Merge commit 'origin/master' into gallium-map-range

66 files changed, 2177 insertions, 1183 deletions

diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
index ca03dc15111..5c3188e7f9d 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.h
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -127,15 +127,9 @@ struct cell_context
    struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
    uint num_vertex_elements;
 
-   struct pipe_transfer *cbuf_transfer[PIPE_MAX_COLOR_BUFS];
-   struct pipe_transfer *zsbuf_transfer;
-
    ubyte *cbuf_map[PIPE_MAX_COLOR_BUFS];
    ubyte *zsbuf_map;
 
-   struct pipe_surface *tex_surf;
-   uint *tex_map;
-
    uint dirty;
    uint dirty_textures;  /* bitmask of texture units */
    uint dirty_samplers;  /* bitmask of sampler units */
diff --git a/src/gallium/drivers/cell/ppu/cell_fence.c b/src/gallium/drivers/cell/ppu/cell_fence.c
index 867b5dcaa09..32dbf5706c7 100644
--- a/src/gallium/drivers/cell/ppu/cell_fence.c
+++ b/src/gallium/drivers/cell/ppu/cell_fence.c
@@ -153,16 +153,12 @@ cell_add_fenced_textures(struct cell_context *cell)
    for (i = 0; i < cell->num_textures; i++) {
       struct cell_texture *ct = cell->texture[i];
       if (ct) {
-         uint level;
-         for (level = 0; level < CELL_MAX_TEXTURE_LEVELS; level++) {
-            if (ct->tiled_buffer[level]) {
 #if 0
-               printf("Adding texture %p buffer %p to list\n",
-                      ct, ct->tiled_buffer[level]);
+         printf("Adding texture %p buffer %p to list\n",
+                ct, ct->tiled_buffer[level]);
 #endif
-               cell_add_buffer_to_list(cell, list, ct->tiled_buffer[level]);
-            }
-         }
+         if (ct->buffer)
+            cell_add_buffer_to_list(cell, list, ct->buffer);
       }
    }
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_pipe_state.c b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
index facd9551fe7..ccd0fef6e84 100644
--- a/src/gallium/drivers/cell/ppu/cell_pipe_state.c
+++ b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
@@ -285,28 +285,22 @@ cell_map_surfaces(struct cell_context *cell)
    for (i = 0; i < 1; i++) {
       struct pipe_surface *ps = cell->framebuffer.cbufs[i];
       if (ps) {
-         cell->cbuf_transfer[i] =
-            screen->get_tex_transfer(screen, ps->texture, ps->face,
-                                     ps->level, ps->zslice,
-                                     PIPE_TRANSFER_READ_WRITE,
-                                     0, 0, ps->width, ps->height);
-
-         cell->cbuf_map[i] =
-            screen->transfer_map(screen, cell->cbuf_transfer[i]);
+         struct cell_texture *ct = cell_texture(ps->texture);
+         cell->cbuf_map[i] = screen->buffer_map(screen,
+                                                ct->buffer,
+                                                (PIPE_BUFFER_USAGE_GPU_READ |
+                                                 PIPE_BUFFER_USAGE_GPU_WRITE));
       }
    }
 
    {
       struct pipe_surface *ps = cell->framebuffer.zsbuf;
       if (ps) {
-         cell->zsbuf_transfer =
-            screen->get_tex_transfer(screen, ps->texture, ps->face,
-                                     ps->level, ps->zslice,
-                                     PIPE_TRANSFER_READ_WRITE,
-                                     0, 0, ps->width, ps->height);
-
-         cell->zsbuf_map =
-            screen->transfer_map(screen, cell->zsbuf_transfer);
+         struct cell_texture *ct = cell_texture(ps->texture);
+         cell->zsbuf_map = screen->buffer_map(screen,
+                                              ct->buffer,
+                                              (PIPE_BUFFER_USAGE_GPU_READ |
+                                               PIPE_BUFFER_USAGE_GPU_WRITE));
       }
    }
 }
@@ -322,24 +316,24 @@ cell_unmap_surfaces(struct cell_context *cell)
    uint i;
 
    for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
-      if (cell->cbuf_transfer[i] && cell->cbuf_map[i]) {
-         /* unmap color buffer/surface [i] */
-         screen->transfer_unmap(screen, cell->cbuf_transfer[i]);
-         cell->cbuf_map[i] = NULL;
+      struct pipe_surface *ps = cell->framebuffer.cbufs[i];
+      if (ps && cell->cbuf_map[i]) {
+         struct cell_texture *ct = cell_texture(ps->texture);
+         assert(ps->texture);
+         assert(ct->buffer);
 
-         /* get rid of transfer object [i] */
-         screen->tex_transfer_release(screen, &cell->cbuf_transfer[i]);
-         assert(cell->cbuf_transfer[i] == NULL);
+         screen->buffer_unmap(screen, ct->buffer);
+         cell->cbuf_map[i] = NULL;
       }
    }
 
-   if (cell->zsbuf_transfer && cell->zsbuf_map) {
-      screen->transfer_unmap(screen, cell->zsbuf_transfer);
-      cell->zsbuf_map = NULL;
-
-      /* get rid of transfer object */
-      screen->tex_transfer_release(screen, &cell->zsbuf_transfer);
-      assert(cell->zsbuf_transfer == NULL);
+   {
+      struct pipe_surface *ps = cell->framebuffer.zsbuf;
+      if (ps && cell->zsbuf_map) {
+         struct cell_texture *ct = cell_texture(ps->texture);
+         screen->buffer_unmap(screen, ct->buffer);
+         cell->zsbuf_map = NULL;
+      }
    }
 }
 
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index ff529fe22cb..9479c0898fd 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -287,19 +287,23 @@ cell_emit_state(struct cell_context *cell)
       for (i = 0;i < CELL_MAX_SAMPLERS; i++) {
          if (cell->dirty_textures & (1 << i)) {
             STATIC_ASSERT(sizeof(struct cell_command_texture) % 16 == 0);
-            struct cell_command_texture *texture
-               =  (struct cell_command_texture *)cell_batch_alloc16(cell, sizeof(*texture));
+            struct cell_command_texture *texture =
+               (struct cell_command_texture *)
+               cell_batch_alloc16(cell, sizeof(*texture));
+
             texture->opcode[0] = CELL_CMD_STATE_TEXTURE;
             texture->unit = i;
             if (cell->texture[i]) {
+               struct cell_texture *ct = cell->texture[i];
                uint level;
                for (level = 0; level < CELL_MAX_TEXTURE_LEVELS; level++) {
-                  texture->start[level] = cell->texture[i]->tiled_mapped[level];
-                  texture->width[level] = cell->texture[i]->base.width[level];
-                  texture->height[level] = cell->texture[i]->base.height[level];
-                  texture->depth[level] = cell->texture[i]->base.depth[level];
+                  texture->start[level] = (ct->mapped +
+                                           ct->level_offset[level]);
+                  texture->width[level] = ct->base.width[level];
+                  texture->height[level] = ct->base.height[level];
+                  texture->depth[level] = ct->base.depth[level];
                }
-               texture->target = cell->texture[i]->base.target;
+               texture->target = ct->base.target;
             }
             else {
                uint level;
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
index fa52e2cbea4..74724393f78 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.c
+++ b/src/gallium/drivers/cell/ppu/cell_texture.c
@@ -62,7 +62,7 @@ cell_texture_layout(struct cell_texture *ct)
 
    ct->buffer_size = 0;
 
-   for ( level = 0 ; level <= pt->last_level ; level++ ) {
+   for (level = 0; level <= pt->last_level; level++) {
       unsigned size;
       unsigned w_tile, h_tile;
 
@@ -90,7 +90,7 @@ cell_texture_layout(struct cell_texture *ct)
 
       ct->buffer_size += size;
 
-      width  = minify(width);
+      width = minify(width);
       height = minify(height);
       depth = minify(depth);
    }
@@ -131,33 +131,19 @@ cell_texture_release(struct pipe_screen *screen,
    if (!*pt)
       return;
 
-   /*
-   DBG("%s %p refcount will be %d\n",
-       __FUNCTION__, (void *) *pt, (*pt)->refcount - 1);
-   */
    if (--(*pt)->refcount <= 0) {
       /* Delete this texture now.
        * But note that the underlying pipe_buffer may linger...
        */
       struct cell_texture *ct = cell_texture(*pt);
-      uint i;
 
-      /*
-      DBG("%s deleting %p\n", __FUNCTION__, (void *) ct);
-      */
+      if (ct->mapped) {
+         pipe_buffer_unmap(screen, ct->buffer);
+         ct->mapped = NULL;
+      }
 
       pipe_buffer_reference(screen, &ct->buffer, NULL);
 
-      for (i = 0; i < CELL_MAX_TEXTURE_LEVELS; i++) {
-         /* Unreference the tiled image buffer.
-          * It may not actually be deleted until a fence is hit.
-          */
-         if (ct->tiled_buffer[i]) {
-            ct->tiled_mapped[i] = NULL;
-            pipe_buffer_reference(screen, &ct->tiled_buffer[i], NULL);
-         }
-      }
-
       FREE(ct);
    }
    *pt = NULL;
@@ -294,107 +280,6 @@ untwiddle_image_uint(uint w, uint h, uint tile_size, uint *dst,
 }
 
 
-/**
- * Convert linear texture image data to tiled format for SPU usage.
- */
-static void
-cell_twiddle_texture(struct pipe_screen *screen,
-                     struct pipe_surface *surface)
-{
-#if 0 // XXX fix me
-   struct cell_texture *ct = cell_texture(surface->texture);
-   const uint level = surface->level;
-   const uint texWidth = ct->base.width[level];
-   const uint texHeight = ct->base.height[level];
-   const uint bufWidth = align(texWidth, TILE_SIZE);
-   const uint bufHeight = align(texHeight, TILE_SIZE);
-   const void *map = screen->surface_map(screen, surface, PIPE_BUFFER_USAGE_CPU_READ);
-   const uint *src = (const uint *) map;
-
-   switch (ct->base.format) {
-   case PIPE_FORMAT_A8R8G8B8_UNORM:
-   case PIPE_FORMAT_B8G8R8A8_UNORM:
-   case PIPE_FORMAT_S8Z24_UNORM:
-      {
-         int numFaces = ct->base.target == PIPE_TEXTURE_CUBE ? 6 : 1;
-         int offset = bufWidth * bufHeight * 4 * surface->face;
-         uint *dst;
-
-         if (!ct->tiled_buffer[level]) {
-            /* allocate buffer for tiled data now */
-            struct pipe_winsys *ws = screen->winsys;
-            uint bytes = bufWidth * bufHeight * 4 * numFaces;
-            ct->tiled_buffer[level] =
-               ws->buffer_create(ws, 16, PIPE_BUFFER_USAGE_PIXEL, bytes);
-            /* and map it */
-            ct->tiled_mapped[level] =
-               ws->buffer_map(ws, ct->tiled_buffer[level],
-                              PIPE_BUFFER_USAGE_GPU_READ);
-         }
-         dst = (uint *) ((ubyte *) ct->tiled_mapped[level] + offset);
-
-         twiddle_image_uint(texWidth, texHeight, TILE_SIZE, dst,
-                            surface->stride, src);
-      }
-      break;
-   default:
-      printf("Cell: twiddle unsupported texture format %s\n",
-             pf_name(ct->base.format));
-   }
-
-   screen->surface_unmap(screen, surface);
-#endif
-}
-
-
-/**
- * Convert SPU tiled texture image data to linear format for app usage.
- */
-static void
-cell_untwiddle_texture(struct pipe_screen *screen,
-                     struct pipe_surface *surface)
-{
-#if 0 // XXX fix me
-   struct cell_texture *ct = cell_texture(surface->texture);
-   const uint level = surface->level;
-   const uint texWidth = ct->base.width[level];
-   const uint texHeight = ct->base.height[level];
-   const void *map = screen->surface_map(screen, surface, PIPE_BUFFER_USAGE_CPU_READ);
-   const uint *src = (const uint *) ((const ubyte *) map + surface->offset);
-
-   switch (ct->base.format) {
-   case PIPE_FORMAT_A8R8G8B8_UNORM:
-   case PIPE_FORMAT_B8G8R8A8_UNORM:
-   case PIPE_FORMAT_S8Z24_UNORM:
-      {
-         int numFaces = ct->base.target == PIPE_TEXTURE_CUBE ? 6 : 1;
-         int offset = surface->stride * texHeight * 4 * surface->face;
-         uint *dst;
-
-         if (!ct->untiled_data[level]) {
-            ct->untiled_data[level] =
-               align_malloc(surface->stride * texHeight * 4 * numFaces, 16);
-         }
-
-         dst = (uint *) ((ubyte *) ct->untiled_data[level] + offset);
-
-         untwiddle_image_uint(texWidth, texHeight, TILE_SIZE, dst,
-                              surface->stride, src);
-      }
-      break;
-   default:
-      {
-         ct->untiled_data[level] = NULL;
-         printf("Cell: untwiddle unsupported texture format %s\n",
-                pf_name(ct->base.format));
-      }
-   }
-
-   screen->surface_unmap(screen, surface);
-#endif
-}
-
-
 static struct pipe_surface *
 cell_get_tex_surface(struct pipe_screen *screen,
                      struct pipe_texture *pt,
@@ -409,38 +294,25 @@ cell_get_tex_surface(struct pipe_screen *screen,
       ps->refcount = 1;
       pipe_texture_reference(&ps->texture, pt);
       ps->format = pt->format;
-      //ps->block = pt->block;
       ps->width = pt->width[level];
       ps->height = pt->height[level];
-      //ps->nblocksx = pt->nblocksx[level];
-      //ps->nblocksy = pt->nblocksy[level];
-      //ps->stride = ct->stride[level];
       ps->offset = ct->level_offset[level];
-      ps->usage = usage;
-
       /* XXX may need to override usage flags (see sp_texture.c) */
-
-      pipe_texture_reference(&ps->texture, pt); 
+      ps->usage = usage;
       ps->face = face;
       ps->level = level;
       ps->zslice = zslice;
 
-      if (pt->target == PIPE_TEXTURE_CUBE || pt->target == PIPE_TEXTURE_3D) {
-#if 0 // XXX fix me
-         ps->offset += ((pt->target == PIPE_TEXTURE_CUBE) ? face : zslice) *
-            ps->nblocksy *
-            ps->stride;
-#endif
+      if (pt->target == PIPE_TEXTURE_CUBE) {
+         ps->offset += face * pt->nblocksy[level] * ct->stride[level];
+      }
+      else if (pt->target == PIPE_TEXTURE_3D) {
+         ps->offset += zslice * pt->nblocksy[level] * ct->stride[level];
       }
       else {
          assert(face == 0);
          assert(zslice == 0);
       }
-
-      if (ps->usage & PIPE_BUFFER_USAGE_CPU_READ) {
-         /* convert from tiled to linear layout */
-         cell_untwiddle_texture(screen, ps);
-      }
    }
    return ps;
 }
@@ -450,24 +322,8 @@ static void
 cell_tex_surface_release(struct pipe_screen *screen, 
                          struct pipe_surface **s)
 {
-   struct cell_texture *ct = cell_texture((*s)->texture);
-   const uint level = (*s)->level;
    struct pipe_surface *surf = *s;
 
-   if ((surf->usage & PIPE_BUFFER_USAGE_CPU_READ) && (ct->untiled_data[level]))
-   {
-      align_free(ct->untiled_data[level]);
-      ct->untiled_data[level] = NULL;
-   }
-
-   if ((ct->base.tex_usage & PIPE_TEXTURE_USAGE_SAMPLER) &&
-       (surf->usage & PIPE_BUFFER_USAGE_CPU_WRITE)) {
-      /* convert from linear to tiled layout */
-      cell_twiddle_texture(screen, surf);
-   }
-
-   /* XXX if done rendering to teximage, re-tile */
-
    if (--surf->refcount == 0) {
       pipe_texture_reference(&surf->texture, NULL);
       FREE(surf);
@@ -476,6 +332,11 @@ cell_tex_surface_release(struct pipe_screen *screen,
 }
 
 
+/**
+ * Create new pipe_transfer object.
+ * This is used by the user to put tex data into a texture (and get it
+ * back out for glGetTexImage).
+ */
 static struct pipe_transfer *
 cell_get_tex_transfer(struct pipe_screen *screen,
                       struct pipe_texture *texture,
@@ -485,14 +346,13 @@ cell_get_tex_transfer(struct pipe_screen *screen,
 {
    struct cell_texture *ct = cell_texture(texture);
    struct cell_transfer *ctrans;
-   struct pipe_transfer *pt;
 
    assert(texture);
    assert(level <= texture->last_level);
 
    ctrans = CALLOC_STRUCT(cell_transfer);
-   pt = &ctrans->base;
    if (ctrans) {
+      struct pipe_transfer *pt = &ctrans->base;
       pt->refcount = 1;
       pipe_texture_reference(&pt->texture, texture);
       pt->format = texture->format;
@@ -504,23 +364,26 @@ cell_get_tex_transfer(struct pipe_screen *screen,
       pt->nblocksx = texture->nblocksx[level];
       pt->nblocksy = texture->nblocksy[level];
       pt->stride = ct->stride[level];
-      ctrans->offset = ct->level_offset[level];
       pt->usage = usage;
       pt->face = face;
       pt->level = level;
       pt->zslice = zslice;
 
-      if (texture->target == PIPE_TEXTURE_CUBE ||
-          texture->target == PIPE_TEXTURE_3D) {
-         ctrans->offset += ((texture->target == PIPE_TEXTURE_CUBE) ? face :
-                         zslice) * pt->nblocksy * pt->stride;
+      ctrans->offset = ct->level_offset[level];
+
+      if (texture->target == PIPE_TEXTURE_CUBE) {
+         ctrans->offset += face * pt->nblocksy * pt->stride;
+      }
+      else if (texture->target == PIPE_TEXTURE_3D) {
+         ctrans->offset += zslice * pt->nblocksy * pt->stride;
       }
       else {
          assert(face == 0);
          assert(zslice == 0);
       }
+      return pt;
    }
-   return pt;
+   return NULL;
 }
 
 
@@ -542,16 +405,23 @@ cell_tex_transfer_release(struct pipe_screen *screen,
 }
 
 
+/**
+ * Return pointer to texture image data in linear layout.
+ */
 static void *
-cell_transfer_map( struct pipe_screen *screen,
-                       struct pipe_transfer *transfer )
+cell_transfer_map(struct pipe_screen *screen, struct pipe_transfer *transfer)
 {
-   ubyte *map;
-   struct cell_texture *spt;
-   unsigned flags = 0;
+   struct cell_transfer *ctrans = cell_transfer(transfer);
+   struct pipe_texture *pt = transfer->texture;
+   struct cell_texture *ct = cell_texture(pt);
+   const uint level = ctrans->base.level;
+   const uint texWidth = pt->width[level];
+   const uint texHeight = pt->height[level];
+   const uint stride = ct->stride[level];
+   unsigned flags = 0x0;
+   unsigned size;
 
    assert(transfer->texture);
-   spt = cell_texture(transfer->texture);
 
    if (transfer->usage != PIPE_TRANSFER_READ) {
       flags |= PIPE_BUFFER_USAGE_CPU_WRITE;
@@ -561,89 +431,82 @@ cell_transfer_map( struct pipe_screen *screen,
       flags |= PIPE_BUFFER_USAGE_CPU_READ;
    }
 
-   map = pipe_buffer_map(screen, spt->buffer, flags);
-   if (map == NULL)
-      return NULL;
+   if (!ct->mapped) {
+      /* map now */
+      ct->mapped = pipe_buffer_map(screen, ct->buffer, flags);
+   }
 
-   /* May want to different things here depending on read/write nature
-    * of the map:
+   /*
+    * Create a buffer of ordinary memory for the linear texture.
+    * This is the memory that the user will read/write.
     */
-   if (transfer->texture && transfer->usage != PIPE_TRANSFER_READ) 
-   {
-      /* Do something to notify sharing contexts of a texture change.
-       * In cell, that would mean flushing the texture cache.
-       */
-#if 00
-      cell_screen(screen)->timestamp++;
-#endif
+   size = pt->nblocksx[level] * pt->nblocksy[level] * pt->block.size;
+
+   ctrans->map = align_malloc(size, 16);
+   if (!ctrans->map)
+      return NULL; /* out of memory */
+
+   if (transfer->usage & PIPE_TRANSFER_READ) {
+      /* need to untwiddle the texture to make a linear version */
+      const uint bpp = pf_get_size(ct->base.format);
+      if (bpp == 4) {
+         const uint *src = (uint *) (ct->mapped + ctrans->offset);
+         uint *dst = ctrans->map;
+         untwiddle_image_uint(texWidth, texHeight, TILE_SIZE,
+                              dst, stride, src);
+      }
+      else {
+         // xxx fix
+      }
    }
-   
-   return map + cell_transfer(transfer)->offset +
-      transfer->y / transfer->block.height * transfer->stride +
-      transfer->x / transfer->block.width * transfer->block.size;
+
+   return ctrans->map;
 }
 
 
+/**
+ * Called when user is done reading/writing texture data.
+ * If new data was written, this is where we convert the linear data
+ * to tiled data.
+ */
 static void
 cell_transfer_unmap(struct pipe_screen *screen,
-                       struct pipe_transfer *transfer)
+                    struct pipe_transfer *transfer)
 {
-   struct cell_texture *spt;
-
-   assert(transfer->texture);
-   spt = cell_texture(transfer->texture);
-
-   pipe_buffer_unmap( screen, spt->buffer );
-}
-
-
-static void *
-cell_surface_map(struct pipe_screen *screen,
-                 struct pipe_surface *surface,
-                 unsigned flags)
-{
-   ubyte *map;
-   struct cell_texture *ct = cell_texture(surface->texture);
-   const uint level = surface->level;
-
-   assert(ct);
-
-#if 0
-   if (flags & ~surface->usage) {
-      assert(0);
-      return NULL;
+   struct cell_transfer *ctrans = cell_transfer(transfer);
+   struct pipe_texture *pt = transfer->texture;
+   struct cell_texture *ct = cell_texture(pt);
+   const uint level = ctrans->base.level;
+   const uint texWidth = pt->width[level];
+   const uint texHeight = pt->height[level];
+   const uint stride = ct->stride[level];
+
+   if (!ct->mapped) {
+      /* map now */
+      ct->mapped = pipe_buffer_map(screen, ct->buffer,
+                                   PIPE_BUFFER_USAGE_CPU_READ);
    }
-#endif
 
-   map = pipe_buffer_map( screen, ct->buffer, flags );
-   if (map == NULL) {
-      return NULL;
-   }
-   else {
-      if ((surface->usage & PIPE_BUFFER_USAGE_CPU_READ) &&
-          (ct->untiled_data[level])) {
-         return (void *) ((ubyte *) ct->untiled_data[level] + surface->offset);
+   if (transfer->usage & PIPE_TRANSFER_WRITE) {
+      /* The user wrote new texture data into the mapped buffer.
+       * We need to convert the new linear data into the twiddled/tiled format.
+       */
+      const uint bpp = pf_get_size(ct->base.format);
+      if (bpp == 4) {
+         const uint *src = ctrans->map;
+         uint *dst = (uint *) (ct->mapped + ctrans->offset);
+         twiddle_image_uint(texWidth, texHeight, TILE_SIZE, dst, stride, src);
       }
       else {
-         return (void *) (map + surface->offset);
+         // xxx fix
       }
    }
-}
-
-
-static void
-cell_surface_unmap(struct pipe_screen *screen,
-                   struct pipe_surface *surface)
-{
-   struct cell_texture *ct = cell_texture(surface->texture);
-
-   assert(ct);
 
-   pipe_buffer_unmap( screen, ct->buffer );
+   align_free(ctrans->map);
+   ctrans->map = NULL;
 }
 
 
-
 void
 cell_init_screen_texture_funcs(struct pipe_screen *screen)
 {
@@ -655,6 +518,7 @@ cell_init_screen_texture_funcs(struct pipe_screen *screen)
 
    screen->get_tex_transfer = cell_get_tex_transfer;
    screen->tex_transfer_release = cell_tex_transfer_release;
+
    screen->transfer_map = cell_transfer_map;
    screen->transfer_unmap = cell_transfer_unmap;
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.h b/src/gallium/drivers/cell/ppu/cell_texture.h
index fc6486adbe6..3ffc0bfdb51 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.h
+++ b/src/gallium/drivers/cell/ppu/cell_texture.h
@@ -43,20 +43,14 @@ struct cell_texture
    unsigned long level_offset[CELL_MAX_TEXTURE_LEVELS];
    unsigned long stride[CELL_MAX_TEXTURE_LEVELS];
 
-   /* The data is held here:
-    */
+   /** The tiled texture data is held in this buffer */
    struct pipe_buffer *buffer;
    unsigned long buffer_size;
 
-   /** Texture data in tiled layout is held here */
-   struct pipe_buffer *tiled_buffer[CELL_MAX_TEXTURE_LEVELS];
-   /** Mapped, tiled texture data */
-   void *tiled_mapped[CELL_MAX_TEXTURE_LEVELS];
-
-   struct pipe_transfer *transfer;
-
-   /** The original, linear texture data */
-   void *untiled_data[CELL_MAX_TEXTURE_LEVELS];
+   /** The buffer above, mapped.  This is the memory from which the
+    * SPUs will fetch texels.  This texture data is in the tiled layout.
+    */
+   ubyte *mapped;
 };
 
 
@@ -65,6 +59,7 @@ struct cell_transfer
    struct pipe_transfer base;
 
    unsigned long offset;
+   void *map;
 };
 
 
diff --git a/src/gallium/drivers/i915simple/i915_screen.c b/src/gallium/drivers/i915simple/i915_screen.c
index b7bd3b3b748..49471287a2e 100644
--- a/src/gallium/drivers/i915simple/i915_screen.c
+++ b/src/gallium/drivers/i915simple/i915_screen.c
@@ -36,6 +36,7 @@
 #include "i915_context.h"
 #include "i915_screen.h"
 #include "i915_texture.h"
+#include "i915_winsys.h"
 
 
 static const char *
diff --git a/src/gallium/drivers/i915simple/i915_screen.h b/src/gallium/drivers/i915simple/i915_screen.h
index a371663453c..5284c325951 100644
--- a/src/gallium/drivers/i915simple/i915_screen.h
+++ b/src/gallium/drivers/i915simple/i915_screen.h
@@ -75,10 +75,6 @@ i915_transfer( struct pipe_transfer *transfer )
 }
 
 
-extern struct pipe_screen *
-i915_create_screen(struct pipe_winsys *winsys, uint pci_id);
-
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/gallium/drivers/i915simple/i915_state_emit.c b/src/gallium/drivers/i915simple/i915_state_emit.c
index 26e03f5127f..1e1fb968b47 100644
--- a/src/gallium/drivers/i915simple/i915_state_emit.c
+++ b/src/gallium/drivers/i915simple/i915_state_emit.c
@@ -214,7 +214,6 @@ i915_emit_hardware_state(struct i915_context *i915 )
 	 unsigned ctile = BUF_3D_USE_FENCE;
          struct i915_texture *tex = (struct i915_texture *)
                                     cbuf_surface->texture;
-         struct pipe_buffer *buffer = tex->buffer;
          assert(tex);
 
 	 if (tex && tex->tiled) {
@@ -238,7 +237,6 @@ i915_emit_hardware_state(struct i915_context *i915 )
 	 unsigned ztile = BUF_3D_USE_FENCE;
          struct i915_texture *tex = (struct i915_texture *)
                                     depth_surface->texture;
-         struct pipe_buffer *buffer = tex->buffer;
          assert(tex);
 
 	 if (tex && tex->tiled) {
diff --git a/src/gallium/drivers/i915simple/i915_texture.c b/src/gallium/drivers/i915simple/i915_texture.c
index 957726523fa..6aead3e75e9 100644
--- a/src/gallium/drivers/i915simple/i915_texture.c
+++ b/src/gallium/drivers/i915simple/i915_texture.c
@@ -42,6 +42,7 @@
 #include "i915_texture.h"
 #include "i915_debug.h"
 #include "i915_screen.h"
+#include "i915_winsys.h"
 
 /*
  * Helper function and arrays
@@ -765,3 +766,20 @@ i915_init_screen_texture_functions(struct pipe_screen *screen)
    screen->texture_blanket = i915_texture_blanket;
    screen->tex_surface_release = i915_tex_surface_release;
 }
+
+boolean i915_get_texture_buffer( struct pipe_texture *texture,
+                                 struct pipe_buffer **buf,
+                                 unsigned *stride )
+{
+   struct i915_texture *tex = (struct i915_texture *)texture;
+
+   if (!tex)
+      return FALSE;
+
+   pipe_buffer_reference(texture->screen, buf, tex->buffer);
+
+   if (stride)
+      *stride = tex->stride;
+
+   return TRUE;
+}
diff --git a/src/gallium/drivers/i915simple/i915_winsys.h b/src/gallium/drivers/i915simple/i915_winsys.h
index 81904c2a742..ff5b34f193a 100644
--- a/src/gallium/drivers/i915simple/i915_winsys.h
+++ b/src/gallium/drivers/i915simple/i915_winsys.h
@@ -30,7 +30,8 @@
  * This is the interface that i915simple requires any window system
  * hosting it to implement.  This is the only include file in i915simple
  * which is public.
- * 
+ *
+ * This isn't currently true as the winsys needs i915_batchbuffer.h
  */
 
 #ifndef I915_WINSYS_H
@@ -45,10 +46,9 @@ extern "C" {
 #endif
 
    
-/* Pipe drivers are (meant to be!) independent of both GL and the
- * window system.  The window system provides a buffer manager and a
- * set of additional hooks for things like command buffer submission,
- * etc.
+/* Pipe drivers are independent of both GL and the window system.
+ * The window system provides a buffer manager and a set of additional
+ * hooks for things like command buffer submission, etc.
  *
  * There clearly has to be some agreement between the window system
  * driver and the hardware driver about the format of command buffers,
@@ -56,6 +56,7 @@ extern "C" {
  */
 
 struct i915_batchbuffer;
+struct pipe_texture;
 struct pipe_buffer;
 struct pipe_fence_handle;
 struct pipe_winsys;
@@ -64,7 +65,7 @@ struct pipe_screen;
 
 /**
  * Additional winsys interface for i915simple.
- * 
+ *
  * It is an over-simple batchbuffer mechanism.  Will want to improve the
  * performance of this, perhaps based on the cmdstream stuff.  It
  * would be pretty impossible to implement swz on top of this
@@ -110,12 +111,33 @@ struct i915_winsys {
 #define I915_BUFFER_USAGE_LIT_VERTEX  (PIPE_BUFFER_USAGE_CUSTOM << 0)
 
 
-struct pipe_context *i915_create_context( struct pipe_screen *,
-                                          struct pipe_winsys *,
-                                          struct i915_winsys * );
+/**
+ * Create i915 pipe_screen.
+ */
+struct pipe_screen *i915_create_screen( struct pipe_winsys *winsys,
+                                        uint pci_id );
+
+/**
+ * Create a i915 pipe_context.
+ */
+struct pipe_context *i915_create_context( struct pipe_screen *screen,
+                                          struct pipe_winsys *winsys,
+                                          struct i915_winsys *i915 );
+
+/**
+ * Used for the winsys to get the buffer used for a texture
+ * and also the stride used for the texture.
+ *
+ * Buffer is referenced for you so you need to unref after use.
+ *
+ * This is needed for example kms.
+ */
+boolean i915_get_texture_buffer( struct pipe_texture *texture,
+                                 struct pipe_buffer **buf,
+                                 unsigned *stride );
 
 #ifdef __cplusplus
 }
 #endif
 
-#endif 
+#endif
diff --git a/src/gallium/drivers/nv04/Makefile b/src/gallium/drivers/nv04/Makefile
index cf9deea8519..7c14bacb1de 100644
--- a/src/gallium/drivers/nv04/Makefile
+++ b/src/gallium/drivers/nv04/Makefile
@@ -15,6 +15,7 @@ C_SOURCES = \
 	nv04_state.c \
 	nv04_state_emit.c \
 	nv04_surface.c \
+	nv04_transfer.c \
 	nv04_vbo.c
 
 include ../../Makefile.template
diff --git a/src/gallium/drivers/nv04/nv04_miptree.c b/src/gallium/drivers/nv04/nv04_miptree.c
index 993c5ef5dd2..9acd613e2e4 100644
--- a/src/gallium/drivers/nv04/nv04_miptree.c
+++ b/src/gallium/drivers/nv04/nv04_miptree.c
@@ -50,8 +50,6 @@ nv04_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	mt->base = *pt;
 	mt->base.refcount = 1;
 	mt->base.screen = pscreen;
-	mt->shadow_tex = NULL;
-	mt->shadow_surface = NULL;
 
 	//mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 
@@ -111,12 +109,6 @@ nv04_miptree_release(struct pipe_screen *pscreen, struct pipe_texture **ppt)
 			FREE(mt->level[l].image_offset);
 	}
 
-	if (mt->shadow_tex) {
-		assert(mt->shadow_surface);
-		pscreen->tex_surface_release(pscreen, &mt->shadow_surface);
-		nv04_miptree_release(pscreen, &mt->shadow_tex);
-	}
-
 	FREE(mt);
 }
 
@@ -126,29 +118,26 @@ nv04_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 			 unsigned flags)
 {
 	struct nv04_miptree *nv04mt = (struct nv04_miptree *)pt;
-	struct pipe_surface *ps;
+	struct nv04_surface *ns;
 
-	ps = CALLOC_STRUCT(pipe_surface);
-	if (!ps)
+	ns = CALLOC_STRUCT(nv04_surface);
+	if (!ns)
 		return NULL;
-	pipe_texture_reference(&ps->texture, pt);
-	ps->format = pt->format;
-	ps->width = pt->width[level];
-	ps->height = pt->height[level];
-	ps->block = pt->block;
-	ps->nblocksx = pt->nblocksx[level];
-	ps->nblocksy = pt->nblocksy[level];
-	ps->stride = nv04mt->level[level].pitch;
-	ps->usage = flags;
-	ps->status = PIPE_SURFACE_STATUS_DEFINED;
-	ps->refcount = 1;
-	ps->face = face;
-	ps->level = level;
-	ps->zslice = zslice;
-
-	ps->offset = nv04mt->level[level].image_offset;
-
-	return ps;
+	pipe_texture_reference(&ns->base.texture, pt);
+	ns->base.format = pt->format;
+	ns->base.width = pt->width[level];
+	ns->base.height = pt->height[level];
+	ns->base.usage = flags;
+	ns->base.status = PIPE_SURFACE_STATUS_DEFINED;
+	ns->base.refcount = 1;
+	ns->base.face = face;
+	ns->base.level = level;
+	ns->base.zslice = zslice;
+	ns->pitch = nv04mt->level[level].pitch;
+
+	ns->base.offset = nv04mt->level[level].image_offset;
+
+	return &ns->base;
 }
 
 static void
diff --git a/src/gallium/drivers/nv04/nv04_screen.c b/src/gallium/drivers/nv04/nv04_screen.c
index 9ef38bc244c..f9f6d974264 100644
--- a/src/gallium/drivers/nv04/nv04_screen.c
+++ b/src/gallium/drivers/nv04/nv04_screen.c
@@ -119,28 +119,6 @@ nv04_screen_is_format_supported(struct pipe_screen *screen,
 	return FALSE;
 }
 
-static void *
-nv04_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
-		 unsigned flags )
-{
-	void *map;
-	struct nv04_miptree *nv04mt = (struct nv04_miptree *)surface->texture;
-
-	map = pipe_buffer_map(screen, nv04mt->buffer, flags);
-	if (!map)
-		return NULL;
-
-	return map + surface->offset;
-}
-
-static void
-nv04_surface_unmap(struct pipe_screen *screen, struct pipe_surface *surface)
-{
-	struct nv04_miptree *nv04mt = (struct nv04_miptree *)surface->texture;
-
-	pipe_buffer_unmap(screen, nv04mt->buffer);
-}
-
 static void
 nv04_screen_destroy(struct pipe_screen *pscreen)
 {
@@ -226,10 +204,8 @@ nv04_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 
 	screen->pipe.is_format_supported = nv04_screen_is_format_supported;
 
-	screen->pipe.surface_map = nv04_surface_map;
-	screen->pipe.surface_unmap = nv04_surface_unmap;
-
 	nv04_screen_init_miptree_functions(&screen->pipe);
+	nv04_screen_init_transfer_functions(&screen->pipe);
 	u_simple_screen_init(&screen->pipe);
 
 	return &screen->pipe;
diff --git a/src/gallium/drivers/nv04/nv04_screen.h b/src/gallium/drivers/nv04/nv04_screen.h
index 540aec907bf..ee6fb6db44a 100644
--- a/src/gallium/drivers/nv04/nv04_screen.h
+++ b/src/gallium/drivers/nv04/nv04_screen.h
@@ -24,4 +24,7 @@ nv04_screen(struct pipe_screen *screen)
 	return (struct nv04_screen *)screen;
 }
 
+void
+nv04_screen_init_transfer_functions(struct pipe_screen *pscreen);
+
 #endif
diff --git a/src/gallium/drivers/nv04/nv04_state.h b/src/gallium/drivers/nv04/nv04_state.h
index 15d4685ec1a..0d51439e3ff 100644
--- a/src/gallium/drivers/nv04/nv04_state.h
+++ b/src/gallium/drivers/nv04/nv04_state.h
@@ -35,9 +35,6 @@ struct nv04_miptree {
 	struct pipe_buffer *buffer;
 	uint total_size;
 
-	struct pipe_texture *shadow_tex;
-	struct pipe_surface *shadow_surface;
-
 	struct {
 		uint pitch;
 		uint image_offset;
diff --git a/src/gallium/drivers/nv04/nv04_state_emit.c b/src/gallium/drivers/nv04/nv04_state_emit.c
index bd8ef1adbff..eb2c1c57c67 100644
--- a/src/gallium/drivers/nv04/nv04_state_emit.c
+++ b/src/gallium/drivers/nv04/nv04_state_emit.c
@@ -93,7 +93,7 @@ static void nv04_emit_sampler(struct nv04_context *nv04, int unit)
 static void nv04_state_emit_framebuffer(struct nv04_context* nv04)
 {
 	struct pipe_framebuffer_state* fb = nv04->framebuffer;
-	struct pipe_surface *rt, *zeta;
+	struct nv04_surface *rt, *zeta;
 	uint32_t rt_format, w, h;
 	int colour_format = 0, zeta_format = 0;
 	struct nv04_miptree *nv04mt = 0;
@@ -101,7 +101,7 @@ static void nv04_state_emit_framebuffer(struct nv04_context* nv04)
 	w = fb->cbufs[0]->width;
 	h = fb->cbufs[0]->height;
 	colour_format = fb->cbufs[0]->format;
-	rt = fb->cbufs[0];
+	rt = (struct nv04_surface *)fb->cbufs[0];
 
 	if (fb->zsbuf) {
 		if (colour_format) {
@@ -113,7 +113,7 @@ static void nv04_state_emit_framebuffer(struct nv04_context* nv04)
 		}
 
 		zeta_format = fb->zsbuf->format;
-		zeta = fb->zsbuf;
+		zeta = (struct nv04_surface *)fb->zsbuf;
 	}
 
 	switch (colour_format) {
@@ -131,13 +131,13 @@ static void nv04_state_emit_framebuffer(struct nv04_context* nv04)
 	BEGIN_RING(context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_FORMAT, 1);
 	OUT_RING(rt_format);
 
-	nv04mt = (struct nv04_miptree *)rt->texture;
+	nv04mt = (struct nv04_miptree *)rt->base.texture;
 	/* FIXME pitches have to be aligned ! */
 	BEGIN_RING(context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_PITCH, 2);
-	OUT_RING(rt->stride|(zeta->stride<<16));
+	OUT_RING(rt->pitch|(zeta->pitch<<16));
 	OUT_RELOCl(nv04mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	if (fb->zsbuf) {
-		nv04mt = (struct nv04_miptree *)zeta->texture;
+		nv04mt = (struct nv04_miptree *)zeta->base.texture;
 		BEGIN_RING(context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_OFFSET_ZETA, 1);
 		OUT_RELOCl(nv04mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	}
@@ -202,8 +202,11 @@ nv04_emit_hw_state(struct nv04_context *nv04)
 	 */
 
 	/* Render target */
+	unsigned rt_pitch = ((struct nv04_surface *)nv04->rt)->pitch;
+	unsigned zeta_pitch = ((struct nv04_surface *)nv04->zeta)->pitch;
+
 	BEGIN_RING(context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_PITCH, 2);
-	OUT_RING(nv04->rt->stride|(nv04->zeta->stride<<16));
+	OUT_RING(rt_pitch|(zeta_pitch<<16));
 	OUT_RELOCl(nv04->rt, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	if (nv04->zeta) {
 		BEGIN_RING(context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_OFFSET_ZETA, 1);
diff --git a/src/gallium/drivers/nv04/nv04_surface_2d.c b/src/gallium/drivers/nv04/nv04_surface_2d.c
index 230cfd17dd9..e8fd3166ab8 100644
--- a/src/gallium/drivers/nv04/nv04_surface_2d.c
+++ b/src/gallium/drivers/nv04/nv04_surface_2d.c
@@ -101,6 +101,7 @@ nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx,
 	struct nouveau_grobj *sifm = ctx->sifm;
 	struct nouveau_bo *src_bo = ctx->nvws->get_bo(ctx->buf(src));
 	struct nouveau_bo *dst_bo = ctx->nvws->get_bo(ctx->buf(dst));
+	const unsigned src_pitch = ((struct nv04_surface *)src)->pitch;
 	const unsigned max_w = 1024;
 	const unsigned max_h = 1024;
 	const unsigned sub_w = w > max_w ? max_w : w;
@@ -110,6 +111,8 @@ nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx,
 
 	/* POT or GTFO */
 	assert(!(w & (w - 1)) && !(h & (h - 1)));
+	/* That's the way she likes it */
+	assert(src_pitch == ((struct nv04_surface *)dst)->pitch);
 
 	BEGIN_RING(chan, swzsurf, NV04_SWIZZLED_SURFACE_DMA_IMAGE, 1);
 	OUT_RELOCo(chan, dst_bo,
@@ -130,7 +133,7 @@ nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx,
 	  for (cx = 0; cx < w; cx += sub_w) {
 	    BEGIN_RING(chan, swzsurf, NV04_SWIZZLED_SURFACE_OFFSET, 1);
 	    OUT_RELOCl(chan, dst_bo, dst->offset + nv04_swizzle_bits(cx, cy) *
-			     dst->block.size, NOUVEAU_BO_GART |
+			     dst->texture->block.size, NOUVEAU_BO_GART |
 			     NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 
 	    BEGIN_RING(chan, sifm, NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_CONVERSION, 9);
@@ -146,11 +149,11 @@ nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx,
 
 	    BEGIN_RING(chan, sifm, NV04_SCALED_IMAGE_FROM_MEMORY_SIZE, 4);
 	    OUT_RING  (chan, sub_h << 16 | sub_w);
-	    OUT_RING  (chan, src->stride |
+	    OUT_RING  (chan, src_pitch |
 			     NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_ORIGIN_CENTER |
 			     NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_FILTER_POINT_SAMPLE);
-	    OUT_RELOCl(chan, src_bo, src->offset + cy * src->stride +
-			     cx * src->block.size, NOUVEAU_BO_GART |
+	    OUT_RELOCl(chan, src_bo, src->offset + cy * src_pitch +
+			     cx * src->texture->block.size, NOUVEAU_BO_GART |
 			     NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
 	    OUT_RING  (chan, 0);
 	  }
@@ -168,10 +171,12 @@ nv04_surface_copy_m2mf(struct nv04_surface_2d *ctx,
 	struct nouveau_grobj *m2mf = ctx->m2mf;
 	struct nouveau_bo *src_bo = ctx->nvws->get_bo(ctx->buf(src));
 	struct nouveau_bo *dst_bo = ctx->nvws->get_bo(ctx->buf(dst));
-	unsigned dst_offset, src_offset;
-
-	dst_offset = dst->offset + (dy * dst->stride) + (dx * dst->block.size);
-	src_offset = src->offset + (sy * src->stride) + (sx * src->block.size);
+	unsigned src_pitch = ((struct nv04_surface *)src)->pitch;
+	unsigned dst_pitch = ((struct nv04_surface *)dst)->pitch;
+	unsigned dst_offset = dst->offset + dy * dst_pitch +
+	                      dx * dst->texture->block.size;
+	unsigned src_offset = src->offset + sy * src_pitch +
+	                      sx * src->texture->block.size;
 
 	WAIT_RING (chan, 3 + ((h / 2047) + 1) * 9);
 	BEGIN_RING(chan, m2mf, NV04_MEMORY_TO_MEMORY_FORMAT_DMA_BUFFER_IN, 2);
@@ -188,16 +193,16 @@ nv04_surface_copy_m2mf(struct nv04_surface_2d *ctx,
 			   NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD);
 		OUT_RELOCl(chan, dst_bo, dst_offset,
 			   NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_WR);
-		OUT_RING  (chan, src->stride);
-		OUT_RING  (chan, dst->stride);
-		OUT_RING  (chan, w * src->block.size);
+		OUT_RING  (chan, src_pitch);
+		OUT_RING  (chan, dst_pitch);
+		OUT_RING  (chan, w * src->texture->block.size);
 		OUT_RING  (chan, count);
 		OUT_RING  (chan, 0x0101);
 		OUT_RING  (chan, 0);
 
 		h -= count;
-		src_offset += src->stride * count;
-		dst_offset += dst->stride * count;
+		src_offset += src_pitch * count;
+		dst_offset += dst_pitch * count;
 	}
 
 	return 0;
@@ -213,6 +218,8 @@ nv04_surface_copy_blit(struct nv04_surface_2d *ctx, struct pipe_surface *dst,
 	struct nouveau_grobj *blit = ctx->blit;
 	struct nouveau_bo *src_bo = ctx->nvws->get_bo(ctx->buf(src));
 	struct nouveau_bo *dst_bo = ctx->nvws->get_bo(ctx->buf(dst));
+	unsigned src_pitch = ((struct nv04_surface *)src)->pitch;
+	unsigned dst_pitch = ((struct nv04_surface *)dst)->pitch;
 	int format;
 
 	format = nv04_surface_format(dst->format);
@@ -225,7 +232,7 @@ nv04_surface_copy_blit(struct nv04_surface_2d *ctx, struct pipe_surface *dst,
 	OUT_RELOCo(chan, dst_bo, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	BEGIN_RING(chan, surf2d, NV04_CONTEXT_SURFACES_2D_FORMAT, 4);
 	OUT_RING  (chan, format);
-	OUT_RING  (chan, (dst->stride << 16) | src->stride);
+	OUT_RING  (chan, (dst_pitch << 16) | src_pitch);
 	OUT_RELOCl(chan, src_bo, src->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
 	OUT_RELOCl(chan, dst_bo, dst->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 
@@ -242,6 +249,8 @@ nv04_surface_copy(struct nv04_surface_2d *ctx, struct pipe_surface *dst,
 		  int dx, int dy, struct pipe_surface *src, int sx, int sy,
 		  int w, int h)
 {
+	unsigned src_pitch = ((struct nv04_surface *)src)->pitch;
+	unsigned dst_pitch = ((struct nv04_surface *)dst)->pitch;
 	int src_linear = src->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR;
 	int dst_linear = dst->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR;
 
@@ -257,7 +266,8 @@ nv04_surface_copy(struct nv04_surface_2d *ctx, struct pipe_surface *dst,
 	 * to NV_MEMORY_TO_MEMORY_FORMAT in this case.
 	 */
 	if ((src->offset & 63) || (dst->offset & 63) ||
-	    (src->stride & 63) || (dst->stride & 63)) {
+	    (src_pitch & 63) || (dst_pitch & 63) ||
+	    debug_get_bool_option("NOUVEAU_NO_COPYBLIT", FALSE)) {
 		nv04_surface_copy_m2mf(ctx, dst, dx, dy, src, sx, sy, w, h);
 		return;
 	}
@@ -273,6 +283,7 @@ nv04_surface_fill(struct nv04_surface_2d *ctx, struct pipe_surface *dst,
 	struct nouveau_grobj *surf2d = ctx->surf2d;
 	struct nouveau_grobj *rect = ctx->rect;
 	struct nouveau_bo *dst_bo = ctx->nvws->get_bo(ctx->buf(dst));
+	unsigned dst_pitch = ((struct nv04_surface *)dst)->pitch;
 	int cs2d_format, gdirect_format;
 
 	cs2d_format = nv04_surface_format(dst->format);
@@ -287,7 +298,7 @@ nv04_surface_fill(struct nv04_surface_2d *ctx, struct pipe_surface *dst,
 	OUT_RELOCo(chan, dst_bo, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	BEGIN_RING(chan, surf2d, NV04_CONTEXT_SURFACES_2D_FORMAT, 4);
 	OUT_RING  (chan, cs2d_format);
-	OUT_RING  (chan, (dst->stride << 16) | dst->stride);
+	OUT_RING  (chan, (dst_pitch << 16) | dst_pitch);
 	OUT_RELOCl(chan, dst_bo, dst->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	OUT_RELOCl(chan, dst_bo, dst->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 
diff --git a/src/gallium/drivers/nv04/nv04_surface_2d.h b/src/gallium/drivers/nv04/nv04_surface_2d.h
index 21b8f869606..82ce7189c8b 100644
--- a/src/gallium/drivers/nv04/nv04_surface_2d.h
+++ b/src/gallium/drivers/nv04/nv04_surface_2d.h
@@ -1,6 +1,11 @@
 #ifndef __NV04_SURFACE_2D_H__
 #define __NV04_SURFACE_2D_H__
 
+struct nv04_surface {
+	struct pipe_surface base;
+	unsigned pitch;
+};
+
 struct nv04_surface_2d {
 	struct nouveau_winsys *nvws;
 	struct nouveau_notifier *ntfy;
diff --git a/src/gallium/drivers/nv04/nv04_transfer.c b/src/gallium/drivers/nv04/nv04_transfer.c
new file mode 100644
index 00000000000..314d2045f76
--- /dev/null
+++ b/src/gallium/drivers/nv04/nv04_transfer.c
@@ -0,0 +1,201 @@
+#include <pipe/p_state.h>
+#include <pipe/p_defines.h>
+#include <pipe/p_inlines.h>
+#include <util/u_memory.h>
+#include <nouveau/nouveau_winsys.h>
+#include "nv04_context.h"
+#include "nv04_screen.h"
+#include "nv04_state.h"
+
+struct nv04_transfer {
+	struct pipe_transfer base;
+	struct pipe_surface *surface;
+	bool direct;
+};
+
+static unsigned nv04_usage_tx_to_buf(unsigned tx_usage)
+{
+	switch (tx_usage) {
+		case PIPE_TRANSFER_READ:
+			return PIPE_BUFFER_USAGE_CPU_READ;
+		case PIPE_TRANSFER_WRITE:
+			return PIPE_BUFFER_USAGE_CPU_WRITE;
+		case PIPE_TRANSFER_READ_WRITE:
+			return PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+		default:
+			assert(0);
+	}
+
+	return -1;
+}
+
+static void
+nv04_compatible_transfer_tex(struct pipe_texture *pt, unsigned level,
+                             struct pipe_texture *template)
+{
+	memset(template, 0, sizeof(struct pipe_texture));
+	template->target = pt->target;
+	template->format = pt->format;
+	template->width[0] = pt->width[level];
+	template->height[0] = pt->height[level];
+	template->depth[0] = 1;
+	template->block = pt->block;
+	template->nblocksx[0] = pt->nblocksx[level];
+	template->nblocksy[0] = pt->nblocksx[level];
+	template->last_level = 0;
+	template->compressed = pt->compressed;
+	template->nr_samples = pt->nr_samples;
+
+	template->tex_usage = PIPE_TEXTURE_USAGE_DYNAMIC |
+	                      NOUVEAU_TEXTURE_USAGE_LINEAR;
+}
+
+static struct pipe_transfer *
+nv04_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
+		  unsigned face, unsigned level, unsigned zslice,
+		  enum pipe_transfer_usage usage,
+		  unsigned x, unsigned y, unsigned w, unsigned h)
+{
+	struct nv04_miptree *mt = (struct nv04_miptree *)pt;
+	struct nv04_transfer *tx;
+	struct pipe_texture tx_tex_template, *tx_tex;
+
+	tx = CALLOC_STRUCT(nv04_transfer);
+	if (!tx)
+		return NULL;
+
+	tx->base.refcount = 1;
+	pipe_texture_reference(&tx->base.texture, pt);
+	tx->base.format = pt->format;
+	tx->base.x = x;
+	tx->base.y = y;
+	tx->base.width = w;
+	tx->base.height = h;
+	tx->base.block = pt->block;
+	tx->base.nblocksx = pt->nblocksx[level];
+	tx->base.nblocksy = pt->nblocksy[level];
+	tx->base.stride = mt->level[level].pitch;
+	tx->base.usage = usage;
+	tx->base.face = face;
+	tx->base.level = level;
+	tx->base.zslice = zslice;
+
+	/* Direct access to texture */
+	if ((pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC ||
+	     debug_get_bool_option("NOUVEAU_NO_TRANSFER", TRUE/*XXX:FALSE*/)) &&
+	    pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)
+	{
+		tx->direct = true;
+		tx->surface = pscreen->get_tex_surface(pscreen, pt,
+	                                               face, level, zslice,
+	                                               nv04_usage_tx_to_buf(usage));
+		return &tx->base;
+	}
+
+	tx->direct = false;
+
+	nv04_compatible_transfer_tex(pt, level, &tx_tex_template);
+
+	tx_tex = pscreen->texture_create(pscreen, &tx_tex_template);
+	if (!tx_tex)
+	{
+		FREE(tx);
+		return NULL;
+	}
+
+	tx->surface = pscreen->get_tex_surface(pscreen, tx_tex,
+	                                       face, level, zslice,
+	                                       nv04_usage_tx_to_buf(usage));
+
+	pipe_texture_reference(&tx_tex, NULL);
+
+	if (!tx->surface)
+	{
+		pipe_surface_reference(&tx->surface, NULL);
+		FREE(tx);
+		return NULL;
+	}
+
+	if (usage != PIPE_TRANSFER_WRITE) {
+		struct nv04_screen *nvscreen = nv04_screen(pscreen);
+		struct pipe_surface *src;
+
+		src = pscreen->get_tex_surface(pscreen, pt,
+	                                       face, level, zslice,
+	                                       PIPE_BUFFER_USAGE_GPU_READ);
+
+		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
+		/* TODO: Check if SIFM can un-swizzle */
+		nvscreen->eng2d->copy(nvscreen->eng2d,
+		                      tx->surface, 0, 0,
+		                      src, 0, 0,
+		                      src->width, src->height);
+
+		pipe_surface_reference(&src, NULL);
+	}
+
+	return &tx->base;
+}
+
+static void
+nv04_transfer_del(struct pipe_screen *pscreen, struct pipe_transfer **pptx)
+{
+	struct pipe_transfer *ptx = *pptx;
+	struct nv04_transfer *tx = (struct nv04_transfer *)ptx;
+
+	if (!tx->direct && ptx->usage != PIPE_TRANSFER_READ) {
+		struct nv04_screen *nvscreen = nv04_screen(pscreen);
+		struct pipe_surface *dst;
+
+		dst = pscreen->get_tex_surface(pscreen, ptx->texture,
+	                                       ptx->face, ptx->level, ptx->zslice,
+	                                       PIPE_BUFFER_USAGE_GPU_WRITE);
+
+		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
+		nvscreen->eng2d->copy(nvscreen->eng2d,
+		                      dst, 0, 0,
+		                      tx->surface, 0, 0,
+		                      dst->width, dst->height);
+
+		pipe_surface_reference(&dst, NULL);
+	}
+
+	*pptx = NULL;
+	if (--ptx->refcount)
+		return;
+
+	pipe_surface_reference(&tx->surface, NULL);
+	pipe_texture_reference(&ptx->texture, NULL);
+	FREE(ptx);
+}
+
+static void *
+nv04_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv04_transfer *tx = (struct nv04_transfer *)ptx;
+	struct nv04_surface *ns = (struct nv04_surface *)tx->surface;
+	struct nv04_miptree *mt = (struct nv04_miptree *)tx->surface->texture;
+	void *map = pipe_buffer_map(pscreen, mt->buffer,
+	                            nv04_usage_tx_to_buf(ptx->usage));
+
+	return map + ns->base.offset +
+	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
+}
+
+static void
+nv04_transfer_unmap(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv04_transfer *tx = (struct nv04_transfer *)ptx;
+	struct nv04_miptree *mt = (struct nv04_miptree *)tx->surface->texture;
+
+	pipe_buffer_unmap(pscreen, mt->buffer);
+}
+
+void
+nv04_screen_init_transfer_functions(struct pipe_screen *pscreen)
+{
+	pscreen->get_tex_transfer = nv04_transfer_new;
+	pscreen->tex_transfer_release = nv04_transfer_del;
+	pscreen->transfer_map = nv04_transfer_map;
+	pscreen->transfer_unmap = nv04_transfer_unmap;
+}
diff --git a/src/gallium/drivers/nv10/Makefile b/src/gallium/drivers/nv10/Makefile
index 2b5fbd4f5a6..62677f5194a 100644
--- a/src/gallium/drivers/nv10/Makefile
+++ b/src/gallium/drivers/nv10/Makefile
@@ -14,6 +14,7 @@ C_SOURCES = \
 	nv10_state.c \
 	nv10_state_emit.c \
 	nv10_surface.c \
+	nv10_transfer.c \
 	nv10_vbo.c
 
 include ../../Makefile.template
diff --git a/src/gallium/drivers/nv10/nv10_miptree.c b/src/gallium/drivers/nv10/nv10_miptree.c
index 96161354617..4747868a508 100644
--- a/src/gallium/drivers/nv10/nv10_miptree.c
+++ b/src/gallium/drivers/nv10/nv10_miptree.c
@@ -131,30 +131,31 @@ nv10_miptree_surface_get(struct pipe_screen *screen, struct pipe_texture *pt,
 			 unsigned face, unsigned level, unsigned zslice,
 			 unsigned flags)
 {
-	struct pipe_winsys *ws = screen->winsys;
 	struct nv10_miptree *nv10mt = (struct nv10_miptree *)pt;
-	struct pipe_surface *ps;
+	struct nv04_surface *ns;
 
-	ps = CALLOC_STRUCT(pipe_surface);
-	if (!ps)
+	ns = CALLOC_STRUCT(nv04_surface);
+	if (!ns)
 		return NULL;
-	pipe_texture_reference(&ps->texture, pt);
-	ps->format = pt->format;
-	ps->width = pt->width[level];
-	ps->height = pt->height[level];
-	ps->block = pt->block;
-	ps->nblocksx = pt->nblocksx[level];
-	ps->nblocksy = pt->nblocksy[level];
-	ps->stride = nv10mt->level[level].pitch;
-	ps->refcount = 1;
+	pipe_texture_reference(&ns->base.texture, pt);
+	ns->base.format = pt->format;
+	ns->base.width = pt->width[level];
+	ns->base.height = pt->height[level];
+	ns->base.usage = flags;
+	ns->base.status = PIPE_SURFACE_STATUS_DEFINED;
+	ns->base.refcount = 1;
+	ns->base.face = face;
+	ns->base.level = level;
+	ns->base.zslice = zslice;
+	ns->pitch = nv10mt->level[level].pitch;
 
 	if (pt->target == PIPE_TEXTURE_CUBE) {
-		ps->offset = nv10mt->level[level].image_offset[face];
+		ns->base.offset = nv10mt->level[level].image_offset[face];
 	} else {
-		ps->offset = nv10mt->level[level].image_offset[0];
+		ns->base.offset = nv10mt->level[level].image_offset[0];
 	}
 
-	return ps;
+	return &ns->base;
 }
 
 static void
diff --git a/src/gallium/drivers/nv10/nv10_screen.c b/src/gallium/drivers/nv10/nv10_screen.c
index f417b06c944..6532a93c7ba 100644
--- a/src/gallium/drivers/nv10/nv10_screen.c
+++ b/src/gallium/drivers/nv10/nv10_screen.c
@@ -116,30 +116,6 @@ nv10_screen_is_format_supported(struct pipe_screen *screen,
 	return FALSE;
 }
 
-static void *
-nv10_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
-		 unsigned flags )
-{
-	struct pipe_winsys *ws = screen->winsys;
-	void *map;
-        struct nv10_miptree *nv10mt = (struct nv10_miptree *)surface->texture;
-
-	map = ws->buffer_map(ws, nv10mt->buffer, flags);
-	if (!map)
-		return NULL;
-
-	return map + surface->offset;
-}
-
-static void
-nv10_surface_unmap(struct pipe_screen *screen, struct pipe_surface *surface)
-{
-	struct pipe_winsys *ws = screen->winsys;
-        struct nv10_miptree *nv10mt = (struct nv10_miptree *)surface->texture;
-
-	ws->buffer_unmap(ws, nv10mt->buffer);
-}
-
 static void
 nv10_screen_destroy(struct pipe_screen *pscreen)
 {
@@ -215,10 +191,8 @@ nv10_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 
 	screen->pipe.is_format_supported = nv10_screen_is_format_supported;
 
-	screen->pipe.surface_map = nv10_surface_map;
-	screen->pipe.surface_unmap = nv10_surface_unmap;
-
 	nv10_screen_init_miptree_functions(&screen->pipe);
+	nv10_screen_init_transfer_functions(&screen->pipe);
 	u_simple_screen_init(&screen->pipe);
 
 	return &screen->pipe;
diff --git a/src/gallium/drivers/nv10/nv10_screen.h b/src/gallium/drivers/nv10/nv10_screen.h
index 60102a369a9..ad829ee3fd4 100644
--- a/src/gallium/drivers/nv10/nv10_screen.h
+++ b/src/gallium/drivers/nv10/nv10_screen.h
@@ -21,4 +21,8 @@ nv10_screen(struct pipe_screen *screen)
 	return (struct nv10_screen *)screen;
 }
 
+
+void
+nv10_screen_init_transfer_functions(struct pipe_screen *pscreen);
+
 #endif
diff --git a/src/gallium/drivers/nv10/nv10_state_emit.c b/src/gallium/drivers/nv10/nv10_state_emit.c
index 5dec618b933..d8691ef9c67 100644
--- a/src/gallium/drivers/nv10/nv10_state_emit.c
+++ b/src/gallium/drivers/nv10/nv10_state_emit.c
@@ -103,7 +103,7 @@ static void nv10_state_emit_scissor(struct nv10_context* nv10)
 static void nv10_state_emit_framebuffer(struct nv10_context* nv10)
 {
 	struct pipe_framebuffer_state* fb = nv10->framebuffer;
-	struct pipe_surface *rt, *zeta = NULL;
+	struct nv04_surface *rt, *zeta = NULL;
 	uint32_t rt_format, w, h;
 	int colour_format = 0, zeta_format = 0;
         struct nv10_miptree *nv10mt = 0;
@@ -111,7 +111,7 @@ static void nv10_state_emit_framebuffer(struct nv10_context* nv10)
 	w = fb->cbufs[0]->width;
 	h = fb->cbufs[0]->height;
 	colour_format = fb->cbufs[0]->format;
-	rt = fb->cbufs[0];
+	rt = (struct nv04_surface *)fb->cbufs[0];
 
 	if (fb->zsbuf) {
 		if (colour_format) {
@@ -123,7 +123,7 @@ static void nv10_state_emit_framebuffer(struct nv10_context* nv10)
 		}
 
 		zeta_format = fb->zsbuf->format;
-		zeta = fb->zsbuf;
+		zeta = (struct nv04_surface *)fb->zsbuf;
 	}
 
 	rt_format = NV10TCL_RT_FORMAT_TYPE_LINEAR;
@@ -142,18 +142,18 @@ static void nv10_state_emit_framebuffer(struct nv10_context* nv10)
 
 	if (zeta) {
 		BEGIN_RING(celsius, NV10TCL_RT_PITCH, 1);
-		OUT_RING  (rt->stride | (zeta->stride << 16));
+		OUT_RING  (rt->pitch | (zeta->pitch << 16));
 	} else {
 		BEGIN_RING(celsius, NV10TCL_RT_PITCH, 1);
-		OUT_RING  (rt->stride | (rt->stride << 16));
+		OUT_RING  (rt->pitch | (rt->pitch << 16));
 	}
 
-	nv10mt = (struct nv10_miptree *)rt->texture;
+	nv10mt = (struct nv10_miptree *)rt->base.texture;
 	nv10->rt[0] = nv10mt->buffer;
 
 	if (zeta_format)
 	{
-		nv10mt = (struct nv10_miptree *)zeta->texture;
+		nv10mt = (struct nv10_miptree *)zeta->base.texture;
 		nv10->zeta = nv10mt->buffer;
 	}
 
diff --git a/src/gallium/drivers/nv10/nv10_transfer.c b/src/gallium/drivers/nv10/nv10_transfer.c
new file mode 100644
index 00000000000..967e2cc20ce
--- /dev/null
+++ b/src/gallium/drivers/nv10/nv10_transfer.c
@@ -0,0 +1,201 @@
+#include <pipe/p_state.h>
+#include <pipe/p_defines.h>
+#include <pipe/p_inlines.h>
+#include <util/u_memory.h>
+#include <nouveau/nouveau_winsys.h>
+#include "nv10_context.h"
+#include "nv10_screen.h"
+#include "nv10_state.h"
+
+struct nv10_transfer {
+	struct pipe_transfer base;
+	struct pipe_surface *surface;
+	bool direct;
+};
+
+static unsigned nv10_usage_tx_to_buf(unsigned tx_usage)
+{
+	switch (tx_usage) {
+		case PIPE_TRANSFER_READ:
+			return PIPE_BUFFER_USAGE_CPU_READ;
+		case PIPE_TRANSFER_WRITE:
+			return PIPE_BUFFER_USAGE_CPU_WRITE;
+		case PIPE_TRANSFER_READ_WRITE:
+			return PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+		default:
+			assert(0);
+	}
+
+	return -1;
+}
+
+static void
+nv10_compatible_transfer_tex(struct pipe_texture *pt, unsigned level,
+                             struct pipe_texture *template)
+{
+	memset(template, 0, sizeof(struct pipe_texture));
+	template->target = pt->target;
+	template->format = pt->format;
+	template->width[0] = pt->width[level];
+	template->height[0] = pt->height[level];
+	template->depth[0] = 1;
+	template->block = pt->block;
+	template->nblocksx[0] = pt->nblocksx[level];
+	template->nblocksy[0] = pt->nblocksx[level];
+	template->last_level = 0;
+	template->compressed = pt->compressed;
+	template->nr_samples = pt->nr_samples;
+
+	template->tex_usage = PIPE_TEXTURE_USAGE_DYNAMIC |
+	                      NOUVEAU_TEXTURE_USAGE_LINEAR;
+}
+
+static struct pipe_transfer *
+nv10_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
+		  unsigned face, unsigned level, unsigned zslice,
+		  enum pipe_transfer_usage usage,
+		  unsigned x, unsigned y, unsigned w, unsigned h)
+{
+	struct nv10_miptree *mt = (struct nv10_miptree *)pt;
+	struct nv10_transfer *tx;
+	struct pipe_texture tx_tex_template, *tx_tex;
+
+	tx = CALLOC_STRUCT(nv10_transfer);
+	if (!tx)
+		return NULL;
+
+	tx->base.refcount = 1;
+	pipe_texture_reference(&tx->base.texture, pt);
+	tx->base.format = pt->format;
+	tx->base.x = x;
+	tx->base.y = y;
+	tx->base.width = w;
+	tx->base.height = h;
+	tx->base.block = pt->block;
+	tx->base.nblocksx = pt->nblocksx[level];
+	tx->base.nblocksy = pt->nblocksy[level];
+	tx->base.stride = mt->level[level].pitch;
+	tx->base.usage = usage;
+	tx->base.face = face;
+	tx->base.level = level;
+	tx->base.zslice = zslice;
+
+	/* Direct access to texture */
+	if ((pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC ||
+	     debug_get_bool_option("NOUVEAU_NO_TRANSFER", TRUE/*XXX:FALSE*/)) &&
+	    pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)
+	{
+		tx->direct = true;
+		tx->surface = pscreen->get_tex_surface(pscreen, pt,
+	                                               face, level, zslice,
+	                                               nv10_usage_tx_to_buf(usage));
+		return &tx->base;
+	}
+
+	tx->direct = false;
+
+	nv10_compatible_transfer_tex(pt, level, &tx_tex_template);
+
+	tx_tex = pscreen->texture_create(pscreen, &tx_tex_template);
+	if (!tx_tex)
+	{
+		FREE(tx);
+		return NULL;
+	}
+
+	tx->surface = pscreen->get_tex_surface(pscreen, tx_tex,
+	                                       face, level, zslice,
+	                                       nv10_usage_tx_to_buf(usage));
+
+	pipe_texture_reference(&tx_tex, NULL);
+
+	if (!tx->surface)
+	{
+		pipe_surface_reference(&tx->surface, NULL);
+		FREE(tx);
+		return NULL;
+	}
+
+	if (usage != PIPE_TRANSFER_WRITE) {
+		struct nv10_screen *nvscreen = nv10_screen(pscreen);
+		struct pipe_surface *src;
+
+		src = pscreen->get_tex_surface(pscreen, pt,
+	                                       face, level, zslice,
+	                                       PIPE_BUFFER_USAGE_GPU_READ);
+
+		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
+		/* TODO: Check if SIFM can un-swizzle */
+		nvscreen->eng2d->copy(nvscreen->eng2d,
+		                      tx->surface, 0, 0,
+		                      src, 0, 0,
+		                      src->width, src->height);
+
+		pipe_surface_reference(&src, NULL);
+	}
+
+	return &tx->base;
+}
+
+static void
+nv10_transfer_del(struct pipe_screen *pscreen, struct pipe_transfer **pptx)
+{
+	struct pipe_transfer *ptx = *pptx;
+	struct nv10_transfer *tx = (struct nv10_transfer *)ptx;
+
+	if (!tx->direct && ptx->usage != PIPE_TRANSFER_READ) {
+		struct nv10_screen *nvscreen = nv10_screen(pscreen);
+		struct pipe_surface *dst;
+
+		dst = pscreen->get_tex_surface(pscreen, ptx->texture,
+	                                       ptx->face, ptx->level, ptx->zslice,
+	                                       PIPE_BUFFER_USAGE_GPU_WRITE);
+
+		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
+		nvscreen->eng2d->copy(nvscreen->eng2d,
+		                      dst, 0, 0,
+		                      tx->surface, 0, 0,
+		                      dst->width, dst->height);
+
+		pipe_surface_reference(&dst, NULL);
+	}
+
+	*pptx = NULL;
+	if (--ptx->refcount)
+		return;
+
+	pipe_surface_reference(&tx->surface, NULL);
+	pipe_texture_reference(&ptx->texture, NULL);
+	FREE(ptx);
+}
+
+static void *
+nv10_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv10_transfer *tx = (struct nv10_transfer *)ptx;
+	struct nv04_surface *ns = (struct nv04_surface *)tx->surface;
+	struct nv10_miptree *mt = (struct nv10_miptree *)tx->surface->texture;
+	void *map = pipe_buffer_map(pscreen, mt->buffer,
+	                            nv10_usage_tx_to_buf(ptx->usage));
+
+	return map + ns->base.offset +
+	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
+}
+
+static void
+nv10_transfer_unmap(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv10_transfer *tx = (struct nv10_transfer *)ptx;
+	struct nv10_miptree *mt = (struct nv10_miptree *)tx->surface->texture;
+
+	pipe_buffer_unmap(pscreen, mt->buffer);
+}
+
+void
+nv10_screen_init_transfer_functions(struct pipe_screen *pscreen)
+{
+	pscreen->get_tex_transfer = nv10_transfer_new;
+	pscreen->tex_transfer_release = nv10_transfer_del;
+	pscreen->transfer_map = nv10_transfer_map;
+	pscreen->transfer_unmap = nv10_transfer_unmap;
+}
diff --git a/src/gallium/drivers/nv20/Makefile b/src/gallium/drivers/nv20/Makefile
index 93e34f8e92a..1305f26c591 100644
--- a/src/gallium/drivers/nv20/Makefile
+++ b/src/gallium/drivers/nv20/Makefile
@@ -14,6 +14,7 @@ C_SOURCES = \
 	nv20_state.c \
 	nv20_state_emit.c \
 	nv20_surface.c \
+	nv20_transfer.c \
 	nv20_vbo.c
 #	nv20_vertprog.c
 
diff --git a/src/gallium/drivers/nv20/nv20_miptree.c b/src/gallium/drivers/nv20/nv20_miptree.c
index ef7e9c5428e..29462408970 100644
--- a/src/gallium/drivers/nv20/nv20_miptree.c
+++ b/src/gallium/drivers/nv20/nv20_miptree.c
@@ -9,10 +9,14 @@ static void
 nv20_miptree_layout(struct nv20_miptree *nv20mt)
 {
 	struct pipe_texture *pt = &nv20mt->base;
-	boolean swizzled = FALSE;
 	uint width = pt->width[0], height = pt->height[0];
 	uint offset = 0;
 	int nr_faces, l, f;
+	uint wide_pitch = pt->tex_usage & (PIPE_TEXTURE_USAGE_SAMPLER |
+		                           PIPE_TEXTURE_USAGE_DEPTH_STENCIL |
+		                           PIPE_TEXTURE_USAGE_RENDER_TARGET |
+		                           PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
+		                           PIPE_TEXTURE_USAGE_PRIMARY);
 
 	if (pt->target == PIPE_TEXTURE_CUBE) {
 		nr_faces = 6;
@@ -26,25 +30,31 @@ nv20_miptree_layout(struct nv20_miptree *nv20mt)
 		pt->nblocksx[l] = pf_get_nblocksx(&pt->block, width);
 		pt->nblocksy[l] = pf_get_nblocksy(&pt->block, height);
 
-		if (swizzled)
-			nv20mt->level[l].pitch = pt->nblocksx[l] * pt->block.size;
+		if (wide_pitch && (pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR))
+			nv20mt->level[l].pitch = align(pt->width[0] * pt->block.size, 64);
 		else
-			nv20mt->level[l].pitch = pt->nblocksx[0] * pt->block.size;
-		nv20mt->level[l].pitch = (nv20mt->level[l].pitch + 63) & ~63;
+			nv20mt->level[l].pitch = pt->width[l] * pt->block.size;
 
 		nv20mt->level[l].image_offset =
 			CALLOC(nr_faces, sizeof(unsigned));
 
 		width  = MAX2(1, width  >> 1);
 		height = MAX2(1, height >> 1);
-
 	}
 
 	for (f = 0; f < nr_faces; f++) {
-		for (l = 0; l <= pt->last_level; l++) {
+		for (l = 0; l < pt->last_level; l++) {
 			nv20mt->level[l].image_offset[f] = offset;
-			offset += nv20mt->level[l].pitch * pt->height[l];
+
+			if (!(pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR) &&
+			    pt->width[l + 1] > 1 && pt->height[l + 1] > 1)
+				offset += align(nv20mt->level[l].pitch * pt->height[l], 64);
+			else
+				offset += nv20mt->level[l].pitch * pt->height[l];
 		}
+
+		nv20mt->level[l].image_offset[f] = offset;
+		offset += nv20mt->level[l].pitch * pt->height[l];
 	}
 
 	nv20mt->total_size = offset;
@@ -96,7 +106,8 @@ nv20_miptree_create(struct pipe_screen *screen, const struct pipe_texture *pt)
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 	else
 	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_PRIMARY |
-	                     PIPE_TEXTURE_USAGE_DISPLAY_TARGET))
+	                     PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
+	                     PIPE_TEXTURE_USAGE_DEPTH_STENCIL))
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 	else
 	if (pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC)
@@ -107,7 +118,11 @@ nv20_miptree_create(struct pipe_screen *screen, const struct pipe_texture *pt)
 		case PIPE_FORMAT_A8R8G8B8_UNORM:
 		case PIPE_FORMAT_X8R8G8B8_UNORM:
 		case PIPE_FORMAT_R16_SNORM:
+		{
+			if (debug_get_bool_option("NOUVEAU_NO_SWIZZLE", FALSE))
+				mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 			break;
+		}
 		default:
 			mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 		}
@@ -152,33 +167,33 @@ nv20_miptree_surface_get(struct pipe_screen *screen, struct pipe_texture *pt,
 			 unsigned flags)
 {
 	struct nv20_miptree *nv20mt = (struct nv20_miptree *)pt;
-	struct pipe_surface *ps;
+	struct nv04_surface *ns;
 
-	ps = CALLOC_STRUCT(pipe_surface);
-	if (!ps)
+	ns = CALLOC_STRUCT(nv04_surface);
+	if (!ns)
 		return NULL;
-	pipe_texture_reference(&ps->texture, pt);
-	ps->format = pt->format;
-	ps->width = pt->width[level];
-	ps->height = pt->height[level];
-	ps->block = pt->block;
-	ps->nblocksx = pt->nblocksx[level];
-	ps->nblocksy = pt->nblocksy[level];
-	ps->stride = nv20mt->level[level].pitch;
-	ps->usage = flags;
-	ps->status = PIPE_SURFACE_STATUS_DEFINED;
-	ps->refcount = 1;
+	pipe_texture_reference(&ns->base.texture, pt);
+	ns->base.format = pt->format;
+	ns->base.width = pt->width[level];
+	ns->base.height = pt->height[level];
+	ns->base.usage = flags;
+	ns->base.status = PIPE_SURFACE_STATUS_DEFINED;
+	ns->base.refcount = 1;
+	ns->base.face = face;
+	ns->base.level = level;
+	ns->base.zslice = zslice;
+	ns->pitch = nv20mt->level[level].pitch;
 
 	if (pt->target == PIPE_TEXTURE_CUBE) {
-		ps->offset = nv20mt->level[level].image_offset[face];
+		ns->base.offset = nv20mt->level[level].image_offset[face];
 	} else
 	if (pt->target == PIPE_TEXTURE_3D) {
-		ps->offset = nv20mt->level[level].image_offset[zslice];
+		ns->base.offset = nv20mt->level[level].image_offset[zslice];
 	} else {
-		ps->offset = nv20mt->level[level].image_offset[0];
+		ns->base.offset = nv20mt->level[level].image_offset[0];
 	}
 
-	return ps;
+	return &ns->base;
 }
 
 static void
diff --git a/src/gallium/drivers/nv20/nv20_screen.c b/src/gallium/drivers/nv20/nv20_screen.c
index 5f2b7b4f71f..7760ae27c0e 100644
--- a/src/gallium/drivers/nv20/nv20_screen.c
+++ b/src/gallium/drivers/nv20/nv20_screen.c
@@ -116,30 +116,6 @@ nv20_screen_is_format_supported(struct pipe_screen *screen,
 	return FALSE;
 }
 
-static void *
-nv20_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
-		 unsigned flags )
-{
-	struct pipe_winsys *ws = screen->winsys;
-	void *map;
-	struct nv20_miptree *nv20mt = (struct nv20_miptree *)surface->texture;
-
-	map = ws->buffer_map(ws, nv20mt->buffer, flags);
-	if (!map)
-		return NULL;
-
-	return map + surface->offset;
-}
-
-static void
-nv20_surface_unmap(struct pipe_screen *screen, struct pipe_surface *surface)
-{
-	struct pipe_winsys *ws = screen->winsys;
-	struct nv20_miptree *nv20mt = (struct nv20_miptree *)surface->texture;
-
-	ws->buffer_unmap(ws, nv20mt->buffer);
-}
-
 static void
 nv20_screen_destroy(struct pipe_screen *pscreen)
 {
@@ -211,10 +187,8 @@ nv20_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 
 	screen->pipe.is_format_supported = nv20_screen_is_format_supported;
 
-	screen->pipe.surface_map = nv20_surface_map;
-	screen->pipe.surface_unmap = nv20_surface_unmap;
-
 	nv20_screen_init_miptree_functions(&screen->pipe);
+	nv20_screen_init_transfer_functions(&screen->pipe);
 	u_simple_screen_init(&screen->pipe);
 
 	return &screen->pipe;
diff --git a/src/gallium/drivers/nv20/nv20_screen.h b/src/gallium/drivers/nv20/nv20_screen.h
index bf2f2c0d9fb..d9fce2bced8 100644
--- a/src/gallium/drivers/nv20/nv20_screen.h
+++ b/src/gallium/drivers/nv20/nv20_screen.h
@@ -21,4 +21,8 @@ nv20_screen(struct pipe_screen *screen)
 	return (struct nv20_screen *)screen;
 }
 
+
+void
+nv20_screen_init_transfer_functions(struct pipe_screen *pscreen);
+
 #endif
diff --git a/src/gallium/drivers/nv20/nv20_state_emit.c b/src/gallium/drivers/nv20/nv20_state_emit.c
index 0f4df9ca317..4042f46d053 100644
--- a/src/gallium/drivers/nv20/nv20_state_emit.c
+++ b/src/gallium/drivers/nv20/nv20_state_emit.c
@@ -109,7 +109,7 @@ static void nv20_state_emit_scissor(struct nv20_context* nv20)
 static void nv20_state_emit_framebuffer(struct nv20_context* nv20)
 {
 	struct pipe_framebuffer_state* fb = nv20->framebuffer;
-	struct pipe_surface *rt, *zeta = NULL;
+	struct nv04_surface *rt, *zeta = NULL;
 	uint32_t rt_format, w, h;
 	int colour_format = 0, zeta_format = 0;
 	struct nv20_miptree *nv20mt = 0;
@@ -117,7 +117,7 @@ static void nv20_state_emit_framebuffer(struct nv20_context* nv20)
 	w = fb->cbufs[0]->width;
 	h = fb->cbufs[0]->height;
 	colour_format = fb->cbufs[0]->format;
-	rt = fb->cbufs[0];
+	rt = (struct nv04_surface *)fb->cbufs[0];
 
 	if (fb->zsbuf) {
 		if (colour_format) {
@@ -129,7 +129,7 @@ static void nv20_state_emit_framebuffer(struct nv20_context* nv20)
 		}
 
 		zeta_format = fb->zsbuf->format;
-		zeta = fb->zsbuf;
+		zeta = (struct nv04_surface *)fb->zsbuf;
 	}
 
 	rt_format = NV20TCL_RT_FORMAT_TYPE_LINEAR | 0x20;
@@ -148,18 +148,18 @@ static void nv20_state_emit_framebuffer(struct nv20_context* nv20)
 
 	if (zeta) {
 		BEGIN_RING(kelvin, NV20TCL_RT_PITCH, 1);
-		OUT_RING  (rt->stride | (zeta->stride << 16));
+		OUT_RING  (rt->pitch | (zeta->pitch << 16));
 	} else {
 		BEGIN_RING(kelvin, NV20TCL_RT_PITCH, 1);
-		OUT_RING  (rt->stride | (rt->stride << 16));
+		OUT_RING  (rt->pitch | (rt->pitch << 16));
 	}
 
-	nv20mt = (struct nv20_miptree *)rt->texture;
+	nv20mt = (struct nv20_miptree *)rt->base.texture;
 	nv20->rt[0] = nv20mt->buffer;
 
 	if (zeta_format)
 	{
-		nv20mt = (struct nv20_miptree *)zeta->texture;
+		nv20mt = (struct nv20_miptree *)zeta->base.texture;
 		nv20->zeta = nv20mt->buffer;
 	}
 
diff --git a/src/gallium/drivers/nv20/nv20_transfer.c b/src/gallium/drivers/nv20/nv20_transfer.c
new file mode 100644
index 00000000000..19de09486d3
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_transfer.c
@@ -0,0 +1,201 @@
+#include <pipe/p_state.h>
+#include <pipe/p_defines.h>
+#include <pipe/p_inlines.h>
+#include <util/u_memory.h>
+#include <nouveau/nouveau_winsys.h>
+#include "nv20_context.h"
+#include "nv20_screen.h"
+#include "nv20_state.h"
+
+struct nv20_transfer {
+	struct pipe_transfer base;
+	struct pipe_surface *surface;
+	bool direct;
+};
+
+static unsigned nv20_usage_tx_to_buf(unsigned tx_usage)
+{
+	switch (tx_usage) {
+		case PIPE_TRANSFER_READ:
+			return PIPE_BUFFER_USAGE_CPU_READ;
+		case PIPE_TRANSFER_WRITE:
+			return PIPE_BUFFER_USAGE_CPU_WRITE;
+		case PIPE_TRANSFER_READ_WRITE:
+			return PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+		default:
+			assert(0);
+	}
+
+	return -1;
+}
+
+static void
+nv20_compatible_transfer_tex(struct pipe_texture *pt, unsigned level,
+                             struct pipe_texture *template)
+{
+	memset(template, 0, sizeof(struct pipe_texture));
+	template->target = pt->target;
+	template->format = pt->format;
+	template->width[0] = pt->width[level];
+	template->height[0] = pt->height[level];
+	template->depth[0] = 1;
+	template->block = pt->block;
+	template->nblocksx[0] = pt->nblocksx[level];
+	template->nblocksy[0] = pt->nblocksx[level];
+	template->last_level = 0;
+	template->compressed = pt->compressed;
+	template->nr_samples = pt->nr_samples;
+
+	template->tex_usage = PIPE_TEXTURE_USAGE_DYNAMIC |
+	                      NOUVEAU_TEXTURE_USAGE_LINEAR;
+}
+
+static struct pipe_transfer *
+nv20_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
+		  unsigned face, unsigned level, unsigned zslice,
+		  enum pipe_transfer_usage usage,
+		  unsigned x, unsigned y, unsigned w, unsigned h)
+{
+	struct nv20_miptree *mt = (struct nv20_miptree *)pt;
+	struct nv20_transfer *tx;
+	struct pipe_texture tx_tex_template, *tx_tex;
+
+	tx = CALLOC_STRUCT(nv20_transfer);
+	if (!tx)
+		return NULL;
+
+	tx->base.refcount = 1;
+	pipe_texture_reference(&tx->base.texture, pt);
+	tx->base.format = pt->format;
+	tx->base.x = x;
+	tx->base.y = y;
+	tx->base.width = w;
+	tx->base.height = h;
+	tx->base.block = pt->block;
+	tx->base.nblocksx = pt->nblocksx[level];
+	tx->base.nblocksy = pt->nblocksy[level];
+	tx->base.stride = mt->level[level].pitch;
+	tx->base.usage = usage;
+	tx->base.face = face;
+	tx->base.level = level;
+	tx->base.zslice = zslice;
+
+	/* Direct access to texture */
+	if ((pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC ||
+	     debug_get_bool_option("NOUVEAU_NO_TRANSFER", TRUE/*XXX:FALSE*/)) &&
+	    pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)
+	{
+		tx->direct = true;
+		tx->surface = pscreen->get_tex_surface(pscreen, pt,
+	                                               face, level, zslice,
+	                                               nv20_usage_tx_to_buf(usage));
+		return &tx->base;
+	}
+
+	tx->direct = false;
+
+	nv20_compatible_transfer_tex(pt, level, &tx_tex_template);
+
+	tx_tex = pscreen->texture_create(pscreen, &tx_tex_template);
+	if (!tx_tex)
+	{
+		FREE(tx);
+		return NULL;
+	}
+
+	tx->surface = pscreen->get_tex_surface(pscreen, tx_tex,
+	                                       face, level, zslice,
+	                                       nv20_usage_tx_to_buf(usage));
+
+	pipe_texture_reference(&tx_tex, NULL);
+
+	if (!tx->surface)
+	{
+		pipe_surface_reference(&tx->surface, NULL);
+		FREE(tx);
+		return NULL;
+	}
+
+	if (usage != PIPE_TRANSFER_WRITE) {
+		struct nv20_screen *nvscreen = nv20_screen(pscreen);
+		struct pipe_surface *src;
+
+		src = pscreen->get_tex_surface(pscreen, pt,
+	                                       face, level, zslice,
+	                                       PIPE_BUFFER_USAGE_GPU_READ);
+
+		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
+		/* TODO: Check if SIFM can un-swizzle */
+		nvscreen->eng2d->copy(nvscreen->eng2d,
+		                      tx->surface, 0, 0,
+		                      src, 0, 0,
+		                      src->width, src->height);
+
+		pipe_surface_reference(&src, NULL);
+	}
+
+	return &tx->base;
+}
+
+static void
+nv20_transfer_del(struct pipe_screen *pscreen, struct pipe_transfer **pptx)
+{
+	struct pipe_transfer *ptx = *pptx;
+	struct nv20_transfer *tx = (struct nv20_transfer *)ptx;
+
+	if (!tx->direct && ptx->usage != PIPE_TRANSFER_READ) {
+		struct nv20_screen *nvscreen = nv20_screen(pscreen);
+		struct pipe_surface *dst;
+
+		dst = pscreen->get_tex_surface(pscreen, ptx->texture,
+	                                       ptx->face, ptx->level, ptx->zslice,
+	                                       PIPE_BUFFER_USAGE_GPU_WRITE);
+
+		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
+		nvscreen->eng2d->copy(nvscreen->eng2d,
+		                      dst, 0, 0,
+		                      tx->surface, 0, 0,
+		                      dst->width, dst->height);
+
+		pipe_surface_reference(&dst, NULL);
+	}
+
+	*pptx = NULL;
+	if (--ptx->refcount)
+		return;
+
+	pipe_surface_reference(&tx->surface, NULL);
+	pipe_texture_reference(&ptx->texture, NULL);
+	FREE(ptx);
+}
+
+static void *
+nv20_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv20_transfer *tx = (struct nv20_transfer *)ptx;
+	struct nv04_surface *ns = (struct nv04_surface *)tx->surface;
+	struct nv20_miptree *mt = (struct nv20_miptree *)tx->surface->texture;
+	void *map = pipe_buffer_map(pscreen, mt->buffer,
+	                            nv20_usage_tx_to_buf(ptx->usage));
+
+	return map + ns->base.offset +
+	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
+}
+
+static void
+nv20_transfer_unmap(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv20_transfer *tx = (struct nv20_transfer *)ptx;
+	struct nv20_miptree *mt = (struct nv20_miptree *)tx->surface->texture;
+
+	pipe_buffer_unmap(pscreen, mt->buffer);
+}
+
+void
+nv20_screen_init_transfer_functions(struct pipe_screen *pscreen)
+{
+	pscreen->get_tex_transfer = nv20_transfer_new;
+	pscreen->tex_transfer_release = nv20_transfer_del;
+	pscreen->transfer_map = nv20_transfer_map;
+	pscreen->transfer_unmap = nv20_transfer_unmap;
+}
diff --git a/src/gallium/drivers/nv30/Makefile b/src/gallium/drivers/nv30/Makefile
index 4c29e2eab31..364c80d8f3e 100644
--- a/src/gallium/drivers/nv30/Makefile
+++ b/src/gallium/drivers/nv30/Makefile
@@ -22,6 +22,7 @@ C_SOURCES = \
 	nv30_state_viewport.c \
 	nv30_state_zsa.c \
 	nv30_surface.c \
+	nv30_transfer.c \
 	nv30_vbo.c \
 	nv30_vertprog.c
 
diff --git a/src/gallium/drivers/nv30/nv30_miptree.c b/src/gallium/drivers/nv30/nv30_miptree.c
index b11ed8c24e2..ec0a8b8438c 100644
--- a/src/gallium/drivers/nv30/nv30_miptree.c
+++ b/src/gallium/drivers/nv30/nv30_miptree.c
@@ -69,6 +69,8 @@ nv30_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 {
 	struct pipe_winsys *ws = pscreen->winsys;
 	struct nv30_miptree *mt;
+	unsigned buf_usage = PIPE_BUFFER_USAGE_PIXEL |
+	                     NOUVEAU_BUFFER_USAGE_TEXTURE;
 
 	mt = MALLOC(sizeof(struct nv30_miptree));
 	if (!mt)
@@ -76,8 +78,6 @@ nv30_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	mt->base = *pt;
 	mt->base.refcount = 1;
 	mt->base.screen = pscreen;
-	mt->shadow_tex = NULL;
-	mt->shadow_surface = NULL;
 
 	/* Swizzled textures must be POT */
 	if (pt->width[0] & (pt->width[0] - 1) ||
@@ -107,11 +107,12 @@ nv30_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 		}
 	}
 
+	if (pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC)
+		buf_usage |= PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+
 	nv30_miptree_layout(mt);
 
-	mt->buffer = ws->buffer_create(ws, 256,
-				       PIPE_BUFFER_USAGE_PIXEL |
-				       NOUVEAU_BUFFER_USAGE_TEXTURE,
+	mt->buffer = ws->buffer_create(ws, 256, buf_usage,
 				       mt->total_size);
 	if (!mt->buffer) {
 		FREE(mt);
@@ -163,12 +164,6 @@ nv30_miptree_release(struct pipe_screen *pscreen, struct pipe_texture **ppt)
 			FREE(mt->level[l].image_offset);
 	}
 
-	if (mt->shadow_tex) {
-		if (mt->shadow_surface)
-			pscreen->tex_surface_release(pscreen, &mt->shadow_surface);
-		nv30_miptree_release(pscreen, &mt->shadow_tex);
-	}
-
 	FREE(mt);
 }
 
@@ -178,36 +173,33 @@ nv30_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 			 unsigned flags)
 {
 	struct nv30_miptree *nv30mt = (struct nv30_miptree *)pt;
-	struct pipe_surface *ps;
+	struct nv04_surface *ns;
 
-	ps = CALLOC_STRUCT(pipe_surface);
-	if (!ps)
+	ns = CALLOC_STRUCT(nv04_surface);
+	if (!ns)
 		return NULL;
-	pipe_texture_reference(&ps->texture, pt);
-	ps->format = pt->format;
-	ps->width = pt->width[level];
-	ps->height = pt->height[level];
-	ps->block = pt->block;
-	ps->nblocksx = pt->nblocksx[level];
-	ps->nblocksy = pt->nblocksy[level];
-	ps->stride = nv30mt->level[level].pitch;
-	ps->usage = flags;
-	ps->status = PIPE_SURFACE_STATUS_DEFINED;
-	ps->refcount = 1;
-	ps->face = face;
-	ps->level = level;
-	ps->zslice = zslice;
+	pipe_texture_reference(&ns->base.texture, pt);
+	ns->base.format = pt->format;
+	ns->base.width = pt->width[level];
+	ns->base.height = pt->height[level];
+	ns->base.usage = flags;
+	ns->base.status = PIPE_SURFACE_STATUS_DEFINED;
+	ns->base.refcount = 1;
+	ns->base.face = face;
+	ns->base.level = level;
+	ns->base.zslice = zslice;
+	ns->pitch = nv30mt->level[level].pitch;
 
 	if (pt->target == PIPE_TEXTURE_CUBE) {
-		ps->offset = nv30mt->level[level].image_offset[face];
+		ns->base.offset = nv30mt->level[level].image_offset[face];
 	} else
 	if (pt->target == PIPE_TEXTURE_3D) {
-		ps->offset = nv30mt->level[level].image_offset[zslice];
+		ns->base.offset = nv30mt->level[level].image_offset[zslice];
 	} else {
-		ps->offset = nv30mt->level[level].image_offset[0];
+		ns->base.offset = nv30mt->level[level].image_offset[0];
 	}
 
-	return ps;
+	return &ns->base;
 }
 
 static void
diff --git a/src/gallium/drivers/nv30/nv30_screen.c b/src/gallium/drivers/nv30/nv30_screen.c
index c97a73f0b1a..d395c5e1b7f 100644
--- a/src/gallium/drivers/nv30/nv30_screen.c
+++ b/src/gallium/drivers/nv30/nv30_screen.c
@@ -135,82 +135,6 @@ nv30_surface_buffer(struct pipe_surface *surf)
 	return mt->buffer;
 }
 
-static void *
-nv30_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
-		 unsigned flags )
-{
-	struct pipe_winsys	*ws = screen->winsys;
-	struct pipe_surface	*surface_to_map;
-	void			*map;
-
-	if (!(surface->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
-		struct nv30_miptree *mt = (struct nv30_miptree *)surface->texture;
-
-		if (!mt->shadow_tex) {
-			unsigned old_tex_usage = surface->texture->tex_usage;
-			surface->texture->tex_usage = NOUVEAU_TEXTURE_USAGE_LINEAR |
-			                              PIPE_TEXTURE_USAGE_DYNAMIC;
-			mt->shadow_tex = screen->texture_create(screen, surface->texture);
-			surface->texture->tex_usage = old_tex_usage;
-
-			assert(mt->shadow_tex->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR);
-		}
-
-		mt->shadow_surface = screen->get_tex_surface
-		(
-			screen, mt->shadow_tex,
-			surface->face, surface->level, surface->zslice,
-			surface->usage
-		);
-
-		surface_to_map = mt->shadow_surface;
-	}
-	else
-		surface_to_map = surface;
-
-	assert(surface_to_map);
-
-	map = ws->buffer_map(ws, nv30_surface_buffer(surface_to_map), flags);
-	if (!map)
-		return NULL;
-
-	return map + surface_to_map->offset;
-}
-
-static void
-nv30_surface_unmap(struct pipe_screen *screen, struct pipe_surface *surface)
-{
-	struct pipe_winsys	*ws = screen->winsys;
-	struct pipe_surface	*surface_to_unmap;
-
-	/* TODO: Copy from shadow just before push buffer is flushed instead.
-	         There are probably some programs that map/unmap excessively
-	         before rendering. */
-	if (!(surface->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
-		struct nv30_miptree *mt = (struct nv30_miptree *)surface->texture;
-
-		assert(mt->shadow_tex);
-
-		surface_to_unmap = mt->shadow_surface;
-	}
-	else
-		surface_to_unmap = surface;
-
-	assert(surface_to_unmap);
-
-	ws->buffer_unmap(ws, nv30_surface_buffer(surface_to_unmap));
-
-	if (surface_to_unmap != surface) {
-		struct nv30_screen *nvscreen = nv30_screen(screen);
-
-		nvscreen->eng2d->copy(nvscreen->eng2d, surface, 0, 0,
-		                      surface_to_unmap, 0, 0,
-		                      surface->width, surface->height);
-
-		screen->tex_surface_release(screen, &surface_to_unmap);
-	}
-}
-
 static void
 nv30_screen_destroy(struct pipe_screen *pscreen)
 {
@@ -391,10 +315,8 @@ nv30_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 
 	screen->pipe.is_format_supported = nv30_screen_surface_format_supported;
 
-	screen->pipe.surface_map = nv30_surface_map;
-	screen->pipe.surface_unmap = nv30_surface_unmap;
-
 	nv30_screen_init_miptree_functions(&screen->pipe);
+	nv30_screen_init_transfer_functions(&screen->pipe);
 	u_simple_screen_init(&screen->pipe);
 
 	return &screen->pipe;
diff --git a/src/gallium/drivers/nv30/nv30_screen.h b/src/gallium/drivers/nv30/nv30_screen.h
index b11e470f949..8e368839759 100644
--- a/src/gallium/drivers/nv30/nv30_screen.h
+++ b/src/gallium/drivers/nv30/nv30_screen.h
@@ -34,4 +34,7 @@ nv30_screen(struct pipe_screen *screen)
 	return (struct nv30_screen *)screen;
 }
 
+void
+nv30_screen_init_transfer_functions(struct pipe_screen *pscreen);
+
 #endif
diff --git a/src/gallium/drivers/nv30/nv30_state.h b/src/gallium/drivers/nv30/nv30_state.h
index 2023278e377..e6f23bf1667 100644
--- a/src/gallium/drivers/nv30/nv30_state.h
+++ b/src/gallium/drivers/nv30/nv30_state.h
@@ -76,9 +76,6 @@ struct nv30_miptree {
 	struct pipe_buffer *buffer;
 	uint total_size;
 
-	struct pipe_texture *shadow_tex;
-	struct pipe_surface *shadow_surface;
-
 	struct {
 		uint pitch;
 		uint *image_offset;
diff --git a/src/gallium/drivers/nv30/nv30_state_fb.c b/src/gallium/drivers/nv30/nv30_state_fb.c
index 77368cb2055..ee0326011ca 100644
--- a/src/gallium/drivers/nv30/nv30_state_fb.c
+++ b/src/gallium/drivers/nv30/nv30_state_fb.c
@@ -5,7 +5,7 @@ static boolean
 nv30_state_framebuffer_validate(struct nv30_context *nv30)
 {
 	struct pipe_framebuffer_state *fb = &nv30->framebuffer;
-	struct pipe_surface *rt[2], *zeta = NULL;
+	struct nv04_surface *rt[2], *zeta = NULL;
 	uint32_t rt_enable, rt_format;
 	int i, colour_format = 0, zeta_format = 0;
 	struct nouveau_stateobj *so = so_new(64, 10);
@@ -21,7 +21,7 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30)
 		} else {
 			colour_format = fb->cbufs[i]->format;
 			rt_enable |= (NV34TCL_RT_ENABLE_COLOR0 << i);
-			rt[i] = fb->cbufs[i];
+			rt[i] = (struct nv04_surface *)fb->cbufs[i];
 		}
 	}
 
@@ -30,13 +30,13 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30)
 
 	if (fb->zsbuf) {
 		zeta_format = fb->zsbuf->format;
-		zeta = fb->zsbuf;
+		zeta = (struct nv04_surface *)fb->zsbuf;
 	}
 
-	if (!(rt[0]->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
+	if (!(rt[0]->base.texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
 		assert(!(fb->width & (fb->width - 1)) && !(fb->height & (fb->height - 1)));
 		for (i = 1; i < fb->nr_cbufs; i++)
-			assert(!(rt[i]->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR));
+			assert(!(rt[i]->base.texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR));
 
 		/* FIXME: NV34TCL_RT_FORMAT_LOG2_[WIDTH/HEIGHT] */
 		rt_format = NV34TCL_RT_FORMAT_TYPE_SWIZZLED |
@@ -71,44 +71,44 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30)
 	}
 
 	if (rt_enable & NV34TCL_RT_ENABLE_COLOR0) {
-		uint32_t pitch = rt[0]->stride;
+		uint32_t pitch = rt[0]->pitch;
 		if (zeta) {
-			pitch |= (zeta->stride << 16);
+			pitch |= (zeta->pitch << 16);
 		} else {
 			pitch |= (pitch << 16);
 		}
 
-		nv30mt = (struct nv30_miptree *)rt[0]->texture;
+		nv30mt = (struct nv30_miptree *)rt[0]->base.texture;
 		so_method(so, nv30->screen->rankine, NV34TCL_DMA_COLOR0, 1);
 		so_reloc (so, nv30mt->buffer, 0, rt_flags | NOUVEAU_BO_OR,
 			  nv30->nvws->channel->vram->handle,
 			  nv30->nvws->channel->gart->handle);
 		so_method(so, nv30->screen->rankine, NV34TCL_COLOR0_PITCH, 2);
 		so_data  (so, pitch);
-		so_reloc (so, nv30mt->buffer, rt[0]->offset, rt_flags |
+		so_reloc (so, nv30mt->buffer, rt[0]->base.offset, rt_flags |
 			  NOUVEAU_BO_LOW, 0, 0);
 	}
 
 	if (rt_enable & NV34TCL_RT_ENABLE_COLOR1) {
-		nv30mt = (struct nv30_miptree *)rt[1]->texture;
+		nv30mt = (struct nv30_miptree *)rt[1]->base.texture;
 		so_method(so, nv30->screen->rankine, NV34TCL_DMA_COLOR1, 1);
 		so_reloc (so, nv30mt->buffer, 0, rt_flags | NOUVEAU_BO_OR,
 			  nv30->nvws->channel->vram->handle,
 			  nv30->nvws->channel->gart->handle);
 		so_method(so, nv30->screen->rankine, NV34TCL_COLOR1_OFFSET, 2);
-		so_reloc (so, nv30mt->buffer, rt[1]->offset, rt_flags |
+		so_reloc (so, nv30mt->buffer, rt[1]->base.offset, rt_flags |
 			  NOUVEAU_BO_LOW, 0, 0);
-		so_data  (so, rt[1]->stride);
+		so_data  (so, rt[1]->pitch);
 	}
 
 	if (zeta_format) {
-		nv30mt = (struct nv30_miptree *)zeta->texture;
+		nv30mt = (struct nv30_miptree *)zeta->base.texture;
 		so_method(so, nv30->screen->rankine, NV34TCL_DMA_ZETA, 1);
 		so_reloc (so, nv30mt->buffer, 0, rt_flags | NOUVEAU_BO_OR,
 			  nv30->nvws->channel->vram->handle,
 			  nv30->nvws->channel->gart->handle);
 		so_method(so, nv30->screen->rankine, NV34TCL_ZETA_OFFSET, 1);
-		so_reloc (so, nv30mt->buffer, zeta->offset, rt_flags |
+		so_reloc (so, nv30mt->buffer, zeta->base.offset, rt_flags |
 			  NOUVEAU_BO_LOW, 0, 0);
 		/* TODO: allocate LMA depth buffer */
 	}
diff --git a/src/gallium/drivers/nv30/nv30_transfer.c b/src/gallium/drivers/nv30/nv30_transfer.c
new file mode 100644
index 00000000000..df4dc4b1f68
--- /dev/null
+++ b/src/gallium/drivers/nv30/nv30_transfer.c
@@ -0,0 +1,201 @@
+#include <pipe/p_state.h>
+#include <pipe/p_defines.h>
+#include <pipe/p_inlines.h>
+#include <util/u_memory.h>
+#include <nouveau/nouveau_winsys.h>
+#include "nv30_context.h"
+#include "nv30_screen.h"
+#include "nv30_state.h"
+
+struct nv30_transfer {
+	struct pipe_transfer base;
+	struct pipe_surface *surface;
+	bool direct;
+};
+
+static unsigned nv30_usage_tx_to_buf(unsigned tx_usage)
+{
+	switch (tx_usage) {
+		case PIPE_TRANSFER_READ:
+			return PIPE_BUFFER_USAGE_CPU_READ;
+		case PIPE_TRANSFER_WRITE:
+			return PIPE_BUFFER_USAGE_CPU_WRITE;
+		case PIPE_TRANSFER_READ_WRITE:
+			return PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+		default:
+			assert(0);
+	}
+
+	return -1;
+}
+
+static void
+nv30_compatible_transfer_tex(struct pipe_texture *pt, unsigned level,
+                             struct pipe_texture *template)
+{
+	memset(template, 0, sizeof(struct pipe_texture));
+	template->target = pt->target;
+	template->format = pt->format;
+	template->width[0] = pt->width[level];
+	template->height[0] = pt->height[level];
+	template->depth[0] = 1;
+	template->block = pt->block;
+	template->nblocksx[0] = pt->nblocksx[level];
+	template->nblocksy[0] = pt->nblocksx[level];
+	template->last_level = 0;
+	template->compressed = pt->compressed;
+	template->nr_samples = pt->nr_samples;
+
+	template->tex_usage = PIPE_TEXTURE_USAGE_DYNAMIC |
+	                      NOUVEAU_TEXTURE_USAGE_LINEAR;
+}
+
+static struct pipe_transfer *
+nv30_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
+		  unsigned face, unsigned level, unsigned zslice,
+		  enum pipe_transfer_usage usage,
+		  unsigned x, unsigned y, unsigned w, unsigned h)
+{
+	struct nv30_miptree *mt = (struct nv30_miptree *)pt;
+	struct nv30_transfer *tx;
+	struct pipe_texture tx_tex_template, *tx_tex;
+
+	tx = CALLOC_STRUCT(nv30_transfer);
+	if (!tx)
+		return NULL;
+
+	tx->base.refcount = 1;
+	pipe_texture_reference(&tx->base.texture, pt);
+	tx->base.format = pt->format;
+	tx->base.x = x;
+	tx->base.y = y;
+	tx->base.width = w;
+	tx->base.height = h;
+	tx->base.block = pt->block;
+	tx->base.nblocksx = pt->nblocksx[level];
+	tx->base.nblocksy = pt->nblocksy[level];
+	tx->base.stride = mt->level[level].pitch;
+	tx->base.usage = usage;
+	tx->base.face = face;
+	tx->base.level = level;
+	tx->base.zslice = zslice;
+
+	/* Direct access to texture */
+	if ((pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC ||
+	     debug_get_bool_option("NOUVEAU_NO_TRANSFER", TRUE/*XXX:FALSE*/)) &&
+	    pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)
+	{
+		tx->direct = true;
+		tx->surface = pscreen->get_tex_surface(pscreen, pt,
+	                                               face, level, zslice,
+	                                               nv30_usage_tx_to_buf(usage));
+		return &tx->base;
+	}
+
+	tx->direct = false;
+
+	nv30_compatible_transfer_tex(pt, level, &tx_tex_template);
+
+	tx_tex = pscreen->texture_create(pscreen, &tx_tex_template);
+	if (!tx_tex)
+	{
+		FREE(tx);
+		return NULL;
+	}
+
+	tx->surface = pscreen->get_tex_surface(pscreen, tx_tex,
+	                                       face, level, zslice,
+	                                       nv30_usage_tx_to_buf(usage));
+
+	pipe_texture_reference(&tx_tex, NULL);
+
+	if (!tx->surface)
+	{
+		pipe_surface_reference(&tx->surface, NULL);
+		FREE(tx);
+		return NULL;
+	}
+
+	if (usage != PIPE_TRANSFER_WRITE) {
+		struct nv30_screen *nvscreen = nv30_screen(pscreen);
+		struct pipe_surface *src;
+
+		src = pscreen->get_tex_surface(pscreen, pt,
+	                                       face, level, zslice,
+	                                       PIPE_BUFFER_USAGE_GPU_READ);
+
+		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
+		/* TODO: Check if SIFM can un-swizzle */
+		nvscreen->eng2d->copy(nvscreen->eng2d,
+		                      tx->surface, 0, 0,
+		                      src, 0, 0,
+		                      src->width, src->height);
+
+		pipe_surface_reference(&src, NULL);
+	}
+
+	return &tx->base;
+}
+
+static void
+nv30_transfer_del(struct pipe_screen *pscreen, struct pipe_transfer **pptx)
+{
+	struct pipe_transfer *ptx = *pptx;
+	struct nv30_transfer *tx = (struct nv30_transfer *)ptx;
+
+	if (!tx->direct && ptx->usage != PIPE_TRANSFER_READ) {
+		struct nv30_screen *nvscreen = nv30_screen(pscreen);
+		struct pipe_surface *dst;
+
+		dst = pscreen->get_tex_surface(pscreen, ptx->texture,
+	                                       ptx->face, ptx->level, ptx->zslice,
+	                                       PIPE_BUFFER_USAGE_GPU_WRITE);
+
+		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
+		nvscreen->eng2d->copy(nvscreen->eng2d,
+		                      dst, 0, 0,
+		                      tx->surface, 0, 0,
+		                      dst->width, dst->height);
+
+		pipe_surface_reference(&dst, NULL);
+	}
+
+	*pptx = NULL;
+	if (--ptx->refcount)
+		return;
+
+	pipe_surface_reference(&tx->surface, NULL);
+	pipe_texture_reference(&ptx->texture, NULL);
+	FREE(ptx);
+}
+
+static void *
+nv30_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv30_transfer *tx = (struct nv30_transfer *)ptx;
+	struct nv04_surface *ns = (struct nv04_surface *)tx->surface;
+	struct nv30_miptree *mt = (struct nv30_miptree *)tx->surface->texture;
+	void *map = pipe_buffer_map(pscreen, mt->buffer,
+	                            nv30_usage_tx_to_buf(ptx->usage));
+
+	return map + ns->base.offset +
+	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
+}
+
+static void
+nv30_transfer_unmap(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv30_transfer *tx = (struct nv30_transfer *)ptx;
+	struct nv30_miptree *mt = (struct nv30_miptree *)tx->surface->texture;
+
+	pipe_buffer_unmap(pscreen, mt->buffer);
+}
+
+void
+nv30_screen_init_transfer_functions(struct pipe_screen *pscreen)
+{
+	pscreen->get_tex_transfer = nv30_transfer_new;
+	pscreen->tex_transfer_release = nv30_transfer_del;
+	pscreen->transfer_map = nv30_transfer_map;
+	pscreen->transfer_unmap = nv30_transfer_unmap;
+}
diff --git a/src/gallium/drivers/nv40/Makefile b/src/gallium/drivers/nv40/Makefile
index 8c738aefa6a..0ecae2b4913 100644
--- a/src/gallium/drivers/nv40/Makefile
+++ b/src/gallium/drivers/nv40/Makefile
@@ -22,6 +22,7 @@ C_SOURCES = \
 	nv40_state_viewport.c \
 	nv40_state_zsa.c \
 	nv40_surface.c \
+	nv40_transfer.c \
 	nv40_vbo.c \
 	nv40_vertprog.c
 
diff --git a/src/gallium/drivers/nv40/nv40_miptree.c b/src/gallium/drivers/nv40/nv40_miptree.c
index e38b1e7f5ca..638d279aa5a 100644
--- a/src/gallium/drivers/nv40/nv40_miptree.c
+++ b/src/gallium/drivers/nv40/nv40_miptree.c
@@ -78,8 +78,6 @@ nv40_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	mt->base = *pt;
 	mt->base.refcount = 1;
 	mt->base.screen = pscreen;
-	mt->shadow_tex = NULL;
-	mt->shadow_surface = NULL;
 
 	/* Swizzled textures must be POT */
 	if (pt->width[0] & (pt->width[0] - 1) ||
@@ -165,12 +163,6 @@ nv40_miptree_release(struct pipe_screen *pscreen, struct pipe_texture **ppt)
 			FREE(mt->level[l].image_offset);
 	}
 
-	if (mt->shadow_tex) {
-		if (mt->shadow_surface)
-			pscreen->tex_surface_release(pscreen, &mt->shadow_surface);
-		nv40_miptree_release(pscreen, &mt->shadow_tex);
-	}
-
 	FREE(mt);
 }
 
@@ -180,36 +172,33 @@ nv40_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 			 unsigned flags)
 {
 	struct nv40_miptree *mt = (struct nv40_miptree *)pt;
-	struct pipe_surface *ps;
+	struct nv04_surface *ns;
 
-	ps = CALLOC_STRUCT(pipe_surface);
-	if (!ps)
+	ns = CALLOC_STRUCT(nv04_surface);
+	if (!ns)
 		return NULL;
-	pipe_texture_reference(&ps->texture, pt);
-	ps->format = pt->format;
-	ps->width = pt->width[level];
-	ps->height = pt->height[level];
-	ps->block = pt->block;
-	ps->nblocksx = pt->nblocksx[level];
-	ps->nblocksy = pt->nblocksy[level];
-	ps->stride = mt->level[level].pitch;
-	ps->usage = flags;
-	ps->status = PIPE_SURFACE_STATUS_DEFINED;
-	ps->refcount = 1;
-	ps->face = face;
-	ps->level = level;
-	ps->zslice = zslice;
+	pipe_texture_reference(&ns->base.texture, pt);
+	ns->base.format = pt->format;
+	ns->base.width = pt->width[level];
+	ns->base.height = pt->height[level];
+	ns->base.usage = flags;
+	ns->base.status = PIPE_SURFACE_STATUS_DEFINED;
+	ns->base.refcount = 1;
+	ns->base.face = face;
+	ns->base.level = level;
+	ns->base.zslice = zslice;
+	ns->pitch = mt->level[level].pitch;
 
 	if (pt->target == PIPE_TEXTURE_CUBE) {
-		ps->offset = mt->level[level].image_offset[face];
+		ns->base.offset = mt->level[level].image_offset[face];
 	} else
 	if (pt->target == PIPE_TEXTURE_3D) {
-		ps->offset = mt->level[level].image_offset[zslice];
+		ns->base.offset = mt->level[level].image_offset[zslice];
 	} else {
-		ps->offset = mt->level[level].image_offset[0];
+		ns->base.offset = mt->level[level].image_offset[0];
 	}
 
-	return ps;
+	return &ns->base;
 }
 
 static void
diff --git a/src/gallium/drivers/nv40/nv40_screen.c b/src/gallium/drivers/nv40/nv40_screen.c
index 2372bc84410..0d4baefaea3 100644
--- a/src/gallium/drivers/nv40/nv40_screen.c
+++ b/src/gallium/drivers/nv40/nv40_screen.c
@@ -144,81 +144,6 @@ nv40_surface_buffer(struct pipe_surface *surf)
 	return mt->buffer;
 }
 
-static void *
-nv40_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
-		 unsigned flags )
-{
-	struct pipe_winsys	*ws = screen->winsys;
-	struct pipe_surface	*surface_to_map;
-	void			*map;
-
-	if (!(surface->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
-		struct nv40_miptree *mt = (struct nv40_miptree *)surface->texture;
-
-		if (!mt->shadow_tex) {
-			unsigned old_tex_usage = surface->texture->tex_usage;
-			surface->texture->tex_usage = NOUVEAU_TEXTURE_USAGE_LINEAR |
-			                              PIPE_TEXTURE_USAGE_DYNAMIC;
-			mt->shadow_tex = screen->texture_create(screen, surface->texture);
-			surface->texture->tex_usage = old_tex_usage;
-
-			assert(mt->shadow_tex->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR);
-		}
-
-		mt->shadow_surface = screen->get_tex_surface
-		(
-			screen, mt->shadow_tex,
-			surface->face, surface->level, surface->zslice,
-			surface->usage
-		);
-
-		surface_to_map = mt->shadow_surface;
-	}
-	else
-		surface_to_map = surface;
-
-	assert(surface_to_map);
-	map = ws->buffer_map(ws, nv40_surface_buffer(surface_to_map), flags);
-	if (!map)
-		return NULL;
-
-	return map + surface_to_map->offset;
-}
-
-static void
-nv40_surface_unmap(struct pipe_screen *screen, struct pipe_surface *surface)
-{
-	struct pipe_winsys	*ws = screen->winsys;
-	struct pipe_surface	*surface_to_unmap;
-
-	/* TODO: Copy from shadow just before push buffer is flushed instead.
-	         There are probably some programs that map/unmap excessively
-	         before rendering. */
-	if (!(surface->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
-		struct nv40_miptree *mt = (struct nv40_miptree *)surface->texture;
-
-		assert(mt->shadow_tex);
-
-		surface_to_unmap = mt->shadow_surface;
-	}
-	else
-		surface_to_unmap = surface;
-
-	assert(surface_to_unmap);
-
-	ws->buffer_unmap(ws, nv40_surface_buffer(surface_to_unmap));
-
-	if (surface_to_unmap != surface) {
-		struct nv40_screen *nvscreen = nv40_screen(screen);
-
-		nvscreen->eng2d->copy(nvscreen->eng2d, surface, 0, 0,
-		                      surface_to_unmap, 0, 0,
-		                      surface->width, surface->height);
-
-		screen->tex_surface_release(screen, &surface_to_unmap);
-	}
-}
-
 static void
 nv40_screen_destroy(struct pipe_screen *pscreen)
 {
@@ -240,7 +165,7 @@ nv40_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 {
 	struct nv40_screen *screen = CALLOC_STRUCT(nv40_screen);
 	struct nouveau_stateobj *so;
-	unsigned curie_class;
+	unsigned curie_class = 0;
 	unsigned chipset = nvws->channel->device->chipset;
 	int ret;
 
@@ -265,8 +190,6 @@ nv40_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 		if (NV6X_GRCLASS4497_CHIPSETS & (1 << (chipset & 0x0f)))
 			curie_class = NV44TCL;
 		break;
-	default:
-		break;
 	}
 
 	if (!curie_class) {
@@ -372,10 +295,8 @@ nv40_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 
 	screen->pipe.is_format_supported = nv40_screen_surface_format_supported;
 
-	screen->pipe.surface_map = nv40_surface_map;
-	screen->pipe.surface_unmap = nv40_surface_unmap;
-
 	nv40_screen_init_miptree_functions(&screen->pipe);
+	nv40_screen_init_transfer_functions(&screen->pipe);
 	u_simple_screen_init(&screen->pipe);
 
 	return &screen->pipe;
diff --git a/src/gallium/drivers/nv40/nv40_screen.h b/src/gallium/drivers/nv40/nv40_screen.h
index 4500aa0e5cc..7b503bd207d 100644
--- a/src/gallium/drivers/nv40/nv40_screen.h
+++ b/src/gallium/drivers/nv40/nv40_screen.h
@@ -34,4 +34,7 @@ nv40_screen(struct pipe_screen *screen)
 	return (struct nv40_screen *)screen;
 }
 
+void
+nv40_screen_init_transfer_functions(struct pipe_screen *pscreen);
+
 #endif
diff --git a/src/gallium/drivers/nv40/nv40_state.h b/src/gallium/drivers/nv40/nv40_state.h
index 9c55903ae30..8a9d8c8fdf6 100644
--- a/src/gallium/drivers/nv40/nv40_state.h
+++ b/src/gallium/drivers/nv40/nv40_state.h
@@ -79,9 +79,6 @@ struct nv40_miptree {
 	struct pipe_buffer *buffer;
 	uint total_size;
 
-	struct pipe_texture *shadow_tex;
-	struct pipe_surface *shadow_surface;
-
 	struct {
 		uint pitch;
 		uint *image_offset;
diff --git a/src/gallium/drivers/nv40/nv40_state_fb.c b/src/gallium/drivers/nv40/nv40_state_fb.c
index 454abad31f4..5ebd3a1a56b 100644
--- a/src/gallium/drivers/nv40/nv40_state_fb.c
+++ b/src/gallium/drivers/nv40/nv40_state_fb.c
@@ -12,7 +12,7 @@ static boolean
 nv40_state_framebuffer_validate(struct nv40_context *nv40)
 {
 	struct pipe_framebuffer_state *fb = &nv40->framebuffer;
-	struct pipe_surface *rt[4], *zeta;
+	struct nv04_surface *rt[4], *zeta;
 	uint32_t rt_enable, rt_format;
 	int i, colour_format = 0, zeta_format = 0;
 	struct nouveau_stateobj *so = so_new(64, 10);
@@ -27,7 +27,7 @@ nv40_state_framebuffer_validate(struct nv40_context *nv40)
 		} else {
 			colour_format = fb->cbufs[i]->format;
 			rt_enable |= (NV40TCL_RT_ENABLE_COLOR0 << i);
-			rt[i] = fb->cbufs[i];
+			rt[i] = (struct nv04_surface *)fb->cbufs[i];
 		}
 	}
 
@@ -37,13 +37,13 @@ nv40_state_framebuffer_validate(struct nv40_context *nv40)
 
 	if (fb->zsbuf) {
 		zeta_format = fb->zsbuf->format;
-		zeta = fb->zsbuf;
+		zeta = (struct nv04_surface *)fb->zsbuf;
 	}
 
-	if (!(rt[0]->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
+	if (!(rt[0]->base.texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
 		assert(!(fb->width & (fb->width - 1)) && !(fb->height & (fb->height - 1)));
 		for (i = 1; i < fb->nr_cbufs; i++)
-			assert(!(rt[i]->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR));
+			assert(!(rt[i]->base.texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR));
 
 		rt_format = NV40TCL_RT_FORMAT_TYPE_SWIZZLED |
 		            log2i(fb->width) << NV40TCL_RT_FORMAT_LOG2_WIDTH_SHIFT |
@@ -78,60 +78,60 @@ nv40_state_framebuffer_validate(struct nv40_context *nv40)
 
 	if (rt_enable & NV40TCL_RT_ENABLE_COLOR0) {
 		so_method(so, nv40->screen->curie, NV40TCL_DMA_COLOR0, 1);
-		so_reloc (so, nv40_surface_buffer(rt[0]), 0, rt_flags | NOUVEAU_BO_OR,
+		so_reloc (so, nv40_surface_buffer(&rt[0]->base), 0, rt_flags | NOUVEAU_BO_OR,
 			  nv40->nvws->channel->vram->handle,
 			  nv40->nvws->channel->gart->handle);
 		so_method(so, nv40->screen->curie, NV40TCL_COLOR0_PITCH, 2);
-		so_data  (so, rt[0]->stride);
-		so_reloc (so, nv40_surface_buffer(rt[0]), rt[0]->offset, rt_flags |
+		so_data  (so, rt[0]->pitch);
+		so_reloc (so, nv40_surface_buffer(&rt[0]->base), rt[0]->base.offset, rt_flags |
 			  NOUVEAU_BO_LOW, 0, 0);
 	}
 
 	if (rt_enable & NV40TCL_RT_ENABLE_COLOR1) {
 		so_method(so, nv40->screen->curie, NV40TCL_DMA_COLOR1, 1);
-		so_reloc (so, nv40_surface_buffer(rt[1]), 0, rt_flags | NOUVEAU_BO_OR,
+		so_reloc (so, nv40_surface_buffer(&rt[1]->base), 0, rt_flags | NOUVEAU_BO_OR,
 			  nv40->nvws->channel->vram->handle,
 			  nv40->nvws->channel->gart->handle);
 		so_method(so, nv40->screen->curie, NV40TCL_COLOR1_OFFSET, 2);
-		so_reloc (so, nv40_surface_buffer(rt[1]), rt[1]->offset, rt_flags |
+		so_reloc (so, nv40_surface_buffer(&rt[1]->base), rt[1]->base.offset, rt_flags |
 			  NOUVEAU_BO_LOW, 0, 0);
-		so_data  (so, rt[1]->stride);
+		so_data  (so, rt[1]->pitch);
 	}
 
 	if (rt_enable & NV40TCL_RT_ENABLE_COLOR2) {
 		so_method(so, nv40->screen->curie, NV40TCL_DMA_COLOR2, 1);
-		so_reloc (so, nv40_surface_buffer(rt[2]), 0, rt_flags | NOUVEAU_BO_OR,
+		so_reloc (so, nv40_surface_buffer(&rt[2]->base), 0, rt_flags | NOUVEAU_BO_OR,
 			  nv40->nvws->channel->vram->handle,
 			  nv40->nvws->channel->gart->handle);
 		so_method(so, nv40->screen->curie, NV40TCL_COLOR2_OFFSET, 1);
-		so_reloc (so, nv40_surface_buffer(rt[2]), rt[2]->offset, rt_flags |
+		so_reloc (so, nv40_surface_buffer(&rt[2]->base), rt[2]->base.offset, rt_flags |
 			  NOUVEAU_BO_LOW, 0, 0);
 		so_method(so, nv40->screen->curie, NV40TCL_COLOR2_PITCH, 1);
-		so_data  (so, rt[2]->stride);
+		so_data  (so, rt[2]->pitch);
 	}
 
 	if (rt_enable & NV40TCL_RT_ENABLE_COLOR3) {
 		so_method(so, nv40->screen->curie, NV40TCL_DMA_COLOR3, 1);
-		so_reloc (so, nv40_surface_buffer(rt[3]), 0, rt_flags | NOUVEAU_BO_OR,
+		so_reloc (so, nv40_surface_buffer(&rt[3]->base), 0, rt_flags | NOUVEAU_BO_OR,
 			  nv40->nvws->channel->vram->handle,
 			  nv40->nvws->channel->gart->handle);
 		so_method(so, nv40->screen->curie, NV40TCL_COLOR3_OFFSET, 1);
-		so_reloc (so, nv40_surface_buffer(rt[3]), rt[3]->offset, rt_flags |
+		so_reloc (so, nv40_surface_buffer(&rt[3]->base), rt[3]->base.offset, rt_flags |
 			  NOUVEAU_BO_LOW, 0, 0);
 		so_method(so, nv40->screen->curie, NV40TCL_COLOR3_PITCH, 1);
-		so_data  (so, rt[3]->stride);
+		so_data  (so, rt[3]->pitch);
 	}
 
 	if (zeta_format) {
 		so_method(so, nv40->screen->curie, NV40TCL_DMA_ZETA, 1);
-		so_reloc (so, nv40_surface_buffer(zeta), 0, rt_flags | NOUVEAU_BO_OR,
+		so_reloc (so, nv40_surface_buffer(&zeta->base), 0, rt_flags | NOUVEAU_BO_OR,
 			  nv40->nvws->channel->vram->handle,
 			  nv40->nvws->channel->gart->handle);
 		so_method(so, nv40->screen->curie, NV40TCL_ZETA_OFFSET, 1);
-		so_reloc (so, nv40_surface_buffer(zeta), zeta->offset, rt_flags |
+		so_reloc (so, nv40_surface_buffer(&zeta->base), zeta->base.offset, rt_flags |
 			  NOUVEAU_BO_LOW, 0, 0);
 		so_method(so, nv40->screen->curie, NV40TCL_ZETA_PITCH, 1);
-		so_data  (so, zeta->stride);
+		so_data  (so, zeta->pitch);
 	}
 
 	so_method(so, nv40->screen->curie, NV40TCL_RT_ENABLE, 1);
diff --git a/src/gallium/drivers/nv40/nv40_transfer.c b/src/gallium/drivers/nv40/nv40_transfer.c
new file mode 100644
index 00000000000..b090f2238e4
--- /dev/null
+++ b/src/gallium/drivers/nv40/nv40_transfer.c
@@ -0,0 +1,201 @@
+#include <pipe/p_state.h>
+#include <pipe/p_defines.h>
+#include <pipe/p_inlines.h>
+#include <util/u_memory.h>
+#include <nouveau/nouveau_winsys.h>
+#include "nv40_context.h"
+#include "nv40_screen.h"
+#include "nv40_state.h"
+
+struct nv40_transfer {
+	struct pipe_transfer base;
+	struct pipe_surface *surface;
+	bool direct;
+};
+
+static unsigned nv40_usage_tx_to_buf(unsigned tx_usage)
+{
+	switch (tx_usage) {
+		case PIPE_TRANSFER_READ:
+			return PIPE_BUFFER_USAGE_CPU_READ;
+		case PIPE_TRANSFER_WRITE:
+			return PIPE_BUFFER_USAGE_CPU_WRITE;
+		case PIPE_TRANSFER_READ_WRITE:
+			return PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+		default:
+			assert(0);
+	}
+
+	return -1;
+}
+
+static void
+nv40_compatible_transfer_tex(struct pipe_texture *pt, unsigned level,
+                             struct pipe_texture *template)
+{
+	memset(template, 0, sizeof(struct pipe_texture));
+	template->target = pt->target;
+	template->format = pt->format;
+	template->width[0] = pt->width[level];
+	template->height[0] = pt->height[level];
+	template->depth[0] = 1;
+	template->block = pt->block;
+	template->nblocksx[0] = pt->nblocksx[level];
+	template->nblocksy[0] = pt->nblocksx[level];
+	template->last_level = 0;
+	template->compressed = pt->compressed;
+	template->nr_samples = pt->nr_samples;
+
+	template->tex_usage = PIPE_TEXTURE_USAGE_DYNAMIC |
+	                      NOUVEAU_TEXTURE_USAGE_LINEAR;
+}
+
+static struct pipe_transfer *
+nv40_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
+		  unsigned face, unsigned level, unsigned zslice,
+		  enum pipe_transfer_usage usage,
+		  unsigned x, unsigned y, unsigned w, unsigned h)
+{
+	struct nv40_miptree *mt = (struct nv40_miptree *)pt;
+	struct nv40_transfer *tx;
+	struct pipe_texture tx_tex_template, *tx_tex;
+
+	tx = CALLOC_STRUCT(nv40_transfer);
+	if (!tx)
+		return NULL;
+
+	tx->base.refcount = 1;
+	pipe_texture_reference(&tx->base.texture, pt);
+	tx->base.format = pt->format;
+	tx->base.x = x;
+	tx->base.y = y;
+	tx->base.width = w;
+	tx->base.height = h;
+	tx->base.block = pt->block;
+	tx->base.nblocksx = pt->nblocksx[level];
+	tx->base.nblocksy = pt->nblocksy[level];
+	tx->base.stride = mt->level[level].pitch;
+	tx->base.usage = usage;
+	tx->base.face = face;
+	tx->base.level = level;
+	tx->base.zslice = zslice;
+
+	/* Direct access to texture */
+	if ((pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC ||
+	     debug_get_bool_option("NOUVEAU_NO_TRANSFER", TRUE/*XXX:FALSE*/)) &&
+	    pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)
+	{
+		tx->direct = true;
+		tx->surface = pscreen->get_tex_surface(pscreen, pt,
+	                                               face, level, zslice,
+	                                               nv40_usage_tx_to_buf(usage));
+		return &tx->base;
+	}
+
+	tx->direct = false;
+
+	nv40_compatible_transfer_tex(pt, level, &tx_tex_template);
+
+	tx_tex = pscreen->texture_create(pscreen, &tx_tex_template);
+	if (!tx_tex)
+	{
+		FREE(tx);
+		return NULL;
+	}
+
+	tx->surface = pscreen->get_tex_surface(pscreen, tx_tex,
+	                                       face, level, zslice,
+	                                       nv40_usage_tx_to_buf(usage));
+
+	pipe_texture_reference(&tx_tex, NULL);
+
+	if (!tx->surface)
+	{
+		pipe_surface_reference(&tx->surface, NULL);
+		FREE(tx);
+		return NULL;
+	}
+
+	if (usage != PIPE_TRANSFER_WRITE) {
+		struct nv40_screen *nvscreen = nv40_screen(pscreen);
+		struct pipe_surface *src;
+
+		src = pscreen->get_tex_surface(pscreen, pt,
+	                                       face, level, zslice,
+	                                       PIPE_BUFFER_USAGE_GPU_READ);
+
+		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
+		/* TODO: Check if SIFM can un-swizzle */
+		nvscreen->eng2d->copy(nvscreen->eng2d,
+		                      tx->surface, 0, 0,
+		                      src, 0, 0,
+		                      src->width, src->height);
+
+		pipe_surface_reference(&src, NULL);
+	}
+
+	return &tx->base;
+}
+
+static void
+nv40_transfer_del(struct pipe_screen *pscreen, struct pipe_transfer **pptx)
+{
+	struct pipe_transfer *ptx = *pptx;
+	struct nv40_transfer *tx = (struct nv40_transfer *)ptx;
+
+	if (!tx->direct && ptx->usage != PIPE_TRANSFER_READ) {
+		struct nv40_screen *nvscreen = nv40_screen(pscreen);
+		struct pipe_surface *dst;
+
+		dst = pscreen->get_tex_surface(pscreen, ptx->texture,
+	                                       ptx->face, ptx->level, ptx->zslice,
+	                                       PIPE_BUFFER_USAGE_GPU_WRITE);
+
+		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
+		nvscreen->eng2d->copy(nvscreen->eng2d,
+		                      dst, 0, 0,
+		                      tx->surface, 0, 0,
+		                      dst->width, dst->height);
+
+		pipe_surface_reference(&dst, NULL);
+	}
+
+	*pptx = NULL;
+	if (--ptx->refcount)
+		return;
+
+	pipe_surface_reference(&tx->surface, NULL);
+	pipe_texture_reference(&ptx->texture, NULL);
+	FREE(ptx);
+}
+
+static void *
+nv40_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv40_transfer *tx = (struct nv40_transfer *)ptx;
+	struct nv04_surface *ns = (struct nv04_surface *)tx->surface;
+	struct nv40_miptree *mt = (struct nv40_miptree *)tx->surface->texture;
+	void *map = pipe_buffer_map(pscreen, mt->buffer,
+	                            nv40_usage_tx_to_buf(ptx->usage));
+
+	return map + ns->base.offset +
+	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
+}
+
+static void
+nv40_transfer_unmap(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv40_transfer *tx = (struct nv40_transfer *)ptx;
+	struct nv40_miptree *mt = (struct nv40_miptree *)tx->surface->texture;
+
+	pipe_buffer_unmap(pscreen, mt->buffer);
+}
+
+void
+nv40_screen_init_transfer_functions(struct pipe_screen *pscreen)
+{
+	pscreen->get_tex_transfer = nv40_transfer_new;
+	pscreen->tex_transfer_release = nv40_transfer_del;
+	pscreen->transfer_map = nv40_transfer_map;
+	pscreen->transfer_unmap = nv40_transfer_unmap;
+}
diff --git a/src/gallium/drivers/r300/Makefile b/src/gallium/drivers/r300/Makefile
index 9b7524b523a..a0fd17bd59b 100644
--- a/src/gallium/drivers/r300/Makefile
+++ b/src/gallium/drivers/r300/Makefile
@@ -12,6 +12,7 @@ C_SOURCES = \
 	r300_screen.c \
 	r300_state.c \
 	r300_state_derived.c \
+	r300_state_invariant.c \
 	r300_state_shader.c \
 	r300_surface.c \
 	r300_swtcl_emit.c \
diff --git a/src/gallium/drivers/r300/r300_chipset.c b/src/gallium/drivers/r300/r300_chipset.c
index 196537a432d..e01a0546b22 100644
--- a/src/gallium/drivers/r300/r300_chipset.c
+++ b/src/gallium/drivers/r300/r300_chipset.c
@@ -30,7 +30,7 @@
 void r300_parse_chipset(struct r300_capabilities* caps)
 {
     /* Reasonable defaults */
-    caps->has_tcl = TRUE;
+    caps->has_tcl = getenv("RADEON_NO_TCL") ? TRUE : FALSE;
     caps->is_r500 = FALSE;
     caps->num_vert_fpus = 4;
 
@@ -204,6 +204,13 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->has_tcl = FALSE;
             break;
 
+        case 0x793F:
+        case 0x7941:
+        case 0x7942:
+            caps->family = CHIP_FAMILY_RS600;
+            caps->has_tcl = FALSE;
+            break;
+
         case 0x796C:
         case 0x796D:
         case 0x796E:
diff --git a/src/gallium/drivers/r300/r300_chipset.h b/src/gallium/drivers/r300/r300_chipset.h
index a9cd372ec55..21eebeae600 100644
--- a/src/gallium/drivers/r300/r300_chipset.h
+++ b/src/gallium/drivers/r300/r300_chipset.h
@@ -64,6 +64,7 @@ enum {
     CHIP_FAMILY_RC410,
     CHIP_FAMILY_RS480,
     CHIP_FAMILY_RS482,
+    CHIP_FAMILY_RS600,
     CHIP_FAMILY_RS690,
     CHIP_FAMILY_RS740,
     CHIP_FAMILY_RV515,
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index 15a87515494..a9811501435 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -73,7 +73,7 @@ static boolean r300_draw_range_elements(struct pipe_context* pipe,
                                              start + count - 1, NULL);
     }
 
-    return true;
+    return TRUE;
 }
 
 static boolean r300_draw_elements(struct pipe_context* pipe,
@@ -97,6 +97,7 @@ static void r300_destroy_context(struct pipe_context* context) {
     draw_destroy(r300->draw);
 
     FREE(r300->blend_color_state);
+    FREE(r300->rs_block);
     FREE(r300->scissor_state);
     FREE(r300);
 }
@@ -126,6 +127,7 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
     draw_set_rasterize_stage(r300->draw, r300_draw_swtcl_stage(r300));
 
     r300->blend_color_state = CALLOC_STRUCT(r300_blend_color_state);
+    r300->rs_block = CALLOC_STRUCT(r300_rs_block);
     r300->scissor_state = CALLOC_STRUCT(r300_scissor_state);
 
     r300_init_flush_functions(r300);
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index aaab1dd2bcb..5247901be5b 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -75,6 +75,13 @@ struct r300_rs_state {
     uint32_t line_stipple_value;    /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */
 };
 
+struct r300_rs_block {
+    uint32_t ip[8]; /* R300_RS_IP_[0-7], R500_RS_IP_[0-7] */
+    uint32_t count; /* R300_RS_COUNT */
+    uint32_t inst_count; /* R300_RS_INST_COUNT */
+    uint32_t inst[8]; /* R300_RS_INST_[0-7] */
+};
+
 struct r300_sampler_state {
     uint32_t filter0;      /* R300_TX_FILTER0: 0x4400 */
     uint32_t filter1;      /* R300_TX_FILTER1: 0x4440 */
@@ -96,12 +103,13 @@ struct r300_texture_state {
 #define R300_NEW_FRAMEBUFFERS    0x0000010
 #define R300_NEW_FRAGMENT_SHADER 0x0000020
 #define R300_NEW_RASTERIZER      0x0000040
-#define R300_NEW_SAMPLER         0x0000080
-#define R300_NEW_SCISSOR         0x0008000
-#define R300_NEW_TEXTURE         0x0010000
-#define R300_NEW_VERTEX_FORMAT   0x1000000
-#define R300_NEW_VERTEX_SHADER   0x2000000
-#define R300_NEW_KITCHEN_SINK    0x3ffffff
+#define R300_NEW_RS_BLOCK        0x0000080
+#define R300_NEW_SAMPLER         0x0000100
+#define R300_NEW_SCISSOR         0x0010000
+#define R300_NEW_TEXTURE         0x0020000
+#define R300_NEW_VERTEX_FORMAT   0x2000000
+#define R300_NEW_VERTEX_SHADER   0x4000000
+#define R300_NEW_KITCHEN_SINK    0x7ffffff
 
 /* The next several objects are not pure Radeon state; they inherit from
  * various Gallium classes. */
@@ -224,6 +232,8 @@ struct r300_context {
     struct pipe_framebuffer_state framebuffer_state;
     /* Rasterizer state. */
     struct r300_rs_state* rs_state;
+    /* RS block state. */
+    struct r300_rs_block* rs_block;
     /* Sampler states. */
     struct r300_sampler_state* sampler_states[8];
     int sampler_count;
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index 960f45f651f..91fac62cbe6 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -40,9 +40,9 @@ void r300_emit_blend_state(struct r300_context* r300,
 void r300_emit_blend_color_state(struct r300_context* r300,
                                  struct r300_blend_color_state* bc)
 {
-    struct r300_screen* r300screen =
-        (struct r300_screen*)r300->context.screen;
+    struct r300_screen* r300screen = r300_screen(r300->context.screen);
     CS_LOCALS(r300);
+
     if (r300screen->caps->is_r500) {
         BEGIN_CS(3);
         OUT_CS_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2);
@@ -59,9 +59,9 @@ void r300_emit_blend_color_state(struct r300_context* r300,
 void r300_emit_dsa_state(struct r300_context* r300,
                            struct r300_dsa_state* dsa)
 {
-    struct r300_screen* r300screen =
-        (struct r300_screen*)r300->context.screen;
+    struct r300_screen* r300screen = r300_screen(r300->context.screen);
     CS_LOCALS(r300);
+
     BEGIN_CS(r300screen->caps->is_r500 ? 8 : 8);
     OUT_CS_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function);
     /* XXX figure out the r300 counterpart for this */
@@ -84,6 +84,7 @@ void r300_emit_fragment_shader(struct r300_context* r300,
 {
     CS_LOCALS(r300);
     int i;
+
     BEGIN_CS(22);
 
     OUT_CS_REG(R300_US_CONFIG, MAX2(fs->indirections - 1, 0));
@@ -114,7 +115,8 @@ void r500_emit_fragment_shader(struct r300_context* r300,
                                struct r500_fragment_shader* fs)
 {
     CS_LOCALS(r300);
-    int i = 0;
+    int i;
+
     BEGIN_CS(9 + (fs->instruction_count * 6));
     OUT_CS_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO);
     OUT_CS_REG(R500_US_PIXSIZE, fs->shader.stack_size);
@@ -191,9 +193,9 @@ void r300_emit_fb_state(struct r300_context* r300,
 
 void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs)
 {
-    struct r300_screen* r300screen =
-        (struct r300_screen*)r300->context.screen;
+    struct r300_screen* r300screen = r300_screen(r300->context.screen);
     CS_LOCALS(r300);
+
     BEGIN_CS(13);
     OUT_CS_REG(R300_VAP_CNTL_STATUS, rs->vap_control_status);
     OUT_CS_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 6);
@@ -208,10 +210,44 @@ void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs)
     END_CS;
 }
 
+void r300_emit_rs_block_state(struct r300_context* r300,
+                              struct r300_rs_block* rs)
+{
+    struct r300_screen* r300screen = r300_screen(r300->context.screen);
+    CS_LOCALS(r300);
+    int i;
+
+    BEGIN_CS(21);
+    if (r300screen->caps->is_r500) {
+        OUT_CS_REG_SEQ(R500_RS_IP_0, 8);
+    } else {
+        OUT_CS_REG_SEQ(R300_RS_IP_0, 8);
+    }
+    for (i = 0; i < 8; i++) {
+        OUT_CS(rs->ip[i]);
+    }
+
+    OUT_CS_REG_SEQ(R300_RS_COUNT, 2);
+    OUT_CS(rs->count);
+    OUT_CS(rs->inst_count);
+
+    if (r300screen->caps->is_r500) {
+        OUT_CS_REG_SEQ(R500_RS_INST_0, 8);
+    } else {
+        OUT_CS_REG_SEQ(R300_RS_INST_0, 8);
+    }
+    for (i = 0; i < 8; i++) {
+        OUT_CS(rs->inst[i]);
+    }
+
+    END_CS;
+}
+
 void r300_emit_scissor_state(struct r300_context* r300,
                              struct r300_scissor_state* scissor)
 {
     CS_LOCALS(r300);
+
     BEGIN_CS(3);
     OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
     OUT_CS(scissor->scissor_top_left);
@@ -248,8 +284,7 @@ void r300_emit_vertex_format_state(struct r300_context* r300)
 /* Emit all dirty state. */
 void r300_emit_dirty_state(struct r300_context* r300)
 {
-    struct r300_screen* r300screen =
-        (struct r300_screen*)r300->context.screen;
+    struct r300_screen* r300screen = r300_screen(r300->context.screen);
     CS_LOCALS(r300);
 
     if (!(r300->dirty_state) && !(r300->dirty_hw)) {
diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h
index 8888b39a2f9..e0da9d361e4 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -732,6 +732,10 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R500_RS_IP_TEX_PTR_Q_SHIFT 			18
 #define R500_RS_IP_COL_PTR_SHIFT 			24
 #define R500_RS_IP_COL_FMT_SHIFT 			27
+#       define R500_RS_SEL_S(x)                         ((x) << 0)
+#       define R500_RS_SEL_T(x)                         ((x) << 6)
+#       define R500_RS_SEL_R(x)                         ((x) << 12)
+#       define R500_RS_SEL_Q(x)                         ((x) << 18)
 #	define R500_RS_COL_PTR(x)		        ((x) << 24)
 #       define R500_RS_COL_FMT(x)                       ((x) << 27)
 /* gap */
@@ -1216,14 +1220,18 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R500_RS_INST_14					0x4358
 #define R500_RS_INST_15					0x435c
 #define R500_RS_INST_TEX_ID_SHIFT			0
+#        define R500_RS_INST_TEX_ID(x)                  ((x) << 0)
 #define R500_RS_INST_TEX_CN_WRITE			(1 << 4)
 #define R500_RS_INST_TEX_ADDR_SHIFT			5
+#        define R500_RS_INST_TEX_ADDR(x)                ((x) << 0)
 #define R500_RS_INST_COL_ID_SHIFT			12
+#        define R500_RS_INST_COL_ID(x)                  ((x) << 12)
 #define R500_RS_INST_COL_CN_NO_WRITE			(0 << 16)
 #define R500_RS_INST_COL_CN_WRITE			(1 << 16)
 #define R500_RS_INST_COL_CN_WRITE_FBUFFER		(2 << 16)
 #define R500_RS_INST_COL_CN_WRITE_BACKFACE		(3 << 16)
 #define R500_RS_INST_COL_ADDR_SHIFT			18
+#        define R500_RS_INST_COL_ADDR(x)                ((x) << 18)
 #define R500_RS_INST_TEX_ADJ				(1 << 25)
 #define R500_RS_INST_W_CN				(1 << 26)
 
@@ -1240,9 +1248,11 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R300_RS_INST_7                     0x434C
 #	define R300_RS_INST_TEX_ID(x)  		((x) << 0)
 #	define R300_RS_INST_TEX_CN_WRITE 	(1 << 3)
+#	define R300_RS_INST_TEX_ADDR(x)		((x) << 6)
 #	define R300_RS_INST_TEX_ADDR_SHIFT 	6
 #	define R300_RS_INST_COL_ID(x)		((x) << 11)
 #	define R300_RS_INST_COL_CN_WRITE	(1 << 14)
+#	define R300_RS_INST_COL_ADDR(x)		((x) << 17)
 #	define R300_RS_INST_COL_ADDR_SHIFT	17
 #	define R300_RS_INST_TEX_ADJ		(1 << 22)
 #	define R300_RS_COL_BIAS_UNUSED_SHIFT    23
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 5ff9015a7bf..2fcd5048128 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -50,6 +50,7 @@ static const char* chip_families[] = {
     "RC410",
     "RS480",
     "RS482",
+    "RS600",
     "RS690",
     "RS740",
     "RV515",
@@ -178,8 +179,11 @@ static boolean check_tex_2d_format(enum pipe_format format, boolean is_r500)
 {
     switch (format) {
         /* Colorbuffer */
+        case PIPE_FORMAT_A4R4G4B4_UNORM:
+        case PIPE_FORMAT_R5G6B5_UNORM:
+        case PIPE_FORMAT_A1R5G5B5_UNORM:
         case PIPE_FORMAT_A8R8G8B8_UNORM:
-        /* Texture */
+        /* Colorbuffer or texture */
         case PIPE_FORMAT_I8_UNORM:
         /* Z buffer */
         case PIPE_FORMAT_Z16_UNORM:
@@ -187,29 +191,25 @@ static boolean check_tex_2d_format(enum pipe_format format, boolean is_r500)
         case PIPE_FORMAT_Z24S8_UNORM:
             return TRUE;
 
-        /* XXX Supported yet unimplemented formats: */
-        case PIPE_FORMAT_A1R5G5B5_UNORM:
-        case PIPE_FORMAT_R5G6B5_UNORM:
         /* XXX These don't even exist
         case PIPE_FORMAT_A32R32G32B32:
         case PIPE_FORMAT_A16R16G16B16: */
         /* XXX Insert YUV422 packed VYUY and YVYU here */
-        /* XXX What the deuce is UV88? (r3xx accel page 14) */
-        case PIPE_FORMAT_A4R4G4B4_UNORM:
+        /* XXX What the deuce is UV88? (r3xx accel page 14)
             debug_printf("r300: Warning: Got unimplemented format: %s in %s\n",
                 pf_name(format), __FUNCTION__);
-            return FALSE;
+            return FALSE; */
 
         /* XXX Supported yet unimplemented r5xx formats: */
         /* XXX Again, what is UV1010 this time? (r5xx accel page 148) */
         /* XXX Even more that don't exist
         case PIPE_FORMAT_A10R10G10B10_UNORM:
         case PIPE_FORMAT_A2R10G10B10_UNORM:
-        case PIPE_FORMAT_I10_UNORM: */
+        case PIPE_FORMAT_I10_UNORM:
             debug_printf(
                 "r300: Warning: Got unimplemented r500 format: %s in %s\n",
                 pf_name(format), __FUNCTION__);
-            return FALSE;
+            return FALSE; */
 
         default:
             debug_printf("r300: Warning: Got unsupported format: %s in %s\n",
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index da99a3be6b6..6e64ad2dc35 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -28,89 +28,11 @@
 
 #include "r300_context.h"
 #include "r300_reg.h"
+#include "r300_state_inlines.h"
 #include "r300_state_shader.h"
 
 /* r300_state: Functions used to intialize state context by translating
- * Gallium state objects into semi-native r300 state objects.
- *
- * XXX break this file up into pieces if it gets too big! */
-
-/* Pack a float into a dword. */
-static uint32_t pack_float_32(float f)
-{
-    union {
-        float f;
-        uint32_t u;
-    } u;
-
-    u.f = f;
-    return u.u;
-}
-
-static uint32_t translate_blend_function(int blend_func) {
-    switch (blend_func) {
-        case PIPE_BLEND_ADD:
-            return R300_COMB_FCN_ADD_CLAMP;
-        case PIPE_BLEND_SUBTRACT:
-            return R300_COMB_FCN_SUB_CLAMP;
-        case PIPE_BLEND_REVERSE_SUBTRACT:
-            return R300_COMB_FCN_RSUB_CLAMP;
-        case PIPE_BLEND_MIN:
-            return R300_COMB_FCN_MIN;
-        case PIPE_BLEND_MAX:
-            return R300_COMB_FCN_MAX;
-        default:
-            debug_printf("r300: Unknown blend function %d\n", blend_func);
-            break;
-    }
-    return 0;
-}
-
-/* XXX we can also offer the D3D versions of some of these... */
-static uint32_t translate_blend_factor(int blend_fact) {
-    switch (blend_fact) {
-        case PIPE_BLENDFACTOR_ONE:
-            return R300_BLEND_GL_ONE;
-        case PIPE_BLENDFACTOR_SRC_COLOR:
-            return R300_BLEND_GL_SRC_COLOR;
-        case PIPE_BLENDFACTOR_SRC_ALPHA:
-            return R300_BLEND_GL_SRC_ALPHA;
-        case PIPE_BLENDFACTOR_DST_ALPHA:
-            return R300_BLEND_GL_DST_ALPHA;
-        case PIPE_BLENDFACTOR_DST_COLOR:
-            return R300_BLEND_GL_DST_COLOR;
-        case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
-            return R300_BLEND_GL_SRC_ALPHA_SATURATE;
-        case PIPE_BLENDFACTOR_CONST_COLOR:
-            return R300_BLEND_GL_CONST_COLOR;
-        case PIPE_BLENDFACTOR_CONST_ALPHA:
-            return R300_BLEND_GL_CONST_ALPHA;
-        /* XXX WTF are these?
-        case PIPE_BLENDFACTOR_SRC1_COLOR:
-        case PIPE_BLENDFACTOR_SRC1_ALPHA: */
-        case PIPE_BLENDFACTOR_ZERO:
-            return R300_BLEND_GL_ZERO;
-        case PIPE_BLENDFACTOR_INV_SRC_COLOR:
-            return R300_BLEND_GL_ONE_MINUS_SRC_COLOR;
-        case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
-            return R300_BLEND_GL_ONE_MINUS_SRC_ALPHA;
-        case PIPE_BLENDFACTOR_INV_DST_ALPHA:
-            return R300_BLEND_GL_ONE_MINUS_DST_ALPHA;
-        case PIPE_BLENDFACTOR_INV_DST_COLOR:
-            return R300_BLEND_GL_ONE_MINUS_DST_COLOR;
-        case PIPE_BLENDFACTOR_INV_CONST_COLOR:
-            return R300_BLEND_GL_ONE_MINUS_CONST_COLOR;
-        case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
-            return R300_BLEND_GL_ONE_MINUS_CONST_ALPHA;
-        /* XXX see above
-        case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
-        case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: */
-        default:
-            debug_printf("r300: Unknown blend factor %d\n", blend_fact);
-            break;
-    }
-    return 0;
-}
+ * Gallium state objects into semi-native r300 state objects. */
 
 /* Create a new blend state based on the CSO blend state.
  *
@@ -126,16 +48,16 @@ static void* r300_create_blend_state(struct pipe_context* pipe,
         blend->blend_control = R300_ALPHA_BLEND_ENABLE |
                 R300_SEPARATE_ALPHA_ENABLE |
                 R300_READ_ENABLE |
-                translate_blend_function(state->rgb_func) |
-                (translate_blend_factor(state->rgb_src_factor) <<
+                r300_translate_blend_function(state->rgb_func) |
+                (r300_translate_blend_factor(state->rgb_src_factor) <<
                     R300_SRC_BLEND_SHIFT) |
-                (translate_blend_factor(state->rgb_dst_factor) <<
+                (r300_translate_blend_factor(state->rgb_dst_factor) <<
                     R300_DST_BLEND_SHIFT);
         blend->alpha_blend_control =
-                translate_blend_function(state->alpha_func) |
-                (translate_blend_factor(state->alpha_src_factor) <<
+                r300_translate_blend_function(state->alpha_func) |
+                (r300_translate_blend_factor(state->alpha_src_factor) <<
                     R300_SRC_BLEND_SHIFT) |
-                (translate_blend_factor(state->alpha_dst_factor) <<
+                (r300_translate_blend_factor(state->alpha_dst_factor) <<
                     R300_DST_BLEND_SHIFT);
     }
 
@@ -178,21 +100,17 @@ static void r300_set_blend_color(struct pipe_context* pipe,
                                  const struct pipe_blend_color* color)
 {
     struct r300_context* r300 = r300_context(pipe);
-    uint32_t r, g, b, a;
     ubyte ur, ug, ub, ua;
 
-    r = util_iround(color->color[0] * 1023.0f);
-    g = util_iround(color->color[1] * 1023.0f);
-    b = util_iround(color->color[2] * 1023.0f);
-    a = util_iround(color->color[3] * 1023.0f);
-
     ur = float_to_ubyte(color->color[0]);
     ug = float_to_ubyte(color->color[1]);
     ub = float_to_ubyte(color->color[2]);
     ua = float_to_ubyte(color->color[3]);
 
-    r300->blend_color_state->blend_color = (a << 24) | (r << 16) | (g << 8) | b;
+    util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM,
+            &r300->blend_color_state->blend_color);
 
+    /* XXX this is wrong */
     r300->blend_color_state->blend_color_red_alpha = ur | (ua << 16);
     r300->blend_color_state->blend_color_green_blue = ub | (ug << 16);
 
@@ -233,82 +151,6 @@ static void
     r300->dirty_state |= R300_NEW_CONSTANTS;
 }
 
-static uint32_t translate_depth_stencil_function(int zs_func) {
-    switch (zs_func) {
-        case PIPE_FUNC_NEVER:
-            return R300_ZS_NEVER;
-        case PIPE_FUNC_LESS:
-            return R300_ZS_LESS;
-        case PIPE_FUNC_EQUAL:
-            return R300_ZS_EQUAL;
-        case PIPE_FUNC_LEQUAL:
-            return R300_ZS_LEQUAL;
-        case PIPE_FUNC_GREATER:
-            return R300_ZS_GREATER;
-        case PIPE_FUNC_NOTEQUAL:
-            return R300_ZS_NOTEQUAL;
-        case PIPE_FUNC_GEQUAL:
-            return R300_ZS_GEQUAL;
-        case PIPE_FUNC_ALWAYS:
-            return R300_ZS_ALWAYS;
-        default:
-            debug_printf("r300: Unknown depth/stencil function %d\n",
-                zs_func);
-            break;
-    }
-    return 0;
-}
-
-static uint32_t translate_stencil_op(int s_op) {
-    switch (s_op) {
-        case PIPE_STENCIL_OP_KEEP:
-            return R300_ZS_KEEP;
-        case PIPE_STENCIL_OP_ZERO:
-            return R300_ZS_ZERO;
-        case PIPE_STENCIL_OP_REPLACE:
-            return R300_ZS_REPLACE;
-        case PIPE_STENCIL_OP_INCR:
-            return R300_ZS_INCR;
-        case PIPE_STENCIL_OP_DECR:
-            return R300_ZS_DECR;
-        case PIPE_STENCIL_OP_INCR_WRAP:
-            return R300_ZS_INCR_WRAP;
-        case PIPE_STENCIL_OP_DECR_WRAP:
-            return R300_ZS_DECR_WRAP;
-        case PIPE_STENCIL_OP_INVERT:
-            return R300_ZS_INVERT;
-        default:
-            debug_printf("r300: Unknown stencil op %d", s_op);
-            break;
-    }
-    return 0;
-}
-
-static uint32_t translate_alpha_function(int alpha_func) {
-    switch (alpha_func) {
-        case PIPE_FUNC_NEVER:
-            return R300_FG_ALPHA_FUNC_NEVER;
-        case PIPE_FUNC_LESS:
-            return R300_FG_ALPHA_FUNC_LESS;
-        case PIPE_FUNC_EQUAL:
-            return R300_FG_ALPHA_FUNC_EQUAL;
-        case PIPE_FUNC_LEQUAL:
-            return R300_FG_ALPHA_FUNC_LE;
-        case PIPE_FUNC_GREATER:
-            return R300_FG_ALPHA_FUNC_GREATER;
-        case PIPE_FUNC_NOTEQUAL:
-            return R300_FG_ALPHA_FUNC_NOTEQUAL;
-        case PIPE_FUNC_GEQUAL:
-            return R300_FG_ALPHA_FUNC_GE;
-        case PIPE_FUNC_ALWAYS:
-            return R300_FG_ALPHA_FUNC_ALWAYS;
-        default:
-            debug_printf("r300: Unknown alpha function %d", alpha_func);
-            break;
-    }
-    return 0;
-}
-
 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
  *
  * This contains the depth buffer, stencil buffer, alpha test, and such.
@@ -329,7 +171,7 @@ static void*
         }
 
         dsa->z_stencil_control |=
-            (translate_depth_stencil_function(state->depth.func) <<
+            (r300_translate_depth_stencil_function(state->depth.func) <<
                 R300_Z_FUNC_SHIFT);
     }
 
@@ -337,14 +179,14 @@ static void*
     if (state->stencil[0].enabled) {
         dsa->z_buffer_control |= R300_STENCIL_ENABLE;
         dsa->z_stencil_control |=
-                (translate_depth_stencil_function(state->stencil[0].func) <<
-                    R300_S_FRONT_FUNC_SHIFT) |
-                (translate_stencil_op(state->stencil[0].fail_op) <<
-                    R300_S_FRONT_SFAIL_OP_SHIFT) |
-                (translate_stencil_op(state->stencil[0].zpass_op) <<
-                    R300_S_FRONT_ZPASS_OP_SHIFT) |
-                (translate_stencil_op(state->stencil[0].zfail_op) <<
-                    R300_S_FRONT_ZFAIL_OP_SHIFT);
+            (r300_translate_depth_stencil_function(state->stencil[0].func) <<
+                R300_S_FRONT_FUNC_SHIFT) |
+            (r300_translate_stencil_op(state->stencil[0].fail_op) <<
+                R300_S_FRONT_SFAIL_OP_SHIFT) |
+            (r300_translate_stencil_op(state->stencil[0].zpass_op) <<
+                R300_S_FRONT_ZPASS_OP_SHIFT) |
+            (r300_translate_stencil_op(state->stencil[0].zfail_op) <<
+                R300_S_FRONT_ZFAIL_OP_SHIFT);
 
         dsa->stencil_ref_mask = (state->stencil[0].ref_value) |
                 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) |
@@ -353,14 +195,14 @@ static void*
         if (state->stencil[1].enabled) {
             dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK;
             dsa->z_stencil_control |=
-                (translate_depth_stencil_function(state->stencil[1].func) <<
-                    R300_S_BACK_FUNC_SHIFT) |
-                (translate_stencil_op(state->stencil[1].fail_op) <<
-                    R300_S_BACK_SFAIL_OP_SHIFT) |
-                (translate_stencil_op(state->stencil[1].zpass_op) <<
-                    R300_S_BACK_ZPASS_OP_SHIFT) |
-                (translate_stencil_op(state->stencil[1].zfail_op) <<
-                    R300_S_BACK_ZFAIL_OP_SHIFT);
+            (r300_translate_depth_stencil_function(state->stencil[1].func) <<
+                R300_S_BACK_FUNC_SHIFT) |
+            (r300_translate_stencil_op(state->stencil[1].fail_op) <<
+                R300_S_BACK_SFAIL_OP_SHIFT) |
+            (r300_translate_stencil_op(state->stencil[1].zpass_op) <<
+                R300_S_BACK_ZPASS_OP_SHIFT) |
+            (r300_translate_stencil_op(state->stencil[1].zfail_op) <<
+                R300_S_BACK_ZFAIL_OP_SHIFT);
 
             dsa->stencil_ref_bf = (state->stencil[1].ref_value) |
                 (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) |
@@ -370,7 +212,8 @@ static void*
 
     /* Alpha test setup. */
     if (state->alpha.enabled) {
-        dsa->alpha_function = translate_alpha_function(state->alpha.func) |
+        dsa->alpha_function =
+            r300_translate_alpha_function(state->alpha.func) |
             R300_FG_ALPHA_FUNC_ENABLE;
         dsa->alpha_reference = CLAMP(state->alpha.ref_value * 1023.0f,
                                      0, 1023);
@@ -457,7 +300,7 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
         }
     }
 
-    fs->translated = true;
+    fs->translated = TRUE;
     r300->fs = fs;
 
     r300->dirty_state |= R300_NEW_FRAGMENT_SHADER;
@@ -475,10 +318,6 @@ static void r300_set_polygon_stipple(struct pipe_context* pipe,
     /* XXX */
 }
 
-static INLINE int pack_float_16_6x(float f) {
-    return ((int)(f * 6.0) & 0xffff);
-}
-
 /* Create a new rasterizer state based on the CSO rasterizer state.
  *
  * This is a very large chunk of state, and covers most of the graphics
@@ -530,15 +369,15 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
 
     if (rs->polygon_offset_enable) {
         rs->depth_offset_front = rs->depth_offset_back =
-                pack_float_32(state->offset_units);
+            fui(state->offset_units);
         rs->depth_scale_front = rs->depth_scale_back =
-                pack_float_32(state->offset_scale);
+            fui(state->offset_scale);
     }
 
     if (state->line_stipple_enable) {
         rs->line_stipple_config =
             R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE |
-            (pack_float_32((float)state->line_stipple_factor) &
+            (fui((float)state->line_stipple_factor) &
                 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK);
         /* XXX this might need to be scaled up */
         rs->line_stipple_value = state->line_stipple_pattern;
@@ -567,83 +406,6 @@ static void r300_delete_rs_state(struct pipe_context* pipe, void* state)
     FREE(state);
 }
 
-static uint32_t translate_wrap(int wrap) {
-    switch (wrap) {
-        case PIPE_TEX_WRAP_REPEAT:
-            return R300_TX_REPEAT;
-        case PIPE_TEX_WRAP_CLAMP:
-            return R300_TX_CLAMP;
-        case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-            return R300_TX_CLAMP_TO_EDGE;
-        case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-            return R300_TX_CLAMP_TO_BORDER;
-        case PIPE_TEX_WRAP_MIRROR_REPEAT:
-            return R300_TX_REPEAT | R300_TX_MIRRORED;
-        case PIPE_TEX_WRAP_MIRROR_CLAMP:
-            return R300_TX_CLAMP | R300_TX_MIRRORED;
-        case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
-            return R300_TX_CLAMP_TO_EDGE | R300_TX_MIRRORED;
-        case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
-            return R300_TX_CLAMP_TO_EDGE | R300_TX_MIRRORED;
-        default:
-            debug_printf("r300: Unknown texture wrap %d", wrap);
-            return 0;
-    }
-}
-
-static uint32_t translate_tex_filters(int min, int mag, int mip) {
-    uint32_t retval = 0;
-    switch (min) {
-        case PIPE_TEX_FILTER_NEAREST:
-            retval |= R300_TX_MIN_FILTER_NEAREST;
-        case PIPE_TEX_FILTER_LINEAR:
-            retval |= R300_TX_MIN_FILTER_LINEAR;
-        case PIPE_TEX_FILTER_ANISO:
-            retval |= R300_TX_MIN_FILTER_ANISO;
-        default:
-            debug_printf("r300: Unknown texture filter %d", min);
-            break;
-    }
-    switch (mag) {
-        case PIPE_TEX_FILTER_NEAREST:
-            retval |= R300_TX_MAG_FILTER_NEAREST;
-        case PIPE_TEX_FILTER_LINEAR:
-            retval |= R300_TX_MAG_FILTER_LINEAR;
-        case PIPE_TEX_FILTER_ANISO:
-            retval |= R300_TX_MAG_FILTER_ANISO;
-        default:
-            debug_printf("r300: Unknown texture filter %d", mag);
-            break;
-    }
-    switch (mip) {
-        case PIPE_TEX_MIPFILTER_NONE:
-            retval |= R300_TX_MIN_FILTER_MIP_NONE;
-        case PIPE_TEX_MIPFILTER_NEAREST:
-            retval |= R300_TX_MIN_FILTER_MIP_NEAREST;
-        case PIPE_TEX_MIPFILTER_LINEAR:
-            retval |= R300_TX_MIN_FILTER_MIP_LINEAR;
-        default:
-            debug_printf("r300: Unknown texture filter %d", mip);
-            break;
-    }
-
-    return retval;
-}
-
-static uint32_t anisotropy(float max_aniso) {
-    if (max_aniso >= 16.0f) {
-        return R300_TX_MAX_ANISO_16_TO_1;
-    } else if (max_aniso >= 8.0f) {
-        return R300_TX_MAX_ANISO_8_TO_1;
-    } else if (max_aniso >= 4.0f) {
-        return R300_TX_MAX_ANISO_4_TO_1;
-    } else if (max_aniso >= 2.0f) {
-        return R300_TX_MAX_ANISO_2_TO_1;
-    } else {
-        return R300_TX_MAX_ANISO_1_TO_1;
-    }
-}
-
 static void*
         r300_create_sampler_state(struct pipe_context* pipe,
                                   const struct pipe_sampler_state* state)
@@ -653,19 +415,19 @@ static void*
     int lod_bias;
 
     sampler->filter0 |=
-        (translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) |
-        (translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) |
-        (translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT);
+        (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) |
+        (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) |
+        (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT);
 
-    sampler->filter0 |= translate_tex_filters(state->min_img_filter,
-                                              state->mag_img_filter,
-                                              state->min_mip_filter);
+    sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter,
+                                                   state->mag_img_filter,
+                                                   state->min_mip_filter);
 
     lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1);
 
     sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT;
 
-    sampler->filter1 |= anisotropy(state->max_anisotropy);
+    sampler->filter1 |= r300_anisotropy(state->max_anisotropy);
 
     util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM,
                     &sampler->border_color);
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c
index a51904096f2..548a840f25d 100644
--- a/src/gallium/drivers/r300/r300_state_derived.c
+++ b/src/gallium/drivers/r300/r300_state_derived.c
@@ -57,34 +57,40 @@ static uint32_t translate_vertex_data_type(int type) {
 /* XXX this function should be able to handle vert shaders as well as draw */
 static void r300_update_vertex_layout(struct r300_context* r300)
 {
+    struct r300_vertex_format vformat;
     struct vertex_info vinfo;
-    boolean pos = false, psize = false, fog = false;
+    boolean pos = FALSE, psize = FALSE, fog = FALSE;
     int i, texs = 0, cols = 0;
+    int tab[16];
 
     struct tgsi_shader_info* info = &r300->fs->info;
+
     memset(&vinfo, 0, sizeof(vinfo));
+    for (i = 0; i < 16; i++) {
+        tab[i] = -1;
+    }
 
     assert(info->num_inputs <= 16);
 
-    /* This is rather lame. Since draw_find_vs_output doesn't return an error
-     * when it can't find an output, we have to pre-iterate and count each
-     * output ourselves. */
     for (i = 0; i < info->num_inputs; i++) {
         switch (info->input_semantic_name[i]) {
             case TGSI_SEMANTIC_POSITION:
-                pos = true;
+                pos = TRUE;
+                tab[i] = 0;
                 break;
             case TGSI_SEMANTIC_COLOR:
-                cols++;
+                tab[i] = 2 + cols++;
                 break;
             case TGSI_SEMANTIC_FOG:
-                fog = true;
+                fog = TRUE;
+                tab[i] = 6 + texs++;
                 break;
             case TGSI_SEMANTIC_PSIZE:
-                psize = true;
+                psize = TRUE;
+                tab[i] = 1;
                 break;
             case TGSI_SEMANTIC_GENERIC:
-                texs++;
+                tab[i] = 6 + texs++;
                 break;
             default:
                 debug_printf("r300: Unknown vertex input %d\n",
@@ -105,6 +111,12 @@ static void r300_update_vertex_layout(struct r300_context* r300)
 
     if (!pos) {
         debug_printf("r300: Forcing vertex position attribute emit...\n");
+        /* Make room for the position attribute
+         * at the beginning of the tab. */
+        for (i = 1; i < 16; i++) {
+            tab[i] = tab[i-1];
+        }
+        tab[0] = 0;
     }
 
     draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_POS,
@@ -125,16 +137,17 @@ static void r300_update_vertex_layout(struct r300_context* r300)
         vinfo.hwfmt[2] |= (R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i);
     }
 
-    if (fog) {
+    for (i = 0; i < texs; i++) {
         draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_PERSPECTIVE,
-            draw_find_vs_output(r300->draw, TGSI_SEMANTIC_FOG, 0));
-        vinfo.hwfmt[2] |=
-            (R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << cols);
+            draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i));
+        vinfo.hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
+        vinfo.hwfmt[3] |= (4 << (3 * i));
     }
 
-    for (i = 0; i < texs; i++) {
-        draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_LINEAR,
-            draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i));
+    if (fog) {
+        i++;
+        draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_PERSPECTIVE,
+            draw_find_vs_output(r300->draw, TGSI_SEMANTIC_FOG, 0));
         vinfo.hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
         vinfo.hwfmt[3] |= (4 << (3 * i));
     }
@@ -152,18 +165,29 @@ static void r300_update_vertex_layout(struct r300_context* r300)
          (0xf << R300_WRITE_ENA_SHIFT))
 
         for (i = 0; i < vinfo.num_attribs; i++) {
+            /* Make sure we have a proper destination for our attribute */
+            if (tab[i] == -1) {
+                debug_printf("attrib count: %d, fp input count: %d\n",
+                        vinfo.num_attribs, info->num_inputs);
+                for (i = 0; i < vinfo.num_attribs; i++) {
+                    debug_printf("attrib: offset %d, interp %d, size %d,"
+                           " tab %d\n", vinfo.attrib[i].src_index,
+                           vinfo.attrib[i].interp_mode, vinfo.attrib[i].emit,
+                           tab[i]);
+                }
+                assert(0);
+            }
+
             temp = translate_vertex_data_type(vinfo.attrib[i].emit) |
-                R300_SIGNED;
+                (tab[i] << R300_DST_VEC_LOC_SHIFT) | R300_SIGNED;
             if (i & 1) {
                 r300->vertex_info.vap_prog_stream_cntl[i >> 1] &= 0xffff0000;
                 r300->vertex_info.vap_prog_stream_cntl[i >> 1] |=
-                        (translate_vertex_data_type(vinfo.attrib[i].emit) |
-                        R300_SIGNED) << 16;
+                        temp << 16;
             } else {
                 r300->vertex_info.vap_prog_stream_cntl[i >> 1] &= 0xffff;
                 r300->vertex_info.vap_prog_stream_cntl[i >> 1] |=
-                    translate_vertex_data_type(vinfo.attrib[i].emit) |
-                    R300_SIGNED;
+                    temp;
             }
 
             r300->vertex_info.vap_prog_stream_cntl_ext[i >> 1] |=
@@ -180,8 +204,85 @@ static void r300_update_vertex_layout(struct r300_context* r300)
 /* Set up the RS block. This is the part of the chipset that actually does
  * the rasterization of vertices into fragments. This is also the part of the
  * chipset that locks up if any part of it is even slightly wrong. */
-void r300_update_rs_block(struct r300_context* r300)
+static void r300_update_rs_block(struct r300_context* r300)
 {
+    struct r300_rs_block* rs = r300->rs_block;
+    struct vertex_info* vinfo = &r300->vertex_info.vinfo;
+    int col_count = 0, fp_offset = 0, i, tex_count = 0;
+
+    memset(rs, 0, sizeof(struct r300_rs_block));
+
+    if (r300_screen(r300->context.screen)->caps->is_r500) {
+        for (i = 0; i < vinfo->num_attribs; i++) {
+            switch (vinfo->attrib[i].interp_mode) {
+                case INTERP_LINEAR:
+                    rs->ip[col_count] |=
+                        R500_RS_COL_PTR(vinfo->attrib[i].src_index) |
+                        R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
+                    col_count++;
+                    break;
+                case INTERP_PERSPECTIVE:
+                    rs->ip[tex_count] |=
+                        R500_RS_SEL_S(vinfo->attrib[i].src_index) |
+                        R500_RS_SEL_T(vinfo->attrib[i].src_index + 1) |
+                        R500_RS_SEL_R(vinfo->attrib[i].src_index + 2) |
+                        R500_RS_SEL_Q(vinfo->attrib[i].src_index + 3);
+                    tex_count++;
+                    break;
+            }
+        }
+
+        for (i = 0; i < tex_count; i++) {
+            rs->inst[i] |= R500_RS_INST_TEX_ID(i) | R500_RS_INST_TEX_CN_WRITE |
+                R500_RS_INST_TEX_ADDR(fp_offset);
+            fp_offset++;
+        }
+
+        for (i = 0; i < col_count; i++) {
+            rs->inst[i] |= R500_RS_INST_COL_ID(i) | R500_RS_INST_COL_CN_WRITE |
+                R500_RS_INST_COL_ADDR(fp_offset);
+            fp_offset++;
+        }
+
+        rs->inst_count = MAX2(col_count, tex_count);
+    } else {
+        for (i = 0; i < vinfo->num_attribs; i++) {
+            switch (vinfo->attrib[i].interp_mode) {
+                case INTERP_LINEAR:
+                    rs->ip[col_count] |=
+                        R300_RS_COL_PTR(vinfo->attrib[i].src_index) |
+                        R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
+                    col_count++;
+                    break;
+                case INTERP_PERSPECTIVE:
+                    rs->ip[tex_count] |=
+                        R300_RS_TEX_PTR(vinfo->attrib[i].src_index) |
+                        R300_RS_SEL_S(R300_RS_SEL_C0) |
+                        R300_RS_SEL_T(R300_RS_SEL_C1) |
+                        R300_RS_SEL_R(R300_RS_SEL_C2) |
+                        R300_RS_SEL_Q(R300_RS_SEL_C3);
+                    tex_count += 4;
+                    break;
+            }
+        }
+
+        for (i = 0; i < tex_count; i++) {
+            rs->inst[i] |= R300_RS_INST_TEX_ID(i) | R300_RS_INST_TEX_CN_WRITE |
+                R300_RS_INST_TEX_ADDR(fp_offset);
+            fp_offset++;
+        }
+
+        for (i = 0; i < col_count; i++) {
+            rs->inst[i] |= R300_RS_INST_COL_ID(i) | R300_RS_INST_COL_CN_WRITE |
+                R300_RS_INST_COL_ADDR(fp_offset);
+            fp_offset++;
+        }
+    }
+
+    rs->count = (tex_count * 4) | (col_count << R300_IC_COUNT_SHIFT) |
+        R300_HIRES_EN;
+
+    rs->inst_count = MAX2(col_count, tex_count);
 }
 
 void r300_update_derived_state(struct r300_context* r300)
diff --git a/src/gallium/drivers/r300/r300_state_inlines.h b/src/gallium/drivers/r300/r300_state_inlines.h
index 005fb74ed6c..630ac3b2fc4 100644
--- a/src/gallium/drivers/r300/r300_state_inlines.h
+++ b/src/gallium/drivers/r300/r300_state_inlines.h
@@ -1,5 +1,6 @@
 /*
  * Copyright 2009 Joakim Sindholt <[email protected]>
+ *                Corbin Simpson <[email protected]>
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -27,56 +28,319 @@
 
 #include "r300_reg.h"
 
+/* Some maths. These should probably find their way to u_math, if needed. */
+
+static INLINE int pack_float_16_6x(float f) {
+    return ((int)(f * 6.0) & 0xffff);
+}
+
+/* Blend state. */
+
+static INLINE uint32_t r300_translate_blend_function(int blend_func)
+{
+    switch (blend_func) {
+        case PIPE_BLEND_ADD:
+            return R300_COMB_FCN_ADD_CLAMP;
+        case PIPE_BLEND_SUBTRACT:
+            return R300_COMB_FCN_SUB_CLAMP;
+        case PIPE_BLEND_REVERSE_SUBTRACT:
+            return R300_COMB_FCN_RSUB_CLAMP;
+        case PIPE_BLEND_MIN:
+            return R300_COMB_FCN_MIN;
+        case PIPE_BLEND_MAX:
+            return R300_COMB_FCN_MAX;
+        default:
+            debug_printf("r300: Unknown blend function %d\n", blend_func);
+            break;
+    }
+    return 0;
+}
+
+/* XXX we can also offer the D3D versions of some of these... */
+static INLINE uint32_t r300_translate_blend_factor(int blend_fact)
+{
+    switch (blend_fact) {
+        case PIPE_BLENDFACTOR_ONE:
+            return R300_BLEND_GL_ONE;
+        case PIPE_BLENDFACTOR_SRC_COLOR:
+            return R300_BLEND_GL_SRC_COLOR;
+        case PIPE_BLENDFACTOR_SRC_ALPHA:
+            return R300_BLEND_GL_SRC_ALPHA;
+        case PIPE_BLENDFACTOR_DST_ALPHA:
+            return R300_BLEND_GL_DST_ALPHA;
+        case PIPE_BLENDFACTOR_DST_COLOR:
+            return R300_BLEND_GL_DST_COLOR;
+        case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+            return R300_BLEND_GL_SRC_ALPHA_SATURATE;
+        case PIPE_BLENDFACTOR_CONST_COLOR:
+            return R300_BLEND_GL_CONST_COLOR;
+        case PIPE_BLENDFACTOR_CONST_ALPHA:
+            return R300_BLEND_GL_CONST_ALPHA;
+        /* XXX WTF are these?
+        case PIPE_BLENDFACTOR_SRC1_COLOR:
+        case PIPE_BLENDFACTOR_SRC1_ALPHA: */
+        case PIPE_BLENDFACTOR_ZERO:
+            return R300_BLEND_GL_ZERO;
+        case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+            return R300_BLEND_GL_ONE_MINUS_SRC_COLOR;
+        case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+            return R300_BLEND_GL_ONE_MINUS_SRC_ALPHA;
+        case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+            return R300_BLEND_GL_ONE_MINUS_DST_ALPHA;
+        case PIPE_BLENDFACTOR_INV_DST_COLOR:
+            return R300_BLEND_GL_ONE_MINUS_DST_COLOR;
+        case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+            return R300_BLEND_GL_ONE_MINUS_CONST_COLOR;
+        case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+            return R300_BLEND_GL_ONE_MINUS_CONST_ALPHA;
+        /* XXX see above
+        case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
+        case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: */
+        default:
+            debug_printf("r300: Unknown blend factor %d\n", blend_fact);
+            break;
+    }
+    return 0;
+}
+
+/* DSA state. */
+
+static INLINE uint32_t r300_translate_depth_stencil_function(int zs_func)
+{
+    switch (zs_func) {
+        case PIPE_FUNC_NEVER:
+            return R300_ZS_NEVER;
+        case PIPE_FUNC_LESS:
+            return R300_ZS_LESS;
+        case PIPE_FUNC_EQUAL:
+            return R300_ZS_EQUAL;
+        case PIPE_FUNC_LEQUAL:
+            return R300_ZS_LEQUAL;
+        case PIPE_FUNC_GREATER:
+            return R300_ZS_GREATER;
+        case PIPE_FUNC_NOTEQUAL:
+            return R300_ZS_NOTEQUAL;
+        case PIPE_FUNC_GEQUAL:
+            return R300_ZS_GEQUAL;
+        case PIPE_FUNC_ALWAYS:
+            return R300_ZS_ALWAYS;
+        default:
+            debug_printf("r300: Unknown depth/stencil function %d\n",
+                zs_func);
+            break;
+    }
+    return 0;
+}
+
+static INLINE uint32_t r300_translate_stencil_op(int s_op)
+{
+    switch (s_op) {
+        case PIPE_STENCIL_OP_KEEP:
+            return R300_ZS_KEEP;
+        case PIPE_STENCIL_OP_ZERO:
+            return R300_ZS_ZERO;
+        case PIPE_STENCIL_OP_REPLACE:
+            return R300_ZS_REPLACE;
+        case PIPE_STENCIL_OP_INCR:
+            return R300_ZS_INCR;
+        case PIPE_STENCIL_OP_DECR:
+            return R300_ZS_DECR;
+        case PIPE_STENCIL_OP_INCR_WRAP:
+            return R300_ZS_INCR_WRAP;
+        case PIPE_STENCIL_OP_DECR_WRAP:
+            return R300_ZS_DECR_WRAP;
+        case PIPE_STENCIL_OP_INVERT:
+            return R300_ZS_INVERT;
+        default:
+            debug_printf("r300: Unknown stencil op %d", s_op);
+            break;
+    }
+    return 0;
+}
+
+static INLINE uint32_t r300_translate_alpha_function(int alpha_func)
+{
+    switch (alpha_func) {
+        case PIPE_FUNC_NEVER:
+            return R300_FG_ALPHA_FUNC_NEVER;
+        case PIPE_FUNC_LESS:
+            return R300_FG_ALPHA_FUNC_LESS;
+        case PIPE_FUNC_EQUAL:
+            return R300_FG_ALPHA_FUNC_EQUAL;
+        case PIPE_FUNC_LEQUAL:
+            return R300_FG_ALPHA_FUNC_LE;
+        case PIPE_FUNC_GREATER:
+            return R300_FG_ALPHA_FUNC_GREATER;
+        case PIPE_FUNC_NOTEQUAL:
+            return R300_FG_ALPHA_FUNC_NOTEQUAL;
+        case PIPE_FUNC_GEQUAL:
+            return R300_FG_ALPHA_FUNC_GE;
+        case PIPE_FUNC_ALWAYS:
+            return R300_FG_ALPHA_FUNC_ALWAYS;
+        default:
+            debug_printf("r300: Unknown alpha function %d", alpha_func);
+            break;
+    }
+    return 0;
+}
+
+/* Texture sampler state. */
+
+static INLINE uint32_t r300_translate_wrap(int wrap)
+{
+    switch (wrap) {
+        case PIPE_TEX_WRAP_REPEAT:
+            return R300_TX_REPEAT;
+        case PIPE_TEX_WRAP_CLAMP:
+            return R300_TX_CLAMP;
+        case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+            return R300_TX_CLAMP_TO_EDGE;
+        case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+            return R300_TX_CLAMP_TO_BORDER;
+        case PIPE_TEX_WRAP_MIRROR_REPEAT:
+            return R300_TX_REPEAT | R300_TX_MIRRORED;
+        case PIPE_TEX_WRAP_MIRROR_CLAMP:
+            return R300_TX_CLAMP | R300_TX_MIRRORED;
+        case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+            return R300_TX_CLAMP_TO_EDGE | R300_TX_MIRRORED;
+        case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+            return R300_TX_CLAMP_TO_EDGE | R300_TX_MIRRORED;
+        default:
+            debug_printf("r300: Unknown texture wrap %d", wrap);
+            return 0;
+    }
+}
+
+static INLINE uint32_t r300_translate_tex_filters(int min, int mag, int mip)
+{
+    uint32_t retval = 0;
+    switch (min) {
+        case PIPE_TEX_FILTER_NEAREST:
+            retval |= R300_TX_MIN_FILTER_NEAREST;
+        case PIPE_TEX_FILTER_LINEAR:
+            retval |= R300_TX_MIN_FILTER_LINEAR;
+        case PIPE_TEX_FILTER_ANISO:
+            retval |= R300_TX_MIN_FILTER_ANISO;
+        default:
+            debug_printf("r300: Unknown texture filter %d", min);
+            break;
+    }
+    switch (mag) {
+        case PIPE_TEX_FILTER_NEAREST:
+            retval |= R300_TX_MAG_FILTER_NEAREST;
+        case PIPE_TEX_FILTER_LINEAR:
+            retval |= R300_TX_MAG_FILTER_LINEAR;
+        case PIPE_TEX_FILTER_ANISO:
+            retval |= R300_TX_MAG_FILTER_ANISO;
+        default:
+            debug_printf("r300: Unknown texture filter %d", mag);
+            break;
+    }
+    switch (mip) {
+        case PIPE_TEX_MIPFILTER_NONE:
+            retval |= R300_TX_MIN_FILTER_MIP_NONE;
+        case PIPE_TEX_MIPFILTER_NEAREST:
+            retval |= R300_TX_MIN_FILTER_MIP_NEAREST;
+        case PIPE_TEX_MIPFILTER_LINEAR:
+            retval |= R300_TX_MIN_FILTER_MIP_LINEAR;
+        default:
+            debug_printf("r300: Unknown texture filter %d", mip);
+            break;
+    }
+
+    return retval;
+}
+
+static INLINE uint32_t r300_anisotropy(float max_aniso)
+{
+    if (max_aniso >= 16.0f) {
+        return R300_TX_MAX_ANISO_16_TO_1;
+    } else if (max_aniso >= 8.0f) {
+        return R300_TX_MAX_ANISO_8_TO_1;
+    } else if (max_aniso >= 4.0f) {
+        return R300_TX_MAX_ANISO_4_TO_1;
+    } else if (max_aniso >= 2.0f) {
+        return R300_TX_MAX_ANISO_2_TO_1;
+    } else {
+        return R300_TX_MAX_ANISO_1_TO_1;
+    }
+}
+
+/* Buffer formats. */
+
 static INLINE uint32_t r300_translate_colorformat(enum pipe_format format)
 {
     switch (format) {
-    case PIPE_FORMAT_A8R8G8B8_UNORM:
-        return R300_COLOR_FORMAT_ARGB8888;
-    case PIPE_FORMAT_I8_UNORM:
-        return R300_COLOR_FORMAT_I8;
-    case PIPE_FORMAT_A1R5G5B5_UNORM:
-        return R300_COLOR_FORMAT_ARGB1555;
-    case PIPE_FORMAT_R5G6B5_UNORM:
-        return R300_COLOR_FORMAT_RGB565;
-    /* XXX Not in pipe_format
-    case PIPE_FORMAT_A32R32G32B32:
-        return R300_COLOR_FORMAT_ARGB32323232;
-    case PIPE_FORMAT_A16R16G16B16:
-        return R300_COLOR_FORMAT_ARGB16161616; */
-    case PIPE_FORMAT_A4R4G4B4_UNORM:
-        return R300_COLOR_FORMAT_ARGB4444;
-    /* XXX Not in pipe_format
-    case PIPE_FORMAT_A10R10G10B10_UNORM:
-        return R500_COLOR_FORMAT_ARGB10101010;
-    case PIPE_FORMAT_A2R10G10B10_UNORM:
-        return R500_COLOR_FORMAT_ARGB2101010;
-    case PIPE_FORMAT_I10_UNORM:
-        return R500_COLOR_FORMAT_I10; */
-    default:
-        debug_printf("r300: Implementation error: " \
-            "Got unsupported color format %s in %s\n",
-            pf_name(format), __FUNCTION__);
-        break;
+        /* 8-bit buffers */
+        case PIPE_FORMAT_I8_UNORM:
+            return R300_COLOR_FORMAT_I8;
+        /* 16-bit buffers */
+        case PIPE_FORMAT_R5G6B5_UNORM:
+            return R300_COLOR_FORMAT_RGB565;
+        case PIPE_FORMAT_A1R5G5B5_UNORM:
+            return R300_COLOR_FORMAT_ARGB1555;
+        case PIPE_FORMAT_A4R4G4B4_UNORM:
+            return R300_COLOR_FORMAT_ARGB4444;
+        /* 32-bit buffers */
+        case PIPE_FORMAT_A8R8G8B8_UNORM:
+            return R300_COLOR_FORMAT_ARGB8888;
+        /* XXX Not in pipe_format
+        case PIPE_FORMAT_A32R32G32B32:
+            return R300_COLOR_FORMAT_ARGB32323232;
+        case PIPE_FORMAT_A16R16G16B16:
+            return R300_COLOR_FORMAT_ARGB16161616; */
+        /* XXX Not in pipe_format
+        case PIPE_FORMAT_A10R10G10B10_UNORM:
+            return R500_COLOR_FORMAT_ARGB10101010;
+        case PIPE_FORMAT_A2R10G10B10_UNORM:
+            return R500_COLOR_FORMAT_ARGB2101010;
+        case PIPE_FORMAT_I10_UNORM:
+            return R500_COLOR_FORMAT_I10; */
+        default:
+            debug_printf("r300: Implementation error: " \
+                "Got unsupported color format %s in %s\n",
+                pf_name(format), __FUNCTION__);
+            break;
     }
-    
     return 0;
 }
 
 static INLINE uint32_t r300_translate_zsformat(enum pipe_format format)
 {
     switch (format) {
-    case PIPE_FORMAT_Z16_UNORM:
-        return R300_DEPTHFORMAT_16BIT_INT_Z;
-    /* XXX R300_DEPTHFORMAT_16BIT_13E3 anyone? */
-    case PIPE_FORMAT_Z24S8_UNORM:
-        return R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL;
-    default:
-        debug_printf("r300: Implementation error: " \
-            "Got unsupported ZS format %s in %s\n",
-            pf_name(format), __FUNCTION__);
-        break;
+        /* 16-bit depth, no stencil */
+        case PIPE_FORMAT_Z16_UNORM:
+            return R300_DEPTHFORMAT_16BIT_INT_Z;
+        /* 24-bit depth, 8-bit stencil */
+        case PIPE_FORMAT_Z24S8_UNORM:
+            return R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL;
+        default:
+            debug_printf("r300: Implementation error: " \
+                "Got unsupported ZS format %s in %s\n",
+                pf_name(format), __FUNCTION__);
+            break;
+    }
+    return 0;
+}
+
+/* Non-CSO state. (For now.) */
+
+static INLINE uint32_t r300_translate_gb_pipes(int pipe_count)
+{
+    switch (pipe_count) {
+        case 1:
+            return R300_GB_TILE_PIPE_COUNT_RV300;
+            break;
+        case 2:
+            return R300_GB_TILE_PIPE_COUNT_R300;
+            break;
+        case 3:
+            return R300_GB_TILE_PIPE_COUNT_R420_3P;
+            break;
+        case 4:
+            return R300_GB_TILE_PIPE_COUNT_R420;
+            break;
     }
-    
     return 0;
 }
 
diff --git a/src/gallium/drivers/r300/r300_state_invariant.c b/src/gallium/drivers/r300/r300_state_invariant.c
new file mode 100644
index 00000000000..7fd7aefeb72
--- /dev/null
+++ b/src/gallium/drivers/r300/r300_state_invariant.c
@@ -0,0 +1,54 @@
+/*
+ * Copyright 2009 Joakim Sindholt <[email protected]>
+ *                Corbin Simpson <[email protected]>
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE. */
+
+#include "r300_state_invariant.h"
+
+/* Calculate and emit invariant state. This is data that the 3D engine
+ * will probably want at the beginning of every CS, but it's not currently
+ * handled by any CSO setup, and in addition it doesn't really change much.
+ *
+ * Note that eventually this should be empty, but it's useful for development
+ * and general unduplication of code. */
+void r300_emit_invariant_state(struct r300_context* r300)
+{
+    struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps;
+    CS_LOCALS(r300);
+
+    BEGIN_CS(14);
+    /* Amount of time to wait for vertex fetches in PVS */
+    OUT_CS_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
+    /* Various GB enables */
+    OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
+        R300_GB_LINE_STUFF_ENABLE | R300_GB_TRIANGLE_STUFF_ENABLE);
+    /* Subpixel multisampling for AA */
+    OUT_CS_REG(R300_GB_MSPOS0, 0x66666666);
+    OUT_CS_REG(R300_GB_MSPOS1, 0x66666666);
+    /* GB tile config and pipe setup */
+    OUT_CS_REG(R300_GB_TILE_CONFIG, R300_GB_TILE_DISABLE |
+        r300_translate_gb_pipes(caps->num_frag_pipes));
+    /* Source of fog depth */
+    OUT_CS_REG(R300_GB_SELECT, R300_GB_FOG_SELECT_1_1_W);
+    /* AA enable */
+    OUT_CS_REG(R300_GB_AA_CONFIG, 0x0);
+    END_CS;
+}
diff --git a/src/gallium/drivers/r300/r300_state_invariant.h b/src/gallium/drivers/r300/r300_state_invariant.h
new file mode 100644
index 00000000000..8204bf9588b
--- /dev/null
+++ b/src/gallium/drivers/r300/r300_state_invariant.h
@@ -0,0 +1,33 @@
+/*
+ * Copyright 2008 Corbin Simpson <[email protected]>
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE. */
+
+#ifndef R300_STATE_INVARIANT_H
+#define R300_STATE_INVARIANT_H
+
+#include "r300_context.h"
+#include "r300_cs.h"
+#include "r300_reg.h"
+#include "r300_state_inlines.h"
+
+void r300_emit_invariant_state(struct r300_context* r300);
+
+#endif /* R300_STATE_INVARIANT_H */
diff --git a/src/gallium/drivers/r300/r300_surface.c b/src/gallium/drivers/r300/r300_surface.c
index 49e4a96f833..288d8dea15e 100644
--- a/src/gallium/drivers/r300/r300_surface.c
+++ b/src/gallium/drivers/r300/r300_surface.c
@@ -33,9 +33,10 @@ static void r300_surface_fill(struct pipe_context* pipe,
 {
     struct r300_context* r300 = r300_context(pipe);
     CS_LOCALS(r300);
-    struct r300_capabilities* caps = ((struct r300_screen*)pipe->screen)->caps;
+    struct r300_capabilities* caps = r300_screen(pipe->screen)->caps;
     struct r300_texture* tex = (struct r300_texture*)dest->texture;
     int i;
+
     float r, g, b, a;
     unsigned pixpitch = tex->stride / tex->tex.block.size;
     r = (float)((color >> 16) & 0xff) / 255.0f;
@@ -55,7 +56,14 @@ static void r300_surface_fill(struct pipe_context* pipe,
         return;
     }*/
 
-    BEGIN_CS(163 + (caps->is_r500 ? 22 : 14) + (caps->has_tcl ? 4 : 2));
+    r300_emit_invariant_state(r300);
+
+    r300_emit_blend_state(r300, &blend_clear_state);
+    r300_emit_blend_color_state(r300, &blend_color_clear_state);
+    r300_emit_dsa_state(r300, &dsa_clear_state);
+    r300_emit_rs_state(r300, &rs_clear_state);
+
+    BEGIN_CS(129 + (caps->is_r500 ? 22 : 14) + (caps->has_tcl ? 4 : 2));
     /* Flush PVS. */
     OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0);
 
@@ -69,7 +77,12 @@ static void r300_surface_fill(struct pipe_context* pipe,
     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, 0xFFFFFF);
     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0x0);
     /* XXX endian */
-    OUT_CS_REG(R300_VAP_CNTL_STATUS, R300_VC_NO_SWAP);
+    if (caps->has_tcl) {
+        OUT_CS_REG(R300_VAP_CNTL_STATUS, R300_VC_NO_SWAP);
+    } else {
+        OUT_CS_REG(R300_VAP_CNTL_STATUS, R300_VC_NO_SWAP |
+                R300_VAP_TCL_BYPASS);
+    }
     OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0, 0x0);
     /* XXX magic number not in r300_reg */
     OUT_CS_REG(R300_VAP_PSC_SGN_NORM_CNTL, 0xAAAAAAAA);
@@ -79,18 +92,6 @@ static void r300_surface_fill(struct pipe_context* pipe,
     OUT_CS_32F(1.0);
     OUT_CS_32F(1.0);
     OUT_CS_32F(1.0);
-    /* XXX is this too long? */
-    OUT_CS_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xFFFF);
-    OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
-        R300_GB_LINE_STUFF_ENABLE | R300_GB_TRIANGLE_STUFF_ENABLE);
-    /* XXX more magic numbers */
-    OUT_CS_REG(R300_GB_MSPOS0, 0x66666666);
-    OUT_CS_REG(R300_GB_MSPOS1, 0x66666666);
-    /* XXX why doesn't classic Mesa write the number of pipes, too? */
-    OUT_CS_REG(R300_GB_TILE_CONFIG, R300_GB_TILE_ENABLE |
-        R300_GB_TILE_SIZE_16);
-    OUT_CS_REG(R300_GB_SELECT, R300_GB_FOG_SELECT_1_1_W);
-    OUT_CS_REG(R300_GB_AA_CONFIG, 0x0);
     /* XXX point tex stuffing */
     OUT_CS_REG_SEQ(R300_GA_POINT_S0, 1);
     OUT_CS_32F(0.0);
@@ -102,9 +103,6 @@ static void r300_surface_fill(struct pipe_context* pipe,
     OUT_CS_REG(R300_GA_POINT_MINMAX, 0x6 |
         (0x1800 << R300_GA_POINT_MINMAX_MAX_SHIFT));
     /* XXX this big chunk should be refactored into rs_state */
-    OUT_CS_REG(R300_GA_LINE_CNTL, 0x00030006);
-    OUT_CS_REG(R300_GA_LINE_STIPPLE_CONFIG, 0x3BAAAAAB);
-    OUT_CS_REG(R300_GA_LINE_STIPPLE_VALUE, 0x00000000);
     OUT_CS_REG(R300_GA_LINE_S0, 0x00000000);
     OUT_CS_REG(R300_GA_LINE_S1, 0x3F800000);
     OUT_CS_REG(R300_GA_ENHANCE, 0x00000002);
@@ -117,12 +115,6 @@ static void r300_surface_fill(struct pipe_context* pipe,
     OUT_CS_REG(R300_GA_FOG_SCALE, 0x3DBF1412);
     OUT_CS_REG(R300_GA_FOG_OFFSET, 0x00000000);
     OUT_CS_REG(R300_SU_TEX_WRAP, 0x00000000);
-    OUT_CS_REG(R300_SU_POLY_OFFSET_FRONT_SCALE, 0x00000000);
-    OUT_CS_REG(R300_SU_POLY_OFFSET_FRONT_OFFSET, 0x00000000);
-    OUT_CS_REG(R300_SU_POLY_OFFSET_BACK_SCALE, 0x00000000);
-    OUT_CS_REG(R300_SU_POLY_OFFSET_BACK_OFFSET, 0x00000000);
-    OUT_CS_REG(R300_SU_POLY_OFFSET_ENABLE, 0x00000000);
-    OUT_CS_REG(R300_SU_CULL_MODE, 0x00000000);
     OUT_CS_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF);
     OUT_CS_REG(R300_SU_DEPTH_OFFSET, 0x00000000);
     OUT_CS_REG(R300_SC_HYPERZ, 0x0000001C);
@@ -190,11 +182,6 @@ static void r300_surface_fill(struct pipe_context* pipe,
             R300_PS_UCP_MODE_CLIP_AS_TRIFAN);
     }
 
-    /* The size of the point we're about to draw, in sixths of pixels */
-    OUT_CS_REG(R300_GA_POINT_SIZE,
-        ((h * 6) & R300_POINTSIZE_Y_MASK) |
-        ((w * 6) << R300_POINTSIZE_X_SHIFT));
-
     /* XXX */
     OUT_CS_REG(R300_SC_CLIP_RULE, 0xaaaa);
 
@@ -240,7 +227,7 @@ static void r300_surface_fill(struct pipe_context* pipe,
         r300_emit_fragment_shader(r300, &r300_passthrough_fragment_shader);
     }
 
-    BEGIN_CS(8 + (caps->has_tcl ? 20 : 2));
+    BEGIN_CS(7 + (caps->has_tcl ? 21 : 2));
     OUT_CS_REG_SEQ(R300_US_OUT_FMT_0, 4);
     OUT_CS(R300_C0_SEL_B | R300_C1_SEL_G | R300_C2_SEL_R | R300_C3_SEL_A);
     OUT_CS(R300_US_OUT_FMT_UNUSED);
@@ -277,9 +264,10 @@ static void r300_surface_fill(struct pipe_context* pipe,
     }
     END_CS;
 
-    r300_emit_blend_state(r300, &blend_clear_state);
-    r300_emit_blend_color_state(r300, &blend_color_clear_state);
-    r300_emit_dsa_state(r300, &dsa_clear_state);
+    /* The size of the point we're about to draw, in sixths of pixels */
+    OUT_CS_REG(R300_GA_POINT_SIZE,
+        ((h * 6) & R300_POINTSIZE_Y_MASK) |
+        ((w * 6) << R300_POINTSIZE_X_SHIFT));
 
     BEGIN_CS(24);
     /* Flush colorbuffer and blend caches. */
diff --git a/src/gallium/drivers/r300/r300_surface.h b/src/gallium/drivers/r300/r300_surface.h
index 442eac2cf2e..807aad39e49 100644
--- a/src/gallium/drivers/r300/r300_surface.h
+++ b/src/gallium/drivers/r300/r300_surface.h
@@ -57,4 +57,16 @@ const struct r300_dsa_state dsa_clear_state = {
     .stencil_ref_bf = 0x0,
 };
 
+const struct r300_rs_state rs_clear_state = {
+    .line_control = 0x00030006,
+    .depth_scale_front = 0x0,
+    .depth_offset_front = 0x0,
+    .depth_scale_back = 0x0,
+    .depth_offset_back = 0x0,
+    .polygon_offset_enable = 0x0,
+    .cull_mode = 0x0,
+    .line_stipple_config = 0x3BAAAAAB,
+    .line_stipple_value = 0x0,
+};
+
 #endif /* R300_SURFACE_H */
diff --git a/src/gallium/drivers/r300/r300_swtcl_emit.c b/src/gallium/drivers/r300/r300_swtcl_emit.c
index 5b028aaf7b7..f9baaade1e3 100644
--- a/src/gallium/drivers/r300/r300_swtcl_emit.c
+++ b/src/gallium/drivers/r300/r300_swtcl_emit.c
@@ -27,6 +27,7 @@
 #include "r300_cs.h"
 #include "r300_context.h"
 #include "r300_reg.h"
+#include "r300_state_derived.h"
 
 /* r300_swtcl_emit: Vertex and index buffer primitive emission. No HW TCL. */
 
@@ -91,9 +92,9 @@ static boolean r300_swtcl_render_allocate_vertices(struct vbuf_render* render,
     r300render->vertex_size = vertex_size;
 
     if (r300render->vbo) {
-        return true;
+        return TRUE;
     } else {
-        return false;
+        return FALSE;
     }
 }
 
@@ -167,11 +168,11 @@ static boolean r300_swtcl_render_set_primitive(struct vbuf_render* render,
             r300render->hwprim = R300_VAP_VF_CNTL__PRIM_POLYGON;
             break;
         default:
-            return false;
+            return FALSE;
             break;
     }
 
-    return true;
+    return TRUE;
 }
 
 static void prepare_render(struct r300_swtcl_render* render)
@@ -191,12 +192,15 @@ static void prepare_render(struct r300_swtcl_render* render)
      * PACKET3 [3D_LOAD_VBPNTR]
      * COUNT   [1]
      * FORMAT  [size | stride << 8]
+     * OFFSET  [0]
      * VBPNTR  [relocated BO]
      */
     BEGIN_CS(5);
     OUT_CS(CP_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, 3));
     OUT_CS(1);
-    OUT_CS(r300->vertex_info.vinfo.size | (r300->vertex_info.vinfo.size << 8));
+    OUT_CS(r300->vertex_info.vinfo.size |
+            (r300->vertex_info.vinfo.size << 8));
+    OUT_CS(render->vbo_offset);
     OUT_CS_RELOC(render->vbo, 0, RADEON_GEM_DOMAIN_GTT, 0, 0);
     END_CS;
 }
@@ -218,7 +222,7 @@ static void r300_swtcl_render_draw_arrays(struct vbuf_render* render,
     BEGIN_CS(2);
     OUT_CS(CP_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0));
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
-           r300render->hwprim | R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
+           r300render->hwprim);
     END_CS;
 }
 
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index edd43706634..8251a597eab 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -31,8 +31,9 @@ static void r300_setup_miptree(struct r300_texture* tex)
 {
     struct pipe_texture* base = &tex->tex;
     int stride, size, offset;
+    int i;
 
-    for (int i = 0; i <= base->last_level; i++) {
+    for (i = 0; i <= base->last_level; i++) {
         if (i > 0) {
             base->width[i] = minify(base->width[i-1]);
             base->height[i] = minify(base->height[i-1]);
diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
index d56eed80a4f..eef6e5806c9 100644
--- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c
+++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
@@ -60,6 +60,7 @@ struct softpipe_vbuf_render
    struct softpipe_context *softpipe;
    uint prim;
    uint vertex_size;
+   uint nr_vertices;
    uint vertex_buffer_size;
    void *vertex_buffer;
 };
@@ -95,12 +96,44 @@ sp_vbuf_allocate_vertices(struct vbuf_render *vbr,
    }
 
    cvbr->vertex_size = vertex_size;
+   cvbr->nr_vertices = nr_vertices;
+   
    return cvbr->vertex_buffer != NULL;
 }
 
 static void
 sp_vbuf_release_vertices(struct vbuf_render *vbr)
 {
+#if 0
+   {
+      struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
+      const struct vertex_info *info = 
+         softpipe_get_vbuf_vertex_info(cvbr->softpipe);
+      const float *vtx = (const float *) cvbr->vertex_buffer;
+      uint i, j;
+      debug_printf("%s (vtx_size = %u,  vtx_used = %u)\n",
+             __FUNCTION__, cvbr->vertex_size, cvbr->nr_vertices);
+      for (i = 0; i < cvbr->nr_vertices; i++) {
+         for (j = 0; j < info->num_attribs; j++) {
+            uint k;
+            switch (info->attrib[j].emit) {
+            case EMIT_4F:  k = 4;   break;
+            case EMIT_3F:  k = 3;   break;
+            case EMIT_2F:  k = 2;   break;
+            case EMIT_1F:  k = 1;   break;
+            default: assert(0);
+            }
+            debug_printf("Vert %u attr %u: ", i, j);
+            while (k-- > 0) {
+               debug_printf("%g ", vtx[0]);
+               vtx++;
+            }
+            debug_printf("\n");
+         }
+      }
+   }
+#endif
+
    /* keep the old allocation for next time */
 }
 
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
index 28a9784b164..142ce230fc0 100644
--- a/src/gallium/drivers/softpipe/sp_texture.c
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -243,9 +243,11 @@ softpipe_get_tex_surface(struct pipe_screen *screen,
       ps->level = level;
       ps->zslice = zslice;
 
-      if (pt->target == PIPE_TEXTURE_CUBE || pt->target == PIPE_TEXTURE_3D) {
-         ps->offset += ((pt->target == PIPE_TEXTURE_CUBE) ? face : zslice) *
-            pt->nblocksy[level] * spt->stride[level];
+      if (pt->target == PIPE_TEXTURE_CUBE) {
+         ps->offset += face * pt->nblocksy[level] * spt->stride[level];
+      }
+      else if (pt->target == PIPE_TEXTURE_3D) {
+         ps->offset += zslice * pt->nblocksy[level] * spt->stride[level];
       }
       else {
          assert(face == 0);
@@ -283,14 +285,13 @@ softpipe_get_tex_transfer(struct pipe_screen *screen,
 {
    struct softpipe_texture *sptex = softpipe_texture(texture);
    struct softpipe_transfer *spt;
-   struct pipe_transfer *pt;
 
    assert(texture);
    assert(level <= texture->last_level);
 
    spt = CALLOC_STRUCT(softpipe_transfer);
-   pt = &spt->base;
    if (spt) {
+      struct pipe_transfer *pt = &spt->base;
       pt->refcount = 1;
       pipe_texture_reference(&pt->texture, texture);
       pt->format = texture->format;
@@ -302,23 +303,26 @@ softpipe_get_tex_transfer(struct pipe_screen *screen,
       pt->nblocksx = texture->nblocksx[level];
       pt->nblocksy = texture->nblocksy[level];
       pt->stride = sptex->stride[level];
-      spt->offset = sptex->level_offset[level];
       pt->usage = usage;
       pt->face = face;
       pt->level = level;
       pt->zslice = zslice;
 
-      if (texture->target == PIPE_TEXTURE_CUBE ||
-          texture->target == PIPE_TEXTURE_3D) {
-         spt->offset += ((texture->target == PIPE_TEXTURE_CUBE) ? face :
-                         zslice) * pt->nblocksy * pt->stride;
+      spt->offset = sptex->level_offset[level];
+
+      if (texture->target == PIPE_TEXTURE_CUBE) {
+         spt->offset += face * pt->nblocksy * pt->stride;
+      }
+      else if (texture->target == PIPE_TEXTURE_3D) {
+         spt->offset += zslice * pt->nblocksy * pt->stride;
       }
       else {
          assert(face == 0);
          assert(zslice == 0);
       }
+      return pt;
    }
-   return pt;
+   return NULL;
 }
 
 
diff --git a/src/gallium/drivers/trace/tr_dump.c b/src/gallium/drivers/trace/tr_dump.c
index d98cef221b7..8757ac8bec9 100644
--- a/src/gallium/drivers/trace/tr_dump.c
+++ b/src/gallium/drivers/trace/tr_dump.c
@@ -40,7 +40,7 @@
 
 #include "pipe/p_config.h"
 
-#if defined(PIPE_OS_LINUX)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD)
 #include <stdlib.h>
 #endif
 
@@ -236,7 +236,7 @@ boolean trace_dump_trace_begin()
       trace_dump_writes("<?xml-stylesheet type='text/xsl' href='trace.xsl'?>\n");
       trace_dump_writes("<trace version='0.1'>\n");
       
-#if defined(PIPE_OS_LINUX)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD)
       /* Linux applications rarely cleanup GL / Gallium resources so catch 
        * application exit here */ 
       atexit(trace_dump_trace_close);