st/mesa: properly implement MapTextureImage with multiple mapped slices (v2)
[mesa.git] / src / mesa / state_tracker / st_cb_texture.c
index 3cea2df07d9462a850d6049332a79ec6ed160205..97bba8b7da59f38c663f8c7462269da3e99121d5 100644 (file)
@@ -1,6 +1,6 @@
 /**************************************************************************
  * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2007 VMware, Inc.
  * All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  * 
  **************************************************************************/
 
-#include "main/mfeatures.h"
 #include "main/bufferobj.h"
 #include "main/enums.h"
 #include "main/fbobject.h"
@@ -59,7 +58,6 @@
 #include "util/u_inlines.h"
 #include "pipe/p_shader_tokens.h"
 #include "util/u_tile.h"
-#include "util/u_blit.h"
 #include "util/u_format.h"
 #include "util/u_surface.h"
 #include "util/u_sampler.h"
@@ -69,7 +67,7 @@
 #define DBG if (0) printf
 
 
-static enum pipe_texture_target
+enum pipe_texture_target
 gl_target_to_pipe(GLenum target)
 {
    switch (target) {
@@ -79,6 +77,8 @@ gl_target_to_pipe(GLenum target)
    case GL_TEXTURE_2D:
    case GL_PROXY_TEXTURE_2D:
    case GL_TEXTURE_EXTERNAL_OES:
+   case GL_TEXTURE_2D_MULTISAMPLE:
+   case GL_PROXY_TEXTURE_2D_MULTISAMPLE:
       return PIPE_TEXTURE_2D;
    case GL_TEXTURE_RECTANGLE_NV:
    case GL_PROXY_TEXTURE_RECTANGLE_NV:
@@ -100,6 +100,8 @@ gl_target_to_pipe(GLenum target)
       return PIPE_TEXTURE_1D_ARRAY;
    case GL_TEXTURE_2D_ARRAY_EXT:
    case GL_PROXY_TEXTURE_2D_ARRAY_EXT:
+   case GL_TEXTURE_2D_MULTISAMPLE_ARRAY:
+   case GL_PROXY_TEXTURE_2D_MULTISAMPLE_ARRAY:
       return PIPE_TEXTURE_2D_ARRAY;
    case GL_TEXTURE_BUFFER:
       return PIPE_BUFFER;
@@ -139,7 +141,7 @@ st_NewTextureObject(struct gl_context * ctx, GLuint name, GLenum target)
    struct st_texture_object *obj = ST_CALLOC_STRUCT(st_texture_object);
 
    DBG("%s\n", __FUNCTION__);
-   _mesa_initialize_texture_object(&obj->base, name, target);
+   _mesa_initialize_texture_object(ctx, &obj->base, name, target);
 
    return &obj->base;
 }
@@ -149,13 +151,11 @@ static void
 st_DeleteTextureObject(struct gl_context *ctx,
                        struct gl_texture_object *texObj)
 {
-   struct st_context *st = st_context(ctx);
    struct st_texture_object *stObj = st_texture_object(texObj);
-   if (stObj->pt)
-      pipe_resource_reference(&stObj->pt, NULL);
-   if (stObj->sampler_view) {
-      pipe_sampler_view_release(st->pipe, &stObj->sampler_view);
-   }
+
+   pipe_resource_reference(&stObj->pt, NULL);
+   st_texture_release_all_sampler_views(stObj);
+   st_texture_free_sampler_views(stObj);
    _mesa_delete_texture_object(ctx, texObj);
 }
 
@@ -173,10 +173,12 @@ st_FreeTextureImageBuffer(struct gl_context *ctx,
       pipe_resource_reference(&stImage->pt, NULL);
    }
 
-   if (stImage->TexData) {
-      _mesa_align_free(stImage->TexData);
-      stImage->TexData = NULL;
-   }
+   _mesa_align_free(stImage->TexData);
+   stImage->TexData = NULL;
+
+   free(stImage->transfer);
+   stImage->transfer = NULL;
+   stImage->num_transfers = 0;
 }
 
 
@@ -192,6 +194,7 @@ st_MapTextureImage(struct gl_context *ctx,
    struct st_texture_image *stImage = st_texture_image(texImage);
    unsigned pipeMode;
    GLubyte *map;
+   struct pipe_transfer *transfer;
 
    pipeMode = 0x0;
    if (mode & GL_MAP_READ_BIT)
@@ -201,10 +204,11 @@ st_MapTextureImage(struct gl_context *ctx,
    if (mode & GL_MAP_INVALIDATE_RANGE_BIT)
       pipeMode |= PIPE_TRANSFER_DISCARD_RANGE;
 
-   map = st_texture_image_map(st, stImage, pipeMode, x, y, slice, w, h, 1);
+   map = st_texture_image_map(st, stImage, pipeMode, x, y, slice, w, h, 1,
+                              &transfer);
    if (map) {
       *mapOut = map;
-      *rowStrideOut = stImage->transfer->stride;
+      *rowStrideOut = transfer->stride;
    }
    else {
       *mapOut = NULL;
@@ -221,7 +225,7 @@ st_UnmapTextureImage(struct gl_context *ctx,
 {
    struct st_context *st = st_context(ctx);
    struct st_texture_image *stImage  = st_texture_image(texImage);
-   st_texture_image_unmap(st, stImage);
+   st_texture_image_unmap(st, stImage, slice);
 }
 
 
@@ -409,7 +413,7 @@ guess_and_alloc_texture(struct st_context *st,
                                  ptWidth,
                                  ptHeight,
                                  ptDepth,
-                                 ptLayers,
+                                 ptLayers, 0,
                                  bindings);
 
    stObj->lastLevel = lastLevel;
@@ -454,7 +458,7 @@ st_AllocTextureImageBuffer(struct gl_context *ctx,
    /* The parent texture object does not have space for this image */
 
    pipe_resource_reference(&stObj->pt, NULL);
-   pipe_sampler_view_release(st->pipe, &stObj->sampler_view);
+   st_texture_release_all_sampler_views(stObj);
 
    if (!guess_and_alloc_texture(st, stObj, stImage)) {
       /* Probably out of memory.
@@ -497,7 +501,7 @@ st_AllocTextureImageBuffer(struct gl_context *ctx,
                                       ptWidth,
                                       ptHeight,
                                       ptDepth,
-                                      ptLayers,
+                                      ptLayers, 0,
                                       bindings);
       return stImage->pt != NULL;
    }
@@ -519,7 +523,7 @@ prep_teximage(struct gl_context *ctx, struct gl_texture_image *texImage,
    if (stObj->surface_based) {
       const GLenum target = texObj->Target;
       const GLuint level = texImage->Level;
-      gl_format texFormat;
+      mesa_format texFormat;
 
       _mesa_clear_texture_object(ctx, texObj);
       pipe_resource_reference(&stObj->pt, NULL);
@@ -539,14 +543,278 @@ prep_teximage(struct gl_context *ctx, struct gl_texture_image *texImage,
 }
 
 
+/**
+ * Return a writemask for the gallium blit. The parameters can be base
+ * formats or "format" from glDrawPixels/glTexImage/glGetTexImage.
+ */
+unsigned
+st_get_blit_mask(GLenum srcFormat, GLenum dstFormat)
+{
+   switch (dstFormat) {
+   case GL_DEPTH_STENCIL:
+      switch (srcFormat) {
+      case GL_DEPTH_STENCIL:
+         return PIPE_MASK_ZS;
+      case GL_DEPTH_COMPONENT:
+         return PIPE_MASK_Z;
+      case GL_STENCIL_INDEX:
+         return PIPE_MASK_S;
+      default:
+         assert(0);
+         return 0;
+      }
+
+   case GL_DEPTH_COMPONENT:
+      switch (srcFormat) {
+      case GL_DEPTH_STENCIL:
+      case GL_DEPTH_COMPONENT:
+         return PIPE_MASK_Z;
+      default:
+         assert(0);
+         return 0;
+      }
+
+   case GL_STENCIL_INDEX:
+      switch (srcFormat) {
+      case GL_STENCIL_INDEX:
+         return PIPE_MASK_S;
+      default:
+         assert(0);
+         return 0;
+      }
+
+   default:
+      return PIPE_MASK_RGBA;
+   }
+}
+
+
+static void
+st_TexSubImage(struct gl_context *ctx, GLuint dims,
+               struct gl_texture_image *texImage,
+               GLint xoffset, GLint yoffset, GLint zoffset,
+               GLint width, GLint height, GLint depth,
+               GLenum format, GLenum type, const void *pixels,
+               const struct gl_pixelstore_attrib *unpack)
+{
+   struct st_context *st = st_context(ctx);
+   struct st_texture_image *stImage = st_texture_image(texImage);
+   struct st_texture_object *stObj = st_texture_object(texImage->TexObject);
+   struct pipe_context *pipe = st->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_resource *dst = stImage->pt;
+   struct pipe_resource *src = NULL;
+   struct pipe_resource src_templ;
+   struct pipe_transfer *transfer;
+   struct pipe_blit_info blit;
+   enum pipe_format src_format, dst_format;
+   mesa_format mesa_src_format;
+   GLenum gl_target = texImage->TexObject->Target;
+   unsigned bind;
+   GLubyte *map;
+
+   if (!st->prefer_blit_based_texture_transfer) {
+      goto fallback;
+   }
+
+   if (!dst) {
+      goto fallback;
+   }
+
+   /* XXX Fallback for depth-stencil formats due to an incomplete stencil
+    * blit implementation in some drivers. */
+   if (format == GL_DEPTH_STENCIL) {
+      goto fallback;
+   }
+
+   /* If the base internal format and the texture format don't match,
+    * we can't use blit-based TexSubImage. */
+   if (texImage->_BaseFormat !=
+       _mesa_get_format_base_format(texImage->TexFormat)) {
+      goto fallback;
+   }
+
+   /* See if the texture format already matches the format and type,
+    * in which case the memcpy-based fast path will likely be used and
+    * we don't have to blit. */
+   if (_mesa_format_matches_format_and_type(texImage->TexFormat, format,
+                                            type, unpack->SwapBytes)) {
+      goto fallback;
+   }
+
+   if (format == GL_DEPTH_COMPONENT || format == GL_DEPTH_STENCIL)
+      bind = PIPE_BIND_DEPTH_STENCIL;
+   else
+      bind = PIPE_BIND_RENDER_TARGET;
+
+   /* See if the destination format is supported.
+    * For luminance and intensity, only the red channel is stored there. */
+   dst_format = util_format_linear(dst->format);
+   dst_format = util_format_luminance_to_red(dst_format);
+   dst_format = util_format_intensity_to_red(dst_format);
+
+   if (!dst_format ||
+       !screen->is_format_supported(screen, dst_format, dst->target,
+                                    dst->nr_samples, bind)) {
+      goto fallback;
+   }
+
+   /* Choose the source format. */
+   src_format = st_choose_matching_format(screen, PIPE_BIND_SAMPLER_VIEW,
+                                          format, type, unpack->SwapBytes);
+   if (!src_format) {
+      goto fallback;
+   }
+
+   mesa_src_format = st_pipe_format_to_mesa_format(src_format);
+
+   /* There is no reason to do this if we cannot use memcpy for the temporary
+    * source texture at least. This also takes transfer ops into account,
+    * etc. */
+   if (!_mesa_texstore_can_use_memcpy(ctx,
+                             _mesa_get_format_base_format(mesa_src_format),
+                             mesa_src_format, format, type, unpack)) {
+      goto fallback;
+   }
+
+   /* TexSubImage only sets a single cubemap face. */
+   if (gl_target == GL_TEXTURE_CUBE_MAP) {
+      gl_target = GL_TEXTURE_2D;
+   }
+
+   /* Initialize the source texture description. */
+   memset(&src_templ, 0, sizeof(src_templ));
+   src_templ.target = gl_target_to_pipe(gl_target);
+   src_templ.format = src_format;
+   src_templ.bind = PIPE_BIND_SAMPLER_VIEW;
+   src_templ.usage = PIPE_USAGE_STAGING;
+
+   st_gl_texture_dims_to_pipe_dims(gl_target, width, height, depth,
+                                   &src_templ.width0, &src_templ.height0,
+                                   &src_templ.depth0, &src_templ.array_size);
+
+   /* Check for NPOT texture support. */
+   if (!screen->get_param(screen, PIPE_CAP_NPOT_TEXTURES) &&
+       (!util_is_power_of_two(src_templ.width0) ||
+        !util_is_power_of_two(src_templ.height0) ||
+        !util_is_power_of_two(src_templ.depth0))) {
+      goto fallback;
+   }
+
+   /* Create the source texture. */
+   src = screen->resource_create(screen, &src_templ);
+   if (!src) {
+      goto fallback;
+   }
+
+   /* Map source pixels. */
+   pixels = _mesa_validate_pbo_teximage(ctx, dims, width, height, depth,
+                                        format, type, pixels, unpack,
+                                        "glTexSubImage");
+   if (!pixels) {
+      /* This is a GL error. */
+      pipe_resource_reference(&src, NULL);
+      return;
+   }
+
+   /* From now on, we need the gallium representation of dimensions. */
+   if (gl_target == GL_TEXTURE_1D_ARRAY) {
+      depth = height;
+      height = 1;
+   }
+
+   map = pipe_transfer_map_3d(pipe, src, 0, PIPE_TRANSFER_WRITE, 0, 0, 0,
+                              width, height, depth, &transfer);
+   if (!map) {
+      _mesa_unmap_teximage_pbo(ctx, unpack);
+      pipe_resource_reference(&src, NULL);
+      goto fallback;
+   }
+
+   /* Upload pixels (just memcpy). */
+   {
+      const uint bytesPerRow = width * util_format_get_blocksize(src_format);
+      GLuint row, slice;
+
+      for (slice = 0; slice < (unsigned) depth; slice++) {
+         if (gl_target == GL_TEXTURE_1D_ARRAY) {
+            /* 1D array textures.
+             * We need to convert gallium coords to GL coords.
+             */
+            GLvoid *src = _mesa_image_address3d(unpack, pixels,
+                                                width, depth, format,
+                                                type, 0, slice, 0);
+            memcpy(map, src, bytesPerRow);
+         }
+         else {
+            ubyte *slice_map = map;
+
+            for (row = 0; row < (unsigned) height; row++) {
+               GLvoid *src = _mesa_image_address3d(unpack, pixels,
+                                                   width, height, format,
+                                                   type, slice, row, 0);
+               memcpy(slice_map, src, bytesPerRow);
+               slice_map += transfer->stride;
+            }
+         }
+         map += transfer->layer_stride;
+      }
+   }
+
+   pipe_transfer_unmap(pipe, transfer);
+   _mesa_unmap_teximage_pbo(ctx, unpack);
+
+   /* Blit. */
+   blit.src.resource = src;
+   blit.src.level = 0;
+   blit.src.format = src_format;
+   blit.dst.resource = dst;
+   blit.dst.level = stObj->pt != stImage->pt ? 0 : texImage->Level;
+   blit.dst.format = dst_format;
+   blit.src.box.x = blit.src.box.y = blit.src.box.z = 0;
+   blit.dst.box.x = xoffset;
+   blit.dst.box.y = yoffset;
+   blit.dst.box.z = zoffset + texImage->Face;
+   blit.src.box.width = blit.dst.box.width = width;
+   blit.src.box.height = blit.dst.box.height = height;
+   blit.src.box.depth = blit.dst.box.depth = depth;
+   blit.mask = st_get_blit_mask(format, texImage->_BaseFormat);
+   blit.filter = PIPE_TEX_FILTER_NEAREST;
+   blit.scissor_enable = FALSE;
+
+   st->pipe->blit(st->pipe, &blit);
+
+   pipe_resource_reference(&src, NULL);
+   return;
+
+fallback:
+   _mesa_store_texsubimage(ctx, dims, texImage, xoffset, yoffset, zoffset,
+                           width, height, depth, format, type, pixels,
+                           unpack);
+}
+
 static void
 st_TexImage(struct gl_context * ctx, GLuint dims,
             struct gl_texture_image *texImage,
             GLenum format, GLenum type, const void *pixels,
             const struct gl_pixelstore_attrib *unpack)
 {
+   assert(dims == 1 || dims == 2 || dims == 3);
+
    prep_teximage(ctx, texImage, format, type);
-   _mesa_store_teximage(ctx, dims, texImage, format, type, pixels, unpack);
+
+   if (texImage->Width == 0 || texImage->Height == 0 || texImage->Depth == 0)
+      return;
+
+   /* allocate storage for texture data */
+   if (!ctx->Driver.AllocTextureImageBuffer(ctx, texImage)) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage%uD", dims);
+      return;
+   }
+
+   st_TexSubImage(ctx, dims, texImage, 0, 0, 0,
+                  texImage->Width, texImage->Height, texImage->Depth,
+                  format, type, pixels, unpack);
 }
 
 
@@ -561,53 +829,157 @@ st_CompressedTexImage(struct gl_context *ctx, GLuint dims,
 
 
 
+
 /**
- * glGetTexImage() helper: decompress a compressed texture by rendering
- * a textured quad.  Store the results in the user's buffer.
+ * Called via ctx->Driver.GetTexImage()
+ *
+ * This uses a blit to copy the texture to a texture format which matches
+ * the format and type combo and then a fast read-back is done using memcpy.
+ * We can do arbitrary X/Y/Z/W/0/1 swizzling here as long as there is
+ * a format which matches the swizzling.
+ *
+ * If such a format isn't available, it falls back to _mesa_get_teximage.
+ *
+ * NOTE: Drivers usually do a blit to convert between tiled and linear
+ *       texture layouts during texture uploads/downloads, so the blit
+ *       we do here should be free in such cases.
  */
 static void
-decompress_with_blit(struct gl_context * ctx,
-                     GLenum format, GLenum type, GLvoid *pixels,
-                     struct gl_texture_image *texImage)
+st_GetTexImage(struct gl_context * ctx,
+               GLenum format, GLenum type, GLvoid * pixels,
+               struct gl_texture_image *texImage)
 {
    struct st_context *st = st_context(ctx);
    struct pipe_context *pipe = st->pipe;
    struct pipe_screen *screen = pipe->screen;
-   const GLuint width = texImage->Width;
-   const GLuint height = texImage->Height;
-   const GLuint depth = texImage->Depth;
+   GLuint width = texImage->Width;
+   GLuint height = texImage->Height;
+   GLuint depth = texImage->Depth;
+   struct st_texture_image *stImage = st_texture_image(texImage);
    struct pipe_resource *src = st_texture_object(texImage->TexObject)->pt;
-   struct pipe_resource *dst;
+   struct pipe_resource *dst = NULL;
    struct pipe_resource dst_templ;
-   enum pipe_format pipe_format;
-   gl_format mesa_format;
+   enum pipe_format dst_format, src_format;
+   mesa_format mesa_format;
    GLenum gl_target = texImage->TexObject->Target;
    enum pipe_texture_target pipe_target;
    struct pipe_blit_info blit;
-   unsigned bind = (PIPE_BIND_RENDER_TARGET | PIPE_BIND_TRANSFER_READ);
+   unsigned bind = PIPE_BIND_TRANSFER_READ;
    struct pipe_transfer *tex_xfer;
-   ubyte *map;
+   ubyte *map = NULL;
+   boolean done = FALSE;
+
+   if (!st->prefer_blit_based_texture_transfer &&
+       !_mesa_is_format_compressed(texImage->TexFormat)) {
+      /* Try to avoid the fallback if we're doing texture decompression here */
+      goto fallback;
+   }
+
+   if (!stImage->pt || !src) {
+      goto fallback;
+   }
+
+   /* XXX Fallback to _mesa_get_teximage for depth-stencil formats
+    * due to an incomplete stencil blit implementation in some drivers. */
+   if (format == GL_DEPTH_STENCIL) {
+      goto fallback;
+   }
+
+   /* If the base internal format and the texture format don't match, we have
+    * to fall back to _mesa_get_teximage. */
+   if (texImage->_BaseFormat !=
+       _mesa_get_format_base_format(texImage->TexFormat)) {
+      goto fallback;
+   }
+
+   /* See if the texture format already matches the format and type,
+    * in which case the memcpy-based fast path will be used. */
+   if (_mesa_format_matches_format_and_type(texImage->TexFormat, format,
+                                            type, ctx->Pack.SwapBytes)) {
+      goto fallback;
+   }
+
+   /* Convert the source format to what is expected by GetTexImage
+    * and see if it's supported.
+    *
+    * This only applies to glGetTexImage:
+    * - Luminance must be returned as (L,0,0,1).
+    * - Luminance alpha must be returned as (L,0,0,A).
+    * - Intensity must be returned as (I,0,0,1)
+    */
+   src_format = util_format_linear(src->format);
+   src_format = util_format_luminance_to_red(src_format);
+   src_format = util_format_intensity_to_red(src_format);
+
+   if (!src_format ||
+       !screen->is_format_supported(screen, src_format, src->target,
+                                    src->nr_samples,
+                                    PIPE_BIND_SAMPLER_VIEW)) {
+      goto fallback;
+   }
+
+   if (format == GL_DEPTH_COMPONENT || format == GL_DEPTH_STENCIL)
+      bind |= PIPE_BIND_DEPTH_STENCIL;
+   else
+      bind |= PIPE_BIND_RENDER_TARGET;
 
    /* GetTexImage only returns a single face for cubemaps. */
    if (gl_target == GL_TEXTURE_CUBE_MAP) {
       gl_target = GL_TEXTURE_2D;
    }
-
    pipe_target = gl_target_to_pipe(gl_target);
 
-   /* Find the best match for the format+type combo. */
-   pipe_format = st_choose_format(pipe->screen, GL_RGBA8, format, type,
-                                  pipe_target, 0, bind);
-   if (pipe_format == PIPE_FORMAT_NONE) {
-      /* unable to get an rgba format!?! */
-      _mesa_problem(ctx, "%s: cannot find a supported format", __func__);
-      return;
+   /* Choose the destination format by finding the best match
+    * for the format+type combo. */
+   dst_format = st_choose_matching_format(screen, bind, format, type,
+                                         ctx->Pack.SwapBytes);
+
+   if (dst_format == PIPE_FORMAT_NONE) {
+      GLenum dst_glformat;
+
+      /* Fall back to _mesa_get_teximage except for compressed formats,
+       * where decompression with a blit is always preferred. */
+      if (!util_format_is_compressed(src->format)) {
+         goto fallback;
+      }
+
+      /* Set the appropriate format for the decompressed texture.
+       * Luminance and sRGB formats shouldn't appear here.*/
+      switch (src_format) {
+      case PIPE_FORMAT_DXT1_RGB:
+      case PIPE_FORMAT_DXT1_RGBA:
+      case PIPE_FORMAT_DXT3_RGBA:
+      case PIPE_FORMAT_DXT5_RGBA:
+      case PIPE_FORMAT_RGTC1_UNORM:
+      case PIPE_FORMAT_RGTC2_UNORM:
+      case PIPE_FORMAT_ETC1_RGB8:
+         dst_glformat = GL_RGBA8;
+         break;
+      case PIPE_FORMAT_RGTC1_SNORM:
+      case PIPE_FORMAT_RGTC2_SNORM:
+         if (!ctx->Extensions.EXT_texture_snorm)
+            goto fallback;
+         dst_glformat = GL_RGBA8_SNORM;
+         break;
+      /* TODO: for BPTC_*FLOAT, set RGBA32F and check for ARB_texture_float */
+      default:
+         assert(0);
+         goto fallback;
+      }
+
+      dst_format = st_choose_format(st, dst_glformat, format, type,
+                                    pipe_target, 0, bind, FALSE);
+
+      if (dst_format == PIPE_FORMAT_NONE) {
+         /* unable to get an rgba format!?! */
+         goto fallback;
+      }
    }
 
    /* create the destination texture */
    memset(&dst_templ, 0, sizeof(dst_templ));
    dst_templ.target = pipe_target;
-   dst_templ.format = pipe_format;
+   dst_templ.format = dst_format;
    dst_templ.bind = bind;
    dst_templ.usage = PIPE_USAGE_STAGING;
 
@@ -617,13 +989,18 @@ decompress_with_blit(struct gl_context * ctx,
 
    dst = screen->resource_create(screen, &dst_templ);
    if (!dst) {
-      _mesa_problem(ctx, "%s: cannot create a temporary texture", __func__);
-      return;
+      goto fallback;
+   }
+
+   /* From now on, we need the gallium representation of dimensions. */
+   if (gl_target == GL_TEXTURE_1D_ARRAY) {
+      depth = height;
+      height = 1;
    }
 
    blit.src.resource = src;
    blit.src.level = texImage->Level;
-   blit.src.format = util_format_linear(src->format);
+   blit.src.format = src_format;
    blit.dst.resource = dst;
    blit.dst.level = 0;
    blit.dst.format = dst->format;
@@ -634,7 +1011,7 @@ decompress_with_blit(struct gl_context * ctx,
    blit.src.box.width = blit.dst.box.width = width;
    blit.src.box.height = blit.dst.box.height = height;
    blit.src.box.depth = blit.dst.box.depth = depth;
-   blit.mask = PIPE_MASK_RGBA;
+   blit.mask = st_get_blit_mask(texImage->_BaseFormat, format);
    blit.filter = PIPE_TEX_FILTER_NEAREST;
    blit.scissor_enable = FALSE;
 
@@ -649,24 +1026,35 @@ decompress_with_blit(struct gl_context * ctx,
       goto end;
    }
 
-   mesa_format = st_pipe_format_to_mesa_format(pipe_format);
+   mesa_format = st_pipe_format_to_mesa_format(dst_format);
 
    /* copy/pack data into user buffer */
    if (_mesa_format_matches_format_and_type(mesa_format, format, type,
                                             ctx->Pack.SwapBytes)) {
       /* memcpy */
-      const uint bytesPerRow = width * util_format_get_blocksize(pipe_format);
+      const uint bytesPerRow = width * util_format_get_blocksize(dst_format);
       GLuint row, slice;
 
       for (slice = 0; slice < depth; slice++) {
-         ubyte *slice_map = map;
-
-         for (row = 0; row < height; row++) {
+         if (gl_target == GL_TEXTURE_1D_ARRAY) {
+            /* 1D array textures.
+             * We need to convert gallium coords to GL coords.
+             */
             GLvoid *dest = _mesa_image_address3d(&ctx->Pack, pixels,
-                                                 width, height, format,
-                                                 type, slice, row, 0);
-            memcpy(dest, slice_map, bytesPerRow);
-            slice_map += tex_xfer->stride;
+                                                 width, depth, format,
+                                                 type, 0, slice, 0);
+            memcpy(dest, map, bytesPerRow);
+         }
+         else {
+            ubyte *slice_map = map;
+
+            for (row = 0; row < height; row++) {
+               GLvoid *dest = _mesa_image_address3d(&ctx->Pack, pixels,
+                                                    width, height, format,
+                                                    type, slice, row, 0);
+               memcpy(dest, slice_map, bytesPerRow);
+               slice_map += tex_xfer->stride;
+            }
          }
          map += tex_xfer->layer_stride;
       }
@@ -674,37 +1062,54 @@ decompress_with_blit(struct gl_context * ctx,
    else {
       /* format translation via floats */
       GLuint row, slice;
-      enum pipe_format pformat = util_format_linear(dst->format);
       GLfloat *rgba;
 
+      assert(util_format_is_compressed(src->format));
+
       rgba = malloc(width * 4 * sizeof(GLfloat));
       if (!rgba) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glGetTexImage()");
          goto end;
       }
 
+      if (ST_DEBUG & DEBUG_FALLBACK)
+         debug_printf("%s: fallback format translation\n", __FUNCTION__);
+
       for (slice = 0; slice < depth; slice++) {
-         for (row = 0; row < height; row++) {
-            const GLbitfield transferOps = 0x0; /* bypassed for glGetTexImage() */
+         if (gl_target == GL_TEXTURE_1D_ARRAY) {
+            /* 1D array textures.
+             * We need to convert gallium coords to GL coords.
+             */
             GLvoid *dest = _mesa_image_address3d(&ctx->Pack, pixels,
-                                                 width, height, format,
-                                                 type, slice, row, 0);
-
-            if (ST_DEBUG & DEBUG_FALLBACK)
-               debug_printf("%s: fallback format translation\n", __FUNCTION__);
+                                                 width, depth, format,
+                                                 type, 0, slice, 0);
 
             /* get float[4] rgba row from surface */
-            pipe_get_tile_rgba_format(tex_xfer, map, 0, row, width, 1,
-                                      pformat, rgba);
+            pipe_get_tile_rgba_format(tex_xfer, map, 0, 0, width, 1,
+                                      dst_format, rgba);
 
             _mesa_pack_rgba_span_float(ctx, width, (GLfloat (*)[4]) rgba, format,
-                                       type, dest, &ctx->Pack, transferOps);
+                                       type, dest, &ctx->Pack, 0);
+         }
+         else {
+            for (row = 0; row < height; row++) {
+               GLvoid *dest = _mesa_image_address3d(&ctx->Pack, pixels,
+                                                    width, height, format,
+                                                    type, slice, row, 0);
+
+               /* get float[4] rgba row from surface */
+               pipe_get_tile_rgba_format(tex_xfer, map, 0, row, width, 1,
+                                         dst_format, rgba);
+
+               _mesa_pack_rgba_span_float(ctx, width, (GLfloat (*)[4]) rgba, format,
+                                          type, dest, &ctx->Pack, 0);
+            }
          }
          map += tex_xfer->layer_stride;
       }
 
       free(rgba);
    }
+   done = TRUE;
 
 end:
    if (map)
@@ -712,29 +1117,9 @@ end:
 
    _mesa_unmap_pbo_dest(ctx, &ctx->Pack);
    pipe_resource_reference(&dst, NULL);
-}
-
-
-
-/**
- * Called via ctx->Driver.GetTexImage()
- */
-static void
-st_GetTexImage(struct gl_context * ctx,
-               GLenum format, GLenum type, GLvoid * pixels,
-               struct gl_texture_image *texImage)
-{
-   struct st_texture_image *stImage = st_texture_image(texImage);
 
-   if (stImage->pt && util_format_is_s3tc(stImage->pt->format)) {
-      /* Need to decompress the texture.
-       * We'll do this by rendering a textured quad (which is hopefully
-       * faster than using the fallback code in texcompress.c).
-       * Note that we only expect RGBA formats (no Z/depth formats).
-       */
-      decompress_with_blit(ctx, format, type, pixels, texImage);
-   }
-   else {
+fallback:
+   if (!done) {
       _mesa_get_teximage(ctx, format, type, pixels, texImage);
    }
 }
@@ -752,7 +1137,7 @@ fallback_copy_texsubimage(struct gl_context *ctx,
                           struct st_renderbuffer *strb,
                           struct st_texture_image *stImage,
                           GLenum baseFormat,
-                          GLint destX, GLint destY, GLint destZ,
+                          GLint destX, GLint destY, GLint slice,
                           GLint srcX, GLint srcY,
                           GLsizei width, GLsizei height)
 {
@@ -765,6 +1150,7 @@ fallback_copy_texsubimage(struct gl_context *ctx,
    unsigned dst_width = width;
    unsigned dst_height = height;
    unsigned dst_depth = 1;
+   struct pipe_transfer *transfer;
 
    if (ST_DEBUG & DEBUG_FALLBACK)
       debug_printf("%s: fallback processing\n", __FUNCTION__);
@@ -773,18 +1159,10 @@ fallback_copy_texsubimage(struct gl_context *ctx,
       srcY = strb->Base.Height - srcY - height;
    }
 
-   if (stImage->pt->target == PIPE_TEXTURE_1D_ARRAY) {
-      /* Move y/height to z/depth for 1D array textures.  */
-      destZ = destY;
-      destY = 0;
-      dst_depth = dst_height;
-      dst_height = 1;
-   }
-
    map = pipe_transfer_map(pipe,
                            strb->texture,
-                           strb->rtt_level,
-                           strb->rtt_face + strb->rtt_slice,
+                           strb->surface->u.tex.level,
+                           strb->surface->u.tex.first_layer,
                            PIPE_TRANSFER_READ,
                            srcX, srcY,
                            width, height, &src_trans);
@@ -797,8 +1175,9 @@ fallback_copy_texsubimage(struct gl_context *ctx,
       transfer_usage = PIPE_TRANSFER_WRITE;
 
    texDest = st_texture_image_map(st, stImage, transfer_usage,
-                                  destX, destY, destZ,
-                                  dst_width, dst_height, dst_depth);
+                                  destX, destY, slice,
+                                  dst_width, dst_height, dst_depth,
+                                  &transfer);
 
    if (baseFormat == GL_DEPTH_COMPONENT ||
        baseFormat == GL_DEPTH_STENCIL) {
@@ -828,13 +1207,11 @@ fallback_copy_texsubimage(struct gl_context *ctx,
             }
 
             if (stImage->pt->target == PIPE_TEXTURE_1D_ARRAY) {
-               pipe_put_tile_z(stImage->transfer,
-                               texDest + row*stImage->transfer->layer_stride,
+               pipe_put_tile_z(transfer, texDest + row*transfer->layer_stride,
                                0, 0, width, 1, data);
             }
             else {
-               pipe_put_tile_z(stImage->transfer, texDest, 0, row, width, 1,
-                               data);
+               pipe_put_tile_z(transfer, texDest, 0, row, width, 1, data);
             }
          }
       }
@@ -860,10 +1237,10 @@ fallback_copy_texsubimage(struct gl_context *ctx,
          }
 
          if (stImage->pt->target == PIPE_TEXTURE_1D_ARRAY) {
-            dstRowStride = stImage->transfer->layer_stride;
+            dstRowStride = transfer->layer_stride;
          }
          else {
-            dstRowStride = stImage->transfer->stride;
+            dstRowStride = transfer->stride;
          }
 
          /* get float/RGBA image from framebuffer */
@@ -896,115 +1273,11 @@ fallback_copy_texsubimage(struct gl_context *ctx,
       free(tempSrc);
    }
 
-   st_texture_image_unmap(st, stImage);
+   st_texture_image_unmap(st, stImage, slice);
    pipe->transfer_unmap(pipe, src_trans);
 }
 
 
-
-/**
- * If the format of the src renderbuffer and the format of the dest
- * texture are compatible (in terms of blitting), return a TGSI writemask
- * to be used during the blit.
- * If the src/dest are incompatible, return 0.
- */
-static unsigned
-compatible_src_dst_formats(struct gl_context *ctx,
-                           const struct gl_renderbuffer *src,
-                           const struct gl_texture_image *dst)
-{
-   /* Get logical base formats for the src and dest.
-    * That is, use the user-requested formats and not the actual, device-
-    * chosen formats.
-    * For example, the user may have requested an A8 texture but the
-    * driver may actually be using an RGBA texture format.  When we
-    * copy/blit to that texture, we only want to copy the Alpha channel
-    * and not the RGB channels.
-    *
-    * Similarly, when the src FBO was created an RGB format may have been
-    * requested but the driver actually chose an RGBA format.  In that case,
-    * we don't want to copy the undefined Alpha channel to the dest texture
-    * (it should be 1.0).
-    */
-   const GLenum srcFormat = _mesa_base_fbo_format(ctx, src->InternalFormat);
-   const GLenum dstFormat = _mesa_base_tex_format(ctx, dst->InternalFormat);
-
-   /**
-    * XXX when we have red-only and red/green renderbuffers we'll need
-    * to add more cases here (or implement a general-purpose routine that
-    * queries the existance of the R,G,B,A channels in the src and dest).
-    */
-   if (srcFormat == dstFormat) {
-      /* This is the same as matching_base_formats, which should
-       * always pass, as it did previously.
-       */
-      return TGSI_WRITEMASK_XYZW;
-   }
-   else if (srcFormat == GL_RGB && dstFormat == GL_RGBA) {
-      /* Make sure that A in the dest is 1.  The actual src format
-       * may be RGBA and have undefined A values.
-       */
-      return TGSI_WRITEMASK_XYZ;
-   }
-   else if (srcFormat == GL_RGBA && dstFormat == GL_RGB) {
-      /* Make sure that A in the dest is 1.  The actual dst format
-       * may be RGBA and will need A=1 to provide proper alpha values
-       * when sampled later.
-       */
-      return TGSI_WRITEMASK_XYZ;
-   }
-   else {
-      if (ST_DEBUG & DEBUG_FALLBACK)
-         debug_printf("%s failed for src %s, dst %s\n",
-                      __FUNCTION__, 
-                      _mesa_lookup_enum_by_nr(srcFormat),
-                      _mesa_lookup_enum_by_nr(dstFormat));
-
-      /* Otherwise fail.
-       */
-      return 0;
-   }
-}
-
-
-/**
- * Do pipe->blit. Return FALSE if the blitting is unsupported
- * for the given formats.
- */
-static GLboolean
-st_pipe_blit(struct pipe_context *pipe, struct pipe_blit_info *blit)
-{
-   struct pipe_screen *screen = pipe->screen;
-   unsigned dst_usage;
-
-   if (util_format_is_depth_or_stencil(blit->dst.format)) {
-      dst_usage = PIPE_BIND_DEPTH_STENCIL;
-   }
-   else {
-      dst_usage = PIPE_BIND_RENDER_TARGET;
-   }
-
-   /* try resource_copy_region in case the format is not supported
-    * for rendering */
-   if (util_try_blit_via_copy_region(pipe, blit)) {
-      return GL_TRUE; /* done */
-   }
-
-   /* check the format support */
-   if (!screen->is_format_supported(screen, blit->src.format,
-                                    PIPE_TEXTURE_2D, 0,
-                                    PIPE_BIND_SAMPLER_VIEW) ||
-       !screen->is_format_supported(screen, blit->dst.format,
-                                    PIPE_TEXTURE_2D, 0,
-                                    dst_usage)) {
-      return GL_FALSE;
-   }
-
-   pipe->blit(pipe, blit);
-   return GL_TRUE;
-}
-
-
 /**
  * Do a CopyTex[Sub]Image1/2/3D() using a hardware (blit) path if possible.
  * Note that the region to copy has already been clipped so we know we
@@ -1015,212 +1288,103 @@ st_pipe_blit(struct pipe_context *pipe, struct pipe_blit_info *blit)
 static void
 st_CopyTexSubImage(struct gl_context *ctx, GLuint dims,
                    struct gl_texture_image *texImage,
-                   GLint destX, GLint destY, GLint destZ,
+                   GLint destX, GLint destY, GLint slice,
                    struct gl_renderbuffer *rb,
                    GLint srcX, GLint srcY, GLsizei width, GLsizei height)
 {
    struct st_texture_image *stImage = st_texture_image(texImage);
-   const GLenum texBaseFormat = texImage->_BaseFormat;
+   struct st_texture_object *stObj = st_texture_object(texImage->TexObject);
    struct st_renderbuffer *strb = st_renderbuffer(rb);
    struct st_context *st = st_context(ctx);
    struct pipe_context *pipe = st->pipe;
    struct pipe_screen *screen = pipe->screen;
-   enum pipe_format dest_format, src_format;
-   GLuint color_writemask;
-   struct pipe_surface *dest_surface = NULL;
+   struct pipe_blit_info blit;
+   enum pipe_format dst_format;
    GLboolean do_flip = (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP);
-   struct pipe_surface surf_tmpl;
-   unsigned dst_usage;
-   unsigned blit_mask;
-   GLint srcY0, srcY1, yStep;
-
-   /* make sure finalize_textures has been called? 
-    */
-   if (0) st_validate_state(st);
+   unsigned bind;
+   GLint srcY0, srcY1;
 
    if (!strb || !strb->surface || !stImage->pt) {
       debug_printf("%s: null strb or stImage\n", __FUNCTION__);
       return;
    }
 
-   assert(strb);
-   assert(strb->surface);
-   assert(stImage->pt);
-
-   src_format = strb->surface->format;
-   dest_format = stImage->pt->format;
-
-   if (do_flip) {
-      srcY1 = strb->Base.Height - srcY - height;
-      srcY0 = srcY1 + height;
-      yStep = -1;
-   }
-   else {
-      srcY0 = srcY;
-      srcY1 = srcY0 + height;
-      yStep = 1;
-   }
-
-   if (ctx->_ImageTransferState) {
+   if (_mesa_texstore_needs_transfer_ops(ctx, texImage->_BaseFormat,
+                                         texImage->TexFormat)) {
       goto fallback;
    }
 
-   /* Compressed and subsampled textures aren't supported for blitting. */
-   if (!util_format_is_plain(dest_format)) {
+   /* The base internal format must match the mesa format, so make sure
+    * e.g. an RGB internal format is really allocated as RGB and not as RGBA.
+    */
+   if (texImage->_BaseFormat !=
+       _mesa_get_format_base_format(texImage->TexFormat) ||
+       rb->_BaseFormat != _mesa_get_format_base_format(rb->Format)) {
       goto fallback;
    }
 
-   /* Set the blit writemask. */
-   switch (texBaseFormat) {
-   case GL_DEPTH_STENCIL:
-      switch (strb->Base._BaseFormat) {
-      case GL_DEPTH_STENCIL:
-         blit_mask = PIPE_MASK_ZS;
-         break;
-      case GL_DEPTH_COMPONENT:
-         blit_mask = PIPE_MASK_Z;
-         break;
-      case GL_STENCIL_INDEX:
-         blit_mask = PIPE_MASK_S;
-         break;
-      default:
-         assert(0);
-         return;
-      }
-      dst_usage = PIPE_BIND_DEPTH_STENCIL;
-      break;
-
-   case GL_DEPTH_COMPONENT:
-      blit_mask = PIPE_MASK_Z;
-      dst_usage = PIPE_BIND_DEPTH_STENCIL;
-      break;
+   /* Choose the destination format to match the TexImage behavior. */
+   dst_format = util_format_linear(stImage->pt->format);
+   dst_format = util_format_luminance_to_red(dst_format);
+   dst_format = util_format_intensity_to_red(dst_format);
 
-   default:
-      /* Colorbuffers.
-       *
-       * Determine if the src framebuffer and dest texture have the same
-       * base format.  We need this to detect a case such as the framebuffer
-       * being GL_RGBA but the texture being GL_RGB.  If the actual hardware
-       * texture format stores RGBA we need to set A=1 (overriding the
-       * framebuffer's alpha values).
-       *
-       * XXX util_blit_pixels doesn't support MSAA resolve, so always use
-       *     pipe->blit
-       */
-      if (texBaseFormat == strb->Base._BaseFormat ||
-          strb->texture->nr_samples > 1) {
-         blit_mask = PIPE_MASK_RGBA;
-      }
-      else {
-         blit_mask = 0;
-      }
-      dst_usage = PIPE_BIND_RENDER_TARGET;
+   /* See if the destination format is supported. */
+   if (texImage->_BaseFormat == GL_DEPTH_STENCIL ||
+       texImage->_BaseFormat == GL_DEPTH_COMPONENT) {
+      bind = PIPE_BIND_DEPTH_STENCIL;
    }
-
-   /* Blit the texture.
-    * This supports flipping, format conversions, and downsampling.
-    */
-   if (blit_mask) {
-      /* If stImage->pt is an independent image (not a pointer into a full
-       * mipmap) stImage->pt.last_level will be zero and we need to use that
-       * as the dest level.
-       */
-      unsigned dstLevel = MIN2(stImage->base.Level, stImage->pt->last_level);
-      struct pipe_blit_info blit;
-
-      memset(&blit, 0, sizeof(blit));
-      blit.src.resource = strb->texture;
-      blit.src.format = src_format;
-      blit.src.level = strb->surface->u.tex.level;
-      blit.src.box.x = srcX;
-      blit.src.box.y = srcY0;
-      blit.src.box.z = strb->surface->u.tex.first_layer;
-      blit.src.box.width = width;
-      blit.src.box.height = srcY1 - srcY0;
-      blit.src.box.depth = 1;
-      blit.dst.resource = stImage->pt;
-      blit.dst.format = dest_format;
-      blit.dst.level = dstLevel;
-      blit.dst.box.x = destX;
-      blit.dst.box.y = destY;
-      blit.dst.box.z = stImage->base.Face + destZ;
-      blit.dst.box.width = width;
-      blit.dst.box.height = height;
-      blit.dst.box.depth = 1;
-      blit.mask = blit_mask;
-      blit.filter = PIPE_TEX_FILTER_NEAREST;
-
-      /* 1D array textures need special treatment.
-       * Blit rows from the source to layers in the destination. */
-      if (texImage->TexObject->Target == GL_TEXTURE_1D_ARRAY) {
-         int y, layer;
-
-         for (y = srcY0, layer = 0; layer < height; y += yStep, layer++) {
-            blit.src.box.y = y;
-            blit.src.box.height = 1;
-            blit.dst.box.y = 0;
-            blit.dst.box.height = 1;
-            blit.dst.box.z = destY + layer;
-
-            if (!st_pipe_blit(pipe, &blit)) {
-               goto fallback;
-            }
-         }
-      }
-      else {
-         /* All the other texture targets. */
-         if (!st_pipe_blit(pipe, &blit)) {
-            goto fallback;
-         }
-      }
-      return;
+   else {
+      bind = PIPE_BIND_RENDER_TARGET;
    }
 
-   /* try u_blit */
-   if (texImage->TexObject->Target == GL_TEXTURE_1D_ARRAY) {
-      /* u_blit cannot copy 1D array textures as required by CopyTexSubImage */
+   if (!dst_format ||
+       !screen->is_format_supported(screen, dst_format, stImage->pt->target,
+                                    stImage->pt->nr_samples, bind)) {
       goto fallback;
    }
 
-   color_writemask = compatible_src_dst_formats(ctx, &strb->Base, texImage);
-
-   if (!color_writemask ||
-       !screen->is_format_supported(screen, src_format,
-                                    PIPE_TEXTURE_2D, 0,
-                                    PIPE_BIND_SAMPLER_VIEW) ||
-       !screen->is_format_supported(screen, dest_format,
-                                    PIPE_TEXTURE_2D, 0,
-                                    dst_usage)) {
-      goto fallback;
+   /* Y flipping for the main framebuffer. */
+   if (do_flip) {
+      srcY1 = strb->Base.Height - srcY - height;
+      srcY0 = srcY1 + height;
+   }
+   else {
+      srcY0 = srcY;
+      srcY1 = srcY0 + height;
    }
 
-   memset(&surf_tmpl, 0, sizeof(surf_tmpl));
-   surf_tmpl.format = util_format_linear(stImage->pt->format);
-   surf_tmpl.u.tex.level = stImage->base.Level;
-   surf_tmpl.u.tex.first_layer = stImage->base.Face + destZ;
-   surf_tmpl.u.tex.last_layer = stImage->base.Face + destZ;
-
-   dest_surface = pipe->create_surface(pipe, stImage->pt,
-                                       &surf_tmpl);
-   util_blit_pixels(st->blit,
-                    strb->texture,
-                    strb->surface->u.tex.level,
-                    srcX, srcY0,
-                    srcX + width, srcY1,
-                    strb->surface->u.tex.first_layer,
-                    dest_surface,
-                    destX, destY,
-                    destX + width, destY + height,
-                    0.0, PIPE_TEX_MIPFILTER_NEAREST,
-                    color_writemask, 0);
-   pipe_surface_reference(&dest_surface, NULL);
+   /* Blit the texture.
+    * This supports flipping, format conversions, and downsampling.
+    */
+   memset(&blit, 0, sizeof(blit));
+   blit.src.resource = strb->texture;
+   blit.src.format = util_format_linear(strb->surface->format);
+   blit.src.level = strb->surface->u.tex.level;
+   blit.src.box.x = srcX;
+   blit.src.box.y = srcY0;
+   blit.src.box.z = strb->surface->u.tex.first_layer;
+   blit.src.box.width = width;
+   blit.src.box.height = srcY1 - srcY0;
+   blit.src.box.depth = 1;
+   blit.dst.resource = stImage->pt;
+   blit.dst.format = dst_format;
+   blit.dst.level = stObj->pt != stImage->pt ? 0 : texImage->Level;
+   blit.dst.box.x = destX;
+   blit.dst.box.y = destY;
+   blit.dst.box.z = stImage->base.Face + slice;
+   blit.dst.box.width = width;
+   blit.dst.box.height = height;
+   blit.dst.box.depth = 1;
+   blit.mask = st_get_blit_mask(rb->_BaseFormat, texImage->_BaseFormat);
+   blit.filter = PIPE_TEX_FILTER_NEAREST;
+   pipe->blit(pipe, &blit);
    return;
 
 fallback:
    /* software fallback */
    fallback_copy_texsubimage(ctx,
-                             strb, stImage, texBaseFormat,
-                             destX, destY, destZ,
+                             strb, stImage, texImage->_BaseFormat,
+                             destX, destY, slice,
                              srcX, srcY, width, height);
 }
 
@@ -1306,7 +1470,7 @@ st_finalize_texture(struct gl_context *ctx,
    GLuint face;
    struct st_texture_image *firstImage;
    enum pipe_format firstImageFormat;
-   GLuint ptWidth, ptHeight, ptDepth, ptLayers;
+   GLuint ptWidth, ptHeight, ptDepth, ptLayers, ptNumSamples;
 
    if (_mesa_is_texture_complete(tObj, &tObj->Sampler)) {
       /* The texture is complete and we know exactly how many mipmap levels
@@ -1325,9 +1489,15 @@ st_finalize_texture(struct gl_context *ctx,
    if (tObj->Target == GL_TEXTURE_BUFFER) {
       struct st_buffer_object *st_obj = st_buffer_object(tObj->BufferObject);
 
+      if (!st_obj) {
+         pipe_resource_reference(&stObj->pt, NULL);
+         st_texture_release_all_sampler_views(stObj);
+         return GL_TRUE;
+      }
+
       if (st_obj->buffer != stObj->pt) {
          pipe_resource_reference(&stObj->pt, st_obj->buffer);
-         pipe_sampler_view_release(st->pipe, &stObj->sampler_view);
+         st_texture_release_all_sampler_views(stObj);
          stObj->width0 = stObj->pt->width0 / _mesa_get_format_bytes(tObj->_BufferObjectFormat);
          stObj->height0 = 1;
          stObj->depth0 = 1;
@@ -1348,7 +1518,12 @@ st_finalize_texture(struct gl_context *ctx,
        firstImage->pt != stObj->pt &&
        (!stObj->pt || firstImage->pt->last_level >= stObj->pt->last_level)) {
       pipe_resource_reference(&stObj->pt, firstImage->pt);
-      pipe_sampler_view_release(st->pipe, &stObj->sampler_view);
+      st_texture_release_all_sampler_views(stObj);
+   }
+
+   /* If this texture comes from a window system, there is nothing else to do. */
+   if (stObj->surface_based) {
+      return GL_TRUE;
    }
 
    /* Find gallium format for the Mesa texture */
@@ -1370,6 +1545,7 @@ st_finalize_texture(struct gl_context *ctx,
       /* convert GL dims to Gallium dims */
       st_gl_texture_dims_to_pipe_dims(stObj->base.Target, width, height, depth,
                                       &ptWidth, &ptHeight, &ptDepth, &ptLayers);
+      ptNumSamples = firstImage->base.NumSamples;
    }
 
    /* If we already have a gallium texture, check that it matches the texture
@@ -1377,18 +1553,19 @@ st_finalize_texture(struct gl_context *ctx,
     */
    if (stObj->pt) {
       if (stObj->pt->target != gl_target_to_pipe(stObj->base.Target) ||
-          !st_sampler_compat_formats(stObj->pt->format, firstImageFormat) ||
+          stObj->pt->format != firstImageFormat ||
           stObj->pt->last_level < stObj->lastLevel ||
           stObj->pt->width0 != ptWidth ||
           stObj->pt->height0 != ptHeight ||
           stObj->pt->depth0 != ptDepth ||
+          stObj->pt->nr_samples != ptNumSamples ||
           stObj->pt->array_size != ptLayers)
       {
          /* The gallium texture does not match the Mesa texture so delete the
           * gallium texture now.  We'll make a new one below.
           */
          pipe_resource_reference(&stObj->pt, NULL);
-         pipe_sampler_view_release(st->pipe, &stObj->sampler_view);
+         st_texture_release_all_sampler_views(stObj);
          st->dirty.st |= ST_NEW_FRAMEBUFFER;
       }
    }
@@ -1405,7 +1582,7 @@ st_finalize_texture(struct gl_context *ctx,
                                     ptWidth,
                                     ptHeight,
                                     ptDepth,
-                                    ptLayers,
+                                    ptLayers, ptNumSamples,
                                     bindings);
 
       if (!stObj->pt) {
@@ -1451,11 +1628,14 @@ st_AllocTextureStorage(struct gl_context *ctx,
                        GLsizei height, GLsizei depth)
 {
    const GLuint numFaces = _mesa_num_tex_faces(texObj->Target);
+   struct gl_texture_image *texImage = texObj->Image[0][0];
    struct st_context *st = st_context(ctx);
    struct st_texture_object *stObj = st_texture_object(texObj);
+   struct pipe_screen *screen = st->pipe->screen;
    GLuint ptWidth, ptHeight, ptDepth, ptLayers, bindings;
    enum pipe_format fmt;
    GLint level;
+   GLuint num_samples = texImage->NumSamples;
 
    assert(levels > 0);
 
@@ -1465,10 +1645,30 @@ st_AllocTextureStorage(struct gl_context *ctx,
    stObj->depth0 = depth;
    stObj->lastLevel = levels - 1;
 
-   fmt = st_mesa_format_to_pipe_format(texObj->Image[0][0]->TexFormat);
+   fmt = st_mesa_format_to_pipe_format(texImage->TexFormat);
 
    bindings = default_bindings(st, fmt);
 
+   /* Raise the sample count if the requested one is unsupported. */
+   if (num_samples > 1) {
+      boolean found = FALSE;
+
+      for (; num_samples <= ctx->Const.MaxSamples; num_samples++) {
+         if (screen->is_format_supported(screen, fmt, PIPE_TEXTURE_2D,
+                                         num_samples,
+                                         PIPE_BIND_SAMPLER_VIEW)) {
+            /* Update the sample count in gl_texture_image as well. */
+            texImage->NumSamples = num_samples;
+            found = TRUE;
+            break;
+         }
+      }
+
+      if (!found) {
+         return GL_FALSE;
+      }
+   }
+
    st_gl_texture_dims_to_pipe_dims(texObj->Target,
                                    width, height, depth,
                                    &ptWidth, &ptHeight, &ptDepth, &ptLayers);
@@ -1476,11 +1676,11 @@ st_AllocTextureStorage(struct gl_context *ctx,
    stObj->pt = st_texture_create(st,
                                  gl_target_to_pipe(texObj->Target),
                                  fmt,
-                                 levels,
+                                 levels - 1,
                                  ptWidth,
                                  ptHeight,
                                  ptDepth,
-                                 ptLayers,
+                                 ptLayers, num_samples,
                                  bindings);
    if (!stObj->pt)
       return GL_FALSE;
@@ -1501,7 +1701,7 @@ st_AllocTextureStorage(struct gl_context *ctx,
 
 static GLboolean
 st_TestProxyTexImage(struct gl_context *ctx, GLenum target,
-                     GLint level, gl_format format,
+                     GLint level, mesa_format format,
                      GLint width, GLint height,
                      GLint depth, GLint border)
 {
@@ -1557,7 +1757,7 @@ st_init_texture_functions(struct dd_function_table *functions)
    functions->ChooseTextureFormat = st_ChooseTextureFormat;
    functions->QuerySamplesForFormat = st_QuerySamplesForFormat;
    functions->TexImage = st_TexImage;
-   functions->TexSubImage = _mesa_store_texsubimage;
+   functions->TexSubImage = st_TexSubImage;
    functions->CompressedTexSubImage = _mesa_store_compressed_texsubimage;
    functions->CopyTexSubImage = st_CopyTexSubImage;
    functions->GenerateMipmap = st_generate_mipmap;