Merge commit 'origin/mesa_7_7_branch'

Conflicts:
	src/gallium/drivers/svga/svga_screen_texture.c
	src/gallium/state_trackers/xorg/xorg_composite.c
	src/gallium/state_trackers/xorg/xorg_exa.c
	src/gallium/state_trackers/xorg/xorg_renderer.c
	src/gallium/state_trackers/xorg/xorg_xv.c
	src/mesa/main/texgetimage.c
	src/mesa/main/version.h

16 files changed, 256 insertions, 120 deletions

diff --git a/src/gallium/drivers/svga/Makefile b/src/gallium/drivers/svga/Makefile
index 38b63394e38..f3619081875 100644
--- a/src/gallium/drivers/svga/Makefile
+++ b/src/gallium/drivers/svga/Makefile
@@ -50,6 +50,9 @@ C_SOURCES = \
 LIBRARY_INCLUDES = \
 	-I$(TOP)/src/gallium/drivers/svga/include
 
+# With linux-debug we get a lots of warnings, filter out the bad flags.
+CFLAGS := $(filter-out -pedantic, $(filter-out -ansi, $(CFLAGS)))
+
 LIBRARY_DEFINES = \
 	-std=gnu99 -fvisibility=hidden \
 	-DHAVE_STDINT_H -DHAVE_SYS_TYPES_H
diff --git a/src/gallium/drivers/svga/svga_context.c b/src/gallium/drivers/svga/svga_context.c
index 73233957f36..c3de12b4a39 100644
--- a/src/gallium/drivers/svga/svga_context.c
+++ b/src/gallium/drivers/svga/svga_context.c
@@ -230,7 +230,9 @@ void svga_context_flush( struct svga_context *svga,
                          struct pipe_fence_handle **pfence )
 {
    struct svga_screen *svgascreen = svga_screen(svga->pipe.screen);
-   
+
+   svga->curr.nr_fbs = 0;
+
    /* Unmap upload manager buffers: 
     */
    u_upload_flush(svga->upload_vb);
diff --git a/src/gallium/drivers/svga/svga_context.h b/src/gallium/drivers/svga/svga_context.h
index 9a3e92fd8d1..e650a251d19 100644
--- a/src/gallium/drivers/svga/svga_context.h
+++ b/src/gallium/drivers/svga/svga_context.h
@@ -191,6 +191,11 @@ struct svga_state
    struct pipe_framebuffer_state framebuffer;
    float depthscale;
 
+   /* Hack to limit the number of different render targets between
+    * flushes.  Helps avoid blowing out our surface cache in EXA.
+    */
+   int nr_fbs;
+
    struct pipe_poly_stipple poly_stipple;
    struct pipe_scissor_state scissor;
    struct pipe_blend_color blend_color;
diff --git a/src/gallium/drivers/svga/svga_debug.h b/src/gallium/drivers/svga/svga_debug.h
index b7bb5686ed3..3a3fcd8fae2 100644
--- a/src/gallium/drivers/svga/svga_debug.h
+++ b/src/gallium/drivers/svga/svga_debug.h
@@ -43,6 +43,7 @@
 #define DEBUG_FLUSH    0x1000   /* flush after every draw */
 #define DEBUG_SYNC     0x2000   /* sync after every flush */
 #define DEBUG_QUERY    0x4000
+#define DEBUG_CACHE    0x8000
 
 #ifdef DEBUG
 extern int SVGA_DEBUG;
diff --git a/src/gallium/drivers/svga/svga_draw.c b/src/gallium/drivers/svga/svga_draw.c
index 1b371cecc61..8db40d0fd57 100644
--- a/src/gallium/drivers/svga/svga_draw.c
+++ b/src/gallium/drivers/svga/svga_draw.c
@@ -29,10 +29,13 @@
 #include "util/u_memory.h"
 #include "util/u_math.h"
 
+#include "svga_context.h"
 #include "svga_draw.h"
 #include "svga_draw_private.h"
+#include "svga_debug.h"
 #include "svga_screen.h"
 #include "svga_screen_buffer.h"
+#include "svga_screen_texture.h"
 #include "svga_winsys.h"
 #include "svga_cmd.h"
 
@@ -160,6 +163,10 @@ svga_hwtnl_flush( struct svga_hwtnl *hwtnl )
          ib_handle[i] = handle;
       }
 
+      SVGA_DBG(DEBUG_DMA, "draw to sid %p, %d prims\n",
+               svga_surface(svga->curr.framebuffer.cbufs[0])->handle,
+               hwtnl->cmd.prim_count);
+
       ret = SVGA3D_BeginDrawPrimitives(swc, 
                                        &vdecl, 
                                        hwtnl->cmd.vdecl_count, 
diff --git a/src/gallium/drivers/svga/svga_pipe_blit.c b/src/gallium/drivers/svga/svga_pipe_blit.c
index 5a4a8c0f5f1..4f575b06e62 100644
--- a/src/gallium/drivers/svga/svga_pipe_blit.c
+++ b/src/gallium/drivers/svga/svga_pipe_blit.c
@@ -25,6 +25,7 @@
 
 #include "svga_screen_texture.h"
 #include "svga_context.h"
+#include "svga_debug.h"
 #include "svga_cmd.h"
 
 #define FILE_DEBUG_FLAG DEBUG_BLIT
@@ -43,6 +44,13 @@ static void svga_surface_copy(struct pipe_context *pipe,
 
    svga_hwtnl_flush_retry( svga );
 
+   SVGA_DBG(DEBUG_DMA, "blit to sid %p (%d,%d), from sid %p (%d,%d) sz %dx%d\n",
+            svga_surface(dest)->handle,
+            destx, desty,
+            svga_surface(src)->handle,
+            srcx, srcy,
+            width, height);
+
    ret = SVGA3D_BeginSurfaceCopy(svga->swc,
                                  src,
                                  dest,
diff --git a/src/gallium/drivers/svga/svga_pipe_clear.c b/src/gallium/drivers/svga/svga_pipe_clear.c
index 8977d26541c..6195c3897ed 100644
--- a/src/gallium/drivers/svga/svga_pipe_clear.c
+++ b/src/gallium/drivers/svga/svga_pipe_clear.c
@@ -24,12 +24,14 @@
  **********************************************************/
 
 #include "svga_cmd.h"
+#include "svga_debug.h"
 
 #include "pipe/p_defines.h"
 #include "util/u_pack_color.h"
 
 #include "svga_context.h"
 #include "svga_state.h"
+#include "svga_screen_texture.h"
 
 
 static enum pipe_error
@@ -98,6 +100,10 @@ svga_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba,
 {
    struct svga_context *svga = svga_context( pipe );
    int ret;
+   
+   if (buffers & PIPE_CLEAR_COLOR)
+      SVGA_DBG(DEBUG_DMA, "clear sid %p\n",
+               svga_surface(svga->curr.framebuffer.cbufs[0])->handle);
 
    ret = try_clear( svga, buffers, rgba, depth, stencil );
 
diff --git a/src/gallium/drivers/svga/svga_pipe_flush.c b/src/gallium/drivers/svga/svga_pipe_flush.c
index 942366de721..0becb0765ac 100644
--- a/src/gallium/drivers/svga/svga_pipe_flush.c
+++ b/src/gallium/drivers/svga/svga_pipe_flush.c
@@ -59,6 +59,9 @@ static void svga_flush( struct pipe_context *pipe,
    /* Flush command queue.
     */
    svga_context_flush(svga, fence);
+
+   SVGA_DBG(DEBUG_DMA|DEBUG_PERF, "%s flags %x fence_ptr %p\n",
+            __FUNCTION__, flags, fence ? *fence : 0x0);
 }
 
 
diff --git a/src/gallium/drivers/svga/svga_screen.c b/src/gallium/drivers/svga/svga_screen.c
index 3afcaffff55..fc1b3c980ef 100644
--- a/src/gallium/drivers/svga/svga_screen.c
+++ b/src/gallium/drivers/svga/svga_screen.c
@@ -57,6 +57,7 @@ static const struct debug_named_value svga_debug_flags[] = {
    { "perf",     DEBUG_PERF },
    { "flush",    DEBUG_FLUSH },
    { "sync",     DEBUG_SYNC },
+   { "cache",    DEBUG_CACHE },
    {NULL, 0}
 };
 #endif
@@ -297,6 +298,10 @@ svga_fence_finish(struct pipe_screen *screen,
                   unsigned flag)
 {
    struct svga_winsys_screen *sws = svga_screen(screen)->sws;
+
+   SVGA_DBG(DEBUG_DMA|DEBUG_PERF, "%s fence_ptr %p\n",
+            __FUNCTION__, fence);
+
    return sws->fence_finish(sws, fence, flag);
 }
 
diff --git a/src/gallium/drivers/svga/svga_screen_buffer.c b/src/gallium/drivers/svga/svga_screen_buffer.c
index 3b7811734ed..1f8a8896723 100644
--- a/src/gallium/drivers/svga/svga_screen_buffer.c
+++ b/src/gallium/drivers/svga/svga_screen_buffer.c
@@ -71,7 +71,10 @@ svga_buffer_create_host_surface(struct svga_screen *ss,
       
       sbuf->key.numFaces = 1;
       sbuf->key.numMipLevels = 1;
+      sbuf->key.cachable = 1;
       
+      SVGA_DBG(DEBUG_DMA, "surface_create for buffer sz %d\n", sbuf->base.size);
+
       sbuf->handle = svga_screen_surface_create(ss, &sbuf->key);
       if(!sbuf->handle)
          return PIPE_ERROR_OUT_OF_MEMORY;
@@ -82,7 +85,7 @@ svga_buffer_create_host_surface(struct svga_screen *ss,
        */
       sbuf->hw.flags.discard = TRUE;
 
-      SVGA_DBG(DEBUG_DMA, "   grab sid %p sz %d\n", sbuf->handle, sbuf->base.size);
+      SVGA_DBG(DEBUG_DMA, "   --> got sid %p sz %d (buffer)\n", sbuf->handle, sbuf->base.size);
    }
    
    return PIPE_OK;
@@ -444,7 +447,7 @@ svga_buffer_map_range( struct pipe_screen *screen,
             enum pipe_error ret;
             struct pipe_fence_handle *fence = NULL;
             
-            SVGA_DBG(DEBUG_DMA|DEBUG_PERF, "dma from sid %p, bytes %u - %u\n", 
+            SVGA_DBG(DEBUG_DMA|DEBUG_PERF, "dma from sid %p (buffer), bytes %u - %u\n", 
                      sbuf->handle, 0, sbuf->base.size);
 
             memset(&flags, 0, sizeof flags);
@@ -776,12 +779,11 @@ svga_screen_buffer_wrap_surface(struct pipe_screen *screen,
 
    /*
     * We are not the creator of this surface and therefore we must not
-    * cache it for reuse. The caching code only caches SVGA3D_BUFFER surfaces
-    * so make sure this isn't one of those.
+    * cache it for reuse. Set the cacheable flag to zero in the key to
+    * prevent this.
     */
-
-   assert(format != SVGA3D_BUFFER);
    sbuf->key.format = format;
+   sbuf->key.cachable = 0;
    sws->surface_reference(sws, &sbuf->handle, srf);
 
    return buf;
@@ -794,6 +796,8 @@ svga_screen_buffer_get_winsys_surface(struct pipe_buffer *buffer)
    struct svga_winsys_screen *sws = svga_winsys_screen(buffer->screen);
    struct svga_winsys_surface *vsurf = NULL;
 
+   assert(svga_buffer(buffer)->key.cachable == 0);
+   svga_buffer(buffer)->key.cachable = 0;
    sws->surface_reference(sws, &vsurf, svga_buffer(buffer)->handle);
    return vsurf;
 }
diff --git a/src/gallium/drivers/svga/svga_screen_cache.c b/src/gallium/drivers/svga/svga_screen_cache.c
index 7360c1688bb..8a06383f61e 100644
--- a/src/gallium/drivers/svga/svga_screen_cache.c
+++ b/src/gallium/drivers/svga/svga_screen_cache.c
@@ -24,6 +24,7 @@
  **********************************************************/
 
 #include "util/u_memory.h"
+#include "util/u_hash.h"
 
 #include "svga_debug.h"
 #include "svga_winsys.h"
@@ -36,24 +37,11 @@
 
 /** 
  * Compute the bucket for this key. 
- * 
- * We simply compute log2(width) for now, but
  */
 static INLINE unsigned
 svga_screen_cache_bucket(const struct svga_host_surface_cache_key *key)
 {
-   unsigned bucket = 0;
-   unsigned size = key->size.width;
-   
-   while ((size >>= 1))
-      ++bucket;
-   
-   if(key->flags & SVGA3D_SURFACE_HINT_INDEXBUFFER)
-      bucket += 32;
-   
-   assert(bucket < SVGA_HOST_SURFACE_CACHE_BUCKETS);
-   
-   return bucket;
+   return util_hash_crc32( key, sizeof *key ) % SVGA_HOST_SURFACE_CACHE_BUCKETS;
 }
 
 
@@ -69,6 +57,8 @@ svga_screen_cache_lookup(struct svga_screen *svgascreen,
    unsigned bucket;
    unsigned tries = 0;
 
+   assert(key->cachable);
+
    bucket = svga_screen_cache_bucket(key);
 
    pipe_mutex_lock(cache->mutex);
@@ -104,11 +94,9 @@ svga_screen_cache_lookup(struct svga_screen *svgascreen,
 
    pipe_mutex_unlock(cache->mutex);
    
-#if 0
-   _debug_printf("%s: cache %s after %u tries\n", __FUNCTION__, handle ? "hit" : "miss", tries);
-#else
-   (void)tries;
-#endif
+   if (SVGA_DEBUG & DEBUG_DMA)
+      debug_printf("%s: cache %s after %u tries (bucket %d)\n", __FUNCTION__, 
+                   handle ? "hit" : "miss", tries, bucket);
    
    return handle;
 }
@@ -128,6 +116,7 @@ svga_screen_cache_add(struct svga_screen *svgascreen,
    struct svga_host_surface_cache_entry *entry = NULL;
    struct svga_winsys_surface *handle = *p_handle;
    
+   assert(key->cachable);
 
    assert(handle);
    if(!handle)
@@ -137,15 +126,16 @@ svga_screen_cache_add(struct svga_screen *svgascreen,
    pipe_mutex_lock(cache->mutex);
    
    if(!LIST_IS_EMPTY(&cache->empty)) {
-        /* use the first empty entry */
-        entry = LIST_ENTRY(struct svga_host_surface_cache_entry, cache->empty.next, head);
+      /* use the first empty entry */
+      entry = LIST_ENTRY(struct svga_host_surface_cache_entry, cache->empty.next, head);
         
-        LIST_DEL(&entry->head);
-     }
+      LIST_DEL(&entry->head);
+   }
    else if(!LIST_IS_EMPTY(&cache->unused)) {
       /* free the last used buffer and reuse its entry */
       entry = LIST_ENTRY(struct svga_host_surface_cache_entry, cache->unused.prev, head);
-      SVGA_DBG(DEBUG_DMA, "unref sid %p\n", entry->handle);
+      SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
+               "unref sid %p (make space)\n", entry->handle);
       sws->surface_reference(sws, &entry->handle, NULL);
 
       LIST_DEL(&entry->bucket_head);
@@ -157,11 +147,14 @@ svga_screen_cache_add(struct svga_screen *svgascreen,
       entry->handle = handle;
       memcpy(&entry->key, key, sizeof entry->key);
    
+      SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
+               "cache sid %p\n", entry->handle);
       LIST_ADD(&entry->head, &cache->validated);
    }
    else {
       /* Couldn't cache the buffer -- this really shouldn't happen */
-      SVGA_DBG(DEBUG_DMA, "unref sid %p\n", handle);
+      SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
+               "unref sid %p (couldn't find space)\n", handle);
       sws->surface_reference(sws, &handle, NULL);
    }
    
@@ -220,7 +213,8 @@ svga_screen_cache_cleanup(struct svga_screen *svgascreen)
    
    for(i = 0; i < SVGA_HOST_SURFACE_CACHE_SIZE; ++i) {
       if(cache->entries[i].handle) {
-	 SVGA_DBG(DEBUG_DMA, "unref sid %p\n", cache->entries[i].handle);
+	 SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
+                  "unref sid %p (shutdown)\n", cache->entries[i].handle);
 	 sws->surface_reference(sws, &cache->entries[i].handle, NULL);
       }
 
@@ -261,18 +255,45 @@ svga_screen_surface_create(struct svga_screen *svgascreen,
 {
    struct svga_winsys_screen *sws = svgascreen->sws;
    struct svga_winsys_surface *handle = NULL;
+   boolean cachable = SVGA_SURFACE_CACHE_ENABLED && key->cachable;
+
+   SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
+            "%s sz %dx%dx%d mips %d faces %d cachable %d\n", 
+            __FUNCTION__,
+            key->size.width,
+            key->size.height,
+            key->size.depth,
+            key->numMipLevels,
+            key->numFaces,
+            key->cachable);
+
+   if (cachable) {
+      if (key->format == SVGA3D_BUFFER) {
+         /* For buffers, round the buffer size up to the nearest power
+          * of two to increase the probability of cache hits.  Keep
+          * texture surface dimensions unchanged.
+          */
+         uint32_t size = 1;
+         while(size < key->size.width)
+            size <<= 1;
+         key->size.width = size;
+      }
 
-   if (SVGA_SURFACE_CACHE_ENABLED && key->format == SVGA3D_BUFFER) {
-      /* round the buffer size up to the nearest power of two to increase the
-       * probability of cache hits */
-      uint32_t size = 1;
-      while(size < key->size.width)
-         size <<= 1;
-      key->size.width = size;
-      
       handle = svga_screen_cache_lookup(svgascreen, key);
-      if (handle)
-         SVGA_DBG(DEBUG_DMA, "  reuse sid %p sz %d\n", handle, size);
+      if (handle) {
+         if (key->format == SVGA3D_BUFFER)
+            SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
+                     "reuse sid %p sz %d (buffer)\n", handle, 
+                     key->size.width);
+         else
+            SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
+                     "reuse sid %p sz %dx%dx%d mips %d faces %d\n", handle, 
+                     key->size.width,
+                     key->size.height,
+                     key->size.depth,
+                     key->numMipLevels,
+                     key->numFaces);
+      }
    }
 
    if (!handle) {
@@ -283,7 +304,12 @@ svga_screen_surface_create(struct svga_screen *svgascreen,
                                    key->numFaces, 
                                    key->numMipLevels);
       if (handle)
-         SVGA_DBG(DEBUG_DMA, "create sid %p sz %d\n", handle, key->size);
+         SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
+                  "  CREATE sid %p sz %dx%dx%d\n", 
+                  handle, 
+                  key->size.width,
+                  key->size.height,
+                  key->size.depth);
    }
 
    return handle;
@@ -297,11 +323,16 @@ svga_screen_surface_destroy(struct svga_screen *svgascreen,
 {
    struct svga_winsys_screen *sws = svgascreen->sws;
    
-   if(SVGA_SURFACE_CACHE_ENABLED && key->format == SVGA3D_BUFFER) {
+   /* We only set the cachable flag for surfaces of which we are the
+    * exclusive owner.  So just hold onto our existing reference in
+    * that case.
+    */
+   if(SVGA_SURFACE_CACHE_ENABLED && key->cachable) {
       svga_screen_cache_add(svgascreen, key, p_handle);
    }
    else {
-      SVGA_DBG(DEBUG_DMA, "unref sid %p\n", *p_handle);
+      SVGA_DBG(DEBUG_DMA,
+               "unref sid %p (uncachable)\n", *p_handle);
       sws->surface_reference(sws, p_handle, NULL);
    }
 }
diff --git a/src/gallium/drivers/svga/svga_screen_cache.h b/src/gallium/drivers/svga/svga_screen_cache.h
index 1bbe9877688..f5aa740d408 100644
--- a/src/gallium/drivers/svga/svga_screen_cache.h
+++ b/src/gallium/drivers/svga/svga_screen_cache.h
@@ -36,10 +36,18 @@
 #include "util/u_double_list.h"
 
 
-/* TODO: Reduce this once we don't allocate an index buffer per draw call */ 
+/* Guess the storage size of cached surfaces and try and keep it under
+ * this amount:
+ */ 
+#define SVGA_HOST_SURFACE_CACHE_BYTES 16*1024*1024
+
+/* Maximum number of discrete surfaces in the cache:
+ */
 #define SVGA_HOST_SURFACE_CACHE_SIZE 1024
 
-#define SVGA_HOST_SURFACE_CACHE_BUCKETS 64
+/* Number of hash buckets:
+ */
+#define SVGA_HOST_SURFACE_CACHE_BUCKETS 256
 
 
 struct svga_winsys_surface;
@@ -53,8 +61,9 @@ struct svga_host_surface_cache_key
    SVGA3dSurfaceFlags flags;
    SVGA3dSurfaceFormat format;
    SVGA3dSize size;
-   uint32_t numFaces;
-   uint32_t numMipLevels;
+   uint32_t numFaces:24;
+   uint32_t numMipLevels:7;
+   uint32_t cachable:1;         /* False if this is a shared surface */
 };
 
 
diff --git a/src/gallium/drivers/svga/svga_screen_texture.c b/src/gallium/drivers/svga/svga_screen_texture.c
index fb11b80dcf9..6e10d65a205 100644
--- a/src/gallium/drivers/svga/svga_screen_texture.c
+++ b/src/gallium/drivers/svga/svga_screen_texture.c
@@ -266,14 +266,8 @@ svga_texture_create(struct pipe_screen *screen,
                     const struct pipe_texture *templat)
 {
    struct svga_screen *svgascreen = svga_screen(screen);
-   struct svga_winsys_screen *sws = svgascreen->sws;
    struct svga_texture *tex = CALLOC_STRUCT(svga_texture);
    unsigned width, height, depth;
-   SVGA3dSurfaceFlags flags = 0;
-   SVGA3dSurfaceFormat format;
-   SVGA3dSize size;
-   uint32 numFaces;
-   uint32 numMipLevels;
    unsigned level;
    
    if (!tex)
@@ -298,23 +292,24 @@ svga_texture_create(struct pipe_screen *screen,
       depth = u_minify(depth, 1);
    }
    
-   size.width = templat->width0;
-   size.height = templat->height0;
-   size.depth = templat->depth0;
+   tex->key.flags = 0;
+   tex->key.size.width = templat->width0;
+   tex->key.size.height = templat->height0;
+   tex->key.size.depth = templat->depth0;
    
    if(templat->target == PIPE_TEXTURE_CUBE) {
-      flags |= SVGA3D_SURFACE_CUBEMAP;
-      numFaces = 6;
+      tex->key.flags |= SVGA3D_SURFACE_CUBEMAP;
+      tex->key.numFaces = 6;
    }
    else {
-      numFaces = 1;
+      tex->key.numFaces = 1;
    }
 
    if(templat->tex_usage & PIPE_TEXTURE_USAGE_SAMPLER)
-      flags |= SVGA3D_SURFACE_HINT_TEXTURE;
+      tex->key.flags |= SVGA3D_SURFACE_HINT_TEXTURE;
 
    if(templat->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY)
-      flags |= SVGA3D_SURFACE_HINT_SCANOUT;
+      tex->key.flags |= SVGA3D_SURFACE_HINT_SCANOUT;
    
    /* 
     * XXX: Never pass the SVGA3D_SURFACE_HINT_RENDERTARGET hint. Mesa cannot
@@ -325,21 +320,24 @@ svga_texture_create(struct pipe_screen *screen,
 #if 0
    if((templat->tex_usage & PIPE_TEXTURE_USAGE_RENDER_TARGET) &&
       !pf_is_compressed(templat->format))
-      flags |= SVGA3D_SURFACE_HINT_RENDERTARGET;
+      tex->key.flags |= SVGA3D_SURFACE_HINT_RENDERTARGET;
 #endif
    
    if(templat->tex_usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL)
-      flags |= SVGA3D_SURFACE_HINT_DEPTHSTENCIL;
+      tex->key.flags |= SVGA3D_SURFACE_HINT_DEPTHSTENCIL;
    
-   numMipLevels = templat->last_level + 1;
+   tex->key.numMipLevels = templat->last_level + 1;
    
-   format = svga_translate_format(templat->format);
-   if(format == SVGA3D_FORMAT_INVALID)
+   tex->key.format = svga_translate_format(templat->format);
+   if(tex->key.format == SVGA3D_FORMAT_INVALID)
       goto error2;
+
+   tex->key.cachable = 1;
    
-   tex->handle = sws->surface_create(sws, flags, format, size, numFaces, numMipLevels);
+   SVGA_DBG(DEBUG_DMA, "surface_create for texture\n", tex->handle);
+   tex->handle = svga_screen_surface_create(svgascreen, &tex->key);
    if (tex->handle)
-      SVGA_DBG(DEBUG_DMA, "create sid %p (texture)\n", tex->handle);
+      SVGA_DBG(DEBUG_DMA, "  --> got sid %p (texture)\n", tex->handle);
 
    return &tex->base;
 
@@ -395,6 +393,7 @@ svga_texture_blanket(struct pipe_screen * screen,
       return NULL;
 
    tex->base = *base;
+   
 
    if (sbuf->key.format == 1)
       tex->base.format = PIPE_FORMAT_X8R8G8B8_UNORM;
@@ -404,6 +403,12 @@ svga_texture_blanket(struct pipe_screen * screen,
    pipe_reference_init(&tex->base.reference, 1);
    tex->base.screen = screen;
 
+   SVGA_DBG(DEBUG_DMA, "blanket sid %p\n", sbuf->handle);
+
+   /* We don't own this storage, so don't try to cache it.
+    */
+   assert(sbuf->key.cachable == 0);
+   tex->key.cachable = 0;
    sws->surface_reference(sws, &tex->handle, sbuf->handle);
 
    return &tex->base;
@@ -424,7 +429,7 @@ svga_texture_destroy(struct pipe_texture *pt)
      DBG("%s deleting %p\n", __FUNCTION__, (void *) tex);
    */
    SVGA_DBG(DEBUG_DMA, "unref sid %p (texture)\n", tex->handle);
-   ss->sws->surface_reference(ss->sws, &tex->handle, NULL);
+   svga_screen_surface_destroy(ss, &tex->key, &tex->handle);
 
    FREE(tex);
 }
@@ -515,43 +520,47 @@ svga_texture_view_surface(struct pipe_context *pipe,
                           unsigned start_mip,
                           unsigned num_mip,
                           int face_pick,
-                          int zslice_pick)
+                          int zslice_pick,
+                          struct svga_host_surface_cache_key *key) /* OUT */
 {
    struct svga_screen *ss = svga_screen(tex->base.screen);
-   struct svga_winsys_screen *sws = ss->sws;
    struct svga_winsys_surface *handle;
    int i, j;
-   SVGA3dSurfaceFlags flags = 0;
-   SVGA3dSize size;
-   uint32 numFaces;
-   uint32 numMipLevels = num_mip;
    unsigned z_offset = 0;
 
    SVGA_DBG(DEBUG_PERF, 
             "svga: Create surface view: face %d zslice %d mips %d..%d\n",
             face_pick, zslice_pick, start_mip, start_mip+num_mip-1);
 
-   size.width = u_minify(tex->base.width0, start_mip);
-   size.height = u_minify(tex->base.height0, start_mip);
-   size.depth = zslice_pick < 0 ? u_minify(tex->base.depth0, start_mip) : 1;
-   assert(size.depth == 1);
+   key->flags = 0;
+   key->format = format;
+   key->numMipLevels = num_mip;
+   key->size.width = u_minify(tex->base.width0, start_mip);
+   key->size.height = u_minify(tex->base.height0, start_mip);
+   key->size.depth = zslice_pick < 0 ? u_minify(tex->base.depth0, start_mip) : 1;
+   key->cachable = 1;
+   assert(key->size.depth == 1);
    
    if(tex->base.target == PIPE_TEXTURE_CUBE && face_pick < 0) {
-      flags |= SVGA3D_SURFACE_CUBEMAP;
-      numFaces = 6;
+      key->flags |= SVGA3D_SURFACE_CUBEMAP;
+      key->numFaces = 6;
    } else {
-      numFaces = 1;
+      key->numFaces = 1;
    }
 
-   if(format == SVGA3D_FORMAT_INVALID)
+   if(key->format == SVGA3D_FORMAT_INVALID) {
+      key->cachable = 0;
       return NULL;
+   }
 
-   handle = sws->surface_create(sws, flags, format, size, numFaces, numMipLevels);
-
-   if (!handle)
+   SVGA_DBG(DEBUG_DMA, "surface_create for texture view\n");
+   handle = svga_screen_surface_create(ss, key);
+   if (!handle) {
+      key->cachable = 0;
       return NULL;
+   }
 
-   SVGA_DBG(DEBUG_DMA, "create sid %p (texture view)\n", handle);
+   SVGA_DBG(DEBUG_DMA, " --> got sid %p (texture view)\n", handle);
 
    if (face_pick < 0)
       face_pick = 0;
@@ -559,15 +568,23 @@ svga_texture_view_surface(struct pipe_context *pipe,
    if (zslice_pick >= 0)
        z_offset = zslice_pick;
 
-   for (i = 0; i < num_mip; i++) {
-      for (j = 0; j < numFaces; j++) {
+   for (i = 0; i < key->numMipLevels; i++) {
+      for (j = 0; j < key->numFaces; j++) {
          if(tex->defined[j + face_pick][i + start_mip]) {
-            unsigned depth = zslice_pick < 0 ? u_minify(tex->base.depth0, i + start_mip) : 1;
-            svga_texture_copy_handle(svga_context(pipe), ss,
-                                     tex->handle, 0, 0, z_offset, i + start_mip, j + face_pick,
+            unsigned depth = (zslice_pick < 0 ?
+                              u_minify(tex->base.depth0, i + start_mip) :
+                              1);
+
+            svga_texture_copy_handle(svga_context(pipe),
+                                     ss,
+                                     tex->handle, 
+                                     0, 0, z_offset, 
+                                     i + start_mip, 
+                                     j + face_pick,
                                      handle, 0, 0, 0, i, j,
                                      u_minify(tex->base.width0, i + start_mip),
-                                     u_minify(tex->base.height0, i + start_mip), depth);
+                                     u_minify(tex->base.height0, i + start_mip),
+                                     depth);
          }
       }
    }
@@ -584,25 +601,23 @@ svga_get_tex_surface(struct pipe_screen *screen,
 {
    struct svga_texture *tex = svga_texture(pt);
    struct svga_surface *s;
-   struct pipe_surface *ps;
    boolean render = flags & PIPE_BUFFER_USAGE_GPU_WRITE ? TRUE : FALSE;
    boolean view = FALSE;
    SVGA3dSurfaceFormat format;
 
    s = CALLOC_STRUCT(svga_surface);
-   ps = &s->base;
-   if (!ps)
+   if (!s)
       return NULL;
 
-   pipe_reference_init(&ps->reference, 1);
-   pipe_texture_reference(&ps->texture, pt);
-   ps->format = pt->format;
-   ps->width = u_minify(pt->width0, level);
-   ps->height = u_minify(pt->height0, level);
-   ps->usage = flags;
-   ps->level = level;
-   ps->face = face;
-   ps->zslice = zslice;
+   pipe_reference_init(&s->base.reference, 1);
+   pipe_texture_reference(&s->base.texture, pt);
+   s->base.format = pt->format;
+   s->base.width = u_minify(pt->width0, level);
+   s->base.height = u_minify(pt->height0, level);
+   s->base.usage = flags;
+   s->base.level = level;
+   s->base.face = face;
+   s->base.zslice = zslice;
 
    if (!render)
       format = svga_translate_format(pt->format);
@@ -617,11 +632,13 @@ svga_get_tex_surface(struct pipe_screen *screen,
       view = TRUE;
 
    /* Currently only used for compressed textures */
-   if (render && (format != svga_translate_format(pt->format))) {
+   if (render && 
+       format != svga_translate_format(pt->format)) {
       view = TRUE;
    }
 
-   if (level != 0 && svga_screen(screen)->debug.force_level_surface_view)
+   if (level != 0 && 
+       svga_screen(screen)->debug.force_level_surface_view)
       view = TRUE;
 
    if (pt->target == PIPE_TEXTURE_3D)
@@ -632,9 +649,10 @@ svga_get_tex_surface(struct pipe_screen *screen,
 
    if (view) {
       SVGA_DBG(DEBUG_VIEWS, "svga: Surface view: yes %p, level %u face %u z %u, %p\n",
-               pt, level, face, zslice, ps);
+               pt, level, face, zslice, s);
 
-      s->handle = svga_texture_view_surface(NULL, tex, format, level, 1, face, zslice);
+      s->handle = svga_texture_view_surface(NULL, tex, format, level, 1, face, zslice,
+                                            &s->key);
       s->real_face = 0;
       s->real_level = 0;
       s->real_zslice = 0;
@@ -642,15 +660,16 @@ svga_get_tex_surface(struct pipe_screen *screen,
       struct svga_winsys_screen *sws = svga_winsys_screen(screen);
 
       SVGA_DBG(DEBUG_VIEWS, "svga: Surface view: no %p, level %u, face %u, z %u, %p\n",
-               pt, level, face, zslice, ps);
+               pt, level, face, zslice, s);
 
+      memset(&s->key, 0, sizeof s->key);
       sws->surface_reference(sws, &s->handle, tex->handle);
       s->real_face = face;
       s->real_level = level;
       s->real_zslice = zslice;
    }
 
-   return ps;
+   return &s->base;
 }
 
 
@@ -661,7 +680,8 @@ svga_tex_surface_destroy(struct pipe_surface *surf)
    struct svga_screen *ss = svga_screen(surf->texture->screen);
 
    SVGA_DBG(DEBUG_DMA, "unref sid %p (tex surface)\n", s->handle);
-   ss->sws->surface_reference(ss->sws, &s->handle, NULL);
+   assert(s->key.cachable == 0);
+   svga_screen_surface_destroy(ss, &s->key, &s->handle);
    pipe_texture_reference(&surf->texture, NULL);
    FREE(surf);
 }
@@ -956,6 +976,7 @@ svga_get_tex_sampler_view(struct pipe_context *pipe, struct pipe_texture *pt,
                pt->height0,
                pt->depth0,
                pt->last_level);
+      sv->key.cachable = 0;
       sws->surface_reference(sws, &sv->handle, tex->handle);
       return sv;
    }
@@ -973,10 +994,12 @@ svga_get_tex_sampler_view(struct pipe_context *pipe, struct pipe_texture *pt,
    sv->handle = svga_texture_view_surface(pipe, tex, format,
                                           min_lod,
                                           max_lod - min_lod + 1,
-                                          -1, -1);
+                                          -1, -1,
+                                          &sv->key);
 
    if (!sv->handle) {
       assert(0);
+      sv->key.cachable = 0;
       sws->surface_reference(sws, &sv->handle, tex->handle);
       return sv;
    }
@@ -1029,7 +1052,7 @@ svga_destroy_sampler_view_priv(struct svga_sampler_view *v)
    struct svga_screen *ss = svga_screen(v->texture->base.screen);
 
    SVGA_DBG(DEBUG_DMA, "unref sid %p (sampler view)\n", v->handle);
-   ss->sws->surface_reference(ss->sws, &v->handle, NULL);
+   svga_screen_surface_destroy(ss, &v->key, &v->handle);
 
    FREE(v);
 }
@@ -1059,6 +1082,8 @@ svga_screen_texture_get_winsys_surface(struct pipe_texture *texture)
    struct svga_winsys_screen *sws = svga_winsys_screen(texture->screen);
    struct svga_winsys_surface *vsurf = NULL;
 
+   assert(svga_texture(texture)->key.cachable == 0);
+   svga_texture(texture)->key.cachable = 0;
    sws->surface_reference(sws, &vsurf, svga_texture(texture)->handle);
    return vsurf;
 }
diff --git a/src/gallium/drivers/svga/svga_screen_texture.h b/src/gallium/drivers/svga/svga_screen_texture.h
index 1e6fef59a39..1cc4063e653 100644
--- a/src/gallium/drivers/svga/svga_screen_texture.h
+++ b/src/gallium/drivers/svga/svga_screen_texture.h
@@ -29,7 +29,7 @@
 
 #include "pipe/p_compiler.h"
 #include "pipe/p_state.h"
-
+#include "svga_screen_cache.h"
 
 struct pipe_context;
 struct pipe_screen;
@@ -68,6 +68,7 @@ struct svga_sampler_view
 
    unsigned age;
 
+   struct svga_host_surface_cache_key key;
    struct svga_winsys_surface *handle;
 };
 
@@ -76,8 +77,6 @@ struct svga_texture
 {
    struct pipe_texture base;
 
-   struct svga_winsys_surface *handle;
-
    boolean defined[6][PIPE_MAX_TEXTURE_LEVELS];
    
    struct svga_sampler_view *cached_view;
@@ -86,6 +85,16 @@ struct svga_texture
    unsigned age;
 
    boolean views_modified;
+
+   /**
+    * Creation key for the host surface handle.
+    * 
+    * This structure describes all the host surface characteristics so that it 
+    * can be looked up in cache, since creating a host surface is often a slow
+    * operation.
+    */
+   struct svga_host_surface_cache_key key;
+   struct svga_winsys_surface *handle;
 };
 
 
@@ -93,6 +102,7 @@ struct svga_surface
 {
    struct pipe_surface base;
 
+   struct svga_host_surface_cache_key key;
    struct svga_winsys_surface *handle;
 
    unsigned real_face;
diff --git a/src/gallium/drivers/svga/svga_state_framebuffer.c b/src/gallium/drivers/svga/svga_state_framebuffer.c
index 7d7f93d8e3c..cfdcae4ee4a 100644
--- a/src/gallium/drivers/svga/svga_state_framebuffer.c
+++ b/src/gallium/drivers/svga/svga_state_framebuffer.c
@@ -54,6 +54,9 @@ static int emit_framebuffer( struct svga_context *svga,
    
    for(i = 0; i < PIPE_MAX_COLOR_BUFS; ++i) {
       if (curr->cbufs[i] != hw->cbufs[i]) {
+         if (svga->curr.nr_fbs++ > 8)
+            return PIPE_ERROR_OUT_OF_MEMORY;
+
          ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_COLOR0 + i, curr->cbufs[i]);
          if (ret != PIPE_OK)
             return ret;
diff --git a/src/gallium/drivers/svga/svga_tgsi.c b/src/gallium/drivers/svga/svga_tgsi.c
index 81eea1a145a..b8ef137c015 100644
--- a/src/gallium/drivers/svga/svga_tgsi.c
+++ b/src/gallium/drivers/svga/svga_tgsi.c
@@ -222,6 +222,20 @@ svga_tgsi_translate( const struct svga_shader *shader,
    result->nr_tokens = (emit.ptr - emit.buf) / sizeof(unsigned);
    memcpy(&result->key, &key, sizeof key);
 
+   if (SVGA_DEBUG & DEBUG_TGSI) 
+   {
+      debug_printf( "#####################################\n" );
+      debug_printf( "Shader %u below\n", shader->id );
+      tgsi_dump( shader->tokens, 0 );
+      if (SVGA_DEBUG & DEBUG_TGSI) {
+         debug_printf( "Shader %u compiled below\n", shader->id );
+         svga_shader_dump( result->tokens,
+                           result->nr_tokens ,
+                           FALSE );
+      }
+      debug_printf( "#####################################\n" );
+   }
+
    return result;
 
 fail: