, tdfxMemRange *range)
-{
-    /* insert at head of list */
-    range->next = shared->tmPool;
-    shared->tmPool = range;
+      FREE( tss );
+      ctx->Shared->DriverData = NULL;
+   }
 }
-#endif
 
-#define DELETE_RANGE_NODE(shared, range) \
-    (range)->next = (shared)->tmPool;    \
-    (shared)->tmPool = (range)
 
 
+/* Delete a tdfxMemRange struct.
+ * We keep a linked list of free/available tdfxMemRange structs to
+ * avoid extra malloc/free calls.
+ */
+#define DELETE_RANGE_NODE( tss, range )					\
+do {									\
+   (range)->next = (tss)->rangePool;					\
+   (tss)->rangePool = (range);						\
+} while (0)
 
-/*
- * When we've run out of texture memory we have to throw out an
+/* When we've run out of texture memory we have to throw out an
  * existing texture to make room for the new one.  This function
  * determins the texture to throw out.
  */
 static struct gl_texture_object *
-FindOldestObject(tdfxContextPtr fxMesa, FxU32 tmu)
+tdfxTMFindOldestObject( tdfxContextPtr fxMesa, FxU32 unit )
 {
-    const GLuint bindnumber = fxMesa->texBindNumber;
-    struct gl_texture_object *oldestObj, *obj, *lowestPriorityObj;
-    GLfloat lowestPriority;
-    GLuint oldestAge;
-
-    oldestObj = NULL;
-    oldestAge = 0;
-
-    lowestPriority = 1.0F;
-    lowestPriorityObj = NULL;
-
-    for (obj = fxMesa->glCtx->Shared->TexObjectList; obj; obj = obj->Next) {
-        tdfxTexInfo *info = TDFX_TEXTURE_DATA(obj);
-
-        if (info && info->isInTM &&
-            ((info->whichTMU == tmu) || (info->whichTMU == TDFX_TMU_BOTH) ||
-             (info->whichTMU == TDFX_TMU_SPLIT))) {
-            GLuint age, lasttime;
-
-            assert(info->tm[0]);
-            lasttime = info->lastTimeUsed;
-
-            if (lasttime > bindnumber)
-                age = bindnumber + (UINT_MAX - lasttime + 1); /* TO DO: check wrap around */
-            else
-                age = bindnumber - lasttime;
-
-            if (age >= oldestAge) {
-                oldestAge = age;
-                oldestObj = obj;
-            }
-
-            /* examine priority */
-            if (obj->Priority < lowestPriority) {
-                lowestPriority = obj->Priority;
-                lowestPriorityObj = obj;
-            }
-        }
-    }
-
-    if (lowestPriority < 1.0) {
-        ASSERT(lowestPriorityObj);
-        /*
-        printf("discard %d pri=%f\n", lowestPriorityObj->Name, lowestPriority);
-        */
-        return lowestPriorityObj;
-    }
-    else {
-        /*
-        printf("discard %d age=%d\n", oldestObj->Name, oldestAge);
-        */
-        return oldestObj;
-    }
-}
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   const GLuint bindNumber = fxMesa->texBindNumber;
+   struct gl_texture_object *oldestObj, *texObj, *lowestPriorityObj;
+   GLfloat lowestPriority;
+   GLuint oldestAge;
+
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+      fprintf( stderr, __FUNCTION__ "\n" );
+
+   oldestObj = NULL;
+   oldestAge = 0;
+
+   lowestPriority = 1.0F;
+   lowestPriorityObj = NULL;
+
+   for ( texObj = ss->TexObjectList ; texObj ; texObj = texObj->Next ) {
+      tdfxTexObjPtr t = TDFX_TEXTURE_DATA(texObj);
+
+      if ( t && t->isInTM &&
+	   ( ( t->whichTMU == unit ) ||
+	     ( t->whichTMU == TDFX_TMU_BOTH ) ||
+	     ( t->whichTMU == TDFX_TMU_SPLIT ) ) ) {
+	 GLuint age, lastTime;
+
+	 assert( t->range[0] );
+	 lastTime = t->lastTimeUsed;
+
+	 if ( lastTime > bindNumber ) {
+	    /* TODO: check wrap around */
+	    age = bindNumber + (UINT_MAX - lastTime + 1);
+	 } else {
+	    age = bindNumber - lastTime;
+	 }
+	 if ( age >= oldestAge ) {
+	    oldestAge = age;
+	    oldestObj = texObj;
+	 }
 
+	 /* examine priority */
+	 if ( texObj->Priority < lowestPriority ) {
+	    lowestPriority = texObj->Priority;
+	    lowestPriorityObj = texObj;
+	 }
+      }
+   }
 
-#if 0
-static void
-FlushTexMemory(tdfxContextPtr fxMesa)
-{
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    struct gl_texture_object *obj;
-
-    for (obj = mesaShared->TexObjectList; obj; obj = obj->Next) {
-       if (obj->RefCount < 2) {
-          /* don't flush currently bound textures */
-          tdfxTMMoveOutTM_NoLock(fxMesa, obj);
-       }
-    }
+   if ( lowestPriority < 1.0 ) {
+      ASSERT( lowestPriorityObj );
+      if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	 fprintf( stderr, "discard %d pri=%f\n",
+		  lowestPriorityObj->Name, lowestPriority );
+      return lowestPriorityObj;
+   } else {
+      if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	 fprintf( stderr, "discard %d age=%d\n",
+		  oldestObj->Name, oldestAge );
+      return oldestObj;
+   }
 }
-#endif
 
 
-/*
- * Find the address (offset?) at which we can store a new texture.
- * <tmu> is the texture unit.
+/* Find the address (offset?) at which we can store a new texture.
+ * <unit> is the texture unit.
  * <size> is the texture size in bytes.
  */
-static FxU32
-FindStartAddr(tdfxContextPtr fxMesa, FxU32 tmu, FxU32 size)
+static FxU32 tdfxTMFindStartAddr( tdfxContextPtr fxMesa,
+				  FxU32 unit, FxU32 size )
 {
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    tdfxMemRange *prev, *block;
-    FxU32 result;
-#if 0
-    int discardedCount = 0;
-#define MAX_DISCARDS 10
-#endif
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+   struct gl_texture_object *texObj;
+   tdfxMemRange *prev, *block;
+   FxU32 result;
+
+   if ( tss->umaTexMemory ) {
+      assert( unit == TDFX_TMU0 );
+   }
 
-    if (shared->umaTexMemory) {
-        assert(tmu == TDFX_TMU0);
-    }
-
-    _glthread_LOCK_MUTEX(mesaShared->Mutex);
-    while (1) {
-        prev = NULL;
-        block = shared->tmFree[tmu];
-        while (block) {
-            if (block->endAddr - block->startAddr >= size) {
-                /* The texture will fit here */
-                result = block->startAddr;
-                block->startAddr += size;
-                if (block->startAddr == block->endAddr) {
-                    /* Remove this node since it's empty */
-                    if (prev) {
-                        prev->next = block->next;
-                    }
-                    else {
-                        shared->tmFree[tmu] = block->next;
-                    }
-                    DELETE_RANGE_NODE(shared, block);
-                }
-                shared->freeTexMem[tmu] -= size;
-                _glthread_UNLOCK_MUTEX(mesaShared->Mutex);
-                return result;
-            }
-            prev = block;
-            block = block->next;
-        }
-        /* We failed to find a block large enough to accomodate <size> bytes.
-         * Find the oldest texObject and free it.
-         */
-#if 0
-        discardedCount++;
-        if (discardedCount > MAX_DISCARDS + 1) {
-            gl_problem(NULL, "tdfx driver: extreme texmem fragmentation");
-            _glthread_UNLOCK_MUTEX(mesaShared->Mutex);
-            return BAD_ADDRESS;
-        }
-        else if (discardedCount > MAX_DISCARDS) {
-            /* texture memory is probably really fragmented, flush it */
-            FlushTexMemory(fxMesa);
-        }
-        else
-#endif
-        {
-            struct gl_texture_object *obj = FindOldestObject(fxMesa, tmu);
-            if (obj) {
-                tdfxTMMoveOutTM_NoLock(fxMesa, obj);
-                fxMesa->stats.texSwaps++;
-            }
-            else {
-                gl_problem(NULL, "tdfx driver: extreme texmem fragmentation");
-                _glthread_UNLOCK_MUTEX(mesaShared->Mutex);
-                return BAD_ADDRESS;
-            }
-        }
-    }
-
-    /* never get here, but play it safe */
-    _glthread_UNLOCK_MUTEX(mesaShared->Mutex);
-    return BAD_ADDRESS;
+   _glthread_LOCK_MUTEX( ss->Mutex );
+   while ( 1 ) {
+      prev = NULL;
+      block = tss->freeRanges[unit];
+
+      while ( block ) {
+	 if ( block->endAddr - block->startAddr >= size ) {
+	    /* The texture will fit here */
+	    result = block->startAddr;
+	    block->startAddr += size;
+	    if ( block->startAddr == block->endAddr ) {
+	       /* Remove this node since it's empty */
+	       if ( prev ) {
+		  prev->next = block->next;
+	       } else {
+		  tss->freeRanges[unit] = block->next;
+	       }
+	       DELETE_RANGE_NODE( tss, block );
+	    }
+	    tss->freeTexMem[unit] -= size;
+	    _glthread_UNLOCK_MUTEX( ss->Mutex );
+	    return result;
+	 }
+	 prev = block;
+	 block = block->next;
+      }
+
+      /* We failed to find a block large enough to accomodate <size> bytes.
+       * Find the oldest texObject and free it.
+       */
+      texObj = tdfxTMFindOldestObject( fxMesa, unit );
+      if ( texObj ) {
+	 tdfxTMMoveOutTMLocked( fxMesa, texObj );
+	 fxMesa->stats.texSwaps++;
+      } else {
+	 gl_problem( NULL, "tdfx driver: extreme texmem fragmentation" );
+	 _glthread_UNLOCK_MUTEX( ss->Mutex );
+	 return BAD_ADDRESS;
+      }
+   }
+
+   /* never get here, but play it safe */
+   _glthread_UNLOCK_MUTEX( ss->Mutex );
+   return BAD_ADDRESS;
 }
 
 
-/*
- * Remove the given tdfxMemRange node from hardware texture memory.
+/* Remove the given tdfxMemRange node from hardware texture memory.
  */
-static void
-RemoveRange_NoLock(tdfxContextPtr fxMesa, FxU32 tmu, tdfxMemRange *range)
+static void tdfxTMRemoveRangeLocked( tdfxContextPtr fxMesa,
+				     FxU32 unit, tdfxMemRange *range )
 {
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    tdfxMemRange *block, *prev;
-
-    if (shared->umaTexMemory) {
-       assert(tmu == TDFX_TMU0);
-    }
-
-    if (!range)
-        return;
-
-    if (range->startAddr == range->endAddr) {
-        DELETE_RANGE_NODE(shared, range);
-        return;
-    }
-    shared->freeTexMem[tmu] += range->endAddr - range->startAddr;
-
-    /* find position in linked list to insert this tdfxMemRange node */
-    prev = NULL;
-    block = shared->tmFree[tmu];
-    while (block) {
-        assert(range->startAddr != block->startAddr);
-        if (range->startAddr > block->startAddr) {
-            prev = block;
-            block = block->next;
-        }
-        else {
-            break;
-        }
-    }
-
-    /* Insert the free block, combine with adjacent blocks when possible */
-    range->next = block;
-    if (block) {
-        if (range->endAddr == block->startAddr) {
-            /* Combine */
-            block->startAddr = range->startAddr;
-            DELETE_RANGE_NODE(shared, range);
-            range = block;
-        }
-    }
-    if (prev) {
-        if (prev->endAddr == range->startAddr) {
-            /* Combine */
-            prev->endAddr = range->endAddr;
-            prev->next = range->next;
-            DELETE_RANGE_NODE(shared, range);
-        }
-        else {
-            prev->next = range;
-        }
-    }
-    else {
-        shared->tmFree[tmu] = range;
-    }
-}
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+   tdfxMemRange *block, *prev;
 
+   if ( tss->umaTexMemory ) {
+      assert( unit == TDFX_TMU0 );
+   }
 
-#if 0 /* NOT USED */
-static void
-RemoveRange(tdfxContextPtr fxMesa, FxU32 tmu, tdfxMemRange *range)
-{
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    _glthread_LOCK_MUTEX(mesaShared->Mutex);
-    RemoveRange_NoLock(fxMesa, tmu, range);
-    _glthread_UNLOCK_MUTEX(mesaShared->Mutex);
+   if ( !range )
+      return;
+
+   if ( range->startAddr == range->endAddr ) {
+      DELETE_RANGE_NODE( tss, range );
+      return;
+   }
+   tss->freeTexMem[unit] += range->endAddr - range->startAddr;
+
+   /* find position in linked list to insert this tdfxMemRange node */
+   prev = NULL;
+   block = tss->freeRanges[unit];
+   while ( block ) {
+      assert( range->startAddr != block->startAddr );
+      if ( range->startAddr > block->startAddr ) {
+	 prev = block;
+	 block = block->next;
+      } else {
+	 break;
+      }
+   }
+
+   /* Insert the free block, combine with adjacent blocks when possible */
+   range->next = block;
+   if ( block ) {
+      if ( range->endAddr == block->startAddr ) {
+	 /* Combine */
+	 block->startAddr = range->startAddr;
+	 DELETE_RANGE_NODE( tss, range );
+	 range = block;
+      }
+   }
+   if ( prev ) {
+      if ( prev->endAddr == range->startAddr ) {
+	 /* Combine */
+	 prev->endAddr = range->endAddr;
+	 prev->next = range->next;
+	 DELETE_RANGE_NODE( tss, range );
+      } else {
+	 prev->next = range;
+      }
+   } else {
+      tss->freeRanges[unit] = range;
+   }
 }
-#endif
 
 
-/*
- * Allocate space for a texture image.
+/* Allocate space for a texture image.
  * <tmu> is the texture unit
  * <texmemsize> is the number of bytes to allocate
  */
 static tdfxMemRange *
-AllocTexMem(tdfxContextPtr fxMesa, FxU32 tmu, FxU32 texmemsize)
+tdfxTMAllocTexMem( tdfxContextPtr fxMesa, FxU32 unit, FxU32 size )
 {
-    FxU32 startAddr;
-    startAddr = FindStartAddr(fxMesa, tmu, texmemsize);
-    if (startAddr == BAD_ADDRESS) {
-       printf("AllocTexMem returned NULL!  tmu=%d texmemsize=%d\n",
-              tmu, texmemsize);
-        return NULL;
-    }
-    else {
-        tdfxMemRange *range;
-        range = NewRangeNode(fxMesa, startAddr, startAddr + texmemsize);
-        return range;
-    }
+   tdfxMemRange *range = NULL;
+   FxU32 start;
+
+   start = tdfxTMFindStartAddr( fxMesa, unit, size );
+
+   if ( start != BAD_ADDRESS ) {
+      range = tdfxTMNewRangeNode( fxMesa, start, start + size );
+   } else {
+      fprintf( stderr,
+	       "tdfxTMAllocTexMem returned NULL!  unit=%ld size=%ld\n",
+	       unit, size );
+   }
+   return range;
 }
 
 
-/*
- * Download (copy) the given texture data (all mipmap levels) into the
- * Voodoo's texture memory.
- * The texture memory must have already been allocated.
+/* Download (copy) the given texture data (all mipmap levels) into the
+ * Voodoo's texture memory.  The texture memory must have already been
+ * allocated.
  */
-void
-tdfxTMDownloadTexture(tdfxContextPtr fxMesa, struct gl_texture_object *tObj)
+void tdfxTMDownloadTextureLocked( tdfxContextPtr fxMesa,
+				  struct gl_texture_object *tObj )
 {
-    tdfxTexInfo *ti;
-    GLint l;
-    FxU32 targetTMU;
-
-    assert(tObj);
-    ti = TDFX_TEXTURE_DATA(tObj);
-    assert(ti);
-    targetTMU = ti->whichTMU;
-
-    switch (targetTMU) {
-    case TDFX_TMU0:
-    case TDFX_TMU1:
-        if (ti->tm[targetTMU]) {
-            for (l = ti->minLevel; l <= ti->maxLevel
-                    && ti->mipmapLevel[l].data; l++) {
-                GrLOD_t glideLod = ti->info.largeLodLog2 - l + tObj->BaseLevel;
-                FX_grTexDownloadMipMapLevel_NoLock(targetTMU,
-                                                  ti->tm[targetTMU]->startAddr,
-                                                  glideLod,
-                                                  ti->info.largeLodLog2,
-                                                  ti->info.aspectRatioLog2,
-                                                  ti->info.format,
-                                                  GR_MIPMAPLEVELMASK_BOTH,
-                                                  ti->mipmapLevel[l].data);
-            }
-        }
-        break;
-    case TDFX_TMU_SPLIT:
-        if (ti->tm[TDFX_TMU0] && ti->tm[TDFX_TMU1]) {
-            for (l = ti->minLevel; l <= ti->maxLevel
-                    && ti->mipmapLevel[l].data; l++) {
-                GrLOD_t glideLod = ti->info.largeLodLog2 - l + tObj->BaseLevel;
-                FX_grTexDownloadMipMapLevel_NoLock(GR_TMU0,
-                                                  ti->tm[TDFX_TMU0]->startAddr,
-                                                  glideLod,
-                                                  ti->info.largeLodLog2,
-                                                  ti->info.aspectRatioLog2,
-                                                  ti->info.format,
-                                                  GR_MIPMAPLEVELMASK_ODD,
-                                                  ti->mipmapLevel[l].data);
-
-                FX_grTexDownloadMipMapLevel_NoLock(GR_TMU1,
-                                                  ti->tm[TDFX_TMU1]->startAddr,
-                                                  glideLod,
-                                                  ti->info.largeLodLog2,
-                                                  ti->info.aspectRatioLog2,
-                                                  ti->info.format,
-                                                  GR_MIPMAPLEVELMASK_EVEN,
-                                                  ti->mipmapLevel[l].data);
-            }
-        }
-        break;
-    case TDFX_TMU_BOTH:
-        if (ti->tm[TDFX_TMU0] && ti->tm[TDFX_TMU1]) {
-            for (l = ti->minLevel; l <= ti->maxLevel
-                    && ti->mipmapLevel[l].data; l++) {
-                GrLOD_t glideLod = ti->info.largeLodLog2 - l + tObj->BaseLevel;
-                FX_grTexDownloadMipMapLevel_NoLock(GR_TMU0,
-                                                  ti->tm[TDFX_TMU0]->startAddr,
-                                                  glideLod,
-                                                  ti->info.largeLodLog2,
-                                                  ti->info.aspectRatioLog2,
-                                                  ti->info.format,
-                                                  GR_MIPMAPLEVELMASK_BOTH,
-                                                  ti->mipmapLevel[l].data);
-
-                FX_grTexDownloadMipMapLevel_NoLock(GR_TMU1,
-                                                  ti->tm[TDFX_TMU1]->startAddr,
-                                                  glideLod,
-                                                  ti->info.largeLodLog2,
-                                                  ti->info.aspectRatioLog2,
-                                                  ti->info.format,
-                                                  GR_MIPMAPLEVELMASK_BOTH,
-                                                  ti->mipmapLevel[l].data);
-            }
-        }
-        break;
-    default:
-        gl_problem(NULL, "error in tdfxTMDownloadTexture: bad tmu");
-        return;
-    }
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+   FxU32 targetTMU;
+   GLint l;
+
+   assert( tObj );
+   assert( t );
+
+   targetTMU = t->whichTMU;
+
+   switch ( targetTMU ) {
+   case TDFX_TMU0:
+   case TDFX_TMU1:
+      if ( t->range[targetTMU] ) {
+	 for ( l = t->minLevel ; l <= t->maxLevel && t->image[l].data ; l++ ) {
+	    GrLOD_t glideLod = t->info.largeLodLog2 - l + tObj->BaseLevel;
+
+	    grTexDownloadMipMapLevel( targetTMU,
+				      t->range[targetTMU]->startAddr,
+				      glideLod,
+				      t->info.largeLodLog2,
+				      t->info.aspectRatioLog2,
+				      t->info.format,
+				      GR_MIPMAPLEVELMASK_BOTH,
+				      t->image[l].data );
+	 }
+      }
+      break;
+
+   case TDFX_TMU_SPLIT:
+      if ( t->range[TDFX_TMU0] && t->range[TDFX_TMU1] ) {
+	 for ( l = t->minLevel ; l <= t->maxLevel && t->image[l].data ; l++ ) {
+	    GrLOD_t glideLod = t->info.largeLodLog2 - l + tObj->BaseLevel;
+
+	    grTexDownloadMipMapLevel( GR_TMU0,
+				      t->range[TDFX_TMU0]->startAddr,
+				      glideLod,
+				      t->info.largeLodLog2,
+				      t->info.aspectRatioLog2,
+				      t->info.format,
+				      GR_MIPMAPLEVELMASK_ODD,
+				      t->image[l].data );
+
+	    grTexDownloadMipMapLevel( GR_TMU1,
+				      t->range[TDFX_TMU1]->startAddr,
+				      glideLod,
+				      t->info.largeLodLog2,
+				      t->info.aspectRatioLog2,
+				      t->info.format,
+				      GR_MIPMAPLEVELMASK_EVEN,
+				      t->image[l].data );
+	 }
+      }
+      break;
+
+   case TDFX_TMU_BOTH:
+      if ( t->range[TDFX_TMU0] && t->range[TDFX_TMU1] ) {
+	 for ( l = t->minLevel ; l <= t->maxLevel && t->image[l].data ; l++ ) {
+	    GrLOD_t glideLod = t->info.largeLodLog2 - l + tObj->BaseLevel;
+
+	    grTexDownloadMipMapLevel( GR_TMU0,
+				      t->range[TDFX_TMU0]->startAddr,
+				      glideLod,
+				      t->info.largeLodLog2,
+				      t->info.aspectRatioLog2,
+				      t->info.format,
+				      GR_MIPMAPLEVELMASK_BOTH,
+				      t->image[l].data );
+
+	    grTexDownloadMipMapLevel( GR_TMU1,
+				      t->range[TDFX_TMU1]->startAddr,
+				      glideLod,
+				      t->info.largeLodLog2,
+				      t->info.aspectRatioLog2,
+				      t->info.format,
+				      GR_MIPMAPLEVELMASK_BOTH,
+				      t->image[l].data );
+	 }
+      }
+      break;
+
+   default:
+      gl_problem( NULL, "error in tdfxTMDownloadTexture: bad unit" );
+      return;
+   }
 }
 
 
-void
-tdfxTMReloadMipMapLevel(GLcontext *ctx, struct gl_texture_object *tObj,
-                        GLint level)
+void tdfxTMReloadMipMapLevelLocked( GLcontext *ctx,
+				    struct gl_texture_object *tObj,
+				    GLint level )
 {
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-    GrLOD_t glideLod;
-    FxU32 tmu;
-
-    tmu = ti->whichTMU;
-    glideLod =  ti->info.largeLodLog2 - level + tObj->BaseLevel;
-    ASSERT(ti->isInTM);
-
-    switch (tmu) {
-    case TDFX_TMU0:
-    case TDFX_TMU1:
-        FX_grTexDownloadMipMapLevel(fxMesa, tmu,
-                                    ti->tm[tmu]->startAddr,
-                                    glideLod,
-                                    ti->info.largeLodLog2,
-                                    ti->info.aspectRatioLog2,
-                                    ti->info.format,
-                                    GR_MIPMAPLEVELMASK_BOTH,
-                                    ti->mipmapLevel[level].data);
-        break;
-    case TDFX_TMU_SPLIT:
-        FX_grTexDownloadMipMapLevel(fxMesa, GR_TMU0,
-                                    ti->tm[GR_TMU0]->startAddr,
-                                    glideLod,
-                                    ti->info.largeLodLog2,
-                                    ti->info.aspectRatioLog2,
-                                    ti->info.format,
-                                    GR_MIPMAPLEVELMASK_ODD,
-                                    ti->mipmapLevel[level].data);
-
-        FX_grTexDownloadMipMapLevel(fxMesa, GR_TMU1,
-                                    ti->tm[GR_TMU1]->startAddr,
-                                    glideLod,
-                                    ti->info.largeLodLog2,
-                                    ti->info.aspectRatioLog2,
-                                    ti->info.format,
-                                    GR_MIPMAPLEVELMASK_EVEN,
-                                    ti->mipmapLevel[level].data);
-        break;
-    case TDFX_TMU_BOTH:
-        FX_grTexDownloadMipMapLevel(fxMesa, GR_TMU0,
-                                    ti->tm[GR_TMU0]->startAddr,
-                                    glideLod,
-                                    ti->info.largeLodLog2,
-                                    ti->info.aspectRatioLog2,
-                                    ti->info.format,
-                                    GR_MIPMAPLEVELMASK_BOTH,
-                                    ti->mipmapLevel[level].data);
-
-        FX_grTexDownloadMipMapLevel(fxMesa, GR_TMU1,
-                                    ti->tm[GR_TMU1]->startAddr,
-                                    glideLod,
-                                    ti->info.largeLodLog2,
-                                    ti->info.aspectRatioLog2,
-                                    ti->info.format,
-                                    GR_MIPMAPLEVELMASK_BOTH,
-                                    ti->mipmapLevel[level].data);
-        break;
-
-    default:
-        gl_problem(ctx, "error in tdfxTMReloadMipMapLevel(): wrong tmu");
-        break;
-    }
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+   GrLOD_t glideLod;
+   FxU32 unit;
+
+   ASSERT( t->isInTM );
+
+   unit = t->whichTMU;
+   glideLod =  t->info.largeLodLog2 - level + tObj->BaseLevel;
+
+   switch ( unit ) {
+   case TDFX_TMU0:
+   case TDFX_TMU1:
+      grTexDownloadMipMapLevel( unit,
+				t->range[unit]->startAddr,
+				glideLod,
+				t->info.largeLodLog2,
+				t->info.aspectRatioLog2,
+				t->info.format,
+				GR_MIPMAPLEVELMASK_BOTH,
+				t->image[level].data );
+      break;
+
+   case TDFX_TMU_SPLIT:
+      grTexDownloadMipMapLevel( GR_TMU0,
+				t->range[GR_TMU0]->startAddr,
+				glideLod,
+				t->info.largeLodLog2,
+				t->info.aspectRatioLog2,
+				t->info.format,
+				GR_MIPMAPLEVELMASK_ODD,
+				t->image[level].data );
+
+      grTexDownloadMipMapLevel( GR_TMU1,
+				t->range[GR_TMU1]->startAddr,
+				glideLod,
+				t->info.largeLodLog2,
+				t->info.aspectRatioLog2,
+				t->info.format,
+				GR_MIPMAPLEVELMASK_EVEN,
+				t->image[level].data );
+      break;
+
+   case TDFX_TMU_BOTH:
+      grTexDownloadMipMapLevel( GR_TMU0,
+				t->range[GR_TMU0]->startAddr,
+				glideLod,
+				t->info.largeLodLog2,
+				t->info.aspectRatioLog2,
+				t->info.format,
+				GR_MIPMAPLEVELMASK_BOTH,
+				t->image[level].data );
+
+      grTexDownloadMipMapLevel( GR_TMU1,
+				t->range[GR_TMU1]->startAddr,
+				glideLod,
+				t->info.largeLodLog2,
+				t->info.aspectRatioLog2,
+				t->info.format,
+				GR_MIPMAPLEVELMASK_BOTH,
+				t->image[level].data );
+      break;
+
+   default:
+      gl_problem( ctx, "error in tdfxTMReloadMipMapLevel(): wrong unit" );
+      break;
+   }
 }
 
 
-/*
- * Allocate space for the given texture in texture memory then
+/* Allocate space for the given texture in texture memory then
  * download (copy) it into that space.
  */
-void
-tdfxTMMoveInTM_NoLock( tdfxContextPtr fxMesa, struct gl_texture_object *tObj,
-                       FxU32 targetTMU )
+void tdfxTMMoveInTMLocked( tdfxContextPtr fxMesa,
+			    struct gl_texture_object *tObj, FxU32 targetTMU )
 {
-    tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-    FxU32 texmemsize;
-
-    fxMesa->stats.reqTexUpload++;
-
-    if (ti->isInTM) {
-        if (ti->whichTMU == targetTMU)
-            return;
-        if (targetTMU == TDFX_TMU_SPLIT || ti->whichTMU == TDFX_TMU_SPLIT) {
-            tdfxTMMoveOutTM_NoLock(fxMesa, tObj);
-        }
-        else {
-            if (ti->whichTMU == TDFX_TMU_BOTH)
-                return;
-            targetTMU = TDFX_TMU_BOTH;
-        }
-    }
-
-    ti->whichTMU = targetTMU;
-
-    switch (targetTMU) {
-    case TDFX_TMU0:
-    case TDFX_TMU1:
-        texmemsize = FX_grTexTextureMemRequired_NoLock(GR_MIPMAPLEVELMASK_BOTH,
-                                                       &(ti->info));
-        ti->tm[targetTMU] = AllocTexMem(fxMesa, targetTMU, texmemsize);
-        break;
-    case TDFX_TMU_SPLIT:
-        texmemsize = FX_grTexTextureMemRequired_NoLock(GR_MIPMAPLEVELMASK_ODD,
-                                                       &(ti->info));
-        ti->tm[TDFX_TMU0] = AllocTexMem(fxMesa, TDFX_TMU0, texmemsize);
-        if (ti->tm[TDFX_TMU0])
-           fxMesa->stats.memTexUpload += texmemsize;
-
-        texmemsize = FX_grTexTextureMemRequired_NoLock(GR_MIPMAPLEVELMASK_EVEN,
-                                                       &(ti->info));
-        ti->tm[TDFX_TMU1] = AllocTexMem(fxMesa, TDFX_TMU1, texmemsize);
-        break;
-    case TDFX_TMU_BOTH:
-        texmemsize = FX_grTexTextureMemRequired_NoLock(GR_MIPMAPLEVELMASK_BOTH,
-                                                       &(ti->info));
-        ti->tm[TDFX_TMU0] = AllocTexMem(fxMesa, TDFX_TMU0, texmemsize);
-        if (ti->tm[TDFX_TMU0])
-           fxMesa->stats.memTexUpload += texmemsize;
-
-        texmemsize = FX_grTexTextureMemRequired_NoLock(GR_MIPMAPLEVELMASK_BOTH,
-                                                       &(ti->info));
-        ti->tm[TDFX_TMU1] = AllocTexMem(fxMesa, TDFX_TMU1, texmemsize);
-        break;
-    default:
-        gl_problem(NULL, "error in tdfxTMMoveInTM() -> bad tmu (%d)");
-        return;
-    }
-
-    ti->reloadImages = GL_TRUE;
-    ti->isInTM = GL_TRUE;
-
-    fxMesa->stats.texUpload++;
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+   FxU32 size;
+
+   fxMesa->stats.reqTexUpload++;
+
+   if ( t->isInTM ) {
+      if ( t->whichTMU == targetTMU )
+	 return;
+
+      if ( targetTMU == TDFX_TMU_SPLIT || t->whichTMU == TDFX_TMU_SPLIT ) {
+	 tdfxTMMoveOutTMLocked( fxMesa, tObj );
+      } else {
+	 if ( t->whichTMU == TDFX_TMU_BOTH )
+	    return;
+	 targetTMU = TDFX_TMU_BOTH;
+      }
+   }
+
+   t->whichTMU = targetTMU;
+
+   switch ( targetTMU ) {
+   case TDFX_TMU0:
+   case TDFX_TMU1:
+      size = grTexTextureMemRequired( GR_MIPMAPLEVELMASK_BOTH, &t->info );
+      t->range[targetTMU] = tdfxTMAllocTexMem(fxMesa, targetTMU, size);
+      break;
+
+   case TDFX_TMU_SPLIT:
+      size = grTexTextureMemRequired( GR_MIPMAPLEVELMASK_ODD, &t->info );
+      t->range[TDFX_TMU0] = tdfxTMAllocTexMem( fxMesa, TDFX_TMU0, size );
+      if ( t->range[TDFX_TMU0] )
+	 fxMesa->stats.memTexUpload += size;
+
+      size = grTexTextureMemRequired( GR_MIPMAPLEVELMASK_EVEN, &t->info );
+      t->range[TDFX_TMU1] = tdfxTMAllocTexMem( fxMesa, TDFX_TMU1, size );
+      break;
+
+   case TDFX_TMU_BOTH:
+      size = grTexTextureMemRequired( GR_MIPMAPLEVELMASK_BOTH, &t->info );
+      t->range[TDFX_TMU0] = tdfxTMAllocTexMem( fxMesa, TDFX_TMU0, size );
+      if ( t->range[TDFX_TMU0] )
+	 fxMesa->stats.memTexUpload += size;
+
+      size = grTexTextureMemRequired( GR_MIPMAPLEVELMASK_BOTH, &t->info );
+      t->range[TDFX_TMU1] = tdfxTMAllocTexMem( fxMesa, TDFX_TMU1, size );
+      break;
+
+   default:
+      gl_problem( NULL, "error in tdfxTMMoveInTM() -> bad unit (%d)" );
+      return;
+   }
+
+   t->reloadImages = GL_TRUE;
+   t->isInTM = GL_TRUE;
+
+   fxMesa->stats.texUpload++;
 }
 
 
-/*
- * Move the given texture out of hardware texture memory.
+/* Move the given texture out of hardware texture memory.
  * This deallocates the texture's memory space.
  */
-void
-tdfxTMMoveOutTM_NoLock( tdfxContextPtr fxMesa, struct gl_texture_object *tObj )
+void tdfxTMMoveOutTMLocked( tdfxContextPtr fxMesa,
+			    struct gl_texture_object *tObj )
 {
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-
-    if (MESA_VERBOSE & VERBOSE_DRIVER) {
-        fprintf(stderr, "fxmesa: fxTMMoveOutTM(%p (%d))\n", tObj, tObj->Name);
-    }
-
-    /*
-    VerifyFreeList(fxMesa, 0);
-    VerifyFreeList(fxMesa, 1);
-    */
-
-    if (!ti || !ti->isInTM)
-        return;
-
-    switch (ti->whichTMU) {
-    case TDFX_TMU0:
-    case TDFX_TMU1:
-        RemoveRange_NoLock(fxMesa, ti->whichTMU, ti->tm[ti->whichTMU]);
-        break;
-    case TDFX_TMU_SPLIT:
-    case TDFX_TMU_BOTH:
-        assert(!shared->umaTexMemory);
-        RemoveRange_NoLock(fxMesa, TDFX_TMU0, ti->tm[TDFX_TMU0]);
-        RemoveRange_NoLock(fxMesa, TDFX_TMU1, ti->tm[TDFX_TMU1]);
-        break;
-    default:
-        gl_problem(NULL, "tdfx driver: bad tmu in tdfxTMMOveOutTM()");
-        return;
-    }
-
-    ti->isInTM = GL_FALSE;
-    ti->tm[0] = NULL;
-    ti->tm[1] = NULL;
-    ti->whichTMU = TDFX_TMU_NONE;
-
-    /*
-    VerifyFreeList(fxMesa, 0);
-    VerifyFreeList(fxMesa, 1);
-    */
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE ) {
+      fprintf( stderr, __FUNCTION__ "( %p (%d) )\n", tObj, tObj->Name );
+      tdfxTMVerifyFreeList( fxMesa, 0 );
+      tdfxTMVerifyFreeList( fxMesa, 1 );
+   }
+
+   if ( !t || !t->isInTM )
+      return;
+
+   switch ( t->whichTMU ) {
+   case TDFX_TMU0:
+   case TDFX_TMU1:
+      tdfxTMRemoveRangeLocked( fxMesa, t->whichTMU, t->range[t->whichTMU] );
+      break;
+
+   case TDFX_TMU_SPLIT:
+   case TDFX_TMU_BOTH:
+      assert( !tss->umaTexMemory );
+      tdfxTMRemoveRangeLocked( fxMesa, TDFX_TMU0, t->range[TDFX_TMU0] );
+      tdfxTMRemoveRangeLocked( fxMesa, TDFX_TMU1, t->range[TDFX_TMU1] );
+      break;
+
+   default:
+      gl_problem( NULL, "tdfx driver: bad unit in tdfxTMMOveOutTM()" );
+      return;
+   }
+
+   t->isInTM = GL_FALSE;
+   t->range[0] = NULL;
+   t->range[1] = NULL;
+   t->whichTMU = TDFX_TMU_NONE;
+
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE ) {
+      tdfxTMVerifyFreeList( fxMesa, 0 );
+      tdfxTMVerifyFreeList( fxMesa, 1 );
+   }
 }
 
 
-/*
- * Called via glDeleteTexture to delete a texture object.
- */
-void
-tdfxTMFreeTexture(tdfxContextPtr fxMesa, struct gl_texture_object *tObj)
+void tdfxTMFreeTextureLocked( tdfxContextPtr fxMesa,
+			      struct gl_texture_object *tObj )
 {
-    tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-    if (ti) {
-        int i;
-        tdfxTMMoveOutTM(fxMesa, tObj);
-        for (i = 0; i < MAX_TEXTURE_LEVELS; i++) {
-            if (ti->mipmapLevel[i].data) {
-                FREE(ti->mipmapLevel[i].data);
-                ti->mipmapLevel[i].data = NULL;
-            }
-        }
-        FREE(ti);
-        tObj->DriverData = NULL;
-    }
-    /*
-    VerifyFreeList(fxMesa, 0);
-    VerifyFreeList(fxMesa, 1);
-    */
-}
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+
+   if ( t ) {
+      int i;
+      tdfxTMMoveOutTMLocked( fxMesa, tObj );
+      for ( i = 0 ; i < MAX_TEXTURE_LEVELS ; i++ ) {
+	 if ( t->image[i].original.data ) {
+	    FREE( t->image[i].original.data );
+	    t->image[i].original.data = NULL;
+	    t->image[i].original.width = 0;
+	    t->image[i].original.height = 0;
+	    t->image[i].original.size = 0;
+	 }
+	 if ( t->image[i].rescaled.data ) {
+	    FREE( t->image[i].rescaled.data );
+	    t->image[i].rescaled.data = NULL;
+	    t->image[i].rescaled.width = 0;
+	    t->image[i].rescaled.height = 0;
+	    t->image[i].rescaled.size = 0;
+	 }
+      }
+      FREE( t );
+      tObj->DriverData = NULL;
+   }
 
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE ) {
+      tdfxTMVerifyFreeList( fxMesa, 0 );
+      tdfxTMVerifyFreeList( fxMesa, 1 );
+   }
+}
 
 
-/*
- * After a context switch this function will be called to restore
+/* After a context switch this function will be called to restore
  * texture memory for the new context.
  */
-void tdfxTMRestoreTextures_NoLock( tdfxContextPtr fxMesa )
+void tdfxTMRestoreTexturesLocked( tdfxContextPtr fxMesa )
 {
    GLcontext *ctx = fxMesa->glCtx;
    struct gl_texture_object *tObj;
    int i;
 
    for ( tObj = ctx->Shared->TexObjectList ; tObj ; tObj = tObj->Next ) {
-      tdfxTexInfo *ti = TDFX_TEXTURE_DATA( tObj );
-      if ( ti && ti->isInTM ) {
+      tdfxTexObjPtr t = TDFX_TEXTURE_DATA( tObj );
+      if ( t && t->isInTM ) {
 	 for ( i = 0 ; i < MAX_TEXTURE_UNITS ; i++ ) {
 	    if ( ctx->Texture.Unit[i].Current == tObj ) {
-	       tdfxTMDownloadTexture( fxMesa, tObj );
+	       tdfxTMDownloadTextureLocked( fxMesa, tObj );
 	       break;
 	    }
 	 }
 	 if ( i == MAX_TEXTURE_UNITS ) {
-	    tdfxTMMoveOutTM_NoLock( fxMesa, tObj );
+	    tdfxTMMoveOutTMLocked( fxMesa, tObj );
 	 }
       }
    }
-   /*
-   VerifyFreeList(fxMesa, 0);
-   VerifyFreeList(fxMesa, 1);
-   */
+
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE ) {
+      tdfxTMVerifyFreeList( fxMesa, 0 );
+      tdfxTMVerifyFreeList( fxMesa, 1 );
+   }
 }
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.h b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.h
index 1965aaa84..6979367a0 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.h
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.h
@@ -38,47 +38,39 @@
 #ifndef __TDFX_TEXMAN_H__
 #define __TDFX_TEXMAN_H__
 
-
 #include "tdfx_lock.h"
 
-
 extern void tdfxTMInit( tdfxContextPtr fxMesa );
-
 extern void tdfxTMClose( tdfxContextPtr fxMesa );
 
-extern void tdfxTMDownloadTexture(tdfxContextPtr fxMesa,
-                                  struct gl_texture_object *tObj);
-
-extern void tdfxTMReloadMipMapLevel( GLcontext *ctx,
-				     struct gl_texture_object *tObj,
-				     GLint level );
-
-extern void tdfxTMMoveInTM_NoLock( tdfxContextPtr fxMesa,
-                                   struct gl_texture_object *tObj,
-                                   FxU32 targetTMU );
-
-extern void tdfxTMMoveOutTM_NoLock( tdfxContextPtr fxMesa,
-                                    struct gl_texture_object *tObj );
-
-extern void tdfxTMFreeTexture( tdfxContextPtr fxMesa,
-			       struct gl_texture_object *tObj );
-
-extern void tdfxTMRestoreTextures_NoLock( tdfxContextPtr fxMesa );
-
-
-#define tdfxTMMoveInTM( fxMesa, tObj, targetTMU )		\
-   do {								\
-      LOCK_HARDWARE( fxMesa );					\
-      tdfxTMMoveInTM_NoLock( fxMesa, tObj, targetTMU );		\
-      UNLOCK_HARDWARE( fxMesa );				\
-   } while (0)
-
-#define tdfxTMMoveOutTM( fxMesa, tObj )				\
-   do {								\
-      LOCK_HARDWARE( fxMesa );					\
-      tdfxTMMoveOutTM_NoLock( fxMesa, tObj );			\
-      UNLOCK_HARDWARE( fxMesa );				\
-   } while (0)
-
+extern void tdfxTMDownloadTextureLocked( tdfxContextPtr fxMesa,
+					 struct gl_texture_object *tObj );
+extern void tdfxTMReloadMipMapLevelLocked( GLcontext *ctx,
+					   struct gl_texture_object *tObj,
+					   GLint level );
+extern void tdfxTMMoveInTMLocked( tdfxContextPtr fxMesa,
+				  struct gl_texture_object *tObj,
+				  FxU32 targetTMU );
+extern void tdfxTMMoveOutTMLocked( tdfxContextPtr fxMesa,
+				   struct gl_texture_object *tObj );
+extern void tdfxTMRestoreTexturesLocked( tdfxContextPtr fxMesa );
+
+extern void tdfxTMFreeTextureLocked( tdfxContextPtr fxMesa,
+				     struct gl_texture_object *tObj );
+
+
+#define tdfxTMMoveInTM( fxMesa, tObj, targetTMU )			\
+do {									\
+   LOCK_HARDWARE( fxMesa );						\
+   tdfxTMMoveInTMLocked( fxMesa, tObj, targetTMU );			\
+   UNLOCK_HARDWARE( fxMesa );						\
+} while (0)
+
+#define tdfxTMMoveOutTM( fxMesa, tObj )					\
+do {									\
+   LOCK_HARDWARE( fxMesa );						\
+   tdfxTMMoveOutTMLocked( fxMesa, tObj );				\
+   UNLOCK_HARDWARE( fxMesa );						\
+} while (0)
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.c
index 5493eea8a..368f051a3 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.c
@@ -38,7 +38,6 @@
 #include "tdfx_state.h"
 #include "tdfx_tex.h"
 #include "tdfx_texman.h"
-#include "tdfx_texstate.h"
 
 
 /* =============================================================
@@ -1270,105 +1269,105 @@ SetupDoubleTexEnvVoodoo3(GLcontext *ctx, int tmu0,
 static void
 setupSingleTMU(tdfxContextPtr fxMesa, struct gl_texture_object *tObj)
 {
-   struct tdfxSharedState *shared = (struct tdfxSharedState *) fxMesa->glCtx->Shared->DriverData;
-   tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
    const GLcontext *ctx = fxMesa->glCtx;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ctx->Shared->DriverData;
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
 
    /* Make sure we're not loaded incorrectly */
-   if (ti->isInTM && !shared->umaTexMemory) {
+   if (t->isInTM && !tss->umaTexMemory) {
       /* if doing filtering between mipmap levels, alternate mipmap levels
        * must be in alternate TMUs.
        */
-      if (ti->LODblend) {
-         if (ti->whichTMU != TDFX_TMU_SPLIT)
-            tdfxTMMoveOutTM_NoLock(fxMesa, tObj);
+      if (t->LODblend) {
+         if (t->whichTMU != TDFX_TMU_SPLIT)
+            tdfxTMMoveOutTMLocked(fxMesa, tObj);
       }
       else {
-         if (ti->whichTMU == TDFX_TMU_SPLIT)
-            tdfxTMMoveOutTM_NoLock(fxMesa, tObj);
+         if (t->whichTMU == TDFX_TMU_SPLIT)
+            tdfxTMMoveOutTMLocked(fxMesa, tObj);
       }
    }
 
    /* Make sure we're loaded correctly */
-   if (!ti->isInTM) {
+   if (!t->isInTM) {
       /* Have to download the texture */
-      if (shared->umaTexMemory) {
-         tdfxTMMoveInTM_NoLock(fxMesa, tObj, TDFX_TMU0);
+      if (tss->umaTexMemory) {
+         tdfxTMMoveInTMLocked(fxMesa, tObj, TDFX_TMU0);
       }
       else {
          /* Voodoo3 (split texture memory) */
-         if (ti->LODblend) {
-            tdfxTMMoveInTM_NoLock(fxMesa, tObj, TDFX_TMU_SPLIT);
+         if (t->LODblend) {
+            tdfxTMMoveInTMLocked(fxMesa, tObj, TDFX_TMU_SPLIT);
          }
          else {
 #if 0
             /* XXX putting textures into the second memory bank when the
              * first bank is full is not working at this time.
              */
-            if (fxMesa->haveTwoTMUs) {
-               GLint memReq = FX_grTexTextureMemRequired_NoLock(
-                                       GR_MIPMAPLEVELMASK_BOTH, &(ti->info));
-               if (shared->freeTexMem[TDFX_TMU0] > memReq) {
-                  tdfxTMMoveInTM_NoLock(fxMesa, tObj, TDFX_TMU0);
+            if (fxMesa->numTMUs > 1) {
+               GLint memReq =
+		  grTexTextureMemRequired( GR_MIPMAPLEVELMASK_BOTH, &t->info );
+               if (tss->freeTexMem[TDFX_TMU0] > memReq) {
+                  tdfxTMMoveInTMLocked(fxMesa, tObj, TDFX_TMU0);
                }
                else {
-                  tdfxTMMoveInTM_NoLock(fxMesa, tObj, TDFX_TMU1);
+                  tdfxTMMoveInTMLocked(fxMesa, tObj, TDFX_TMU1);
                }
             }
             else
 #endif
             {
-               tdfxTMMoveInTM_NoLock(fxMesa, tObj, TDFX_TMU0);
+               tdfxTMMoveInTMLocked( fxMesa, tObj, TDFX_TMU0 );
             }
          }
       }
    }
 
-   if (ti->LODblend && ti->whichTMU == TDFX_TMU_SPLIT) {
+   if (t->LODblend && t->whichTMU == TDFX_TMU_SPLIT) {
       /* mipmap levels split between texture banks */
       GLint u;
 
-      if (ti->info.format == GR_TEXFMT_P_8 && !ctx->Texture.SharedPalette) {
+      if (t->info.format == GR_TEXFMT_P_8 && !ctx->Texture.SharedPalette) {
          fxMesa->TexPalette.Type = GR_TEXTABLE_PALETTE_6666_EXT;
-         fxMesa->TexPalette.Data = &(ti->palette);
+         fxMesa->TexPalette.Data = &(t->palette);
          fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PALETTE;
       }
 
       for (u = 0; u < 2; u++) {
-         fxMesa->TexParams[u].sClamp = ti->sClamp;
-         fxMesa->TexParams[u].tClamp = ti->tClamp;
-         fxMesa->TexParams[u].minFilt = ti->minFilt;
-         fxMesa->TexParams[u].magFilt = ti->magFilt;
-         fxMesa->TexParams[u].mmMode = ti->mmMode;
-         fxMesa->TexParams[u].LODblend = ti->LODblend;
+         fxMesa->TexParams[u].sClamp = t->sClamp;
+         fxMesa->TexParams[u].tClamp = t->tClamp;
+         fxMesa->TexParams[u].minFilt = t->minFilt;
+         fxMesa->TexParams[u].magFilt = t->magFilt;
+         fxMesa->TexParams[u].mmMode = t->mmMode;
+         fxMesa->TexParams[u].LODblend = t->LODblend;
          fxMesa->TexParams[u].LodBias = ctx->Texture.Unit[u].LodBias;
       }
       fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PARAMS;
 
-      fxMesa->TexSource[0].StartAddress = ti->tm[TDFX_TMU0]->startAddr;
+      fxMesa->TexSource[0].StartAddress = t->range[TDFX_TMU0]->startAddr;
       fxMesa->TexSource[0].EvenOdd = GR_MIPMAPLEVELMASK_ODD;
-      fxMesa->TexSource[0].Info = &(ti->info);
-      fxMesa->TexSource[1].StartAddress = ti->tm[TDFX_TMU1]->startAddr;
+      fxMesa->TexSource[0].Info = &(t->info);
+      fxMesa->TexSource[1].StartAddress = t->range[TDFX_TMU1]->startAddr;
       fxMesa->TexSource[1].EvenOdd = GR_MIPMAPLEVELMASK_EVEN;
-      fxMesa->TexSource[1].Info = &(ti->info);
+      fxMesa->TexSource[1].Info = &(t->info);
       fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_SOURCE;
    }
    else {
       FxU32 tmu;
 
-      if (ti->whichTMU == TDFX_TMU_BOTH)
+      if (t->whichTMU == TDFX_TMU_BOTH)
          tmu = TDFX_TMU0;
       else
-         tmu = ti->whichTMU;
+         tmu = t->whichTMU;
 
-      if (shared->umaTexMemory) {
-         assert(ti->whichTMU == TDFX_TMU0);
+      if (tss->umaTexMemory) {
+         assert(t->whichTMU == TDFX_TMU0);
          assert(tmu == TDFX_TMU0);
       }
 
-      if (ti->info.format == GR_TEXFMT_P_8 && !ctx->Texture.SharedPalette) {
+      if (t->info.format == GR_TEXFMT_P_8 && !ctx->Texture.SharedPalette) {
          fxMesa->TexPalette.Type = GR_TEXTABLE_PALETTE_6666_EXT;
-         fxMesa->TexPalette.Data = &(ti->palette);
+         fxMesa->TexPalette.Data = &(t->palette);
          fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PALETTE;
       }
 
@@ -1377,18 +1376,18 @@ setupSingleTMU(tdfxContextPtr fxMesa, struct gl_texture_object *tObj)
        * texture memory, so perhaps it's not a good idea.
        */
 
-      if (fxMesa->TexParams[tmu].sClamp != ti->sClamp ||
-          fxMesa->TexParams[tmu].tClamp != ti->tClamp ||
-          fxMesa->TexParams[tmu].minFilt != ti->minFilt ||
-          fxMesa->TexParams[tmu].magFilt != ti->magFilt ||
-          fxMesa->TexParams[tmu].mmMode != ti->mmMode ||
+      if (fxMesa->TexParams[tmu].sClamp != t->sClamp ||
+          fxMesa->TexParams[tmu].tClamp != t->tClamp ||
+          fxMesa->TexParams[tmu].minFilt != t->minFilt ||
+          fxMesa->TexParams[tmu].magFilt != t->magFilt ||
+          fxMesa->TexParams[tmu].mmMode != t->mmMode ||
           fxMesa->TexParams[tmu].LODblend != FXFALSE ||
           fxMesa->TexParams[tmu].LodBias != ctx->Texture.Unit[tmu].LodBias) {
-         fxMesa->TexParams[tmu].sClamp = ti->sClamp;
-         fxMesa->TexParams[tmu].tClamp = ti->tClamp;
-         fxMesa->TexParams[tmu].minFilt = ti->minFilt;
-         fxMesa->TexParams[tmu].magFilt = ti->magFilt;
-         fxMesa->TexParams[tmu].mmMode = ti->mmMode;
+         fxMesa->TexParams[tmu].sClamp = t->sClamp;
+         fxMesa->TexParams[tmu].tClamp = t->tClamp;
+         fxMesa->TexParams[tmu].minFilt = t->minFilt;
+         fxMesa->TexParams[tmu].magFilt = t->magFilt;
+         fxMesa->TexParams[tmu].mmMode = t->mmMode;
          fxMesa->TexParams[tmu].LODblend = FXFALSE;
          fxMesa->TexParams[tmu].LodBias = ctx->Texture.Unit[tmu].LodBias;
          fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PARAMS;
@@ -1397,16 +1396,16 @@ setupSingleTMU(tdfxContextPtr fxMesa, struct gl_texture_object *tObj)
       /* Glide texture source info */
       fxMesa->TexSource[0].Info = NULL;
       fxMesa->TexSource[1].Info = NULL;
-      if (ti->tm[tmu]) {
-         fxMesa->TexSource[tmu].StartAddress = ti->tm[tmu]->startAddr;
+      if (t->range[tmu]) {
+         fxMesa->TexSource[tmu].StartAddress = t->range[tmu]->startAddr;
          fxMesa->TexSource[tmu].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[tmu].Info = &(ti->info);
+         fxMesa->TexSource[tmu].Info = &(t->info);
          fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_SOURCE;
       }
    }
 
-   fxMesa->sScale0 = ti->sScale;
-   fxMesa->tScale0 = ti->tScale;
+   fxMesa->sScale0 = t->sScale;
+   fxMesa->tScale0 = t->tScale;
 }
 
 static void
@@ -1420,12 +1419,12 @@ selectSingleTMUSrc(tdfxContextPtr fxMesa, GLint tmu, FxBool LODblend)
       fxMesa->TexCombine[0].InvertRGB = FXFALSE;
       fxMesa->TexCombine[0].InvertAlpha = FXFALSE;
 
-      if (fxMesa->haveTwoTMUs) {
-         const struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-         const struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
+      if ( fxMesa->numTMUs > 1 ) {
+         const struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+         const tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
          int tmu;
 
-         if (shared->umaTexMemory)
+         if (tss->umaTexMemory)
             tmu = GR_TMU0;
          else
             tmu = GR_TMU1;
@@ -1447,7 +1446,7 @@ selectSingleTMUSrc(tdfxContextPtr fxMesa, GLint tmu, FxBool LODblend)
          fxMesa->TexCombine[0].FactorAlpha = GR_COMBINE_FACTOR_NONE;
          fxMesa->TexCombine[0].InvertRGB = FXFALSE;
          fxMesa->TexCombine[0].InvertAlpha = FXFALSE;
-         if (fxMesa->haveTwoTMUs) {
+         if ( fxMesa->numTMUs > 1 ) {
             fxMesa->TexCombine[1].FunctionRGB = GR_COMBINE_FUNCTION_ZERO;
             fxMesa->TexCombine[1].FactorRGB = GR_COMBINE_FACTOR_NONE;
             fxMesa->TexCombine[1].FunctionAlpha = GR_COMBINE_FUNCTION_ZERO;
@@ -1511,7 +1510,7 @@ static void print_state(tdfxContextPtr fxMesa)
 static void setupTextureSingleTMU(GLcontext * ctx, GLuint unit)
 {
    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-   tdfxTexInfo *ti;
+   tdfxTexObjPtr t;
    struct gl_texture_object *tObj;
    int tmu;
    GLenum envMode, baseFormat;
@@ -1530,17 +1529,17 @@ static void setupTextureSingleTMU(GLcontext * ctx, GLuint unit)
 
    setupSingleTMU(fxMesa, tObj);
 
-   ti = TDFX_TEXTURE_DATA(tObj);
-   if (ti->whichTMU == TDFX_TMU_BOTH)
+   t = TDFX_TEXTURE_DATA(tObj);
+   if (t->whichTMU == TDFX_TMU_BOTH)
       tmu = TDFX_TMU0;
    else
-      tmu = ti->whichTMU;
+      tmu = t->whichTMU;
 
    if (fxMesa->tmuSrc != tmu) {
-      selectSingleTMUSrc(fxMesa, tmu, ti->LODblend);
+      selectSingleTMUSrc(fxMesa, tmu, t->LODblend);
    }
 
-   if (ti->reloadImages)
+   if (t->reloadImages)
       fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_IMAGES;
 
    /* Some texture environments not supported */
@@ -1627,35 +1626,35 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
 #define T0_IN_TMU1     0x10
 #define T1_IN_TMU1     0x20
 
-    const struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    const struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
+    const struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+    const tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
     const GLcontext *ctx = fxMesa->glCtx;
-    tdfxTexInfo *ti0 = TDFX_TEXTURE_DATA(tObj0);
-    tdfxTexInfo *ti1 = TDFX_TEXTURE_DATA(tObj1);
+    tdfxTexObjPtr t0 = TDFX_TEXTURE_DATA(tObj0);
+    tdfxTexObjPtr t1 = TDFX_TEXTURE_DATA(tObj1);
     GLuint tstate = 0;
     int tmu0 = 0, tmu1 = 1;
 
-    if (shared->umaTexMemory) {
-       if (!ti0->isInTM) {
-          tdfxTMMoveInTM_NoLock(fxMesa, tObj0, TDFX_TMU0);
-          assert(ti0->isInTM);
+    if (tss->umaTexMemory) {
+       if (!t0->isInTM) {
+          tdfxTMMoveInTMLocked(fxMesa, tObj0, TDFX_TMU0);
+          assert(t0->isInTM);
        }
-       if (!ti1->isInTM) {
-          tdfxTMMoveInTM_NoLock(fxMesa, tObj1, TDFX_TMU0);
-          assert(ti1->isInTM);
+       if (!t1->isInTM) {
+          tdfxTMMoveInTMLocked(fxMesa, tObj1, TDFX_TMU0);
+          assert(t1->isInTM);
        }
     }
     else {
        /* We shouldn't need to do this. There is something wrong with
           multitexturing when the TMUs are swapped. So, we're forcing
           them to always be loaded correctly. !!! */
-       if (ti0->whichTMU == TDFX_TMU1)
-           tdfxTMMoveOutTM_NoLock(fxMesa, tObj0);
-       if (ti1->whichTMU == TDFX_TMU0)
-           tdfxTMMoveOutTM_NoLock(fxMesa, tObj1);
+       if (t0->whichTMU == TDFX_TMU1)
+           tdfxTMMoveOutTMLocked(fxMesa, tObj0);
+       if (t1->whichTMU == TDFX_TMU0)
+           tdfxTMMoveOutTMLocked(fxMesa, tObj1);
 
-       if (ti0->isInTM) {
-           switch (ti0->whichTMU) {
+       if (t0->isInTM) {
+           switch (t0->whichTMU) {
            case TDFX_TMU0:
                tstate |= T0_IN_TMU0;
                break;
@@ -1673,8 +1672,8 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
        else
            tstate |= T0_NOT_IN_TMU;
 
-       if (ti1->isInTM) {
-           switch (ti1->whichTMU) {
+       if (t1->isInTM) {
+           switch (t1->whichTMU) {
            case TDFX_TMU0:
                tstate |= T1_IN_TMU0;
                break;
@@ -1697,7 +1696,7 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
        if (!(((tstate & T0_IN_TMU0) && (tstate & T1_IN_TMU1)) ||
              ((tstate & T0_IN_TMU1) && (tstate & T1_IN_TMU0)))) {
            if (tObj0 == tObj1) {
-              tdfxTMMoveInTM_NoLock(fxMesa, tObj1, TDFX_TMU_BOTH);
+              tdfxTMMoveInTMLocked(fxMesa, tObj1, TDFX_TMU_BOTH);
            }
            else {
                /* Find the minimal way to correct the situation */
@@ -1705,10 +1704,10 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
                    /* We have one in the standard order, setup the other */
                    if (tstate & T0_IN_TMU0) {
                       /* T0 is in TMU0, put T1 in TMU1 */
-                      tdfxTMMoveInTM_NoLock(fxMesa, tObj1, TDFX_TMU1);
+                      tdfxTMMoveInTMLocked(fxMesa, tObj1, TDFX_TMU1);
                    }
                    else {
-                       tdfxTMMoveInTM_NoLock(fxMesa, tObj0, TDFX_TMU0);
+                       tdfxTMMoveInTMLocked(fxMesa, tObj0, TDFX_TMU0);
                    }
                    /* tmu0 and tmu1 are setup */
                }
@@ -1716,36 +1715,36 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
                    /* we have one in the reverse order, setup the other */
                    if (tstate & T1_IN_TMU0) {
                       /* T1 is in TMU0, put T0 in TMU1 */
-                      tdfxTMMoveInTM_NoLock(fxMesa, tObj0, TDFX_TMU1);
+                      tdfxTMMoveInTMLocked(fxMesa, tObj0, TDFX_TMU1);
                    }
                    else {
-                       tdfxTMMoveInTM_NoLock(fxMesa, tObj1, TDFX_TMU0);
+                       tdfxTMMoveInTMLocked(fxMesa, tObj1, TDFX_TMU0);
                    }
                    tmu0 = 1;
                    tmu1 = 0;
                }
                else {              /* Nothing is loaded */
-                   tdfxTMMoveInTM_NoLock(fxMesa, tObj0, TDFX_TMU0);
-                   tdfxTMMoveInTM_NoLock(fxMesa, tObj1, TDFX_TMU1);
+                   tdfxTMMoveInTMLocked(fxMesa, tObj0, TDFX_TMU0);
+                   tdfxTMMoveInTMLocked(fxMesa, tObj1, TDFX_TMU1);
                    /* tmu0 and tmu1 are setup */
                }
            }
        }
     }
 
-    ti0->lastTimeUsed = fxMesa->texBindNumber;
-    ti1->lastTimeUsed = fxMesa->texBindNumber;
+    t0->lastTimeUsed = fxMesa->texBindNumber;
+    t1->lastTimeUsed = fxMesa->texBindNumber;
 
 
     if (!ctx->Texture.SharedPalette) {
-        if (ti0->info.format == GR_TEXFMT_P_8) {
+        if (t0->info.format == GR_TEXFMT_P_8) {
             fxMesa->TexPalette.Type = GR_TEXTABLE_PALETTE_6666_EXT;
-            fxMesa->TexPalette.Data = &(ti0->palette);
+            fxMesa->TexPalette.Data = &(t0->palette);
             fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PALETTE;
         }
-        else if (ti1->info.format == GR_TEXFMT_P_8) {
+        else if (t1->info.format == GR_TEXFMT_P_8) {
             fxMesa->TexPalette.Type = GR_TEXTABLE_PALETTE_6666_EXT;
-            fxMesa->TexPalette.Data = &(ti1->palette);
+            fxMesa->TexPalette.Data = &(t1->palette);
             fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PALETTE;
         }
         else {
@@ -1756,25 +1755,25 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
     /*
      * Setup Unit 0
      */
-    assert(ti0->isInTM);
-    assert(ti0->tm[tmu0]);
-    fxMesa->TexSource[tmu0].StartAddress = ti0->tm[tmu0]->startAddr;
+    assert(t0->isInTM);
+    assert(t0->range[tmu0]);
+    fxMesa->TexSource[tmu0].StartAddress = t0->range[tmu0]->startAddr;
     fxMesa->TexSource[tmu0].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-    fxMesa->TexSource[tmu0].Info = &(ti0->info);
+    fxMesa->TexSource[tmu0].Info = &(t0->info);
     fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_SOURCE;
 
-    if (fxMesa->TexParams[tmu0].sClamp != ti0->sClamp ||
-        fxMesa->TexParams[tmu0].tClamp != ti0->tClamp ||
-        fxMesa->TexParams[tmu0].minFilt != ti0->minFilt ||
-        fxMesa->TexParams[tmu0].magFilt != ti0->magFilt ||
-        fxMesa->TexParams[tmu0].mmMode != ti0->mmMode ||
+    if (fxMesa->TexParams[tmu0].sClamp != t0->sClamp ||
+        fxMesa->TexParams[tmu0].tClamp != t0->tClamp ||
+        fxMesa->TexParams[tmu0].minFilt != t0->minFilt ||
+        fxMesa->TexParams[tmu0].magFilt != t0->magFilt ||
+        fxMesa->TexParams[tmu0].mmMode != t0->mmMode ||
         fxMesa->TexParams[tmu0].LODblend != FXFALSE ||
         fxMesa->TexParams[tmu0].LodBias != ctx->Texture.Unit[tmu0].LodBias) {
-       fxMesa->TexParams[tmu0].sClamp = ti0->sClamp;
-       fxMesa->TexParams[tmu0].tClamp = ti0->tClamp;
-       fxMesa->TexParams[tmu0].minFilt = ti0->minFilt;
-       fxMesa->TexParams[tmu0].magFilt = ti0->magFilt;
-       fxMesa->TexParams[tmu0].mmMode = ti0->mmMode;
+       fxMesa->TexParams[tmu0].sClamp = t0->sClamp;
+       fxMesa->TexParams[tmu0].tClamp = t0->tClamp;
+       fxMesa->TexParams[tmu0].minFilt = t0->minFilt;
+       fxMesa->TexParams[tmu0].magFilt = t0->magFilt;
+       fxMesa->TexParams[tmu0].mmMode = t0->mmMode;
        fxMesa->TexParams[tmu0].LODblend = FXFALSE;
        fxMesa->TexParams[tmu0].LodBias = ctx->Texture.Unit[tmu0].LodBias;
        fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PARAMS;
@@ -1783,42 +1782,42 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
     /*
      * Setup Unit 1
      */
-    if (shared->umaTexMemory) {
-        ASSERT(ti1->isInTM);
-        ASSERT(ti1->tm[0]);
-        fxMesa->TexSource[tmu1].StartAddress = ti1->tm[0]->startAddr;
+    if (tss->umaTexMemory) {
+        ASSERT(t1->isInTM);
+        ASSERT(t1->range[0]);
+        fxMesa->TexSource[tmu1].StartAddress = t1->range[0]->startAddr;
         fxMesa->TexSource[tmu1].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-        fxMesa->TexSource[tmu1].Info = &(ti1->info);
+        fxMesa->TexSource[tmu1].Info = &(t1->info);
     }
     else {
-        ASSERT(ti1->isInTM);
-        ASSERT(ti1->tm[tmu1]);
-        fxMesa->TexSource[tmu1].StartAddress = ti1->tm[tmu1]->startAddr;
+        ASSERT(t1->isInTM);
+        ASSERT(t1->range[tmu1]);
+        fxMesa->TexSource[tmu1].StartAddress = t1->range[tmu1]->startAddr;
         fxMesa->TexSource[tmu1].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-        fxMesa->TexSource[tmu1].Info = &(ti1->info);
+        fxMesa->TexSource[tmu1].Info = &(t1->info);
     }
 
-    if (fxMesa->TexParams[tmu1].sClamp != ti1->sClamp ||
-        fxMesa->TexParams[tmu1].tClamp != ti1->tClamp ||
-        fxMesa->TexParams[tmu1].minFilt != ti1->minFilt ||
-        fxMesa->TexParams[tmu1].magFilt != ti1->magFilt ||
-        fxMesa->TexParams[tmu1].mmMode != ti1->mmMode ||
+    if (fxMesa->TexParams[tmu1].sClamp != t1->sClamp ||
+        fxMesa->TexParams[tmu1].tClamp != t1->tClamp ||
+        fxMesa->TexParams[tmu1].minFilt != t1->minFilt ||
+        fxMesa->TexParams[tmu1].magFilt != t1->magFilt ||
+        fxMesa->TexParams[tmu1].mmMode != t1->mmMode ||
         fxMesa->TexParams[tmu1].LODblend != FXFALSE ||
         fxMesa->TexParams[tmu1].LodBias != ctx->Texture.Unit[tmu1].LodBias) {
-       fxMesa->TexParams[tmu1].sClamp = ti1->sClamp;
-       fxMesa->TexParams[tmu1].tClamp = ti1->tClamp;
-       fxMesa->TexParams[tmu1].minFilt = ti1->minFilt;
-       fxMesa->TexParams[tmu1].magFilt = ti1->magFilt;
-       fxMesa->TexParams[tmu1].mmMode = ti1->mmMode;
+       fxMesa->TexParams[tmu1].sClamp = t1->sClamp;
+       fxMesa->TexParams[tmu1].tClamp = t1->tClamp;
+       fxMesa->TexParams[tmu1].minFilt = t1->minFilt;
+       fxMesa->TexParams[tmu1].magFilt = t1->magFilt;
+       fxMesa->TexParams[tmu1].mmMode = t1->mmMode;
        fxMesa->TexParams[tmu1].LODblend = FXFALSE;
        fxMesa->TexParams[tmu1].LodBias = ctx->Texture.Unit[tmu1].LodBias;
        fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PARAMS;
     }
 
-    fxMesa->sScale0 = ti0->sScale;
-    fxMesa->tScale0 = ti0->tScale;
-    fxMesa->sScale1 = ti1->sScale;
-    fxMesa->tScale1 = ti1->tScale;
+    fxMesa->sScale0 = t0->sScale;
+    fxMesa->tScale0 = t0->tScale;
+    fxMesa->sScale1 = t1->sScale;
+    fxMesa->tScale1 = t1->tScale;
 
 #undef T0_NOT_IN_TMU
 #undef T1_NOT_IN_TMU
@@ -1833,8 +1832,8 @@ static void setupTextureDoubleTMU(GLcontext * ctx)
    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
    struct gl_texture_object *tObj0 = ctx->Texture.Unit[0].CurrentD[2];
    struct gl_texture_object *tObj1 = ctx->Texture.Unit[1].CurrentD[2];
-   tdfxTexInfo *ti0 = TDFX_TEXTURE_DATA(tObj0);
-   tdfxTexInfo *ti1 = TDFX_TEXTURE_DATA(tObj1);
+   tdfxTexObjPtr t0 = TDFX_TEXTURE_DATA(tObj0);
+   tdfxTexObjPtr t1 = TDFX_TEXTURE_DATA(tObj1);
    struct gl_texture_image *baseImage0 = tObj0->Image[tObj0->BaseLevel];
    struct gl_texture_image *baseImage1 = tObj1->Image[tObj1->BaseLevel];
    const GLenum envMode0 = ctx->Texture.Unit[0].EnvMode;
@@ -1847,7 +1846,7 @@ static void setupTextureDoubleTMU(GLcontext * ctx)
 
    setupDoubleTMU(fxMesa, tObj0, tObj1);
 
-   if (ti0->reloadImages || ti1->reloadImages)
+   if (t0->reloadImages || t1->reloadImages)
       fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_IMAGES;
 
    fxMesa->tmuSrc = TDFX_TMU_BOTH;
@@ -1893,7 +1892,7 @@ static void setupTextureDoubleTMU(GLcontext * ctx)
    }
    else {
       int unit0, unit1;
-      if ((ti0->whichTMU == TDFX_TMU1) || (ti1->whichTMU == TDFX_TMU0))
+      if ((t0->whichTMU == TDFX_TMU1) || (t1->whichTMU == TDFX_TMU0))
          unit0 = 1;
       else
          unit0 = 0;
@@ -1924,34 +1923,34 @@ static void setupTextureDoubleTMU(GLcontext * ctx)
 }
 
 
-void
-tdfxUpdateTextureState( GLcontext *ctx )
+void tdfxUpdateTextureState( GLcontext *ctx )
 {
    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-   GLuint tex2Denabled = ctx->Texture.ReallyEnabled;
+   GLuint enabled = ctx->Texture.ReallyEnabled;
 
-   if (!fxMesa->haveTwoTMUs)
-      tex2Denabled &= TEXTURE0_2D;
+   if ( fxMesa->numTMUs == 1 )
+      enabled &= TEXTURE0_2D;
 
-   switch (tex2Denabled) {
+   switch ( enabled ) {
    case TEXTURE0_2D:
       LOCK_HARDWARE( fxMesa );  /* XXX remove locking eventually */
-      setupTextureSingleTMU(ctx, 0);
+      setupTextureSingleTMU( ctx, 0 );
       UNLOCK_HARDWARE( fxMesa );
       break;
    case TEXTURE1_2D:
       LOCK_HARDWARE( fxMesa );
-      setupTextureSingleTMU(ctx, 1);
+      setupTextureSingleTMU( ctx, 1 );
       UNLOCK_HARDWARE( fxMesa );
       break;
-   case (TEXTURE0_2D | TEXTURE1_2D):
+   case TEXTURE0_2D | TEXTURE1_2D:
       LOCK_HARDWARE( fxMesa );
-      setupTextureDoubleTMU(ctx);
+      setupTextureDoubleTMU( ctx );
       UNLOCK_HARDWARE( fxMesa );
       break;
+
    default:
-      /* disable hardware texturing */
-      if (TDFX_IS_NAPALM(fxMesa)) {
+      /* Disable hardware texturing */
+      if ( TDFX_IS_NAPALM( fxMesa ) ) {
          fxMesa->ColorCombineExt.SourceA = GR_CMBX_ITRGB;
          fxMesa->ColorCombineExt.ModeA = GR_FUNC_MODE_X;
          fxMesa->ColorCombineExt.SourceB = GR_CMBX_ZERO;
@@ -1972,9 +1971,8 @@ tdfxUpdateTextureState( GLcontext *ctx )
          fxMesa->AlphaCombineExt.InvertD = FXFALSE;
          fxMesa->AlphaCombineExt.Shift = 0;
          fxMesa->AlphaCombineExt.Invert = FXFALSE;
-      }
-      else {
-         /* Voodoo 3*/
+      } else {
+         /* Voodoo 3 */
          fxMesa->ColorCombine.Function = GR_COMBINE_FUNCTION_LOCAL;
          fxMesa->ColorCombine.Factor = GR_COMBINE_FACTOR_NONE;
          fxMesa->ColorCombine.Local = GR_COMBINE_LOCAL_ITERATED;
@@ -1997,106 +1995,103 @@ tdfxUpdateTextureState( GLcontext *ctx )
 }
 
 
-
-/*
- * This is a special case of texture state update.
+/* This is a special case of texture state update.
  * It's used when we've simply bound a new texture to a texture
  * unit and the new texture has the exact same attributes as the
  * previously bound texture.
  * This is very common in Quake3.
  */
-void
-tdfxUpdateTextureBinding( GLcontext *ctx )
+void tdfxUpdateTextureBinding( GLcontext *ctx )
 {
    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   const struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   const tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
    struct gl_texture_object *tObj0 = ctx->Texture.Unit[0].CurrentD[2];
    struct gl_texture_object *tObj1 = ctx->Texture.Unit[1].CurrentD[2];
-   tdfxTexInfo *ti0 = TDFX_TEXTURE_DATA(tObj0);
-   tdfxTexInfo *ti1 = TDFX_TEXTURE_DATA(tObj1);
-
-    const struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    const struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
+   tdfxTexObjPtr t0 = TDFX_TEXTURE_DATA(tObj0);
+   tdfxTexObjPtr t1 = TDFX_TEXTURE_DATA(tObj1);
 
-   if (ti0) {
-      fxMesa->sScale0 = ti0->sScale;
-      fxMesa->tScale0 = ti0->tScale;
-      if (ti0->info.format == GR_TEXFMT_P_8) {
+   if ( t0 ) {
+      fxMesa->sScale0 = t0->sScale;
+      fxMesa->tScale0 = t0->tScale;
+      if ( t0->info.format == GR_TEXFMT_P_8 ) {
          fxMesa->TexPalette.Type = GR_TEXTABLE_PALETTE_6666_EXT;
-         fxMesa->TexPalette.Data = &(ti0->palette);
+         fxMesa->TexPalette.Data = &t0->palette;
          fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PALETTE;
-      }
-      else if (ti1 && ti1->info.format == GR_TEXFMT_P_8) {
+      } else if ( t1 && t1->info.format == GR_TEXFMT_P_8 ) {
          fxMesa->TexPalette.Type = GR_TEXTABLE_PALETTE_6666_EXT;
-         fxMesa->TexPalette.Data = &(ti1->palette);
+         fxMesa->TexPalette.Data = &t1->palette;
          fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PALETTE;
       }
    }
-   if (ti1) {
-      fxMesa->sScale1 = ti1->sScale;
-      fxMesa->tScale1 = ti1->tScale;
+   if ( t1 ) {
+      fxMesa->sScale1 = t1->sScale;
+      fxMesa->tScale1 = t1->tScale;
    }
 
-   if (ctx->Texture.ReallyEnabled == TEXTURE0_2D) {
-      if (shared->umaTexMemory) {
-         fxMesa->TexSource[0].StartAddress = ti0->tm[0]->startAddr;
+   switch  ( ctx->Texture.ReallyEnabled ) {
+   case TEXTURE0_2D:
+      if ( tss->umaTexMemory ) {
+         fxMesa->TexSource[0].StartAddress = t0->range[0]->startAddr;
          fxMesa->TexSource[0].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[0].Info = &(ti0->info);
-      }
-      else {
-         if (ti0->LODblend && ti0->whichTMU == TDFX_TMU_SPLIT) {
-            fxMesa->TexSource[0].StartAddress = ti0->tm[TDFX_TMU0]->startAddr;
+         fxMesa->TexSource[0].Info = &(t0->info);
+      } else {
+         if ( t0->LODblend && t0->whichTMU == TDFX_TMU_SPLIT ) {
+            fxMesa->TexSource[0].StartAddress = t0->range[TDFX_TMU0]->startAddr;
             fxMesa->TexSource[0].EvenOdd = GR_MIPMAPLEVELMASK_ODD;
-            fxMesa->TexSource[0].Info = &(ti0->info);
-            fxMesa->TexSource[1].StartAddress = ti0->tm[TDFX_TMU1]->startAddr;
+            fxMesa->TexSource[0].Info = &t0->info;
+            fxMesa->TexSource[1].StartAddress = t0->range[TDFX_TMU1]->startAddr;
             fxMesa->TexSource[1].EvenOdd = GR_MIPMAPLEVELMASK_EVEN;
-            fxMesa->TexSource[1].Info = &(ti0->info);
-         }
-         else {
-            FxU32 tmu;
-            if (ti0->whichTMU == TDFX_TMU_BOTH)
-               tmu = TDFX_TMU0;
-            else
-               tmu = ti0->whichTMU;
+            fxMesa->TexSource[1].Info = &t0->info;
+         } else {
+            FxU32 unit;
+            if ( t0->whichTMU == TDFX_TMU_BOTH ) {
+               unit = TDFX_TMU0;
+            } else {
+               unit = t0->whichTMU;
+	    }
             fxMesa->TexSource[0].Info = NULL;
             fxMesa->TexSource[1].Info = NULL;
-            if (ti0->tm[tmu]) {
-               fxMesa->TexSource[tmu].StartAddress = ti0->tm[tmu]->startAddr;
-               fxMesa->TexSource[tmu].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-               fxMesa->TexSource[tmu].Info = &(ti0->info);
+            if ( t0->range[unit] ) {
+               fxMesa->TexSource[unit].StartAddress = t0->range[unit]->startAddr;
+               fxMesa->TexSource[unit].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
+               fxMesa->TexSource[unit].Info = &t0->info;
             }
          }
       }
-   }
-   else if (ctx->Texture.ReallyEnabled == TEXTURE1_2D) {
-      if (shared->umaTexMemory) {
-         fxMesa->TexSource[0].StartAddress = ti1->tm[0]->startAddr;
+      break;
+
+   case TEXTURE1_2D:
+      if ( tss->umaTexMemory ) {
+         fxMesa->TexSource[0].StartAddress = t1->range[0]->startAddr;
          fxMesa->TexSource[0].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[0].Info = &(ti1->info);
+         fxMesa->TexSource[0].Info = &t1->info;
       }
-   }
-   else if (ctx->Texture.ReallyEnabled == (TEXTURE0_2D | TEXTURE1_2D)) {
-      if (shared->umaTexMemory) {
+      break;
+
+   case TEXTURE0_2D | TEXTURE1_2D:
+      if ( tss->umaTexMemory ) {
          const FxU32 tmu0 = 0, tmu1 = 1;
-         fxMesa->TexSource[tmu0].StartAddress = ti0->tm[0]->startAddr;
+         fxMesa->TexSource[tmu0].StartAddress = t0->range[0]->startAddr;
          fxMesa->TexSource[tmu0].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[tmu0].Info = &(ti0->info);
+         fxMesa->TexSource[tmu0].Info = &t0->info;
 
-         fxMesa->TexSource[tmu1].StartAddress = ti1->tm[0]->startAddr;
+         fxMesa->TexSource[tmu1].StartAddress = t1->range[0]->startAddr;
          fxMesa->TexSource[tmu1].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[tmu1].Info = &(ti1->info);
+         fxMesa->TexSource[tmu1].Info = &t1->info;
       }
       else {
          const FxU32 tmu0 = 0, tmu1 = 1;
-         fxMesa->TexSource[tmu0].StartAddress = ti0->tm[tmu0]->startAddr;
+         fxMesa->TexSource[tmu0].StartAddress = t0->range[tmu0]->startAddr;
          fxMesa->TexSource[tmu0].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[tmu0].Info = &(ti0->info);
+         fxMesa->TexSource[tmu0].Info = &t0->info;
 
-         fxMesa->TexSource[tmu1].StartAddress = ti1->tm[tmu1]->startAddr;
+         fxMesa->TexSource[tmu1].StartAddress = t1->range[tmu1]->startAddr;
          fxMesa->TexSource[tmu1].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[tmu1].Info = &(ti1->info);
+         fxMesa->TexSource[tmu1].Info = &t1->info;
       }
+      break;
    }
 
-
    fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_SOURCE;
 }
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.h b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.h
deleted file mode 100644
index 0c0d4adcb..000000000
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.h
+++ /dev/null
@@ -1,44 +0,0 @@
-/* -*- mode: c; c-basic-offset: 3 -*-
- *
- * Copyright 2000 VA Linux Systems Inc., Fremont, California.
- *
- * 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
- * VA LINUX SYSTEMS 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.
- */
-/* $XFree86$ */
-
-/*
- * Original rewrite:
- *	Gareth Hughes <gareth@valinux.com>, 29 Sep - 1 Oct 2000
- *
- * Authors:
- *	Gareth Hughes <gareth@valinux.com>
- *	Brian Paul <brianp@valinux.com>
- *
- */
-
-#ifndef __TDFX_TEXSTATE_H__
-#define __TDFX_TEXSTATE_H__
-
-extern void tdfxUpdateTextureState( GLcontext *ctx );
-extern void tdfxUpdateTextureBinding( GLcontext *ctx );
-
-#endif
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_wrapper.h b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_wrapper.h
index 20e5530c5..0adb85b41 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_wrapper.h
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_wrapper.h
@@ -552,8 +552,6 @@ extern FxBool FX_grLfbLock(tdfxContextPtr fxMesa,
     UNLOCK_HARDWARE(fxMesa);						\
   } while (0)
 
-#define FX_grTexDownloadMipMapLevel_NoLock grTexDownloadMipMapLevel
-
 #define FX_grTexDownloadMipMapLevelPartial(fxMesa, t, sa, tlod, llod, ar, f, eo, d, s, e);	\
   do {									    \
     LOCK_HARDWARE(fxMesa);						    \
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_xmesa.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_xmesa.c
index a0b4304e4..00935e2de 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_xmesa.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_xmesa.c
@@ -52,19 +52,6 @@
 #include "tdfx_texman.h"
 
 
-#ifndef TDFX_DEBUG
-int TDFX_DEBUG = (0
-/*  		  | DEBUG_ALWAYS_SYNC */
-/*		  | DEBUG_VERBOSE_API */
-/*		  | DEBUG_VERBOSE_MSG */
-/*		  | DEBUG_VERBOSE_LRU */
-/*  		  | DEBUG_VERBOSE_DRI */
-/*  		  | DEBUG_VERBOSE_IOCTL */
-/*   		  | DEBUG_VERBOSE_2D */
-   );
-#endif
-
-
 GLboolean
 XMesaInitDriver( __DRIscreenPrivate *sPriv )
 {
@@ -393,11 +380,12 @@ XMesaMakeCurrent( __DRIcontextPrivate *driContextPriv,
 GLboolean
 XMesaOpenFullScreen(__DRIcontextPrivate *driContextPriv)
 {
-    fprintf(stderr,"***** XMesaOpenFullScreen *****\n");
+   if ( 0 )
+      fprintf( stderr, "***** XMesaOpenFullScreen *****\n" );
 #if 0 /* When new glide3 calls exist */
-    return((GLboolean)grDRISetupFullScreen(GL_TRUE));
+   return (GLboolean)grDRISetupFullScreen( GL_TRUE );
 #else
-    return GL_TRUE;
+   return GL_TRUE;
 #endif
 }
 
@@ -405,11 +393,12 @@ XMesaOpenFullScreen(__DRIcontextPrivate *driContextPriv)
 GLboolean
 XMesaCloseFullScreen(__DRIcontextPrivate *driContextPriv)
 {
-    fprintf(stderr,"***** XMesaCloseFullScreen *****\n");
+   if ( 0 )
+      fprintf( stderr, "***** XMesaCloseFullScreen *****\n" );
 #if 0 /* When new glide3 calls exist */
-    return((GLboolean)grDRISetupFullScreen(GL_FALSE));
+   return (GLboolean)grDRISetupFullScreen( GL_FALSE );
 #else
-    return GL_TRUE;
+   return GL_TRUE;
 #endif
 }
 
diff --git a/xc/programs/Xserver/GL/mesa/src/Imakefile b/xc/programs/Xserver/GL/mesa/src/Imakefile
index c2ead442c..05f2cfe1f 100644
--- a/xc/programs/Xserver/GL/mesa/src/Imakefile
+++ b/xc/programs/Xserver/GL/mesa/src/Imakefile
@@ -147,6 +147,8 @@ LinkSourceFile(state.c,$(MESASRCDIR)/src)
 LinkSourceFile(state.h,$(MESASRCDIR)/src)
 LinkSourceFile(stencil.c,$(MESASRCDIR)/src)
 LinkSourceFile(stencil.h,$(MESASRCDIR)/src)
+LinkSourceFile(texformat.c, $(MESASRCDIR)/src)
+LinkSourceFile(texformat.h, $(MESASRCDIR)/src)
 LinkSourceFile(texgen_tmp.h,$(MESASRCDIR)/src)
 LinkSourceFile(teximage.c,$(MESASRCDIR)/src)
 LinkSourceFile(teximage.h,$(MESASRCDIR)/src)
@@ -245,6 +247,7 @@ LinkSourceFile(zoom.h,$(MESASRCDIR)/src)
 		stages.c \
 		state.c \
 		stencil.c \
+		texformat.c \
 		teximage.c \
 		texobj.c \
 		texstate.c \
@@ -321,6 +324,7 @@ LinkSourceFile(zoom.h,$(MESASRCDIR)/src)
 		stages.o \
 		state.o \
 		stencil.o \
+		texformat.o \
 		teximage.o \
 		texobj.o \
 		texstate.o \
@@ -384,4 +388,3 @@ InstallDriverSDKLibraryModule(GLcore,$(DRIVERSDKMODULEDIR),extensions)
 DependSubdirs($(SUBDIRS))
 MakeLintLibSubdirs($(SUBDIRS))
 LintSubdirs($(SUBDIRS))
-
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/ati/r128_dri.h b/xc/programs/Xserver/hw/xfree86/drivers/ati/r128_dri.h
index 5651f0d59..01857a0e3 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/ati/r128_dri.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/ati/r128_dri.h
@@ -45,7 +45,7 @@
 /* DRI Driver defaults */
 #define R128_DEFAULT_CCE_PIO_MODE R128_PM4_64PIO_64VCBM_64INDBM
 #define R128_DEFAULT_CCE_BM_MODE  R128_PM4_64BM_64VCBM_64INDBM
-#define R128_DEFAULT_AGP_MODE     2
+#define R128_DEFAULT_AGP_MODE     1
 #define R128_DEFAULT_AGP_SIZE     8 /* MB (must be a power of 2 and > 4MB) */
 #define R128_DEFAULT_RING_SIZE    1 /* MB (must be page aligned) */
 #define R128_DEFAULT_BUFFER_SIZE  2 /* MB (must be page aligned) */
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_dri.h b/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_dri.h
index 11e8a3e09..bfbc07b40 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_dri.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_dri.h
@@ -43,7 +43,7 @@
 /* DRI Driver defaults */
 #define RADEON_DEFAULT_CP_PIO_MODE  RADEON_CSQ_PRIPIO_INDPIO
 #define RADEON_DEFAULT_CP_BM_MODE   RADEON_CSQ_PRIBM_INDBM
-#define RADEON_DEFAULT_AGP_MODE     2
+#define RADEON_DEFAULT_AGP_MODE     1
 #define RADEON_DEFAULT_AGP_SIZE     8 /* MB (must be a power of 2 and > 4MB) */
 #define RADEON_DEFAULT_RING_SIZE    1 /* MB (must be page aligned) */
 #define RADEON_DEFAULT_BUFFER_SIZE  2 /* MB (must be page aligned) */
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_dripriv.h b/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_dripriv.h
index 852027c17..ffbd869c5 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_dripriv.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_dripriv.h
@@ -52,8 +52,13 @@ typedef struct {
 } RADEONConfigPrivRec, *RADEONConfigPrivPtr;
 
 typedef struct {
+#ifdef PER_CONTEXT_SAREA
     drmContext ctx_id;
     drmHandle sarea_handle;
+#else
+    /* Nothing here yet */
+    int dummy;
+#endif
 } RADEONDRIContextRec, *RADEONDRIContextPtr;
 
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_reg.h b/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_reg.h
index c29db273b..20ad35cbc 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_reg.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_reg.h
@@ -1068,36 +1068,36 @@
 #define RADEON_PP_TXFORMAT_0                0x1c58
 #define RADEON_PP_TXFORMAT_1                0x1c70
 #define RADEON_PP_TXFORMAT_2                0x1c88
-#       define RADEON_TXF_8BPP_I             (0  <<  0)
-#       define RADEON_TXF_16BPP_AI88         (1  <<  0)
-#       define RADEON_TXF_8BPP_RGB332        (2  <<  0)
-#       define RADEON_TXF_16BPP_ARGB1555     (3  <<  0)
-#       define RADEON_TXF_16BPP_RGB565       (4  <<  0)
-#       define RADEON_TXF_16BPP_ARGB4444     (5  <<  0)
-#       define RADEON_TXF_32BPP_ARGB8888     (6  <<  0)
-#       define RADEON_TXF_32BPP_RGBA8888     (7  <<  0)
-#       define RADEON_TXF_8BPP_Y             (8  <<  0)
-#       define RADEON_TXF_FORMAT_MASK        (31 <<  0)
-#       define RADEON_TXF_FORMAT_SHIFT       0
-#       define RADEON_TXF_APPLE_YUV_MODE     (1  <<  5)
-#       define RADEON_TXF_ALPHA_IN_MAP       (1  <<  6)
-#       define RADEON_TXF_NON_POWER2         (1  <<  7)
-#       define RADEON_TXF_WIDTH_MASK         (15 <<  8)
-#       define RADEON_TXF_WIDTH_SHIFT        8
-#       define RADEON_TXF_HEIGHT_MASK        (15 << 12)
-#       define RADEON_TXF_HEIGHT_SHIFT       12
-#       define RADEON_TXF_ST_ROUTE_STQ0      (0  << 24)
-#       define RADEON_TXF_ST_ROUTE_MASK      (3  << 24)
-#       define RADEON_TXF_ST_ROUTE_STQ1      (1  << 24)
-#       define RADEON_TXF_ST_ROUTE_STQ2      (2  << 24)
-#       define RADEON_TXF_ENDIAN_NO_SWAP     (0  << 26)
-#       define RADEON_TXF_ENDIAN_16BPP_SWAP  (1  << 26)
-#       define RADEON_TXF_ENDIAN_32BPP_SWAP  (2  << 26)
-#       define RADEON_TXF_ENDIAN_HALFDW_SWAP (3  << 26)
-#       define RADEON_TXF_ALPHA_MASK_ENABLE  (1  << 28)
-#       define RADEON_TXF_CHROMA_KEY_ENABLE  (1  << 29)
-#       define RADEON_TXF_CUBIC_MAP_ENABLE   (1  << 30)
-#       define RADEON_TXF_PERSPECTIVE_ENABLE (1  << 31)
+#       define RADEON_TXFORMAT_I8                 (0  <<  0)
+#       define RADEON_TXFORMAT_AI88               (1  <<  0)
+#       define RADEON_TXFORMAT_RGB332             (2  <<  0)
+#       define RADEON_TXFORMAT_ARGB1555           (3  <<  0)
+#       define RADEON_TXFORMAT_RGB565             (4  <<  0)
+#       define RADEON_TXFORMAT_ARGB4444           (5  <<  0)
+#       define RADEON_TXFORMAT_ARGB8888           (6  <<  0)
+#       define RADEON_TXFORMAT_RGBA8888           (7  <<  0)
+#       define RADEON_TXFORMAT_Y8                 (8  <<  0)
+#       define RADEON_TXFORMAT_FORMAT_MASK        (31 <<  0)
+#       define RADEON_TXFORMAT_FORMAT_SHIFT       0
+#       define RADEON_TXFORMAT_APPLE_YUV_MODE     (1  <<  5)
+#       define RADEON_TXFORMAT_ALPHA_IN_MAP       (1  <<  6)
+#       define RADEON_TXFORMAT_NON_POWER2         (1  <<  7)
+#       define RADEON_TXFORMAT_WIDTH_MASK         (15 <<  8)
+#       define RADEON_TXFORMAT_WIDTH_SHIFT        8
+#       define RADEON_TXFORMAT_HEIGHT_MASK        (15 << 12)
+#       define RADEON_TXFORMAT_HEIGHT_SHIFT       12
+#       define RADEON_TXFORMAT_ST_ROUTE_STQ0      (0  << 24)
+#       define RADEON_TXFORMAT_ST_ROUTE_MASK      (3  << 24)
+#       define RADEON_TXFORMAT_ST_ROUTE_STQ1      (1  << 24)
+#       define RADEON_TXFORMAT_ST_ROUTE_STQ2      (2  << 24)
+#       define RADEON_TXFORMAT_ENDIAN_NO_SWAP     (0  << 26)
+#       define RADEON_TXFORMAT_ENDIAN_16BPP_SWAP  (1  << 26)
+#       define RADEON_TXFORMAT_ENDIAN_32BPP_SWAP  (2  << 26)
+#       define RADEON_TXFORMAT_ENDIAN_HALFDW_SWAP (3  << 26)
+#       define RADEON_TXFORMAT_ALPHA_MASK_ENABLE  (1  << 28)
+#       define RADEON_TXFORMAT_CHROMA_KEY_ENABLE  (1  << 29)
+#       define RADEON_TXFORMAT_CUBIC_MAP_ENABLE   (1  << 30)
+#       define RADEON_TXFORMAT_PERSPECTIVE_ENABLE (1  << 31)
 #define RADEON_PP_TXOFFSET_0                0x1c5c
 #define RADEON_PP_TXOFFSET_1                0x1c74
 #define RADEON_PP_TXOFFSET_2                0x1c8c
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_sarea.h b/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_sarea.h
index 14cb002fd..b80f35e52 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_sarea.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_sarea.h
@@ -81,7 +81,7 @@
 
 /* Vertex/indirect buffer size
  */
-#define RADEON_BUFFER_SIZE		16384
+#define RADEON_BUFFER_SIZE		65536
 
 /* Byte offsets for indirect buffer data
  */
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.c b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.c
index cb17e0533..3b34acf08 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.c
@@ -251,7 +251,7 @@ static Bool MGAInitVisualConfigs( ScreenPtr pScreen )
                pConfigs[i].redSize		= 8;
                pConfigs[i].greenSize		= 8;
                pConfigs[i].blueSize		= 8;
-               pConfigs[i].alphaSize		= 0;
+               pConfigs[i].alphaSize		= 8;
                pConfigs[i].redMask		= 0x00FF0000;
                pConfigs[i].greenMask		= 0x0000FF00;
                pConfigs[i].blueMask		= 0x000000FF;
@@ -736,6 +736,8 @@ static Bool MGADRIAgpInit(ScreenPtr pScreen)
 	       "[drm] Added %d %d byte DMA buffers\n",
 	       count, MGA_BUFFER_SIZE );
 
+   xf86EnableBusMaster( pMga->PciTag );
+
    return TRUE;
 }
 
@@ -845,8 +847,7 @@ static Bool MGADRIKernelInit( ScreenPtr pScreen )
    /* FIXME: This is just here to clean up after the engine reset test
     * in the kernel module.  Please remove it later...
     */
-   pMga->haveQuiescense = 0;
-   CHECK_DMA_QUIESCENT( pMGA, pScrn );
+   pMga->GetQuiescence( pScrn );
 #endif
 
    return TRUE;
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.h b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.h
index ab39b1701..56069492f 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.h
@@ -34,7 +34,7 @@
 #include <xf86drm.h>
 #include <xf86drmMga.h>
 
-#define MGA_DEFAULT_AGP_MODE     2
+#define MGA_DEFAULT_AGP_MODE     1
 #define MGA_MAX_AGP_MODE         4
 
 /* Buffer are aligned on 4096 byte boundaries.
@@ -61,7 +61,7 @@ typedef struct {
    unsigned int depthPitch;
 
    unsigned int textureOffset;
-   unsigned int textureSize;
+   int textureSize;
 
    drmRegion agp;
 
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/tdfx/tdfx_dri.c b/xc/programs/Xserver/hw/xfree86/drivers/tdfx/tdfx_dri.c
index c70cb35e5..5d160998d 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/tdfx/tdfx_dri.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/tdfx/tdfx_dri.c
@@ -16,20 +16,20 @@
 static char TDFXKernelDriverName[] = "tdfx";
 static char TDFXClientDriverName[] = "tdfx";
 
-static Bool TDFXCreateContext(ScreenPtr pScreen, VisualPtr visual, 
+static Bool TDFXCreateContext(ScreenPtr pScreen, VisualPtr visual,
 			      drmContext hwContext, void *pVisualConfigPriv,
 			      DRIContextType contextStore);
 static void TDFXDestroyContext(ScreenPtr pScreen, drmContext hwContext,
 			       DRIContextType contextStore);
-static void TDFXDRISwapContext(ScreenPtr pScreen, DRISyncType syncType, 
-			       DRIContextType readContextType, 
+static void TDFXDRISwapContext(ScreenPtr pScreen, DRISyncType syncType,
+			       DRIContextType readContextType,
 			       void *readContextStore,
-			       DRIContextType writeContextType, 
+			       DRIContextType writeContextType,
 			       void *writeContextStore);
 static Bool TDFXDRIOpenFullScreen(ScreenPtr pScreen);
 static Bool TDFXDRICloseFullScreen(ScreenPtr pScreen);
 static void TDFXDRIInitBuffers(WindowPtr pWin, RegionPtr prgn, CARD32 index);
-static void TDFXDRIMoveBuffers(WindowPtr pParent, DDXPointRec ptOldOrg, 
+static void TDFXDRIMoveBuffers(WindowPtr pParent, DDXPointRec ptOldOrg,
 			       RegionPtr prgnSrc, CARD32 index);
 static void TDFXDRITransitionTo2d(ScreenPtr pScreen);
 static void TDFXDRITransitionTo3d(ScreenPtr pScreen);
@@ -65,7 +65,7 @@ TDFXInitVisualConfigs(ScreenPtr pScreen)
       xfree(pTDFXConfigs);
       return FALSE;
     }
-    for (i=0; i<numConfigs; i++) 
+    for (i=0; i<numConfigs; i++)
       pTDFXConfigPtrs[i] = &pTDFXConfigs[i];
 
     i=0;
@@ -104,7 +104,7 @@ TDFXInitVisualConfigs(ScreenPtr pScreen)
 	    if (depth) {
 	      if (pTDFX->cpp>2)
 		pConfigs[i].depthSize = 24;
-	      else 
+	      else
 		pConfigs[i].depthSize = 16;
 	    } else {
 	      pConfigs[i].depthSize = 0;
@@ -158,7 +158,7 @@ TDFXInitVisualConfigs(ScreenPtr pScreen)
       return FALSE;
     }
 
-    for (i = 0; i < numConfigs; i++) 
+    for (i = 0; i < numConfigs; i++)
       pTDFXConfigPtrs[i] = &pTDFXConfigs[i];
 
     i=0;
@@ -199,7 +199,7 @@ TDFXInitVisualConfigs(ScreenPtr pScreen)
 	    if (depth) {
 	      if (pTDFX->cpp > 2)
 		pConfigs[i].depthSize = 24;
-	      else 
+	      else
 		pConfigs[i].depthSize = 16;
 	    } else {
 	      pConfigs[i].depthSize = 0;
@@ -249,7 +249,7 @@ TDFXDoWakeupHandler(int screenNum, pointer wakeupData, unsigned long result,
   TDFXNeedSync(pScrn);
 }
 
-static void 
+static void
 TDFXDoBlockHandler(int screenNum, pointer blockData, pointer pTimeout,
 		  pointer pReadmask)
 {
@@ -343,10 +343,10 @@ Bool TDFXDRIScreenInit(ScreenPtr pScreen)
     pDRIInfo->maxDrawableTableEntry = TDFX_MAX_DRAWABLES;
 
 #ifdef NOT_DONE
-  /* FIXME need to extend DRI protocol to pass this size back to client 
+  /* FIXME need to extend DRI protocol to pass this size back to client
    * for SAREA mapping that includes a device private record
    */
-  pDRIInfo->SAREASize = 
+  pDRIInfo->SAREASize =
     ((sizeof(XF86DRISAREARec) + 0xfff) & 0x1000); /* round to page */
   /* + shared memory device private rec */
 #else
@@ -415,7 +415,7 @@ Bool TDFXDRIScreenInit(ScreenPtr pScreen)
   }
 
   pTDFXDRI->regsSize=TDFXIOMAPSIZE;
-  if (drmAddMap(pTDFX->drmSubFD, (drmHandle)pTDFX->MMIOAddr[0], 
+  if (drmAddMap(pTDFX->drmSubFD, (drmHandle)pTDFX->MMIOAddr[0],
 		pTDFXDRI->regsSize, DRM_REGISTERS, 0, &pTDFXDRI->regs)<0) {
     TDFXDRICloseScreen(pScreen);
     xf86DrvMsg(pScreen->myNum, X_ERROR, "drmAddMap failed, disabling DRI.\n");
@@ -455,7 +455,7 @@ TDFXDRICloseScreen(ScreenPtr pScreen)
 }
 
 static Bool
-TDFXCreateContext(ScreenPtr pScreen, VisualPtr visual, 
+TDFXCreateContext(ScreenPtr pScreen, VisualPtr visual,
 		  drmContext hwContext, void *pVisualConfigPriv,
 		  DRIContextType contextStore)
 {
@@ -463,7 +463,7 @@ TDFXCreateContext(ScreenPtr pScreen, VisualPtr visual,
 }
 
 static void
-TDFXDestroyContext(ScreenPtr pScreen, drmContext hwContext, 
+TDFXDestroyContext(ScreenPtr pScreen, drmContext hwContext,
 		   DRIContextType contextStore)
 {
 }
@@ -496,7 +496,7 @@ TDFXDRIFinishScreenInit(ScreenPtr pScreen)
 }
 
 static void
-TDFXDRISwapContext(ScreenPtr pScreen, DRISyncType syncType, 
+TDFXDRISwapContext(ScreenPtr pScreen, DRISyncType syncType,
 		   DRIContextType oldContextType, void *oldContext,
 		   DRIContextType newContextType, void *newContext)
 {
@@ -518,10 +518,10 @@ TDFXDRIInitBuffers(WindowPtr pWin, RegionPtr prgn, CARD32 index)
   TDFXSetupForSolidFill(pScrn, 0, GXcopy, -1);
   while (nbox--) {
     TDFXSelectBuffer(pTDFX, TDFX_BACK);
-    TDFXSubsequentSolidFillRect(pScrn, pbox->x1, pbox->y1, 
+    TDFXSubsequentSolidFillRect(pScrn, pbox->x1, pbox->y1,
 				pbox->x2-pbox->x1, pbox->y2-pbox->y1);
     TDFXSelectBuffer(pTDFX, TDFX_DEPTH);
-    TDFXSubsequentSolidFillRect(pScrn, pbox->x1, pbox->y1, 
+    TDFXSubsequentSolidFillRect(pScrn, pbox->x1, pbox->y1,
 				pbox->x2-pbox->x1, pbox->y2-pbox->y1);
     pbox++;
   }
@@ -537,7 +537,7 @@ TDFXDRIInitBuffers(WindowPtr pWin, RegionPtr prgn, CARD32 index)
   are reversed.
 */
 static void
-TDFXDRIMoveBuffers(WindowPtr pParent, DDXPointRec ptOldOrg, 
+TDFXDRIMoveBuffers(WindowPtr pParent, DDXPointRec ptOldOrg,
 		   RegionPtr prgnSrc, CARD32 index)
 {
   ScreenPtr pScreen = pParent->drawable.pScreen;
@@ -684,11 +684,11 @@ TDFXDRIMoveBuffers(WindowPtr pParent, DDXPointRec ptOldOrg,
 static Bool
 TDFXDRIOpenFullScreen(ScreenPtr pScreen)
 {
+#if 0
   ScrnInfoPtr pScrn;
   TDFXPtr pTDFX;
 
   xf86DrvMsg(pScreen->myNum, X_INFO, "OpenFullScreen\n");
-#if 0
   pScrn = xf86Screens[pScreen->myNum];
   pTDFX=TDFXPTR(pScrn);
   if (pTDFX->numChips>1) {
@@ -701,10 +701,10 @@ TDFXDRIOpenFullScreen(ScreenPtr pScreen)
 static Bool
 TDFXDRICloseFullScreen(ScreenPtr pScreen)
 {
+#if 0
   ScrnInfoPtr pScrn;
 
   xf86DrvMsg(pScreen->myNum, X_INFO, "CloseFullScreen\n");
-#if 0
   pScrn = xf86Screens[pScreen->myNum];
   TDFXDisableSLI(pScrn);
 #endif
@@ -718,7 +718,7 @@ TDFXDRITransitionTo2d(ScreenPtr pScreen)
   ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
   TDFXPtr pTDFX = TDFXPTR(pScrn);
 
-  xf86FreeOffscreenArea(pTDFX->reservedArea); 
+  xf86FreeOffscreenArea(pTDFX->reservedArea);
 }
 
 
@@ -740,7 +740,7 @@ TDFXDRITransitionTo3d(ScreenPtr pScreen)
   }
 
   xf86PurgeUnlockedOffscreenAreas(pScreen);
-  
+
   /* this is an ugly hack! isnt there a resize? */
   pArea = xf86AllocateOffscreenArea(pScreen, pScrn->displayWidth,
 				    pTDFX->pixmapCacheLinesMin,
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/Makefile.linux b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/Makefile.linux
index 26eed9a5f..d022557d8 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/Makefile.linux
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/Makefile.linux
@@ -47,38 +47,34 @@
 
 # **** End of SMP/MODVERSIONS detection
 
-#MODS=           gamma.o tdfx.o
-MODS=           tdfx.o
-LIBS=
+MODS =		gamma.o tdfx.o
+LIBS =
 
-#DRMOBJS=	init.o memory.o proc.o auth.o context.o drawable.o bufs.o \
-#		lists.o lock.o ioctl.o fops.o vm.o dma.o ctxbitmap.o
-DRMOBS=
-DRMTEMPLATES=	drm_init.h drm_memory.h drm_proc.h drm_auth.h drm_context.h \
-		drm_drawable.h drm_bufs.h drm_lists.h drm_lock.h drm_ioctl.h \
-		drm_fops.h drm_vm.h drm_dma.h drm_stub.h
-DRMHEADERS=	drm.h drmP.h compat-pre24.h
+DRMTEMPLATES =	drm_auth.h drm_bufs.h drm_context.h drm_dma.h drm_drawable.h \
+		drm_drv.h drm_fops.h drm_init.h drm_ioctl.h drm_lists.h \
+		drm_lock.h drm_memory.h drm_proc.h drm_stub.h drm_vm.h
+DRMHEADERS =	drm.h drmP.h compat-pre24.h
 
-#GAMMAOBJS=      gamma_drv.o gamma_drm.o gamma_dma.o gamma_context.o
-#GAMMAHEADERS=   gamma_drv.h $(DRMHEADERS) $(DRMTEMPLATES)
+GAMMAOBJS =	gamma_drv.o gamma_dma.o
+GAMMAHEADERS =	gamma_drv.h $(DRMHEADERS) $(DRMTEMPLATES)
 
-TDFXOBJS=       tdfx_drv.o
-TDFXHEADERS=    tdfx.h $(DRMHEADERS)
+TDFXOBJS =	tdfx_drv.o
+TDFXHEADERS =	tdfx.h $(DRMHEADERS)
 
-INC=		/usr/include
+INC =		/usr/include
 
-CFLAGS=		-O2 $(WARNINGS)
-WARNINGS=	-Wall -Wwrite-strings -Wpointer-arith -Wcast-align \
+CFLAGS =	-O2 $(WARNINGS)
+WARNINGS =	-Wall -Wwrite-strings -Wpointer-arith -Wcast-align \
 		-Wstrict-prototypes -Wnested-externs \
 		-Wpointer-arith
 # -Wshadow -Winline -- make output too noisy
-MODCFLAGS=      $(CFLAGS) -D__KERNEL__ -DMODULE -fomit-frame-pointer
-PRGCFLAGS=      $(CFLAGS) -g -ansi -pedantic -DPOSIX_C_SOURCE=199309L \
+MODCFLAGS =	$(CFLAGS) -D__KERNEL__ -DMODULE -fomit-frame-pointer
+PRGCFLAGS =	$(CFLAGS) -g -ansi -pedantic -DPOSIX_C_SOURCE=199309L \
 		-D_POSIX_SOURCE -D_XOPEN_SOURCE -D_BSD_SOURCE -D_SVID_SOURCE \
 		-I../../../../../../include -I../../../../../../../../include \
 		-I../../../../../../../../programs/Xserver/hw/xfree86/common \
 		-I. -I../../.. -I../../../../../../../../lib/X11
-PRGLIBS=
+PRGLIBS =
 
 # **** Start of SMP/MODVERSIONS detection
 
@@ -134,7 +130,8 @@ endif
 
 ifeq ($(AGP),1)
 MODCFLAGS += -DCONFIG_AGP -DCONFIG_AGP_MODULE
-#DRMOBJS += agpsupport.o
+DRMTEMPLATES += drm_agpsupport.h
+DRMHEADERS += agpsupport-pre24.h
 MODS += mga.o r128.o radeon.o
 ifeq ($(MACHINE),i386)
 MODS += i810.o
@@ -144,17 +141,17 @@ MODS += i810.o
 endif
 
 
-MGAOBJS=	mga_drv.o mga_dma.o mga_state.o mga_warp.o
-MGAHEADERS=	mga.h mga_drv.h $(DRMHEADERS) $(DRMTEMPLATES)
+MGAOBJS =	mga_drv.o mga_dma.o mga_state.o mga_warp.o
+MGAHEADERS =	mga.h mga_drv.h mga_drm.h $(DRMHEADERS) $(DRMTEMPLATES)
 
-I810OBJS=	i810_drv.o i810_dma.o
-I810HEADERS=	i810.h i810_drv.h $(DRMHEADERS) $(DRMTEMPLATES)
+I810OBJS =	i810_drv.o i810_dma.o
+I810HEADERS =	i810.h i810_drv.h i810_drm.h $(DRMHEADERS) $(DRMTEMPLATES)
 
-R128OBJS=	r128_drv.o r128_cce.o r128_state.o
-R128HEADERS=	r128.h r128_drv.h r128_drm.h $(DRMHEADERS) $(DRMTEMPLATES)
+R128OBJS =	r128_drv.o r128_cce.o r128_state.o
+R128HEADERS =	r128.h r128_drv.h r128_drm.h $(DRMHEADERS) $(DRMTEMPLATES)
 
-RADEONOBJS=	radeon_drv.o radeon_cp.o radeon_state.o
-RADEONHEADERS=	radeon.h radeon_drv.h radeon_drm.h $(DRMHEADERS) \
+RADEONOBJS =	radeon_drv.o radeon_cp.o radeon_state.o
+RADEONHEADERS =	radeon.h radeon_drv.h radeon_drm.h $(DRMHEADERS) \
 		$(DRMTEMPLATES)
 endif
 
@@ -177,7 +174,7 @@ all::;@echo === SMP=${SMP} MODULES=${MODULES} MODVERSIONS=${MODVERSIONS} AGP=${A
 all::;@echo === kill_fasync has $(PARAMS) parameters
 all::;@echo === Compiling for machine $(MACHINE)
 all::;@echo === WARNING
-all::;@echo === WARNING 2.4.0 kernels before test11 DONT WORK
+all::;@echo === WARNING 2.4.0 kernels before 2.4.0-test11 DO NOT WORK
 all::;@echo === WARNING
 
 ifeq ($(MODULES),0)
@@ -205,17 +202,13 @@ endif
 
 # **** End of configuration
 
-libdrm.a: $(DRMOBJS)
-	-$(RM) -f $@
-	$(AR) rcs $@ $(DRMOBJS)
-
 dristat: dristat.c
 	$(CC) $(PRGCFLAGS) $< -o $@
 
-#gamma_drv.o: gamma_drv.c
-#	$(CC) $(MODCFLAGS) -DEXPORT_SYMTAB -I$(TREE) -c $< -o $@
-#gamma.o: $(GAMMAOBJS)
-#	$(LD) -r $^ -o $@
+gamma_drv.o: gamma_drv.c
+	$(CC) $(MODCFLAGS) -DEXPORT_SYMTAB -I$(TREE) -c $< -o $@
+gamma.o: $(GAMMAOBJS)
+	$(LD) -r $^ -o $@
 
 tdfx_drv.o: tdfx_drv.c
 	$(CC) $(MODCFLAGS) -DEXPORT_SYMTAB -I$(TREE) -c $< -o $@
@@ -260,8 +253,7 @@ ChangeLog:
 %.o: %.c
 	$(CC) $(MODCFLAGS) -I$(TREE) -c $< -o $@
 
-$(DRMOBJS): $(DRMHEADERS)
-#$(GAMMAOBJS): $(GAMMAHEADERS)
+$(GAMMAOBJS): $(GAMMAHEADERS)
 $(TDFXOBJS): $(TDFXHEADERS)
 ifeq ($(AGP),1)
 $(MGAOBJS): $(MGAHEADERS)
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/compat-pre24.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/compat-pre24.h
index f6dae008d..0ad96b12a 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/compat-pre24.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/compat-pre24.h
@@ -45,6 +45,7 @@
 				/* This is a hack that only works for
                                    this code base -- because we always
                                    call this with dev->tq.* */
+#undef INIT_LIST_HEAD
 #define INIT_LIST_HEAD(pointer) dev->tq.next = NULL
 
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/dristat.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/dristat.c
index f429efd3f..47193ab46 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/dristat.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/dristat.c
@@ -35,28 +35,11 @@
 #include "../xf86drmHash.c"
 #include "../xf86drm.c"
 
-#define DRM_DIR_NAME "/dev/dri"
-#define DRM_DEV_NAME "%s/card%d"
-
 #define DRM_VERSION 0x00000001
 #define DRM_MEMORY  0x00000002
 #define DRM_CLIENTS 0x00000004
 #define DRM_STATS   0x00000008
-
-typedef struct drmStatsS {
-    unsigned long count;
-    struct {
-	unsigned long value;
-	const char    *long_format;
-	const char    *long_name;
-	const char    *rate_format;
-	const char    *rate_name;
-	int           isvalue;
-	const char    *mult_names;
-	int           mult;
-	int           verbose;
-    } data[15];
-} drmStatsT;
+#define DRM_BUSID   0x00000010
 
 static void getversion(int fd)
 {
@@ -78,6 +61,15 @@ static void getversion(int fd)
     }
 }
 
+static void getbusid(int fd)
+{
+    const char *busid = drmGetBusid(fd);
+    
+    printf("  Busid: %s\n", *busid ? busid : "(not set)");
+    drmFreeBusid(busid);
+}
+
+#if 0
 typedef struct {
     unsigned long	offset;	 /* Requested physical address (0 for SAREA)*/
     unsigned long	size;	 /* Requested physical size (bytes)	    */
@@ -88,160 +80,7 @@ typedef struct {
     int		mtrr;	 /* MTRR slot used			    */
 				 /* Private data			    */
 } drmVmRec, *drmVmPtr;
-
-int drmGetMap(int fd, int idx, drmHandle *offset, drmSize *size,
-	      drmMapType *type, drmMapFlags *flags, drmHandle *handle,
-	      int *mtrr)
-{
-    drm_map_t map;
-
-    map.offset = idx;
-    if (ioctl(fd, DRM_IOCTL_GET_MAP, &map)) return -errno;
-    *offset = map.offset;
-    *size   = map.size;
-    *type   = map.type;
-    *flags  = map.flags;
-    *handle = (unsigned long)map.handle;
-    *mtrr   = map.mtrr;
-    return 0;
-}
-
-int drmGetClient(int fd, int idx, int *auth, int *pid, int *uid,
-		 unsigned long *magic, unsigned long *iocs)
-{
-    drm_client_t client;
-
-    client.idx = idx;
-    if (ioctl(fd, DRM_IOCTL_GET_CLIENT, &client)) return -errno;
-    *auth      = client.auth;
-    *pid       = client.pid;
-    *uid       = client.uid;
-    *magic     = client.magic;
-    *iocs      = client.iocs;
-    return 0;
-}
-
-int drmGetStats(int fd, drmStatsT *stats)
-{
-    drm_stats_t s;
-    int         i;
-
-    if (ioctl(fd, DRM_IOCTL_GET_STATS, &s)) return -errno;
-
-    stats->count = 0;
-    memset(stats, 0, sizeof(*stats));
-    if (s.count > sizeof(stats->data)/sizeof(stats->data[0]))
-	return -1;
-
-#define SET_VALUE                              \
-    stats->data[i].long_format = "%-20.20s";   \
-    stats->data[i].rate_format = "%8.8s";      \
-    stats->data[i].isvalue     = 1;            \
-    stats->data[i].verbose     = 0
-
-#define SET_COUNT                              \
-    stats->data[i].long_format = "%-20.20s";     \
-    stats->data[i].rate_format = "%5.5s";      \
-    stats->data[i].isvalue     = 0;            \
-    stats->data[i].mult_names  = "kgm";        \
-    stats->data[i].mult        = 1000;         \
-    stats->data[i].verbose     = 0
-
-#define SET_BYTE                             \
-    stats->data[i].long_format = "%-9.9s";   \
-    stats->data[i].rate_format = "%5.5s";    \
-    stats->data[i].isvalue     = 0;          \
-    stats->data[i].mult_names  = "KGM";      \
-    stats->data[i].mult        = 1024;       \
-    stats->data[i].verbose     = 0
-
-
-    stats->count = s.count;
-    for (i = 0; i < s.count; i++) {
-	stats->data[i].value = s.data[i].value;
-	switch (s.data[i].type) {
-	case _DRM_STAT_LOCK:
-	    stats->data[i].long_name = "Lock";
-	    stats->data[i].rate_name = "Lock";
-	    SET_VALUE;
-	    break;
-	case _DRM_STAT_OPENS:
-	    stats->data[i].long_name = "Opens";
-	    stats->data[i].rate_name = "O";
-	    SET_COUNT;
-	    stats->data[i].verbose   = 1;
-	    break;
-	case _DRM_STAT_CLOSES:
-	    stats->data[i].long_name = "Closes";
-	    stats->data[i].rate_name = "Lock";
-	    SET_COUNT;
-	    stats->data[i].verbose   = 1;
-	    break;
-	case _DRM_STAT_IOCTLS:
-	    stats->data[i].long_name = "Ioctls";
-	    stats->data[i].rate_name = "Ioc/s";
-	    SET_COUNT;
-	    break;
-	case _DRM_STAT_LOCKS:
-	    stats->data[i].long_name = "Locks";
-	    stats->data[i].rate_name = "Lck/s";
-	    SET_COUNT;
-	    break;
-	case _DRM_STAT_UNLOCKS:
-	    stats->data[i].long_name = "Unlocks";
-	    stats->data[i].rate_name = "Unl/s";
-	    SET_COUNT;
-	    break;
-	case _DRM_STAT_IRQ:
-	    stats->data[i].long_name = "IRQs";
-	    stats->data[i].rate_name = "IRQ/s";
-	    SET_COUNT;
-	    break;
-	case _DRM_STAT_PRIMARY:
-	    stats->data[i].long_name = "Primary Bytes";
-	    stats->data[i].rate_name = "PB/s";
-	    SET_BYTE;
-	    break;
-	case _DRM_STAT_SECONDARY:
-	    stats->data[i].long_name = "Secondary Bytes";
-	    stats->data[i].rate_name = "SB/s";
-	    SET_BYTE;
-	    break;
-	case _DRM_STAT_DMA:
-	    stats->data[i].long_name = "DMA";
-	    stats->data[i].rate_name = "DMA/s";
-	    SET_COUNT;
-	    break;
-	case _DRM_STAT_SPECIAL:
-	    stats->data[i].long_name = "Special DMA";
-	    stats->data[i].rate_name = "dma/s";
-	    SET_COUNT;
-	    break;
-	case _DRM_STAT_MISSED:
-	    stats->data[i].long_name = "Miss";
-	    stats->data[i].rate_name = "Ms/s";
-	    SET_COUNT;
-	    break;
-	case _DRM_STAT_VALUE:
-	    stats->data[i].long_name = "Value";
-	    stats->data[i].rate_name = "Value";
-	    SET_VALUE;
-	    break;
-	case _DRM_STAT_BYTE:
-	    stats->data[i].long_name = "Bytes";
-	    stats->data[i].rate_name = "B/s";
-	    SET_BYTE;
-	    break;
-	case _DRM_STAT_COUNT:
-	default:
-	    stats->data[i].long_name = "Count";
-	    stats->data[i].rate_name = "Cnt/s";
-	    SET_COUNT;
-	    break;
-	}
-    }
-    return 0;
-}
+#endif
 
 static void getvm(int fd)
 {
@@ -255,7 +94,7 @@ static void getvm(int fd)
     drmHandle       handle;
     int             mtrr;
 
-    printf("  VM map information:\n");
+    printf("  VM map information (Restricted locked kernel WC Lock):\n");
     printf("    slot     offset       size type flags    address mtrr\n");
 
     for (i = 0;
@@ -298,7 +137,7 @@ static void getclients(int fd)
     int           procfd;
     
     printf("  DRI client information:\n");
-    printf("    a   pid    uid      magic     ioctls  prog\n");
+    printf("    a   pid   uid      magic     ioctls   prog\n");
 
     for (i = 0; !drmGetClient(fd, i, &auth, &pid, &uid, &magic, &iocs); i++) {
 	sprintf(buf, "/proc/%d/cmdline", pid);
@@ -307,12 +146,16 @@ static void getclients(int fd)
 	    read(procfd, cmd, sizeof(cmd)-1);
 	    close(procfd);
 	}
-	if (*cmd)
+	if (*cmd) {
+	    char *pt;
+
+	    for (pt = cmd; *pt; pt++) if (!isprint(*pt)) *pt = ' ';
 	    printf("    %c %5d %5d %10lu %10lu   %s\n",
 		   auth ? 'y' : 'n', pid, uid, magic, iocs, cmd);
-	else
+	} else {
 	    printf("    %c %5d %5d %10lu %10lu\n",
 		   auth ? 'y' : 'n', pid, uid, magic, iocs);
+	}
     }
 }
 
@@ -394,44 +237,6 @@ static void getstats(int fd, int i)
     
 }
 
-static int drmOpenMinor(int minor, uid_t user, gid_t group,
-			mode_t dirmode, mode_t devmode, int force)
-{
-    struct stat st;
-    char        buf[64];
-    long        dev    = makedev(DRM_MAJOR, minor);
-    int         setdir = 0;
-    int         setdev = 0;
-    int         fd;
-
-    if (stat(DRM_DIR_NAME, &st) || !S_ISDIR(st.st_mode)) {
-	remove(DRM_DIR_NAME);
-	mkdir(DRM_DIR_NAME, dirmode);
-	++setdir;
-    }
-
-    if (force || setdir) {
-	chown(DRM_DIR_NAME, user, group);
-	chmod(DRM_DIR_NAME, dirmode);
-    }
-
-    sprintf(buf, DRM_DEV_NAME, DRM_DIR_NAME, minor);
-    if (stat(buf, &st) || st.st_rdev != dev) {
-	remove(buf);
-	mknod(buf, S_IFCHR, dev);
-	++setdev;
-    }
-
-    if (force || setdev) {
-	chown(buf, user, group);
-	chmod(buf, devmode);
-    }
-
-    if ((fd = open(buf, O_RDWR, 0)) >= 0) return fd;
-    if (setdev) remove(buf);
-    return -errno;
-}
-	
 int main(int argc, char **argv)
 {
     int  c;
@@ -442,13 +247,14 @@ int main(int argc, char **argv)
     char buf[64];
     int  i;
 
-    while ((c = getopt(argc, argv, "avmcsM:i:")) != EOF)
+    while ((c = getopt(argc, argv, "avmcsbM:i:")) != EOF)
 	switch (c) {
 	case 'a': mask = ~0;                          break;
 	case 'v': mask |= DRM_VERSION;                break;
 	case 'm': mask |= DRM_MEMORY;                 break;
 	case 'c': mask |= DRM_CLIENTS;                break;
 	case 's': mask |= DRM_STATS;                  break;
+	case 'b': mask |= DRM_BUSID;                  break;
 	case 'i': interval = strtol(optarg, NULL, 0); break;
 	case 'M': minor = strtol(optarg, NULL, 0);    break;
 	default:
@@ -458,9 +264,10 @@ int main(int argc, char **argv)
 
     for (i = 0; i < 16; i++) if (!minor || i == minor) {
 	sprintf(buf, DRM_DEV_NAME, DRM_DIR_NAME, i);
-	fd = drmOpenMinor(i, 0, 0, 0700, 0600, 0);
+	fd = drmOpenMinor(i, 1);
 	if (fd >= 0) {
 	    printf("%s\n", buf);
+	    if (mask & DRM_BUSID)   getbusid(fd);
 	    if (mask & DRM_VERSION) getversion(fd);
 	    if (mask & DRM_MEMORY)  getvm(fd);
 	    if (mask & DRM_CLIENTS) getclients(fd);
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm.h
index 7c65f5fc3..ddabc2771 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm.h
@@ -44,11 +44,8 @@
 #define DRM_IOCTL_NR(n)	     ((n) & 0xff)
 #endif
 
-#define DRM_DEV_MODE	 (S_IRUSR|S_IWUSR|S_IRGRP|S_IWGRP)
-#define DRM_DEV_UID	 0
-#define DRM_DEV_GID	 0
-
 #define DRM_MAJOR       226
+#define DRM_MAX_MINOR   15
 #define DRM_NAME	"drm"	  /* Name in kernel, /dev, and /proc	    */
 #define DRM_MIN_ORDER	5	  /* At least 2^5 bytes = 32 bytes	    */
 #define DRM_MAX_ORDER	22	  /* Up to 2^22 bytes = 4MB		    */
@@ -453,7 +450,7 @@ typedef struct drm_agp_info {
 #define DRM_IOCTL_RADEON_CLEAR		DRM_IOW( 0x48, drm_radeon_clear_t)
 #define DRM_IOCTL_RADEON_VERTEX		DRM_IOW( 0x49, drm_radeon_vertex_t)
 #define DRM_IOCTL_RADEON_INDICES	DRM_IOW( 0x4a, drm_radeon_indices_t)
-#define DRM_IOCTL_RADEON_BLIT		DRM_IOW( 0x4b, drm_radeon_blit_t)
+#define DRM_IOCTL_RADEON_TEXTURE	DRM_IOWR(0x4b, drm_radeon_texture_t)
 #define DRM_IOCTL_RADEON_STIPPLE	DRM_IOW( 0x4c, drm_radeon_stipple_t)
 #define DRM_IOCTL_RADEON_INDIRECT	DRM_IOWR(0x4d, drm_radeon_indirect_t)
 
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmP.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmP.h
index 804df4095..a7e4d5732 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmP.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmP.h
@@ -84,15 +84,24 @@
 #ifndef __HAVE_MTRR
 #define __HAVE_MTRR		0
 #endif
+#ifndef __HAVE_CTX_BITMAP
+#define __HAVE_CTX_BITMAP	0
+#endif
 #ifndef __HAVE_DMA
 #define __HAVE_DMA		0
 #endif
+#ifndef __HAVE_DMA_IRQ
+#define __HAVE_DMA_IRQ		0
+#endif
 #ifndef __HAVE_DMA_WAITLIST
 #define __HAVE_DMA_WAITLIST	0
 #endif
 #ifndef __HAVE_DMA_FREELIST
 #define __HAVE_DMA_FREELIST	0
 #endif
+#ifndef __HAVE_DMA_HISTOGRAM
+#define __HAVE_DMA_HISTOGRAM	0
+#endif
 
 #define __REALLY_HAVE_AGP	(__HAVE_AGP && (defined(CONFIG_AGP) || \
 						defined(CONFIG_AGP_MODULE)))
@@ -104,7 +113,6 @@
 
 #define DRM_DEBUG_CODE 2	  /* Include debugging code (if > 1, then
 				     also include looping detection. */
-#define DRM_DMA_HISTOGRAM 1	  /* Make histogram of DMA latency. */
 
 #define DRM_HASH_SIZE	      16 /* Size of key hash table		  */
 #define DRM_KERNEL_CONTEXT    0	 /* Change drm_resctx if changed	  */
@@ -643,7 +651,7 @@ typedef struct drm_device {
 	struct tq_struct  tq;
 	cycles_t	  ctx_start;
 	cycles_t	  lck_start;
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
 	drm_histogram_t	  histo;
 #endif
 
@@ -667,9 +675,11 @@ typedef struct drm_device {
 } drm_device_t;
 
 
-				/* Internal function definitions */
+/* ================================================================
+ * Internal function definitions
+ */
 
-				/* Misc. support (init.c) */
+				/* Misc. support (drm_init.h) */
 extern int	     DRM(flags);
 extern void	     DRM(parse_options)( char *s );
 extern int           DRM(cpu_valid)( void );
@@ -686,7 +696,7 @@ extern int           DRM(lock)(struct inode *inode, struct file *filp,
 extern int           DRM(unlock)(struct inode *inode, struct file *filp,
 				 unsigned int cmd, unsigned long arg);
 
-				/* Device support (fops.c) */
+				/* Device support (drm_fops.h) */
 extern int	     DRM(open_helper)(struct inode *inode, struct file *filp,
 				      drm_device_t *dev);
 extern int	     DRM(flush)(struct file *filp);
@@ -698,7 +708,7 @@ extern int	     DRM(write_string)(drm_device_t *dev, const char *s);
 extern unsigned int  DRM(poll)(struct file *filp,
 			       struct poll_table_struct *wait);
 
-				/* Mapping support (vm.c) */
+				/* Mapping support (drm_vm.h) */
 #if LINUX_VERSION_CODE < 0x020317
 extern unsigned long DRM(vm_nopage)(struct vm_area_struct *vma,
 				    unsigned long address,
@@ -728,17 +738,7 @@ extern int	     DRM(mmap_dma)(struct file *filp,
 				   struct vm_area_struct *vma);
 extern int	     DRM(mmap)(struct file *filp, struct vm_area_struct *vma);
 
-
-				/* Proc support (proc.c) */
-extern struct proc_dir_entry *drm_proc_init(drm_device_t *dev,
-					    int minor,
-					    struct proc_dir_entry *root,
-					    struct proc_dir_entry **dev_root);
-extern int          drm_proc_cleanup(int minor,
-                                     struct proc_dir_entry *root,
-                                     struct proc_dir_entry *dev_root);
-
-				/* Memory management support (memory.c) */
+				/* Memory management support (drm_memory.h) */
 extern void	     DRM(mem_init)(void);
 extern int	     DRM(mem_info)(char *buf, char **start, off_t offset,
 				   int request, int *eof, void *data);
@@ -761,7 +761,7 @@ extern int           DRM(bind_agp)(agp_memory *handle, unsigned int start);
 extern int           DRM(unbind_agp)(agp_memory *handle);
 #endif
 
-				/* Misc. IOCTL support (ioctl.c) */
+				/* Misc. IOCTL support (drm_ioctl.h) */
 extern int	     DRM(irq_busid)(struct inode *inode, struct file *filp,
 				    unsigned int cmd, unsigned long arg);
 extern int	     DRM(getunique)(struct inode *inode, struct file *filp,
@@ -775,7 +775,7 @@ extern int	     DRM(getclient)(struct inode *inode, struct file *filp,
 extern int	     DRM(getstats)(struct inode *inode, struct file *filp,
 				   unsigned int cmd, unsigned long arg);
 
-				/* Context IOCTL support (context.c) */
+				/* Context IOCTL support (drm_context.h) */
 extern int	     DRM(resctx)( struct inode *inode, struct file *filp,
 				  unsigned int cmd, unsigned long arg );
 extern int	     DRM(addctx)( struct inode *inode, struct file *filp,
@@ -794,14 +794,24 @@ extern int	     DRM(rmctx)( struct inode *inode, struct file *filp,
 extern int	     DRM(context_switch)(drm_device_t *dev, int old, int new);
 extern int	     DRM(context_switch_complete)(drm_device_t *dev, int new);
 
-				/* Drawable IOCTL support (drawable.c) */
+#if __HAVE_CTX_BITMAP
+extern int	     DRM(ctxbitmap_init)( drm_device_t *dev );
+extern void	     DRM(ctxbitmap_cleanup)( drm_device_t *dev );
+#endif
+
+extern int	     DRM(setsareactx)( struct inode *inode, struct file *filp,
+				       unsigned int cmd, unsigned long arg );
+extern int	     DRM(getsareactx)( struct inode *inode, struct file *filp,
+				       unsigned int cmd, unsigned long arg );
+
+				/* Drawable IOCTL support (drm_drawable.h) */
 extern int	     DRM(adddraw)(struct inode *inode, struct file *filp,
 				  unsigned int cmd, unsigned long arg);
 extern int	     DRM(rmdraw)(struct inode *inode, struct file *filp,
 				 unsigned int cmd, unsigned long arg);
 
 
-				/* Authentication IOCTL support (auth.c) */
+				/* Authentication IOCTL support (drm_auth.h) */
 extern int	     DRM(add_magic)(drm_device_t *dev, drm_file_t *priv,
 				    drm_magic_t magic);
 extern int	     DRM(remove_magic)(drm_device_t *dev, drm_magic_t magic);
@@ -811,7 +821,7 @@ extern int	     DRM(authmagic)(struct inode *inode, struct file *filp,
 				    unsigned int cmd, unsigned long arg);
 
 
-				/* Locking IOCTL support (lock.c) */
+				/* Locking IOCTL support (drm_lock.h) */
 extern int	     DRM(block)(struct inode *inode, struct file *filp,
 				unsigned int cmd, unsigned long arg);
 extern int	     DRM(unblock)(struct inode *inode, struct file *filp,
@@ -832,20 +842,13 @@ extern int	     DRM(flush_block_and_flush)(drm_device_t *dev, int context,
 						drm_lock_flags_t flags);
 extern int           DRM(notifier)(void *priv);
 
-				/* Context Bitmap support (ctxbitmap.c) */
-extern int	     DRM(ctxbitmap_init)( drm_device_t *dev );
-extern void	     DRM(ctxbitmap_cleanup)( drm_device_t *dev );
-extern int	     DRM(setsareactx)( struct inode *inode, struct file *filp,
-				       unsigned int cmd, unsigned long arg );
-extern int	     DRM(getsareactx)( struct inode *inode, struct file *filp,
-				       unsigned int cmd, unsigned long arg );
-
-				/* Buffer management support (bufs.c) */
+				/* Buffer management support (drm_bufs.h) */
 extern int	     DRM(order)( unsigned long size );
 extern int	     DRM(addmap)( struct inode *inode, struct file *filp,
 				  unsigned int cmd, unsigned long arg );
 extern int	     DRM(rmmap)( struct inode *inode, struct file *filp,
 				 unsigned int cmd, unsigned long arg );
+#if __HAVE_DMA
 extern int	     DRM(addbufs)( struct inode *inode, struct file *filp,
 				   unsigned int cmd, unsigned long arg );
 extern int	     DRM(infobufs)( struct inode *inode, struct file *filp,
@@ -857,27 +860,37 @@ extern int	     DRM(freebufs)( struct inode *inode, struct file *filp,
 extern int	     DRM(mapbufs)( struct inode *inode, struct file *filp,
 				   unsigned int cmd, unsigned long arg );
 
-
-#if __HAVE_DMA
-				/* DMA support (dma.c) */
+				/* DMA support (drm_dma.h) */
 extern int	     DRM(dma_setup)(drm_device_t *dev);
 extern void	     DRM(dma_takedown)(drm_device_t *dev);
 extern void	     DRM(free_buffer)(drm_device_t *dev, drm_buf_t *buf);
 extern void	     DRM(reclaim_buffers)(drm_device_t *dev, pid_t pid);
+#if __HAVE_OLD_DMA
+/* GH: This is a dirty hack for now...
+ */
 extern void	     DRM(clear_next_buffer)(drm_device_t *dev);
 extern int	     DRM(select_queue)(drm_device_t *dev,
 				       void (*wrapper)(unsigned long));
 extern int	     DRM(dma_enqueue)(drm_device_t *dev, drm_dma_t *dma);
-#if 0
 extern int	     DRM(dma_get_buffers)(drm_device_t *dev, drm_dma_t *dma);
 #endif
+#if __HAVE_DMA_IRQ
+extern int           DRM(control)( struct inode *inode, struct file *filp,
+				   unsigned int cmd, unsigned long arg );
+extern int           DRM(irq_install)( drm_device_t *dev, int irq );
+extern int           DRM(irq_uninstall)( drm_device_t *dev );
+extern void          DRM(dma_service)( int irq, void *device,
+				       struct pt_regs *regs );
+#if __HAVE_DMA_IRQ_BH
+extern void          DRM(dma_immediate_bh)( void *dev );
+#endif
+#endif
 #if DRM_DMA_HISTOGRAM
 extern int	     DRM(histogram_slot)(unsigned long count);
 extern void	     DRM(histogram_compute)(drm_device_t *dev, drm_buf_t *buf);
 #endif
-#endif
 
-				/* Buffer list management support (lists.c) */
+				/* Buffer list support (drm_lists.h) */
 #if __HAVE_DMA_WAITLIST
 extern int	     DRM(waitlist_create)(drm_waitlist_t *bl, int count);
 extern int	     DRM(waitlist_destroy)(drm_waitlist_t *bl);
@@ -891,9 +904,10 @@ extern int	     DRM(freelist_put)(drm_device_t *dev, drm_freelist_t *bl,
 				       drm_buf_t *buf);
 extern drm_buf_t     *DRM(freelist_get)(drm_freelist_t *bl, int block);
 #endif
+#endif /* __HAVE_DMA */
 
 #if __REALLY_HAVE_AGP
-				/* AGP/GART support (agpsupport.c) */
+				/* AGP/GART support (drm_agpsupport.h) */
 extern drm_agp_head_t *DRM(agp_init)(void);
 extern void           DRM(agp_uninit)(void);
 extern int            DRM(agp_acquire)(struct inode *inode, struct file *filp,
@@ -917,6 +931,7 @@ extern agp_memory     *DRM(agp_allocate_memory)(size_t pages, u32 type);
 extern int            DRM(agp_free_memory)(agp_memory *handle);
 extern int            DRM(agp_bind_memory)(agp_memory *handle, off_t start);
 extern int            DRM(agp_unbind_memory)(agp_memory *handle);
+#endif
 
 				/* Stub support (drm_stub.h) */
 int                   DRM(stub_register)(const char *name,
@@ -924,6 +939,14 @@ int                   DRM(stub_register)(const char *name,
 					 drm_device_t *dev);
 int                   DRM(stub_unregister)(int minor);
 
-#endif
-#endif
+				/* Proc support (drm_proc.h) */
+extern struct proc_dir_entry *DRM(proc_init)(drm_device_t *dev,
+					     int minor,
+					     struct proc_dir_entry *root,
+					     struct proc_dir_entry **dev_root);
+extern int            DRM(proc_cleanup)(int minor,
+					struct proc_dir_entry *root,
+					struct proc_dir_entry *dev_root);
+
+#endif /* __KERNEL__ */
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_agpsupport.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_agpsupport.h
index b070a59e6..dfd0d8fc2 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_agpsupport.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_agpsupport.h
@@ -77,7 +77,7 @@ int DRM(agp_acquire)(struct inode *inode, struct file *filp,
 	drm_device_t	 *dev	 = priv->dev;
 	int              retcode;
 
-	if (dev->agp->acquired || !drm_agp->acquire) return -EINVAL;
+	if (!dev->agp|| dev->agp->acquired || !drm_agp->acquire) return -EINVAL;
 	if ((retcode = drm_agp->acquire())) return retcode;
 	dev->agp->acquired = 1;
 	return 0;
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_bufs.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_bufs.h
index 02694f46d..1d5403225 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_bufs.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_bufs.h
@@ -135,6 +135,7 @@ int DRM(addmap)( struct inode *inode, struct file *filp,
 #if __REALLY_HAVE_AGP
 	case _DRM_AGP:
 		map->offset = map->offset + dev->agp->base;
+		map->mtrr   = dev->agp->agp_mtrr; /* for getmap */
 		break;
 #endif
 	default:
@@ -332,7 +333,7 @@ int DRM(addbufs_agp)( struct inode *inode, struct file *filp,
 		buf->order   = order;
 		buf->used    = 0;
 
-		buf->offset  = (dma->byte_count + offset); /* ******** */
+		buf->offset  = (dma->byte_count + offset);
 		buf->bus_address = agp_offset + offset;
 		buf->address = (void *)(agp_offset + offset);
 		buf->next    = NULL;
@@ -346,7 +347,7 @@ int DRM(addbufs_agp)( struct inode *inode, struct file *filp,
 					       DRM_MEM_BUFS );
 		memset( buf->dev_private, 0, buf->dev_priv_size );
 
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
 		buf->time_queued = 0;
 		buf->time_dispatched = 0;
 		buf->time_completed = 0;
@@ -377,17 +378,12 @@ int DRM(addbufs_agp)( struct inode *inode, struct file *filp,
 	DRM_DEBUG( "dma->buf_count : %d\n", dma->buf_count );
 	DRM_DEBUG( "entry->buf_count : %d\n", entry->buf_count );
 
-/* GH: Please leave this disabled for now.  I need to fix this properly...
- */
-#if 0
-	/* FIXME: work this mess out...
-	 */
+#if __HAVE_DMA_FREELIST
 	DRM(freelist_create)( &entry->freelist, entry->buf_count );
 	for ( i = 0 ; i < entry->buf_count ; i++ ) {
 		DRM(freelist_put)( dev, &entry->freelist, &entry->buflist[i] );
 	}
 #endif
-
 	up( &dev->struct_sem );
 
 	request.count = entry->buf_count;
@@ -523,11 +519,11 @@ int DRM(addbufs_pci)( struct inode *inode, struct file *filp,
 			buf->pending = 0;
 			init_waitqueue_head( &buf->dma_wait );
 			buf->pid     = 0;
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
 			buf->time_queued     = 0;
 			buf->time_dispatched = 0;
 			buf->time_completed  = 0;
-			buf->time_freed	     = 0;
+			buf->time_freed      = 0;
 #endif
 			DRM_DEBUG( "buffer %d @ %p\n",
 				   entry->buf_count, buf->address );
@@ -549,17 +545,12 @@ int DRM(addbufs_pci)( struct inode *inode, struct file *filp,
 	dma->page_count += entry->seg_count << page_order;
 	dma->byte_count += PAGE_SIZE * (entry->seg_count << page_order);
 
-/* GH: Please leave this disabled for now.  I need to fix this properly...
- */
-#if 0
-	/* FIXME: work this mess out...
-	 */
+#if __HAVE_DMA_FREELIST
 	DRM(freelist_create)( &entry->freelist, entry->buf_count );
 	for ( i = 0 ; i < entry->buf_count ; i++ ) {
 		DRM(freelist_put)( dev, &entry->freelist, &entry->buflist[i] );
 	}
 #endif
-
 	up( &dev->struct_sem );
 
 	request.count = entry->buf_count;
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_context.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_context.h
index 8f891fd2a..4ac896ef5 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_context.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_context.h
@@ -29,6 +29,11 @@
  *    Gareth Hughes <gareth@valinux.com>
  */
 
+#define __NO_VERSION__
+#include "drmP.h"
+
+#if __HAVE_CTX_BITMAP
+
 /* ================================================================
  * Context bitmap support
  */
@@ -210,7 +215,7 @@ int DRM(context_switch)( drm_device_t *dev, int old, int new )
                 return -EBUSY;
         }
 
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
         dev->ctx_start = get_cycles();
 #endif
 
@@ -243,7 +248,7 @@ int DRM(context_switch_complete)( drm_device_t *dev, int new )
 				/* If a context switch is ever initiated
                                    when the kernel holds the lock, release
                                    that lock here. */
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
         atomic_inc( &dev->histo.ctx[DRM(histogram_slot)(get_cycles()
 							- dev->ctx_start)] );
 
@@ -380,3 +385,376 @@ int DRM(rmctx)( struct inode *inode, struct file *filp,
 
 	return 0;
 }
+
+
+#else /* __HAVE_CTX_BITMAP */
+
+/* ================================================================
+ * Old-style context support
+ */
+
+
+int DRM(context_switch)(drm_device_t *dev, int old, int new)
+{
+	char	    buf[64];
+	drm_queue_t *q;
+
+#if 0
+	atomic_inc(&dev->total_ctx);
+#endif
+
+	if (test_and_set_bit(0, &dev->context_flag)) {
+		DRM_ERROR("Reentering -- FIXME\n");
+		return -EBUSY;
+	}
+
+#if __HAVE_DMA_HISTOGRAM
+	dev->ctx_start = get_cycles();
+#endif
+
+	DRM_DEBUG("Context switch from %d to %d\n", old, new);
+
+	if (new >= dev->queue_count) {
+		clear_bit(0, &dev->context_flag);
+		return -EINVAL;
+	}
+
+	if (new == dev->last_context) {
+		clear_bit(0, &dev->context_flag);
+		return 0;
+	}
+
+	q = dev->queuelist[new];
+	atomic_inc(&q->use_count);
+	if (atomic_read(&q->use_count) == 1) {
+		atomic_dec(&q->use_count);
+		clear_bit(0, &dev->context_flag);
+		return -EINVAL;
+	}
+
+	if (DRM(flags) & DRM_FLAG_NOCTX) {
+		DRM(context_switch_complete)(dev, new);
+	} else {
+		sprintf(buf, "C %d %d\n", old, new);
+		DRM(write_string)(dev, buf);
+	}
+
+	atomic_dec(&q->use_count);
+
+	return 0;
+}
+
+int DRM(context_switch_complete)(drm_device_t *dev, int new)
+{
+	drm_device_dma_t *dma = dev->dma;
+
+	dev->last_context = new;  /* PRE/POST: This is the _only_ writer. */
+	dev->last_switch  = jiffies;
+
+	if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("Lock isn't held after context switch\n");
+	}
+
+	if (!dma || !(dma->next_buffer && dma->next_buffer->while_locked)) {
+		if (DRM(lock_free)(dev, &dev->lock.hw_lock->lock,
+				  DRM_KERNEL_CONTEXT)) {
+			DRM_ERROR("Cannot free lock\n");
+		}
+	}
+
+#if __HAVE_DMA_HISTOGRAM
+	atomic_inc(&dev->histo.ctx[DRM(histogram_slot)(get_cycles()
+						      - dev->ctx_start)]);
+
+#endif
+	clear_bit(0, &dev->context_flag);
+	wake_up_interruptible(&dev->context_wait);
+
+	return 0;
+}
+
+static int DRM(init_queue)(drm_device_t *dev, drm_queue_t *q, drm_ctx_t *ctx)
+{
+	DRM_DEBUG("\n");
+
+	if (atomic_read(&q->use_count) != 1
+	    || atomic_read(&q->finalization)
+	    || atomic_read(&q->block_count)) {
+		DRM_ERROR("New queue is already in use: u%d f%d b%d\n",
+			  atomic_read(&q->use_count),
+			  atomic_read(&q->finalization),
+			  atomic_read(&q->block_count));
+	}
+
+	atomic_set(&q->finalization,  0);
+	atomic_set(&q->block_count,   0);
+	atomic_set(&q->block_read,    0);
+	atomic_set(&q->block_write,   0);
+	atomic_set(&q->total_queued,  0);
+	atomic_set(&q->total_flushed, 0);
+	atomic_set(&q->total_locks,   0);
+
+	init_waitqueue_head(&q->write_queue);
+	init_waitqueue_head(&q->read_queue);
+	init_waitqueue_head(&q->flush_queue);
+
+	q->flags = ctx->flags;
+
+	DRM(waitlist_create)(&q->waitlist, dev->dma->buf_count);
+
+	return 0;
+}
+
+
+/* drm_alloc_queue:
+PRE: 1) dev->queuelist[0..dev->queue_count] is allocated and will not
+	disappear (so all deallocation must be done after IOCTLs are off)
+     2) dev->queue_count < dev->queue_slots
+     3) dev->queuelist[i].use_count == 0 and
+	dev->queuelist[i].finalization == 0 if i not in use
+POST: 1) dev->queuelist[i].use_count == 1
+      2) dev->queue_count < dev->queue_slots */
+
+static int DRM(alloc_queue)(drm_device_t *dev)
+{
+	int	    i;
+	drm_queue_t *queue;
+	int	    oldslots;
+	int	    newslots;
+				/* Check for a free queue */
+	for (i = 0; i < dev->queue_count; i++) {
+		atomic_inc(&dev->queuelist[i]->use_count);
+		if (atomic_read(&dev->queuelist[i]->use_count) == 1
+		    && !atomic_read(&dev->queuelist[i]->finalization)) {
+			DRM_DEBUG("%d (free)\n", i);
+			return i;
+		}
+		atomic_dec(&dev->queuelist[i]->use_count);
+	}
+				/* Allocate a new queue */
+	down(&dev->struct_sem);
+
+	queue = gamma_alloc(sizeof(*queue), DRM_MEM_QUEUES);
+	memset(queue, 0, sizeof(*queue));
+	atomic_set(&queue->use_count, 1);
+
+	++dev->queue_count;
+	if (dev->queue_count >= dev->queue_slots) {
+		oldslots = dev->queue_slots * sizeof(*dev->queuelist);
+		if (!dev->queue_slots) dev->queue_slots = 1;
+		dev->queue_slots *= 2;
+		newslots = dev->queue_slots * sizeof(*dev->queuelist);
+
+		dev->queuelist = DRM(realloc)(dev->queuelist,
+					      oldslots,
+					      newslots,
+					      DRM_MEM_QUEUES);
+		if (!dev->queuelist) {
+			up(&dev->struct_sem);
+			DRM_DEBUG("out of memory\n");
+			return -ENOMEM;
+		}
+	}
+	dev->queuelist[dev->queue_count-1] = queue;
+
+	up(&dev->struct_sem);
+	DRM_DEBUG("%d (new)\n", dev->queue_count - 1);
+	return dev->queue_count - 1;
+}
+
+int DRM(resctx)(struct inode *inode, struct file *filp,
+		unsigned int cmd, unsigned long arg)
+{
+	drm_ctx_res_t	res;
+	drm_ctx_t	ctx;
+	int		i;
+
+	DRM_DEBUG("%d\n", DRM_RESERVED_CONTEXTS);
+	if (copy_from_user(&res, (drm_ctx_res_t *)arg, sizeof(res)))
+		return -EFAULT;
+	if (res.count >= DRM_RESERVED_CONTEXTS) {
+		memset(&ctx, 0, sizeof(ctx));
+		for (i = 0; i < DRM_RESERVED_CONTEXTS; i++) {
+			ctx.handle = i;
+			if (copy_to_user(&res.contexts[i],
+					 &i,
+					 sizeof(i)))
+				return -EFAULT;
+		}
+	}
+	res.count = DRM_RESERVED_CONTEXTS;
+	if (copy_to_user((drm_ctx_res_t *)arg, &res, sizeof(res)))
+		return -EFAULT;
+	return 0;
+}
+
+int DRM(addctx)(struct inode *inode, struct file *filp,
+		unsigned int cmd, unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+
+	if (copy_from_user(&ctx, (drm_ctx_t *)arg, sizeof(ctx)))
+		return -EFAULT;
+	if ((ctx.handle = DRM(alloc_queue)(dev)) == DRM_KERNEL_CONTEXT) {
+				/* Init kernel's context and get a new one. */
+		DRM(init_queue)(dev, dev->queuelist[ctx.handle], &ctx);
+		ctx.handle = DRM(alloc_queue)(dev);
+	}
+	DRM(init_queue)(dev, dev->queuelist[ctx.handle], &ctx);
+	DRM_DEBUG("%d\n", ctx.handle);
+	if (copy_to_user((drm_ctx_t *)arg, &ctx, sizeof(ctx)))
+		return -EFAULT;
+	return 0;
+}
+
+int DRM(modctx)(struct inode *inode, struct file *filp,
+		unsigned int cmd, unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+	drm_queue_t	*q;
+
+	if (copy_from_user(&ctx, (drm_ctx_t *)arg, sizeof(ctx)))
+		return -EFAULT;
+
+	DRM_DEBUG("%d\n", ctx.handle);
+
+	if (ctx.handle < 0 || ctx.handle >= dev->queue_count) return -EINVAL;
+	q = dev->queuelist[ctx.handle];
+
+	atomic_inc(&q->use_count);
+	if (atomic_read(&q->use_count) == 1) {
+				/* No longer in use */
+		atomic_dec(&q->use_count);
+		return -EINVAL;
+	}
+
+	if (DRM_BUFCOUNT(&q->waitlist)) {
+		atomic_dec(&q->use_count);
+		return -EBUSY;
+	}
+
+	q->flags = ctx.flags;
+
+	atomic_dec(&q->use_count);
+	return 0;
+}
+
+int DRM(getctx)(struct inode *inode, struct file *filp,
+		unsigned int cmd, unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+	drm_queue_t	*q;
+
+	if (copy_from_user(&ctx, (drm_ctx_t *)arg, sizeof(ctx)))
+		return -EFAULT;
+
+	DRM_DEBUG("%d\n", ctx.handle);
+
+	if (ctx.handle >= dev->queue_count) return -EINVAL;
+	q = dev->queuelist[ctx.handle];
+
+	atomic_inc(&q->use_count);
+	if (atomic_read(&q->use_count) == 1) {
+				/* No longer in use */
+		atomic_dec(&q->use_count);
+		return -EINVAL;
+	}
+
+	ctx.flags = q->flags;
+	atomic_dec(&q->use_count);
+
+	if (copy_to_user((drm_ctx_t *)arg, &ctx, sizeof(ctx)))
+		return -EFAULT;
+
+	return 0;
+}
+
+int DRM(switchctx)(struct inode *inode, struct file *filp,
+		   unsigned int cmd, unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+
+	if (copy_from_user(&ctx, (drm_ctx_t *)arg, sizeof(ctx)))
+		return -EFAULT;
+	DRM_DEBUG("%d\n", ctx.handle);
+	return DRM(context_switch)(dev, dev->last_context, ctx.handle);
+}
+
+int DRM(newctx)(struct inode *inode, struct file *filp,
+		unsigned int cmd, unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+
+	if (copy_from_user(&ctx, (drm_ctx_t *)arg, sizeof(ctx)))
+		return -EFAULT;
+	DRM_DEBUG("%d\n", ctx.handle);
+	DRM(context_switch_complete)(dev, ctx.handle);
+
+	return 0;
+}
+
+int DRM(rmctx)(struct inode *inode, struct file *filp,
+	       unsigned int cmd, unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+	drm_queue_t	*q;
+	drm_buf_t	*buf;
+
+	if (copy_from_user(&ctx, (drm_ctx_t *)arg, sizeof(ctx)))
+		return -EFAULT;
+	DRM_DEBUG("%d\n", ctx.handle);
+
+	if (ctx.handle >= dev->queue_count) return -EINVAL;
+	q = dev->queuelist[ctx.handle];
+
+	atomic_inc(&q->use_count);
+	if (atomic_read(&q->use_count) == 1) {
+				/* No longer in use */
+		atomic_dec(&q->use_count);
+		return -EINVAL;
+	}
+
+	atomic_inc(&q->finalization); /* Mark queue in finalization state */
+	atomic_sub(2, &q->use_count); /* Mark queue as unused (pending
+					 finalization) */
+
+	while (test_and_set_bit(0, &dev->interrupt_flag)) {
+		schedule();
+		if (signal_pending(current)) {
+			clear_bit(0, &dev->interrupt_flag);
+			return -EINTR;
+		}
+	}
+				/* Remove queued buffers */
+	while ((buf = DRM(waitlist_get)(&q->waitlist))) {
+		DRM(free_buffer)(dev, buf);
+	}
+	clear_bit(0, &dev->interrupt_flag);
+
+				/* Wakeup blocked processes */
+	wake_up_interruptible(&q->read_queue);
+	wake_up_interruptible(&q->write_queue);
+	wake_up_interruptible(&q->flush_queue);
+
+				/* Finalization over.  Queue is made
+				   available when both use_count and
+				   finalization become 0, which won't
+				   happen until all the waiting processes
+				   stop waiting. */
+	atomic_dec(&q->finalization);
+	return 0;
+}
+
+#endif /* __HAVE_CTX_BITMAP */
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_dma.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_dma.h
index 35eb896e2..e715bd416 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_dma.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_dma.h
@@ -118,7 +118,7 @@ void DRM(dma_takedown)(drm_device_t *dev)
 }
 
 
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
 /* This is slow, but is useful for debugging. */
 int DRM(histogram_slot)(unsigned long count)
 {
@@ -182,7 +182,7 @@ void DRM(free_buffer)(drm_device_t *dev, drm_buf_t *buf)
 	buf->pending  = 0;
 	buf->pid      = 0;
 	buf->used     = 0;
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
 	buf->time_completed = get_cycles();
 #endif
 
@@ -231,85 +231,6 @@ void DRM(reclaim_buffers)(drm_device_t *dev, pid_t pid)
  */
 #if __HAVE_OLD_DMA
 
-int drm_context_switch(drm_device_t *dev, int old, int new)
-{
-	char	    buf[64];
-	drm_queue_t *q;
-
-#if 0
-	atomic_inc(&dev->total_ctx);
-#endif
-
-	if (test_and_set_bit(0, &dev->context_flag)) {
-		DRM_ERROR("Reentering -- FIXME\n");
-		return -EBUSY;
-	}
-
-#if DRM_DMA_HISTOGRAM
-	dev->ctx_start = get_cycles();
-#endif
-
-	DRM_DEBUG("Context switch from %d to %d\n", old, new);
-
-	if (new >= dev->queue_count) {
-		clear_bit(0, &dev->context_flag);
-		return -EINVAL;
-	}
-
-	if (new == dev->last_context) {
-		clear_bit(0, &dev->context_flag);
-		return 0;
-	}
-
-	q = dev->queuelist[new];
-	atomic_inc(&q->use_count);
-	if (atomic_read(&q->use_count) == 1) {
-		atomic_dec(&q->use_count);
-		clear_bit(0, &dev->context_flag);
-		return -EINVAL;
-	}
-
-	if (DRM(flags) & DRM_FLAG_NOCTX) {
-		drm_context_switch_complete(dev, new);
-	} else {
-		sprintf(buf, "C %d %d\n", old, new);
-		DRM(write_string)(dev, buf);
-	}
-
-	atomic_dec(&q->use_count);
-
-	return 0;
-}
-
-int drm_context_switch_complete(drm_device_t *dev, int new)
-{
-	drm_device_dma_t *dma = dev->dma;
-
-	dev->last_context = new;  /* PRE/POST: This is the _only_ writer. */
-	dev->last_switch  = jiffies;
-
-	if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
-		DRM_ERROR("Lock isn't held after context switch\n");
-	}
-
-	if (!dma || !(dma->next_buffer && dma->next_buffer->while_locked)) {
-		if (DRM(lock_free)(dev, &dev->lock.hw_lock->lock,
-				  DRM_KERNEL_CONTEXT)) {
-			DRM_ERROR("Cannot free lock\n");
-		}
-	}
-
-#if DRM_DMA_HISTOGRAM
-	atomic_inc(&dev->histo.ctx[DRM(histogram_slot)(get_cycles()
-						      - dev->ctx_start)]);
-
-#endif
-	clear_bit(0, &dev->context_flag);
-	wake_up_interruptible(&dev->context_wait);
-
-	return 0;
-}
-
 void DRM(clear_next_buffer)(drm_device_t *dev)
 {
 	drm_device_dma_t *dma = dev->dma;
@@ -571,6 +492,103 @@ int DRM(dma_get_buffers)(drm_device_t *dev, drm_dma_t *dma)
 	return 0;
 }
 
-#endif
+#endif /* __HAVE_OLD_DMA */
+
+
+#if __HAVE_DMA_IRQ
+
+int DRM(irq_install)( drm_device_t *dev, int irq )
+{
+	int ret;
+
+	if ( !irq )
+		return -EINVAL;
+
+	down( &dev->struct_sem );
+	if ( dev->irq ) {
+		up( &dev->struct_sem );
+		return -EBUSY;
+	}
+	dev->irq = irq;
+	up( &dev->struct_sem );
+
+	DRM_DEBUG( "%s: irq=%d\n", __FUNCTION__, irq );
+
+	dev->context_flag = 0;
+	dev->interrupt_flag = 0;
+	dev->dma_flag = 0;
+
+	dev->dma->next_buffer = NULL;
+	dev->dma->next_queue = NULL;
+	dev->dma->this_buffer = NULL;
 
+#if __HAVE_DMA_IRQ_BH
+	INIT_LIST_HEAD( &dev->tq.list );
+	dev->tq.sync = 0;
+	dev->tq.routine = DRM(dma_immediate_bh);
+	dev->tq.data = dev;
 #endif
+
+				/* Before installing handler */
+	DRIVER_PREINSTALL();
+
+				/* Install handler */
+	ret = request_irq( dev->irq, DRM(dma_service),
+			   0, dev->devname, dev );
+	if ( ret < 0 ) {
+		down( &dev->struct_sem );
+		dev->irq = 0;
+		up( &dev->struct_sem );
+		return ret;
+	}
+
+				/* After installing handler */
+	DRIVER_POSTINSTALL();
+
+	return 0;
+}
+
+int DRM(irq_uninstall)( drm_device_t *dev )
+{
+	int irq;
+
+	down( &dev->struct_sem );
+	irq = dev->irq;
+	dev->irq = 0;
+	up( &dev->struct_sem );
+
+	if ( !irq )
+		return -EINVAL;
+
+	DRM_DEBUG( "%s: irq=%d\n", __FUNCTION__, irq );
+
+	DRIVER_UNINSTALL();
+
+	free_irq( irq, dev );
+
+	return 0;
+}
+
+int DRM(control)( struct inode *inode, struct file *filp,
+		  unsigned int cmd, unsigned long arg )
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
+	drm_control_t ctl;
+
+	if ( copy_from_user( &ctl, (drm_control_t *)arg, sizeof(ctl) ) )
+		return -EFAULT;
+
+	switch ( ctl.func ) {
+	case DRM_INST_HANDLER:
+		return DRM(irq_install)( dev, ctl.irq );
+	case DRM_UNINST_HANDLER:
+		return DRM(irq_uninstall)( dev );
+	default:
+		return -EINVAL;
+	}
+}
+
+#endif /* __HAVE_DMA_IRQ */
+
+#endif /* __HAVE_DMA */
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_drv.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_drv.h
index 05923b50b..7447ca6d4 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_drv.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_drv.h
@@ -492,7 +492,8 @@ static int __init drm_init( void )
 	}
 #endif
 #if __REALLY_HAVE_MTRR
-	dev->agp->agp_mtrr = mtrr_add( dev->agp->agp_info.aper_base,
+	if (dev->agp)
+		dev->agp->agp_mtrr = mtrr_add( dev->agp->agp_info.aper_base,
 				       dev->agp->agp_info.aper_size*1024*1024,
 				       MTRR_TYPE_WRCOMB,
 				       1 );
@@ -646,7 +647,7 @@ int DRM(release)( struct inode *inode, struct file *filp )
 	if ( dev->lock.hw_lock &&
 	     _DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock) &&
 	     dev->lock.pid == current->pid ) {
-		DRM_ERROR( "Process %d dead, freeing lock for context %d\n",
+		DRM_DEBUG( "Process %d dead, freeing lock for context %d\n",
 			   current->pid,
 			   _DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock) );
 #if __HAVE_RELEASE
@@ -697,7 +698,7 @@ int DRM(release)( struct inode *inode, struct file *filp )
 					DRM_KERNEL_CONTEXT );
 		}
 	}
-#else
+#elif __HAVE_DMA
 	DRM(reclaim_buffers)( dev, priv->pid );
 #endif
 
@@ -804,7 +805,7 @@ int DRM(lock)( struct inode *inode, struct file *filp,
 #if __HAVE_MULTIPLE_DMA_QUEUES
 	drm_queue_t *q;
 #endif
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
         cycles_t start;
 
         dev->lck_start = start = get_cycles();
@@ -892,7 +893,7 @@ int DRM(lock)( struct inode *inode, struct file *filp,
 
         DRM_DEBUG( "%d %s\n", lock.context, ret ? "interrupted" : "has lock" );
 
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
         atomic_inc(&dev->histo.lacq[DRM(histogram_slot)(get_cycles()-start)]);
 #endif
         return ret;
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_fops.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_fops.h
index d646f3188..dd5747669 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_fops.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_fops.h
@@ -83,50 +83,6 @@ int DRM(flush)(struct file *filp)
 	return 0;
 }
 
-#if 0
-/* drm_release is called whenever a process closes /dev/drm*.  Linux calls
-   this only if any mappings have been closed. */
-
-int drm_release(struct inode *inode, struct file *filp)
-{
-	drm_file_t    *priv   = filp->private_data;
-	drm_device_t  *dev    = priv->dev;
-
-	DRM_DEBUG("pid = %d, device = 0x%x, open_count = %d\n",
-		  current->pid, dev->device, dev->open_count);
-
-	if (dev->lock.hw_lock
-	    && _DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)
-	    && dev->lock.pid == current->pid) {
-		DRM_ERROR("Process %d dead, freeing lock for context %d\n",
-			  current->pid,
-			  _DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock));
-		drm_lock_free(dev,
-			      &dev->lock.hw_lock->lock,
-			      _DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock));
-
-				/* FIXME: may require heavy-handed reset of
-                                   hardware at this point, possibly
-                                   processed via a callback to the X
-                                   server. */
-	}
-	drm_reclaim_buffers(dev, priv->pid);
-
-	drm_fasync(-1, filp, 0);
-
-	down(&dev->struct_sem);
-	if (priv->prev) priv->prev->next = priv->next;
-	else		dev->file_first	 = priv->next;
-	if (priv->next) priv->next->prev = priv->prev;
-	else		dev->file_last	 = priv->prev;
-	up(&dev->struct_sem);
-
-	drm_free(priv, sizeof(*priv), DRM_MEM_FILES);
-
-	return 0;
-}
-#endif
-
 int DRM(fasync)(int fd, struct file *filp, int on)
 {
 	drm_file_t    *priv   = filp->private_data;
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_lists.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_lists.h
index b597ef770..f8dbaaa71 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_lists.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_lists.h
@@ -74,7 +74,7 @@ int DRM(waitlist_put)(drm_waitlist_t *bl, drm_buf_t *buf)
 			  buf->idx, buf->pid);
 		return -EINVAL;
 	}
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
 	buf->time_queued = get_cycles();
 #endif
 	buf->list	 = DRM_LIST_WAIT;
@@ -106,6 +106,7 @@ drm_buf_t *DRM(waitlist_get)(drm_waitlist_t *bl)
 
 #endif /* __HAVE_DMA_WAITLIST */
 
+
 #if __HAVE_DMA_FREELIST
 
 int DRM(freelist_create)(drm_freelist_t *bl, int count)
@@ -142,7 +143,7 @@ int DRM(freelist_put)(drm_device_t *dev, drm_freelist_t *bl, drm_buf_t *buf)
 			  buf->idx, buf->waiting, buf->pending, buf->list);
 	}
 	if (!bl) return 1;
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
 	buf->time_freed = get_cycles();
 	DRM(histogram_compute)(dev, buf);
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_memory.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_memory.h
index caf053943..1763d9b41 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_memory.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_memory.h
@@ -351,8 +351,7 @@ void DRM(ioremapfree)(void *pt, unsigned long size)
 	}
 }
 
-#if defined(CONFIG_AGP) || defined(CONFIG_AGP_MODULE)
-#if __MUST_HAVE_AGP
+#if __REALLY_HAVE_AGP
 
 agp_memory *DRM(alloc_agp)(int pages, u32 type)
 {
@@ -457,4 +456,3 @@ int DRM(unbind_agp)(agp_memory *handle)
 	return retcode;
 }
 #endif
-#endif /* defined(CONFIG_AGP) || defined(CONFIG_AGP_MODULE) */
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_proc.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_proc.h
index 606286607..f65f42b71 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_proc.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_proc.h
@@ -50,7 +50,7 @@ static int	   DRM(bufs_info)(char *buf, char **start, off_t offset,
 static int	   DRM(vma_info)(char *buf, char **start, off_t offset,
 				 int request, int *eof, void *data);
 #endif
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
 static int	   DRM(histo_info)(char *buf, char **start, off_t offset,
 				   int request, int *eof, void *data);
 #endif
@@ -68,7 +68,7 @@ struct drm_proc_list {
 #if DRM_DEBUG_CODE
 	{ "vma",     DRM(vma_info)     },
 #endif
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
 	{ "histo",   DRM(histo_info)   },
 #endif
 };
@@ -488,7 +488,7 @@ static int DRM(vma_info)(char *buf, char **start, off_t offset, int request,
 #endif
 
 
-#if DRM_DMA_HISTOGRAM
+#if __HAVE_DMA_HISTOGRAM
 static int DRM(_histo_info)(char *buf, char **start, off_t offset, int request,
 			    int *eof, void *data)
 {
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_stub.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_stub.h
index ed37af4da..bc9587963 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_stub.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm_stub.h
@@ -1,4 +1,4 @@
-/* drm_stub.c -- -*- linux-c -*-
+/* drm_stub.h -- -*- linux-c -*-
  * Created: Fri Jan 19 10:48:35 2001 by faith@acm.org
  *
  * Copyright 2001 VA Linux Systems, Inc., Sunnyvale, California.
@@ -28,6 +28,13 @@
  *
  */
 
+#define __NO_VERSION__
+#include "drmP.h"
+
+#if LINUX_VERSION_CODE < 0x020400
+#include "stubsupport-pre24.h"
+#endif
+
 #define DRM_STUB_MAXCARDS 16	/* Enough for one machine */
 
 static struct drm_stub_list {
@@ -58,7 +65,7 @@ static int DRM(stub_open)(struct inode *inode, struct file *filp)
 		filp->f_op = fops_get(old_fops);
 	}
 	fops_put(old_fops);
-		
+
 	return err;
 }
 
@@ -73,7 +80,7 @@ static int DRM(stub_getminor)(const char *name, struct file_operations *fops,
 			      drm_device_t *dev)
 {
 	int i;
-	
+
 	if (!DRM(stub_list)) {
 		DRM(stub_list) = DRM(alloc)(sizeof(*DRM(stub_list))
 					    * DRM_STUB_MAXCARDS, DRM_MEM_STUB);
@@ -117,10 +124,13 @@ static int DRM(stub_putminor)(int minor)
 int DRM(stub_register)(const char *name, struct file_operations *fops,
 		       drm_device_t *dev)
 {
-	if (register_chrdev(DRM_MAJOR, "drm", &DRM(stub_fops))) {
-				/* Already registered */
-		struct drm_stub_info *i;
+	struct drm_stub_info *i = NULL;
+	
+	if (register_chrdev(DRM_MAJOR, "drm", &DRM(stub_fops)))
 		i = (struct drm_stub_info *)inter_module_get("drm");
+
+	if (i) {
+				/* Already registered */
 		DRM(stub_info).info_register   = i->info_register;
 		DRM(stub_info).info_unregister = i->info_unregister;
 	} else {
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma.h
index 5ca1f7451..232ed018b 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma.h
@@ -1,3 +1,32 @@
+/* gamma.c -- 3dlabs GMX 2000 driver -*- linux-c -*-
+ * Created: Mon Jan  4 08:58:31 1999 by gareth@valinux.com
+ *
+ * Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
+ * 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
+ * PRECISION INSIGHT 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.
+ *
+ * Authors:
+ *    Gareth Hughes <gareth@valinux.com>
+ */
+
 #ifndef __GAMMA_H__
 #define __GAMMA_H__
 
@@ -5,18 +34,60 @@
  */
 #define DRM(x) gamma_##x
 
-#define __HAVE_MTRR		1
-#define __HAVE_PCI_DMA		1
-#define __HAVE_CTX_BITMAP	0
+/* General customization:
+ */
+#define __HAVE_MTRR			1
+
+/* DMA customization:
+ */
+#define __HAVE_DMA			1
+#define __HAVE_OLD_DMA			1
+#define __HAVE_PCI_DMA			1
+
 #define __HAVE_MULTIPLE_DMA_QUEUES	1
+#define __HAVE_DMA_WAITQUEUE		1
+
+#define __HAVE_DMA_WAITLIST		1
+#define __HAVE_DMA_FREELIST		1
+
 #define __HAVE_DMA_FLUSH		1
-#define __HAVE_DMA_QUEUE		0
 #define __HAVE_DMA_SCHEDULE		1
-#define __HAVE_DMA_WAITQUEUE		1
-#define __HAVE_DMA_WAITLIST	1
-#define __HAVE_DMA_FREELIST	1
-#define __HAVE_DMA		1
-#define __HAVE_OLD_DMA		1
-#define __HAVE_DMA_IRQ		1
 
-#endif
+#define __HAVE_DMA_READY		1
+#define DRIVER_DMA_READY() do {						\
+	gamma_dma_ready(dev);						\
+} while (0)
+
+#define __HAVE_DMA_QUIESCENT		1
+#define DRIVER_DMA_QUIESCENT() do {					\
+	/* FIXME ! */ 							\
+	gamma_dma_quiescent_dual(dev);					\
+	return 0;							\
+} while (0)
+
+#define __HAVE_DMA_IRQ			1
+#define __HAVE_DMA_IRQ_BH		1
+#define DRIVER_PREINSTALL() do {					\
+	drm_gamma_private_t *dev_priv =					\
+				(drm_gamma_private_t *)dev->dev_private;\
+	GAMMA_WRITE( GAMMA_GCOMMANDMODE,	0x00000000 );		\
+	GAMMA_WRITE( GAMMA_GDMACONTROL,		0x00000000 );		\
+} while (0)
+
+#define DRIVER_POSTINSTALL() do {					\
+	drm_gamma_private_t *dev_priv =					\
+				(drm_gamma_private_t *)dev->dev_private;\
+	GAMMA_WRITE( GAMMA_GINTENABLE,		0x00002001 );		\
+	GAMMA_WRITE( GAMMA_COMMANDINTENABLE,	0x00000008 );		\
+	GAMMA_WRITE( GAMMA_GDELAYTIMER,		0x00039090 );		\
+} while (0)
+
+#define DRIVER_UNINSTALL() do {						\
+	drm_gamma_private_t *dev_priv =					\
+				(drm_gamma_private_t *)dev->dev_private;\
+	GAMMA_WRITE( GAMMA_GDELAYTIMER,		0x00000000 );		\
+	GAMMA_WRITE( GAMMA_COMMANDINTENABLE,	0x00000000 );		\
+	GAMMA_WRITE( GAMMA_GINTENABLE,		0x00000000 );		\
+} while (0)
+
+#endif /* __GAMMA_H__ */
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_context.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_context.c
deleted file mode 100644
index d94054a29..000000000
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_context.c
+++ /dev/null
@@ -1,319 +0,0 @@
-/* context.c -- IOCTLs for contexts and DMA queues -*- linux-c -*-
- * Created: Tue Feb  2 08:37:54 1999 by faith@precisioninsight.com
- *
- * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
- * Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
- * 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
- * PRECISION INSIGHT 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.
- * 
- * Authors:
- *    Rickard E. (Rik) Faith <faith@valinux.com>
- *
- */
-
-#define __NO_VERSION__
-#include "gamma.h"
-#include "drmP.h"
-
-static int drm_init_queue(drm_device_t *dev, drm_queue_t *q, drm_ctx_t *ctx)
-{
-	DRM_DEBUG("\n");
-	
-	if (atomic_read(&q->use_count) != 1
-	    || atomic_read(&q->finalization)
-	    || atomic_read(&q->block_count)) {
-		DRM_ERROR("New queue is already in use: u%d f%d b%d\n",
-			  atomic_read(&q->use_count),
-			  atomic_read(&q->finalization),
-			  atomic_read(&q->block_count));
-	}
-		  
-	atomic_set(&q->finalization,  0);
-	atomic_set(&q->block_count,   0);
-	atomic_set(&q->block_read,    0);
-	atomic_set(&q->block_write,   0);
-	atomic_set(&q->total_queued,  0);
-	atomic_set(&q->total_flushed, 0);
-	atomic_set(&q->total_locks,   0);
-
-	init_waitqueue_head(&q->write_queue);
-	init_waitqueue_head(&q->read_queue);
-	init_waitqueue_head(&q->flush_queue);
-
-	q->flags = ctx->flags;
-
-	gamma_waitlist_create(&q->waitlist, dev->dma->buf_count);
-
-	return 0;
-}
-
-
-/* drm_alloc_queue:
-PRE: 1) dev->queuelist[0..dev->queue_count] is allocated and will not
-	disappear (so all deallocation must be done after IOCTLs are off)
-     2) dev->queue_count < dev->queue_slots
-     3) dev->queuelist[i].use_count == 0 and
-	dev->queuelist[i].finalization == 0 if i not in use 
-POST: 1) dev->queuelist[i].use_count == 1
-      2) dev->queue_count < dev->queue_slots */
-		
-static int drm_alloc_queue(drm_device_t *dev)
-{
-	int	    i;
-	drm_queue_t *queue;
-	int	    oldslots;
-	int	    newslots;
-				/* Check for a free queue */
-	for (i = 0; i < dev->queue_count; i++) {
-		atomic_inc(&dev->queuelist[i]->use_count);
-		if (atomic_read(&dev->queuelist[i]->use_count) == 1
-		    && !atomic_read(&dev->queuelist[i]->finalization)) {
-			DRM_DEBUG("%d (free)\n", i);
-			return i;
-		}
-		atomic_dec(&dev->queuelist[i]->use_count);
-	}
-				/* Allocate a new queue */
-	down(&dev->struct_sem);
-	
-	queue = gamma_alloc(sizeof(*queue), DRM_MEM_QUEUES);
-	memset(queue, 0, sizeof(*queue));
-	atomic_set(&queue->use_count, 1);
-	
-	++dev->queue_count;
-	if (dev->queue_count >= dev->queue_slots) {
-		oldslots = dev->queue_slots * sizeof(*dev->queuelist);
-		if (!dev->queue_slots) dev->queue_slots = 1;
-		dev->queue_slots *= 2;
-		newslots = dev->queue_slots * sizeof(*dev->queuelist);
-
-		dev->queuelist = gamma_realloc(dev->queuelist,
-					     oldslots,
-					     newslots,
-					     DRM_MEM_QUEUES);
-		if (!dev->queuelist) {
-			up(&dev->struct_sem);
-			DRM_DEBUG("out of memory\n");
-			return -ENOMEM;
-		}
-	}
-	dev->queuelist[dev->queue_count-1] = queue;
-	
-	up(&dev->struct_sem);
-	DRM_DEBUG("%d (new)\n", dev->queue_count - 1);
-	return dev->queue_count - 1;
-}
-
-int gamma_resctx(struct inode *inode, struct file *filp, unsigned int cmd,
-	       unsigned long arg)
-{
-	drm_ctx_res_t	res;
-	drm_ctx_t	ctx;
-	int		i;
-
-	DRM_DEBUG("%d\n", DRM_RESERVED_CONTEXTS);
-	if (copy_from_user(&res, (drm_ctx_res_t *)arg, sizeof(res)))
-		return -EFAULT;
-	if (res.count >= DRM_RESERVED_CONTEXTS) {
-		memset(&ctx, 0, sizeof(ctx));
-		for (i = 0; i < DRM_RESERVED_CONTEXTS; i++) {
-			ctx.handle = i;
-			if (copy_to_user(&res.contexts[i],
-					 &i,
-					 sizeof(i)))
-				return -EFAULT;
-		}
-	}
-	res.count = DRM_RESERVED_CONTEXTS;
-	if (copy_to_user((drm_ctx_res_t *)arg, &res, sizeof(res)))
-		return -EFAULT;
-	return 0;
-}
-
-
-int gamma_addctx(struct inode *inode, struct file *filp, unsigned int cmd,
-	       unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_t	ctx;
-
-	if (copy_from_user(&ctx, (drm_ctx_t *)arg, sizeof(ctx)))
-		return -EFAULT;
-	if ((ctx.handle = drm_alloc_queue(dev)) == DRM_KERNEL_CONTEXT) {
-				/* Init kernel's context and get a new one. */
-		drm_init_queue(dev, dev->queuelist[ctx.handle], &ctx);
-		ctx.handle = drm_alloc_queue(dev);
-	}
-	drm_init_queue(dev, dev->queuelist[ctx.handle], &ctx);
-	DRM_DEBUG("%d\n", ctx.handle);
-	if (copy_to_user((drm_ctx_t *)arg, &ctx, sizeof(ctx)))
-		return -EFAULT;
-	return 0;
-}
-
-int gamma_modctx(struct inode *inode, struct file *filp, unsigned int cmd,
-	       unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_t	ctx;
-	drm_queue_t	*q;
-		
-	if (copy_from_user(&ctx, (drm_ctx_t *)arg, sizeof(ctx)))
-		return -EFAULT;
-	
-	DRM_DEBUG("%d\n", ctx.handle);
-	
-	if (ctx.handle < 0 || ctx.handle >= dev->queue_count) return -EINVAL;
-	q = dev->queuelist[ctx.handle];
-	
-	atomic_inc(&q->use_count);
-	if (atomic_read(&q->use_count) == 1) {
-				/* No longer in use */
-		atomic_dec(&q->use_count);
-		return -EINVAL;
-	}
-
-	if (DRM_BUFCOUNT(&q->waitlist)) {
-		atomic_dec(&q->use_count);
-		return -EBUSY;
-	}
-	
-	q->flags = ctx.flags;
-	
-	atomic_dec(&q->use_count);
-	return 0;
-}
-
-int gamma_getctx(struct inode *inode, struct file *filp, unsigned int cmd,
-	       unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_t	ctx;
-	drm_queue_t	*q;
-		
-	if (copy_from_user(&ctx, (drm_ctx_t *)arg, sizeof(ctx)))
-		return -EFAULT;
-	
-	DRM_DEBUG("%d\n", ctx.handle);
-	
-	if (ctx.handle >= dev->queue_count) return -EINVAL;
-	q = dev->queuelist[ctx.handle];
-	
-	atomic_inc(&q->use_count);
-	if (atomic_read(&q->use_count) == 1) {
-				/* No longer in use */
-		atomic_dec(&q->use_count);
-		return -EINVAL;
-	}
-	
-	ctx.flags = q->flags;
-	atomic_dec(&q->use_count);
-	
-	if (copy_to_user((drm_ctx_t *)arg, &ctx, sizeof(ctx)))
-		return -EFAULT;
-	
-	return 0;
-}
-
-int gamma_switchctx(struct inode *inode, struct file *filp, unsigned int cmd,
-		  unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_t	ctx;
-
-	if (copy_from_user(&ctx, (drm_ctx_t *)arg, sizeof(ctx)))
-		return -EFAULT;
-	DRM_DEBUG("%d\n", ctx.handle);
-	return drm_context_switch(dev, dev->last_context, ctx.handle);
-}
-
-int gamma_newctx(struct inode *inode, struct file *filp, unsigned int cmd,
-	       unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_t	ctx;
-
-	if (copy_from_user(&ctx, (drm_ctx_t *)arg, sizeof(ctx)))
-		return -EFAULT;
-	DRM_DEBUG("%d\n", ctx.handle);
-	drm_context_switch_complete(dev, ctx.handle);
-
-	return 0;
-}
-
-int gamma_rmctx(struct inode *inode, struct file *filp, unsigned int cmd,
-	      unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_t	ctx;
-	drm_queue_t	*q;
-	drm_buf_t	*buf;
-
-	if (copy_from_user(&ctx, (drm_ctx_t *)arg, sizeof(ctx)))
-		return -EFAULT;
-	DRM_DEBUG("%d\n", ctx.handle);
-	
-	if (ctx.handle >= dev->queue_count) return -EINVAL;
-	q = dev->queuelist[ctx.handle];
-	
-	atomic_inc(&q->use_count);
-	if (atomic_read(&q->use_count) == 1) {
-				/* No longer in use */
-		atomic_dec(&q->use_count);
-		return -EINVAL;
-	}
-	
-	atomic_inc(&q->finalization); /* Mark queue in finalization state */
-	atomic_sub(2, &q->use_count); /* Mark queue as unused (pending
-					 finalization) */
-
-	while (test_and_set_bit(0, &dev->interrupt_flag)) {
-		schedule();
-		if (signal_pending(current)) {
-			clear_bit(0, &dev->interrupt_flag);
-			return -EINTR;
-		}
-	}
-				/* Remove queued buffers */
-	while ((buf = gamma_waitlist_get(&q->waitlist))) {
-		gamma_free_buffer(dev, buf);
-	}
-	clear_bit(0, &dev->interrupt_flag);
-	
-				/* Wakeup blocked processes */
-	wake_up_interruptible(&q->read_queue);
-	wake_up_interruptible(&q->write_queue);
-	wake_up_interruptible(&q->flush_queue);
-	
-				/* Finalization over.  Queue is made
-				   available when both use_count and
-				   finalization become 0, which won't
-				   happen until all the waiting processes
-				   stop waiting. */
-	atomic_dec(&q->finalization);
-	return 0;
-}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_dma.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_dma.c
index e2ac60616..d3a9414d7 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_dma.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_dma.c
@@ -11,11 +11,11 @@
  * 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
@@ -23,7 +23,7 @@
  * 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.
- * 
+ *
  * Authors:
  *    Rickard E. (Rik) Faith <faith@valinux.com>
  *
@@ -37,76 +37,13 @@
 #include <linux/interrupt.h>	/* For task queue support */
 #include <linux/delay.h>
 
-#if 0
-#define DO_IOREMAP( _map )						\
-do {									\
-	(_map)->handle = DRM(ioremap)( (_map)->offset, (_map)->size );	\
-} while (0)
-
-#define DO_IOREMAPFREE( _map )						\
-do {									\
-	if ( (_map)->handle && (_map)->size )				\
-		DRM(ioremapfree)( (_map)->handle, (_map)->size );	\
-} while (0)
-
-#define DO_FIND_MAP( _map, _offset )					\
-do {									\
-	int _i;								\
-	for ( _i = 0 ; _i < dev->map_count ; _i++ ) {			\
-		if ( dev->maplist[_i]->offset == _offset ) {		\
-			_map = dev->maplist[_i];			\
-			break;						\
-		}							\
-	}								\
-} while (0)
-#endif
-
-/* WARNING!!! MAGIC NUMBER!!!  The number of regions already added to the
-   kernel must be specified here.  Currently, the number is 2.	This must
-   match the order the X server uses for instantiating register regions ,
-   or must be passed in a new ioctl. */
-#define GAMMA_REG(reg)						   \
-	(2							   \
-	 + ((reg < 0x1000)					   \
-	    ? 0							   \
-	    : ((reg < 0x10000) ? 1 : ((reg < 0x11000) ? 2 : 3))))
-
-#define GAMMA_OFF(reg)						   \
-	((reg < 0x1000)						   \
-	 ? reg							   \
-	 : ((reg < 0x10000)					   \
-	    ? (reg - 0x1000)					   \
-	    : ((reg < 0x11000)					   \
-	       ? (reg - 0x10000)				   \
-	       : (reg - 0x11000))))
-
-#define GAMMA_BASE(reg)	 ((unsigned long)dev->maplist[GAMMA_REG(reg)]->handle)
-#define GAMMA_ADDR(reg)	 (GAMMA_BASE(reg) + GAMMA_OFF(reg))
-#define GAMMA_DEREF(reg) *(__volatile__ int *)GAMMA_ADDR(reg)
-#define GAMMA_READ(reg)	 GAMMA_DEREF(reg)
-#define GAMMA_WRITE(reg,val) do { GAMMA_DEREF(reg) = val; } while (0)
-
-#define GAMMA_BROADCASTMASK    0x9378
-#define GAMMA_COMMANDINTENABLE 0x0c48
-#define GAMMA_DMAADDRESS       0x0028
-#define GAMMA_DMACOUNT	       0x0030
-#define GAMMA_FILTERMODE       0x8c00
-#define GAMMA_GCOMMANDINTFLAGS 0x0c50
-#define GAMMA_GCOMMANDMODE     0x0c40
-#define GAMMA_GCOMMANDSTATUS   0x0c60
-#define GAMMA_GDELAYTIMER      0x0c38
-#define GAMMA_GDMACONTROL      0x0060
-#define GAMMA_GINTENABLE       0x0808
-#define GAMMA_GINTFLAGS	       0x0810
-#define GAMMA_INFIFOSPACE      0x0018
-#define GAMMA_OUTFIFOWORDS     0x0020
-#define GAMMA_OUTPUTFIFO       0x2000
-#define GAMMA_SYNC	       0x8c40
-#define GAMMA_SYNC_TAG	       0x0188
 
 static inline void gamma_dma_dispatch(drm_device_t *dev, unsigned long address,
 				      unsigned long length)
 {
+	drm_gamma_private_t *dev_priv =
+		(drm_gamma_private_t *)dev->dev_private;
+
 	GAMMA_WRITE(GAMMA_DMAADDRESS, virt_to_phys((void *)address));
 	while (GAMMA_READ(GAMMA_GCOMMANDSTATUS) != 4)
 		;
@@ -115,6 +52,9 @@ static inline void gamma_dma_dispatch(drm_device_t *dev, unsigned long address,
 
 void gamma_dma_quiescent_single(drm_device_t *dev)
 {
+	drm_gamma_private_t *dev_priv =
+		(drm_gamma_private_t *)dev->dev_private;
+
 	while (GAMMA_READ(GAMMA_DMACOUNT))
 		;
 	while (GAMMA_READ(GAMMA_INFIFOSPACE) < 3)
@@ -122,7 +62,7 @@ void gamma_dma_quiescent_single(drm_device_t *dev)
 
 	GAMMA_WRITE(GAMMA_FILTERMODE, 1 << 10);
 	GAMMA_WRITE(GAMMA_SYNC, 0);
-	
+
 	do {
 		while (!GAMMA_READ(GAMMA_OUTFIFOWORDS))
 			;
@@ -131,6 +71,9 @@ void gamma_dma_quiescent_single(drm_device_t *dev)
 
 void gamma_dma_quiescent_dual(drm_device_t *dev)
 {
+	drm_gamma_private_t *dev_priv =
+		(drm_gamma_private_t *)dev->dev_private;
+
 	while (GAMMA_READ(GAMMA_DMACOUNT))
 		;
 	while (GAMMA_READ(GAMMA_INFIFOSPACE) < 3)
@@ -140,13 +83,13 @@ void gamma_dma_quiescent_dual(drm_device_t *dev)
 
 	GAMMA_WRITE(GAMMA_FILTERMODE, 1 << 10);
 	GAMMA_WRITE(GAMMA_SYNC, 0);
-	
+
 				/* Read from first MX */
 	do {
 		while (!GAMMA_READ(GAMMA_OUTFIFOWORDS))
 			;
 	} while (GAMMA_READ(GAMMA_OUTPUTFIFO) != GAMMA_SYNC_TAG);
-	
+
 				/* Read from second MX */
 	do {
 		while (!GAMMA_READ(GAMMA_OUTFIFOWORDS + 0x10000))
@@ -156,20 +99,28 @@ void gamma_dma_quiescent_dual(drm_device_t *dev)
 
 void gamma_dma_ready(drm_device_t *dev)
 {
+	drm_gamma_private_t *dev_priv =
+		(drm_gamma_private_t *)dev->dev_private;
+
 	while (GAMMA_READ(GAMMA_DMACOUNT))
 		;
 }
 
 static inline int gamma_dma_is_ready(drm_device_t *dev)
 {
+	drm_gamma_private_t *dev_priv =
+		(drm_gamma_private_t *)dev->dev_private;
+
 	return !GAMMA_READ(GAMMA_DMACOUNT);
 }
 
-static void gamma_dma_service(int irq, void *device, struct pt_regs *regs)
+void gamma_dma_service(int irq, void *device, struct pt_regs *regs)
 {
-	drm_device_t	 *dev = (drm_device_t *)device;
-	drm_device_dma_t *dma = dev->dma;
-	
+	drm_device_t	    *dev      = (drm_device_t *)device;
+	drm_device_dma_t    *dma      = dev->dma;
+	drm_gamma_private_t *dev_priv =
+		(drm_gamma_private_t *)dev->dev_private;
+
 	atomic_inc(&dev->counts[6]); /* _DRM_STAT_IRQ */
 	GAMMA_WRITE(GAMMA_GDELAYTIMER, 0xc350/2); /* 0x05S */
 	GAMMA_WRITE(GAMMA_GCOMMANDINTFLAGS, 8);
@@ -202,7 +153,7 @@ static int gamma_do_dma(drm_device_t *dev, int locked)
 #endif
 
 	if (test_and_set_bit(0, &dev->dma_flag)) return -EBUSY;
-	
+
 #if DRM_DMA_HISTOGRAM
 	dma_start = get_cycles();
 #endif
@@ -234,7 +185,7 @@ static int gamma_do_dma(drm_device_t *dev, int locked)
 		clear_bit(0, &dev->dma_flag);
 		return 0;
 	}
-	
+
 	if (!gamma_dma_is_ready(dev)) {
 		clear_bit(0, &dev->dma_flag);
 		return -EBUSY;
@@ -258,13 +209,14 @@ static int gamma_do_dma(drm_device_t *dev, int locked)
 	    && !(dev->queuelist[buf->context]->flags
 		 & _DRM_CONTEXT_PRESERVED)) {
 				/* PRE: dev->last_context != buf->context */
-		if (drm_context_switch(dev, dev->last_context, buf->context)) {
-			gamma_clear_next_buffer(dev);
-			gamma_free_buffer(dev, buf);
+		if (DRM(context_switch)(dev, dev->last_context,
+					buf->context)) {
+			DRM(clear_next_buffer)(dev);
+			DRM(free_buffer)(dev, buf);
 		}
 		retcode = -EBUSY;
 		goto cleanup;
-			
+
 				/* POST: we will wait for the context
 				   switch and will dispatch on a later call
 				   when dev->last_context == buf->context.
@@ -305,12 +257,12 @@ cleanup:
 	return retcode;
 }
 
-static void gamma_dma_schedule_timer_wrapper(unsigned long dev)
+static void gamma_dma_timer_bh(unsigned long dev)
 {
 	gamma_dma_schedule((drm_device_t *)dev, 0);
 }
 
-static void gamma_dma_schedule_tq_wrapper(void *dev)
+void gamma_dma_immediate_bh(void *dev)
 {
 	gamma_dma_schedule(dev, 0);
 }
@@ -354,8 +306,7 @@ again:
 		if (!(retcode = gamma_do_dma(dev, locked))) ++processed;
 	} else {
 		do {
-			next = gamma_select_queue(dev,
-					     gamma_dma_schedule_timer_wrapper);
+			next = gamma_select_queue(dev, gamma_dma_timer_bh);
 			if (next >= 0) {
 				q   = dev->queuelist[next];
 				buf = gamma_waitlist_get(&q->waitlist);
@@ -383,9 +334,9 @@ again:
 			goto again;
 		}
 	}
-	
+
 	clear_bit(0, &dev->interrupt_flag);
-	
+
 #if DRM_DMA_HISTOGRAM
 	atomic_inc(&dev->histo.schedule[gamma_histogram_slot(get_cycles()
 							   - schedule_start)]);
@@ -472,15 +423,15 @@ static int gamma_dma_priority(drm_device_t *dev, drm_dma_t *d)
 			goto cleanup;
 		}
 		buf->pending = 1;
-		
+
 		if (dev->last_context != buf->context
 		    && !(dev->queuelist[buf->context]->flags
 			 & _DRM_CONTEXT_PRESERVED)) {
 			add_wait_queue(&dev->context_wait, &entry);
 			current->state = TASK_INTERRUPTIBLE;
 				/* PRE: dev->last_context != buf->context */
-			drm_context_switch(dev, dev->last_context,
-					   buf->context);
+			DRM(context_switch)(dev, dev->last_context,
+					    buf->context);
 				/* POST: we will wait for the context
 				   switch and will dispatch on a later call
 				   when dev->last_context == buf->context.
@@ -507,7 +458,7 @@ static int gamma_dma_priority(drm_device_t *dev, drm_dma_t *d)
 		gamma_dma_dispatch(dev, address, length);
 		atomic_inc(&dev->counts[9]); /* _DRM_STAT_SPECIAL */
 		atomic_add(length, &dev->counts[8]); /* _DRM_STAT_PRIMARY */
-		
+
 		if (last_buf) {
 			gamma_free_buffer(dev, last_buf);
 		}
@@ -520,7 +471,7 @@ cleanup:
 		gamma_dma_ready(dev);
 		gamma_free_buffer(dev, last_buf);
 	}
-	
+
 	if (must_free && !dev->context_flag) {
 		if (gamma_lock_free(dev, &dev->lock.hw_lock->lock,
 				  DRM_KERNEL_CONTEXT)) {
@@ -542,15 +493,15 @@ static int gamma_dma_send_buffers(drm_device_t *dev, drm_dma_t *d)
 		last_buf = dma->buflist[d->send_indices[d->send_count-1]];
 		add_wait_queue(&last_buf->dma_wait, &entry);
 	}
-	
+
 	if ((retcode = gamma_dma_enqueue(dev, d))) {
 		if (d->flags & _DRM_DMA_BLOCK)
 			remove_wait_queue(&last_buf->dma_wait, &entry);
 		return retcode;
 	}
-	
+
 	gamma_dma_schedule(dev, 0);
-	
+
 	if (d->flags & _DRM_DMA_BLOCK) {
 		DRM_DEBUG("%d waiting\n", current->pid);
 		for (;;) {
@@ -618,7 +569,7 @@ int gamma_dma(struct inode *inode, struct file *filp, unsigned int cmd,
 	if (d.send_count) {
 		if (d.flags & _DRM_DMA_PRIORITY)
 			retcode = gamma_dma_priority(dev, &d);
-		else 
+		else
 			retcode = gamma_dma_send_buffers(dev, &d);
 	}
 
@@ -635,105 +586,3 @@ int gamma_dma(struct inode *inode, struct file *filp, unsigned int cmd,
 
 	return retcode;
 }
-
-int gamma_irq_install(drm_device_t *dev, int irq)
-{
-	int retcode;
-
-	if (!irq)     return -EINVAL;
-	
-	down(&dev->struct_sem);
-	if (dev->irq) {
-		up(&dev->struct_sem);
-		return -EBUSY;
-	}
-	dev->irq = irq;
-	up(&dev->struct_sem);
-	
-	DRM_DEBUG("%d\n", irq);
-
-	dev->context_flag     = 0;
-	dev->interrupt_flag   = 0;
-	dev->dma_flag	      = 0;
-	
-	dev->dma->next_buffer = NULL;
-	dev->dma->next_queue  = NULL;
-	dev->dma->this_buffer = NULL;
-
-	INIT_LIST_HEAD(&dev->tq.list);
-	dev->tq.sync	      = 0;
-	dev->tq.routine	      = gamma_dma_schedule_tq_wrapper;
-	dev->tq.data	      = dev;
-
-
-				/* Before installing handler */
-	GAMMA_WRITE(GAMMA_GCOMMANDMODE, 0);
-	GAMMA_WRITE(GAMMA_GDMACONTROL, 0);
-	
-				/* Install handler */
-	if ((retcode = request_irq(dev->irq,
-				   gamma_dma_service,
-				   0,
-				   dev->devname,
-				   dev))) {
-		down(&dev->struct_sem);
-		dev->irq = 0;
-		up(&dev->struct_sem);
-		return retcode;
-	}
-
-				/* After installing handler */
-	GAMMA_WRITE(GAMMA_GINTENABLE,	    0x2001);
-	GAMMA_WRITE(GAMMA_COMMANDINTENABLE, 0x0008);
-	GAMMA_WRITE(GAMMA_GDELAYTIMER,	   0x39090);
-	
-	return 0;
-}
-
-int gamma_irq_uninstall(drm_device_t *dev)
-{
-	int irq;
-
-	down(&dev->struct_sem);
-	irq	 = dev->irq;
-	dev->irq = 0;
-	up(&dev->struct_sem);
-	
-	if (!irq) return -EINVAL;
-	
-	DRM_DEBUG("%d\n", irq);
-	
-	GAMMA_WRITE(GAMMA_GDELAYTIMER,	    0);
-	GAMMA_WRITE(GAMMA_COMMANDINTENABLE, 0);
-	GAMMA_WRITE(GAMMA_GINTENABLE,	    0);
-	free_irq(irq, dev);
-
-	return 0;
-}
-
-
-int gamma_control(struct inode *inode, struct file *filp, unsigned int cmd,
-		  unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_control_t	ctl;
-	int		retcode;
-	
-	if (copy_from_user(&ctl, (drm_control_t *)arg, sizeof(ctl)))
-		return -EFAULT;
-	
-	switch (ctl.func) {
-	case DRM_INST_HANDLER:
-		if ((retcode = gamma_irq_install(dev, ctl.irq)))
-			return retcode;
-		break;
-	case DRM_UNINST_HANDLER:
-		if ((retcode = gamma_irq_uninstall(dev)))
-			return retcode;
-		break;
-	default:
-		return -EINVAL;
-	}
-	return 0;
-}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drm.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drm.c
deleted file mode 100644
index 4b12d8c47..000000000
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drm.c
+++ /dev/null
@@ -1,33 +0,0 @@
-#define __NO_VERSION__
-#include "gamma.h"
-#include "drmP.h"
-#include "gamma_drv.h"
-
-#define DRIVER_DEV_PRIV_T	drm_gamma_private_t
-#define DRIVER_AGP_BUFFER_MAP	dev_priv->buffers
-
-#include "drm_auth.h"
-
-#include "drm_bufs.h"
-
-#include "drm_dma.h"
-
-#include "drm_drawable.h"
-
-#include "drm_fops.h"
-
-#include "drm_init.h"
-
-#include "drm_ioctl.h"
-
-#include "drm_lists.h"
-
-#include "drm_lock.h"
-
-#include "drm_memory.h"
-
-#include "drm_proc.h"
-
-#include "drm_vm.h"
-
-#include "drm_stub.h"
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.c
index c24d5a33c..98916bc53 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.c
@@ -26,7 +26,7 @@
  *
  * Authors:
  *    Rickard E. (Rik) Faith <faith@valinux.com>
- *
+ *    Gareth Hughes <gareth@valinux.com>
  */
 
 #include <linux/config.h>
@@ -38,29 +38,15 @@
 
 #define DRIVER_NAME		"gamma"
 #define DRIVER_DESC		"3DLabs gamma"
-#define DRIVER_DATE		"20010215"
+#define DRIVER_DATE		"20010216"
 
 #define DRIVER_MAJOR		1
 #define DRIVER_MINOR		0
 #define DRIVER_PATCHLEVEL	0
 
-
 #define DRIVER_IOCTLS							  \
 	[DRM_IOCTL_NR(DRM_IOCTL_DMA)]	     = { gamma_dma,	  1, 0 }
 
-#define __HAVE_MTRR		1
-#define __HAVE_CTX_BITMAP	0
-#define __HAVE_PCI_DMA		1
-#define __HAVE_MULTIPLE_DMA_QUEUES	1
-#define __HAVE_DMA_FLUSH		1
-#define __HAVE_DMA_QUEUE		0
-#define __HAVE_DMA_SCHEDULE		1
-#define __HAVE_DMA_WAITQUEUE		1
-#define __HAVE_DMA_WAITLIST	1
-#define __HAVE_DMA_FREELIST	1
-#define __HAVE_DMA		1
-#define __HAVE_OLD_DMA		1
-#define __HAVE_DMA_IRQ		1
 
 #define __HAVE_COUNTERS		5
 #define __HAVE_COUNTER6		_DRM_STAT_IRQ
@@ -69,35 +55,19 @@
 #define __HAVE_COUNTER9		_DRM_STAT_SPECIAL
 #define __HAVE_COUNTER10	_DRM_STAT_MISSED
 
-#define __HAVE_DMA_READY	1
-#define DRIVER_DMA_READY()						\
-do {									\
-	gamma_dma_ready(dev);						\
-} while (0)
-
-#define __HAVE_DMA_QUIESCENT	1
-#define DRIVER_DMA_QUIESCENT()						\
-do {									\
-	/* FIXME ! */ 							\
-	gamma_dma_quiescent_dual(dev);					\
-	return 0;							\
-} while (0)
-
-#if 0
-#define __HAVE_DRIVER_RELEASE	1
-#define DRIVER_RELEASE() do {						\
-	gamma_reclaim_buffers( dev, priv->pid );			\
-	if ( dev->dev_private ) {					\
-		drm_gamma_private_t *dev_priv = dev->dev_private;	\
-		dev_priv->dispatch_status &= MGA_IN_DISPATCH;		\
-	}								\
-} while (0)
-#endif
-
-#if 0
-#define DRIVER_PRETAKEDOWN() do {					\
-	if ( dev->dev_private ) gamma_do_cleanup_dma( dev );		\
-} while (0)
-#endif
 
+#include "drm_auth.h"
+#include "drm_bufs.h"
+#include "drm_context.h"
+#include "drm_dma.h"
+#include "drm_drawable.h"
 #include "drm_drv.h"
+#include "drm_fops.h"
+#include "drm_init.h"
+#include "drm_ioctl.h"
+#include "drm_lists.h"
+#include "drm_lock.h"
+#include "drm_memory.h"
+#include "drm_proc.h"
+#include "drm_vm.h"
+#include "drm_stub.h"
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.h
index 1bfa26509..68b520700 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.h
@@ -11,11 +11,11 @@
  * 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
@@ -23,10 +23,10 @@
  * 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.
- * 
+ *
  * Authors:
  *    Rickard E. (Rik) Faith <faith@valinux.com>
- * 
+ *
  */
 
 #ifndef _GAMMA_DRV_H_
@@ -35,6 +35,10 @@
 
 typedef struct drm_gamma_private {
 	drm_map_t *buffers;
+	drm_map_t *mmio0;
+	drm_map_t *mmio1;
+	drm_map_t *mmio2;
+	drm_map_t *mmio3;
 } drm_gamma_private_t;
 
 #define LOCK_TEST_WITH_RETURN( dev )					\
@@ -47,30 +51,55 @@ do {									\
 	}								\
 } while (0)
 
+
 extern void gamma_dma_ready(drm_device_t *dev);
 extern void gamma_dma_quiescent_single(drm_device_t *dev);
 extern void gamma_dma_quiescent_dual(drm_device_t *dev);
-				/* gamma_drv.c */
-extern int  gamma_version(struct inode *inode, struct file *filp,
-			  unsigned int cmd, unsigned long arg);
-extern int  gamma_open(struct inode *inode, struct file *filp);
-extern int  gamma_release(struct inode *inode, struct file *filp);
-extern int  gamma_ioctl(struct inode *inode, struct file *filp,
-			unsigned int cmd, unsigned long arg);
-extern int  gamma_lock(struct inode *inode, struct file *filp,
-		       unsigned int cmd, unsigned long arg);
-extern int  gamma_unlock(struct inode *inode, struct file *filp,
-			 unsigned int cmd, unsigned long arg);
 
 				/* gamma_dma.c */
 extern int  gamma_dma_schedule(drm_device_t *dev, int locked);
 extern int  gamma_dma(struct inode *inode, struct file *filp,
 		      unsigned int cmd, unsigned long arg);
-extern int  gamma_irq_install(drm_device_t *dev, int irq);
-extern int  gamma_irq_uninstall(drm_device_t *dev);
-extern int  gamma_control(struct inode *inode, struct file *filp,
-			  unsigned int cmd, unsigned long arg);
 extern int  gamma_find_devices(void);
 extern int  gamma_found(void);
 
+
+#define GAMMA_OFF(reg)						   \
+	((reg < 0x1000)						   \
+	 ? reg							   \
+	 : ((reg < 0x10000)					   \
+	    ? (reg - 0x1000)					   \
+	    : ((reg < 0x11000)					   \
+	       ? (reg - 0x10000)				   \
+	       : (reg - 0x11000))))
+
+#define GAMMA_BASE(reg)	 ((unsigned long)				     \
+			  ((reg < 0x1000)    ? dev_priv->mmio0->handle :     \
+			   ((reg < 0x10000)  ? dev_priv->mmio1->handle :     \
+			    ((reg < 0x11000) ? dev_priv->mmio2->handle :     \
+					       dev_priv->mmio3->handle))))
+	
+#define GAMMA_ADDR(reg)	 (GAMMA_BASE(reg) + GAMMA_OFF(reg))
+#define GAMMA_DEREF(reg) *(__volatile__ int *)GAMMA_ADDR(reg)
+#define GAMMA_READ(reg)	 GAMMA_DEREF(reg)
+#define GAMMA_WRITE(reg,val) do { GAMMA_DEREF(reg) = val; } while (0)
+
+#define GAMMA_BROADCASTMASK    0x9378
+#define GAMMA_COMMANDINTENABLE 0x0c48
+#define GAMMA_DMAADDRESS       0x0028
+#define GAMMA_DMACOUNT	       0x0030
+#define GAMMA_FILTERMODE       0x8c00
+#define GAMMA_GCOMMANDINTFLAGS 0x0c50
+#define GAMMA_GCOMMANDMODE     0x0c40
+#define GAMMA_GCOMMANDSTATUS   0x0c60
+#define GAMMA_GDELAYTIMER      0x0c38
+#define GAMMA_GDMACONTROL      0x0060
+#define GAMMA_GINTENABLE       0x0808
+#define GAMMA_GINTFLAGS	       0x0810
+#define GAMMA_INFIFOSPACE      0x0018
+#define GAMMA_OUTFIFOWORDS     0x0020
+#define GAMMA_OUTPUTFIFO       0x2000
+#define GAMMA_SYNC	       0x8c40
+#define GAMMA_SYNC_TAG	       0x0188
+
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810.h
index 021d7988c..ed25b2a99 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810.h
@@ -34,17 +34,79 @@
  */
 #define DRM(x) i810_##x
 
+/* General customization:
+ */
 #define __HAVE_AGP		1
 #define __MUST_HAVE_AGP		1
-
 #define __HAVE_MTRR		1
-
 #define __HAVE_CTX_BITMAP	1
 
+/* Driver customization:
+ */
+#define __HAVE_RELEASE		1
+#define DRIVER_RELEASE() do {						\
+	i810_reclaim_buffers( dev, priv->pid );				\
+} while (0)
+
+/* DMA customization:
+ */
 #define __HAVE_DMA		1
-#define __HAVE_DMA_IRQ		1
 #define __HAVE_DMA_QUEUE	1
 #define __HAVE_DMA_WAITLIST	1
 #define __HAVE_DMA_RECLAIM	1
 
+#define __HAVE_DMA_QUIESCENT	1
+#define DRIVER_DMA_QUIESCENT() do {					\
+	i810_dma_quiescent( dev );					\
+} while (0)
+
+#define __HAVE_DMA_IRQ		1
+#define __HAVE_DMA_IRQ_BH	1
+#define DRIVER_PREINSTALL() do {					\
+	drm_i810_private_t *dev_priv =					\
+				(drm_i810_private_t *)dev->dev_private;	\
+	u16 tmp;							\
+   	tmp = I810_READ16( I810REG_HWSTAM );				\
+   	tmp = tmp & 0x6000;						\
+   	I810_WRITE16( I810REG_HWSTAM, tmp );				\
+									\
+      	tmp = I810_READ16( I810REG_INT_MASK_R );			\
+   	tmp = tmp & 0x6000;		/* Unmask interrupts */		\
+   	I810_WRITE16( I810REG_INT_MASK_R, tmp );			\
+   	tmp = I810_READ16( I810REG_INT_ENABLE_R );			\
+   	tmp = tmp & 0x6000;		/* Disable all interrupts */	\
+      	I810_WRITE16( I810REG_INT_ENABLE_R, tmp );			\
+} while (0)
+
+#define DRIVER_POSTINSTALL() do {					\
+	drm_i810_private_t *dev_priv =					\
+				(drm_i810_private_t *)dev->dev_private;	\
+	u16 tmp;							\
+   	tmp = I810_READ16( I810REG_INT_ENABLE_R );			\
+   	tmp = tmp & 0x6000;						\
+   	tmp = tmp | 0x0003;	/* Enable bp & user interrupts */	\
+   	I810_WRITE16( I810REG_INT_ENABLE_R, tmp );			\
+} while (0)
+
+#define DRIVER_UNINSTALL() do {						\
+	drm_i810_private_t *dev_priv =					\
+				(drm_i810_private_t *)dev->dev_private;	\
+	u16 tmp;							\
+   	tmp = I810_READ16( I810REG_INT_IDENTITY_R );			\
+   	tmp = tmp & ~(0x6000);		/* Clear all interrupts */	\
+   	if ( tmp != 0 ) I810_WRITE16( I810REG_INT_IDENTITY_R, tmp );	\
+									\
+   	tmp = I810_READ16( I810REG_INT_ENABLE_R );			\
+   	tmp = tmp & 0x6000;		/* Disable all interrupts */	\
+   	I810_WRITE16( I810REG_INT_ENABLE_R, tmp );			\
+} while (0)
+
+/* Buffer customization:
+ */
+
+#define DRIVER_BUF_PRIV_T	drm_i810_buf_priv_t
+
+#define DRIVER_AGP_BUFFERS_MAP( dev )					\
+	((drm_i810_private_t *)((dev)->dev_private))->buffer_map
+
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_dma.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_dma.c
index 8c50b3204..4c90496a7 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_dma.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_dma.c
@@ -48,16 +48,6 @@
 #define I810_BUF_UNMAPPED 0
 #define I810_BUF_MAPPED   1
 
-#define I810_BASE(reg)		((unsigned long) \
-				dev_priv->mmio_map->handle)
-#define I810_ADDR(reg)		(I810_BASE(reg) + reg)
-#define I810_DEREF(reg)		*(__volatile__ int *)I810_ADDR(reg)
-#define I810_READ(reg)		I810_DEREF(reg)
-#define I810_WRITE(reg,val) 	do { I810_DEREF(reg) = val; } while (0)
-#define I810_DEREF16(reg)	*(__volatile__ u16 *)I810_ADDR(reg)
-#define I810_READ16(reg)	I810_DEREF16(reg)
-#define I810_WRITE16(reg,val)	do { I810_DEREF16(reg) = val; } while (0)
-
 #define RING_LOCALS	unsigned int outring, ringmask; volatile char *virt;
 
 #define BEGIN_LP_RING(n) do {				\
@@ -875,7 +865,7 @@ static void i810_dma_dispatch_vertex(drm_device_t *dev,
 
 
 /* Interrupts are only for flushing */
-static void i810_dma_service(int irq, void *device, struct pt_regs *regs)
+void i810_dma_service(int irq, void *device, struct pt_regs *regs)
 {
 	drm_device_t	 *dev = (drm_device_t *)device;
       	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
@@ -893,7 +883,7 @@ static void i810_dma_service(int irq, void *device, struct pt_regs *regs)
    	mark_bh(IMMEDIATE_BH);
 }
 
-static void i810_dma_task_queue(void *device)
+void i810_dma_immediate_bh(void *device)
 {
 	drm_device_t *dev = (drm_device_t *) device;
       	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
@@ -902,131 +892,6 @@ static void i810_dma_task_queue(void *device)
    	wake_up_interruptible(&dev_priv->flush_queue);
 }
 
-int i810_irq_install(drm_device_t *dev, int irq)
-{
-      	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
-	int retcode;
-	u16 temp;
-
-	if (!irq)     return -EINVAL;
-
-	down(&dev->struct_sem);
-	if (dev->irq) {
-		up(&dev->struct_sem);
-		return -EBUSY;
-	}
-	dev->irq = irq;
-	up(&dev->struct_sem);
-
-   	DRM_DEBUG(  "Interrupt Install : %d\n", irq);
-	DRM_DEBUG("%d\n", irq);
-
-	dev->context_flag     = 0;
-	dev->interrupt_flag   = 0;
-	dev->dma_flag	      = 0;
-
-	dev->dma->next_buffer = NULL;
-	dev->dma->next_queue  = NULL;
-	dev->dma->this_buffer = NULL;
-
-	INIT_LIST_HEAD(&dev->tq.list);
-	dev->tq.sync	      = 0;
-	dev->tq.routine	      = i810_dma_task_queue;
-	dev->tq.data	      = dev;
-
-				/* Before installing handler */
-   	temp = I810_READ16(I810REG_HWSTAM);
-   	temp = temp & 0x6000;
-   	I810_WRITE16(I810REG_HWSTAM, temp);
-
-      	temp = I810_READ16(I810REG_INT_MASK_R);
-   	temp = temp & 0x6000;
-   	I810_WRITE16(I810REG_INT_MASK_R, temp); /* Unmask interrupts */
-   	temp = I810_READ16(I810REG_INT_ENABLE_R);
-   	temp = temp & 0x6000;
-      	I810_WRITE16(I810REG_INT_ENABLE_R, temp); /* Disable all interrupts */
-
-				/* Install handler */
-	if ((retcode = request_irq(dev->irq,
-				   i810_dma_service,
-				   SA_SHIRQ,
-				   dev->devname,
-				   dev))) {
-		down(&dev->struct_sem);
-		dev->irq = 0;
-		up(&dev->struct_sem);
-		return retcode;
-	}
-   	temp = I810_READ16(I810REG_INT_ENABLE_R);
-   	temp = temp & 0x6000;
-   	temp = temp | 0x0003;
-   	I810_WRITE16(I810REG_INT_ENABLE_R,
-		     temp); /* Enable bp & user interrupts */
-	return 0;
-}
-
-int i810_irq_uninstall(drm_device_t *dev)
-{
-      	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
-	int irq;
-   	u16 temp;
-
-
-/*  	return 0; */
-
-	down(&dev->struct_sem);
-	irq	 = dev->irq;
-	dev->irq = 0;
-	up(&dev->struct_sem);
-
-	if (!irq) return -EINVAL;
-
-   	DRM_DEBUG(  "Interrupt UnInstall: %d\n", irq);
-	DRM_DEBUG("%d\n", irq);
-
-   	temp = I810_READ16(I810REG_INT_IDENTITY_R);
-   	temp = temp & ~(0x6000);
-   	if(temp != 0) I810_WRITE16(I810REG_INT_IDENTITY_R,
-				   temp); /* Clear all interrupts */
-
-   	temp = I810_READ16(I810REG_INT_ENABLE_R);
-   	temp = temp & 0x6000;
-   	I810_WRITE16(I810REG_INT_ENABLE_R,
-		     temp);                     /* Disable all interrupts */
-
-   	free_irq(irq, dev);
-
-	return 0;
-}
-
-int i810_control(struct inode *inode, struct file *filp, unsigned int cmd,
-		  unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_control_t	ctl;
-	int		retcode;
-
-   	DRM_DEBUG(  "i810_control\n");
-
-	if (copy_from_user(&ctl, (drm_control_t *)arg, sizeof(ctl)))
-		return -EFAULT;
-
-	switch (ctl.func) {
-	case DRM_INST_HANDLER:
-		if ((retcode = i810_irq_install(dev, ctl.irq)))
-			return retcode;
-		break;
-	case DRM_UNINST_HANDLER:
-		if ((retcode = i810_irq_uninstall(dev)))
-			return retcode;
-		break;
-	default:
-		return -EINVAL;
-	}
-	return 0;
-}
-
 static inline void i810_dma_emit_flush(drm_device_t *dev)
 {
    	drm_i810_private_t *dev_priv = dev->dev_private;
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.c
index 86ab394d9..12a59dbf3 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.c
@@ -56,42 +56,26 @@
    	[DRM_IOCTL_NR(DRM_IOCTL_I810_COPY)]   = { i810_copybuf,     1, 0 }, \
    	[DRM_IOCTL_NR(DRM_IOCTL_I810_DOCOPY)] = { i810_docopy,      1, 0 },
 
+
 #define __HAVE_COUNTERS         4
 #define __HAVE_COUNTER6         _DRM_STAT_IRQ
 #define __HAVE_COUNTER7         _DRM_STAT_PRIMARY
 #define __HAVE_COUNTER8         _DRM_STAT_SECONDARY
 #define __HAVE_COUNTER9         _DRM_STAT_DMA
 
-#define __HAVE_DMA_QUIESCENT	1
-#define DRIVER_DMA_QUIESCENT() do {					\
-	i810_dma_quiescent( dev );					\
-} while (0)
-
-#define __HAVE_RELEASE		1
-#define DRIVER_RELEASE() do {						\
-	i810_reclaim_buffers( dev, priv->pid );				\
-} while (0)
-
-#include "drm_drv.h"
-
-
-#define DRIVER_BUF_PRIV_T	drm_i810_buf_priv_t
-
-#define DRIVER_AGP_BUFFERS_MAP( dev )					\
-	((drm_i810_private_t *)((dev)->dev_private))->buffer_map
-
-#include "drm_bufs.h"
-
 
 #include "drm_agpsupport.h"
 #include "drm_auth.h"
+#include "drm_bufs.h"
 #include "drm_context.h"
 #include "drm_dma.h"
 #include "drm_drawable.h"
+#include "drm_drv.h"
 #include "drm_fops.h"
 #include "drm_init.h"
 #include "drm_ioctl.h"
 #include "drm_lock.h"
+#include "drm_lists.h"
 #include "drm_memory.h"
 #include "drm_proc.h"
 #include "drm_vm.h"
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.h
index 1f6e14680..e1b17148e 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.h
@@ -81,12 +81,6 @@ typedef struct drm_i810_private {
 extern int  i810_dma_schedule(drm_device_t *dev, int locked);
 extern int  i810_getbuf(struct inode *inode, struct file *filp,
 			unsigned int cmd, unsigned long arg);
-extern int  i810_irq_install(drm_device_t *dev, int irq);
-extern int  i810_irq_uninstall(drm_device_t *dev);
-extern int  i810_control(struct inode *inode, struct file *filp,
-			  unsigned int cmd, unsigned long arg);
-extern int  i810_lock(struct inode *inode, struct file *filp,
-		       unsigned int cmd, unsigned long arg);
 extern int  i810_dma_init(struct inode *inode, struct file *filp,
 			  unsigned int cmd, unsigned long arg);
 extern int  i810_flush_ioctl(struct inode *inode, struct file *filp,
@@ -114,6 +108,18 @@ int i810_swap_bufs(struct inode *inode, struct file *filp,
 int i810_clear_bufs(struct inode *inode, struct file *filp,
 		    unsigned int cmd, unsigned long arg);
 
+
+#define I810_BASE(reg)		((unsigned long) \
+				dev_priv->mmio_map->handle)
+#define I810_ADDR(reg)		(I810_BASE(reg) + reg)
+#define I810_DEREF(reg)		*(__volatile__ int *)I810_ADDR(reg)
+#define I810_READ(reg)		I810_DEREF(reg)
+#define I810_WRITE(reg,val) 	do { I810_DEREF(reg) = val; } while (0)
+#define I810_DEREF16(reg)	*(__volatile__ u16 *)I810_ADDR(reg)
+#define I810_READ16(reg)	I810_DEREF16(reg)
+#define I810_WRITE16(reg,val)	do { I810_DEREF16(reg) = val; } while (0)
+
+
 #define GFX_OP_USER_INTERRUPT 		((0<<29)|(2<<23))
 #define GFX_OP_BREAKPOINT_INTERRUPT	((0<<29)|(1<<23))
 #define CMD_REPORT_HEAD			(7<<23)
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga.h
index 9b2224048..277aa2029 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga.h
@@ -34,13 +34,34 @@
  */
 #define DRM(x) mga_##x
 
+/* General customization:
+ */
 #define __HAVE_AGP		1
 #define __MUST_HAVE_AGP		1
-
 #define __HAVE_MTRR		1
-
 #define __HAVE_CTX_BITMAP	1
 
+/* Driver customization:
+ */
+#define DRIVER_PRETAKEDOWN() do {					\
+	if ( dev->dev_private ) mga_do_cleanup_dma( dev );		\
+} while (0)
+
+/* DMA customization:
+ */
 #define __HAVE_DMA		1
 
+#define __HAVE_DMA_QUIESCENT	1
+#define DRIVER_DMA_QUIESCENT() do {					\
+	drm_mga_private_t *dev_priv = dev->dev_private;			\
+	return mga_do_wait_for_idle( dev_priv );			\
+} while (0)
+
+/* Buffer customization:
+ */
+#define DRIVER_BUF_PRIV_T	drm_mga_buf_priv_t
+
+#define DRIVER_AGP_BUFFERS_MAP( dev )					\
+	((drm_mga_private_t *)((dev)->dev_private))->buffers
+
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.c
index 275495dc6..21e14f376 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.c
@@ -57,7 +57,6 @@ int mga_do_wait_for_idle( drm_mga_private_t *dev_priv )
 	for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {
 		status = MGA_READ( MGA_STATUS ) & MGA_ENGINE_IDLE_MASK;
 		if ( status == MGA_ENDPRDMASTS ) {
-			/* printk("I'm not busy, flushing pixel cache\n"); */
 			MGA_WRITE8( MGA_CRTC_INDEX, 0 );
 			return 0;
 		}
@@ -273,7 +272,8 @@ static void mga_freelist_print( drm_device_t *dev )
 	DRM_INFO( "\n" );
 	DRM_INFO( "current dispatch: last=0x%x done=0x%x\n",
 		  dev_priv->sarea_priv->last_dispatch,
-		  *dev_priv->prim.head - dev_priv->primary->offset );
+		  (unsigned int)(*dev_priv->prim.head - 
+				 dev_priv->primary->offset) );
 	DRM_INFO( "current freelist:\n" );
 
 	for ( entry = dev_priv->head->next ; entry ; entry = entry->next ) {
@@ -354,7 +354,6 @@ static void mga_freelist_cleanup( drm_device_t *dev )
 static void mga_freelist_reset( drm_device_t *dev )
 {
 	drm_device_dma_t *dma = dev->dma;
-	drm_mga_private_t *dev_priv = dev->dev_private;
 	drm_buf_t *buf;
 	drm_mga_buf_priv_t *buf_priv;
 	int i;
@@ -400,7 +399,7 @@ static drm_buf_t *mga_freelist_get( drm_device_t *dev )
 	return NULL;
 }
 
-static int mga_freelist_put( drm_device_t *dev, drm_buf_t *buf )
+int mga_freelist_put( drm_device_t *dev, drm_buf_t *buf )
 {
 	drm_mga_private_t *dev_priv = dev->dev_private;
 	drm_mga_buf_priv_t *buf_priv = buf->dev_private;
@@ -412,15 +411,17 @@ static int mga_freelist_put( drm_device_t *dev, drm_buf_t *buf )
 		   dev_priv->primary->offset,
 		   buf_priv->list_entry->age.wrap );
 
-	if ( buf_priv->list_entry->age.head == MGA_BUFFER_USED ) {
-		SET_AGE( &next->age, MGA_BUFFER_FREE, 0 );
-	}
-
 	/* Put buffer on the head + 1, as the head is a sentinal.
 	 */
+
 	next = buf_priv->list_entry;
 	head = dev_priv->head;
 	prev = head->next;
+
+	if ( buf_priv->list_entry->age.head == MGA_BUFFER_USED ) {
+		SET_AGE( &next->age, MGA_BUFFER_FREE, 0 );
+	}
+
 	head->next = next;
 	prev->prev = next;
 	next->prev = head;
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.c
index 7898c744d..88fa53a6d 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.c
@@ -38,7 +38,7 @@
 
 #define DRIVER_NAME		"mga"
 #define DRIVER_DESC		"Matrox G200/G400"
-#define DRIVER_DATE		"20010215"
+#define DRIVER_DATE		"20010216"
 
 #define DRIVER_MAJOR		3
 #define DRIVER_MINOR		0
@@ -56,38 +56,20 @@
 	[DRM_IOCTL_NR(DRM_IOCTL_MGA_ILOAD)]   = { mga_dma_iload,   1, 0 }, \
 	[DRM_IOCTL_NR(DRM_IOCTL_MGA_BLIT)]    = { mga_dma_blit,    1, 0 },
 
+
 #define __HAVE_COUNTERS         3
 #define __HAVE_COUNTER6         _DRM_STAT_IRQ
 #define __HAVE_COUNTER7         _DRM_STAT_PRIMARY
 #define __HAVE_COUNTER8         _DRM_STAT_SECONDARY
 
-#define __HAVE_DMA_QUIESCENT	1
-#define DRIVER_DMA_QUIESCENT()						\
-do {									\
-	drm_mga_private_t *dev_priv = dev->dev_private;			\
-	return mga_do_wait_for_idle( dev_priv );			\
-} while (0)
-
-#define DRIVER_PRETAKEDOWN() do {					\
-	if ( dev->dev_private ) mga_do_cleanup_dma( dev );		\
-} while (0)
-
-#include "drm_drv.h"
-
-
-#define DRIVER_BUF_PRIV_T	drm_mga_buf_priv_t
-
-#define DRIVER_AGP_BUFFERS_MAP( dev )					\
-	((drm_mga_private_t *)((dev)->dev_private))->buffers
-
-#include "drm_bufs.h"
-
 
 #include "drm_agpsupport.h"
 #include "drm_auth.h"
+#include "drm_bufs.h"
 #include "drm_context.h"
 #include "drm_dma.h"
 #include "drm_drawable.h"
+#include "drm_drv.h"
 #include "drm_fops.h"
 #include "drm_init.h"
 #include "drm_ioctl.h"
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.h
index ab6d9edbb..f56186e1e 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.h
@@ -102,18 +102,6 @@ typedef struct drm_mga_private {
 	drm_map_t *agp_textures;
 } drm_mga_private_t;
 
-				/* mga_drv.c */
-extern int mga_version( struct inode *inode, struct file *filp,
-			unsigned int cmd, unsigned long arg );
-extern int mga_open( struct inode *inode, struct file *filp );
-extern int mga_release( struct inode *inode, struct file *filp );
-extern int mga_ioctl( struct inode *inode, struct file *filp,
-		      unsigned int cmd, unsigned long arg );
-extern int mga_lock( struct inode *inode, struct file *filp,
-		     unsigned int cmd, unsigned long arg );
-extern int mga_unlock( struct inode *inode, struct file *filp,
-		       unsigned int cmd, unsigned long arg );
-
 				/* mga_dma.c */
 extern int mga_dma_init( struct inode *inode, struct file *filp,
 			 unsigned int cmd, unsigned long arg );
@@ -121,8 +109,6 @@ extern int mga_dma_flush( struct inode *inode, struct file *filp,
 			  unsigned int cmd, unsigned long arg );
 extern int mga_dma_reset( struct inode *inode, struct file *filp,
 			  unsigned int cmd, unsigned long arg );
-extern int mga_control( struct inode *inode, struct file *filp,
-			unsigned int cmd, unsigned long arg );
 extern int mga_dma_buffers( struct inode *inode, struct file *filp,
 			    unsigned int cmd, unsigned long arg );
 
@@ -136,7 +122,8 @@ extern void mga_do_dma_flush( drm_mga_private_t *dev_priv );
 extern void mga_do_dma_wrap_start( drm_mga_private_t *dev_priv );
 extern void mga_do_dma_wrap_end( drm_mga_private_t *dev_priv );
 
-extern int mga_irq_uninstall( drm_device_t *dev );
+extern int mga_freelist_put( drm_device_t *dev, drm_buf_t *buf );
+
 
 				/* mga_state.c */
 extern int  mga_dma_clear( struct inode *inode, struct file *filp,
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128.h
index 0440501ee..83e002afa 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128.h
@@ -34,13 +34,46 @@
  */
 #define DRM(x) r128_##x
 
+/* General customization:
+ */
 #define __HAVE_AGP		1
 #define __MUST_HAVE_AGP		1
-
 #define __HAVE_MTRR		1
-
 #define __HAVE_CTX_BITMAP	1
 
+/* Driver customization:
+ */
+#define DRIVER_PRERELEASE() do {					\
+	if ( dev->dev_private ) {					\
+		drm_r128_private_t *dev_priv = dev->dev_private;	\
+		if ( dev_priv->page_flipping ) {			\
+			r128_do_cleanup_pageflip( dev );		\
+		}							\
+	}								\
+} while (0)
+
+#define DRIVER_PRETAKEDOWN() do {					\
+	if ( dev->dev_private ) r128_do_cleanup_cce( dev );		\
+} while (0)
+
+/* DMA customization:
+ */
 #define __HAVE_DMA		1
 
+#if 0
+/* GH: Remove this for now... */
+#define __HAVE_DMA_QUIESCENT	1
+#define DRIVER_DMA_QUIESCENT() do {					\
+	drm_r128_private_t *dev_priv = dev->dev_private;		\
+	return r128_do_cce_idle( dev_priv );				\
+} while (0)
+#endif
+
+/* Buffer customization:
+ */
+#define DRIVER_BUF_PRIV_T	drm_r128_buf_priv_t
+
+#define DRIVER_AGP_BUFFERS_MAP( dev )					\
+	((drm_r128_private_t *)((dev)->dev_private))->buffers
+
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_cce.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_cce.c
index 5ffc1d0e2..f4b4f0c4a 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_cce.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_cce.c
@@ -25,7 +25,7 @@
  * DEALINGS IN THE SOFTWARE.
  *
  * Authors:
- *   Gareth Hughes <gareth@valinux.com>
+ *    Gareth Hughes <gareth@valinux.com>
  */
 
 #define __NO_VERSION__
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drm.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drm.h
index aaf2ce3e0..86aba175f 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drm.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drm.h
@@ -25,9 +25,8 @@
  * DEALINGS IN THE SOFTWARE.
  *
  * Authors:
- *    Kevin E. Martin <martin@valinux.com>
  *    Gareth Hughes <gareth@valinux.com>
- *
+ *    Kevin E. Martin <martin@valinux.com>
  */
 
 #ifndef __R128_DRM_H__
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.c
index 3bd27ae7d..e42868ed6 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.c
@@ -38,7 +38,7 @@
 
 #define DRIVER_NAME		"r128"
 #define DRIVER_DESC		"ATI Rage 128"
-#define DRIVER_DATE		"20010215"
+#define DRIVER_DATE		"20010216"
 
 #define DRIVER_MAJOR		2
 #define DRIVER_MINOR		1
@@ -62,6 +62,7 @@
    [DRM_IOCTL_NR(DRM_IOCTL_R128_STIPPLE)]    = { r128_cce_stipple,  1, 0 }, \
    [DRM_IOCTL_NR(DRM_IOCTL_R128_INDIRECT)]   = { r128_cce_indirect, 1, 1 },
 
+
 #if 0
 /* GH: Count data sent to card via ring or vertex/indirect buffers.
  */
@@ -71,41 +72,14 @@
 #define __HAVE_COUNTER8         _DRM_STAT_SECONDARY
 #endif
 
-#define __HAVE_DMA_QUIESCENT	1
-#define DRIVER_DMA_QUIESCENT() do {					\
-	drm_r128_private_t *dev_priv = dev->dev_private;		\
-	return r128_do_cce_idle( dev_priv );				\
-} while (0)
-
-#define DRIVER_PRERELEASE() do {					\
-	if ( dev->dev_private ) {					\
-		drm_r128_private_t *dev_priv = dev->dev_private;	\
-		if ( dev_priv->page_flipping ) {			\
-			r128_do_cleanup_pageflip( dev );		\
-		}							\
-	}								\
-} while (0)
-
-#define DRIVER_PRETAKEDOWN() do {					\
-	if ( dev->dev_private ) r128_do_cleanup_cce( dev );		\
-} while (0)
-
-#include "drm_drv.h"
-
-
-#define DRIVER_BUF_PRIV_T	drm_r128_buf_priv_t
-
-#define DRIVER_AGP_BUFFERS_MAP( dev )					\
-	((drm_r128_private_t *)((dev)->dev_private))->buffers
-
-#include "drm_bufs.h"
-
 
 #include "drm_agpsupport.h"
 #include "drm_auth.h"
+#include "drm_bufs.h"
 #include "drm_context.h"
 #include "drm_dma.h"
 #include "drm_drawable.h"
+#include "drm_drv.h"
 #include "drm_fops.h"
 #include "drm_init.h"
 #include "drm_ioctl.h"
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.h
index 950a4711b..9f53746f8 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.h
@@ -25,10 +25,9 @@
  * DEALINGS IN THE SOFTWARE.
  *
  * Authors:
- *   Rickard E. (Rik) Faith <faith@valinux.com>
- *   Kevin E. Martin <martin@valinux.com>
- *   Gareth Hughes <gareth@valinux.com>
- *
+ *    Rickard E. (Rik) Faith <faith@valinux.com>
+ *    Kevin E. Martin <martin@valinux.com>
+ *    Gareth Hughes <gareth@valinux.com>
  */
 
 #ifndef __R128_DRV_H__
@@ -440,8 +439,7 @@ do {									\
 	}								\
 } while (0)
 
-#define R128_WAIT_UNTIL_PAGE_FLIPPED()					\
-do {									\
+#define R128_WAIT_UNTIL_PAGE_FLIPPED() do {				\
 	OUT_RING( CCE_PACKET0( R128_WAIT_UNTIL, 0 ) );			\
 	OUT_RING( R128_EVENT_CRTC_OFFSET );				\
 } while (0)
@@ -456,7 +454,8 @@ do {									\
 
 #define R128_VERBOSE	0
 
-#define RING_LOCALS	int write; unsigned int tail_mask; volatile u32 *ring;
+#define RING_LOCALS							\
+	int write; unsigned int tail_mask; volatile u32 *ring;
 
 #define BEGIN_RING( n ) do {						\
 	if ( R128_VERBOSE ) {						\
@@ -472,11 +471,23 @@ do {									\
 	tail_mask = dev_priv->ring.tail_mask;				\
 } while (0)
 
+/* You can set this to zero if you want.  If the card locks up, you'll
+ * need to keep this set.  It works around a bug in early revs of the
+ * Rage 128 chipset, where the CCE would read 32 dwords past the end of
+ * the ring buffer before wrapping around.
+ */
+#define R128_BROKEN_CCE	1
+
 #define ADVANCE_RING() do {						\
 	if ( R128_VERBOSE ) {						\
 		DRM_INFO( "ADVANCE_RING() tail=0x%06x wr=0x%06x\n",	\
 			  write, dev_priv->ring.tail );			\
 	}								\
+	if ( R128_BROKEN_CCE && write < 32 ) {				\
+		memcpy( dev_priv->ring.end,				\
+			dev_priv->ring.start,				\
+			write * sizeof(u32) );				\
+	}								\
 	r128_flush_write_combine();					\
 	dev_priv->ring.tail = write;					\
 	R128_WRITE( R128_PM4_BUFFER_DL_WPTR, write );			\
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_state.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_state.c
index a4683c4e2..fa16f3ced 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_state.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_state.c
@@ -25,7 +25,6 @@
  *
  * Authors:
  *    Gareth Hughes <gareth@valinux.com>
- *
  */
 
 #define __NO_VERSION__
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon.h
index 6ef992775..db238b1b4 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon.h
@@ -34,13 +34,46 @@
  */
 #define DRM(x) radeon_##x
 
+/* General customization:
+ */
 #define __HAVE_AGP		1
 #define __MUST_HAVE_AGP		1
-
 #define __HAVE_MTRR		1
-
 #define __HAVE_CTX_BITMAP	1
 
+/* Driver customization:
+ */
+#define DRIVER_PRERELEASE() do {					\
+	if ( dev->dev_private ) {					\
+		drm_radeon_private_t *dev_priv = dev->dev_private;	\
+		if ( dev_priv->page_flipping ) {			\
+			radeon_do_cleanup_pageflip( dev );		\
+		}							\
+	}								\
+} while (0)
+
+#define DRIVER_PRETAKEDOWN() do {					\
+	if ( dev->dev_private ) radeon_do_cleanup_cp( dev );		\
+} while (0)
+
+/* DMA customization:
+ */
 #define __HAVE_DMA		1
 
+#if 0
+/* GH: Remove this for now... */
+#define __HAVE_DMA_QUIESCENT	1
+#define DRIVER_DMA_QUIESCENT() do {					\
+	drm_radeon_private_t *dev_priv = dev->dev_private;		\
+	return radeon_do_cp_idle( dev_priv );				\
+} while (0)
+#endif
+
+/* Buffer customization:
+ */
+#define DRIVER_BUF_PRIV_T	drm_radeon_buf_priv_t
+
+#define DRIVER_AGP_BUFFERS_MAP( dev )					\
+	((drm_radeon_private_t *)((dev)->dev_private))->buffers
+
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_cp.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_cp.c
index 28e64284e..c80665caf 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_cp.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_cp.c
@@ -24,9 +24,8 @@
  * DEALINGS IN THE SOFTWARE.
  *
  * Authors:
- *   Kevin E. Martin <martin@valinux.com>
- *   Gareth Hughes <gareth@valinux.com>
- *
+ *    Kevin E. Martin <martin@valinux.com>
+ *    Gareth Hughes <gareth@valinux.com>
  */
 
 #define __NO_VERSION__
@@ -844,11 +843,8 @@ int radeon_cp_start( struct inode *inode, struct file *filp,
 	drm_radeon_private_t *dev_priv = dev->dev_private;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
+
 	if ( dev_priv->cp_running ) {
 		DRM_DEBUG( "%s while CP running\n", __FUNCTION__ );
 		return 0;
@@ -877,11 +873,7 @@ int radeon_cp_stop( struct inode *inode, struct file *filp,
 	int ret;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
 
 	if ( copy_from_user( &stop, (drm_radeon_init_t *)arg, sizeof(stop) ) )
 		return -EFAULT;
@@ -923,11 +915,8 @@ int radeon_cp_reset( struct inode *inode, struct file *filp,
 	drm_radeon_private_t *dev_priv = dev->dev_private;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
+
 	if ( !dev_priv ) {
 		DRM_DEBUG( "%s called before init done\n", __FUNCTION__ );
 		return -EINVAL;
@@ -949,11 +938,7 @@ int radeon_cp_idle( struct inode *inode, struct file *filp,
 	drm_radeon_private_t *dev_priv = dev->dev_private;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
 
 	return radeon_do_cp_idle( dev_priv );
 }
@@ -965,11 +950,7 @@ int radeon_engine_reset( struct inode *inode, struct file *filp,
         drm_device_t *dev = priv->dev;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
 
 	return radeon_do_engine_reset( dev );
 }
@@ -1019,11 +1000,7 @@ int radeon_fullscreen( struct inode *inode, struct file *filp,
         drm_device_t *dev = priv->dev;
 	drm_radeon_fullscreen_t fs;
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
 
 	if ( copy_from_user( &fs, (drm_radeon_fullscreen_t *)arg,
 			     sizeof(fs) ) )
@@ -1247,14 +1224,10 @@ int radeon_cp_buffers( struct inode *inode, struct file *filp,
 	int ret = 0;
 	drm_dma_t d;
 
-	if ( copy_from_user( &d, (drm_dma_t *) arg, sizeof(d) ) )
-		return -EFAULT;
+	LOCK_TEST_WITH_RETURN( dev );
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	if ( copy_from_user( &d, (drm_dma_t *)arg, sizeof(d) ) )
+		return -EFAULT;
 
 	/* Please don't send us buffers.
 	 */
@@ -1278,7 +1251,7 @@ int radeon_cp_buffers( struct inode *inode, struct file *filp,
 		ret = radeon_cp_get_buffers( dev, &d );
 	}
 
-	if ( copy_to_user( (drm_dma_t *) arg, &d, sizeof(d) ) )
+	if ( copy_to_user( (drm_dma_t *)arg, &d, sizeof(d) ) )
 		return -EFAULT;
 
 	return ret;
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_drm.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_drm.h
index 643253d2f..50a7d6ed8 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_drm.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_drm.h
@@ -26,7 +26,6 @@
  * Authors:
  *    Kevin E. Martin <martin@valinux.com>
  *    Gareth Hughes <gareth@valinux.com>
- *
  */
 
 #ifndef __RADEON_DRM_H__
@@ -74,7 +73,7 @@
 
 /* Vertex/indirect buffer size
  */
-#define RADEON_BUFFER_SIZE		16384
+#define RADEON_BUFFER_SIZE		65536
 
 /* Byte offsets for indirect buffer data
  */
@@ -305,14 +304,20 @@ typedef struct drm_radeon_indices {
 	int discard;			/* Client finished with buffer? */
 } drm_radeon_indices_t;
 
-typedef struct drm_radeon_blit {
-	int idx;
-	int pitch;
+typedef struct drm_radeon_tex_image {
+	unsigned int x, y;		/* Blit coordinates */
+	unsigned int width, height;
+	const void *data;
+} drm_radeon_tex_image_t;
+
+typedef struct drm_radeon_texture {
 	int offset;
+	int pitch;
 	int format;
-	unsigned short x, y;
-	unsigned short width, height;
-} drm_radeon_blit_t;
+	int width;			/* Texture image coordinates */
+	int height;
+	drm_radeon_tex_image_t *image;
+} drm_radeon_texture_t;
 
 typedef struct drm_radeon_stipple {
 	unsigned int *mask;
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_drv.c
index 43f6e2e7b..1aa889aee 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_drv.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_drv.c
@@ -36,7 +36,7 @@
 
 #define DRIVER_NAME		"radeon"
 #define DRIVER_DESC		"ATI Radeon"
-#define DRIVER_DATE		"20010215"
+#define DRIVER_DATE		"20010305"
 
 #define DRIVER_MAJOR		1
 #define DRIVER_MINOR		0
@@ -55,10 +55,11 @@
  [DRM_IOCTL_NR(DRM_IOCTL_RADEON_CLEAR)]      = { radeon_cp_clear,    1, 0 }, \
  [DRM_IOCTL_NR(DRM_IOCTL_RADEON_VERTEX)]     = { radeon_cp_vertex,   1, 0 }, \
  [DRM_IOCTL_NR(DRM_IOCTL_RADEON_INDICES)]    = { radeon_cp_indices,  1, 0 }, \
- [DRM_IOCTL_NR(DRM_IOCTL_RADEON_BLIT)]       = { radeon_cp_blit,     1, 0 }, \
+ [DRM_IOCTL_NR(DRM_IOCTL_RADEON_TEXTURE)]    = { radeon_cp_texture,  1, 0 }, \
  [DRM_IOCTL_NR(DRM_IOCTL_RADEON_STIPPLE)]    = { radeon_cp_stipple,  1, 0 }, \
  [DRM_IOCTL_NR(DRM_IOCTL_RADEON_INDIRECT)]   = { radeon_cp_indirect, 1, 1 },
 
+
 #if 0
 /* GH: Count data sent to card via ring or vertex/indirect buffers.
  */
@@ -68,41 +69,14 @@
 #define __HAVE_COUNTER8         _DRM_STAT_SECONDARY
 #endif
 
-#define __HAVE_DMA_QUIESCENT	1
-#define DRIVER_DMA_QUIESCENT() do {					\
-	drm_radeon_private_t *dev_priv = dev->dev_private;		\
-	return radeon_do_cp_idle( dev_priv );				\
-} while (0)
-
-#define DRIVER_PRERELEASE() do {					\
-	if ( dev->dev_private ) {					\
-		drm_radeon_private_t *dev_priv = dev->dev_private;	\
-		if ( dev_priv->page_flipping ) {			\
-			radeon_do_cleanup_pageflip( dev );		\
-		}							\
-	}								\
-} while (0)
-
-#define DRIVER_PRETAKEDOWN() do {					\
-	if ( dev->dev_private ) radeon_do_cleanup_cp( dev );		\
-} while (0)
-
-#include "drm_drv.h"
-
-
-#define DRIVER_BUF_PRIV_T	drm_radeon_buf_priv_t
-
-#define DRIVER_AGP_BUFFERS_MAP( dev )					\
-	((drm_radeon_private_t *)((dev)->dev_private))->buffers
-
-#include "drm_bufs.h"
-
 
 #include "drm_agpsupport.h"
 #include "drm_auth.h"
+#include "drm_bufs.h"
 #include "drm_context.h"
 #include "drm_dma.h"
 #include "drm_drawable.h"
+#include "drm_drv.h"
 #include "drm_fops.h"
 #include "drm_init.h"
 #include "drm_ioctl.h"
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_drv.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_drv.h
index 656d6e03b..d279b0467 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_drv.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_drv.h
@@ -24,10 +24,8 @@
  * DEALINGS IN THE SOFTWARE.
  *
  * Authors:
- *   Rickard E. (Rik) Faith <faith@valinux.com>
- *   Kevin E. Martin <martin@valinux.com>
- *   Gareth Hughes <gareth@valinux.com>
- *
+ *    Kevin E. Martin <martin@valinux.com>
+ *    Gareth Hughes <gareth@valinux.com>
  */
 
 #ifndef __RADEON_DRV_H__
@@ -163,8 +161,8 @@ extern int radeon_cp_vertex( struct inode *inode, struct file *filp,
 			     unsigned int cmd, unsigned long arg );
 extern int radeon_cp_indices( struct inode *inode, struct file *filp,
 			      unsigned int cmd, unsigned long arg );
-extern int radeon_cp_blit( struct inode *inode, struct file *filp,
-			   unsigned int cmd, unsigned long arg );
+extern int radeon_cp_texture( struct inode *inode, struct file *filp,
+			      unsigned int cmd, unsigned long arg );
 extern int radeon_cp_stipple( struct inode *inode, struct file *filp,
 			      unsigned int cmd, unsigned long arg );
 extern int radeon_cp_indirect( struct inode *inode, struct file *filp,
@@ -480,14 +478,14 @@ extern int radeon_cp_indirect( struct inode *inode, struct file *filp,
 #define RADEON_COLOR_FORMAT_RGB8	9
 #define RADEON_COLOR_FORMAT_ARGB4444	15
 
-#define RADEON_TXF_8BPP_I		0
-#define RADEON_TXF_16BPP_AI88		1
-#define RADEON_TXF_8BPP_RGB332		2
-#define RADEON_TXF_16BPP_ARGB1555	3
-#define RADEON_TXF_16BPP_RGB565		4
-#define RADEON_TXF_16BPP_ARGB4444	5
-#define RADEON_TXF_32BPP_ARGB8888	6
-#define RADEON_TXF_32BPP_RGBA8888	7
+#define RADEON_TXFORMAT_I8		0
+#define RADEON_TXFORMAT_AI88		1
+#define RADEON_TXFORMAT_RGB332		2
+#define RADEON_TXFORMAT_ARGB1555	3
+#define RADEON_TXFORMAT_RGB565		4
+#define RADEON_TXFORMAT_ARGB4444	5
+#define RADEON_TXFORMAT_ARGB8888	6
+#define RADEON_TXFORMAT_RGBA8888	7
 
 /* Constants */
 #define RADEON_MAX_USEC_TIMEOUT		100000	/* 100 ms */
@@ -504,25 +502,23 @@ extern int radeon_cp_indirect( struct inode *inode, struct file *filp,
 
 
 #define RADEON_BASE(reg)	((u32)(dev_priv->mmio->handle))
-#define RADEON_ADDR(reg)	(RADEON_BASE(reg) + reg)
+#define RADEON_ADDR(reg)	(RADEON_BASE( reg ) + reg)
 
-#define RADEON_DEREF(reg)	*(volatile u32 *)RADEON_ADDR(reg)
-#define RADEON_READ(reg)	RADEON_DEREF(reg)
-#define RADEON_WRITE(reg,val)	do { RADEON_DEREF(reg) = val; } while (0)
+#define RADEON_DEREF(reg)	*(volatile u32 *)RADEON_ADDR( reg )
+#define RADEON_READ(reg)	RADEON_DEREF( reg )
+#define RADEON_WRITE(reg, val)	do { RADEON_DEREF( reg ) = val; } while (0)
 
-#define RADEON_DEREF8(reg)	*(volatile u8 *)RADEON_ADDR(reg)
-#define RADEON_READ8(reg)	RADEON_DEREF8(reg)
-#define RADEON_WRITE8(reg,val)	do { RADEON_DEREF8(reg) = val; } while (0)
+#define RADEON_DEREF8(reg)	*(volatile u8 *)RADEON_ADDR( reg )
+#define RADEON_READ8(reg)	RADEON_DEREF8( reg )
+#define RADEON_WRITE8(reg, val)	do { RADEON_DEREF8( reg ) = val; } while (0)
 
-#define RADEON_WRITE_PLL(addr,val)                                            \
-do {                                                                          \
-	RADEON_WRITE8(RADEON_CLOCK_CNTL_INDEX,                                \
-		      ((addr) & 0x1f) | RADEON_PLL_WR_EN);                    \
-	RADEON_WRITE(RADEON_CLOCK_CNTL_DATA, (val));                          \
+#define RADEON_WRITE_PLL( addr, val ) do {				\
+	RADEON_WRITE8( RADEON_CLOCK_CNTL_INDEX,				\
+		       ((addr) & 0x1f) | RADEON_PLL_WR_EN );		\
+	RADEON_WRITE( RADEON_CLOCK_CNTL_DATA, (val) );			\
 } while (0)
 
-extern int RADEON_READ_PLL(drm_device_t *dev, int addr);
-
+extern int RADEON_READ_PLL( drm_device_t *dev, int addr );
 
 
 #define CP_PACKET0( reg, n )						\
@@ -541,54 +537,46 @@ extern int RADEON_READ_PLL(drm_device_t *dev, int addr);
  * Engine control helper macros
  */
 
-#define RADEON_WAIT_UNTIL_2D_IDLE()					\
-do {									\
+#define RADEON_WAIT_UNTIL_2D_IDLE() do {				\
 	OUT_RING( CP_PACKET0( RADEON_WAIT_UNTIL, 0 ) );			\
 	OUT_RING( (RADEON_WAIT_2D_IDLECLEAN |				\
 		   RADEON_WAIT_HOST_IDLECLEAN) );			\
 } while (0)
 
-#define RADEON_WAIT_UNTIL_3D_IDLE()					\
-do {									\
+#define RADEON_WAIT_UNTIL_3D_IDLE() do {				\
 	OUT_RING( CP_PACKET0( RADEON_WAIT_UNTIL, 0 ) );			\
 	OUT_RING( (RADEON_WAIT_3D_IDLECLEAN |				\
 		   RADEON_WAIT_HOST_IDLECLEAN) );			\
 } while (0)
 
-#define RADEON_WAIT_UNTIL_IDLE()					\
-do {									\
+#define RADEON_WAIT_UNTIL_IDLE() do {					\
 	OUT_RING( CP_PACKET0( RADEON_WAIT_UNTIL, 0 ) );			\
 	OUT_RING( (RADEON_WAIT_2D_IDLECLEAN |				\
 		   RADEON_WAIT_3D_IDLECLEAN |				\
 		   RADEON_WAIT_HOST_IDLECLEAN) );			\
 } while (0)
 
-#define RADEON_WAIT_UNTIL_PAGE_FLIPPED()				\
-do {									\
+#define RADEON_WAIT_UNTIL_PAGE_FLIPPED() do {				\
 	OUT_RING( CP_PACKET0( RADEON_WAIT_UNTIL, 0 ) );			\
 	OUT_RING( RADEON_WAIT_CRTC_PFLIP );				\
 } while (0)
 
-#define RADEON_FLUSH_CACHE()						\
-do {									\
+#define RADEON_FLUSH_CACHE() do {					\
 	OUT_RING( CP_PACKET0( RADEON_RB2D_DSTCACHE_CTLSTAT, 0 ) );	\
 	OUT_RING( RADEON_RB2D_DC_FLUSH );				\
 } while (0)
 
-#define RADEON_PURGE_CACHE()						\
-do {									\
+#define RADEON_PURGE_CACHE() do {					\
 	OUT_RING( CP_PACKET0( RADEON_RB2D_DSTCACHE_CTLSTAT, 0 ) );	\
 	OUT_RING( RADEON_RB2D_DC_FLUSH_ALL );				\
 } while (0)
 
-#define RADEON_FLUSH_ZCACHE()						\
-do {									\
+#define RADEON_FLUSH_ZCACHE() do {					\
 	OUT_RING( CP_PACKET0( RADEON_RB3D_ZCACHE_CTLSTAT, 0 ) );	\
 	OUT_RING( RADEON_RB3D_ZC_FLUSH );				\
 } while (0)
 
-#define RADEON_PURGE_ZCACHE()						\
-do {									\
+#define RADEON_PURGE_ZCACHE() do {					\
 	OUT_RING( CP_PACKET0( RADEON_RB3D_ZCACHE_CTLSTAT, 0 ) );	\
 	OUT_RING( RADEON_RB3D_ZC_FLUSH_ALL );				\
 } while (0)
@@ -598,6 +586,16 @@ do {									\
  * Misc helper macros
  */
 
+#define LOCK_TEST_WITH_RETURN( dev )					\
+do {									\
+	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||		\
+	     dev->lock.pid != current->pid ) {				\
+		DRM_ERROR( "%s called without lock held\n",		\
+			   __FUNCTION__ );				\
+		return -EINVAL;						\
+	}								\
+} while (0)
+
 #define RING_SPACE_TEST_WITH_RETURN( dev_priv )				\
 do {									\
 	drm_radeon_ring_buffer_t *ring = &dev_priv->ring; int i;	\
@@ -627,20 +625,17 @@ do {									\
 	}								\
 } while (0)
 
-#define RADEON_DISPATCH_AGE( age )					\
-do {									\
+#define RADEON_DISPATCH_AGE( age ) do {					\
 	OUT_RING( CP_PACKET0( RADEON_LAST_DISPATCH_REG, 0 ) );		\
 	OUT_RING( age );						\
 } while (0)
 
-#define RADEON_FRAME_AGE( age )						\
-do {									\
+#define RADEON_FRAME_AGE( age ) do {					\
 	OUT_RING( CP_PACKET0( RADEON_LAST_FRAME_REG, 0 ) );		\
 	OUT_RING( age );						\
 } while (0)
 
-#define RADEON_CLEAR_AGE( age )						\
-do {									\
+#define RADEON_CLEAR_AGE( age ) do {					\
 	OUT_RING( CP_PACKET0( RADEON_LAST_CLEAR_REG, 0 ) );		\
 	OUT_RING( age );						\
 } while (0)
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_state.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_state.c
index 9b84a7397..ff1b3512f 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_state.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/radeon_state.c
@@ -972,50 +972,67 @@ static void radeon_cp_dispatch_indices( drm_device_t *dev,
 	sarea_priv->nbox = 0;
 }
 
-static int radeon_cp_dispatch_blit( drm_device_t *dev,
-				    drm_radeon_blit_t *blit )
+#define RADEON_MAX_TEXTURE_SIZE (RADEON_BUFFER_SIZE - 8 * sizeof(u32))
+
+static int radeon_cp_dispatch_texture( drm_device_t *dev,
+				       drm_radeon_texture_t *tex,
+				       drm_radeon_tex_image_t *image )
 {
 	drm_radeon_private_t *dev_priv = dev->dev_private;
-	drm_device_dma_t *dma = dev->dma;
 	drm_buf_t *buf;
 	drm_radeon_buf_priv_t *buf_priv;
 	u32 format;
-	u32 *data;
-	int dword_shift, dwords;
+	u32 *buffer;
+	u8 *data;
+	int size, dwords, tex_width, blit_width;
+	u32 y, height;
+	int ret = 0, i;
 	RING_LOCALS;
-	DRM_DEBUG( "blit: ofs=0x%x p=%d f=%d x=%hd y=%hd w=%hd h=%hd\n",
-		   blit->offset >> 10, blit->pitch, blit->format,
-		   blit->x, blit->y, blit->width, blit->height );
 
-	radeon_update_ring_snapshot( dev_priv );
+	/* FIXME: Be smarter about this...
+	 */
+	buf = radeon_freelist_get( dev );
+	if ( !buf ) return -EAGAIN;
+
+	DRM_DEBUG( "tex: ofs=0x%x p=%d f=%d x=%hd y=%hd w=%hd h=%hd\n",
+		   tex->offset >> 10, tex->pitch, tex->format,
+		   image->x, image->y, image->width, image->height );
+
+	buf_priv = buf->dev_private;
 
 	/* The compiler won't optimize away a division by a variable,
 	 * even if the only legal values are powers of two.  Thus, we'll
 	 * use a shift instead.
 	 */
-	switch ( blit->format ) {
-	case RADEON_TXF_32BPP_ARGB8888:
-	case RADEON_TXF_32BPP_RGBA8888:
+	switch ( tex->format ) {
+	case RADEON_TXFORMAT_ARGB8888:
+	case RADEON_TXFORMAT_RGBA8888:
 		format = RADEON_COLOR_FORMAT_ARGB8888;
-		dword_shift = 0;
+		tex_width = tex->width * 4;
+		blit_width = image->width * 4;
 		break;
-	case RADEON_TXF_16BPP_AI88:
-	case RADEON_TXF_16BPP_ARGB1555:
-	case RADEON_TXF_16BPP_RGB565:
-	case RADEON_TXF_16BPP_ARGB4444:
+	case RADEON_TXFORMAT_AI88:
+	case RADEON_TXFORMAT_ARGB1555:
+	case RADEON_TXFORMAT_RGB565:
+	case RADEON_TXFORMAT_ARGB4444:
 		format = RADEON_COLOR_FORMAT_RGB565;
-		dword_shift = 1;
+		tex_width = tex->width * 2;
+		blit_width = image->width * 2;
 		break;
-	case RADEON_TXF_8BPP_I:
-	case RADEON_TXF_8BPP_RGB332:
+	case RADEON_TXFORMAT_I8:
+	case RADEON_TXFORMAT_RGB332:
 		format = RADEON_COLOR_FORMAT_CI8;
-		dword_shift = 2;
+		tex_width = tex->width * 1;
+		blit_width = image->width * 1;
 		break;
 	default:
-		DRM_ERROR( "invalid blit format %d\n", blit->format );
+		DRM_ERROR( "invalid texture format %d\n", tex->format );
 		return -EINVAL;
 	}
 
+	DRM_DEBUG( "   tex=%dx%d  blit=%d\n",
+		   tex_width, tex->height, blit_width );
+
 	/* Flush the pixel cache.  This ensures no pixel data gets mixed
 	 * up with the texture data from the host data blit, otherwise
 	 * part of the texture image may be corrupted.
@@ -1027,46 +1044,81 @@ static int radeon_cp_dispatch_blit( drm_device_t *dev,
 
 	ADVANCE_RING();
 
-	/* Dispatch the indirect buffer.
+	/* Make a copy of the parameters in case we have to update them
+	 * for a multi-pass texture blit.
 	 */
-	buf = dma->buflist[blit->idx];
-	buf_priv = buf->dev_private;
+	y = image->y;
+	height = image->height;
+	data = (u8 *)image->data;
 
-	if ( buf->pid != current->pid ) {
-		DRM_ERROR( "process %d using buffer owned by %d\n",
-			   current->pid, buf->pid );
-		return -EINVAL;
-	}
-	if ( buf->pending ) {
-		DRM_ERROR( "sending pending buffer %d\n", blit->idx );
-		return -EINVAL;
-	}
+	size = height * blit_width;
 
-	buf_priv->discard = 1;
+	if ( size > RADEON_MAX_TEXTURE_SIZE ) {
+		/* Texture image is too large, do a multipass upload */
+		ret = -EAGAIN;
 
-	dwords = (blit->width * blit->height) >> dword_shift;
-	if ( !dwords ) dwords = 1;
+		/* Adjust the blit size to fit the indirect buffer */
+		height = RADEON_MAX_TEXTURE_SIZE / blit_width;
+		size = height * blit_width;
 
-	data = (u32 *)((char *)dev_priv->buffers->handle + buf->offset);
+		/* Update the input parameters for next time */
+		image->y += height;
+		image->height -= height;
+		image->data = (char *)image->data + size;
 
-	data[0] = CP_PACKET3( RADEON_CNTL_HOSTDATA_BLT, dwords + 6 );
-	data[1] = (RADEON_GMC_DST_PITCH_OFFSET_CNTL |
-		   RADEON_GMC_BRUSH_NONE |
-		   (format << 8) |
-		   RADEON_GMC_SRC_DATATYPE_COLOR |
-		   RADEON_ROP3_S |
-		   RADEON_DP_SRC_SOURCE_HOST_DATA |
-		   RADEON_GMC_CLR_CMP_CNTL_DIS |
-		   RADEON_GMC_WR_MSK_DIS);
+		if ( copy_to_user( tex->image, image, sizeof(*image) ) )
+			return -EFAULT;
+	} else if ( size < 4 ) {
+		size = 4;
+	}
 
-	data[2] = (blit->pitch << 22) | (blit->offset >> 10);
-	data[3] = 0xffffffff;
-	data[4] = 0xffffffff;
-	data[5] = (blit->y << 16) | blit->x;
-	data[6] = (blit->height << 16) | blit->width;
-	data[7] = dwords;
+	dwords = size / 4;
+
+	/* Dispatch the indirect buffer.
+	 */
+	buffer = (u32 *)((char *)dev_priv->buffers->handle + buf->offset);
+
+	buffer[0] = CP_PACKET3( RADEON_CNTL_HOSTDATA_BLT, dwords + 6 );
+	buffer[1] = (RADEON_GMC_DST_PITCH_OFFSET_CNTL |
+		     RADEON_GMC_BRUSH_NONE |
+		     (format << 8) |
+		     RADEON_GMC_SRC_DATATYPE_COLOR |
+		     RADEON_ROP3_S |
+		     RADEON_DP_SRC_SOURCE_HOST_DATA |
+		     RADEON_GMC_CLR_CMP_CNTL_DIS |
+		     RADEON_GMC_WR_MSK_DIS);
+
+	buffer[2] = (tex->pitch << 22) | (tex->offset >> 10);
+	buffer[3] = 0xffffffff;
+	buffer[4] = 0xffffffff;
+	buffer[5] = (y << 16) | image->x;
+	buffer[6] = (height << 16) | image->width;
+	buffer[7] = dwords;
+
+	buffer += 8;
+
+	if ( tex_width >= 32 ) {
+		/* Texture image width is larger than the minimum, so we
+		 * can upload it directly.
+		 */
+		if ( copy_from_user( buffer, data, dwords * sizeof(u32) ) )
+			return -EFAULT;
+	} else {
+		/* Texture image width is less than the minimum, so we
+		 * need to pad out each image scanline to the minimum
+		 * width.
+		 */
+		for ( i = 0 ; i < tex->height ; i++ ) {
+			if ( copy_from_user( buffer, data, tex_width ) )
+				return -EFAULT;
+			buffer += 8;
+			data += tex_width;
+		}
+	}
 
+	buf->pid = current->pid;
 	buf->used = (dwords + 8) * sizeof(u32);
+	buf_priv->discard = 1;
 
 	radeon_cp_dispatch_indirect( dev, buf, 0, buf->used );
 
@@ -1081,7 +1133,7 @@ static int radeon_cp_dispatch_blit( drm_device_t *dev,
 
 	ADVANCE_RING();
 
-	return 0;
+	return ret;
 }
 
 static void radeon_cp_dispatch_stipple( drm_device_t *dev, u32 *stipple )
@@ -1122,11 +1174,7 @@ int radeon_cp_clear( struct inode *inode, struct file *filp,
 	drm_radeon_clear_rect_t depth_boxes[RADEON_NR_SAREA_CLIPRECTS];
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
 
 	if ( copy_from_user( &clear, (drm_radeon_clear_t *)arg,
 			     sizeof(clear) ) )
@@ -1156,11 +1204,7 @@ int radeon_cp_swap( struct inode *inode, struct file *filp,
 	drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
 
 	RING_SPACE_TEST_WITH_RETURN( dev_priv );
 
@@ -1189,11 +1233,8 @@ int radeon_cp_vertex( struct inode *inode, struct file *filp,
 	drm_radeon_buf_priv_t *buf_priv;
 	drm_radeon_vertex_t vertex;
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
+
 	if ( !dev_priv || dev_priv->is_pci ) {
 		DRM_ERROR( "%s called with a PCI card\n", __FUNCTION__ );
 		return -EINVAL;
@@ -1255,11 +1296,8 @@ int radeon_cp_indices( struct inode *inode, struct file *filp,
 	drm_radeon_indices_t elts;
 	int count;
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
+
 	if ( !dev_priv || dev_priv->is_pci ) {
 		DRM_ERROR( "%s called with a PCI card\n", __FUNCTION__ );
 		return -EINVAL;
@@ -1321,38 +1359,34 @@ int radeon_cp_indices( struct inode *inode, struct file *filp,
 	return 0;
 }
 
-int radeon_cp_blit( struct inode *inode, struct file *filp,
-		    unsigned int cmd, unsigned long arg )
+int radeon_cp_texture( struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg )
 {
 	drm_file_t *priv = filp->private_data;
 	drm_device_t *dev = priv->dev;
 	drm_radeon_private_t *dev_priv = dev->dev_private;
-	drm_device_dma_t *dma = dev->dma;
-	drm_radeon_blit_t blit;
+	drm_radeon_texture_t tex;
+	drm_radeon_tex_image_t image;
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
 
-	if ( copy_from_user( &blit, (drm_radeon_blit_t *)arg,
-			     sizeof(blit) ) )
+	if ( copy_from_user( &tex, (drm_radeon_texture_t *)arg, sizeof(tex) ) )
 		return -EFAULT;
 
-	DRM_DEBUG( "%s: pid=%d index=%d\n",
-		   __FUNCTION__, current->pid, blit.idx );
-
-	if ( blit.idx < 0 || blit.idx > dma->buf_count ) {
-		DRM_ERROR( "sending %d buffers (of %d max)\n",
-			   blit.idx, dma->buf_count );
+	if ( tex.image == NULL ) {
+		DRM_ERROR( "null texture image!\n" );
 		return -EINVAL;
 	}
 
+	if ( copy_from_user( &image,
+			     (drm_radeon_tex_image_t *)tex.image,
+			     sizeof(image) ) )
+		return -EFAULT;
+
 	RING_SPACE_TEST_WITH_RETURN( dev_priv );
 	VB_AGE_TEST_WITH_RETURN( dev_priv );
 
-	return radeon_cp_dispatch_blit( dev, &blit );
+	return radeon_cp_dispatch_texture( dev, &tex, &image );
 }
 
 int radeon_cp_stipple( struct inode *inode, struct file *filp,
@@ -1364,18 +1398,13 @@ int radeon_cp_stipple( struct inode *inode, struct file *filp,
 	drm_radeon_stipple_t stipple;
 	u32 mask[32];
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
 
 	if ( copy_from_user( &stipple, (drm_radeon_stipple_t *)arg,
 			     sizeof(stipple) ) )
 		return -EFAULT;
 
-	if ( copy_from_user( &mask, stipple.mask,
-			     32 * sizeof(u32) ) )
+	if ( copy_from_user( &mask, stipple.mask, 32 * sizeof(u32) ) )
 		return -EFAULT;
 
 	RING_SPACE_TEST_WITH_RETURN( dev_priv );
@@ -1397,11 +1426,8 @@ int radeon_cp_indirect( struct inode *inode, struct file *filp,
 	drm_radeon_indirect_t indirect;
 	RING_LOCALS;
 
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||
-	     dev->lock.pid != current->pid ) {
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );
-		return -EINVAL;
-	}
+	LOCK_TEST_WITH_RETURN( dev );
+
 	if ( !dev_priv || dev_priv->is_pci ) {
 		DRM_ERROR( "%s called with a PCI card\n", __FUNCTION__ );
 		return -EINVAL;
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drm.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/stubsupport-pre24.h
index b4acd70a8..f5046f7f9 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drm.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/stubsupport-pre24.h
@@ -1,9 +1,8 @@
-/* gamma_drm.h -- Public header for the Gamma driver -*- linux-c -*-
- * Created: Tue Jan 25 01:50:01 1999 by jhartmann@precisioninsight.com
+/* stubsupport.h -- -*- linux-c -*-
+ * Created: Fri Jan 19 10:48:35 2001 by faith@acm.org
  *
- * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
- * Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
- * All rights reserved.
+ * Copyright 2001 VA Linux Systems, Inc., Sunnyvale, California.
+ * 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"),
@@ -19,23 +18,27 @@
  * 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
- * VA LINUX SYSTEMS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * PRECISION INSIGHT 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.
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
  *
  * Authors:
- *    Jeff Hartmann <jhartmann@valinux.com>
+ *    Rickard E. (Rik) Faith <faith@valinux.com>
+ *
  */
 
-#ifndef __GAMMA_DRM_H__
-#define __GAMMA_DRM_H__
+#ifndef _STUBSUPPORT_PRE24_H_
+#define _STUBSUPPORT_PRE24_H_
+struct drm_stub_info *DRM(_stub_pointer) = NULL;
+#define inter_module_put(x)
+#define inter_module_unregister(x)
+#define inter_module_get(x) DRM(_stub_pointer)
+#define inter_module_register(x,y,z) do { DRM(_stub_pointer) = z; } while (0)
 
-typedef struct drm_gamma_init {
-	unsigned long hControlRegs0;
-	unsigned long hControlRegs1;
-	unsigned long hControlRegs2;
-	unsigned long hControlRegs3;
-} drm_gamma_init_t;
+				/* This is a kludge for backward compatibility
+                                   that is only useful in DRM(stub_open) */
+#define fops_put(fops) MOD_DEC_USE_COUNT
+#define fops_get(fops) (fops); MOD_INC_USE_COUNT
 
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx.h
index 6af675ff6..40aba8fb2 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx.h
@@ -34,6 +34,8 @@
  */
 #define DRM(x) tdfx_##x
 
+/* General customization:
+ */
 #define __HAVE_MTRR		1
 #define __HAVE_CTX_BITMAP	1
 
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx_drv.c
index e8d7955ca..f478395f8 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx_drv.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx_drv.c
@@ -38,21 +38,19 @@
 
 #define DRIVER_NAME		"tdfx"
 #define DRIVER_DESC		"3dfx Banshee/Voodoo3+"
-#define DRIVER_DATE		"20010215"
+#define DRIVER_DATE		"20010216"
 
 #define DRIVER_MAJOR		1
 #define DRIVER_MINOR		0
 #define DRIVER_PATCHLEVEL	0
 
-#include "drm_drv.h"
-
 
-#include "drm_agpsupport.h"
 #include "drm_auth.h"
 #include "drm_bufs.h"
 #include "drm_context.h"
 #include "drm_dma.h"
 #include "drm_drawable.h"
+#include "drm_drv.h"
 #include "drm_fops.h"
 #include "drm_init.h"
 #include "drm_ioctl.h"
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drm.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drm.c
index f51840201..a640d5987 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drm.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drm.c
@@ -81,7 +81,18 @@ extern unsigned long _bus_base(void);
 #include "xf86drm.h"
 #include "drm.h"
 
-#define DRM_FIXED_DEVICE_MAJOR 145
+#ifndef DRM_MAJOR
+#define DRM_MAJOR 226		/* Linux */
+#endif
+
+#ifndef __linux__
+#undef  DRM_MAJOR
+#define DRM_MAJOR 145		/* Should set in drm.h for *BSD */
+#endif
+
+#ifndef DRM_MAX_MINOR
+#define DRM_MAX_MINOR 16
+#endif
 
 #ifdef __linux__
 #include <sys/sysmacros.h>	/* for makedev() */
@@ -161,93 +172,108 @@ static drmHashEntry *drmGetEntry(int fd)
     return entry;
 }
 
-/* drm_open is used to open the /dev/dri device */
-
-static int drm_open(const char *file)
-{
-    int fd = open(file, O_RDWR, 0);
-
-    if (fd >= 0) return fd;
-    return -errno;
-}
-
-static int drmOpenDevice(const char *path, long dev,
-			 mode_t mode, uid_t user, gid_t group)
+static int drmOpenDevice(long dev, int minor)
 {
 #ifdef XFree86LOADER
     struct xf86stat st;
 #else
     struct stat     st;
 #endif
+    char            buf[64];
+    int             fd;
+    mode_t          dirmode = DRM_DEV_DIRMODE;
+    mode_t          devmode = DRM_DEV_MODE;
+    int             isroot  = !geteuid();
+#if defined(XFree86Server)
+    uid_t           user    = DRM_DEV_UID;
+    gid_t           group   = DRM_DEV_GID;
+#endif
 
-				/* Fiddle mode to remove execute bits */
-    mode &= ~(S_IXUSR|S_IXGRP|S_IXOTH);
+#if defined(XFree86Server)
+    devmode  = xf86ConfigDRI.mode ? xf86ConfigDRI.mode : DRM_DEV_MODE;
+    dirmode  = (devmode & S_IRUSR) ? S_IXUSR : 0;
+    dirmode |= (devmode & S_IRGRP) ? S_IXGRP : 0;
+    dirmode |= (devmode & S_IROTH) ? S_IXOTH : 0;
+    dirmode |= devmode;
+    devmode &= ~(S_IXUSR|S_IXGRP|S_IXOTH);
+    group = (xf86ConfigDRI.group >= 0) ? xf86ConfigDRI.group : DRM_DEV_GID;
+#endif
 
-    if (!stat(path, &st) && st.st_rdev == dev) {
-	if (!geteuid()) {
-	    chown(path, user, group);
-	    chmod(path, mode);
-	}
-	return drm_open(path);
+    if (stat(DRM_DIR_NAME, &st)) {
+	if (!isroot) return DRM_ERR_NOT_ROOT;
+	remove(DRM_DIR_NAME);
+	mkdir(DRM_DIR_NAME, dirmode);
     }
+#if defined(XFree86Server)
+    chown(DRM_DIR_NAME, user, group);
+    chmod(DRM_DIR_NAME, dirmode);
+#endif
 
-    if (geteuid()) return DRM_ERR_NOT_ROOT;
-    remove(path);
-    if (mknod(path, S_IFCHR, dev)) {
-	remove(path);
-	return DRM_ERR_NOT_ROOT;
+    sprintf(buf, DRM_DEV_NAME, DRM_DIR_NAME, minor);
+    if (stat(buf, &st) || st.st_rdev != dev) {
+	if (!isroot) return DRM_ERR_NOT_ROOT;
+	remove(buf);
+	mknod(buf, S_IFCHR | devmode, dev);
     }
-    chown(path, user, group);
-    chmod(path, mode);
-    return drm_open(path);
+#if defined(XFree86Server)
+    chown(buf, user, group);
+    chmod(buf, devmode);
+#endif
+
+    if ((fd = open(buf, O_RDWR, 0)) >= 0) return fd;
+    remove(buf);
+    return -errno;
 }
 
-/* drmAvailable looks for /proc/dri, and returns 1 if it is present.  On
-   OSs that do not have a Linux-like /proc, this information will not be
-   available, and we'll have to create a device and check if the driver is
-   loaded that way. */
+int drmOpenMinor(int minor, int create)
+{
+    int  fd;
+    char buf[64];
+    
+    if (create) return drmOpenDevice(makedev(DRM_MAJOR, minor), minor);
+    
+    sprintf(buf, DRM_DEV_NAME, DRM_DIR_NAME, minor);
+    if ((fd = open(buf, O_RDWR, 0)) >= 0) return fd;
+    return -errno;
+}
+
+/* drmAvailable looks for (DRM_MAJOR, 0) and returns 1 if it returns
+   information for DRM_IOCTL_VERSION.  For backward compatibility with
+   older Linux implementations, /proc/dri is also checked. */
 
 int drmAvailable(void)
 {
-    char          dev_name[64];
     drmVersionPtr version;
     int           retval = 0;
     int           fd;
 
-    if (!access("/proc/dri/0", R_OK)) return 1;
-
-    sprintf(dev_name, "/dev/dri-temp-%d", getpid());
-
-    remove(dev_name);
-    if ((fd = drmOpenDevice(dev_name, makedev(DRM_FIXED_DEVICE_MAJOR, 0),
-			    S_IRUSR, geteuid(), getegid())) >= 0) {
-				/* Read version to make sure this is
-                                   actually a DRI device. */
-	if ((version = drmGetVersion(fd))) {
-	    retval = 1;
-	    drmFreeVersion(version);
-	}
-	close(fd);
+    if ((fd = drmOpenMinor(0, 1)) < 0) {
+				/* Try proc for backward Linux compatibility */
+	if (!access("/proc/dri/0", R_OK)) return 1;
+	return 0;
     }
-    remove(dev_name);
+    
+    if ((version = drmGetVersion(fd))) {
+	retval = 1;
+	drmFreeVersion(version);
+    }
+    close(fd);
 
     return retval;
 }
 
 static int drmOpenByBusid(const char *busid)
 {
-    int    i;
-    char   dev_name[64];
-    char   *buf;
-    int    fd;
-
-    for (i = 0; i < 8; i++) {
-	sprintf(dev_name, "/dev/dri/card%d", i);
-	if ((fd = drm_open(dev_name)) >= 0) {
+    int        i;
+    int        fd;
+    const char *buf;
+    
+    for (i = 0; i < DRM_MAX_MINOR; i++) {
+	if ((fd = drmOpenMinor(i, 0)) >= 0) {
 	    buf = drmGetBusid(fd);
 	    if (buf && !strcmp(buf, busid)) {
-	      drmFreeBusid(buf);
-	      return fd;
+		drmFreeBusid(buf);
+		return fd;
 	    }
 	    if (buf) drmFreeBusid(buf);
 	    close(fd);
@@ -258,54 +284,43 @@ static int drmOpenByBusid(const char *busid)
 
 static int drmOpenByName(const char *name)
 {
-    int    i;
-    char   proc_name[64];
-    char   dev_name[64];
-    char   buf[512];
-    mode_t mode   = DRM_DEV_MODE;
-    mode_t dirmode;
-    gid_t  group  = DRM_DEV_GID;
-    uid_t  user   = DRM_DEV_UID;
-    int    fd;
-    char   *pt;
-    char   *driver = NULL;
-    char   *devstring;
-    long   dev     = 0;
-    int    retcode;
-
-#if defined(XFree86Server)
-    mode  = xf86ConfigDRI.mode ? xf86ConfigDRI.mode : DRM_DEV_MODE;
-    group = (xf86ConfigDRI.group >= 0) ? xf86ConfigDRI.group : DRM_DEV_GID;
-#endif
-
-#if defined(XFree86Server)
+    int           i;
+    int           fd;
+    drmVersionPtr version;
+    
     if (!drmAvailable()) {
+#if !defined(XFree86Server)
+	return -1;
+#else
         /* try to load the kernel module now */
         if (!xf86LoadKernelModule(name)) {
             ErrorF("[drm] failed to load kernel module \"%s\"\n",
 		   name);
             return -1;
         }
-    }
-#else
-    if (!drmAvailable())
-       return -1;
 #endif
+    }
 
-    if (!geteuid()) {
-	dirmode = mode;
-	if (dirmode & S_IRUSR) dirmode |= S_IXUSR;
-	if (dirmode & S_IRGRP) dirmode |= S_IXGRP;
-	if (dirmode & S_IROTH) dirmode |= S_IXOTH;
-	dirmode &= ~(S_IWGRP | S_IWOTH);
-	mkdir("/dev/dri", 0);
-	chown("/dev/dri", user, group);
-	chmod("/dev/dri", dirmode);
+    for (i = 0; i < DRM_MAX_MINOR; i++) {
+	if ((fd = drmOpenMinor(i, 1)) >= 0) {
+	    if ((version = drmGetVersion(fd))) {
+		if (!strcmp(version->name, name)) {
+		    drmFreeVersion(version);
+		    return fd;
+		}
+		drmFreeVersion(version);
+	    }
+	}
     }
 
+#ifdef __linux__
+				/* Backward-compatibility /proc support */
     for (i = 0; i < 8; i++) {
+	char proc_name[64], buf[512];
+	char *driver, *pt, *devstring;
+	int  retcode;
+	
 	sprintf(proc_name, "/proc/dri/%d/name", i);
-	sprintf(dev_name, "/dev/dri/card%d", i);
 	if ((fd = open(proc_name, 0, 0)) >= 0) {
 	    retcode = read(fd, buf, sizeof(buf)-1);
 	    close(fd);
@@ -319,34 +334,17 @@ static int drmOpenByName(const char *name)
 			for (devstring = ++pt; *pt && *pt != ' '; ++pt)
 			    ;
 			if (*pt) { /* Found busid */
-			  return drmOpenByBusid(++pt);
+			    return drmOpenByBusid(++pt);
 			} else {	/* No busid */
-			  dev = strtol(devstring, NULL, 0);
-			  return drmOpenDevice(dev_name, dev,
-					       mode, user, group);
+			    return drmOpenDevice(strtol(devstring, NULL, 0),i);
 			}
 		    }
 		}
 	    }
-	} else {
-	    drmVersionPtr version;
-				/* /proc/dri not available, possibly
-                                   because we aren't on a Linux system.
-                                   So, try to create the next device and
-                                   see if it's active. */
-	    dev = makedev(DRM_FIXED_DEVICE_MAJOR, i);
-	    if ((fd = drmOpenDevice(dev_name, dev, mode, user, group))) {
-		if ((version = drmGetVersion(fd))) {
-		    if (!strcmp(version->name, name)) {
-			drmFreeVersion(version);
-			return fd;
-		    }
-		    drmFreeVersion(version);
-		}
-	    }
-	    remove(dev_name);
 	}
     }
+#endif
+
     return -1;
 }
 
@@ -1123,6 +1121,160 @@ int drmGetContextPrivateMapping(int fd, drmContext ctx_id, drmHandlePtr handle)
     return 0;
 }
 
+int drmGetMap(int fd, int idx, drmHandle *offset, drmSize *size,
+	      drmMapType *type, drmMapFlags *flags, drmHandle *handle,
+	      int *mtrr)
+{
+    drm_map_t map;
+
+    map.offset = idx;
+    if (ioctl(fd, DRM_IOCTL_GET_MAP, &map)) return -errno;
+    *offset = map.offset;
+    *size   = map.size;
+    *type   = map.type;
+    *flags  = map.flags;
+    *handle = (unsigned long)map.handle;
+    *mtrr   = map.mtrr;
+    return 0;
+}
+
+int drmGetClient(int fd, int idx, int *auth, int *pid, int *uid,
+		 unsigned long *magic, unsigned long *iocs)
+{
+    drm_client_t client;
+
+    client.idx = idx;
+    if (ioctl(fd, DRM_IOCTL_GET_CLIENT, &client)) return -errno;
+    *auth      = client.auth;
+    *pid       = client.pid;
+    *uid       = client.uid;
+    *magic     = client.magic;
+    *iocs      = client.iocs;
+    return 0;
+}
+
+int drmGetStats(int fd, drmStatsT *stats)
+{
+    drm_stats_t s;
+    int         i;
+
+    if (ioctl(fd, DRM_IOCTL_GET_STATS, &s)) return -errno;
+
+    stats->count = 0;
+    memset(stats, 0, sizeof(*stats));
+    if (s.count > sizeof(stats->data)/sizeof(stats->data[0]))
+	return -1;
+
+#define SET_VALUE                              \
+    stats->data[i].long_format = "%-20.20s";   \
+    stats->data[i].rate_format = "%8.8s";      \
+    stats->data[i].isvalue     = 1;            \
+    stats->data[i].verbose     = 0
+
+#define SET_COUNT                              \
+    stats->data[i].long_format = "%-20.20s";   \
+    stats->data[i].rate_format = "%5.5s";      \
+    stats->data[i].isvalue     = 0;            \
+    stats->data[i].mult_names  = "kgm";        \
+    stats->data[i].mult        = 1000;         \
+    stats->data[i].verbose     = 0
+
+#define SET_BYTE                               \
+    stats->data[i].long_format = "%-20.20s";   \
+    stats->data[i].rate_format = "%5.5s";      \
+    stats->data[i].isvalue     = 0;            \
+    stats->data[i].mult_names  = "KGM";        \
+    stats->data[i].mult        = 1024;         \
+    stats->data[i].verbose     = 0
+
+
+    stats->count = s.count;
+    for (i = 0; i < s.count; i++) {
+	stats->data[i].value = s.data[i].value;
+	switch (s.data[i].type) {
+	case _DRM_STAT_LOCK:
+	    stats->data[i].long_name = "Lock";
+	    stats->data[i].rate_name = "Lock";
+	    SET_VALUE;
+	    break;
+	case _DRM_STAT_OPENS:
+	    stats->data[i].long_name = "Opens";
+	    stats->data[i].rate_name = "O";
+	    SET_COUNT;
+	    stats->data[i].verbose   = 1;
+	    break;
+	case _DRM_STAT_CLOSES:
+	    stats->data[i].long_name = "Closes";
+	    stats->data[i].rate_name = "Lock";
+	    SET_COUNT;
+	    stats->data[i].verbose   = 1;
+	    break;
+	case _DRM_STAT_IOCTLS:
+	    stats->data[i].long_name = "Ioctls";
+	    stats->data[i].rate_name = "Ioc/s";
+	    SET_COUNT;
+	    break;
+	case _DRM_STAT_LOCKS:
+	    stats->data[i].long_name = "Locks";
+	    stats->data[i].rate_name = "Lck/s";
+	    SET_COUNT;
+	    break;
+	case _DRM_STAT_UNLOCKS:
+	    stats->data[i].long_name = "Unlocks";
+	    stats->data[i].rate_name = "Unl/s";
+	    SET_COUNT;
+	    break;
+	case _DRM_STAT_IRQ:
+	    stats->data[i].long_name = "IRQs";
+	    stats->data[i].rate_name = "IRQ/s";
+	    SET_COUNT;
+	    break;
+	case _DRM_STAT_PRIMARY:
+	    stats->data[i].long_name = "Primary Bytes";
+	    stats->data[i].rate_name = "PB/s";
+	    SET_BYTE;
+	    break;
+	case _DRM_STAT_SECONDARY:
+	    stats->data[i].long_name = "Secondary Bytes";
+	    stats->data[i].rate_name = "SB/s";
+	    SET_BYTE;
+	    break;
+	case _DRM_STAT_DMA:
+	    stats->data[i].long_name = "DMA";
+	    stats->data[i].rate_name = "DMA/s";
+	    SET_COUNT;
+	    break;
+	case _DRM_STAT_SPECIAL:
+	    stats->data[i].long_name = "Special DMA";
+	    stats->data[i].rate_name = "dma/s";
+	    SET_COUNT;
+	    break;
+	case _DRM_STAT_MISSED:
+	    stats->data[i].long_name = "Miss";
+	    stats->data[i].rate_name = "Ms/s";
+	    SET_COUNT;
+	    break;
+	case _DRM_STAT_VALUE:
+	    stats->data[i].long_name = "Value";
+	    stats->data[i].rate_name = "Value";
+	    SET_VALUE;
+	    break;
+	case _DRM_STAT_BYTE:
+	    stats->data[i].long_name = "Bytes";
+	    stats->data[i].rate_name = "B/s";
+	    SET_BYTE;
+	    break;
+	case _DRM_STAT_COUNT:
+	default:
+	    stats->data[i].long_name = "Count";
+	    stats->data[i].rate_name = "Cnt/s";
+	    SET_COUNT;
+	    break;
+	}
+    }
+    return 0;
+}
+
 #if defined(XFree86Server) || defined(DRM_USE_MALLOC)
 static void drmSIGIOHandler(int interrupt, void *closure)
 {
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmGlint.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmGlint.c
deleted file mode 100644
index 0729c8692..000000000
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmGlint.c
+++ /dev/null
@@ -1,96 +0,0 @@
-/* xf86drmGlint.c -- User-level interface to Gamma DRM device
- * Created: Wed Feb  21 08:31:23 2000 by jhartmann@valinux.com
- *
- * Copyright 1999, 2000 Precision Insight, 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, 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
- * PRECISION INSIGHT 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:
- *   Jeff Hartmann <jhartmann@valinux.com>
- *
- */
-/* $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmMga.c,v 1.3 2000/08/24 22:20:17 tsi Exp $ */
-
-#ifdef XFree86Server
-# include "xf86.h"
-# include "xf86_OSproc.h"
-# include "xf86_ansic.h"
-# include "xf86Priv.h"
-# define _DRM_MALLOC xalloc
-# define _DRM_FREE   xfree
-# ifndef XFree86LOADER
-#  include <sys/stat.h>
-#  include <sys/mman.h>
-# endif
-#else
-# include <stdio.h>
-# include <stdlib.h>
-# include <unistd.h>
-# include <string.h>
-# include <ctype.h>
-# include <fcntl.h>
-# include <errno.h>
-# include <signal.h>
-# include <sys/types.h>
-# include <sys/stat.h>
-# include <sys/ioctl.h>
-# include <sys/mman.h>
-# include <sys/time.h>
-# ifdef DRM_USE_MALLOC
-#  define _DRM_MALLOC malloc
-#  define _DRM_FREE   free
-extern int xf86InstallSIGIOHandler(int fd, void (*f)(int, void *), void *);
-extern int xf86RemoveSIGIOHandler(int fd);
-# else
-#  include <Xlibint.h>
-#  define _DRM_MALLOC Xmalloc
-#  define _DRM_FREE   Xfree
-# endif
-#endif
-
-/* Not all systems have MAP_FAILED defined */
-#ifndef MAP_FAILED
-#define MAP_FAILED ((void *)-1)
-#endif
-
-#ifdef __linux__
-#include <sys/sysmacros.h>	/* for makedev() */
-#endif
-#include "xf86drm.h"
-#include "xf86drmGlint.h"
-#include "drm.h"
-
-int
-drmGLINTInitDMA(int drmFD, drmGLINTInit *init)
-{
-    drm_gamma_init_t init;
-    memset(&init, 0, sizeof(drm_gamma_init_t));
-   
-    init.hControlRegs0 = init->hControlRegs0;
-    init.hControlRegs1 = init->hControlRegs1;
-    init.hControlRegs2 = init->hControlRegs2;
-    init.hControlRegs3 = init->hControlRegs3;
-
-    if (ioctl(drmFD, DRM_IOCTL_GAMMA_INIT, &init)) {
-        return -errno;
-    }
-    return 0;
-}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmRadeon.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmRadeon.c
index 62caddadc..944a030ff 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmRadeon.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmRadeon.c
@@ -311,25 +311,32 @@ int drmRadeonFlushIndices( int fd, int prim, int index,
    }
 }
 
-int drmRadeonTextureBlit( int fd, int index,
-			  int offset, int pitch, int format,
-			  int x, int y, int width, int height )
+int drmRadeonLoadTexture( int fd, int offset, int pitch, int format,
+			  int width, int height, drmRadeonTexImage *image )
 {
-   drm_radeon_blit_t blit;
-
-   blit.idx = index;
-   blit.offset = offset;
-   blit.pitch = pitch;
-   blit.format = format;
-   blit.x = x;
-   blit.y = y;
-   blit.width = width;
-   blit.height = height;
-
-   if ( ioctl( fd, DRM_IOCTL_RADEON_BLIT, &blit ) < 0 ) {
-      return -errno;
-   } else {
+   drm_radeon_texture_t tex;
+   drm_radeon_tex_image_t tmp;
+   int ret;
+
+   tex.offset = offset;
+   tex.pitch = pitch;
+   tex.format = format;
+   tex.width = width;
+   tex.height = height;
+   tex.image = &tmp;
+
+   /* This gets updated by the kernel when a multipass blit is needed.
+    */
+   memcpy( &tmp, image, sizeof(drm_radeon_tex_image_t) );
+
+   do {
+      ret = ioctl( fd, DRM_IOCTL_RADEON_TEXTURE, &tex );
+   } while ( ret && errno == EAGAIN );
+
+   if ( ret == 0 ) {
       return 0;
+   } else {
+      return -errno;
    }
 }
 
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/shared/drm/kernel/drm.h b/xc/programs/Xserver/hw/xfree86/os-support/shared/drm/kernel/drm.h
index 7c65f5fc3..ddabc2771 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/shared/drm/kernel/drm.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/shared/drm/kernel/drm.h
@@ -44,11 +44,8 @@
 #define DRM_IOCTL_NR(n)	     ((n) & 0xff)
 #endif
 
-#define DRM_DEV_MODE	 (S_IRUSR|S_IWUSR|S_IRGRP|S_IWGRP)
-#define DRM_DEV_UID	 0
-#define DRM_DEV_GID	 0
-
 #define DRM_MAJOR       226
+#define DRM_MAX_MINOR   15
 #define DRM_NAME	"drm"	  /* Name in kernel, /dev, and /proc	    */
 #define DRM_MIN_ORDER	5	  /* At least 2^5 bytes = 32 bytes	    */
 #define DRM_MAX_ORDER	22	  /* Up to 2^22 bytes = 4MB		    */
@@ -453,7 +450,7 @@ typedef struct drm_agp_info {
 #define DRM_IOCTL_RADEON_CLEAR		DRM_IOW( 0x48, drm_radeon_clear_t)
 #define DRM_IOCTL_RADEON_VERTEX		DRM_IOW( 0x49, drm_radeon_vertex_t)
 #define DRM_IOCTL_RADEON_INDICES	DRM_IOW( 0x4a, drm_radeon_indices_t)
-#define DRM_IOCTL_RADEON_BLIT		DRM_IOW( 0x4b, drm_radeon_blit_t)
+#define DRM_IOCTL_RADEON_TEXTURE	DRM_IOWR(0x4b, drm_radeon_texture_t)
 #define DRM_IOCTL_RADEON_STIPPLE	DRM_IOW( 0x4c, drm_radeon_stipple_t)
 #define DRM_IOCTL_RADEON_INDIRECT	DRM_IOWR(0x4d, drm_radeon_indirect_t)
 
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/xf86drm.h b/xc/programs/Xserver/hw/xfree86/os-support/xf86drm.h
index df5ddd8fb..180e15844 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/xf86drm.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/xf86drm.h
@@ -33,6 +33,16 @@
 #ifndef _XF86DRM_H_
 #define _XF86DRM_H_
 
+				/* Defaults, if nothing set in xf86config */
+#define DRM_DEV_UID	 0
+#define DRM_DEV_GID	 0
+#define DRM_DEV_DIRMODE	 (S_IRUSR|S_IWUSR|S_IXUSR|S_IRGRP|S_IWGRP|S_IXGRP)
+#define DRM_DEV_MODE	 (S_IRUSR|S_IWUSR|S_IRGRP|S_IWGRP)
+
+#define DRM_DIR_NAME  "/dev/dri"
+#define DRM_DEV_NAME  "%s/card%d"
+#define DRM_PROC_NAME "/proc/dri/" /* For backware Linux compatibility */
+
 #define DRM_ERR_NO_DEVICE  (-1001)
 #define DRM_ERR_NO_ACCESS  (-1002)
 #define DRM_ERR_NOT_ROOT   (-1003)
@@ -58,6 +68,21 @@ typedef struct _drmVersion {
     char    *desc;                /* User-space buffer to hold desc         */
 } drmVersion, *drmVersionPtr;
 
+typedef struct _drmStats {
+    unsigned long count;	     /* Number of data                      */
+    struct {
+	unsigned long value;	     /* Value from kernel                   */
+	const char    *long_format;  /* Suggested format for long_name      */
+	const char    *long_name;    /* Long name for value                 */
+	const char    *rate_format;  /* Suggested format for rate_name      */
+	const char    *rate_name;    /* Short name for value per second     */
+	int           isvalue;       /* True if value (vs. counter)         */
+	const char    *mult_names;   /* Multiplier names (e.g., "KGM")      */
+	int           mult;          /* Multiplier value (e.g., 1024)       */
+	int           verbose;       /* Suggest only in verbose output      */
+    } data[15];
+} drmStatsT;
+
 
 				/* All of these enums *MUST* match with the
                                    kernel implementation -- so do *NOT*
@@ -191,14 +216,7 @@ typedef struct _drmTextureRegion {
     unsigned int  age;
 } drmTextureRegion, *drmTextureRegionPtr;
 
-#if 0
-				/* The kernel does this, but it doesn't
-                                   seem necessary with recent gcc's.  */
-typedef struct { unsigned int a[100]; } __drm_dummy_lock_t;
-#define __drm_dummy_lock(lock) (*(__volatile__ __drm_dummy_lock_t *)lock)
-#else
 #define __drm_dummy_lock(lock) (*(__volatile__ unsigned int *)lock)
-#endif
 
 #define DRM_LOCK_HELD  0x80000000 /* Hardware lock is held                 */
 #define DRM_LOCK_CONT  0x40000000 /* Hardware lock is contended            */
@@ -388,6 +406,14 @@ extern int           drmGetMagic(int fd, drmMagicPtr magic);
 extern char          *drmGetBusid(int fd);
 extern int           drmGetInterruptFromBusID(int fd, int busnum, int devnum,
 					      int funcnum);
+extern int           drmGetMap(int fd, int idx, drmHandle *offset,
+			       drmSize *size, drmMapType *type,
+			       drmMapFlags *flags, drmHandle *handle,
+			       int *mtrr);
+extern int           drmGetClient(int fd, int idx, int *auth, int *pid,
+				  int *uid, unsigned long *magic,
+				  unsigned long *iocs);
+extern int           drmGetStats(int fd, drmStatsT *stats);
 
 
 /* General user-level programmer's API: X server (root) only  */
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/xf86drmGlint.h b/xc/programs/Xserver/hw/xfree86/os-support/xf86drmGlint.h
deleted file mode 100644
index 7bd0073a8..000000000
--- a/xc/programs/Xserver/hw/xfree86/os-support/xf86drmGlint.h
+++ /dev/null
@@ -1,13 +0,0 @@
-#ifndef __XF86DRI_GLINT_H__
-#define __XF86DRI_GLINT_H__
-
-typedef struct _drmGLINTInit {
-    drmHandle hControlRegs0;
-    drmHandle hControlRegs1;
-    drmHandle hControlRegs2;
-    drmHandle hControlRegs3;
-} drmGLINTInit;
-
-extern int drmGLINTInitDMA(int drmFD, drmGLINTInit *init);
-
-#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/xf86drmRadeon.h b/xc/programs/Xserver/hw/xfree86/os-support/xf86drmRadeon.h
index 87f70590d..e3bf83fdd 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/xf86drmRadeon.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/xf86drmRadeon.h
@@ -43,27 +43,35 @@
 #define DRM_RADEON_DEPTH	0x4
 
 typedef struct {
-    int sarea_priv_offset;
-    int is_pci;
-    int cp_mode;
-    int agp_size;
-    int ring_size;
-    int usec_timeout;
-
-    unsigned int fb_bpp;
-    unsigned int front_offset, front_pitch;
-    unsigned int back_offset, back_pitch;
-    unsigned int depth_bpp;
-    unsigned int depth_offset, depth_pitch;
-
-    unsigned int fb_offset;
-    unsigned int mmio_offset;
-    unsigned int ring_offset;
-    unsigned int ring_rptr_offset;
-    unsigned int buffers_offset;
-    unsigned int agp_textures_offset;
+   int sarea_priv_offset;
+   int is_pci;
+   int cp_mode;
+   int agp_size;
+   int ring_size;
+   int usec_timeout;
+
+   unsigned int fb_bpp;
+   unsigned int front_offset, front_pitch;
+   unsigned int back_offset, back_pitch;
+   unsigned int depth_bpp;
+   unsigned int depth_offset, depth_pitch;
+
+   unsigned int fb_offset;
+   unsigned int mmio_offset;
+   unsigned int ring_offset;
+   unsigned int ring_rptr_offset;
+   unsigned int buffers_offset;
+   unsigned int agp_textures_offset;
 } drmRadeonInit;
 
+typedef struct {
+   unsigned int x;
+   unsigned int y;
+   unsigned int width;
+   unsigned int height;
+   void *data;
+} drmRadeonTexImage;
+
 extern int drmRadeonInitCP( int fd, drmRadeonInit *info );
 extern int drmRadeonCleanupCP( int fd );
 
@@ -87,9 +95,9 @@ extern int drmRadeonFlushVertexBuffer( int fd, int prim, int index,
 extern int drmRadeonFlushIndices( int fd, int prim, int index,
 				  int start, int end, int discard );
 
-extern int drmRadeonTextureBlit( int fd, int index,
-				 int offset, int pitch, int format,
-				 int x, int y, int width, int height );
+extern int drmRadeonLoadTexture( int fd, int offset, int pitch, int format,
+				 int width, int height,
+				 drmRadeonTexImage *image );
 
 extern int drmRadeonPolygonStipple( int fd, unsigned int *mask );