Merge remote branch 'main/master' into radeon-rewrite

Conflicts:
	src/mesa/drivers/dri/r300/r300_cmdbuf.c
	src/mesa/drivers/dri/r300/r300_state.c
	src/mesa/drivers/dri/r300/r300_swtcl.c
	src/mesa/drivers/dri/r300/radeon_ioctl.c
	src/mesa/drivers/dri/radeon/radeon_screen.c

13 files changed, 418 insertions, 364 deletions

diff --git a/src/gallium/drivers/cell/ppu/cell_clear.c b/src/gallium/drivers/cell/ppu/cell_clear.c
index c2e2769..edc0674 100644
--- a/src/gallium/drivers/cell/ppu/cell_clear.c
+++ b/src/gallium/drivers/cell/ppu/cell_clear.c
@@ -70,18 +70,12 @@ void
 cell_clear_surface(struct pipe_context *pipe, struct pipe_surface *ps,
                    unsigned clearValue)
 {
-   struct pipe_screen *screen = pipe->screen;
    struct cell_context *cell = cell_context(pipe);
    uint surfIndex;
 
    if (cell->dirty)
       cell_update_derived(cell);
 
-
-   if (!cell->cbuf_map[0])
-      cell->cbuf_map[0] = screen->surface_map(screen, ps,
-                                              PIPE_BUFFER_USAGE_GPU_WRITE);
-
    if (ps == cell->framebuffer.zsbuf) {
       /* clear z/stencil buffer */
       surfIndex = 1;
diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
index eb1397b..5c3188e 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.h
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -130,9 +130,6 @@ struct cell_context
    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 867b5dc..13125a9 100644
--- a/src/gallium/drivers/cell/ppu/cell_fence.c
+++ b/src/gallium/drivers/cell/ppu/cell_fence.c
@@ -96,7 +96,7 @@ cell_add_buffer_to_list(struct cell_context *cell,
    struct cell_buffer_node *node = CALLOC_STRUCT(cell_buffer_node);
    /* create new list node which references the buffer, insert at head */
    if (node) {
-      pipe_buffer_reference(ps, &node->buffer, buffer);
+      pipe_buffer_reference(&node->buffer, buffer);
       node->next = list->head;
       list->head = node;
    }
@@ -127,10 +127,10 @@ cell_free_fenced_buffers(struct cell_context *cell,
          pipe_buffer_unmap(ps, node->buffer);
 #if 0
          printf("Unref buffer %p\n", node->buffer);
-         if (node->buffer->refcount == 1)
+         if (node->buffer->reference.count == 1)
             printf("   Delete!\n");
 #endif
-         pipe_buffer_reference(ps, &node->buffer, NULL);
+         pipe_buffer_reference(&node->buffer, NULL);
          FREE(node);
          node = next;
       }
@@ -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 ca358ed..ccd0fef 100644
--- a/src/gallium/drivers/cell/ppu/cell_pipe_state.c
+++ b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
@@ -253,10 +253,13 @@ cell_set_sampler_textures(struct pipe_context *pipe,
    assert(num <= CELL_MAX_SAMPLERS);
 
    for (i = 0; i < CELL_MAX_SAMPLERS; i++) {
-      struct pipe_texture *new_tex = i < num ? texture[i] : NULL;
-      if ((struct pipe_texture *) cell->texture[i] != new_tex) {
+      struct cell_texture *new_tex = cell_texture(i < num ? texture[i] : NULL);
+      struct cell_texture *old_tex = cell->texture[i];
+      if (old_tex != new_tex) {
+
          pipe_texture_reference((struct pipe_texture **) &cell->texture[i],
-                                new_tex);
+                                (struct pipe_texture *) new_tex);
+
          changed |= (1 << i);
       }
    }
@@ -270,6 +273,70 @@ cell_set_sampler_textures(struct pipe_context *pipe,
 }
 
 
+/**
+ * Map color and z/stencil framebuffer surfaces.
+ */
+static void
+cell_map_surfaces(struct cell_context *cell)
+{
+   struct pipe_screen *screen = cell->pipe.screen;
+   uint i;
+
+   for (i = 0; i < 1; i++) {
+      struct pipe_surface *ps = cell->framebuffer.cbufs[i];
+      if (ps) {
+         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) {
+         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));
+      }
+   }
+}
+
+
+/**
+ * Unmap color and z/stencil framebuffer surfaces.
+ */
+static void
+cell_unmap_surfaces(struct cell_context *cell)
+{
+   struct pipe_screen *screen = cell->pipe.screen;
+   uint i;
+
+   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
+      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);
+
+         screen->buffer_unmap(screen, ct->buffer);
+         cell->cbuf_map[i] = 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;
+      }
+   }
+}
+
 
 static void
 cell_set_framebuffer_state(struct pipe_context *pipe,
@@ -278,24 +345,10 @@ cell_set_framebuffer_state(struct pipe_context *pipe,
    struct cell_context *cell = cell_context(pipe);
 
    if (1 /*memcmp(&cell->framebuffer, fb, sizeof(*fb))*/) {
-      struct pipe_surface *csurf = fb->cbufs[0];
-      struct pipe_surface *zsurf = fb->zsbuf;
       uint i;
-      uint flags = (PIPE_BUFFER_USAGE_GPU_WRITE |
-                    PIPE_BUFFER_USAGE_GPU_READ);
 
       /* unmap old surfaces */
-      for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
-         if (cell->framebuffer.cbufs[i] && cell->cbuf_map[i]) {
-            pipe_surface_unmap(cell->framebuffer.cbufs[i]);
-            cell->cbuf_map[i] = NULL;
-         }
-      }
-
-      if (cell->framebuffer.zsbuf && cell->zsbuf_map) {
-         pipe_surface_unmap(cell->framebuffer.zsbuf);
-         cell->zsbuf_map = NULL;
-      }
+      cell_unmap_surfaces(cell);
 
       /* Finish any pending rendering to the current surface before
        * installing a new surface!
@@ -314,11 +367,7 @@ cell_set_framebuffer_state(struct pipe_context *pipe,
       pipe_surface_reference(&cell->framebuffer.zsbuf, fb->zsbuf);
 
       /* map new surfaces */
-      if (csurf)
-         cell->cbuf_map[0] = pipe_surface_map(csurf, flags);
-
-      if (zsurf)
-         cell->zsbuf_map = pipe_surface_map(zsurf, flags);
+      cell_map_surfaces(cell);
 
       cell->dirty |= CELL_NEW_FRAMEBUFFER;
    }
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index ff529fe..9479c08 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_state_shader.c b/src/gallium/drivers/cell/ppu/cell_state_shader.c
index bf517ea..6568c78 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_shader.c
@@ -193,9 +193,7 @@ cell_set_constant_buffer(struct pipe_context *pipe,
    draw_flush(cell->draw);
 
    /* note: reference counting */
-   pipe_buffer_reference(pipe->screen,
-                         &cell->constants[shader].buffer,
-                         buf->buffer);
+   pipe_buffer_reference(&cell->constants[shader].buffer, buf->buffer);
 
    if (shader == PIPE_SHADER_VERTEX)
       cell->dirty |= CELL_NEW_VS_CONSTANTS;
diff --git a/src/gallium/drivers/cell/ppu/cell_surface.c b/src/gallium/drivers/cell/ppu/cell_surface.c
index c9203fe..ffb8595 100644
--- a/src/gallium/drivers/cell/ppu/cell_surface.c
+++ b/src/gallium/drivers/cell/ppu/cell_surface.c
@@ -30,9 +30,21 @@
 #include "cell_surface.h"
 
 
+static void
+cell_surface_copy(struct pipe_context *pipe,
+                  struct pipe_surface *dest, unsigned destx, unsigned desty,
+                  struct pipe_surface *src, unsigned srcx, unsigned srcy,
+                  unsigned width, unsigned height)
+{
+   util_surface_copy(pipe, FALSE,
+                     dest, destx, desty,
+                     src, srcx, srcy,
+                     width, height);
+}
+
 void
 cell_init_surface_functions(struct cell_context *cell)
 {
-   cell->pipe.surface_copy = util_surface_copy;
+   cell->pipe.surface_copy = cell_surface_copy;
    cell->pipe.surface_fill = util_surface_fill;
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
index 9ba995a..e265944 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);
    }
@@ -101,18 +101,17 @@ static struct pipe_texture *
 cell_texture_create(struct pipe_screen *screen,
                     const struct pipe_texture *templat)
 {
-   struct pipe_winsys *ws = screen->winsys;
    struct cell_texture *ct = CALLOC_STRUCT(cell_texture);
    if (!ct)
       return NULL;
 
    ct->base = *templat;
-   ct->base.refcount = 1;
+   pipe_reference_init(&ct->base.reference, 1);
    ct->base.screen = screen;
 
    cell_texture_layout(ct);
 
-   ct->buffer = ws->buffer_create(ws, 32, PIPE_BUFFER_USAGE_PIXEL,
+   ct->buffer = screen->buffer_create(screen, 32, PIPE_BUFFER_USAGE_PIXEL,
                                    ct->buffer_size);
 
    if (!ct->buffer) {
@@ -125,42 +124,18 @@ cell_texture_create(struct pipe_screen *screen,
 
 
 static void
-cell_texture_release(struct pipe_screen *screen,
-                     struct pipe_texture **pt)
+cell_texture_destroy(struct pipe_texture *pt)
 {
-   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;
+   struct cell_texture *ct = cell_texture(pt);
 
-      /*
-      DBG("%s deleting %p\n", __FUNCTION__, (void *) ct);
-      */
+   if (ct->mapped) {
+      pipe_buffer_unmap(ct->buffer->screen, ct->buffer);
+      ct->mapped = NULL;
+   }
 
-      pipe_buffer_reference(screen, &ct->buffer, NULL);
+   pipe_buffer_reference(&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;
+   FREE(ct);
 }
 
 
@@ -294,103 +269,6 @@ 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)
-{
-   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);
-}
-
-
-/**
- * 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)
-{
-   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);
-}
-
-
 static struct pipe_surface *
 cell_get_tex_surface(struct pipe_screen *screen,
                      struct pipe_texture *pt,
@@ -402,130 +280,223 @@ cell_get_tex_surface(struct pipe_screen *screen,
 
    ps = CALLOC_STRUCT(pipe_surface);
    if (ps) {
-      ps->refcount = 1;
+      pipe_reference_init(&ps->reference, 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) {
-         ps->offset += ((pt->target == PIPE_TEXTURE_CUBE) ? face : zslice) *
-            ps->nblocksy *
-            ps->stride;
+      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;
 }
 
 
 static void 
-cell_tex_surface_release(struct pipe_screen *screen, 
-                         struct pipe_surface **s)
+cell_tex_surface_destroy(struct pipe_surface *surf)
 {
-   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;
-   }
+   pipe_texture_reference(&surf->texture, NULL);
+   FREE(surf);
+}
 
-   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);
+
+/**
+ * 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,
+                      unsigned face, unsigned level, unsigned zslice,
+                      enum pipe_transfer_usage usage,
+                      unsigned x, unsigned y, unsigned w, unsigned h)
+{
+   struct cell_texture *ct = cell_texture(texture);
+   struct cell_transfer *ctrans;
+
+   assert(texture);
+   assert(level <= texture->last_level);
+
+   ctrans = CALLOC_STRUCT(cell_transfer);
+   if (ctrans) {
+      struct pipe_transfer *pt = &ctrans->base;
+      pipe_texture_reference(&pt->texture, texture);
+      pt->format = texture->format;
+      pt->block = texture->block;
+      pt->x = x;
+      pt->y = y;
+      pt->width = w;
+      pt->height = h;
+      pt->nblocksx = texture->nblocksx[level];
+      pt->nblocksy = texture->nblocksy[level];
+      pt->stride = ct->stride[level];
+      pt->usage = usage;
+      pt->face = face;
+      pt->level = level;
+      pt->zslice = zslice;
+
+      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 NULL;
+}
 
-   /* XXX if done rendering to teximage, re-tile */
 
-   if (--surf->refcount == 0) {
-      pipe_texture_reference(&surf->texture, NULL);
-      FREE(surf);
-   }
-   *s = NULL;
+static void 
+cell_tex_transfer_destroy(struct pipe_transfer *t)
+{
+   struct cell_transfer *transfer = cell_transfer(t);
+   /* Effectively do the texture_update work here - if texture images
+    * needed post-processing to put them into hardware layout, this is
+    * where it would happen.  For cell, nothing to do.
+    */
+   assert (transfer->base.texture);
+   pipe_texture_reference(&transfer->base.texture, NULL);
+   FREE(transfer);
 }
 
 
+/**
+ * Return pointer to texture image data in linear layout.
+ */
 static void *
-cell_surface_map(struct pipe_screen *screen,
-                 struct pipe_surface *surface,
-                 unsigned flags)
+cell_transfer_map(struct pipe_screen *screen, struct pipe_transfer *transfer)
 {
-   ubyte *map;
-   struct cell_texture *ct = cell_texture(surface->texture);
-   const uint level = surface->level;
+   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(ct);
+   assert(transfer->texture);
 
-#if 0
-   if (flags & ~surface->usage) {
-      assert(0);
-      return NULL;
+   if (transfer->usage != PIPE_TRANSFER_READ) {
+      flags |= PIPE_BUFFER_USAGE_CPU_WRITE;
    }
-#endif
 
-   map = pipe_buffer_map( screen, ct->buffer, flags );
-   if (map == NULL) {
-      return NULL;
+   if (transfer->usage != PIPE_TRANSFER_WRITE) {
+      flags |= PIPE_BUFFER_USAGE_CPU_READ;
+   }
+
+   if (!ct->mapped) {
+      /* map now */
+      ct->mapped = pipe_buffer_map(screen, ct->buffer, flags);
    }
-   else {
-      if ((surface->usage & PIPE_BUFFER_USAGE_CPU_READ) &&
-          (ct->untiled_data[level])) {
-         return (void *) ((ubyte *) ct->untiled_data[level] + surface->offset);
+
+   /*
+    * Create a buffer of ordinary memory for the linear texture.
+    * This is the memory that the user will read/write.
+    */
+   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 {
-         return (void *) (map + surface->offset);
+         // xxx fix
       }
    }
+
+   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_surface_unmap(struct pipe_screen *screen,
-                   struct pipe_surface *surface)
+cell_transfer_unmap(struct pipe_screen *screen,
+                    struct pipe_transfer *transfer)
 {
-   struct cell_texture *ct = cell_texture(surface->texture);
+   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);
+   }
 
-   assert(ct);
+   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 {
+         // xxx fix
+      }
+   }
 
-   pipe_buffer_unmap( screen, ct->buffer );
+   align_free(ctrans->map);
+   ctrans->map = NULL;
 }
 
 
-
 void
 cell_init_screen_texture_funcs(struct pipe_screen *screen)
 {
    screen->texture_create = cell_texture_create;
-   screen->texture_release = cell_texture_release;
+   screen->texture_destroy = cell_texture_destroy;
 
    screen->get_tex_surface = cell_get_tex_surface;
-   screen->tex_surface_release = cell_tex_surface_release;
+   screen->tex_surface_destroy = cell_tex_surface_destroy;
+
+   screen->get_tex_transfer = cell_get_tex_transfer;
+   screen->tex_transfer_destroy = cell_tex_transfer_destroy;
 
-   screen->surface_map = cell_surface_map;
-   screen->surface_unmap = cell_surface_unmap;
+   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 7018b0c..3ffc0bf 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.h
+++ b/src/gallium/drivers/cell/ppu/cell_texture.h
@@ -43,16 +43,23 @@ 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];
-   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;
+};
+
+
+struct cell_transfer
+{
+   struct pipe_transfer base;
+
+   unsigned long offset;
+   void *map;
 };
 
 
@@ -64,6 +71,13 @@ cell_texture(struct pipe_texture *pt)
 }
 
 
+/** cast wrapper */
+static INLINE struct cell_transfer *
+cell_transfer(struct pipe_transfer *pt)
+{
+   return (struct cell_transfer *) pt;
+}
+
 
 extern void
 cell_init_screen_texture_funcs(struct pipe_screen *screen);
diff --git a/src/gallium/drivers/cell/ppu/cell_vbuf.c b/src/gallium/drivers/cell/ppu/cell_vbuf.c
index ab54e79..cfaffb5 100644
--- a/src/gallium/drivers/cell/ppu/cell_vbuf.c
+++ b/src/gallium/drivers/cell/ppu/cell_vbuf.c
@@ -62,6 +62,7 @@ struct cell_vbuf_render
    uint vertex_size;     /**< in bytes */
    void *vertex_buffer;  /**< just for debug, really */
    uint vertex_buf;      /**< in [0, CELL_NUM_BUFFERS-1] */
+   uint vertex_buffer_size;  /**< size in bytes */
 };
 
 
@@ -82,24 +83,26 @@ cell_vbuf_get_vertex_info(struct vbuf_render *vbr)
 }
 
 
-static void *
+static boolean
 cell_vbuf_allocate_vertices(struct vbuf_render *vbr,
                             ushort vertex_size, ushort nr_vertices)
 {
    struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
+   unsigned size = vertex_size * nr_vertices;
    /*printf("Alloc verts %u * %u\n", vertex_size, nr_vertices);*/
 
    assert(cvbr->vertex_buf == ~0);
    cvbr->vertex_buf = cell_get_empty_buffer(cvbr->cell);
    cvbr->vertex_buffer = cvbr->cell->buffer[cvbr->vertex_buf];
+   cvbr->vertex_buffer_size = size;
    cvbr->vertex_size = vertex_size;
-   return cvbr->vertex_buffer;
+
+   return cvbr->vertex_buffer != NULL;
 }
 
 
 static void
-cell_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices, 
-                           unsigned vertex_size, unsigned vertices_used)
+cell_vbuf_release_vertices(struct vbuf_render *vbr)
 {
    struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
    struct cell_context *cell = cvbr->cell;
@@ -127,11 +130,29 @@ cell_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices,
    cvbr->vertex_buf = ~0;
    cell_flush_int(cell, 0x0);
 
-   assert(vertices == cvbr->vertex_buffer);
    cvbr->vertex_buffer = NULL;
 }
 
 
+static void *
+cell_vbuf_map_vertices(struct vbuf_render *vbr)
+{
+   struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
+   return cvbr->vertex_buffer;
+}
+
+
+static void 
+cell_vbuf_unmap_vertices(struct vbuf_render *vbr, 
+                         ushort min_index,
+                         ushort max_index )
+{
+   struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
+   assert( cvbr->vertex_buffer_size >= (max_index+1) * cvbr->vertex_size );
+   /* do nothing */
+}
+
+
 
 static boolean
 cell_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim)
@@ -295,6 +316,8 @@ cell_init_vbuf(struct cell_context *cell)
 
    cell->vbuf_render->base.get_vertex_info = cell_vbuf_get_vertex_info;
    cell->vbuf_render->base.allocate_vertices = cell_vbuf_allocate_vertices;
+   cell->vbuf_render->base.map_vertices = cell_vbuf_map_vertices;
+   cell->vbuf_render->base.unmap_vertices = cell_vbuf_unmap_vertices;
    cell->vbuf_render->base.set_primitive = cell_vbuf_set_primitive;
    cell->vbuf_render->base.draw = cell_vbuf_draw;
    cell->vbuf_render->base.release_vertices = cell_vbuf_release_vertices;
diff --git a/src/gallium/drivers/cell/spu/Makefile b/src/gallium/drivers/cell/spu/Makefile
index 116453b..3cc5230 100644
--- a/src/gallium/drivers/cell/spu/Makefile
+++ b/src/gallium/drivers/cell/spu/Makefile
@@ -33,9 +33,10 @@ OLD_SOURCES = \
 	spu_vertex_shader.c
 
 
-SPU_OBJECTS = $(SOURCES:.c=.o) \
+SPU_OBJECTS = $(SOURCES:.c=.o)
+
+SPU_ASM_OUT = $(SOURCES:.c=.s)
 
-SPU_ASM_OUT = $(SOURCES:.c=.s) \
 
 INCLUDE_DIRS = \
 	-I$(TOP)/src/mesa \
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
index 0d9fcb9..d727268 100644
--- a/src/gallium/drivers/cell/spu/spu_tri.c
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -29,7 +29,6 @@
  * Triangle rendering within a tile.
  */
 
-#include <transpose_matrix4x4.h>
 #include "pipe/p_compiler.h"
 #include "pipe/p_format.h"
 #include "util/u_math.h"
@@ -71,6 +70,12 @@ struct vertex_header {
 #define MASK_ALL          0xf
 
 
+#define CHAN0 0
+#define CHAN1 1
+#define CHAN2 2
+#define CHAN3 3
+
+
 #define DEBUG_VERTS 0
 
 /**
@@ -144,99 +149,97 @@ struct setup_stage {
 static struct setup_stage setup;
 
 
-/**
- * Evaluate attribute coefficients (plane equations) to compute
- * attribute values for the four fragments in a quad.
- * Eg: four colors will be computed (in AoS format).
- */
-static INLINE void
-eval_coeff(uint slot, float x, float y, vector float w, vector float result[4])
+static INLINE vector float
+splatx(vector float v)
 {
-   switch (spu.vertex_info.attrib[slot].interp_mode) {
-   case INTERP_CONSTANT:
-      result[QUAD_TOP_LEFT] =
-      result[QUAD_TOP_RIGHT] =
-      result[QUAD_BOTTOM_LEFT] =
-      result[QUAD_BOTTOM_RIGHT] = setup.coef[slot].a0;
-      break;
-   case INTERP_LINEAR:
-      {
-         vector float dadx = setup.coef[slot].dadx;
-         vector float dady = setup.coef[slot].dady;
-         vector float topLeft =
-            spu_add(setup.coef[slot].a0,
-                    spu_add(spu_mul(spu_splats(x), dadx),
-                            spu_mul(spu_splats(y), dady)));
-
-         result[QUAD_TOP_LEFT] = topLeft;
-         result[QUAD_TOP_RIGHT] = spu_add(topLeft, dadx);
-         result[QUAD_BOTTOM_LEFT] = spu_add(topLeft, dady);
-         result[QUAD_BOTTOM_RIGHT] = spu_add(spu_add(topLeft, dadx), dady);
-      }
-      break;
-   case INTERP_PERSPECTIVE:
-      {
-         vector float dadx = setup.coef[slot].dadx;
-         vector float dady = setup.coef[slot].dady;
-         vector float topLeft =
-            spu_add(setup.coef[slot].a0,
-                    spu_add(spu_mul(spu_splats(x), dadx),
-                            spu_mul(spu_splats(y), dady)));
-
-         vector float wInv = spu_re(w);  /* 1.0 / w */
-
-         result[QUAD_TOP_LEFT] = spu_mul(topLeft, wInv);
-         result[QUAD_TOP_RIGHT] = spu_mul(spu_add(topLeft, dadx), wInv);
-         result[QUAD_BOTTOM_LEFT] = spu_mul(spu_add(topLeft, dady), wInv);
-         result[QUAD_BOTTOM_RIGHT] = spu_mul(spu_add(spu_add(topLeft, dadx), dady), wInv);
-      }
-      break;
-   case INTERP_POS:
-   case INTERP_NONE:
-      break;
-   default:
-      ASSERT(0);
-   }
+   return spu_splats(spu_extract(v, CHAN0));
 }
 
-
-/**
- * As above, but return 4 vectors in SOA format.
- * XXX this will all be re-written someday.
- */
-static INLINE void
-eval_coeff_soa(uint slot, float x, float y, vector float w, vector float result[4])
+static INLINE vector float
+splaty(vector float v)
 {
-   eval_coeff(slot, x, y, w, result);
-   _transpose_matrix4x4(result, result);
+   return spu_splats(spu_extract(v, CHAN1));
 }
 
+static INLINE vector float
+splatz(vector float v)
+{
+   return spu_splats(spu_extract(v, CHAN2));
+}
 
-/** Evalute coefficients to get Z for four pixels in a quad */
 static INLINE vector float
-eval_z(float x, float y)
+splatw(vector float v)
 {
-   const uint slot = 0;
-   const float dzdx = spu_extract(setup.coef[slot].dadx, 2);
-   const float dzdy = spu_extract(setup.coef[slot].dady, 2);
-   const float topLeft = spu_extract(setup.coef[slot].a0, 2) + x * dzdx + y * dzdy;
-   const vector float topLeftv = spu_splats(topLeft);
-   const vector float derivs = (vector float) { 0.0, dzdx, dzdy, dzdx + dzdy };
-   return spu_add(topLeftv, derivs);
+   return spu_splats(spu_extract(v, CHAN3));
 }
 
 
-/** Evalute coefficients to get W for four pixels in a quad */
-static INLINE vector float
-eval_w(float x, float y)
+/**
+ * Setup fragment shader inputs by evaluating triangle's vertex
+ * attribute coefficient info.
+ * \param x  quad x pos
+ * \param y  quad y pos
+ * \param fragZ  returns quad Z values
+ * \param fragInputs  returns fragment program inputs
+ * Note: this code could be incorporated into the fragment program
+ * itself to avoid the loop and switch.
+ */
+static void
+eval_inputs(float x, float y, vector float *fragZ, vector float fragInputs[])
 {
-   const uint slot = 0;
-   const float dwdx = spu_extract(setup.coef[slot].dadx, 3);
-   const float dwdy = spu_extract(setup.coef[slot].dady, 3);
-   const float topLeft = spu_extract(setup.coef[slot].a0, 3) + x * dwdx + y * dwdy;
-   const vector float topLeftv = spu_splats(topLeft);
-   const vector float derivs = (vector float) { 0.0, dwdx, dwdy, dwdx + dwdy };
-   return spu_add(topLeftv, derivs);
+   static const vector float deltaX = (const vector float) {0, 1, 0, 1};
+   static const vector float deltaY = (const vector float) {0, 0, 1, 1};
+
+   const uint posSlot = 0;
+   const vector float pos = setup.coef[posSlot].a0;
+   const vector float dposdx = setup.coef[posSlot].dadx;
+   const vector float dposdy = setup.coef[posSlot].dady;
+   const vector float fragX = spu_splats(x) + deltaX;
+   const vector float fragY = spu_splats(y) + deltaY;
+   vector float fragW, wInv;
+   uint i;
+
+   *fragZ = splatz(pos) + fragX * splatz(dposdx) + fragY * splatz(dposdy);
+   fragW =  splatw(pos) + fragX * splatw(dposdx) + fragY * splatw(dposdy);
+   wInv = spu_re(fragW);  /* 1 / w */
+
+   /* loop over fragment program inputs */
+   for (i = 0; i < spu.vertex_info.num_attribs; i++) {
+      uint attr = i + 1;
+      enum interp_mode interp = spu.vertex_info.attrib[attr].interp_mode;
+
+      /* constant term */
+      vector float a0 = setup.coef[attr].a0;
+      vector float r0 = splatx(a0);
+      vector float r1 = splaty(a0);
+      vector float r2 = splatz(a0);
+      vector float r3 = splatw(a0);
+
+      if (interp == INTERP_LINEAR || interp == INTERP_PERSPECTIVE) {
+         /* linear term */
+         vector float dadx = setup.coef[attr].dadx;
+         vector float dady = setup.coef[attr].dady;
+         /* Use SPU intrinsics here to get slightly better code.
+          * originally: r0 += fragX * splatx(dadx) + fragY * splatx(dady);
+          */
+         r0 = spu_madd(fragX, splatx(dadx), spu_madd(fragY, splatx(dady), r0));
+         r1 = spu_madd(fragX, splaty(dadx), spu_madd(fragY, splaty(dady), r1));
+         r2 = spu_madd(fragX, splatz(dadx), spu_madd(fragY, splatz(dady), r2));
+         r3 = spu_madd(fragX, splatw(dadx), spu_madd(fragY, splatw(dady), r3));
+         if (interp == INTERP_PERSPECTIVE) {
+            /* perspective term */
+            r0 *= wInv;
+            r1 *= wInv;
+            r2 *= wInv;
+            r3 *= wInv;
+         }
+      }
+      fragInputs[CHAN0] = r0;
+      fragInputs[CHAN1] = r1;
+      fragInputs[CHAN2] = r2;
+      fragInputs[CHAN3] = r3;
+      fragInputs += 4;
+   }
 }
 
 
@@ -262,19 +265,11 @@ emit_quad( int x, int y, mask_t mask)
           * Run fragment shader, execute per-fragment ops, update fb/tile.
           */
          vector float inputs[4*4], outputs[2*4];
-         vector float fragZ = eval_z((float) x, (float) y);
-         vector float fragW = eval_w((float) x, (float) y);
          vector unsigned int kill_mask;
+         vector float fragZ;
+
+         eval_inputs((float) x, (float) y, &fragZ, inputs);
 
-         /* setup inputs */
-#if 0
-         eval_coeff_soa(1, (float) x, (float) y, fragW, inputs);
-#else
-         uint i;
-         for (i = 0; i < spu.vertex_info.num_attribs; i++) {
-            eval_coeff_soa(i+1, (float) x, (float) y, fragW, inputs + i * 4);
-         }
-#endif
          ASSERT(spu.fragment_program);
          ASSERT(spu.fragment_ops);
 
diff --git a/src/gallium/drivers/cell/spu/spu_util.c b/src/gallium/drivers/cell/spu/spu_util.c
index b8a0d4a..af25dd3 100644
--- a/src/gallium/drivers/cell/spu/spu_util.c
+++ b/src/gallium/drivers/cell/spu/spu_util.c
@@ -1,7 +1,7 @@
 
 #include "cell/common.h"
 #include "pipe/p_shader_tokens.h"
-#include "pipe/p_debug.h"
+#include "util/u_debug.h"
 #include "tgsi/tgsi_parse.h"
 //#include "tgsi_build.h"
 #include "tgsi/tgsi_util.h"