change gl_buffer_object's Size field to GLsizeiptrARB type
[mesa.git] / src / mesa / main / texstore.c
index 7d37bcb00fd8c0f1aad576430ada4637f4f23ff9..5f30e03d06a7cfc266841c4e78901ac479f2d89b 100644 (file)
@@ -1,10 +1,8 @@
-/* $Id: texstore.c,v 1.37 2002/06/15 03:03:09 brianp Exp $ */
-
 /*
  * Mesa 3-D graphics library
- * Version:  4.1
+ * Version:  6.3
  *
- * Copyright (C) 1999-2002  Brian Paul   All Rights Reserved.
+ * Copyright (C) 1999-2004  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"),
  *   Brian Paul
  */
 
+/*
+ * The GL texture image functions in teximage.c basically just do
+ * error checking and data structure allocation.  They in turn call
+ * device driver functions which actually copy/convert/store the user's
+ * texture image data.
+ *
+ * However, most device drivers will be able to use the fallback functions
+ * in this file.  That is, most drivers will have the following bit of
+ * code:
+ *   ctx->Driver.TexImage1D = _mesa_store_teximage1d;
+ *   ctx->Driver.TexImage2D = _mesa_store_teximage2d;
+ *   ctx->Driver.TexImage3D = _mesa_store_teximage3d;
+ *   etc...
+ *
+ * Texture image processing is actually kind of complicated.  We have to do:
+ *    Format/type conversions
+ *    pixel unpacking
+ *    pixel transfer (scale, bais, lookup, convolution!, etc)
+ *
+ * These functions can handle most everything, including processing full
+ * images and sub-images.
+ */
+
+
+#include "glheader.h"
+#include "bufferobj.h"
 #include "colormac.h"
 #include "context.h"
 #include "convolve.h"
 #include "image.h"
 #include "macros.h"
-#include "mem.h"
+#include "imports.h"
+#include "texcompress.h"
 #include "texformat.h"
 #include "teximage.h"
 #include "texstore.h"
-#include "texutil.h"
 
 
-/*
- * 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 the
- * number of components for the format.  Return -1 if invalid enum.
- *
- * GH: Do we really need this?  We have the number of bytes per texel
- * in the texture format structures, so why don't we just use that?
- */
-static GLint
-components_in_intformat( GLint format )
+static const GLint ZERO = 4, ONE = 5;
+
+static GLboolean can_swizzle(GLenum logicalBaseFormat)
+{
+   switch (logicalBaseFormat) {
+   case GL_RGBA:
+   case GL_RGB:
+   case GL_LUMINANCE_ALPHA:
+   case GL_INTENSITY:
+   case GL_ALPHA:
+   case GL_LUMINANCE:
+      return GL_TRUE;
+   default:
+      return GL_FALSE;
+   }
+}
+
+
+/**
+ * When promoting texture formats (see below) we need to compute the
+ * mapping of dest components back to source components.
+ * This function does that.
+ * \param logicalBaseFormat  the logical format of the texture
+ * \param textureBaseFormat  the final texture format
+ * \return map[4]  the four mapping values
+ */
+static void
+compute_component_mapping(GLenum logicalBaseFormat, GLenum textureBaseFormat,
+                          GLubyte map[6])
+{
+   map[ZERO] = ZERO;
+   map[ONE] = ONE;
+
+   /* compute mapping from dest components back to src components */
+   switch (textureBaseFormat) {
+   case GL_RGB:
+   case GL_RGBA:
+      switch (logicalBaseFormat) {
+      case GL_LUMINANCE:
+         map[0] = map[1] = map[2] = 0;
+         if (textureBaseFormat == GL_RGBA)
+            map[3] = ONE;
+         break;
+      case GL_ALPHA:
+         ASSERT(textureBaseFormat == GL_RGBA);
+         map[0] = map[1] = map[2] = ZERO;
+         map[3] = 0;
+         break;
+      case GL_INTENSITY:
+         map[0] = map[1] = map[2] = 0;
+         if (textureBaseFormat == GL_RGBA)
+            map[3] = 0;
+         break;
+      case GL_LUMINANCE_ALPHA:
+         ASSERT(textureBaseFormat == GL_RGBA);
+         map[0] = map[1] = map[2] = 0;
+         map[3] = 1;
+         break;
+      case GL_RGB:
+         ASSERT(textureBaseFormat == GL_RGBA);
+         map[0] = 0;
+         map[1] = 1;
+         map[2] = 2;
+         map[3] = ONE;
+         break;
+      case GL_RGBA:
+         ASSERT(textureBaseFormat == GL_RGBA);
+         map[0] = 0;
+         map[1] = 1;
+         map[2] = 2;
+         map[3] = 3;
+         break;
+      default:
+         _mesa_problem(NULL, "Unexpected logicalBaseFormat");
+         map[0] = map[1] = map[2] = map[3] = 0;
+      }
+      break;
+   case GL_LUMINANCE_ALPHA:
+      switch (logicalBaseFormat) {
+      case GL_LUMINANCE:
+         map[0] = 0;
+         map[1] = ONE;
+         break;
+      case GL_ALPHA:
+         map[0] = ZERO;
+         map[1] = 0;
+         break;
+      case GL_INTENSITY:
+         map[0] = 0;
+         map[1] = 0;
+         break;
+      default:
+         _mesa_problem(NULL, "Unexpected logicalBaseFormat");
+         map[0] = map[1] = 0;
+      }
+      break;
+   default:
+      _mesa_problem(NULL, "Unexpected logicalBaseFormat");
+      map[0] = map[1] = 0;
+      break;
+   }   
+}
+
+
+/**
+ * Make a temporary (color) texture image with GLfloat components.
+ * Apply all needed pixel unpacking and pixel transfer operations.
+ * Note that there are both logicalBaseFormat and textureBaseFormat parameters.
+ * Suppose the user specifies GL_LUMINANCE as the internal texture format
+ * but the graphics hardware doesn't support luminance textures.  So, might
+ * use an RGB hardware format instead.
+ * If logicalBaseFormat != textureBaseFormat we have some extra work to do.
+ *
+ * \param ctx  the rendering context
+ * \param dims  image dimensions: 1, 2 or 3
+ * \param logicalBaseFormat  basic texture derived from the user's
+ *    internal texture format value
+ * \param textureBaseFormat  the actual basic format of the texture
+ * \param srcWidth  source image width
+ * \param srcHeight  source image height
+ * \param srcDepth  source image depth
+ * \param srcFormat  source image format
+ * \param srcType  source image type
+ * \param srcAddr  source image address
+ * \param srcPacking  source image pixel packing
+ * \return resulting image with format = textureBaseFormat and type = GLfloat.
+ */
+static GLfloat *
+make_temp_float_image(GLcontext *ctx, GLuint dims,
+                      GLenum logicalBaseFormat,
+                      GLenum textureBaseFormat,
+                      GLint srcWidth, GLint srcHeight, GLint srcDepth,
+                      GLenum srcFormat, GLenum srcType,
+                      const GLvoid *srcAddr,
+                      const struct gl_pixelstore_attrib *srcPacking)
+{
+   GLuint transferOps = ctx->_ImageTransferState;
+   GLfloat *tempImage;
+
+   ASSERT(dims >= 1 && dims <= 3);
+
+   ASSERT(logicalBaseFormat == GL_RGBA ||
+          logicalBaseFormat == GL_RGB ||
+          logicalBaseFormat == GL_LUMINANCE_ALPHA ||
+          logicalBaseFormat == GL_LUMINANCE ||
+          logicalBaseFormat == GL_ALPHA ||
+          logicalBaseFormat == GL_INTENSITY ||
+          logicalBaseFormat == GL_COLOR_INDEX ||
+          logicalBaseFormat == GL_DEPTH_COMPONENT);
+
+   ASSERT(textureBaseFormat == GL_RGBA ||
+          textureBaseFormat == GL_RGB ||
+          textureBaseFormat == GL_LUMINANCE_ALPHA ||
+          textureBaseFormat == GL_LUMINANCE ||
+          textureBaseFormat == GL_ALPHA ||
+          textureBaseFormat == GL_INTENSITY ||
+          textureBaseFormat == GL_COLOR_INDEX ||
+          textureBaseFormat == GL_DEPTH_COMPONENT);
+
+   /* conventional color image */
+
+   if ((dims == 1 && ctx->Pixel.Convolution1DEnabled) ||
+       (dims >= 2 && ctx->Pixel.Convolution2DEnabled) ||
+       (dims >= 2 && ctx->Pixel.Separable2DEnabled)) {
+      /* need image convolution */
+      const GLuint preConvTransferOps
+         = (transferOps & IMAGE_PRE_CONVOLUTION_BITS) | IMAGE_CLAMP_BIT;
+      const GLuint postConvTransferOps
+         = (transferOps & IMAGE_POST_CONVOLUTION_BITS) | IMAGE_CLAMP_BIT;
+      GLint img, row;
+      GLint convWidth, convHeight;
+      GLfloat *convImage;
+
+      /* pre-convolution image buffer (3D) */
+      tempImage = (GLfloat *) _mesa_malloc(srcWidth * srcHeight * srcDepth
+                                           * 4 * sizeof(GLfloat));
+      if (!tempImage)
+         return NULL;
+
+      /* post-convolution image buffer (2D) */
+      convImage = (GLfloat *) _mesa_malloc(srcWidth * srcHeight
+                                           * 4 * sizeof(GLfloat));
+      if (!convImage) {
+         _mesa_free(tempImage);
+         return NULL;
+      }
+
+      /* loop over 3D image slices */
+      for (img = 0; img < srcDepth; img++) {
+         GLfloat *dst = tempImage + img * (srcWidth * srcHeight * 4);
+
+         /* unpack and do transfer ops up to convolution */
+         for (row = 0; row < srcHeight; row++) {
+            const GLvoid *src = _mesa_image_address(dims, srcPacking,
+                                              srcAddr, srcWidth, srcHeight,
+                                              srcFormat, srcType, img, row, 0);
+            _mesa_unpack_color_span_float(ctx, srcWidth, GL_RGBA, dst,
+                                          srcFormat, srcType, src,
+                                          srcPacking,
+                                          preConvTransferOps);
+            dst += srcWidth * 4;
+         }
+
+         /* do convolution */
+         {
+            GLfloat *src = tempImage + img * (srcWidth * srcHeight * 4);
+            convWidth = srcWidth;
+            convHeight = srcHeight;
+            if (dims == 1) {
+               ASSERT(ctx->Pixel.Convolution1DEnabled);
+               _mesa_convolve_1d_image(ctx, &convWidth, src, convImage);
+            }
+            else {
+               if (ctx->Pixel.Convolution2DEnabled) {
+                  _mesa_convolve_2d_image(ctx, &convWidth, &convHeight,
+                                          src, convImage);
+               }
+               else {
+                  ASSERT(ctx->Pixel.Separable2DEnabled);
+                  _mesa_convolve_sep_image(ctx, &convWidth, &convHeight,
+                                           src, convImage);
+               }
+            }
+         }
+
+         /* do post-convolution transfer and pack into tempImage */
+         {
+            const GLint logComponents
+               = _mesa_components_in_format(logicalBaseFormat);
+            const GLfloat *src = convImage;
+            GLfloat *dst = tempImage + img * (convWidth * convHeight * 4);
+            for (row = 0; row < convHeight; row++) {
+               _mesa_pack_rgba_span_float(ctx, convWidth,
+                                          (const GLfloat (*)[4]) src,
+                                          logicalBaseFormat, GL_FLOAT,
+                                          dst, &ctx->DefaultPacking,
+                                          postConvTransferOps);
+               src += convWidth * 4;
+               dst += convWidth * logComponents;
+            }
+         }
+      } /* loop over 3D image slices */
+
+      _mesa_free(convImage);
+
+      /* might need these below */
+      srcWidth = convWidth;
+      srcHeight = convHeight;
+   }
+   else {
+      /* no convolution */
+      const GLint components = _mesa_components_in_format(logicalBaseFormat);
+      const GLint srcStride = _mesa_image_row_stride(srcPacking,
+                                                 srcWidth, srcFormat, srcType);
+      GLfloat *dst;
+      GLint img, row;
+
+      tempImage = (GLfloat *) _mesa_malloc(srcWidth * srcHeight * srcDepth
+                                           * components * sizeof(GLfloat));
+      if (!tempImage)
+         return NULL;
+
+      dst = tempImage;
+      for (img = 0; img < srcDepth; img++) {
+         const GLubyte *src
+            = (const GLubyte *) _mesa_image_address(dims, srcPacking, srcAddr,
+                                                    srcWidth, srcHeight,
+                                                    srcFormat, srcType,
+                                                    img, 0, 0);
+         for (row = 0; row < srcHeight; row++) {
+            _mesa_unpack_color_span_float(ctx, srcWidth, logicalBaseFormat,
+                                          dst, srcFormat, srcType, src,
+                                          srcPacking, transferOps);
+            dst += srcWidth * components;
+            src += srcStride;
+         }
+      }
+   }
+
+   if (logicalBaseFormat != textureBaseFormat) {
+      /* more work */
+      GLint texComponents = _mesa_components_in_format(textureBaseFormat);
+      GLint logComponents = _mesa_components_in_format(logicalBaseFormat);
+      GLfloat *newImage;
+      GLint i, n;
+      GLubyte map[6];
+
+      /* we only promote up to RGB, RGBA and LUMINANCE_ALPHA formats for now */
+      ASSERT(textureBaseFormat == GL_RGB || textureBaseFormat == GL_RGBA ||
+             textureBaseFormat == GL_LUMINANCE_ALPHA);
+
+      /* The actual texture format should have at least as many components
+       * as the logical texture format.
+       */
+      ASSERT(texComponents >= logComponents);
+
+      newImage = (GLfloat *) _mesa_malloc(srcWidth * srcHeight * srcDepth
+                                          * texComponents * sizeof(GLfloat));
+      if (!newImage) {
+         _mesa_free(tempImage);
+         return NULL;
+      }
+
+      compute_component_mapping(logicalBaseFormat, textureBaseFormat, map);
+
+      n = srcWidth * srcHeight * srcDepth;
+      for (i = 0; i < n; i++) {
+         GLint k;
+         for (k = 0; k < texComponents; k++) {
+            GLint j = map[k];
+            if (j == ZERO)
+               newImage[i * texComponents + k] = 0.0F;
+            else if (j == ONE)
+               newImage[i * texComponents + k] = 1.0F;
+            else
+               newImage[i * texComponents + k] = tempImage[i * logComponents + j];
+         }
+      }
+
+      _mesa_free(tempImage);
+      tempImage = newImage;
+   }
+
+   return tempImage;
+}
+
+
+/**
+ * Make a temporary (color) texture image with GLchan components.
+ * Apply all needed pixel unpacking and pixel transfer operations.
+ * Note that there are both logicalBaseFormat and textureBaseFormat parameters.
+ * Suppose the user specifies GL_LUMINANCE as the internal texture format
+ * but the graphics hardware doesn't support luminance textures.  So, might
+ * use an RGB hardware format instead.
+ * If logicalBaseFormat != textureBaseFormat we have some extra work to do.
+ *
+ * \param ctx  the rendering context
+ * \param dims  image dimensions: 1, 2 or 3
+ * \param logicalBaseFormat  basic texture derived from the user's
+ *    internal texture format value
+ * \param textureBaseFormat  the actual basic format of the texture
+ * \param srcWidth  source image width
+ * \param srcHeight  source image height
+ * \param srcDepth  source image depth
+ * \param srcFormat  source image format
+ * \param srcType  source image type
+ * \param srcAddr  source image address
+ * \param srcPacking  source image pixel packing
+ * \return resulting image with format = textureBaseFormat and type = GLchan.
+ */
+GLchan *
+_mesa_make_temp_chan_image(GLcontext *ctx, GLuint dims,
+                           GLenum logicalBaseFormat,
+                           GLenum textureBaseFormat,
+                           GLint srcWidth, GLint srcHeight, GLint srcDepth,
+                           GLenum srcFormat, GLenum srcType,
+                           const GLvoid *srcAddr,
+                           const struct gl_pixelstore_attrib *srcPacking)
+{
+   GLuint transferOps = ctx->_ImageTransferState;
+   const GLint components = _mesa_components_in_format(logicalBaseFormat);
+   GLboolean freeSrcImage = GL_FALSE;
+   GLint img, row;
+   GLchan *tempImage, *dst;
+
+   ASSERT(dims >= 1 && dims <= 3);
+
+   ASSERT(logicalBaseFormat == GL_RGBA ||
+          logicalBaseFormat == GL_RGB ||
+          logicalBaseFormat == GL_LUMINANCE_ALPHA ||
+          logicalBaseFormat == GL_LUMINANCE ||
+          logicalBaseFormat == GL_ALPHA ||
+          logicalBaseFormat == GL_INTENSITY);
+
+   ASSERT(textureBaseFormat == GL_RGBA ||
+          textureBaseFormat == GL_RGB ||
+          textureBaseFormat == GL_LUMINANCE_ALPHA ||
+          textureBaseFormat == GL_LUMINANCE ||
+          textureBaseFormat == GL_ALPHA ||
+          textureBaseFormat == GL_INTENSITY);
+
+   if ((dims == 1 && ctx->Pixel.Convolution1DEnabled) ||
+       (dims >= 2 && ctx->Pixel.Convolution2DEnabled) ||
+       (dims >= 2 && ctx->Pixel.Separable2DEnabled)) {
+      /* get convolved image */
+      GLfloat *convImage = make_temp_float_image(ctx, dims,
+                                                 logicalBaseFormat,
+                                                 logicalBaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType,
+                                                 srcAddr, srcPacking);
+      if (!convImage)
+         return NULL;
+      /* the convolved image is our new source image */
+      srcAddr = convImage;
+      srcFormat = logicalBaseFormat;
+      srcType = GL_FLOAT;
+      srcPacking = &ctx->DefaultPacking;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      transferOps = 0;
+      freeSrcImage = GL_TRUE;
+   }
+
+   /* unpack and transfer the source image */
+   tempImage = (GLchan *) _mesa_malloc(srcWidth * srcHeight * srcDepth
+                                       * components * sizeof(GLchan));
+   if (!tempImage)
+      return NULL;
+
+   dst = tempImage;
+   for (img = 0; img < srcDepth; img++) {
+      const GLint srcStride = _mesa_image_row_stride(srcPacking,
+                                                     srcWidth, srcFormat,
+                                                     srcType);
+      const GLubyte *src
+         = (const GLubyte *) _mesa_image_address(dims, srcPacking, srcAddr,
+                                                 srcWidth, srcHeight,
+                                                 srcFormat, srcType,
+                                                 img, 0, 0);
+      for (row = 0; row < srcHeight; row++) {
+         _mesa_unpack_color_span_chan(ctx, srcWidth, logicalBaseFormat, dst,
+                                      srcFormat, srcType, src, srcPacking,
+                                      transferOps);
+         dst += srcWidth * components;
+         src += srcStride;
+      }
+   }
+
+   /* If we made a temporary image for convolution, free it here */
+   if (freeSrcImage) {
+      _mesa_free((void *) srcAddr);
+   }
+
+   if (logicalBaseFormat != textureBaseFormat) {
+      /* one more conversion step */
+      GLint texComponents = _mesa_components_in_format(textureBaseFormat);
+      GLint logComponents = _mesa_components_in_format(logicalBaseFormat);
+      GLchan *newImage;
+      GLint i, n;
+      GLubyte map[6];
+
+      /* we only promote up to RGB, RGBA and LUMINANCE_ALPHA formats for now */
+      ASSERT(textureBaseFormat == GL_RGB || textureBaseFormat == GL_RGBA ||
+             textureBaseFormat == GL_LUMINANCE_ALPHA);
+
+      /* The actual texture format should have at least as many components
+       * as the logical texture format.
+       */
+      ASSERT(texComponents >= logComponents);
+
+      newImage = (GLchan *) _mesa_malloc(srcWidth * srcHeight * srcDepth
+                                          * texComponents * sizeof(GLchan));
+      if (!newImage) {
+         _mesa_free(tempImage);
+         return NULL;
+      }
+
+      compute_component_mapping(logicalBaseFormat, textureBaseFormat, map);
+
+      n = srcWidth * srcHeight * srcDepth;
+      for (i = 0; i < n; i++) {
+         GLint k;
+         for (k = 0; k < texComponents; k++) {
+            GLint j = map[k];
+            if (j == ZERO)
+               newImage[i * texComponents + k] = 0;
+            else if (j == ONE)
+               newImage[i * texComponents + k] = CHAN_MAX;
+            else
+               newImage[i * texComponents + k] = tempImage[i * logComponents + j];
+         }
+      }
+
+      _mesa_free(tempImage);
+      tempImage = newImage;
+   }
+
+   return tempImage;
+}
+
+
+static void swizzle_copy(GLubyte *dst,
+                        GLuint dstComponents,
+                        const GLubyte *src, 
+                        GLuint srcComponents,
+                        GLubyte *map,
+                        GLuint count)
+{
+   GLubyte tmp[8];
+   GLint i;
+
+   tmp[ZERO] = 0x0;
+   tmp[ONE] = 0xff;
+
+   switch (dstComponents) {
+   case 4:
+      for (i = 0; i < count; i++) {
+        COPY_4UBV(tmp, src); 
+        src += srcComponents;      
+        dst[0] = tmp[map[0]];
+        dst[1] = tmp[map[1]];
+        dst[2] = tmp[map[2]];
+        dst[3] = tmp[map[3]];
+        dst += 4;
+      }
+      break;
+   case 3:
+      for (i = 0; i < count; i++) {
+        COPY_4UBV(tmp, src); 
+        src += srcComponents;      
+        dst[0] = tmp[map[0]];
+        dst[1] = tmp[map[1]];
+        dst[2] = tmp[map[2]];
+        dst += 3;
+      }
+      break;
+   case 2:
+      for (i = 0; i < count; i++) {
+        COPY_4UBV(tmp, src); 
+        src += srcComponents;      
+        dst[0] = tmp[map[0]];
+        dst[1] = tmp[map[1]];
+        dst += 2;
+      }
+      break;
+   }
+}
+
+
+static void
+_mesa_swizzle_ubyte_image(GLcontext *ctx, 
+                         GLuint dimensions,
+                         GLenum srcFormat,
+                         const GLubyte *dstmap, GLint dstComponents,
+
+                         GLvoid *dstAddr,
+                         GLint dstXoffset, GLint dstYoffset, GLint dstZoffset,
+                         GLint dstRowStride, GLint dstImageStride,
+
+                         GLint srcWidth, GLint srcHeight, GLint srcDepth,
+                         const GLvoid *srcAddr,
+                         const struct gl_pixelstore_attrib *srcPacking )
+{
+   GLint srcComponents = _mesa_components_in_format(srcFormat);
+   GLubyte srcmap[6], map[4];
+   GLint i;
+
+   const GLint srcRowStride = _mesa_image_row_stride(srcPacking, srcWidth,
+                                                     srcFormat, GL_UNSIGNED_BYTE);
+   const GLint srcImageStride = _mesa_image_image_stride(srcPacking,
+                                      srcWidth, srcHeight, srcFormat, GL_UNSIGNED_BYTE);
+   const GLubyte *srcImage = (const GLubyte *) _mesa_image_address(dimensions,
+        srcPacking, srcAddr, srcWidth, srcHeight, srcFormat, GL_UNSIGNED_BYTE, 0, 0, 0);
+
+   GLubyte *dstImage = (GLubyte *) dstAddr
+                     + dstZoffset * dstImageStride
+                     + dstYoffset * dstRowStride
+                     + dstXoffset * dstComponents;
+
+   compute_component_mapping(srcFormat, GL_RGBA, srcmap);
+
+   for (i = 0; i < 4; i++)
+      map[i] = srcmap[dstmap[i]];
+
+   if (srcRowStride == srcWidth * srcComponents &&
+       (srcImageStride == srcWidth * srcHeight * srcComponents ||
+        srcDepth == 1)) {
+      swizzle_copy(dstImage, dstComponents, srcImage, srcComponents, map, 
+                  srcWidth * srcHeight * srcDepth);
+   }
+   else {
+      GLint img, row;
+      for (img = 0; img < srcDepth; img++) {
+         const GLubyte *srcRow = srcImage;
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+           swizzle_copy(dstRow, dstComponents, srcRow, srcComponents, map, srcWidth);
+            dstRow += dstRowStride;
+            srcRow += srcRowStride;
+         }
+         srcImage += srcImageStride;
+         dstImage += dstImageStride;
+      }
+   }
+}
+
+
+
+
+
+/**
+ * Teximage storage routine for when a simple memcpy will do.
+ * No pixel transfer operations or special texel encodings allowed.
+ * 1D, 2D and 3D images supported.
+ */
+static void
+memcpy_texture(GLuint dimensions,
+               const struct gl_texture_format *dstFormat,
+               GLvoid *dstAddr,
+               GLint dstXoffset, GLint dstYoffset, GLint dstZoffset,
+               GLint dstRowStride, GLint dstImageStride,
+               GLint srcWidth, GLint srcHeight, GLint srcDepth,
+               GLenum srcFormat, GLenum srcType,
+               const GLvoid *srcAddr,
+               const struct gl_pixelstore_attrib *srcPacking)
+{
+   const GLint srcRowStride = _mesa_image_row_stride(srcPacking, srcWidth,
+                                                     srcFormat, srcType);
+   const GLint srcImageStride = _mesa_image_image_stride(srcPacking,
+                                      srcWidth, srcHeight, srcFormat, srcType);
+   const GLubyte *srcImage = (const GLubyte *) _mesa_image_address(dimensions,
+        srcPacking, srcAddr, srcWidth, srcHeight, srcFormat, srcType, 0, 0, 0);
+   const GLint bytesPerRow = srcWidth * dstFormat->TexelBytes;
+   const GLint bytesPerImage = srcHeight * bytesPerRow;
+   const GLint bytesPerTexture = srcDepth * bytesPerImage;
+   GLubyte *dstImage = (GLubyte *) dstAddr
+                     + dstZoffset * dstImageStride
+                     + dstYoffset * dstRowStride
+                     + dstXoffset * dstFormat->TexelBytes;
+
+   if (dstRowStride == srcRowStride &&
+       dstRowStride == bytesPerRow &&
+       ((dstImageStride == srcImageStride &&
+         dstImageStride == bytesPerImage) ||
+        (srcDepth == 1))) {
+      /* one big memcpy */
+      _mesa_memcpy(dstImage, srcImage, bytesPerTexture);
+   }
+   else {
+      GLint img, row;
+      for (img = 0; img < srcDepth; img++) {
+         const GLubyte *srcRow = srcImage;
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            _mesa_memcpy(dstRow, srcRow, bytesPerRow);
+            dstRow += dstRowStride;
+            srcRow += srcRowStride;
+         }
+         srcImage += srcImageStride;
+         dstImage += dstImageStride;
+      }
+   }
+}
+
+
+
+/**
+ * Store an image in any of the formats:
+ *   _mesa_texformat_rgba
+ *   _mesa_texformat_rgb
+ *   _mesa_texformat_alpha
+ *   _mesa_texformat_luminance
+ *   _mesa_texformat_luminance_alpha
+ *   _mesa_texformat_intensity
+ * 
+ * \param dims  either 1 or 2 or 3
+ * \param baseInternalFormat  user-specified base internal format
+ * \param dstFormat  destination Mesa texture format
+ * \param dstAddr  destination image address
+ * \param dstX/Y/Zoffset  destination x/y/z offset (ala TexSubImage), in texels
+ * \param dstRowStride  destination image row stride, in bytes
+ * \param dstImageStride  destination image layer stride, in bytes
+ * \param srcWidth/Height/Depth  source image size, in pixels
+ * \param srcFormat  incoming image format
+ * \param srcType  incoming image data type
+ * \param srcAddr  source image address
+ * \param srcPacking  source image packing parameters
+ */
+GLboolean
+_mesa_texstore_rgba(GLcontext *ctx, GLuint dims,
+                    GLenum baseInternalFormat,
+                    const struct gl_texture_format *dstFormat,
+                    GLvoid *dstAddr,
+                    GLint dstXoffset, GLint dstYoffset, GLint dstZoffset,
+                    GLint dstRowStride, GLint dstImageStride,
+                    GLint srcWidth, GLint srcHeight, GLint srcDepth,
+                    GLenum srcFormat, GLenum srcType,
+                    const GLvoid *srcAddr,
+                    const struct gl_pixelstore_attrib *srcPacking)
+{
+   const GLint components = _mesa_components_in_format(baseInternalFormat);
+
+   ASSERT(dstFormat == &_mesa_texformat_rgba ||
+          dstFormat == &_mesa_texformat_rgb ||
+          dstFormat == &_mesa_texformat_alpha ||
+          dstFormat == &_mesa_texformat_luminance ||
+          dstFormat == &_mesa_texformat_luminance_alpha ||
+          dstFormat == &_mesa_texformat_intensity);
+   ASSERT(baseInternalFormat == GL_RGBA ||
+          baseInternalFormat == GL_RGB ||
+          baseInternalFormat == GL_ALPHA ||
+          baseInternalFormat == GL_LUMINANCE ||
+          baseInternalFormat == GL_LUMINANCE_ALPHA ||
+          baseInternalFormat == GL_INTENSITY);
+   ASSERT(dstFormat->TexelBytes == components * sizeof(GLchan));
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       baseInternalFormat == srcFormat &&
+       srcType == CHAN_TYPE) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else if (!ctx->_ImageTransferState &&
+            !srcPacking->SwapBytes &&
+            dstFormat == &_mesa_texformat_rgb &&
+            srcFormat == GL_RGBA &&
+            srcType == CHAN_TYPE) {
+      /* extract RGB from RGBA */
+      int img, row, col;
+      GLchan *dstImage = (GLchan *) (GLubyte *) dstAddr
+                       + dstZoffset * dstImageStride
+                       + dstYoffset * dstRowStride
+                       + dstXoffset * dstFormat->TexelBytes;
+      for (img = 0; img < srcDepth; img++) {
+         const GLint srcRowStride = _mesa_image_row_stride(srcPacking,
+                                                 srcWidth, srcFormat, srcType);
+         GLchan *srcRow = (GLchan *) _mesa_image_address(dims, srcPacking,
+                  srcAddr, srcWidth, srcHeight, srcFormat, srcType, img, 0, 0);
+         GLchan *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            for (col = 0; col < srcWidth; col++) {
+               dstRow[col * 3 + RCOMP] = srcRow[col * 4 + RCOMP];
+               dstRow[col * 3 + GCOMP] = srcRow[col * 4 + GCOMP];
+               dstRow[col * 3 + BCOMP] = srcRow[col * 4 + BCOMP];
+            }
+            dstRow += dstRowStride;
+            srcRow = (GLchan *) ((GLubyte *) srcRow + srcRowStride);
+         }
+         dstImage += dstImageStride;
+      }
+   }
+   else {
+      /* general path */
+      const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLchan *src = tempImage;
+      GLint bytesPerRow;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      bytesPerRow = srcWidth * components * sizeof(GLchan);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            _mesa_memcpy(dstRow, src, bytesPerRow);
+            dstRow += dstRowStride;
+            src += srcWidth * components;
+         }
+         dstImage += dstImageStride;
+      }
+
+      _mesa_free((void *) tempImage);
+   }
+   return GL_TRUE;
+}
+
+
+/**
+ * Store a floating point depth component texture image.
+ */
+GLboolean
+_mesa_texstore_depth_component_float32(STORE_PARAMS)
+{
+   (void) dims;
+   ASSERT(dstFormat == &_mesa_texformat_depth_component_float32);
+   ASSERT(dstFormat->TexelBytes == sizeof(GLfloat));
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       baseInternalFormat == GL_DEPTH_COMPONENT &&
+       srcFormat == GL_DEPTH_COMPONENT &&
+       srcType == GL_FLOAT) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else {
+      /* general path */
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row;
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            const GLvoid *src = _mesa_image_address(dims, srcPacking,
+                srcAddr, srcWidth, srcHeight, srcFormat, srcType, img, row, 0);
+            _mesa_unpack_depth_span(ctx, srcWidth, (GLfloat *) dstRow,
+                                    srcType, src, srcPacking);
+            dstRow += dstRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+   }
+   return GL_TRUE;
+}
+
+
+/**
+ * Store a 16-bit integer depth component texture image.
+ */
+GLboolean
+_mesa_texstore_depth_component16(STORE_PARAMS)
+{
+   (void) dims;
+   ASSERT(dstFormat == &_mesa_texformat_depth_component16);
+   ASSERT(dstFormat->TexelBytes == sizeof(GLushort));
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       baseInternalFormat == GL_DEPTH_COMPONENT &&
+       srcFormat == GL_DEPTH_COMPONENT &&
+       srcType == GL_UNSIGNED_SHORT) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else {
+      /* general path */
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row, col;
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            GLfloat depthTemp[MAX_WIDTH];
+            const GLvoid *src = _mesa_image_address(dims, srcPacking,
+                srcAddr, srcWidth, srcHeight, srcFormat, srcType, img, row, 0);
+            GLushort *dst16 = (GLushort *) dstRow;
+            _mesa_unpack_depth_span(ctx, srcWidth, depthTemp,
+                                    srcType, src, srcPacking);
+            for (col = 0; col < srcWidth; col++) {
+               dst16[col] = (GLushort) (depthTemp[col] * 65535.0F);
+            }
+            dstRow += dstRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+   }
+   return GL_TRUE;
+}
+
+
+/**
+ * Store an rgb565 or rgb565_rev texture image.
+ */
+GLboolean
+_mesa_texstore_rgb565(STORE_PARAMS)
+{
+   ASSERT(dstFormat == &_mesa_texformat_rgb565 ||
+          dstFormat == &_mesa_texformat_rgb565_rev);
+   ASSERT(dstFormat->TexelBytes == 2);
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       dstFormat == &_mesa_texformat_rgb565 &&
+       baseInternalFormat == GL_RGB &&
+       srcFormat == GL_RGB &&
+       srcType == GL_UNSIGNED_SHORT_5_6_5) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else if (!ctx->_ImageTransferState &&
+            !srcPacking->SwapBytes &&
+            baseInternalFormat == GL_RGB &&
+            srcFormat == GL_RGB &&
+            srcType == GL_UNSIGNED_BYTE &&
+            dims == 2) {
+      /* do optimized tex store */
+      const GLint srcRowStride = _mesa_image_row_stride(srcPacking, srcWidth,
+                                                        srcFormat, srcType);
+      const GLubyte *src = (const GLubyte *)
+         _mesa_image_address(dims, srcPacking, srcAddr, srcWidth, srcHeight,
+                             srcFormat, srcType, 0, 0, 0);
+      GLubyte *dst = (GLubyte *) dstAddr
+                   + dstZoffset * dstImageStride
+                   + dstYoffset * dstRowStride
+                   + dstXoffset * dstFormat->TexelBytes;
+      GLint row, col;
+      for (row = 0; row < srcHeight; row++) {
+         const GLubyte *srcUB = (const GLubyte *) src;
+         GLushort *dstUS = (GLushort *) dst;
+         /* check for byteswapped format */
+         if (dstFormat == &_mesa_texformat_rgb565) {
+            for (col = 0; col < srcWidth; col++) {
+               dstUS[col] = PACK_COLOR_565( srcUB[0], srcUB[1], srcUB[2] );
+               srcUB += 3;
+            }
+         }
+         else {
+            for (col = 0; col < srcWidth; col++) {
+               dstUS[col] = PACK_COLOR_565_REV( srcUB[0], srcUB[1], srcUB[2] );
+               srcUB += 3;
+            }
+         }
+         dst += dstRowStride;
+         src += srcRowStride;
+      }
+   }
+   else {
+      /* general path */
+      const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLchan *src = tempImage;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row, col;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            GLushort *dstUS = (GLushort *) dstRow;
+            /* check for byteswapped format */
+            if (dstFormat == &_mesa_texformat_rgb565) {
+               for (col = 0; col < srcWidth; col++) {
+                  dstUS[col] = PACK_COLOR_565( CHAN_TO_UBYTE(src[RCOMP]),
+                                               CHAN_TO_UBYTE(src[GCOMP]),
+                                               CHAN_TO_UBYTE(src[BCOMP]) );
+                  src += 3;
+               }
+            }
+            else {
+               for (col = 0; col < srcWidth; col++) {
+                  dstUS[col] = PACK_COLOR_565_REV( CHAN_TO_UBYTE(src[RCOMP]),
+                                                   CHAN_TO_UBYTE(src[GCOMP]),
+                                                   CHAN_TO_UBYTE(src[BCOMP]) );
+                  src += 3;
+               }
+            }
+            dstRow += dstRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+      _mesa_free((void *) tempImage);
+   }
+   return GL_TRUE;
+}
+
+
+GLboolean
+_mesa_texstore_rgba8888(STORE_PARAMS)
+{
+   const GLuint ui = 1;
+   const GLubyte littleEndian = *((const GLubyte *) &ui);
+
+   (void)littleEndian;
+   ASSERT(dstFormat == &_mesa_texformat_rgba8888 ||
+          dstFormat == &_mesa_texformat_rgba8888_rev);
+   ASSERT(dstFormat->TexelBytes == 4);
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       dstFormat == &_mesa_texformat_rgba8888 &&
+       baseInternalFormat == GL_RGBA &&
+      ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8) ||
+       (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8_REV))) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+#if 0
+   else if (!ctx->_ImageTransferState &&
+           !srcPacking->SwapBytes &&
+           srcType == GL_UNSIGNED_BYTE && 
+           dstFormat == &_mesa_texformat_rgba8888 &&
+           littleEndian &&
+           can_swizzle(srcFormat)) {
+      GLubyte dstmap[4];
+
+      /* dstmap - how to swizzle from GL_RGBA to dst format:
+       *
+       * FIXME - add !litteEndian and _rev varients:
+       */
+      dstmap[3] = 0;
+      dstmap[2] = 1;
+      dstmap[1] = 2;
+      dstmap[0] = 3;
+      
+      _mesa_swizzle_ubyte_image(ctx, dims,
+                               srcFormat,
+                               dstmap, 4,
+                               dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                               dstRowStride, dstImageStride,
+                               srcWidth, srcHeight, srcDepth, srcAddr,
+                               srcPacking);      
+   }
+#endif
+   else {
+      /* general path */
+      const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLchan *src = tempImage;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row, col;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            GLuint *dstUI = (GLuint *) dstRow;
+            if (dstFormat == &_mesa_texformat_rgba8888) {
+               for (col = 0; col < srcWidth; col++) {
+                  dstUI[col] = PACK_COLOR_8888( CHAN_TO_UBYTE(src[RCOMP]),
+                                                CHAN_TO_UBYTE(src[GCOMP]),
+                                                CHAN_TO_UBYTE(src[BCOMP]),
+                                                CHAN_TO_UBYTE(src[ACOMP]) );
+                  src += 4;
+               }
+            }
+            else {
+               for (col = 0; col < srcWidth; col++) {
+                  dstUI[col] = PACK_COLOR_8888_REV( CHAN_TO_UBYTE(src[RCOMP]),
+                                                    CHAN_TO_UBYTE(src[GCOMP]),
+                                                    CHAN_TO_UBYTE(src[BCOMP]),
+                                                    CHAN_TO_UBYTE(src[ACOMP]) );
+                  src += 4;
+               }
+            }
+            dstRow += dstRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+      _mesa_free((void *) tempImage);
+   }
+   return GL_TRUE;
+}
+
+
+GLboolean
+_mesa_texstore_argb8888(STORE_PARAMS)
+{
+   const GLuint ui = 1;
+   const GLubyte littleEndian = *((const GLubyte *) &ui);
+
+   ASSERT(dstFormat == &_mesa_texformat_argb8888 ||
+          dstFormat == &_mesa_texformat_argb8888_rev);
+   ASSERT(dstFormat->TexelBytes == 4);
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       dstFormat == &_mesa_texformat_argb8888 &&
+       baseInternalFormat == GL_RGBA &&
+       srcFormat == GL_BGRA &&
+       ((srcType == GL_UNSIGNED_BYTE && littleEndian) ||
+        srcType == GL_UNSIGNED_INT_8_8_8_8_REV)) {
+      /* simple memcpy path (little endian) */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       dstFormat == &_mesa_texformat_argb8888_rev &&
+       baseInternalFormat == GL_RGBA &&
+       srcFormat == GL_BGRA &&
+       ((srcType == GL_UNSIGNED_BYTE && !littleEndian) ||
+        srcType == GL_UNSIGNED_INT_8_8_8_8)) {
+      /* simple memcpy path (big endian) */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else if (!ctx->_ImageTransferState &&
+            !srcPacking->SwapBytes &&
+           dstFormat == &_mesa_texformat_argb8888 &&
+            srcFormat == GL_RGB &&
+            srcType == GL_UNSIGNED_BYTE) {
+
+      int img, row, col;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      for (img = 0; img < srcDepth; img++) {
+         const GLint srcRowStride = _mesa_image_row_stride(srcPacking,
+                                                 srcWidth, srcFormat, srcType);
+         GLubyte *srcRow = (GLubyte *) _mesa_image_address(dims, srcPacking,
+                  srcAddr, srcWidth, srcHeight, srcFormat, srcType, img, 0, 0);
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            for (col = 0; col < srcWidth; col++) {
+               dstRow[col * 4 + 0] = srcRow[col * 3 + BCOMP];
+               dstRow[col * 4 + 1] = srcRow[col * 3 + GCOMP];
+               dstRow[col * 4 + 2] = srcRow[col * 3 + RCOMP];
+               dstRow[col * 4 + 3] = 0xff;
+            }
+            dstRow += dstRowStride;
+            srcRow += srcRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+   }
+   else if (!ctx->_ImageTransferState &&
+            !srcPacking->SwapBytes &&
+           dstFormat == &_mesa_texformat_argb8888 &&
+            srcFormat == GL_RGBA &&
+            srcType == GL_UNSIGNED_BYTE) {
+
+      int img, row, col;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      for (img = 0; img < srcDepth; img++) {
+         const GLint srcRowStride = _mesa_image_row_stride(srcPacking,
+                                                 srcWidth, srcFormat, srcType);
+         GLubyte *srcRow = (GLubyte *) _mesa_image_address(dims, srcPacking,
+                  srcAddr, srcWidth, srcHeight, srcFormat, srcType, img, 0, 0);
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            for (col = 0; col < srcWidth; col++) {
+               dstRow[col * 4 + 0] = srcRow[col * 4 + BCOMP];
+               dstRow[col * 4 + 1] = srcRow[col * 4 + GCOMP];
+               dstRow[col * 4 + 2] = srcRow[col * 4 + RCOMP];
+               dstRow[col * 4 + 3] = srcRow[col * 4 + ACOMP];
+            }
+            dstRow += dstRowStride;
+            srcRow += srcRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+   }
+   else if (!ctx->_ImageTransferState &&
+           !srcPacking->SwapBytes &&
+           dstFormat == &_mesa_texformat_argb8888 &&
+           srcType == GL_UNSIGNED_BYTE && 
+           littleEndian &&
+           can_swizzle(srcFormat)) {
+
+      GLubyte dstmap[4];
+
+      /* dstmap - how to swizzle from GL_RGBA to dst format:
+       */
+      dstmap[3] = 3;           /* alpha */
+      dstmap[2] = 0;           /* red */
+      dstmap[1] = 1;           /* green */
+      dstmap[0] = 2;           /* blue */
+      _mesa_swizzle_ubyte_image(ctx, dims,
+                               srcFormat,
+                               dstmap, 4,
+                               dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                               dstRowStride, dstImageStride,
+                               srcWidth, srcHeight, srcDepth, srcAddr,
+                               srcPacking);      
+   }
+   else {
+      /* general path */
+      const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLchan *src = tempImage;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row, col;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            GLuint *dstUI = (GLuint *) dstRow;
+            if (dstFormat == &_mesa_texformat_argb8888) {
+               for (col = 0; col < srcWidth; col++) {
+                  dstUI[col] = PACK_COLOR_8888( CHAN_TO_UBYTE(src[ACOMP]),
+                                                CHAN_TO_UBYTE(src[RCOMP]),
+                                                CHAN_TO_UBYTE(src[GCOMP]),
+                                                CHAN_TO_UBYTE(src[BCOMP]) );
+                  src += 4;
+               }
+            }
+            else {
+               for (col = 0; col < srcWidth; col++) {
+                  dstUI[col] = PACK_COLOR_8888_REV( CHAN_TO_UBYTE(src[ACOMP]),
+                                                    CHAN_TO_UBYTE(src[RCOMP]),
+                                                    CHAN_TO_UBYTE(src[GCOMP]),
+                                                    CHAN_TO_UBYTE(src[BCOMP]) );
+                  src += 4;
+               }
+            }
+            dstRow += dstRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+      _mesa_free((void *) tempImage);
+   }
+   return GL_TRUE;
+}
+
+
+GLboolean
+_mesa_texstore_rgb888(STORE_PARAMS)
+{
+   const GLuint ui = 1;
+   const GLubyte littleEndian = *((const GLubyte *) &ui);
+
+   ASSERT(dstFormat == &_mesa_texformat_rgb888);
+   ASSERT(dstFormat->TexelBytes == 3);
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       baseInternalFormat == GL_RGB &&
+       srcFormat == GL_BGR &&
+       srcType == GL_UNSIGNED_BYTE &&
+       littleEndian) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else if (!ctx->_ImageTransferState &&
+            !srcPacking->SwapBytes &&
+            srcFormat == GL_RGBA &&
+            srcType == GL_UNSIGNED_BYTE) {
+      /* extract RGB from RGBA */
+      int img, row, col;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      for (img = 0; img < srcDepth; img++) {
+         const GLint srcRowStride = _mesa_image_row_stride(srcPacking,
+                                                 srcWidth, srcFormat, srcType);
+         GLubyte *srcRow = (GLubyte *) _mesa_image_address(dims, srcPacking,
+                  srcAddr, srcWidth, srcHeight, srcFormat, srcType, img, 0, 0);
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            for (col = 0; col < srcWidth; col++) {
+               dstRow[col * 3 + 0] = srcRow[col * 4 + BCOMP];
+               dstRow[col * 3 + 1] = srcRow[col * 4 + GCOMP];
+               dstRow[col * 3 + 2] = srcRow[col * 4 + RCOMP];
+            }
+            dstRow += dstRowStride;
+            srcRow += srcRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+   }
+   else {
+      /* general path */
+      const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLchan *src = (const GLchan *) tempImage;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row, col;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+#if 0
+            if (littleEndian) {
+               for (col = 0; col < srcWidth; col++) {
+                  dstRow[col * 3 + 0] = CHAN_TO_UBYTE(src[RCOMP]);
+                  dstRow[col * 3 + 1] = CHAN_TO_UBYTE(src[GCOMP]);
+                  dstRow[col * 3 + 2] = CHAN_TO_UBYTE(src[BCOMP]);
+                  srcUB += 3;
+               }
+            }
+            else {
+               for (col = 0; col < srcWidth; col++) {
+                  dstRow[col * 3 + 0] = srcUB[BCOMP];
+                  dstRow[col * 3 + 1] = srcUB[GCOMP];
+                  dstRow[col * 3 + 2] = srcUB[RCOMP];
+                  srcUB += 3;
+               }
+            }
+#else
+            for (col = 0; col < srcWidth; col++) {
+               dstRow[col * 3 + 0] = CHAN_TO_UBYTE(src[BCOMP]);
+               dstRow[col * 3 + 1] = CHAN_TO_UBYTE(src[GCOMP]);
+               dstRow[col * 3 + 2] = CHAN_TO_UBYTE(src[RCOMP]);
+               src += 3;
+            }
+#endif
+            dstRow += dstRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+      _mesa_free((void *) tempImage);
+   }
+   return GL_TRUE;
+}
+
+
+GLboolean
+_mesa_texstore_bgr888(STORE_PARAMS)
+{
+   const GLuint ui = 1;
+   const GLubyte littleEndian = *((const GLubyte *) &ui);
+
+   ASSERT(dstFormat == &_mesa_texformat_bgr888);
+   ASSERT(dstFormat->TexelBytes == 3);
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       baseInternalFormat == GL_RGB &&
+       srcFormat == GL_RGB &&
+       srcType == GL_UNSIGNED_BYTE &&
+       littleEndian) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else if (!ctx->_ImageTransferState &&
+            !srcPacking->SwapBytes &&
+            srcFormat == GL_RGBA &&
+            srcType == GL_UNSIGNED_BYTE) {
+      /* extract BGR from RGBA */
+      int img, row, col;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      for (img = 0; img < srcDepth; img++) {
+         const GLint srcRowStride = _mesa_image_row_stride(srcPacking,
+                                                 srcWidth, srcFormat, srcType);
+         GLubyte *srcRow = (GLubyte *) _mesa_image_address(dims, srcPacking,
+                  srcAddr, srcWidth, srcHeight, srcFormat, srcType, img, 0, 0);
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            for (col = 0; col < srcWidth; col++) {
+               dstRow[col * 3 + 0] = srcRow[col * 4 + RCOMP];
+               dstRow[col * 3 + 1] = srcRow[col * 4 + GCOMP];
+               dstRow[col * 3 + 2] = srcRow[col * 4 + BCOMP];
+            }
+            dstRow += dstRowStride;
+            srcRow += srcRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+   }
+   else {
+      /* general path */
+      const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLchan *src = (const GLchan *) tempImage;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row, col;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            for (col = 0; col < srcWidth; col++) {
+               dstRow[col * 3 + 0] = CHAN_TO_UBYTE(src[RCOMP]);
+               dstRow[col * 3 + 1] = CHAN_TO_UBYTE(src[GCOMP]);
+               dstRow[col * 3 + 2] = CHAN_TO_UBYTE(src[BCOMP]);
+               src += 3;
+            }
+            dstRow += dstRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+      _mesa_free((void *) tempImage);
+   }
+   return GL_TRUE;
+}
+
+
+GLboolean
+_mesa_texstore_argb4444(STORE_PARAMS)
+{
+   ASSERT(dstFormat == &_mesa_texformat_argb4444 ||
+          dstFormat == &_mesa_texformat_argb4444_rev);
+   ASSERT(dstFormat->TexelBytes == 2);
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       dstFormat == &_mesa_texformat_argb4444 &&
+       baseInternalFormat == GL_RGBA &&
+       srcFormat == GL_BGRA &&
+       srcType == GL_UNSIGNED_SHORT_4_4_4_4_REV) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else {
+      /* general path */
+      const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLchan *src = tempImage;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row, col;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            GLushort *dstUS = (GLushort *) dstRow;
+            if (dstFormat == &_mesa_texformat_argb4444) {
+               for (col = 0; col < srcWidth; col++) {
+                  dstUS[col] = PACK_COLOR_4444( CHAN_TO_UBYTE(src[ACOMP]),
+                                                CHAN_TO_UBYTE(src[RCOMP]),
+                                                CHAN_TO_UBYTE(src[GCOMP]),
+                                                CHAN_TO_UBYTE(src[BCOMP]) );
+                  src += 4;
+               }
+            }
+            else {
+               for (col = 0; col < srcWidth; col++) {
+                  dstUS[col] = PACK_COLOR_4444_REV( CHAN_TO_UBYTE(src[ACOMP]),
+                                                    CHAN_TO_UBYTE(src[RCOMP]),
+                                                    CHAN_TO_UBYTE(src[GCOMP]),
+                                                    CHAN_TO_UBYTE(src[BCOMP]) );
+                  src += 4;
+               }
+            }
+            dstRow += dstRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+      _mesa_free((void *) tempImage);
+   }
+   return GL_TRUE;
+}
+
+
+
+GLboolean
+_mesa_texstore_argb1555(STORE_PARAMS)
+{
+   ASSERT(dstFormat == &_mesa_texformat_argb1555 ||
+          dstFormat == &_mesa_texformat_argb1555_rev);
+   ASSERT(dstFormat->TexelBytes == 2);
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       dstFormat == &_mesa_texformat_argb1555 &&
+       baseInternalFormat == GL_RGBA &&
+       srcFormat == GL_BGRA &&
+       srcType == GL_UNSIGNED_SHORT_1_5_5_5_REV) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else {
+      /* general path */
+      const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLchan *src =tempImage;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row, col;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            GLushort *dstUS = (GLushort *) dstRow;
+            if (dstFormat == &_mesa_texformat_argb1555) {
+               for (col = 0; col < srcWidth; col++) {
+                  dstUS[col] = PACK_COLOR_1555( CHAN_TO_UBYTE(src[ACOMP]),
+                                                CHAN_TO_UBYTE(src[RCOMP]),
+                                                CHAN_TO_UBYTE(src[GCOMP]),
+                                                CHAN_TO_UBYTE(src[BCOMP]) );
+                  src += 4;
+               }
+            }
+            else {
+               for (col = 0; col < srcWidth; col++) {
+                  dstUS[col] = PACK_COLOR_1555_REV( CHAN_TO_UBYTE(src[ACOMP]),
+                                                    CHAN_TO_UBYTE(src[RCOMP]),
+                                                    CHAN_TO_UBYTE(src[GCOMP]),
+                                                    CHAN_TO_UBYTE(src[BCOMP]) );
+                  src += 4;
+               }
+            }
+            dstRow += dstRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+      _mesa_free((void *) tempImage);
+   }
+   return GL_TRUE;
+}
+
+
+GLboolean
+_mesa_texstore_al88(STORE_PARAMS)
 {
-   switch (format) {
-      case GL_ALPHA:
-      case GL_ALPHA4:
-      case GL_ALPHA8:
-      case GL_ALPHA12:
-      case GL_ALPHA16:
-         return 1;
-      case 1:
-      case GL_LUMINANCE:
-      case GL_LUMINANCE4:
-      case GL_LUMINANCE8:
-      case GL_LUMINANCE12:
-      case GL_LUMINANCE16:
-         return 1;
-      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:
-         return 2;
-      case GL_INTENSITY:
-      case GL_INTENSITY4:
-      case GL_INTENSITY8:
-      case GL_INTENSITY12:
-      case GL_INTENSITY16:
-         return 1;
-      case 3:
-      case GL_RGB:
-      case GL_R3_G3_B2:
-      case GL_RGB4:
-      case GL_RGB5:
-      case GL_RGB8:
-      case GL_RGB10:
-      case GL_RGB12:
-      case GL_RGB16:
-         return 3;
-      case 4:
-      case GL_RGBA:
-      case GL_RGBA2:
-      case GL_RGBA4:
-      case GL_RGB5_A1:
-      case GL_RGBA8:
-      case GL_RGB10_A2:
-      case GL_RGBA12:
-      case GL_RGBA16:
-         return 4;
-      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:
-         return 1;
-      case GL_DEPTH_COMPONENT:
-      case GL_DEPTH_COMPONENT16_SGIX:
-      case GL_DEPTH_COMPONENT24_SGIX:
-      case GL_DEPTH_COMPONENT32_SGIX:
-         return 1;
-      default:
-         return -1;  /* error */
+   const GLuint ui = 1;
+   const GLubyte littleEndian = *((const GLubyte *) &ui);
+
+   ASSERT(dstFormat == &_mesa_texformat_al88 ||
+          dstFormat == &_mesa_texformat_al88_rev);
+   ASSERT(dstFormat->TexelBytes == 2);
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       dstFormat == &_mesa_texformat_al88 &&
+       baseInternalFormat == GL_LUMINANCE_ALPHA &&
+       srcFormat == GL_LUMINANCE_ALPHA &&
+       srcType == GL_UNSIGNED_BYTE &&
+       littleEndian) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else {
+      /* general path */
+      const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLchan *src = tempImage;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row, col;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            GLushort *dstUS = (GLushort *) dstRow;
+            if (dstFormat == &_mesa_texformat_al88) {
+               for (col = 0; col < srcWidth; col++) {
+                  /* src[0] is luminance, src[1] is alpha */
+                 dstUS[col] = PACK_COLOR_88( CHAN_TO_UBYTE(src[1]),
+                                             CHAN_TO_UBYTE(src[0]) );
+                 src += 2;
+               }
+            }
+            else {
+               for (col = 0; col < srcWidth; col++) {
+                  /* src[0] is luminance, src[1] is alpha */
+                 dstUS[col] = PACK_COLOR_88_REV( CHAN_TO_UBYTE(src[1]),
+                                                 CHAN_TO_UBYTE(src[0]) );
+                 src += 2;
+               }
+            }
+            dstRow += dstRowStride;
+         }
+         dstImage += dstImageStride;
+      }
+      _mesa_free((void *) tempImage);
    }
+   return GL_TRUE;
 }
 
 
-/*
- * This function is used to transfer the user's image data into a texture
- * image buffer.  We handle both full texture images and subtexture images.
- * We also take care of all image transfer operations here, including
- * convolution, scale/bias, colortables, etc.
- *
- * The destination texel channel type is always GLchan.
- *
- * A hardware driver may use this as a helper routine to unpack and
- * apply pixel transfer ops into a temporary image buffer.  Then,
- * convert the temporary image into the special hardware format.
- *
- * Input:
- *   dimensions - 1, 2, or 3
- *   texFormat - GL_LUMINANCE, GL_INTENSITY, GL_LUMINANCE_ALPHA, GL_ALPHA,
- *               GL_RGB or GL_RGBA
- *   texDestAddr - destination image address
- *   srcWidth, srcHeight, srcDepth - size (in pixels) of src and dest images
- *   dstXoffset, dstYoffset, dstZoffset - position to store the image within
- *      the destination 3D texture
- *   dstRowStride, dstImageStride - dest image strides in bytes
- *   srcFormat - source image format (GL_ALPHA, GL_RED, GL_RGB, etc)
- *   srcType - GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT_5_6_5, GL_FLOAT, etc
- *   srcPacking - describes packing of incoming image.
- *   transferOps - mask of pixel transfer operations
- */
-static void
-transfer_teximage(GLcontext *ctx, GLuint dimensions,
-                  GLenum texDestFormat, GLvoid *texDestAddr,
-                  GLint srcWidth, GLint srcHeight, GLint srcDepth,
-                  GLint dstXoffset, GLint dstYoffset, GLint dstZoffset,
-                  GLint dstRowStride, GLint dstImageStride,
-                  GLenum srcFormat, GLenum srcType,
-                  const GLvoid *srcAddr,
-                  const struct gl_pixelstore_attrib *srcPacking,
-                  GLuint transferOps)
+GLboolean
+_mesa_texstore_rgb332(STORE_PARAMS)
 {
-   GLint texComponents;
-
-   ASSERT(ctx);
-   ASSERT(dimensions >= 1 && dimensions <= 3);
-   ASSERT(texDestAddr);
-   ASSERT(srcWidth >= 1);
-   ASSERT(srcHeight >= 1);
-   ASSERT(srcDepth >= 1);
-   ASSERT(dstXoffset >= 0);
-   ASSERT(dstYoffset >= 0);
-   ASSERT(dstZoffset >= 0);
-   ASSERT(dstRowStride >= 0);
-   ASSERT(dstImageStride >= 0);
-   ASSERT(srcAddr);
-   ASSERT(srcPacking);
-
-   texComponents = components_in_intformat(texDestFormat);
-
-   /* try common 2D texture cases first */
-   if (!transferOps && dimensions == 2 && srcType == CHAN_TYPE) {
-
-      if (srcFormat == texDestFormat) {
-         /* This will cover the common GL_RGB, GL_RGBA, GL_ALPHA,
-          * GL_LUMINANCE_ALPHA, etc. texture formats.  Use memcpy().
-          */
-         const GLchan *src = (const GLchan *) _mesa_image_address(
-                                   srcPacking, srcAddr, srcWidth, srcHeight,
-                                   srcFormat, srcType, 0, 0, 0);
-         const GLint srcRowStride = _mesa_image_row_stride(srcPacking,
-                                               srcWidth, srcFormat, srcType);
-         const GLint widthInBytes = srcWidth * texComponents * sizeof(GLchan);
-         GLchan *dst = (GLchan *) texDestAddr
-                     + dstYoffset * (dstRowStride / sizeof(GLchan))
-                     + dstXoffset * texComponents;
-         if (srcRowStride == widthInBytes && dstRowStride == widthInBytes) {
-            MEMCPY(dst, src, srcHeight * widthInBytes);
-         }
-         else {
-            GLint i;
-            for (i = 0; i < srcHeight; i++) {
-               MEMCPY(dst, src, widthInBytes);
-               src += (srcRowStride / sizeof(GLchan));
-               dst += (dstRowStride / sizeof(GLchan));
+   ASSERT(dstFormat == &_mesa_texformat_rgb332);
+   ASSERT(dstFormat->TexelBytes == 1);
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       baseInternalFormat == GL_RGB &&
+       srcFormat == GL_RGB && srcType == GL_UNSIGNED_BYTE_3_3_2) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else {
+      /* general path */
+      const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLchan *src = tempImage;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row, col;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            for (col = 0; col < srcWidth; col++) {
+               dstRow[col] = PACK_COLOR_332( CHAN_TO_UBYTE(src[RCOMP]),
+                                             CHAN_TO_UBYTE(src[GCOMP]),
+                                             CHAN_TO_UBYTE(src[BCOMP]) );
+               src += 3;
             }
+            dstRow += dstRowStride;
          }
-         return;  /* all done */
+         dstImage += dstImageStride;
       }
-      else if (srcFormat == GL_RGBA && texDestFormat == GL_RGB) {
-         /* commonly used by Quake */
-         const GLchan *src = (const GLchan *) _mesa_image_address(
-                                   srcPacking, srcAddr, srcWidth, srcHeight,
-                                   srcFormat, srcType, 0, 0, 0);
-         const GLint srcRowStride = _mesa_image_row_stride(srcPacking,
-                                               srcWidth, srcFormat, srcType);
-         GLchan *dst = (GLchan *) texDestAddr
-                     + dstYoffset * (dstRowStride / sizeof(GLchan))
-                     + dstXoffset * texComponents;
-         GLint i, j;
-         for (i = 0; i < srcHeight; i++) {
-            const GLchan *s = src;
-            GLchan *d = dst;
-            for (j = 0; j < srcWidth; j++) {
-               *d++ = *s++;  /*red*/
-               *d++ = *s++;  /*green*/
-               *d++ = *s++;  /*blue*/
-               s++;          /*alpha*/
+      _mesa_free((void *) tempImage);
+   }
+   return GL_TRUE;
+}
+
+
+/**
+ * Texstore for _mesa_texformat_a8, _mesa_texformat_l8, _mesa_texformat_i8.
+ */
+GLboolean
+_mesa_texstore_a8(STORE_PARAMS)
+{
+   ASSERT(dstFormat == &_mesa_texformat_a8 ||
+          dstFormat == &_mesa_texformat_l8 ||
+          dstFormat == &_mesa_texformat_i8);
+   ASSERT(dstFormat->TexelBytes == 1);
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       baseInternalFormat == srcFormat &&
+       srcType == GL_UNSIGNED_BYTE) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else {
+      /* general path */
+      const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLchan *src = tempImage;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row, col;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            for (col = 0; col < srcWidth; col++) {
+               dstRow[col] = CHAN_TO_UBYTE(src[col]);
             }
-            src += (srcRowStride / sizeof(GLchan));
-            dst += (dstRowStride / sizeof(GLchan));
+            dstRow += dstRowStride;
+            src += srcWidth;
          }
-         return;  /* all done */
+         dstImage += dstImageStride;
       }
+      _mesa_free((void *) tempImage);
    }
+   return GL_TRUE;
+}
 
-   /*
-    * General case solutions
-    */
-   if (texDestFormat == GL_COLOR_INDEX) {
-      /* color index texture */
-      const GLenum texType = CHAN_TYPE;
+
+
+GLboolean
+_mesa_texstore_ci8(STORE_PARAMS)
+{
+   (void) dims; (void) baseInternalFormat;
+   ASSERT(dstFormat == &_mesa_texformat_ci8);
+   ASSERT(dstFormat->TexelBytes == 1);
+   ASSERT(baseInternalFormat == GL_COLOR_INDEX);
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       srcFormat == GL_COLOR_INDEX &&
+       srcType == GL_UNSIGNED_BYTE) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else {
+      /* general path */
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
       GLint img, row;
-      GLchan *dest = (GLchan *) texDestAddr
-                   + dstZoffset * (dstImageStride / sizeof(GLchan))
-                   + dstYoffset * (dstRowStride / sizeof(GLchan))
-                   + dstXoffset * texComponents;
       for (img = 0; img < srcDepth; img++) {
-         GLchan *destRow = dest;
+         GLubyte *dstRow = dstImage;
          for (row = 0; row < srcHeight; row++) {
-            const GLvoid *src = _mesa_image_address(srcPacking,
+            const GLvoid *src = _mesa_image_address(dims, srcPacking,
                 srcAddr, srcWidth, srcHeight, srcFormat, srcType, img, row, 0);
-            _mesa_unpack_index_span(ctx, srcWidth, texType, destRow,
-                                    srcType, src, srcPacking, transferOps);
-            destRow += (dstRowStride / sizeof(GLchan));
+            _mesa_unpack_index_span(ctx, srcWidth, GL_UNSIGNED_BYTE, dstRow,
+                                    srcType, src, srcPacking,
+                                    ctx->_ImageTransferState);
+            dstRow += dstRowStride;
          }
-         dest += dstImageStride;
+         dstImage += dstImageStride;
       }
    }
-   else if (texDestFormat == GL_DEPTH_COMPONENT) {
-      /* Depth texture (shadow maps) */
+   return GL_TRUE;
+}
+
+
+/**
+ * Texstore for _mesa_texformat_ycbcr or _mesa_texformat_ycbcr_rev.
+ */
+GLboolean
+_mesa_texstore_ycbcr(STORE_PARAMS)
+{
+   const GLuint ui = 1;
+   const GLubyte littleEndian = *((const GLubyte *) &ui);
+   (void) ctx; (void) dims; (void) baseInternalFormat;
+
+   ASSERT((dstFormat == &_mesa_texformat_ycbcr) ||
+          (dstFormat == &_mesa_texformat_ycbcr_rev));
+   ASSERT(dstFormat->TexelBytes == 2);
+   ASSERT(ctx->Extensions.MESA_ycbcr_texture);
+   ASSERT(srcFormat == GL_YCBCR_MESA);
+   ASSERT((srcType == GL_UNSIGNED_SHORT_8_8_MESA) ||
+          (srcType == GL_UNSIGNED_SHORT_8_8_REV_MESA));
+   ASSERT(baseInternalFormat == GL_YCBCR_MESA);
+
+   /* always just memcpy since no pixel transfer ops apply */
+   memcpy_texture(dims,
+                  dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                  dstRowStride, dstImageStride,
+                  srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                  srcAddr, srcPacking);
+
+   /* Check if we need byte swapping */
+   /* XXX the logic here _might_ be wrong */
+   if (srcPacking->SwapBytes ^
+       (srcType == GL_UNSIGNED_SHORT_8_8_REV_MESA) ^
+       (dstFormat == &_mesa_texformat_ycbcr_rev) ^
+       !littleEndian) {
+      GLushort *pImage = (GLushort *) ((GLubyte *) dstAddr
+                                       + dstZoffset * dstImageStride
+                                       + dstYoffset * dstRowStride
+                                       + dstXoffset * dstFormat->TexelBytes);
       GLint img, row;
-      GLubyte *dest = (GLubyte *) texDestAddr
-                    + dstZoffset * dstImageStride
-                    + dstYoffset * (dstRowStride / sizeof(GLchan))
-                    + dstXoffset * texComponents;
       for (img = 0; img < srcDepth; img++) {
-         GLubyte *destRow = dest;
+         GLushort *pRow = pImage;
          for (row = 0; row < srcHeight; row++) {
-            const GLvoid *src = _mesa_image_address(srcPacking,
-                srcAddr, srcWidth, srcHeight, srcFormat, srcType, img, row, 0);
-            _mesa_unpack_depth_span(ctx, srcWidth, (GLfloat *) destRow,
-                                    srcType, src, srcPacking);
-            destRow += (dstRowStride / sizeof(GLchan));
+            _mesa_swap2(pRow, srcWidth);
+            pRow += dstRowStride;
          }
-         dest += dstImageStride;
+         pImage += dstImageStride;
       }
    }
-   else {
-      /* regular, color texture */
-      if ((dimensions == 1 && ctx->Pixel.Convolution1DEnabled) ||
-          (dimensions >= 2 && ctx->Pixel.Convolution2DEnabled) ||
-          (dimensions >= 2 && ctx->Pixel.Separable2DEnabled)) {
-         /*
-          * Fill texture image with convolution
-          */
-         GLint img, row;
-         GLint convWidth = srcWidth, convHeight = srcHeight;
-         GLfloat *tmpImage, *convImage;
-         tmpImage = (GLfloat *) MALLOC(srcWidth * srcHeight * 4 * sizeof(GLfloat));
-         if (!tmpImage) {
-            _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage");
-            return;
-         }
-         convImage = (GLfloat *) MALLOC(srcWidth * srcHeight * 4 * sizeof(GLfloat));
-         if (!convImage) {
-            _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage");
-            FREE(tmpImage);
-            return;
-         }
+   return GL_TRUE;
+}
 
-         for (img = 0; img < srcDepth; img++) {
-            const GLfloat *srcf;
-            GLfloat *dstf = tmpImage;
-            GLchan *dest;
 
-            /* unpack and do transfer ops up to convolution */
-            for (row = 0; row < srcHeight; row++) {
-               const GLvoid *src = _mesa_image_address(srcPacking,
-                                              srcAddr, srcWidth, srcHeight,
-                                              srcFormat, srcType, img, row, 0);
-               _mesa_unpack_float_color_span(ctx, srcWidth, GL_RGBA, dstf,
-                         srcFormat, srcType, src, srcPacking,
-                         transferOps & IMAGE_PRE_CONVOLUTION_BITS,
-                         GL_TRUE);
-               dstf += srcWidth * 4;
-            }
 
-            /* convolve */
-            if (dimensions == 1) {
-               ASSERT(ctx->Pixel.Convolution1DEnabled);
-               _mesa_convolve_1d_image(ctx, &convWidth, tmpImage, convImage);
-            }
-            else {
-               if (ctx->Pixel.Convolution2DEnabled) {
-                  _mesa_convolve_2d_image(ctx, &convWidth, &convHeight,
-                                          tmpImage, convImage);
-               }
-               else {
-                  ASSERT(ctx->Pixel.Separable2DEnabled);
-                  _mesa_convolve_sep_image(ctx, &convWidth, &convHeight,
-                                           tmpImage, convImage);
-               }
-            }
 
-            /* packing and transfer ops after convolution */
-            srcf = convImage;
-            dest = (GLchan *) texDestAddr
-                 + (dstZoffset + img) * (dstImageStride / sizeof(GLchan))
-                 + dstYoffset * (dstRowStride / sizeof(GLchan));
-            for (row = 0; row < convHeight; row++) {
-               _mesa_pack_float_rgba_span(ctx, convWidth,
-                                          (const GLfloat (*)[4]) srcf,
-                                          texDestFormat, CHAN_TYPE,
-                                          dest, &_mesa_native_packing,
-                                          transferOps
-                                          & IMAGE_POST_CONVOLUTION_BITS);
-               srcf += convWidth * 4;
-               dest += (dstRowStride / sizeof(GLchan));
-            }
+/**
+ * Store an image in any of the formats:
+ *   _mesa_texformat_rgba_float32
+ *   _mesa_texformat_rgb_float32
+ *   _mesa_texformat_alpha_float32
+ *   _mesa_texformat_luminance_float32
+ *   _mesa_texformat_luminance_alpha_float32
+ *   _mesa_texformat_intensity_float32
+ */
+GLboolean
+_mesa_texstore_rgba_float32(STORE_PARAMS)
+{
+   const GLint components = _mesa_components_in_format(baseInternalFormat);
+
+   ASSERT(dstFormat == &_mesa_texformat_rgba_float32 ||
+          dstFormat == &_mesa_texformat_rgb_float32 ||
+          dstFormat == &_mesa_texformat_alpha_float32 ||
+          dstFormat == &_mesa_texformat_luminance_float32 ||
+          dstFormat == &_mesa_texformat_luminance_alpha_float32 ||
+          dstFormat == &_mesa_texformat_intensity_float32);
+   ASSERT(baseInternalFormat == GL_RGBA ||
+          baseInternalFormat == GL_RGB ||
+          baseInternalFormat == GL_ALPHA ||
+          baseInternalFormat == GL_LUMINANCE ||
+          baseInternalFormat == GL_LUMINANCE_ALPHA ||
+          baseInternalFormat == GL_INTENSITY);
+   ASSERT(dstFormat->TexelBytes == components * sizeof(GLfloat));
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       baseInternalFormat == srcFormat &&
+       srcType == GL_FLOAT) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else {
+      /* general path */
+      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLfloat *src = tempImage;
+      GLint bytesPerRow;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      bytesPerRow = srcWidth * components * sizeof(GLfloat);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dst = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            _mesa_memcpy(dst, src, bytesPerRow);
+            dst += dstRowStride;
+            src += srcWidth * components;
          }
-
-         FREE(convImage);
-         FREE(tmpImage);
+         dstImage += dstImageStride;
       }
-      else {
-         /*
-          * no convolution
-          */
-         GLint img, row;
-         GLchan *dest = (GLchan *) texDestAddr
-                      + dstZoffset * (dstImageStride / sizeof(GLchan))
-                      + dstYoffset * (dstRowStride / sizeof(GLchan))
-                      + dstXoffset * texComponents;
-         for (img = 0; img < srcDepth; img++) {
-            GLchan *destRow = dest;
-            for (row = 0; row < srcHeight; row++) {
-               const GLvoid *srcRow = _mesa_image_address(srcPacking,
-                                              srcAddr, srcWidth, srcHeight,
-                                              srcFormat, srcType, img, row, 0);
-               _mesa_unpack_chan_color_span(ctx, srcWidth, texDestFormat,
-                                       destRow, srcFormat, srcType, srcRow,
-                                       srcPacking, transferOps);
-               destRow += (dstRowStride / sizeof(GLchan));
+
+      _mesa_free((void *) tempImage);
+   }
+   return GL_TRUE;
+}
+
+
+/**
+ * As above, but store 16-bit floats.
+ */
+GLboolean
+_mesa_texstore_rgba_float16(STORE_PARAMS)
+{
+   const GLint components = _mesa_components_in_format(baseInternalFormat);
+
+   ASSERT(dstFormat == &_mesa_texformat_rgba_float16 ||
+          dstFormat == &_mesa_texformat_rgb_float16 ||
+          dstFormat == &_mesa_texformat_alpha_float16 ||
+          dstFormat == &_mesa_texformat_luminance_float16 ||
+          dstFormat == &_mesa_texformat_luminance_alpha_float16 ||
+          dstFormat == &_mesa_texformat_intensity_float16);
+   ASSERT(baseInternalFormat == GL_RGBA ||
+          baseInternalFormat == GL_RGB ||
+          baseInternalFormat == GL_ALPHA ||
+          baseInternalFormat == GL_LUMINANCE ||
+          baseInternalFormat == GL_LUMINANCE_ALPHA ||
+          baseInternalFormat == GL_INTENSITY);
+   ASSERT(dstFormat->TexelBytes == components * sizeof(GLhalfARB));
+
+   if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       baseInternalFormat == srcFormat &&
+       srcType == GL_HALF_FLOAT_ARB) {
+      /* simple memcpy path */
+      memcpy_texture(dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride, dstImageStride,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else {
+      /* general path */
+      const GLfloat *tempImage = make_temp_float_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 dstFormat->BaseFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+      const GLfloat *src = tempImage;
+      GLubyte *dstImage = (GLubyte *) dstAddr
+                        + dstZoffset * dstImageStride
+                        + dstYoffset * dstRowStride
+                        + dstXoffset * dstFormat->TexelBytes;
+      GLint img, row;
+      if (!tempImage)
+         return GL_FALSE;
+      _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+      for (img = 0; img < srcDepth; img++) {
+         GLubyte *dstRow = dstImage;
+         for (row = 0; row < srcHeight; row++) {
+            GLhalfARB *dstTexel = (GLhalfARB *) dstRow;
+            GLint i;
+            for (i = 0; i < srcWidth * components; i++) {
+               dstTexel[i] = _mesa_float_to_half(src[i]);
             }
-            dest += dstImageStride / sizeof(GLchan);
+            dstRow += dstRowStride;
+            src += srcWidth * components;
          }
+         dstImage += dstImageStride;
       }
+
+      _mesa_free((void *) tempImage);
    }
+   return GL_TRUE;
 }
 
 
 
-/*
- * Transfer a texture image from user space to <destAddr> applying all
- * needed image transfer operations and storing the result in the format
- * specified by <dstFormat>.  <dstFormat> may be any format from texformat.h.
- * Input:
- *   dimensions - 1, 2 or 3
- *   baseInternalFormat - base format of the internal texture format
- *       specified by the user.  This is very important, see below.
- *   dstFormat - destination image format
- *   dstAddr - destination address
- *   srcWidth, srcHeight, srcDepth - size of source iamge
- *   dstX/Y/Zoffset - as specified by glTexSubImage
- *   dstRowStride - stride between dest rows in bytes
- *   dstImageStride - stride between dest images in bytes
- *   srcFormat, srcType - incoming image format and datatype
- *   srcAddr - source image address
- *   srcPacking - packing params of source image
- *
- * XXX this function is a bit more complicated than it should be.  If
- * _mesa_convert_texsubimage[123]d could handle any dest/source formats
- * or if transfer_teximage() could store in any MESA_FORMAT_* format, we
- * could simplify things here.
+/**
+ * Check if an unpack PBO is active prior to fetching a texture image.
+ * If so, do bounds checking and map the buffer into main memory.
+ * Any errors detected will be recorded.
+ * The caller _must_ call _mesa_unmap_teximage_pbo() too!
  */
-void
-_mesa_transfer_teximage(GLcontext *ctx, GLuint dimensions,
-                        GLenum baseInternalFormat,
-                        const struct gl_texture_format *dstFormat,
-                        GLvoid *dstAddr,
-                        GLint srcWidth, GLint srcHeight, GLint srcDepth,
-                        GLint dstXoffset, GLint dstYoffset, GLint dstZoffset,
-                        GLint dstRowStride, GLint dstImageStride,
-                        GLenum srcFormat, GLenum srcType,
-                        const GLvoid *srcAddr,
-                        const struct gl_pixelstore_attrib *srcPacking)
+const GLvoid *
+_mesa_validate_pbo_teximage(GLcontext *ctx, GLuint dimensions,
+                           GLsizei width, GLsizei height, GLsizei depth,
+                           GLenum format, GLenum type, const GLvoid *pixels,
+                           const struct gl_pixelstore_attrib *unpack,
+                           const char *funcName)
 {
-   const GLint dstRowStridePixels = dstRowStride / dstFormat->TexelBytes;
-   const GLint dstImageStridePixels = dstImageStride / dstFormat->TexelBytes;
-   GLboolean makeTemp;
-   GLuint transferOps = ctx->_ImageTransferState;
-   GLboolean freeSourceData = GL_FALSE;
-   GLint postConvWidth = srcWidth, postConvHeight = srcHeight;
+   GLubyte *buf;
 
-   assert(baseInternalFormat > 0);
-
-   if (transferOps & IMAGE_CONVOLUTION_BIT) {
-      _mesa_adjust_image_for_convolution(ctx, dimensions, &postConvWidth,
-                                         &postConvHeight);
+   if (unpack->BufferObj->Name == 0) {
+      /* no PBO */
+      return pixels;
+   }
+   if (!_mesa_validate_pbo_access(dimensions, unpack, width, height, depth,
+                                  format, type, pixels)) {
+      _mesa_error(ctx, GL_INVALID_OPERATION, funcName, "(invalid PBO access");
+      return NULL;
    }
 
-   /*
-    * Consider this scenario:  The user's source image is GL_RGB and the
-    * requested internal format is GL_LUMINANCE.  Now suppose the device
-    * driver doesn't support GL_LUMINANCE and instead uses RGB16 as the
-    * texture format.  In that case we still need to do an intermediate
-    * conversion to luminance format so that the incoming red channel gets
-    * replicated into the dest red, green and blue channels.  The following
-    * code takes care of that.
-    */
-   if (dstFormat->BaseFormat != baseInternalFormat) {
-      /* Allocate storage for temporary image in the baseInternalFormat */
-      const GLint texelSize = _mesa_components_in_format(baseInternalFormat)
-         * sizeof(GLchan);
-      const GLint bytes = texelSize * postConvWidth * postConvHeight *srcDepth;
-      const GLint tmpRowStride = texelSize * postConvWidth;
-      const GLint tmpImgStride = texelSize * postConvWidth * postConvHeight;
-      GLvoid *tmpImage = MALLOC(bytes);
-      if (!tmpImage)
-         return;
-      transfer_teximage(ctx, dimensions, baseInternalFormat, tmpImage,
-                        srcWidth, srcHeight, srcDepth,
-                        0, 0, 0, /* x/y/zoffset */
-                        tmpRowStride, tmpImgStride,
-                        srcFormat, srcType, srcAddr, srcPacking, transferOps);
-
-      /* this is our new source image */
-      srcWidth = postConvWidth;
-      srcHeight = postConvHeight;
-      srcFormat = baseInternalFormat;
-      srcType = CHAN_TYPE;
-      srcAddr = tmpImage;
-      srcPacking = &_mesa_native_packing;
-      freeSourceData = GL_TRUE;
-      transferOps = 0;  /* image transfer ops were completed */
-   }
-
-   /* Let the optimized tex conversion functions take a crack at the
-    * image conversion if the dest format is a h/w format.
-    */
-   if (_mesa_is_hardware_tex_format(dstFormat)) {
-      if (transferOps) {
-         makeTemp = GL_TRUE;
-      }
-      else {
-         if (dimensions == 1) {
-            makeTemp = !_mesa_convert_texsubimage1d(dstFormat->MesaFormat,
-                                                    dstXoffset,
-                                                    srcWidth,
-                                                    srcFormat, srcType,
-                                                    srcPacking, srcAddr,
-                                                    dstAddr);
-         }
-         else if (dimensions == 2) {
-            makeTemp = !_mesa_convert_texsubimage2d(dstFormat->MesaFormat,
-                                                    dstXoffset, dstYoffset,
-                                                    srcWidth, srcHeight,
-                                                    dstRowStridePixels,
-                                                    srcFormat, srcType,
-                                                    srcPacking, srcAddr,
-                                                    dstAddr);
-         }
-         else {
-            assert(dimensions == 3);
-            makeTemp = !_mesa_convert_texsubimage3d(dstFormat->MesaFormat,
-                                      dstXoffset, dstYoffset, dstZoffset,
-                                      srcWidth, srcHeight, srcDepth,
-                                      dstRowStridePixels, dstImageStridePixels,
-                                      srcFormat, srcType,
-                                      srcPacking, srcAddr, dstAddr);
-         }
-         if (!makeTemp) {
-            /* all done! */
-            if (freeSourceData)
-               FREE((void *) srcAddr);
-            return;
-         }
-      }
+   buf = (GLubyte *) ctx->Driver.MapBuffer(ctx, GL_PIXEL_UNPACK_BUFFER_EXT,
+                                          GL_READ_ONLY_ARB, unpack->BufferObj);
+   if (!buf) {
+      _mesa_error(ctx, GL_INVALID_OPERATION, funcName, "(PBO is mapped");
+      return NULL;
    }
-   else {
-      /* software texture format */
-      makeTemp = GL_FALSE;
-   }
-
-   if (makeTemp) {
-      GLint postConvWidth = srcWidth, postConvHeight = srcHeight;
-      GLenum tmpFormat;
-      GLuint tmpComps, tmpTexelSize;
-      GLint tmpRowStride, tmpImageStride;
-      GLubyte *tmpImage;
-
-      if (transferOps & IMAGE_CONVOLUTION_BIT) {
-         _mesa_adjust_image_for_convolution(ctx, dimensions, &postConvWidth,
-                                            &postConvHeight);
-      }
-
-      tmpFormat = dstFormat->BaseFormat;
-      tmpComps = _mesa_components_in_format(tmpFormat);
-      tmpTexelSize = tmpComps * sizeof(GLchan);
-      tmpRowStride = postConvWidth * tmpTexelSize;
-      tmpImageStride = postConvWidth * postConvHeight * tmpTexelSize;
-      tmpImage = (GLubyte *) MALLOC(postConvWidth * postConvHeight *
-                                    srcDepth * tmpTexelSize);
-      if (!tmpImage) {
-         if (freeSourceData)
-            FREE((void *) srcAddr);
-         return;
-      }
 
-      transfer_teximage(ctx, dimensions, tmpFormat, tmpImage,
-                        srcWidth, srcHeight, srcDepth,
-                        0, 0, 0, /* x/y/zoffset */
-                        tmpRowStride, tmpImageStride,
-                        srcFormat, srcType, srcAddr, srcPacking, transferOps);
-
-      if (freeSourceData)
-         FREE((void *) srcAddr);
-
-      /* the temp image is our new source image */
-      srcWidth = postConvWidth;
-      srcHeight = postConvHeight;
-      srcFormat = tmpFormat;
-      srcType = CHAN_TYPE;
-      srcAddr = tmpImage;
-      srcPacking = &_mesa_native_packing;
-      freeSourceData = GL_TRUE;
-   }
-
-   if (_mesa_is_hardware_tex_format(dstFormat)) {
-      assert(makeTemp);
-      if (dimensions == 1) {
-         GLboolean b;
-         b = _mesa_convert_texsubimage1d(dstFormat->MesaFormat,
-                                         dstXoffset,
-                                         srcWidth,
-                                         srcFormat, srcType,
-                                         srcPacking, srcAddr,
-                                         dstAddr);
-         assert(b);
-      }
-      else if (dimensions == 2) {
-         GLboolean b;
-         b = _mesa_convert_texsubimage2d(dstFormat->MesaFormat,
-                                         dstXoffset, dstYoffset,
-                                         srcWidth, srcHeight,
-                                         dstRowStridePixels,
-                                         srcFormat, srcType,
-                                         srcPacking, srcAddr,
-                                         dstAddr);
-         assert(b);
-      }
-      else {
-         GLboolean b;
-         b = _mesa_convert_texsubimage3d(dstFormat->MesaFormat,
-                                      dstXoffset, dstYoffset, dstZoffset,
-                                      srcWidth, srcHeight, srcDepth,
-                                      dstRowStridePixels, dstImageStridePixels,
-                                      srcFormat, srcType,
-                                      srcPacking, srcAddr, dstAddr);
-         assert(b);
-      }
+   return ADD_POINTERS(buf, pixels);
+}
+
+
+/**
+ * Check if an unpack PBO is active prior to fetching a compressed texture
+ * image.
+ * If so, do bounds checking and map the buffer into main memory.
+ * Any errors detected will be recorded.
+ * The caller _must_ call _mesa_unmap_teximage_pbo() too!
+ */
+const GLvoid *
+_mesa_validate_pbo_compressed_teximage(GLcontext *ctx,
+                                 GLsizei imageSize, const GLvoid *pixels,
+                                 const struct gl_pixelstore_attrib *packing,
+                                 const char *funcName)
+{
+   GLubyte *buf;
+
+   if (packing->BufferObj->Name == 0) {
+      /* not using a PBO - return pointer unchanged */
+      return pixels;
    }
-   else {
-      /* software format */
-      assert(!makeTemp);
-      transfer_teximage(ctx, dimensions, dstFormat->BaseFormat, dstAddr,
-                        srcWidth, srcHeight, srcDepth,
-                        dstXoffset, dstYoffset, dstZoffset,
-                        dstRowStride, dstImageStride,
-                        srcFormat, srcType, srcAddr, srcPacking, transferOps);
+   if ((const GLubyte *) pixels + imageSize >
+       (const GLubyte *) packing->BufferObj->Size) {
+      /* out of bounds read! */
+      _mesa_error(ctx, GL_INVALID_OPERATION, funcName, "(invalid PBO access");
+      return NULL;
+   }
+
+   buf = (GLubyte*) ctx->Driver.MapBuffer(ctx, GL_PIXEL_UNPACK_BUFFER_EXT,
+                                         GL_READ_ONLY_ARB, packing->BufferObj);
+   if (!buf) {
+      _mesa_error(ctx, GL_INVALID_OPERATION, funcName, "(PBO is mapped");
+      return NULL;
    }
 
-   if (freeSourceData)
-      FREE((void *) srcAddr);  /* the temp image */
+   return ADD_POINTERS(buf, pixels);
+}
+
+
+/**
+ * This function must be called after either of the validate_pbo_*_teximage()
+ * functions.  It unmaps the PBO buffer if it was mapped earlier.
+ */
+void
+_mesa_unmap_teximage_pbo(GLcontext *ctx, const struct gl_pixelstore_attrib *unpack)
+{
+   if (unpack->BufferObj->Name) {
+      ctx->Driver.UnmapBuffer(ctx, GL_PIXEL_UNPACK_BUFFER_EXT,
+                              unpack->BufferObj);
+   }
 }
 
 
 /*
- * This is the software fallback for Driver.TexImage1D().
- * The texture image type will be GLchan.
- * The texture image format will be GL_COLOR_INDEX, GL_INTENSITY,
- * GL_LUMINANCE, GL_LUMINANCE_ALPHA, GL_ALPHA, GL_RGB or GL_RGBA.
- *
+ * This is the software fallback for Driver.TexImage1D()
+ * and Driver.CopyTexImage1D().
+ * \sa _mesa_store_teximage2d()
  */
 void
 _mesa_store_teximage1d(GLcontext *ctx, GLenum target, GLint level,
@@ -621,7 +2163,8 @@ _mesa_store_teximage1d(GLcontext *ctx, GLenum target, GLint level,
                        struct gl_texture_image *texImage)
 {
    GLint postConvWidth = width;
-   GLint texelBytes, sizeInBytes;
+   GLint sizeInBytes;
+   (void) border;
 
    if (ctx->_ImageTransferState & IMAGE_CONVOLUTION_BIT) {
       _mesa_adjust_image_for_convolution(ctx, 1, &postConvWidth, NULL);
@@ -629,59 +2172,71 @@ _mesa_store_teximage1d(GLcontext *ctx, GLenum target, GLint level,
 
    /* choose the texture format */
    assert(ctx->Driver.ChooseTextureFormat);
-   texImage->TexFormat = (*ctx->Driver.ChooseTextureFormat)(ctx,
-                                          internalFormat, format, type);
+   texImage->TexFormat = ctx->Driver.ChooseTextureFormat(ctx, internalFormat,
+                                                         format, type);
    assert(texImage->TexFormat);
-   texImage->FetchTexel = texImage->TexFormat->FetchTexel1D;
-
-   texelBytes = texImage->TexFormat->TexelBytes;
-
-   /* Compute image size, in bytes */
-   if (texImage->IsCompressed) {
-      assert(ctx->Driver.CompressedTextureSize);
-      sizeInBytes = ctx->Driver.CompressedTextureSize(ctx, texImage);
-      assert(sizeInBytes > 0);
-      texImage->CompressedSize = sizeInBytes;
-   }
-   else {
-      sizeInBytes = postConvWidth * texelBytes;
-   }
+   texImage->FetchTexelc = texImage->TexFormat->FetchTexel1D;
+   texImage->FetchTexelf = texImage->TexFormat->FetchTexel1Df;
 
    /* allocate memory */
+   if (texImage->IsCompressed)
+      sizeInBytes = texImage->CompressedSize;
+   else
+      sizeInBytes = postConvWidth * texImage->TexFormat->TexelBytes;
    texImage->Data = MESA_PBUFFER_ALLOC(sizeInBytes);
    if (!texImage->Data) {
       _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage1D");
       return;
    }
 
-   if (pixels) {
-      /* unpack image, apply transfer ops and store in texImage->Data */
-      _mesa_transfer_teximage(ctx, 1,
-                              _mesa_base_tex_format(ctx, internalFormat),
-                              texImage->TexFormat, texImage->Data,
-                              width, 1, 1, 0, 0, 0,
-                              0, /* dstRowStride */
-                              0, /* dstImageStride */
-                              format, type, pixels, packing);
-
-      /* GL_SGIS_generate_mipmap */
-      if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
-         _mesa_generate_mipmap(ctx,
-                               &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
-                               texObj);
+   pixels = _mesa_validate_pbo_teximage(ctx, 1, width, 1, 1, format, type, pixels,
+                                  packing, "glTexImage1D");
+   if (!pixels) {
+      /* Note: we check for a NULL image pointer here, _after_ we allocated
+       * memory for the texture.  That's what the GL spec calls for.
+       */
+      return;
+   }
+   else {
+      const GLint dstRowStride = 0, dstImageStride = 0;
+      GLboolean success;
+      ASSERT(texImage->TexFormat->StoreImage);
+      success = texImage->TexFormat->StoreImage(ctx, 1, texImage->Format,
+                                                texImage->TexFormat,
+                                                texImage->Data,
+                                                0, 0, 0,  /* dstX/Y/Zoffset */
+                                                dstRowStride, dstImageStride,
+                                                width, 1, 1,
+                                                format, type, pixels, packing);
+      if (!success) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage1D");
       }
    }
+
+   /* GL_SGIS_generate_mipmap */
+   if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+      _mesa_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+                            texObj);
+   }
+
+   _mesa_unmap_teximage_pbo(ctx, packing);
 }
 
 
-/*
- * This is the software fallback for Driver.TexImage2D().
- * The texture image type will be GLchan.
- * The texture image format will be GL_COLOR_INDEX, GL_INTENSITY,
- * GL_LUMINANCE, GL_LUMINANCE_ALPHA, GL_ALPHA, GL_RGB or GL_RGBA.
+/**
+ * This is the software fallback for Driver.TexImage2D()
+ * and Driver.CopyTexImage2D().
+ * We store the image in heap memory.  We know nothing about on-board
+ * VRAM here.  But since most DRI drivers rely on keeping a copy of all
+ * textures in main memory, this routine will typically be used by
+ * hardware drivers too.
  *
- * NOTE: if real texture compression is supported, this whole function
- * will need to be overridden.
+ * Reasons why a driver might override this function:
+ *  - Special memory allocation needs (VRAM, AGP, etc)
+ *  - Unusual row/image strides or padding
+ *  - Special housekeeping
+ *  - Using VRAM-based Pixel Buffer Objects
  */
 void
 _mesa_store_teximage2d(GLcontext *ctx, GLenum target, GLint level,
@@ -694,6 +2249,7 @@ _mesa_store_teximage2d(GLcontext *ctx, GLenum target, GLint level,
 {
    GLint postConvWidth = width, postConvHeight = height;
    GLint texelBytes, sizeInBytes;
+   (void) border;
 
    if (ctx->_ImageTransferState & IMAGE_CONVOLUTION_BIT) {
       _mesa_adjust_image_for_convolution(ctx, 2, &postConvWidth,
@@ -705,55 +2261,68 @@ _mesa_store_teximage2d(GLcontext *ctx, GLenum target, GLint level,
    texImage->TexFormat = (*ctx->Driver.ChooseTextureFormat)(ctx,
                                           internalFormat, format, type);
    assert(texImage->TexFormat);
-   texImage->FetchTexel = texImage->TexFormat->FetchTexel2D;
+   texImage->FetchTexelc = texImage->TexFormat->FetchTexel2D;
+   texImage->FetchTexelf = texImage->TexFormat->FetchTexel2Df;
 
    texelBytes = texImage->TexFormat->TexelBytes;
 
-   /* Compute image size, in bytes */
-   if (texImage->IsCompressed) {
-      assert(ctx->Driver.CompressedTextureSize);
-      sizeInBytes = ctx->Driver.CompressedTextureSize(ctx, texImage);
-      assert(sizeInBytes > 0);
-      texImage->CompressedSize = sizeInBytes;
-   }
-   else {
-      sizeInBytes = postConvWidth * postConvHeight * texelBytes;
-   }
-
    /* allocate memory */
+   if (texImage->IsCompressed)
+      sizeInBytes = texImage->CompressedSize;
+   else
+      sizeInBytes = postConvWidth * postConvHeight * texelBytes;
    texImage->Data = MESA_PBUFFER_ALLOC(sizeInBytes);
    if (!texImage->Data) {
       _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage2D");
       return;
    }
 
-   if (pixels) {
-      /* unpack image, apply transfer ops and store in texImage->Data */
-      _mesa_transfer_teximage(ctx, 2,
-                              _mesa_base_tex_format(ctx, internalFormat),
-                              texImage->TexFormat, texImage->Data,
-                              width, height, 1, 0, 0, 0,
-                              texImage->Width * texelBytes,
-                              0, /* dstImageStride */
-                              format, type, pixels, packing);
-
-      /* GL_SGIS_generate_mipmap */
-      if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
-         _mesa_generate_mipmap(ctx,
-                               &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
-                               texObj);
+   pixels = _mesa_validate_pbo_teximage(ctx, 2, width, height, 1, format, type,
+                                  pixels, packing, "glTexImage2D");
+   if (!pixels) {
+      /* Note: we check for a NULL image pointer here, _after_ we allocated
+       * memory for the texture.  That's what the GL spec calls for.
+       */
+      return;
+   }
+   else {
+      GLint dstRowStride, dstImageStride = 0;
+      GLboolean success;
+      if (texImage->IsCompressed) {
+         dstRowStride = _mesa_compressed_row_stride(texImage->IntFormat,width);
       }
+      else {
+         dstRowStride = postConvWidth * texImage->TexFormat->TexelBytes;
+      }
+      ASSERT(texImage->TexFormat->StoreImage);
+      success = texImage->TexFormat->StoreImage(ctx, 2, texImage->Format,
+                                                texImage->TexFormat,
+                                                texImage->Data,
+                                                0, 0, 0,  /* dstX/Y/Zoffset */
+                                                dstRowStride, dstImageStride,
+                                                width, height, 1,
+                                                format, type, pixels, packing);
+      if (!success) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage2D");
+      }
+   }
+
+   /* GL_SGIS_generate_mipmap */
+   if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+      _mesa_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+                            texObj);
    }
+
+   _mesa_unmap_teximage_pbo(ctx, packing);
 }
 
 
 
-/*
- * This is the software fallback for Driver.TexImage3D().
- * The texture image type will be GLchan.
- * The texture image format will be GL_COLOR_INDEX, GL_INTENSITY,
- * GL_LUMINANCE, GL_LUMINANCE_ALPHA, GL_ALPHA, GL_RGB or GL_RGBA.
- *
+/**
+ * This is the software fallback for Driver.TexImage3D()
+ * and Driver.CopyTexImage3D().
+ * \sa _mesa_store_teximage2d()
  */
 void
 _mesa_store_teximage3d(GLcontext *ctx, GLenum target, GLint level,
@@ -765,58 +2334,77 @@ _mesa_store_teximage3d(GLcontext *ctx, GLenum target, GLint level,
                        struct gl_texture_image *texImage)
 {
    GLint texelBytes, sizeInBytes;
+   (void) border;
 
    /* choose the texture format */
    assert(ctx->Driver.ChooseTextureFormat);
    texImage->TexFormat = (*ctx->Driver.ChooseTextureFormat)(ctx,
                                           internalFormat, format, type);
    assert(texImage->TexFormat);
-   texImage->FetchTexel = texImage->TexFormat->FetchTexel3D;
+   texImage->FetchTexelc = texImage->TexFormat->FetchTexel3D;
+   texImage->FetchTexelf = texImage->TexFormat->FetchTexel3Df;
 
    texelBytes = texImage->TexFormat->TexelBytes;
 
-   /* Compute image size, in bytes */
-   if (texImage->IsCompressed) {
-      assert(ctx->Driver.CompressedTextureSize);
-      sizeInBytes = ctx->Driver.CompressedTextureSize(ctx, texImage);
-      assert(sizeInBytes > 0);
-      texImage->CompressedSize = sizeInBytes;
-   }
-   else {
-      sizeInBytes = width * height * depth * texelBytes;
-   }
-
    /* allocate memory */
+   if (texImage->IsCompressed)
+      sizeInBytes = texImage->CompressedSize;
+   else
+      sizeInBytes = width * height * depth * texelBytes;
    texImage->Data = MESA_PBUFFER_ALLOC(sizeInBytes);
    if (!texImage->Data) {
       _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage3D");
       return;
    }
 
-   if (pixels) {
-      /* unpack image, apply transfer ops and store in texImage->Data */
-      _mesa_transfer_teximage(ctx, 3,
-                              _mesa_base_tex_format(ctx, internalFormat),
-                              texImage->TexFormat, texImage->Data,
-                              width, height, depth, 0, 0, 0,
-                              texImage->Width * texelBytes,
-                              texImage->Width * texImage->Height * texelBytes,
-                              format, type, pixels, packing);
-
-      /* GL_SGIS_generate_mipmap */
-      if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
-         _mesa_generate_mipmap(ctx,
-                               &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
-                               texObj);
+   pixels = _mesa_validate_pbo_teximage(ctx, 3, width, height, depth, format, type,
+                                  pixels, packing, "glTexImage3D");
+   if (!pixels) {
+      /* Note: we check for a NULL image pointer here, _after_ we allocated
+       * memory for the texture.  That's what the GL spec calls for.
+       */
+      return;
+   }
+   else {
+      GLint dstRowStride, dstImageStride;
+      GLboolean success;
+      if (texImage->IsCompressed) {
+         dstRowStride = _mesa_compressed_row_stride(texImage->IntFormat,width);
+         dstImageStride = 0;
+      }
+      else {
+         dstRowStride = width * texImage->TexFormat->TexelBytes;
+         dstImageStride = dstRowStride * height;
       }
+      ASSERT(texImage->TexFormat->StoreImage);
+      success = texImage->TexFormat->StoreImage(ctx, 3, texImage->Format,
+                                                texImage->TexFormat,
+                                                texImage->Data,
+                                                0, 0, 0,  /* dstX/Y/Zoffset */
+                                                dstRowStride, dstImageStride,
+                                                width, height, depth,
+                                                format, type, pixels, packing);
+      if (!success) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage3D");
+      }
+   }
+
+   /* GL_SGIS_generate_mipmap */
+   if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+      _mesa_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+                            texObj);
    }
+
+   _mesa_unmap_teximage_pbo(ctx, packing);
 }
 
 
 
 
 /*
- * This is the software fallback for Driver.TexSubImage1D().
+ * This is the software fallback for Driver.TexSubImage1D()
+ * and Driver.CopyTexSubImage1D().
  */
 void
 _mesa_store_texsubimage1d(GLcontext *ctx, GLenum target, GLint level,
@@ -826,25 +2414,42 @@ _mesa_store_texsubimage1d(GLcontext *ctx, GLenum target, GLint level,
                           struct gl_texture_object *texObj,
                           struct gl_texture_image *texImage)
 {
-   _mesa_transfer_teximage(ctx, 1,
-                           _mesa_base_tex_format(ctx, texImage->IntFormat),
-                           texImage->TexFormat, texImage->Data,
-                           width, 1, 1, /* src size */
-                           xoffset, 0, 0, /* dest offsets */
-                           0, /* dstRowStride */
-                           0, /* dstImageStride */
-                           format, type, pixels, packing);
+   pixels = _mesa_validate_pbo_teximage(ctx, 1, width, 1, 1, format, type, pixels,
+                                  packing, "glTexSubImage1D");
+   if (!pixels)
+      return;
+
+   {
+      const GLint dstRowStride = 0, dstImageStride = 0;
+      GLboolean success;
+      ASSERT(texImage->TexFormat->StoreImage);
+      success = texImage->TexFormat->StoreImage(ctx, 1, texImage->Format,
+                                                texImage->TexFormat,
+                                                texImage->Data,
+                                                xoffset, 0, 0,  /* offsets */
+                                                dstRowStride, dstImageStride,
+                                                width, 1, 1,
+                                                format, type, pixels, packing);
+      if (!success) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage1D");
+      }
+   }
 
    /* GL_SGIS_generate_mipmap */
    if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
-      _mesa_generate_mipmap(ctx, &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+      _mesa_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
                             texObj);
    }
+
+   _mesa_unmap_teximage_pbo(ctx, packing);
 }
 
 
-/*
- * This is the software fallback for Driver.TexSubImage2D().
+
+/**
+ * This is the software fallback for Driver.TexSubImage2D()
+ * and Driver.CopyTexSubImage2D().
  */
 void
 _mesa_store_texsubimage2d(GLcontext *ctx, GLenum target, GLint level,
@@ -855,25 +2460,48 @@ _mesa_store_texsubimage2d(GLcontext *ctx, GLenum target, GLint level,
                           struct gl_texture_object *texObj,
                           struct gl_texture_image *texImage)
 {
-   _mesa_transfer_teximage(ctx, 2,
-                           _mesa_base_tex_format(ctx, texImage->IntFormat),
-                           texImage->TexFormat, texImage->Data,
-                           width, height, 1, /* src size */
-                           xoffset, yoffset, 0, /* dest offsets */
-                           texImage->Width * texImage->TexFormat->TexelBytes,
-                           0, /* dstImageStride */
-                           format, type, pixels, packing);
+   pixels = _mesa_validate_pbo_teximage(ctx, 2, width, height, 1, format, type,
+                                  pixels, packing, "glTexSubImage2D");
+   if (!pixels)
+      return;
+
+   {
+      GLint dstRowStride = 0, dstImageStride = 0;
+      GLboolean success;
+      if (texImage->IsCompressed) {
+         dstRowStride = _mesa_compressed_row_stride(texImage->IntFormat,
+                                                    texImage->Width);
+      }
+      else {
+         dstRowStride = texImage->Width * texImage->TexFormat->TexelBytes;
+      }
+      ASSERT(texImage->TexFormat->StoreImage);
+      success = texImage->TexFormat->StoreImage(ctx, 2, texImage->Format,
+                                                texImage->TexFormat,
+                                                texImage->Data,
+                                                xoffset, yoffset, 0,
+                                                dstRowStride, dstImageStride,
+                                                width, height, 1,
+                                                format, type, pixels, packing);
+      if (!success) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage2D");
+      }
+   }
 
    /* GL_SGIS_generate_mipmap */
    if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
-      _mesa_generate_mipmap(ctx, &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+      _mesa_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
                             texObj);
    }
+
+   _mesa_unmap_teximage_pbo(ctx, packing);
 }
 
 
 /*
  * This is the software fallback for Driver.TexSubImage3D().
+ * and Driver.CopyTexSubImage3D().
  */
 void
 _mesa_store_texsubimage3d(GLcontext *ctx, GLenum target, GLint level,
@@ -884,23 +2512,45 @@ _mesa_store_texsubimage3d(GLcontext *ctx, GLenum target, GLint level,
                           struct gl_texture_object *texObj,
                           struct gl_texture_image *texImage)
 {
-   const GLint texelBytes = texImage->TexFormat->TexelBytes;
-   _mesa_transfer_teximage(ctx, 3,
-                           _mesa_base_tex_format(ctx, texImage->IntFormat),
-                           texImage->TexFormat, texImage->Data,
-                           width, height, depth, /* src size */
-                           xoffset, yoffset, xoffset, /* dest offsets */
-                           texImage->Width * texelBytes,
-                           texImage->Width * texImage->Height * texelBytes,
-                           format, type, pixels, packing);
+   pixels = _mesa_validate_pbo_teximage(ctx, 3, width, height, depth, format, type,
+                                  pixels, packing, "glTexSubImage3D");
+   if (!pixels)
+      return;
+
+   {
+      GLint dstRowStride, dstImageStride;
+      GLboolean success;
+      if (texImage->IsCompressed) {
+         dstRowStride = _mesa_compressed_row_stride(texImage->IntFormat,
+                                                    texImage->Width);
+         dstImageStride = 0; /* XXX fix */
+      }
+      else {
+         dstRowStride = texImage->Width * texImage->TexFormat->TexelBytes;
+         dstImageStride = dstRowStride * texImage->Height;
+      }
+      ASSERT(texImage->TexFormat->StoreImage);
+      success = texImage->TexFormat->StoreImage(ctx, 3, texImage->Format,
+                                                texImage->TexFormat,
+                                                texImage->Data,
+                                                xoffset, yoffset, zoffset,
+                                                dstRowStride, dstImageStride,
+                                                width, height, depth,
+                                                format, type, pixels, packing);
+      if (!success) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage3D");
+      }
+   }
+
    /* GL_SGIS_generate_mipmap */
    if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
-      _mesa_generate_mipmap(ctx, &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+      _mesa_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
                             texObj);
    }
-}
-
 
+   _mesa_unmap_teximage_pbo(ctx, packing);
+}
 
 
 /*
@@ -914,9 +2564,14 @@ _mesa_store_compressed_teximage1d(GLcontext *ctx, GLenum target, GLint level,
                                   struct gl_texture_object *texObj,
                                   struct gl_texture_image *texImage)
 {
-   /* Nothing here.
-    * The device driver has to do it all.
-    */
+   /* this space intentionally left blank */
+   (void) ctx;
+   (void) target; (void) level;
+   (void) internalFormat;
+   (void) width; (void) border;
+   (void) imageSize; (void) data;
+   (void) texObj;
+   (void) texImage;
 }
 
 
@@ -932,9 +2587,50 @@ _mesa_store_compressed_teximage2d(GLcontext *ctx, GLenum target, GLint level,
                                   struct gl_texture_object *texObj,
                                   struct gl_texture_image *texImage)
 {
-   /* Nothing here.
-    * The device driver has to do it all.
+   (void) width; (void) height; (void) border;
+
+   /* This is pretty simple, basically just do a memcpy without worrying
+    * about the usual image unpacking or image transfer operations.
     */
+   ASSERT(texObj);
+   ASSERT(texImage);
+   ASSERT(texImage->Width > 0);
+   ASSERT(texImage->Height > 0);
+   ASSERT(texImage->Depth == 1);
+   ASSERT(texImage->Data == NULL); /* was freed in glCompressedTexImage2DARB */
+
+   /* choose the texture format */
+   assert(ctx->Driver.ChooseTextureFormat);
+   texImage->TexFormat = (*ctx->Driver.ChooseTextureFormat)(ctx,
+                                          internalFormat, 0, 0);
+   assert(texImage->TexFormat);
+   texImage->FetchTexelc = texImage->TexFormat->FetchTexel2D;
+   texImage->FetchTexelf = texImage->TexFormat->FetchTexel2Df;
+
+   /* allocate storage */
+   texImage->Data = MESA_PBUFFER_ALLOC(imageSize);
+   if (!texImage->Data) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexImage2DARB");
+      return;
+   }
+
+   data = _mesa_validate_pbo_compressed_teximage(ctx, imageSize, data, &ctx->Unpack,
+                                           "glCompressedTexImage2D");
+   if (!data)
+      return;
+
+   /* copy the data */
+   ASSERT(texImage->CompressedSize == (GLuint) imageSize);
+   MEMCPY(texImage->Data, data, imageSize);
+
+   /* GL_SGIS_generate_mipmap */
+   if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+      _mesa_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+                            texObj);
+   }
+
+   _mesa_unmap_teximage_pbo(ctx, &ctx->Unpack);
 }
 
 
@@ -951,66 +2647,126 @@ _mesa_store_compressed_teximage3d(GLcontext *ctx, GLenum target, GLint level,
                                   struct gl_texture_object *texObj,
                                   struct gl_texture_image *texImage)
 {
-   /* Nothing here.
-    * The device driver has to do it all.
-    */
+   /* this space intentionally left blank */
+   (void) ctx;
+   (void) target; (void) level;
+   (void) internalFormat;
+   (void) width; (void) height; (void) depth;
+   (void) border;
+   (void) imageSize; (void) data;
+   (void) texObj;
+   (void) texImage;
 }
 
 
 
-/*
- * Fallback for Driver.GetCompressedTexImage3D()
- * This will probably work find for hardware drivers.  That is, hardware
- * drivers won't have to override this function, unless the compressed
- * texture must first be fetched from the TRAM.
+/**
+ * Fallback for Driver.CompressedTexSubImage1D()
  */
 void
-_mesa_get_compressed_teximage(GLcontext *ctx, GLenum target,
-                              GLint level, void *image,
-                              const struct gl_texture_object *texObj,
-                              struct gl_texture_image *texImage)
+_mesa_store_compressed_texsubimage1d(GLcontext *ctx, GLenum target,
+                                     GLint level,
+                                     GLint xoffset, GLsizei width,
+                                     GLenum format,
+                                     GLsizei imageSize, const GLvoid *data,
+                                     struct gl_texture_object *texObj,
+                                     struct gl_texture_image *texImage)
 {
-   assert(texImage->IsCompressed);
-   assert(texImage->CompressedSize > 0);
-   MEMCPY(image, texImage->Data, texImage->CompressedSize);
+   /* this space intentionally left blank */
+   (void) ctx;
+   (void) target; (void) level;
+   (void) xoffset; (void) width;
+   (void) format;
+   (void) imageSize; (void) data;
+   (void) texObj;
+   (void) texImage;
 }
 
 
-
-/*
- * This is the fallback for Driver.TestProxyTexImage().
+/**
+ * Fallback for Driver.CompressedTexSubImage2D()
  */
-GLboolean
-_mesa_test_proxy_teximage(GLcontext *ctx, GLenum target, GLint level,
-                          GLint internalFormat, GLenum format, GLenum type,
-                          GLint width, GLint height, GLint depth, GLint border)
+void
+_mesa_store_compressed_texsubimage2d(GLcontext *ctx, GLenum target,
+                                     GLint level,
+                                     GLint xoffset, GLint yoffset,
+                                     GLsizei width, GLsizei height,
+                                     GLenum format,
+                                     GLsizei imageSize, const GLvoid *data,
+                                     struct gl_texture_object *texObj,
+                                     struct gl_texture_image *texImage)
 {
-   struct gl_texture_unit *texUnit;
-   struct gl_texture_object *texObj;
-   struct gl_texture_image *texImage;
-
+   GLint bytesPerRow, destRowStride, srcRowStride;
+   GLint i, rows;
+   GLubyte *dest;
+   const GLubyte *src;
    (void) format;
-   (void) type;
 
-   texUnit = &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
-   texObj = _mesa_select_tex_object(ctx, texUnit, target);
-   texImage = _mesa_select_tex_image(ctx, texUnit, target, level);
+   /* these should have been caught sooner */
+   ASSERT((width & 3) == 0 || width == 2 || width == 1);
+   ASSERT((height & 3) == 0 || height == 2 || height == 1);
+   ASSERT((xoffset & 3) == 0);
+   ASSERT((yoffset & 3) == 0);
 
-   /* We always pass.
-    * The core Mesa code will have already tested the image size, etc.
-    * If a driver has more stringent texture limits to enforce it will
-    * have to override this function.
-    */
-   /* choose the texture format */
-   assert(ctx->Driver.ChooseTextureFormat);
-   texImage->TexFormat = (*ctx->Driver.ChooseTextureFormat)(ctx,
-                                          internalFormat, format, type);
-   assert(texImage->TexFormat);
+   data = _mesa_validate_pbo_compressed_teximage(ctx, imageSize, data, &ctx->Unpack,
+                                           "glCompressedTexSubImage2D");
+   if (!data)
+      return;
 
-   return GL_TRUE;
+   srcRowStride = _mesa_compressed_row_stride(texImage->IntFormat, width);
+   src = (const GLubyte *) data;
+
+   destRowStride = _mesa_compressed_row_stride(texImage->IntFormat,
+                                               texImage->Width);
+   dest = _mesa_compressed_image_address(xoffset, yoffset, 0,
+                                         texImage->IntFormat,
+                                         texImage->Width,
+                              (GLubyte*) texImage->Data);
+
+   bytesPerRow = srcRowStride;
+   rows = height / 4;
+
+   for (i = 0; i < rows; i++) {
+      MEMCPY(dest, src, bytesPerRow);
+      dest += destRowStride;
+      src += srcRowStride;
+   }
+
+   /* GL_SGIS_generate_mipmap */
+   if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+      _mesa_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+                            texObj);
+   }
+
+   _mesa_unmap_teximage_pbo(ctx, &ctx->Unpack);
 }
 
 
+/**
+ * Fallback for Driver.CompressedTexSubImage3D()
+ */
+void
+_mesa_store_compressed_texsubimage3d(GLcontext *ctx, GLenum target,
+                                GLint level,
+                                GLint xoffset, GLint yoffset, GLint zoffset,
+                                GLsizei width, GLsizei height, GLsizei depth,
+                                GLenum format,
+                                GLsizei imageSize, const GLvoid *data,
+                                struct gl_texture_object *texObj,
+                                struct gl_texture_image *texImage)
+{
+   /* this space intentionally left blank */
+   (void) ctx;
+   (void) target; (void) level;
+   (void) xoffset; (void) yoffset; (void) zoffset;
+   (void) width; (void) height; (void) depth;
+   (void) format;
+   (void) imageSize; (void) data;
+   (void) texObj;
+   (void) texImage;
+}
+
 
 /*
  * Average together two rows of a source image to produce a single new
@@ -1026,7 +2782,9 @@ do_row(const struct gl_texture_format *format, GLint srcWidth,
    const GLuint k0 = (srcWidth == dstWidth) ? 0 : 1;
    const GLuint colStride = (srcWidth == dstWidth) ? 1 : 2;
 
+   /* This assertion is no longer valid with non-power-of-2 textures
    assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth);
+   */
 
    switch (format->MesaFormat) {
    case MESA_FORMAT_RGBA:
@@ -1068,7 +2826,6 @@ do_row(const struct gl_texture_format *format, GLint srcWidth,
    case MESA_FORMAT_ALPHA:
    case MESA_FORMAT_LUMINANCE:
    case MESA_FORMAT_INTENSITY:
-   case MESA_FORMAT_COLOR_INDEX:
       {
          GLuint i, j, k;
          const GLchan *rowA = (const GLchan *) srcRowA;
@@ -1095,7 +2852,7 @@ do_row(const struct gl_texture_format *format, GLint srcWidth,
          }
       }
       return;
-   case MESA_FORMAT_DEPTH_COMPONENT:
+   case MESA_FORMAT_DEPTH_COMPONENT_FLOAT32:
       {
          GLuint i, j, k;
          const GLfloat *rowA = (const GLfloat *) srcRowA;
@@ -1107,9 +2864,23 @@ do_row(const struct gl_texture_format *format, GLint srcWidth,
          }
       }
       return;
+   case MESA_FORMAT_DEPTH_COMPONENT16:
+      {
+         GLuint i, j, k;
+         const GLushort *rowA = (const GLushort *) srcRowA;
+         const GLushort *rowB = (const GLushort *) srcRowB;
+         GLushort *dst = (GLushort *) dstRow;
+         for (i = j = 0, k = k0; i < (GLuint) dstWidth;
+              i++, j += colStride, k += colStride) {
+            dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
+         }
+      }
+      return;
    /* Begin hardware formats */
    case MESA_FORMAT_RGBA8888:
+   case MESA_FORMAT_RGBA8888_REV:
    case MESA_FORMAT_ARGB8888:
+   case MESA_FORMAT_ARGB8888_REV:
       {
          GLuint i, j, k;
          const GLubyte (*rowA)[4] = (const GLubyte (*)[4]) srcRowA;
@@ -1129,6 +2900,7 @@ do_row(const struct gl_texture_format *format, GLint srcWidth,
       }
       return;
    case MESA_FORMAT_RGB888:
+   case MESA_FORMAT_BGR888:
       {
          GLuint i, j, k;
          const GLubyte (*rowA)[3] = (const GLubyte (*)[3]) srcRowA;
@@ -1146,6 +2918,7 @@ do_row(const struct gl_texture_format *format, GLint srcWidth,
       }
       return;
    case MESA_FORMAT_RGB565:
+   case MESA_FORMAT_RGB565_REV:
       {
          GLuint i, j, k;
          const GLushort *rowA = (const GLushort *) srcRowA;
@@ -1165,14 +2938,15 @@ do_row(const struct gl_texture_format *format, GLint srcWidth,
             const GLint rowAb1 = (rowA[k] >> 11) & 0x1f;
             const GLint rowBb0 = (rowB[j] >> 11) & 0x1f;
             const GLint rowBb1 = (rowB[k] >> 11) & 0x1f;
-            const GLint red   = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 4;
-            const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 4;
-            const GLint blue  = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 4;
+            const GLint red   = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
+            const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
+            const GLint blue  = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
             dst[i] = (blue << 11) | (green << 5) | red;
          }
       }
       return;
    case MESA_FORMAT_ARGB4444:
+   case MESA_FORMAT_ARGB4444_REV:
       {
          GLuint i, j, k;
          const GLushort *rowA = (const GLushort *) srcRowA;
@@ -1196,15 +2970,16 @@ do_row(const struct gl_texture_format *format, GLint srcWidth,
             const GLint rowAa1 = (rowA[k] >> 12) & 0xf;
             const GLint rowBa0 = (rowB[j] >> 12) & 0xf;
             const GLint rowBa1 = (rowB[k] >> 12) & 0xf;
-            const GLint red   = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 4;
-            const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 4;
-            const GLint blue  = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 4;
-            const GLint alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 4;
+            const GLint red   = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
+            const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
+            const GLint blue  = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
+            const GLint alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
             dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red;
          }
       }
       return;
    case MESA_FORMAT_ARGB1555:
+   case MESA_FORMAT_ARGB1555_REV: /* XXX broken? */
       {
          GLuint i, j, k;
          const GLushort *rowA = (const GLushort *) srcRowA;
@@ -1228,15 +3003,16 @@ do_row(const struct gl_texture_format *format, GLint srcWidth,
             const GLint rowAa1 = (rowA[k] >> 15) & 0x1;
             const GLint rowBa0 = (rowB[j] >> 15) & 0x1;
             const GLint rowBa1 = (rowB[k] >> 15) & 0x1;
-            const GLint red   = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 4;
-            const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 4;
-            const GLint blue  = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 4;
-            const GLint alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 4;
+            const GLint red   = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
+            const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
+            const GLint blue  = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
+            const GLint alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
             dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red;
          }
       }
       return;
    case MESA_FORMAT_AL88:
+   case MESA_FORMAT_AL88_REV:
       {
          GLuint i, j, k;
          const GLubyte (*rowA)[2] = (const GLubyte (*)[2]) srcRowA;
@@ -1271,9 +3047,9 @@ do_row(const struct gl_texture_format *format, GLint srcWidth,
             const GLint rowAb1 = (rowA[k] >> 5) & 0x7;
             const GLint rowBb0 = (rowB[j] >> 5) & 0x7;
             const GLint rowBb1 = (rowB[k] >> 5) & 0x7;
-            const GLint red   = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 4;
-            const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 4;
-            const GLint blue  = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 4;
+            const GLint red   = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
+            const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
+            const GLint blue  = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
             dst[i] = (blue << 5) | (green << 2) | red;
          }
       }
@@ -1293,6 +3069,148 @@ do_row(const struct gl_texture_format *format, GLint srcWidth,
          }
       }
       return;
+   case MESA_FORMAT_RGBA_FLOAT32:
+      {
+         GLuint i, j, k;
+         const GLfloat (*rowA)[4] = (const GLfloat (*)[4]) srcRowA;
+         const GLfloat (*rowB)[4] = (const GLfloat (*)[4]) srcRowB;
+         GLfloat (*dst)[4] = (GLfloat (*)[4]) dstRow;
+         for (i = j = 0, k = k0; i < (GLuint) dstWidth;
+              i++, j += colStride, k += colStride) {
+            dst[i][0] = (rowA[j][0] + rowA[k][0] +
+                         rowB[j][0] + rowB[k][0]) * 0.25F;
+            dst[i][1] = (rowA[j][1] + rowA[k][1] +
+                         rowB[j][1] + rowB[k][1]) * 0.25F;
+            dst[i][2] = (rowA[j][2] + rowA[k][2] +
+                         rowB[j][2] + rowB[k][2]) * 0.25F;
+            dst[i][3] = (rowA[j][3] + rowA[k][3] +
+                         rowB[j][3] + rowB[k][3]) * 0.25F;
+         }
+      }
+      return;
+   case MESA_FORMAT_RGBA_FLOAT16:
+      {
+         GLuint i, j, k, comp;
+         const GLhalfARB (*rowA)[4] = (const GLhalfARB (*)[4]) srcRowA;
+         const GLhalfARB (*rowB)[4] = (const GLhalfARB (*)[4]) srcRowB;
+         GLhalfARB (*dst)[4] = (GLhalfARB (*)[4]) dstRow;
+         for (i = j = 0, k = k0; i < (GLuint) dstWidth;
+              i++, j += colStride, k += colStride) {
+            for (comp = 0; comp < 4; comp++) {
+               GLfloat aj, ak, bj, bk;
+               aj = _mesa_half_to_float(rowA[j][comp]);
+               ak = _mesa_half_to_float(rowA[k][comp]);
+               bj = _mesa_half_to_float(rowB[j][comp]);
+               bk = _mesa_half_to_float(rowB[k][comp]);
+               dst[i][comp] = _mesa_float_to_half((aj + ak + bj + bk) * 0.25F);
+            }
+         }
+      }
+      return;
+   case MESA_FORMAT_RGB_FLOAT32:
+      {
+         GLuint i, j, k;
+         const GLfloat (*rowA)[3] = (const GLfloat (*)[3]) srcRowA;
+         const GLfloat (*rowB)[3] = (const GLfloat (*)[3]) srcRowB;
+         GLfloat (*dst)[3] = (GLfloat (*)[3]) dstRow;
+         for (i = j = 0, k = k0; i < (GLuint) dstWidth;
+              i++, j += colStride, k += colStride) {
+            dst[i][0] = (rowA[j][0] + rowA[k][0] +
+                         rowB[j][0] + rowB[k][0]) * 0.25F;
+            dst[i][1] = (rowA[j][1] + rowA[k][1] +
+                         rowB[j][1] + rowB[k][1]) * 0.25F;
+            dst[i][2] = (rowA[j][2] + rowA[k][2] +
+                         rowB[j][2] + rowB[k][2]) * 0.25F;
+         }
+      }
+      return;
+   case MESA_FORMAT_RGB_FLOAT16:
+      {
+         GLuint i, j, k, comp;
+         const GLhalfARB (*rowA)[3] = (const GLhalfARB (*)[3]) srcRowA;
+         const GLhalfARB (*rowB)[3] = (const GLhalfARB (*)[3]) srcRowB;
+         GLhalfARB (*dst)[3] = (GLhalfARB (*)[3]) dstRow;
+         for (i = j = 0, k = k0; i < (GLuint) dstWidth;
+              i++, j += colStride, k += colStride) {
+            for (comp = 0; comp < 3; comp++) {
+               GLfloat aj, ak, bj, bk;
+               aj = _mesa_half_to_float(rowA[j][comp]);
+               ak = _mesa_half_to_float(rowA[k][comp]);
+               bj = _mesa_half_to_float(rowB[j][comp]);
+               bk = _mesa_half_to_float(rowB[k][comp]);
+               dst[i][comp] = _mesa_float_to_half((aj + ak + bj + bk) * 0.25F);
+            }
+         }
+      }
+      return;
+   case MESA_FORMAT_LUMINANCE_ALPHA_FLOAT32:
+      {
+         GLuint i, j, k;
+         const GLfloat (*rowA)[2] = (const GLfloat (*)[2]) srcRowA;
+         const GLfloat (*rowB)[2] = (const GLfloat (*)[2]) srcRowB;
+         GLfloat (*dst)[2] = (GLfloat (*)[2]) dstRow;
+         for (i = j = 0, k = k0; i < (GLuint) dstWidth;
+              i++, j += colStride, k += colStride) {
+            dst[i][0] = (rowA[j][0] + rowA[k][0] +
+                         rowB[j][0] + rowB[k][0]) * 0.25F;
+            dst[i][1] = (rowA[j][1] + rowA[k][1] +
+                         rowB[j][1] + rowB[k][1]) * 0.25F;
+         }
+      }
+      return;
+   case MESA_FORMAT_LUMINANCE_ALPHA_FLOAT16:
+      {
+         GLuint i, j, k, comp;
+         const GLhalfARB (*rowA)[2] = (const GLhalfARB (*)[2]) srcRowA;
+         const GLhalfARB (*rowB)[2] = (const GLhalfARB (*)[2]) srcRowB;
+         GLhalfARB (*dst)[2] = (GLhalfARB (*)[2]) dstRow;
+         for (i = j = 0, k = k0; i < (GLuint) dstWidth;
+              i++, j += colStride, k += colStride) {
+            for (comp = 0; comp < 2; comp++) {
+               GLfloat aj, ak, bj, bk;
+               aj = _mesa_half_to_float(rowA[j][comp]);
+               ak = _mesa_half_to_float(rowA[k][comp]);
+               bj = _mesa_half_to_float(rowB[j][comp]);
+               bk = _mesa_half_to_float(rowB[k][comp]);
+               dst[i][comp] = _mesa_float_to_half((aj + ak + bj + bk) * 0.25F);
+            }
+         }
+      }
+      return;
+   case MESA_FORMAT_ALPHA_FLOAT32:
+   case MESA_FORMAT_LUMINANCE_FLOAT32:
+   case MESA_FORMAT_INTENSITY_FLOAT32:
+      {
+         GLuint i, j, k;
+         const GLfloat *rowA = (const GLfloat *) srcRowA;
+         const GLfloat *rowB = (const GLfloat *) srcRowB;
+         GLfloat *dst = (GLfloat *) dstRow;
+         for (i = j = 0, k = k0; i < (GLuint) dstWidth;
+              i++, j += colStride, k += colStride) {
+            dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F;
+         }
+      }
+      return;
+   case MESA_FORMAT_ALPHA_FLOAT16:
+   case MESA_FORMAT_LUMINANCE_FLOAT16:
+   case MESA_FORMAT_INTENSITY_FLOAT16:
+      {
+         GLuint i, j, k;
+         const GLhalfARB *rowA = (const GLhalfARB *) srcRowA;
+         const GLhalfARB *rowB = (const GLhalfARB *) srcRowB;
+         GLhalfARB *dst = (GLhalfARB *) dstRow;
+         for (i = j = 0, k = k0; i < (GLuint) dstWidth;
+              i++, j += colStride, k += colStride) {
+            GLfloat aj, ak, bj, bk;
+            aj = _mesa_half_to_float(rowA[j]);
+            ak = _mesa_half_to_float(rowA[k]);
+            bj = _mesa_half_to_float(rowB[j]);
+            bk = _mesa_half_to_float(rowB[k]);
+            dst[i] = _mesa_float_to_half((aj + ak + bj + bk) * 0.25F);
+         }
+      }
+      return;
+
    default:
       _mesa_problem(NULL, "bad format in do_row()");
    }
@@ -1346,9 +3264,7 @@ make_2d_mipmap(const struct gl_texture_format *format, GLint border,
    const GLint dstRowStride = bpt * dstWidth;
    const GLubyte *srcA, *srcB;
    GLubyte *dst;
-   GLint row, colStride;
-
-   colStride = (srcWidth == dstWidth) ? 1 : 2;
+   GLint row;
 
    /* Compute src and dst pointers, skipping any border */
    srcA = srcPtr + border * ((srcWidth + 1) * bpt);
@@ -1470,7 +3386,7 @@ make_3d_mipmap(const struct gl_texture_format *format, GLint border,
     */
 
    /*
-   printf("mip3d %d x %d x %d  ->  %d x %d x %d\n",
+   _mesa_printf("mip3d %d x %d x %d  ->  %d x %d x %d\n",
           srcWidth, srcHeight, srcDepth, dstWidth, dstHeight, dstDepth);
    */
 
@@ -1603,53 +3519,74 @@ make_3d_mipmap(const struct gl_texture_format *format, GLint border,
  * Stop at texObj's MaxLevel or when we get to the 1x1 texture.
  */
 void
-_mesa_generate_mipmap(GLcontext *ctx,
+_mesa_generate_mipmap(GLcontext *ctx, GLenum target,
                       const struct gl_texture_unit *texUnit,
                       struct gl_texture_object *texObj)
 {
-   const GLenum targets1D[] = { GL_TEXTURE_1D, 0 };
-   const GLenum targets2D[] = { GL_TEXTURE_2D, 0 };
-   const GLenum targets3D[] = { GL_TEXTURE_3D, 0 };
-   const GLenum targetsCube[] = { GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB,
-                                  GL_TEXTURE_CUBE_MAP_NEGATIVE_X_ARB,
-                                  GL_TEXTURE_CUBE_MAP_POSITIVE_Y_ARB,
-                                  GL_TEXTURE_CUBE_MAP_NEGATIVE_Y_ARB,
-                                  GL_TEXTURE_CUBE_MAP_POSITIVE_Z_ARB,
-                                  GL_TEXTURE_CUBE_MAP_NEGATIVE_Z_ARB,
-                                  0 };
-   const GLenum targetsRect[] = { GL_TEXTURE_RECTANGLE_NV, 0 };
-   const GLenum *targets;
-   GLint level;
-   GLint maxLevels = 0;
+   const struct gl_texture_image *srcImage;
+   const struct gl_texture_format *convertFormat;
+   const GLubyte *srcData = NULL;
+   GLubyte *dstData = NULL;
+   GLint level, maxLevels;
 
    ASSERT(texObj);
-   ASSERT(texObj->Image[texObj->BaseLevel]);
+   srcImage = texObj->Image[0][texObj->BaseLevel];
+   ASSERT(srcImage);
 
-   switch (texObj->Target) {
-   case GL_TEXTURE_1D:
-      targets = targets1D;
-      maxLevels = ctx->Const.MaxTextureLevels;
-      break;
-   case GL_TEXTURE_2D:
-      targets = targets2D;
-      maxLevels = ctx->Const.MaxTextureLevels;
-      break;
-   case GL_TEXTURE_3D:
-      targets = targets3D;
-      maxLevels = ctx->Const.Max3DTextureLevels;
-      break;
-   case GL_TEXTURE_CUBE_MAP_ARB:
-      targets = targetsCube;
-      maxLevels = ctx->Const.MaxCubeTextureLevels;
-      break;
-   case GL_TEXTURE_RECTANGLE_NV:
-      targets = targetsRect;
-      maxLevels = 1;
-      break;
-   default:
-      _mesa_problem(ctx,
-                    "Bad texture object dimension in _mesa_generate_mipmaps");
-      return;
+   maxLevels = _mesa_max_texture_levels(ctx, texObj->Target);
+   ASSERT(maxLevels > 0);  /* bad target */
+
+   /* Find convertFormat - the format that do_row() will process */
+   if (srcImage->IsCompressed) {
+      /* setup for compressed textures */
+      GLuint row;
+      GLint  components, size;
+      GLchan *dst;
+
+      assert(texObj->Target == GL_TEXTURE_2D);
+
+      if (srcImage->Format == GL_RGB) {
+         convertFormat = &_mesa_texformat_rgb;
+         components = 3;
+      }
+      else if (srcImage->Format == GL_RGBA) {
+         convertFormat = &_mesa_texformat_rgba;
+         components = 4;
+      }
+      else {
+         _mesa_problem(ctx, "bad srcImage->Format in _mesa_generate_mipmaps");
+         return;
+      }
+
+      /* allocate storage for uncompressed GL_RGB or GL_RGBA images */
+      size = _mesa_bytes_per_pixel(srcImage->Format, CHAN_TYPE)
+         * srcImage->Width * srcImage->Height * srcImage->Depth + 20;
+      /* 20 extra bytes, just be safe when calling last FetchTexel */
+      srcData = (GLubyte *) MALLOC(size);
+      if (!srcData) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "generate mipmaps");
+         return;
+      }
+      dstData = (GLubyte *) MALLOC(size / 2);  /* 1/4 would probably be OK */
+      if (!dstData) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "generate mipmaps");
+         FREE((void *) srcData);
+         return;
+      }
+
+      /* decompress base image here */
+      dst = (GLchan *) srcData;
+      for (row = 0; row < srcImage->Height; row++) {
+         GLuint col;
+         for (col = 0; col < srcImage->Width; col++) {
+            srcImage->FetchTexelc(srcImage, col, row, 0, dst);
+            dst += components;
+         }
+      }
+   }
+   else {
+      /* uncompressed */
+      convertFormat = srcImage->TexFormat;
    }
 
    for (level = texObj->BaseLevel; level < texObj->MaxLevel
@@ -1660,15 +3597,14 @@ _mesa_generate_mipmap(GLcontext *ctx,
       GLint srcWidth, srcHeight, srcDepth;
       GLint dstWidth, dstHeight, dstDepth;
       GLint border, bytesPerTexel;
-      GLint t;
 
-      srcImage = texObj->Image[level];
+      /* get src image parameters */
+      srcImage = _mesa_select_tex_image(ctx, texUnit, target, level);
       ASSERT(srcImage);
       srcWidth = srcImage->Width;
       srcHeight = srcImage->Height;
       srcDepth = srcImage->Depth;
       border = srcImage->Border;
-      bytesPerTexel = srcImage->TexFormat->TexelBytes;
 
       /* compute next (level+1) image size */
       if (srcWidth - 2 * border > 1) {
@@ -1694,65 +3630,86 @@ _mesa_generate_mipmap(GLcontext *ctx,
           dstHeight == srcHeight &&
           dstDepth == srcDepth) {
          /* all done */
+         if (srcImage->IsCompressed) {
+            FREE((void *) srcData);
+            FREE(dstData);
+         }
          return;
       }
 
-      /* Need this loop just because of cubemaps */
-      for (t = 0; targets[t]; t++) {
-         ASSERT(t < 6);
+      /* get dest gl_texture_image */
+      dstImage = _mesa_get_tex_image(ctx, texUnit, target, level + 1);
+      if (!dstImage) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "generating mipmaps");
+         return;
+      }
 
-         dstImage = _mesa_select_tex_image(ctx, texUnit, targets[t], level+1);
-         if (!dstImage) {
-            dstImage = _mesa_alloc_texture_image();
-            if (!dstImage) {
-               _mesa_error(ctx, GL_OUT_OF_MEMORY, "generating mipmaps");
-               return;
-            }
-            _mesa_set_tex_image(texObj, targets[t], level + 1, dstImage);
+      /* Free old image data */
+      if (dstImage->Data)
+         MESA_PBUFFER_FREE(dstImage->Data);
+
+      /* initialize new image */
+      _mesa_init_teximage_fields(ctx, target, dstImage, dstWidth, dstHeight,
+                                 dstDepth, border, srcImage->IntFormat);
+      dstImage->DriverData = NULL;
+      dstImage->TexFormat = srcImage->TexFormat;
+      dstImage->FetchTexelc = srcImage->FetchTexelc;
+      dstImage->FetchTexelf = srcImage->FetchTexelf;
+      ASSERT(dstImage->TexFormat);
+      ASSERT(dstImage->FetchTexelc);
+      ASSERT(dstImage->FetchTexelf);
+
+      /* Alloc new teximage data buffer.
+       * Setup src and dest data pointers.
+       */
+      if (dstImage->IsCompressed) {
+         ASSERT(dstImage->CompressedSize > 0); /* set by init_teximage_fields*/
+         dstImage->Data = MESA_PBUFFER_ALLOC(dstImage->CompressedSize);
+         if (!dstImage->Data) {
+            _mesa_error(ctx, GL_OUT_OF_MEMORY, "generating mipmaps");
+            return;
          }
-
-         /* Free old image data */
-         if (dstImage->Data)
-            MESA_PBUFFER_FREE(dstImage->Data);
-
-         /* initialize new image */
-         _mesa_init_teximage_fields(ctx, t, dstImage, dstWidth, dstHeight,
-                                    dstDepth, border, srcImage->Format);
-         dstImage->DriverData = NULL;
-         dstImage->TexFormat = srcImage->TexFormat;
-         dstImage->FetchTexel = srcImage->FetchTexel;
-         ASSERT(dstImage->TexFormat);
-         ASSERT(dstImage->FetchTexel);
-
+         /* srcData and dstData are already set */
+         ASSERT(srcData);
+         ASSERT(dstData);
+      }
+      else {
+         bytesPerTexel = srcImage->TexFormat->TexelBytes;
          ASSERT(dstWidth * dstHeight * dstDepth * bytesPerTexel > 0);
-
-         /* alloc new image buffer */
          dstImage->Data = MESA_PBUFFER_ALLOC(dstWidth * dstHeight * dstDepth
-                                 * bytesPerTexel);
+                                             * bytesPerTexel);
          if (!dstImage->Data) {
             _mesa_error(ctx, GL_OUT_OF_MEMORY, "generating mipmaps");
             return;
          }
+         srcData = (const GLubyte *) srcImage->Data;
+         dstData = (GLubyte *) dstImage->Data;
+      }
 
-         /*
-          * We use simple 2x2 averaging to compute the next mipmap level.
-          */
-         switch (texObj->Target) {
+      /*
+       * We use simple 2x2 averaging to compute the next mipmap level.
+       */
+      switch (target) {
          case GL_TEXTURE_1D:
-            make_1d_mipmap(srcImage->TexFormat, border,
-               srcWidth, (const GLubyte *) srcImage->Data,
-               dstWidth, (GLubyte *) dstImage->Data);
+            make_1d_mipmap(convertFormat, border,
+                           srcWidth, srcData,
+                           dstWidth, dstData);
             break;
          case GL_TEXTURE_2D:
-         case GL_TEXTURE_CUBE_MAP_ARB:
-            make_2d_mipmap(srcImage->TexFormat, border,
-               srcWidth, srcHeight, (const GLubyte *) srcImage->Data,
-               dstWidth, dstHeight, (GLubyte *) dstImage->Data);
+         case GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB:
+         case GL_TEXTURE_CUBE_MAP_NEGATIVE_X_ARB:
+         case GL_TEXTURE_CUBE_MAP_POSITIVE_Y_ARB:
+         case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y_ARB:
+         case GL_TEXTURE_CUBE_MAP_POSITIVE_Z_ARB:
+         case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z_ARB:
+            make_2d_mipmap(convertFormat, border,
+                           srcWidth, srcHeight, srcData,
+                           dstWidth, dstHeight, dstData);
             break;
          case GL_TEXTURE_3D:
-            make_3d_mipmap(srcImage->TexFormat, border,
-               srcWidth, srcHeight, srcDepth, (const GLubyte *) srcImage->Data,
-               dstWidth, dstHeight, dstDepth, (GLubyte *) dstImage->Data);
+            make_3d_mipmap(convertFormat, border,
+                           srcWidth, srcHeight, srcDepth, srcData,
+                           dstWidth, dstHeight, dstDepth, dstData);
             break;
          case GL_TEXTURE_RECTANGLE_NV:
             /* no mipmaps, do nothing */
@@ -1760,7 +3717,311 @@ _mesa_generate_mipmap(GLcontext *ctx,
          default:
             _mesa_problem(ctx, "bad dimensions in _mesa_generate_mipmaps");
             return;
+      }
+
+      if (dstImage->IsCompressed) {
+         GLubyte *temp;
+         /* compress image from dstData into dstImage->Data */
+         const GLenum srcFormat = convertFormat->BaseFormat;
+         GLint dstRowStride = _mesa_compressed_row_stride(srcImage->IntFormat,
+                                                          dstWidth);
+         ASSERT(srcFormat == GL_RGB || srcFormat == GL_RGBA);
+         dstImage->TexFormat->StoreImage(ctx, 2, dstImage->Format,
+                                         dstImage->TexFormat,
+                                         dstImage->Data,
+                                         0, 0, 0, /* dstX/Y/Zoffset */
+                                         dstRowStride, 0, /* strides */
+                                         dstWidth, dstHeight, 1, /* size */
+                                         srcFormat, CHAN_TYPE,
+                                         dstData, /* src data, actually */
+                                         &ctx->DefaultPacking);
+         /* swap src and dest pointers */
+         temp = (GLubyte *) srcData;
+         srcData = dstData;
+         dstData = temp;
+      }
+
+   } /* loop over mipmap levels */
+}
+
+
+/**
+ * Helper function for drivers which need to rescale texture images to
+ * certain aspect ratios.
+ * 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_rescale_teximage2d (GLuint bytesPerPixel,
+                         GLuint srcStrideInPixels,
+                         GLuint dstRowStride,
+                         GLint srcWidth, GLint srcHeight,
+                         GLint dstWidth, GLint dstHeight,
+                         const GLvoid *srcImage, GLvoid *dstImage)
+{
+   GLint row, col;
+
+#define INNER_LOOP( TYPE, 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[col] = src[srcRow * srcStrideInPixels + srcCol];           \
+      }                                                                        \
+      dst = (TYPE *) ((GLubyte *) dst + dstRowStride);                 \
+   }                                                                   \
+
+#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( TYPE, /, / );                                      \
+      }                                                                        \
+      else {                                                           \
+        const GLint wScale = srcWidth / dstWidth;                      \
+        INNER_LOOP( TYPE, /, * );                                      \
+      }                                                                        \
+   }                                                                   \
+   else {                                                              \
+      const GLint hScale = srcHeight / dstHeight;                      \
+      if ( srcWidth < dstWidth ) {                                     \
+        const GLint wScale = dstWidth / srcWidth;                      \
+        INNER_LOOP( TYPE, *, / );                                      \
+      }                                                                        \
+      else {                                                           \
+        const GLint wScale = srcWidth / dstWidth;                      \
+        INNER_LOOP( TYPE, *, * );                                      \
+      }                                                                        \
+   }                                                                   \
+} while (0)
+
+   switch ( bytesPerPixel ) {
+   case 4:
+      RESCALE_IMAGE( GLuint );
+      break;
+
+   case 2:
+      RESCALE_IMAGE( GLushort );
+      break;
+
+   case 1:
+      RESCALE_IMAGE( GLubyte );
+      break;
+   default:
+      _mesa_problem(NULL,"unexpected bytes/pixel in _mesa_rescale_teximage2d");
+   }
+}
+
+
+/**
+ * Upscale an image by replication, not (typical) stretching.
+ * We use this when the image width or height is less than a
+ * certain size (4, 8) and we need to upscale an image.
+ */
+void
+_mesa_upscale_teximage2d (GLsizei inWidth, GLsizei inHeight,
+                          GLsizei outWidth, GLsizei outHeight,
+                          GLint comps, const GLchan *src, GLint srcRowStride,
+                          GLchan *dest )
+{
+   GLint i, j, k;
+
+   ASSERT(outWidth >= inWidth);
+   ASSERT(outHeight >= inHeight);
+#if 0
+   ASSERT(inWidth == 1 || inWidth == 2 || inHeight == 1 || inHeight == 2);
+   ASSERT((outWidth & 3) == 0);
+   ASSERT((outHeight & 3) == 0);
+#endif
+
+   for (i = 0; i < outHeight; i++) {
+      const GLint ii = i % inHeight;
+      for (j = 0; j < outWidth; j++) {
+         const GLint jj = j % inWidth;
+         for (k = 0; k < comps; k++) {
+            dest[(i * outWidth + j) * comps + k]
+               = src[ii * srcRowStride + jj * comps + k];
          }
-      } /* loop over tex image targets */
-   } /* loop over tex levels */
+      }
+   }
+}
+
+
+
+/**
+ * This is the software fallback for Driver.GetTexImage().
+ * All error checking will have been done before this routine is called.
+ */
+void
+_mesa_get_teximage(GLcontext *ctx, GLenum target, GLint level,
+                   GLenum format, GLenum type, GLvoid *pixels,
+                   const struct gl_texture_object *texObj,
+                   const struct gl_texture_image *texImage)
+{
+   GLuint dimensions = (target == GL_TEXTURE_3D) ? 3 : 2;
+
+   if (ctx->Pack.BufferObj->Name) {
+      /* pack texture image into a PBO */
+      GLubyte *buf;
+      if (!_mesa_validate_pbo_access(dimensions, &ctx->Pack, texImage->Width,
+                                     texImage->Height, texImage->Depth,
+                                     format, type, pixels)) {
+         _mesa_error(ctx, GL_INVALID_OPERATION,
+                     "glGetTexImage(invalid PBO access)");
+         return;
+      }
+      buf = (GLubyte *) ctx->Driver.MapBuffer(ctx, GL_PIXEL_PACK_BUFFER_EXT,
+                                              GL_WRITE_ONLY_ARB,
+                                              ctx->Pack.BufferObj);
+      if (!buf) {
+         /* buffer is already mapped - that's an error */
+         _mesa_error(ctx, GL_INVALID_OPERATION,"glGetTexImage(PBO is mapped)");
+         return;
+      }
+      pixels = ADD_POINTERS(buf, pixels);
+   }
+   else if (!pixels) {
+      /* not an error */
+      return;
+   }
+
+   {
+      const GLint width = texImage->Width;
+      const GLint height = texImage->Height;
+      const GLint depth = texImage->Depth;
+      GLint img, row;
+      for (img = 0; img < depth; img++) {
+         for (row = 0; row < height; row++) {
+            /* compute destination address in client memory */
+            GLvoid *dest = _mesa_image_address( dimensions, &ctx->Pack, pixels,
+                                                width, height, format, type,
+                                                img, row, 0);
+            assert(dest);
+
+            if (format == GL_COLOR_INDEX) {
+               GLuint indexRow[MAX_WIDTH];
+               GLint col;
+               /* Can't use FetchTexel here because that returns RGBA */
+               if (texImage->TexFormat->IndexBits == 8) {
+                  const GLubyte *src = (const GLubyte *) texImage->Data;
+                  for (col = 0; col < width; col++) {
+                     indexRow[col] = src[texImage->Width *
+                                        (img * texImage->Height + row) + col];
+                  }
+               }
+               else if (texImage->TexFormat->IndexBits == 16) {
+                  const GLushort *src = (const GLushort *) texImage->Data;
+                  for (col = 0; col < width; col++) {
+                     indexRow[col] = src[texImage->Width *
+                                        (img * texImage->Height + row) + col];
+                  }
+               }
+               else {
+                  _mesa_problem(ctx,
+                                "Color index problem in _mesa_GetTexImage");
+               }
+               _mesa_pack_index_span(ctx, width, type, dest,
+                                     indexRow, &ctx->Pack,
+                                     0 /* no image transfer */);
+            }
+            else if (format == GL_DEPTH_COMPONENT) {
+               GLfloat depthRow[MAX_WIDTH];
+               GLint col;
+               for (col = 0; col < width; col++) {
+                  (*texImage->FetchTexelf)(texImage, col, row, img,
+                                           depthRow + col);
+               }
+               _mesa_pack_depth_span(ctx, width, dest, type,
+                                     depthRow, &ctx->Pack);
+            }
+            else if (format == GL_YCBCR_MESA) {
+               /* No pixel transfer */
+               const GLint rowstride = texImage->RowStride;
+               MEMCPY(dest,
+                      (const GLushort *) texImage->Data + row * rowstride,
+                      width * sizeof(GLushort));
+               /* check for byte swapping */
+               if ((texImage->TexFormat->MesaFormat == MESA_FORMAT_YCBCR
+                    && type == GL_UNSIGNED_SHORT_8_8_REV_MESA) ||
+                   (texImage->TexFormat->MesaFormat == MESA_FORMAT_YCBCR_REV
+                    && type == GL_UNSIGNED_SHORT_8_8_MESA)) {
+                  if (!ctx->Pack.SwapBytes)
+                     _mesa_swap2((GLushort *) dest, width);
+               }
+               else if (ctx->Pack.SwapBytes) {
+                  _mesa_swap2((GLushort *) dest, width);
+               }
+            }
+            else {
+               /* general case:  convert row to RGBA format */
+               GLfloat rgba[MAX_WIDTH][4];
+               GLint col;
+               for (col = 0; col < width; col++) {
+                  (*texImage->FetchTexelf)(texImage, col, row, img, rgba[col]);
+               }
+               _mesa_pack_rgba_span_float(ctx, width,
+                                          (const GLfloat (*)[4]) rgba,
+                                          format, type, dest, &ctx->Pack,
+                                          0 /* no image transfer */);
+            } /* format */
+         } /* row */
+      } /* img */
+   }
+
+   if (ctx->Pack.BufferObj->Name) {
+      ctx->Driver.UnmapBuffer(ctx, GL_PIXEL_PACK_BUFFER_EXT,
+                              ctx->Pack.BufferObj);
+   }
+}
+
+
+
+/**
+ * This is the software fallback for Driver.GetCompressedTexImage().
+ * All error checking will have been done before this routine is called.
+ */
+void
+_mesa_get_compressed_teximage(GLcontext *ctx, GLenum target, GLint level,
+                              GLvoid *img,
+                              const struct gl_texture_object *texObj,
+                              const struct gl_texture_image *texImage)
+{
+   if (ctx->Pack.BufferObj->Name) {
+      /* pack texture image into a PBO */
+      GLubyte *buf;
+      if ((const GLubyte *) img + texImage->CompressedSize >
+          (const GLubyte *) ctx->Pack.BufferObj->Size) {
+         _mesa_error(ctx, GL_INVALID_OPERATION,
+                     "glGetCompressedTexImage(invalid PBO access)");
+         return;
+      }
+      buf = (GLubyte *) ctx->Driver.MapBuffer(ctx, GL_PIXEL_PACK_BUFFER_EXT,
+                                              GL_WRITE_ONLY_ARB,
+                                              ctx->Pack.BufferObj);
+      if (!buf) {
+         /* buffer is already mapped - that's an error */
+         _mesa_error(ctx, GL_INVALID_OPERATION,
+                     "glGetCompressedTexImage(PBO is mapped)");
+         return;
+      }
+      img = ADD_POINTERS(buf, img);
+   }
+   else if (!img) {
+      /* not an error */
+      return;
+   }
+
+   /* just memcpy, no pixelstore or pixel transfer */
+   MEMCPY(img, texImage->Data, texImage->CompressedSize);
+
+   if (ctx->Pack.BufferObj->Name) {
+      ctx->Driver.UnmapBuffer(ctx, GL_PIXEL_PACK_BUFFER_EXT,
+                              ctx->Pack.BufferObj);
+   }
 }