Merge remote-tracking branch 'origin/released'

Conflicts:
	pixman/pixman/pixman-mmx.c
	xorg-server/glx/glapi_gentable.c

19 files changed, 602 insertions, 632 deletions

diff --git a/mesalib/src/gallium/auxiliary/util/u_blitter.c b/mesalib/src/gallium/auxiliary/util/u_blitter.c
index 6a9141209..d0b9187d1 100644
--- a/mesalib/src/gallium/auxiliary/util/u_blitter.c
+++ b/mesalib/src/gallium/auxiliary/util/u_blitter.c
@@ -733,6 +733,7 @@ static void blitter_draw(struct blitter_context_priv *ctx,
    u_upload_unmap(ctx->upload);
    util_draw_vertex_buffer(ctx->base.pipe, NULL, buf, offset,
                            PIPE_PRIM_TRIANGLE_FAN, 4, 2);
+   pipe_resource_reference(&buf, NULL);
 }
 
 static void blitter_draw_rectangle(struct blitter_context *blitter,
diff --git a/mesalib/src/gallium/auxiliary/util/u_draw_quad.c b/mesalib/src/gallium/auxiliary/util/u_draw_quad.c
index 371fefb9e..590fa0c36 100644
--- a/mesalib/src/gallium/auxiliary/util/u_draw_quad.c
+++ b/mesalib/src/gallium/auxiliary/util/u_draw_quad.c
@@ -1,135 +1,134 @@
-/**************************************************************************

- *

- * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.

- * All Rights Reserved.

- *

- * Permission is hereby granted, free of charge, to any person obtaining a

- * copy of this software and associated documentation files (the

- * "Software"), to deal in the Software without restriction, including

- * without limitation 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 TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR

- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,

- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE

- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

- *

- **************************************************************************/

-

-

-#include "pipe/p_context.h"

-#include "pipe/p_defines.h"

-#include "util/u_inlines.h"

-#include "util/u_draw_quad.h"

-#include "util/u_memory.h"

-#include "cso_cache/cso_context.h"

-

-

-/**

- * Draw a simple vertex buffer / primitive.

- * Limited to float[4] vertex attribs, tightly packed.

- */

-void 

-util_draw_vertex_buffer(struct pipe_context *pipe,

-                        struct cso_context *cso,

-                        struct pipe_resource *vbuf,

-                        uint offset,

-                        uint prim_type,

-                        uint num_verts,

-                        uint num_attribs)

-{

-   struct pipe_vertex_buffer vbuffer;

-

-   assert(num_attribs <= PIPE_MAX_ATTRIBS);

-

-   /* tell pipe about the vertex buffer */

-   memset(&vbuffer, 0, sizeof(vbuffer));

-   vbuffer.buffer = vbuf;

-   vbuffer.stride = num_attribs * 4 * sizeof(float);  /* vertex size */

-   vbuffer.buffer_offset = offset;

-

-   if (cso) {

-      cso_set_vertex_buffers(cso, 1, &vbuffer);

-   } else {

-      pipe->set_vertex_buffers(pipe, 1, &vbuffer);

-   }

-

-   /* note: vertex elements already set by caller */

-

-   /* draw */

-   util_draw_arrays(pipe, prim_type, 0, num_verts);

-}

-

-

-

-/**

- * Draw screen-aligned textured quad.

- * Note: this isn't especially efficient.

- */

-void 

-util_draw_texquad(struct pipe_context *pipe, struct cso_context *cso,

-                  float x0, float y0, float x1, float y1, float z)

-{

-   uint numAttribs = 2, i, j;

-   uint vertexBytes = 4 * (4 * numAttribs * sizeof(float));

-   struct pipe_resource *vbuf = NULL;  

-   float *v = NULL;

-

-   v = MALLOC(vertexBytes);

-   if (v == NULL)

-      goto out;

-

-   /*

-    * Load vertex buffer

-    */

-   for (i = j = 0; i < 4; i++) {

-      v[j + 2] = z;   /* z */

-      v[j + 3] = 1.0; /* w */

-      v[j + 6] = 0.0; /* r */

-      v[j + 7] = 1.0; /* q */

-      j += 8;

-   }

-

-   v[0] = x0;

-   v[1] = y0;

-   v[4] = 0.0; /*s*/

-   v[5] = 0.0; /*t*/

-

-   v[8] = x1;

-   v[9] = y0;

-   v[12] = 1.0;

-   v[13] = 0.0;

-

-   v[16] = x1;

-   v[17] = y1;

-   v[20] = 1.0;

-   v[21] = 1.0;

-

-   v[24] = x0;

-   v[25] = y1;

-   v[28] = 0.0;

-   v[29] = 1.0;

-	 

-   vbuf = pipe_user_buffer_create(pipe->screen, v, vertexBytes,

-				  PIPE_BIND_VERTEX_BUFFER);

-   if (!vbuf)

-      goto out;

-

-   util_draw_vertex_buffer(pipe, cso, vbuf, 0, PIPE_PRIM_TRIANGLE_FAN, 4, 2);

-

-out:

-   if (vbuf)

-      pipe_resource_reference(&vbuf, NULL);

-   

-   if (v)

-      FREE(v);

-}

+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation 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 TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "util/u_inlines.h"
+#include "util/u_draw_quad.h"
+#include "util/u_memory.h"
+#include "cso_cache/cso_context.h"
+
+
+/**
+ * Draw a simple vertex buffer / primitive.
+ * Limited to float[4] vertex attribs, tightly packed.
+ */
+void 
+util_draw_vertex_buffer(struct pipe_context *pipe,
+                        struct cso_context *cso,
+                        struct pipe_resource *vbuf,
+                        uint offset,
+                        uint prim_type,
+                        uint num_verts,
+                        uint num_attribs)
+{
+   struct pipe_vertex_buffer vbuffer;
+
+   assert(num_attribs <= PIPE_MAX_ATTRIBS);
+
+   /* tell pipe about the vertex buffer */
+   memset(&vbuffer, 0, sizeof(vbuffer));
+   vbuffer.buffer = vbuf;
+   vbuffer.stride = num_attribs * 4 * sizeof(float);  /* vertex size */
+   vbuffer.buffer_offset = offset;
+
+   /* note: vertex elements already set by caller */
+
+   if (cso) {
+      cso_set_vertex_buffers(cso, 1, &vbuffer);
+      cso_draw_arrays(cso, prim_type, 0, num_verts);
+   } else {
+      pipe->set_vertex_buffers(pipe, 1, &vbuffer);
+      util_draw_arrays(pipe, prim_type, 0, num_verts);
+   }
+}
+
+
+
+/**
+ * Draw screen-aligned textured quad.
+ * Note: this isn't especially efficient.
+ */
+void 
+util_draw_texquad(struct pipe_context *pipe, struct cso_context *cso,
+                  float x0, float y0, float x1, float y1, float z)
+{
+   uint numAttribs = 2, i, j;
+   uint vertexBytes = 4 * (4 * numAttribs * sizeof(float));
+   struct pipe_resource *vbuf = NULL;  
+   float *v = NULL;
+
+   v = MALLOC(vertexBytes);
+   if (v == NULL)
+      goto out;
+
+   /*
+    * Load vertex buffer
+    */
+   for (i = j = 0; i < 4; i++) {
+      v[j + 2] = z;   /* z */
+      v[j + 3] = 1.0; /* w */
+      v[j + 6] = 0.0; /* r */
+      v[j + 7] = 1.0; /* q */
+      j += 8;
+   }
+
+   v[0] = x0;
+   v[1] = y0;
+   v[4] = 0.0; /*s*/
+   v[5] = 0.0; /*t*/
+
+   v[8] = x1;
+   v[9] = y0;
+   v[12] = 1.0;
+   v[13] = 0.0;
+
+   v[16] = x1;
+   v[17] = y1;
+   v[20] = 1.0;
+   v[21] = 1.0;
+
+   v[24] = x0;
+   v[25] = y1;
+   v[28] = 0.0;
+   v[29] = 1.0;
+	 
+   vbuf = pipe_user_buffer_create(pipe->screen, v, vertexBytes,
+				  PIPE_BIND_VERTEX_BUFFER);
+   if (!vbuf)
+      goto out;
+
+   util_draw_vertex_buffer(pipe, cso, vbuf, 0, PIPE_PRIM_TRIANGLE_FAN, 4, 2);
+
+out:
+   if (vbuf)
+      pipe_resource_reference(&vbuf, NULL);
+   
+   if (v)
+      FREE(v);
+}
diff --git a/mesalib/src/gallium/auxiliary/util/u_vbuf.c b/mesalib/src/gallium/auxiliary/util/u_vbuf.c
index 50c35afd5..401c8a4a8 100644
--- a/mesalib/src/gallium/auxiliary/util/u_vbuf.c
+++ b/mesalib/src/gallium/auxiliary/util/u_vbuf.c
@@ -52,9 +52,23 @@ struct u_vbuf_elements {
    /* This might mean two things:
     * - src_format != native_format, as discussed above.
     * - src_offset % 4 != 0 (if the caps don't allow such an offset). */
-   boolean incompatible_layout;
-   /* Per-element flags. */
-   boolean incompatible_layout_elem[PIPE_MAX_ATTRIBS];
+   uint32_t incompatible_elem_mask; /* each bit describes a corresp. attrib  */
+   /* Which buffer has at least one vertex element referencing it
+    * incompatible. */
+   uint32_t incompatible_vb_mask_any;
+   /* Which buffer has all vertex elements referencing it incompatible. */
+   uint32_t incompatible_vb_mask_all;
+   /* Which buffer has at least one vertex element referencing it
+    * compatible. */
+   uint32_t compatible_vb_mask_any;
+   /* Which buffer has all vertex elements referencing it compatible. */
+   uint32_t compatible_vb_mask_all;
+
+   /* Which buffer has at least one vertex element referencing it
+    * non-instanced. */
+   uint32_t noninstance_vb_mask_any;
+
+   void *driver_cso;
 };
 
 enum {
@@ -64,101 +78,128 @@ enum {
    VB_NUM = 3
 };
 
-struct u_vbuf_priv {
-   struct u_vbuf b;
+struct u_vbuf {
+   struct u_vbuf_caps caps;
+
    struct pipe_context *pipe;
    struct translate_cache *translate_cache;
    struct cso_cache *cso_cache;
+   struct u_upload_mgr *uploader;
 
-   /* Vertex element state bound by the state tracker. */
-   void *saved_ve;
-   /* and its associated helper structure for this module. */
-   struct u_vbuf_elements *ve;
+   /* This is what was set in set_vertex_buffers.
+    * May contain user buffers. */
+   struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
+   unsigned nr_vertex_buffers;
+
+   /* Saved vertex buffers. */
+   struct pipe_vertex_buffer vertex_buffer_saved[PIPE_MAX_ATTRIBS];
+   unsigned nr_vertex_buffers_saved;
+
+   /* Vertex buffers for the driver.
+    * There are no user buffers. */
+   struct pipe_vertex_buffer real_vertex_buffer[PIPE_MAX_ATTRIBS];
+   int nr_real_vertex_buffers;
+   boolean vertex_buffers_dirty;
+
+   /* The index buffer. */
+   struct pipe_index_buffer index_buffer;
+
+   /* Vertex elements. */
+   struct u_vbuf_elements *ve, *ve_saved;
 
    /* Vertex elements used for the translate fallback. */
    struct pipe_vertex_element fallback_velems[PIPE_MAX_ATTRIBS];
    /* If non-NULL, this is a vertex element state used for the translate
     * fallback and therefore used for rendering too. */
-   void *fallback_ve;
+   boolean using_translate;
    /* The vertex buffer slot index where translated vertices have been
     * stored in. */
    unsigned fallback_vbs[VB_NUM];
-   /* When binding the fallback vertex element state, we don't want to
-    * change saved_ve and ve. This is set to TRUE in such cases. */
-   boolean ve_binding_lock;
-
-   /* Whether there is any user buffer. */
-   boolean any_user_vbs;
-   /* Whether there is a buffer with a non-native layout. */
-   boolean incompatible_vb_layout;
-   /* Per-buffer flags. */
-   boolean incompatible_vb[PIPE_MAX_ATTRIBS];
+
+   /* Which buffer is a user buffer. */
+   uint32_t user_vb_mask; /* each bit describes a corresp. buffer */
+   /* Which buffer is incompatible (unaligned). */
+   uint32_t incompatible_vb_mask; /* each bit describes a corresp. buffer */
+   /* Which buffer has a non-zero stride. */
+   uint32_t nonzero_stride_vb_mask; /* each bit describes a corresp. buffer */
 };
 
-static void u_vbuf_init_format_caps(struct u_vbuf_priv *mgr)
-{
-   struct pipe_screen *screen = mgr->pipe->screen;
+static void *
+u_vbuf_create_vertex_elements(struct u_vbuf *mgr, unsigned count,
+                              const struct pipe_vertex_element *attribs);
+static void u_vbuf_delete_vertex_elements(struct u_vbuf *mgr, void *cso);
 
-   mgr->b.caps.format_fixed32 =
+
+void u_vbuf_get_caps(struct pipe_screen *screen, struct u_vbuf_caps *caps)
+{
+   caps->format_fixed32 =
       screen->is_format_supported(screen, PIPE_FORMAT_R32_FIXED, PIPE_BUFFER,
                                   0, PIPE_BIND_VERTEX_BUFFER);
 
-   mgr->b.caps.format_float16 =
+   caps->format_float16 =
       screen->is_format_supported(screen, PIPE_FORMAT_R16_FLOAT, PIPE_BUFFER,
                                   0, PIPE_BIND_VERTEX_BUFFER);
 
-   mgr->b.caps.format_float64 =
+   caps->format_float64 =
       screen->is_format_supported(screen, PIPE_FORMAT_R64_FLOAT, PIPE_BUFFER,
                                   0, PIPE_BIND_VERTEX_BUFFER);
 
-   mgr->b.caps.format_norm32 =
+   caps->format_norm32 =
       screen->is_format_supported(screen, PIPE_FORMAT_R32_UNORM, PIPE_BUFFER,
                                   0, PIPE_BIND_VERTEX_BUFFER) &&
       screen->is_format_supported(screen, PIPE_FORMAT_R32_SNORM, PIPE_BUFFER,
                                   0, PIPE_BIND_VERTEX_BUFFER);
 
-   mgr->b.caps.format_scaled32 =
+   caps->format_scaled32 =
       screen->is_format_supported(screen, PIPE_FORMAT_R32_USCALED, PIPE_BUFFER,
                                   0, PIPE_BIND_VERTEX_BUFFER) &&
       screen->is_format_supported(screen, PIPE_FORMAT_R32_SSCALED, PIPE_BUFFER,
                                   0, PIPE_BIND_VERTEX_BUFFER);
+
+   caps->buffer_offset_unaligned =
+      !screen->get_param(screen,
+                        PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY);
+
+   caps->buffer_stride_unaligned =
+      !screen->get_param(screen,
+                        PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY);
+
+   caps->velem_src_offset_unaligned =
+      !screen->get_param(screen,
+                        PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY);
+
+   caps->user_vertex_buffers =
+      screen->get_param(screen, PIPE_CAP_USER_VERTEX_BUFFERS);
 }
 
 struct u_vbuf *
 u_vbuf_create(struct pipe_context *pipe,
-              unsigned upload_buffer_size,
-              unsigned upload_buffer_alignment,
-              unsigned upload_buffer_bind,
-              enum u_fetch_alignment fetch_alignment)
+              struct u_vbuf_caps *caps)
 {
-   struct u_vbuf_priv *mgr = CALLOC_STRUCT(u_vbuf_priv);
+   struct u_vbuf *mgr = CALLOC_STRUCT(u_vbuf);
 
+   mgr->caps = *caps;
    mgr->pipe = pipe;
    mgr->cso_cache = cso_cache_create();
    mgr->translate_cache = translate_cache_create();
    memset(mgr->fallback_vbs, ~0, sizeof(mgr->fallback_vbs));
 
-   mgr->b.uploader = u_upload_create(pipe, upload_buffer_size,
-                                     upload_buffer_alignment,
-                                     upload_buffer_bind);
-
-   mgr->b.caps.fetch_dword_unaligned =
-         fetch_alignment == U_VERTEX_FETCH_BYTE_ALIGNED;
-
-   u_vbuf_init_format_caps(mgr);
+   mgr->uploader = u_upload_create(pipe, 1024 * 1024, 4,
+                                   PIPE_BIND_VERTEX_BUFFER);
 
-   return &mgr->b;
+   return mgr;
 }
 
-/* XXX I had to fork this off of cso_context. */
-static void *
-u_vbuf_pipe_set_vertex_elements(struct u_vbuf_priv *mgr,
-                                unsigned count,
-                                const struct pipe_vertex_element *states)
+/* u_vbuf uses its own caching for vertex elements, because it needs to keep
+ * its own preprocessed state per vertex element CSO. */
+static struct u_vbuf_elements *
+u_vbuf_set_vertex_elements_internal(struct u_vbuf *mgr, unsigned count,
+                                    const struct pipe_vertex_element *states)
 {
+   struct pipe_context *pipe = mgr->pipe;
    unsigned key_size, hash_key;
    struct cso_hash_iter iter;
-   void *handle;
+   struct u_vbuf_elements *ve;
    struct cso_velems_state velems_state;
 
    /* need to include the count into the stored state data too. */
@@ -173,47 +214,52 @@ u_vbuf_pipe_set_vertex_elements(struct u_vbuf_priv *mgr,
    if (cso_hash_iter_is_null(iter)) {
       struct cso_velements *cso = MALLOC_STRUCT(cso_velements);
       memcpy(&cso->state, &velems_state, key_size);
-      cso->data =
-            mgr->pipe->create_vertex_elements_state(mgr->pipe, count,
-                                                    &cso->state.velems[0]);
-      cso->delete_state =
-            (cso_state_callback)mgr->pipe->delete_vertex_elements_state;
-      cso->context = mgr->pipe;
+      cso->data = u_vbuf_create_vertex_elements(mgr, count, states);
+      cso->delete_state = (cso_state_callback)u_vbuf_delete_vertex_elements;
+      cso->context = (void*)mgr;
 
       iter = cso_insert_state(mgr->cso_cache, hash_key, CSO_VELEMENTS, cso);
-      handle = cso->data;
+      ve = cso->data;
    } else {
-      handle = ((struct cso_velements *)cso_hash_iter_data(iter))->data;
+      ve = ((struct cso_velements *)cso_hash_iter_data(iter))->data;
    }
 
-   mgr->pipe->bind_vertex_elements_state(mgr->pipe, handle);
-   return handle;
+   assert(ve);
+   pipe->bind_vertex_elements_state(pipe, ve->driver_cso);
+   return ve;
+}
+
+void u_vbuf_set_vertex_elements(struct u_vbuf *mgr, unsigned count,
+                               const struct pipe_vertex_element *states)
+{
+   mgr->ve = u_vbuf_set_vertex_elements_internal(mgr, count, states);
 }
 
-void u_vbuf_destroy(struct u_vbuf *mgrb)
+void u_vbuf_destroy(struct u_vbuf *mgr)
 {
-   struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
    unsigned i;
 
-   for (i = 0; i < mgr->b.nr_vertex_buffers; i++) {
-      pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, NULL);
+   mgr->pipe->set_vertex_buffers(mgr->pipe, 0, NULL);
+
+   for (i = 0; i < mgr->nr_vertex_buffers; i++) {
+      pipe_resource_reference(&mgr->vertex_buffer[i].buffer, NULL);
    }
-   for (i = 0; i < mgr->b.nr_real_vertex_buffers; i++) {
-      pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, NULL);
+   for (i = 0; i < mgr->nr_real_vertex_buffers; i++) {
+      pipe_resource_reference(&mgr->real_vertex_buffer[i].buffer, NULL);
    }
 
    translate_cache_destroy(mgr->translate_cache);
-   u_upload_destroy(mgr->b.uploader);
+   u_upload_destroy(mgr->uploader);
    cso_cache_delete(mgr->cso_cache);
    FREE(mgr);
 }
 
 static void
-u_vbuf_translate_buffers(struct u_vbuf_priv *mgr, struct translate_key *key,
+u_vbuf_translate_buffers(struct u_vbuf *mgr, struct translate_key *key,
                          unsigned vb_mask, unsigned out_vb,
                          int start_vertex, unsigned num_vertices,
                          int start_index, unsigned num_indices, int min_index,
-                         bool unroll_indices)
+                         boolean unroll_indices)
 {
    struct translate *tr;
    struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS] = {0};
@@ -225,14 +271,14 @@ u_vbuf_translate_buffers(struct u_vbuf_priv *mgr, struct translate_key *key,
    tr = translate_cache_find(mgr->translate_cache, key);
 
    /* Map buffers we want to translate. */
-   for (i = 0; i < mgr->b.nr_vertex_buffers; i++) {
+   for (i = 0; i < mgr->nr_vertex_buffers; i++) {
       if (vb_mask & (1 << i)) {
-         struct pipe_vertex_buffer *vb = &mgr->b.vertex_buffer[i];
+         struct pipe_vertex_buffer *vb = &mgr->vertex_buffer[i];
          unsigned offset = vb->buffer_offset + vb->stride * start_vertex;
          uint8_t *map;
 
-         if (u_vbuf_resource(vb->buffer)->user_ptr) {
-            map = u_vbuf_resource(vb->buffer)->user_ptr + offset;
+         if (vb->buffer->user_ptr) {
+            map = vb->buffer->user_ptr + offset;
          } else {
             unsigned size = vb->stride ? num_vertices * vb->stride
                                        : sizeof(double)*4;
@@ -256,15 +302,15 @@ u_vbuf_translate_buffers(struct u_vbuf_priv *mgr, struct translate_key *key,
 
    /* Translate. */
    if (unroll_indices) {
-      struct pipe_index_buffer *ib = &mgr->b.index_buffer;
+      struct pipe_index_buffer *ib = &mgr->index_buffer;
       struct pipe_transfer *transfer = NULL;
       unsigned offset = ib->offset + start_index * ib->index_size;
       uint8_t *map;
 
       assert(ib->buffer && ib->index_size);
 
-      if (u_vbuf_resource(ib->buffer)->user_ptr) {
-         map = u_vbuf_resource(ib->buffer)->user_ptr + offset;
+      if (ib->buffer->user_ptr) {
+         map = ib->buffer->user_ptr + offset;
       } else {
          map = pipe_buffer_map_range(mgr->pipe, ib->buffer, offset,
                                      num_indices * ib->index_size,
@@ -272,7 +318,7 @@ u_vbuf_translate_buffers(struct u_vbuf_priv *mgr, struct translate_key *key,
       }
 
       /* Create and map the output buffer. */
-      u_upload_alloc(mgr->b.uploader, 0,
+      u_upload_alloc(mgr->uploader, 0,
                      key->output_stride * num_indices,
                      &out_offset, &out_buffer,
                      (void**)&out_map);
@@ -294,7 +340,7 @@ u_vbuf_translate_buffers(struct u_vbuf_priv *mgr, struct translate_key *key,
       }
    } else {
       /* Create and map the output buffer. */
-      u_upload_alloc(mgr->b.uploader,
+      u_upload_alloc(mgr->uploader,
                      key->output_stride * start_vertex,
                      key->output_stride * num_vertices,
                      &out_offset, &out_buffer,
@@ -306,64 +352,52 @@ u_vbuf_translate_buffers(struct u_vbuf_priv *mgr, struct translate_key *key,
    }
 
    /* Unmap all buffers. */
-   for (i = 0; i < mgr->b.nr_vertex_buffers; i++) {
+   for (i = 0; i < mgr->nr_vertex_buffers; i++) {
       if (vb_transfer[i]) {
          pipe_buffer_unmap(mgr->pipe, vb_transfer[i]);
       }
    }
 
    /* Setup the new vertex buffer. */
-   mgr->b.real_vertex_buffer[out_vb].buffer_offset = out_offset;
-   mgr->b.real_vertex_buffer[out_vb].stride = key->output_stride;
+   mgr->real_vertex_buffer[out_vb].buffer_offset = out_offset;
+   mgr->real_vertex_buffer[out_vb].stride = key->output_stride;
 
    /* Move the buffer reference. */
    pipe_resource_reference(
-      &mgr->b.real_vertex_buffer[out_vb].buffer, NULL);
-   mgr->b.real_vertex_buffer[out_vb].buffer = out_buffer;
+      &mgr->real_vertex_buffer[out_vb].buffer, NULL);
+   mgr->real_vertex_buffer[out_vb].buffer = out_buffer;
 }
 
 static boolean
-u_vbuf_translate_find_free_vb_slots(struct u_vbuf_priv *mgr,
+u_vbuf_translate_find_free_vb_slots(struct u_vbuf *mgr,
                                     unsigned mask[VB_NUM])
 {
-   unsigned i, type;
-   unsigned nr = mgr->ve->count;
-   boolean used_vb[PIPE_MAX_ATTRIBS] = {0};
+   unsigned type;
    unsigned fallback_vbs[VB_NUM];
+   /* Set the bit for each buffer which is incompatible, or isn't set. */
+   uint32_t unused_vb_mask =
+      mgr->ve->incompatible_vb_mask_all | mgr->incompatible_vb_mask |
+      ~((1 << mgr->nr_vertex_buffers) - 1);
 
    memset(fallback_vbs, ~0, sizeof(fallback_vbs));
 
-   /* Mark used vertex buffers as... used. */
-   for (i = 0; i < nr; i++) {
-      if (!mgr->ve->incompatible_layout_elem[i]) {
-         unsigned index = mgr->ve->ve[i].vertex_buffer_index;
-
-         if (!mgr->incompatible_vb[index]) {
-            used_vb[index] = TRUE;
-         }
-      }
-   }
-
    /* Find free slots for each type if needed. */
-   i = 0;
    for (type = 0; type < VB_NUM; type++) {
       if (mask[type]) {
-         for (; i < PIPE_MAX_ATTRIBS; i++) {
-            if (!used_vb[i]) {
-               /*printf("found slot=%i for type=%i\n", i, type);*/
-               fallback_vbs[type] = i;
-               i++;
-               if (i > mgr->b.nr_real_vertex_buffers) {
-                  mgr->b.nr_real_vertex_buffers = i;
-               }
-               break;
-            }
-         }
-         if (i == PIPE_MAX_ATTRIBS) {
+         uint32_t index;
+
+         if (!unused_vb_mask) {
             /* fail, reset the number to its original value */
-            mgr->b.nr_real_vertex_buffers = mgr->b.nr_vertex_buffers;
+            mgr->nr_real_vertex_buffers = mgr->nr_vertex_buffers;
             return FALSE;
          }
+
+         index = ffs(unused_vb_mask) - 1;
+         fallback_vbs[type] = index;
+         if (index >= mgr->nr_real_vertex_buffers) {
+            mgr->nr_real_vertex_buffers = index + 1;
+         }
+         /*printf("found slot=%i for type=%i\n", index, type);*/
       }
    }
 
@@ -372,11 +406,11 @@ u_vbuf_translate_find_free_vb_slots(struct u_vbuf_priv *mgr,
 }
 
 static boolean
-u_vbuf_translate_begin(struct u_vbuf_priv *mgr,
+u_vbuf_translate_begin(struct u_vbuf *mgr,
                        int start_vertex, unsigned num_vertices,
                        int start_instance, unsigned num_instances,
                        int start_index, unsigned num_indices, int min_index,
-                       bool unroll_indices)
+                       boolean unroll_indices)
 {
    unsigned mask[VB_NUM] = {0};
    struct translate_key key[VB_NUM];
@@ -403,22 +437,22 @@ u_vbuf_translate_begin(struct u_vbuf_priv *mgr,
    for (i = 0; i < mgr->ve->count; i++) {
       unsigned vb_index = mgr->ve->ve[i].vertex_buffer_index;
 
-      if (!mgr->b.vertex_buffer[vb_index].stride) {
-         if (!mgr->ve->incompatible_layout_elem[i] &&
-             !mgr->incompatible_vb[vb_index]) {
+      if (!mgr->vertex_buffer[vb_index].stride) {
+         if (!(mgr->ve->incompatible_elem_mask & (1 << i)) &&
+             !(mgr->incompatible_vb_mask & (1 << vb_index))) {
             continue;
          }
          mask[VB_CONST] |= 1 << vb_index;
       } else if (mgr->ve->ve[i].instance_divisor) {
-         if (!mgr->ve->incompatible_layout_elem[i] &&
-             !mgr->incompatible_vb[vb_index]) {
+         if (!(mgr->ve->incompatible_elem_mask & (1 << i)) &&
+             !(mgr->incompatible_vb_mask & (1 << vb_index))) {
             continue;
          }
          mask[VB_INSTANCE] |= 1 << vb_index;
       } else {
          if (!unroll_indices &&
-             !mgr->ve->incompatible_layout_elem[i] &&
-             !mgr->incompatible_vb[vb_index]) {
+             !(mgr->ve->incompatible_elem_mask & (1 << i)) &&
+             !(mgr->incompatible_vb_mask & (1 << vb_index))) {
             continue;
          }
          mask[VB_VERTEX] |= 1 << vb_index;
@@ -439,8 +473,8 @@ u_vbuf_translate_begin(struct u_vbuf_priv *mgr,
       unsigned bit, vb_index = mgr->ve->ve[i].vertex_buffer_index;
       bit = 1 << vb_index;
 
-      if (!mgr->ve->incompatible_layout_elem[i] &&
-          !mgr->incompatible_vb[vb_index] &&
+      if (!(mgr->ve->incompatible_elem_mask & (1 << i)) &&
+          !(mgr->incompatible_vb_mask & (1 << vb_index)) &&
           (!unroll_indices || !(mask[VB_VERTEX] & bit))) {
          continue;
       }
@@ -484,7 +518,7 @@ u_vbuf_translate_begin(struct u_vbuf_priv *mgr,
 
          /* Fixup the stride for constant attribs. */
          if (type == VB_CONST) {
-            mgr->b.real_vertex_buffer[mgr->fallback_vbs[VB_CONST]].stride = 0;
+            mgr->real_vertex_buffer[mgr->fallback_vbs[VB_CONST]].stride = 0;
          }
       }
    }
@@ -512,55 +546,48 @@ u_vbuf_translate_begin(struct u_vbuf_priv *mgr,
       }
    }
 
-   /* Preserve saved_ve. */
-   mgr->ve_binding_lock = TRUE;
-   mgr->fallback_ve = u_vbuf_pipe_set_vertex_elements(mgr, mgr->ve->count,
-                                                      mgr->fallback_velems);
-   mgr->ve_binding_lock = FALSE;
+   u_vbuf_set_vertex_elements_internal(mgr, mgr->ve->count,
+                                       mgr->fallback_velems);
+   mgr->using_translate = TRUE;
    return TRUE;
 }
 
-static void u_vbuf_translate_end(struct u_vbuf_priv *mgr)
+static void u_vbuf_translate_end(struct u_vbuf *mgr)
 {
    unsigned i;
 
    /* Restore vertex elements. */
-   /* Note that saved_ve will be overwritten in bind_vertex_elements_state. */
-   mgr->pipe->bind_vertex_elements_state(mgr->pipe, mgr->saved_ve);
-   mgr->fallback_ve = NULL;
+   mgr->pipe->bind_vertex_elements_state(mgr->pipe, mgr->ve->driver_cso);
+   mgr->using_translate = FALSE;
 
    /* Unreference the now-unused VBOs. */
    for (i = 0; i < VB_NUM; i++) {
       unsigned vb = mgr->fallback_vbs[i];
       if (vb != ~0) {
-         pipe_resource_reference(&mgr->b.real_vertex_buffer[vb].buffer, NULL);
+         pipe_resource_reference(&mgr->real_vertex_buffer[vb].buffer, NULL);
          mgr->fallback_vbs[i] = ~0;
       }
    }
-   mgr->b.nr_real_vertex_buffers = mgr->b.nr_vertex_buffers;
+   mgr->nr_real_vertex_buffers = mgr->nr_vertex_buffers;
 }
 
 #define FORMAT_REPLACE(what, withwhat) \
     case PIPE_FORMAT_##what: format = PIPE_FORMAT_##withwhat; break
 
-struct u_vbuf_elements *
-u_vbuf_create_vertex_elements(struct u_vbuf *mgrb,
-                              unsigned count,
-                              const struct pipe_vertex_element *attribs,
-                              struct pipe_vertex_element *native_attribs)
+static void *
+u_vbuf_create_vertex_elements(struct u_vbuf *mgr, unsigned count,
+                              const struct pipe_vertex_element *attribs)
 {
-   struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
+   struct pipe_context *pipe = mgr->pipe;
    unsigned i;
+   struct pipe_vertex_element driver_attribs[PIPE_MAX_ATTRIBS];
    struct u_vbuf_elements *ve = CALLOC_STRUCT(u_vbuf_elements);
+   uint32_t used_buffers = 0;
 
    ve->count = count;
 
-   if (!count) {
-      return ve;
-   }
-
    memcpy(ve->ve, attribs, sizeof(struct pipe_vertex_element) * count);
-   memcpy(native_attribs, attribs, sizeof(struct pipe_vertex_element) * count);
+   memcpy(driver_attribs, attribs, sizeof(struct pipe_vertex_element) * count);
 
    /* Set the best native format in case the original format is not
     * supported. */
@@ -569,10 +596,16 @@ u_vbuf_create_vertex_elements(struct u_vbuf *mgrb,
 
       ve->src_format_size[i] = util_format_get_blocksize(format);
 
+      used_buffers |= 1 << ve->ve[i].vertex_buffer_index;
+
+      if (!ve->ve[i].instance_divisor) {
+         ve->noninstance_vb_mask_any |= 1 << ve->ve[i].vertex_buffer_index;
+      }
+
       /* Choose a native format.
        * For now we don't care about the alignment, that's going to
        * be sorted out later. */
-      if (!mgr->b.caps.format_fixed32) {
+      if (!mgr->caps.format_fixed32) {
          switch (format) {
             FORMAT_REPLACE(R32_FIXED,           R32_FLOAT);
             FORMAT_REPLACE(R32G32_FIXED,        R32G32_FLOAT);
@@ -581,7 +614,7 @@ u_vbuf_create_vertex_elements(struct u_vbuf *mgrb,
             default:;
          }
       }
-      if (!mgr->b.caps.format_float16) {
+      if (!mgr->caps.format_float16) {
          switch (format) {
             FORMAT_REPLACE(R16_FLOAT,           R32_FLOAT);
             FORMAT_REPLACE(R16G16_FLOAT,        R32G32_FLOAT);
@@ -590,7 +623,7 @@ u_vbuf_create_vertex_elements(struct u_vbuf *mgrb,
             default:;
          }
       }
-      if (!mgr->b.caps.format_float64) {
+      if (!mgr->caps.format_float64) {
          switch (format) {
             FORMAT_REPLACE(R64_FLOAT,           R32_FLOAT);
             FORMAT_REPLACE(R64G64_FLOAT,        R32G32_FLOAT);
@@ -599,7 +632,7 @@ u_vbuf_create_vertex_elements(struct u_vbuf *mgrb,
             default:;
          }
       }
-      if (!mgr->b.caps.format_norm32) {
+      if (!mgr->caps.format_norm32) {
          switch (format) {
             FORMAT_REPLACE(R32_UNORM,           R32_FLOAT);
             FORMAT_REPLACE(R32G32_UNORM,        R32G32_FLOAT);
@@ -612,7 +645,7 @@ u_vbuf_create_vertex_elements(struct u_vbuf *mgrb,
             default:;
          }
       }
-      if (!mgr->b.caps.format_scaled32) {
+      if (!mgr->caps.format_scaled32) {
          switch (format) {
             FORMAT_REPLACE(R32_USCALED,         R32_FLOAT);
             FORMAT_REPLACE(R32G32_USCALED,      R32G32_FLOAT);
@@ -626,115 +659,106 @@ u_vbuf_create_vertex_elements(struct u_vbuf *mgrb,
          }
       }
 
-      native_attribs[i].src_format = format;
+      driver_attribs[i].src_format = format;
       ve->native_format[i] = format;
       ve->native_format_size[i] =
             util_format_get_blocksize(ve->native_format[i]);
 
-      ve->incompatible_layout_elem[i] =
-            ve->ve[i].src_format != ve->native_format[i] ||
-            (!mgr->b.caps.fetch_dword_unaligned && ve->ve[i].src_offset % 4 != 0);
-      ve->incompatible_layout =
-            ve->incompatible_layout ||
-            ve->incompatible_layout_elem[i];
+      if (ve->ve[i].src_format != format ||
+          (!mgr->caps.velem_src_offset_unaligned &&
+           ve->ve[i].src_offset % 4 != 0)) {
+         ve->incompatible_elem_mask |= 1 << i;
+         ve->incompatible_vb_mask_any |= 1 << ve->ve[i].vertex_buffer_index;
+      } else {
+         ve->compatible_vb_mask_any |= 1 << ve->ve[i].vertex_buffer_index;
+      }
    }
 
+   ve->compatible_vb_mask_all = ~ve->incompatible_vb_mask_any & used_buffers;
+   ve->incompatible_vb_mask_all = ~ve->compatible_vb_mask_any & used_buffers;
+
    /* Align the formats to the size of DWORD if needed. */
-   if (!mgr->b.caps.fetch_dword_unaligned) {
+   if (!mgr->caps.velem_src_offset_unaligned) {
       for (i = 0; i < count; i++) {
          ve->native_format_size[i] = align(ve->native_format_size[i], 4);
       }
    }
 
+   ve->driver_cso =
+      pipe->create_vertex_elements_state(pipe, count, driver_attribs);
    return ve;
 }
 
-void u_vbuf_bind_vertex_elements(struct u_vbuf *mgrb,
-                                 void *cso,
-                                 struct u_vbuf_elements *ve)
+static void u_vbuf_delete_vertex_elements(struct u_vbuf *mgr, void *cso)
 {
-   struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
-
-   if (!cso) {
-      return;
-   }
-
-   if (!mgr->ve_binding_lock) {
-      mgr->saved_ve = cso;
-      mgr->ve = ve;
-   }
-}
+   struct pipe_context *pipe = mgr->pipe;
+   struct u_vbuf_elements *ve = cso;
 
-void u_vbuf_destroy_vertex_elements(struct u_vbuf *mgr,
-                                    struct u_vbuf_elements *ve)
-{
+   pipe->delete_vertex_elements_state(pipe, ve->driver_cso);
    FREE(ve);
 }
 
-void u_vbuf_set_vertex_buffers(struct u_vbuf *mgrb,
-                               unsigned count,
+void u_vbuf_set_vertex_buffers(struct u_vbuf *mgr, unsigned count,
                                const struct pipe_vertex_buffer *bufs)
 {
-   struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
    unsigned i;
 
-   mgr->any_user_vbs = FALSE;
-   mgr->incompatible_vb_layout = FALSE;
-   memset(mgr->incompatible_vb, 0, sizeof(mgr->incompatible_vb));
-
-   if (!mgr->b.caps.fetch_dword_unaligned) {
-      /* Check if the strides and offsets are aligned to the size of DWORD. */
-      for (i = 0; i < count; i++) {
-         if (bufs[i].buffer) {
-            if (bufs[i].stride % 4 != 0 ||
-                bufs[i].buffer_offset % 4 != 0) {
-               mgr->incompatible_vb_layout = TRUE;
-               mgr->incompatible_vb[i] = TRUE;
-            }
-         }
-      }
-   }
+   mgr->user_vb_mask = 0;
+   mgr->incompatible_vb_mask = 0;
+   mgr->nonzero_stride_vb_mask = 0;
 
    for (i = 0; i < count; i++) {
       const struct pipe_vertex_buffer *vb = &bufs[i];
+      struct pipe_vertex_buffer *orig_vb = &mgr->vertex_buffer[i];
+      struct pipe_vertex_buffer *real_vb = &mgr->real_vertex_buffer[i];
 
-      pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, vb->buffer);
+      pipe_resource_reference(&orig_vb->buffer, vb->buffer);
 
-      mgr->b.real_vertex_buffer[i].buffer_offset =
-      mgr->b.vertex_buffer[i].buffer_offset = vb->buffer_offset;
+      real_vb->buffer_offset = orig_vb->buffer_offset = vb->buffer_offset;
+      real_vb->stride = orig_vb->stride = vb->stride;
 
-      mgr->b.real_vertex_buffer[i].stride =
-      mgr->b.vertex_buffer[i].stride = vb->stride;
+      if (vb->stride) {
+         mgr->nonzero_stride_vb_mask |= 1 << i;
+      }
+
+      if (!vb->buffer) {
+         pipe_resource_reference(&real_vb->buffer, NULL);
+         continue;
+      }
 
-      if (!vb->buffer ||
-          mgr->incompatible_vb[i]) {
-         pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, NULL);
+      if ((!mgr->caps.buffer_offset_unaligned && vb->buffer_offset % 4 != 0) ||
+          (!mgr->caps.buffer_stride_unaligned && vb->stride % 4 != 0)) {
+         mgr->incompatible_vb_mask |= 1 << i;
+         pipe_resource_reference(&real_vb->buffer, NULL);
          continue;
       }
 
-      if (u_vbuf_resource(vb->buffer)->user_ptr) {
-         pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, NULL);
-         mgr->any_user_vbs = TRUE;
+      if (!mgr->caps.user_vertex_buffers && vb->buffer->user_ptr) {
+         mgr->user_vb_mask |= 1 << i;
+         pipe_resource_reference(&real_vb->buffer, NULL);
          continue;
       }
 
-      pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, vb->buffer);
+      pipe_resource_reference(&real_vb->buffer, vb->buffer);
    }
 
-   for (i = count; i < mgr->b.nr_vertex_buffers; i++) {
-      pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, NULL);
+   for (i = count; i < mgr->nr_vertex_buffers; i++) {
+      pipe_resource_reference(&mgr->vertex_buffer[i].buffer, NULL);
    }
-   for (i = count; i < mgr->b.nr_real_vertex_buffers; i++) {
-      pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, NULL);
+   for (i = count; i < mgr->nr_real_vertex_buffers; i++) {
+      pipe_resource_reference(&mgr->real_vertex_buffer[i].buffer, NULL);
    }
 
-   mgr->b.nr_vertex_buffers = count;
-   mgr->b.nr_real_vertex_buffers = count;
+   mgr->nr_vertex_buffers = count;
+   mgr->nr_real_vertex_buffers = count;
+   mgr->vertex_buffers_dirty = TRUE;
 }
 
 void u_vbuf_set_index_buffer(struct u_vbuf *mgr,
                              const struct pipe_index_buffer *ib)
 {
+   struct pipe_context *pipe = mgr->pipe;
+
    if (ib && ib->buffer) {
       assert(ib->offset % ib->index_size == 0);
       pipe_resource_reference(&mgr->index_buffer.buffer, ib->buffer);
@@ -743,18 +767,20 @@ void u_vbuf_set_index_buffer(struct u_vbuf *mgr,
    } else {
       pipe_resource_reference(&mgr->index_buffer.buffer, NULL);
    }
+
+   pipe->set_index_buffer(pipe, ib);
 }
 
 static void
-u_vbuf_upload_buffers(struct u_vbuf_priv *mgr,
+u_vbuf_upload_buffers(struct u_vbuf *mgr,
                       int start_vertex, unsigned num_vertices,
                       int start_instance, unsigned num_instances)
 {
    unsigned i;
    unsigned nr_velems = mgr->ve->count;
-   unsigned nr_vbufs = mgr->b.nr_vertex_buffers;
+   unsigned nr_vbufs = mgr->nr_vertex_buffers;
    struct pipe_vertex_element *velems =
-         mgr->fallback_ve ? mgr->fallback_velems : mgr->ve->ve;
+         mgr->using_translate ? mgr->fallback_velems : mgr->ve->ve;
    unsigned start_offset[PIPE_MAX_ATTRIBS];
    unsigned end_offset[PIPE_MAX_ATTRIBS] = {0};
 
@@ -762,7 +788,7 @@ u_vbuf_upload_buffers(struct u_vbuf_priv *mgr,
    for (i = 0; i < nr_velems; i++) {
       struct pipe_vertex_element *velem = &velems[i];
       unsigned index = velem->vertex_buffer_index;
-      struct pipe_vertex_buffer *vb = &mgr->b.vertex_buffer[index];
+      struct pipe_vertex_buffer *vb = &mgr->vertex_buffer[index];
       unsigned instance_div, first, size;
 
       /* Skip the buffers generated by translate. */
@@ -774,7 +800,7 @@ u_vbuf_upload_buffers(struct u_vbuf_priv *mgr,
 
       assert(vb->buffer);
 
-      if (!u_vbuf_resource(vb->buffer)->user_ptr) {
+      if (!vb->buffer->user_ptr) {
          continue;
       }
 
@@ -820,129 +846,35 @@ u_vbuf_upload_buffers(struct u_vbuf_priv *mgr,
       start = start_offset[i];
       assert(start < end);
 
-      real_vb = &mgr->b.real_vertex_buffer[i];
-      ptr = u_vbuf_resource(mgr->b.vertex_buffer[i].buffer)->user_ptr;
+      real_vb = &mgr->real_vertex_buffer[i];
+      ptr = mgr->vertex_buffer[i].buffer->user_ptr;
 
-      u_upload_data(mgr->b.uploader, start, end - start, ptr + start,
+      u_upload_data(mgr->uploader, start, end - start, ptr + start,
                     &real_vb->buffer_offset, &real_vb->buffer);
 
       real_vb->buffer_offset -= start;
    }
 }
 
-unsigned u_vbuf_draw_max_vertex_count(struct u_vbuf *mgrb)
-{
-   struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
-   unsigned i, nr = mgr->ve->count;
-   struct pipe_vertex_element *velems =
-         mgr->fallback_ve ? mgr->fallback_velems : mgr->ve->ve;
-   unsigned result = ~0;
-
-   for (i = 0; i < nr; i++) {
-      struct pipe_vertex_buffer *vb =
-            &mgr->b.real_vertex_buffer[velems[i].vertex_buffer_index];
-      unsigned size, max_count, value;
-
-      /* We're not interested in constant and per-instance attribs. */
-      if (!vb->buffer ||
-          !vb->stride ||
-          velems[i].instance_divisor) {
-         continue;
-      }
-
-      size = vb->buffer->width0;
-
-      /* Subtract buffer_offset. */
-      value = vb->buffer_offset;
-      if (value >= size) {
-         return 0;
-      }
-      size -= value;
-
-      /* Subtract src_offset. */
-      value = velems[i].src_offset;
-      if (value >= size) {
-         return 0;
-      }
-      size -= value;
-
-      /* Subtract format_size. */
-      value = mgr->ve->native_format_size[i];
-      if (value >= size) {
-         return 0;
-      }
-      size -= value;
-
-      /* Compute the max count. */
-      max_count = 1 + size / vb->stride;
-      result = MIN2(result, max_count);
-   }
-   return result;
-}
-
-static boolean u_vbuf_need_minmax_index(struct u_vbuf_priv *mgr)
+static boolean u_vbuf_need_minmax_index(struct u_vbuf *mgr)
 {
-   unsigned i, nr = mgr->ve->count;
-
-   for (i = 0; i < nr; i++) {
-      struct pipe_vertex_buffer *vb;
-      unsigned index;
-
-      /* Per-instance attribs don't need min/max_index. */
-      if (mgr->ve->ve[i].instance_divisor) {
-         continue;
-      }
-
-      index = mgr->ve->ve[i].vertex_buffer_index;
-      vb = &mgr->b.vertex_buffer[index];
-
-      /* Constant attribs don't need min/max_index. */
-      if (!vb->stride) {
-         continue;
-      }
-
-      /* Per-vertex attribs need min/max_index. */
-      if (u_vbuf_resource(vb->buffer)->user_ptr ||
-          mgr->ve->incompatible_layout_elem[i] ||
-          mgr->incompatible_vb[index]) {
-         return TRUE;
-      }
-   }
-
-   return FALSE;
+   /* See if there are any per-vertex attribs which will be uploaded or
+    * translated. Use bitmasks to get the info instead of looping over vertex
+    * elements. */
+   return ((mgr->user_vb_mask | mgr->incompatible_vb_mask |
+            mgr->ve->incompatible_vb_mask_any) &
+           mgr->ve->noninstance_vb_mask_any & mgr->nonzero_stride_vb_mask) != 0;
 }
 
-static boolean u_vbuf_mapping_vertex_buffer_blocks(struct u_vbuf_priv *mgr)
+static boolean u_vbuf_mapping_vertex_buffer_blocks(struct u_vbuf *mgr)
 {
-   unsigned i, nr = mgr->ve->count;
-
-   for (i = 0; i < nr; i++) {
-      struct pipe_vertex_buffer *vb;
-      unsigned index;
-
-      /* Per-instance attribs are not per-vertex data. */
-      if (mgr->ve->ve[i].instance_divisor) {
-         continue;
-      }
-
-      index = mgr->ve->ve[i].vertex_buffer_index;
-      vb = &mgr->b.vertex_buffer[index];
-
-      /* Constant attribs are not per-vertex data. */
-      if (!vb->stride) {
-         continue;
-      }
-
-      /* Return true for the hw buffers which don't need to be translated. */
-      /* XXX we could use some kind of a is-busy query. */
-      if (!u_vbuf_resource(vb->buffer)->user_ptr &&
-          !mgr->ve->incompatible_layout_elem[i] &&
-          !mgr->incompatible_vb[index]) {
-         return TRUE;
-      }
-   }
-
-   return FALSE;
+   /* Return true if there are hw buffers which don't need to be translated.
+    *
+    * We could query whether each buffer is busy, but that would
+    * be way more costly than this. */
+   return (~mgr->user_vb_mask & ~mgr->incompatible_vb_mask &
+           mgr->ve->compatible_vb_mask_all & mgr->ve->noninstance_vb_mask_any &
+           mgr->nonzero_stride_vb_mask) != 0;
 }
 
 static void u_vbuf_get_minmax_index(struct pipe_context *pipe,
@@ -956,8 +888,8 @@ static void u_vbuf_get_minmax_index(struct pipe_context *pipe,
    unsigned i;
    unsigned restart_index = info->restart_index;
 
-   if (u_vbuf_resource(ib->buffer)->user_ptr) {
-      indices = u_vbuf_resource(ib->buffer)->user_ptr +
+   if (ib->buffer->user_ptr) {
+      indices = ib->buffer->user_ptr +
                 ib->offset + info->start * ib->index_size;
    } else {
       indices = pipe_buffer_map_range(pipe, ib->buffer,
@@ -1044,38 +976,42 @@ static void u_vbuf_get_minmax_index(struct pipe_context *pipe,
    }
 }
 
-enum u_vbuf_return_flags
-u_vbuf_draw_begin(struct u_vbuf *mgrb,
-                  struct pipe_draw_info *info)
+void u_vbuf_draw_vbo(struct u_vbuf *mgr, const struct pipe_draw_info *info)
 {
-   struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
+   struct pipe_context *pipe = mgr->pipe;
    int start_vertex, min_index;
    unsigned num_vertices;
-   bool unroll_indices = false;
+   boolean unroll_indices = FALSE;
+   uint32_t user_vb_mask = mgr->user_vb_mask;
+
+   /* Normal draw. No fallback and no user buffers. */
+   if (!mgr->incompatible_vb_mask &&
+       !mgr->ve->incompatible_elem_mask &&
+       !user_vb_mask) {
+      /* Set vertex buffers if needed. */
+      if (mgr->vertex_buffers_dirty) {
+         pipe->set_vertex_buffers(pipe, mgr->nr_real_vertex_buffers,
+                                  mgr->real_vertex_buffer);
+         mgr->vertex_buffers_dirty = FALSE;
+      }
 
-   if (!mgr->incompatible_vb_layout &&
-       !mgr->ve->incompatible_layout &&
-       !mgr->any_user_vbs) {
-      return 0;
+      pipe->draw_vbo(pipe, info);
+      return;
    }
 
    if (info->indexed) {
-      int max_index;
-      bool index_bounds_valid = false;
-
-      if (info->max_index != ~0) {
-         min_index = info->min_index;
-         max_index = info->max_index;
-         index_bounds_valid = true;
-      } else if (u_vbuf_need_minmax_index(mgr)) {
-         u_vbuf_get_minmax_index(mgr->pipe, &mgr->b.index_buffer, info,
-                                 &min_index, &max_index);
-         index_bounds_valid = true;
-      }
+      /* See if anything needs to be done for per-vertex attribs. */
+      if (u_vbuf_need_minmax_index(mgr)) {
+         int max_index;
+
+         if (info->max_index != ~0) {
+            min_index = info->min_index;
+            max_index = info->max_index;
+         } else {
+            u_vbuf_get_minmax_index(mgr->pipe, &mgr->index_buffer, info,
+                                    &min_index, &max_index);
+         }
 
-      /* If the index bounds are valid, it means some upload or translation
-       * of per-vertex attribs will be performed. */
-      if (index_bounds_valid) {
          assert(min_index <= max_index);
 
          start_vertex = min_index + info->index_bias;
@@ -1090,7 +1026,9 @@ u_vbuf_draw_begin(struct u_vbuf *mgrb,
              num_vertices-info->count > 32 &&
              !u_vbuf_mapping_vertex_buffer_blocks(mgr)) {
             /*printf("num_vertices=%i count=%i\n", num_vertices, info->count);*/
-            unroll_indices = true;
+            unroll_indices = TRUE;
+            user_vb_mask &= ~(mgr->nonzero_stride_vb_mask &
+                              mgr->ve->noninstance_vb_mask_any);
          }
       } else {
          /* Nothing to do for per-vertex attribs. */
@@ -1106,17 +1044,20 @@ u_vbuf_draw_begin(struct u_vbuf *mgrb,
 
    /* Translate vertices with non-native layouts or formats. */
    if (unroll_indices ||
-       mgr->incompatible_vb_layout ||
-       mgr->ve->incompatible_layout) {
+       mgr->incompatible_vb_mask ||
+       mgr->ve->incompatible_elem_mask) {
       /* XXX check the return value */
       u_vbuf_translate_begin(mgr, start_vertex, num_vertices,
                              info->start_instance, info->instance_count,
                              info->start, info->count, min_index,
                              unroll_indices);
+
+      user_vb_mask &= ~(mgr->incompatible_vb_mask |
+                        mgr->ve->incompatible_vb_mask_all);
    }
 
    /* Upload user buffers. */
-   if (mgr->any_user_vbs) {
+   if (user_vb_mask) {
       u_vbuf_upload_buffers(mgr, start_vertex, num_vertices,
                             info->start_instance, info->instance_count);
    }
@@ -1130,34 +1071,75 @@ u_vbuf_draw_begin(struct u_vbuf *mgrb,
    }
 
    unsigned i;
-   for (i = 0; i < mgr->b.nr_vertex_buffers; i++) {
+   for (i = 0; i < mgr->nr_vertex_buffers; i++) {
       printf("input %i: ", i);
-      util_dump_vertex_buffer(stdout, mgr->b.vertex_buffer+i);
+      util_dump_vertex_buffer(stdout, mgr->vertex_buffer+i);
       printf("\n");
    }
-   for (i = 0; i < mgr->b.nr_real_vertex_buffers; i++) {
+   for (i = 0; i < mgr->nr_real_vertex_buffers; i++) {
       printf("real %i: ", i);
-      util_dump_vertex_buffer(stdout, mgr->b.real_vertex_buffer+i);
+      util_dump_vertex_buffer(stdout, mgr->real_vertex_buffer+i);
       printf("\n");
    }
    */
 
-   if (unroll_indices) {
-      info->indexed = FALSE;
-      info->index_bias = 0;
-      info->min_index = 0;
-      info->max_index = info->count - 1;
-      info->start = 0;
+   u_upload_unmap(mgr->uploader);
+   pipe->set_vertex_buffers(pipe, mgr->nr_real_vertex_buffers,
+                            mgr->real_vertex_buffer);
+
+   if (unlikely(unroll_indices)) {
+      struct pipe_draw_info new_info = *info;
+      new_info.indexed = FALSE;
+      new_info.index_bias = 0;
+      new_info.min_index = 0;
+      new_info.max_index = info->count - 1;
+      new_info.start = 0;
+
+      pipe->draw_vbo(pipe, &new_info);
+   } else {
+      pipe->draw_vbo(pipe, info);
+   }
+
+   if (mgr->using_translate) {
+      u_vbuf_translate_end(mgr);
    }
+   mgr->vertex_buffers_dirty = TRUE;
+}
 
-   return U_VBUF_BUFFERS_UPDATED;
+void u_vbuf_save_vertex_elements(struct u_vbuf *mgr)
+{
+   assert(!mgr->ve_saved);
+   mgr->ve_saved = mgr->ve;
 }
 
-void u_vbuf_draw_end(struct u_vbuf *mgrb)
+void u_vbuf_restore_vertex_elements(struct u_vbuf *mgr)
 {
-   struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
+   if (mgr->ve != mgr->ve_saved) {
+      struct pipe_context *pipe = mgr->pipe;
 
-   if (mgr->fallback_ve) {
-      u_vbuf_translate_end(mgr);
+      mgr->ve = mgr->ve_saved;
+      pipe->bind_vertex_elements_state(pipe,
+                                       mgr->ve ? mgr->ve->driver_cso : NULL);
+   }
+   mgr->ve_saved = NULL;
+}
+
+void u_vbuf_save_vertex_buffers(struct u_vbuf *mgr)
+{
+   util_copy_vertex_buffers(mgr->vertex_buffer_saved,
+                            &mgr->nr_vertex_buffers_saved,
+                            mgr->vertex_buffer,
+                            mgr->nr_vertex_buffers);
+}
+
+void u_vbuf_restore_vertex_buffers(struct u_vbuf *mgr)
+{
+   unsigned i;
+
+   u_vbuf_set_vertex_buffers(mgr, mgr->nr_vertex_buffers_saved,
+                             mgr->vertex_buffer_saved);
+   for (i = 0; i < mgr->nr_vertex_buffers_saved; i++) {
+      pipe_resource_reference(&mgr->vertex_buffer_saved[i].buffer, NULL);
    }
+   mgr->nr_vertex_buffers_saved = 0;
 }
diff --git a/mesalib/src/gallium/auxiliary/util/u_vbuf.h b/mesalib/src/gallium/auxiliary/util/u_vbuf.h
index 3669c9b87..59eb59a11 100644
--- a/mesalib/src/gallium/auxiliary/util/u_vbuf.h
+++ b/mesalib/src/gallium/auxiliary/util/u_vbuf.h
@@ -25,8 +25,8 @@
  *
  **************************************************************************/
 
-#ifndef U_VBUF_MGR_H
-#define U_VBUF_MGR_H
+#ifndef U_VBUF_H
+#define U_VBUF_H
 
 /* This module builds upon u_upload_mgr and translate_cache and takes care of
  * user buffer uploads and vertex format fallbacks. It's designed
@@ -35,7 +35,9 @@
 
 #include "pipe/p_context.h"
 #include "pipe/p_state.h"
-#include "util/u_transfer.h"
+
+struct cso_context;
+struct u_vbuf;
 
 /* Hardware vertex fetcher limitations can be described by this structure. */
 struct u_vbuf_caps {
@@ -47,99 +49,38 @@ struct u_vbuf_caps {
    unsigned format_norm32:1;     /* PIPE_FORMAT_*32*NORM */
    unsigned format_scaled32:1;   /* PIPE_FORMAT_*32*SCALED */
 
-   /* Whether vertex fetches don't have to be dword-aligned. */
+   /* Whether vertex fetches don't have to be 4-byte-aligned. */
    /* TRUE if hardware supports it. */
-   unsigned fetch_dword_unaligned:1;
-};
-
-/* The manager.
- * This structure should also be used to access vertex buffers
- * from a driver. */
-struct u_vbuf {
-   /* This is what was set in set_vertex_buffers.
-    * May contain user buffers. */
-   struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
-   unsigned nr_vertex_buffers;
-
-   /* Contains only real vertex buffers.
-    * Hardware drivers should use real_vertex_buffers[i]
-    * instead of vertex_buffers[i].buffer. */
-   struct pipe_vertex_buffer real_vertex_buffer[PIPE_MAX_ATTRIBS];
-   int nr_real_vertex_buffers;
-
-   /* The index buffer. */
-   struct pipe_index_buffer index_buffer;
-
-   /* This uploader can optionally be used by the driver.
-    *
-    * Allowed functions:
-    * - u_upload_alloc
-    * - u_upload_data
-    * - u_upload_buffer
-    * - u_upload_flush */
-   struct u_upload_mgr *uploader;
-
-   struct u_vbuf_caps caps;
-};
-
-struct u_vbuf_resource {
-   struct u_resource b;
-   uint8_t *user_ptr;
-};
+   unsigned buffer_offset_unaligned:1;
+   unsigned buffer_stride_unaligned:1;
+   unsigned velem_src_offset_unaligned:1;
 
-/* Opaque type containing information about vertex elements for the manager. */
-struct u_vbuf_elements;
-
-enum u_fetch_alignment {
-   U_VERTEX_FETCH_BYTE_ALIGNED,
-   U_VERTEX_FETCH_DWORD_ALIGNED
+   /* Whether the driver supports user vertex buffers. */
+   unsigned user_vertex_buffers:1;
 };
 
-enum u_vbuf_return_flags {
-   U_VBUF_BUFFERS_UPDATED = 1
-};
 
+void u_vbuf_get_caps(struct pipe_screen *screen, struct u_vbuf_caps *caps);
 
 struct u_vbuf *
 u_vbuf_create(struct pipe_context *pipe,
-              unsigned upload_buffer_size,
-              unsigned upload_buffer_alignment,
-              unsigned upload_buffer_bind,
-              enum u_fetch_alignment fetch_alignment);
+              struct u_vbuf_caps *caps);
 
 void u_vbuf_destroy(struct u_vbuf *mgr);
 
-struct u_vbuf_elements *
-u_vbuf_create_vertex_elements(struct u_vbuf *mgr,
-                              unsigned count,
-                              const struct pipe_vertex_element *attrs,
-                              struct pipe_vertex_element *native_attrs);
-
-void u_vbuf_bind_vertex_elements(struct u_vbuf *mgr,
-                                 void *cso,
-                                 struct u_vbuf_elements *ve);
-
-void u_vbuf_destroy_vertex_elements(struct u_vbuf *mgr,
-                                    struct u_vbuf_elements *ve);
-
-void u_vbuf_set_vertex_buffers(struct u_vbuf *mgr,
-                               unsigned count,
+/* State and draw functions. */
+void u_vbuf_set_vertex_elements(struct u_vbuf *mgr, unsigned count,
+                                const struct pipe_vertex_element *states);
+void u_vbuf_set_vertex_buffers(struct u_vbuf *mgr, unsigned count,
                                const struct pipe_vertex_buffer *bufs);
-
 void u_vbuf_set_index_buffer(struct u_vbuf *mgr,
                              const struct pipe_index_buffer *ib);
+void u_vbuf_draw_vbo(struct u_vbuf *mgr, const struct pipe_draw_info *info);
 
-enum u_vbuf_return_flags u_vbuf_draw_begin(struct u_vbuf *mgr,
-                                           struct pipe_draw_info *info);
-
-unsigned u_vbuf_draw_max_vertex_count(struct u_vbuf *mgr);
-
-void u_vbuf_draw_end(struct u_vbuf *mgr);
-
-
-static INLINE struct u_vbuf_resource *u_vbuf_resource(struct pipe_resource *r)
-{
-   return (struct u_vbuf_resource*)r;
-}
+/* Save/restore functionality. */
+void u_vbuf_save_vertex_elements(struct u_vbuf *mgr);
+void u_vbuf_restore_vertex_elements(struct u_vbuf *mgr);
+void u_vbuf_save_vertex_buffers(struct u_vbuf *mgr);
+void u_vbuf_restore_vertex_buffers(struct u_vbuf *mgr);
 
 #endif
diff --git a/mesalib/src/glsl/Makefile b/mesalib/src/glsl/Makefile
index f6c722997..7c80c95d8 100644
--- a/mesalib/src/glsl/Makefile
+++ b/mesalib/src/glsl/Makefile
@@ -165,7 +165,7 @@ glcpp/glcpp-parse.c: glcpp/glcpp-parse.y
 builtin_compiler: $(GLSL2_OBJECTS) $(OBJECTS) builtin_stubs.o
 	$(APP_CXX) $(INCLUDES) $(CXXFLAGS) $(LDFLAGS) $(OBJECTS) $(GLSL2_OBJECTS) builtin_stubs.o -o $@
 
-builtin_function.cpp: builtins/profiles/* builtins/ir/* builtins/tools/generate_builtins.py builtins/tools/texture_builtins.py builtin_compiler
+builtin_function.cpp: builtins/profiles/* builtins/ir/* builtins/glsl/* builtins/tools/generate_builtins.py builtins/tools/texture_builtins.py builtin_compiler
 	@echo Regenerating builtin_function.cpp...
 	$(PYTHON2) $(PYTHON_FLAGS) builtins/tools/generate_builtins.py ./builtin_compiler > builtin_function.cpp || rm -f builtin_function.cpp
 
diff --git a/mesalib/src/glsl/ir_reader.cpp b/mesalib/src/glsl/ir_reader.cpp
index 7ce683ef7..03dbb67c3 100644
--- a/mesalib/src/glsl/ir_reader.cpp
+++ b/mesalib/src/glsl/ir_reader.cpp
@@ -407,6 +407,8 @@ ir_reader::read_declaration(s_expression *expr)
 	 var->mode = ir_var_out;
       } else if (strcmp(qualifier->value(), "inout") == 0) {
 	 var->mode = ir_var_inout;
+      } else if (strcmp(qualifier->value(), "temporary") == 0) {
+	 var->mode = ir_var_temporary;
       } else if (strcmp(qualifier->value(), "smooth") == 0) {
 	 var->interpolation = INTERP_QUALIFIER_SMOOTH;
       } else if (strcmp(qualifier->value(), "flat") == 0) {
diff --git a/mesalib/src/glsl/main.cpp b/mesalib/src/glsl/main.cpp
index 8c651b8c0..fbcc4aebb 100644
--- a/mesalib/src/glsl/main.cpp
+++ b/mesalib/src/glsl/main.cpp
@@ -33,6 +33,15 @@
 #define O_RDONLY _O_RDONLY
 #endif
 
+/** @file main.cpp
+ *
+ * This file is the main() routine and scaffolding for producing
+ * builtin_compiler (which doesn't include builtins itself and is used
+ * to generate the profile information for builtin_function.cpp), and
+ * for glsl_compiler (which does include builtins and can be used to
+ * offline compile GLSL code and examine the resulting GLSL IR.
+ */
+
 #include "ast.h"
 #include "glsl_parser_extras.h"
 #include "ir_optimization.h"
diff --git a/mesalib/src/glsl/opt_dead_functions.cpp b/mesalib/src/glsl/opt_dead_functions.cpp
index 017a28639..f50349395 100644
--- a/mesalib/src/glsl/opt_dead_functions.cpp
+++ b/mesalib/src/glsl/opt_dead_functions.cpp
@@ -62,8 +62,6 @@ public:
 
    signature_entry *get_signature_entry(ir_function_signature *var);
 
-   bool (*predicate)(ir_instruction *ir);
-
    /* List of signature_entry */
    exec_list signature_list;
    void *mem_ctx;
diff --git a/mesalib/src/glsl/opt_structure_splitting.cpp b/mesalib/src/glsl/opt_structure_splitting.cpp
index 6dd228e4e..a21238dd0 100644
--- a/mesalib/src/glsl/opt_structure_splitting.cpp
+++ b/mesalib/src/glsl/opt_structure_splitting.cpp
@@ -201,7 +201,6 @@ public:
    variable_entry2 *get_splitting_entry(ir_variable *var);
 
    exec_list *variable_list;
-   void *mem_ctx;
 };
 
 variable_entry2 *
diff --git a/mesalib/src/mesa/main/version.c b/mesalib/src/mesa/main/version.c
index 5b8b65ddf..09a930cf2 100644
--- a/mesalib/src/mesa/main/version.c
+++ b/mesalib/src/mesa/main/version.c
@@ -124,6 +124,7 @@ compute_version(struct gl_context *ctx)
                               ctx->Extensions.EXT_texture_sRGB);
    const GLboolean ver_3_0 = (ver_2_1 &&
                               ctx->Const.GLSLVersion >= 130 &&
+                              ctx->Const.MaxSamples >= 4 &&
                               ctx->Extensions.ARB_color_buffer_float &&
                               ctx->Extensions.ARB_depth_buffer_float &&
                               ctx->Extensions.ARB_half_float_pixel &&
diff --git a/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c b/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c
index 5e078a85e..9a3f22465 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c
@@ -1509,7 +1509,15 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
    readY = srcy;
    readW = width;
    readH = height;
-   _mesa_clip_readpixels(ctx, &readX, &readY, &readW, &readH, &pack);
+   if (!_mesa_clip_readpixels(ctx, &readX, &readY, &readW, &readH, &pack)) {
+      /* The source region is completely out of bounds.  Do nothing.
+       * The GL spec says "Results of copies from outside the window,
+       * or from regions of the window that are not exposed, are
+       * hardware dependent and undefined."
+       */
+      return;
+   }
+
    readW = MAX2(0, readW);
    readH = MAX2(0, readH);
 
diff --git a/mesalib/src/mesa/state_tracker/st_context.c b/mesalib/src/mesa/state_tracker/st_context.c
index 0245fd92b..19d9da131 100644
--- a/mesalib/src/mesa/state_tracker/st_context.c
+++ b/mesalib/src/mesa/state_tracker/st_context.c
@@ -65,6 +65,7 @@
 #include "util/u_inlines.h"
 #include "util/u_upload_mgr.h"
 #include "cso_cache/cso_context.h"
+#include "util/u_vbuf.h"
 
 
 DEBUG_GET_ONCE_BOOL_OPTION(mesa_mvp_dp4, "MESA_MVP_DP4", FALSE)
@@ -111,6 +112,29 @@ st_get_msaa(void)
 }
 
 
+static void st_init_vbuf(struct st_context *st)
+{
+   struct u_vbuf_caps caps;
+
+   u_vbuf_get_caps(st->pipe->screen, &caps);
+
+   /* Create u_vbuf if there is anything unsupported. */
+   if (!caps.buffer_offset_unaligned ||
+       !caps.buffer_stride_unaligned ||
+       !caps.velem_src_offset_unaligned ||
+       !caps.format_fixed32 ||
+       !caps.format_float16 ||
+       !caps.format_float64 ||
+       !caps.format_norm32 ||
+       !caps.format_scaled32 ||
+       !caps.user_vertex_buffers) {
+      /* XXX user vertex buffers are always uploaded regardless of the CAP. */
+      st->vbuf = u_vbuf_create(st->pipe, &caps);
+      cso_install_vbuf(st->cso_context, st->vbuf);
+   }
+}
+
+
 static struct st_context *
 st_create_context_priv( struct gl_context *ctx, struct pipe_context *pipe )
 {
@@ -134,6 +158,7 @@ st_create_context_priv( struct gl_context *ctx, struct pipe_context *pipe )
    st->uploader = u_upload_create(st->pipe, 65536, 4, PIPE_BIND_VERTEX_BUFFER);
    st->cso_context = cso_create_context(pipe);
 
+   st_init_vbuf(st);
    st_init_atoms( st );
    st_init_bitmap(st);
    st_init_clear(st);
@@ -245,6 +270,7 @@ static void st_destroy_context_priv( struct st_context *st )
 void st_destroy_context( struct st_context *st )
 {
    struct pipe_context *pipe = st->pipe;
+   struct u_vbuf *vbuf = st->vbuf;
    struct cso_context *cso = st->cso_context;
    struct gl_context *ctx = st->ctx;
    GLuint i;
@@ -275,7 +301,13 @@ void st_destroy_context( struct st_context *st )
 
    _mesa_free_context_data(ctx);
 
+   /* This will free the st_context too, so 'st' must not be accessed
+    * afterwards. */
    st_destroy_context_priv(st);
+   st = NULL;
+
+   if (vbuf)
+      u_vbuf_destroy(vbuf);
 
    cso_destroy_context(cso);
 
diff --git a/mesalib/src/mesa/state_tracker/st_context.h b/mesalib/src/mesa/state_tracker/st_context.h
index e3d65d97a..3ec98ada1 100644
--- a/mesalib/src/mesa/state_tracker/st_context.h
+++ b/mesalib/src/mesa/state_tracker/st_context.h
@@ -41,6 +41,7 @@ struct gen_mipmap_state;
 struct st_context;
 struct st_fragment_program;
 struct u_upload_mgr;
+struct u_vbuf;
 
 
 #define ST_NEW_MESA                    (1 << 0) /* Mesa state has changed */
@@ -73,6 +74,8 @@ struct st_context
    struct pipe_context *pipe;
 
    struct u_upload_mgr *uploader;
+   struct u_vbuf *vbuf;
+
    struct draw_context *draw;  /**< For selection/feedback/rastpos only */
    struct draw_stage *feedback_stage;  /**< For GL_FEEDBACK rendermode */
    struct draw_stage *selection_stage;  /**< For GL_SELECT rendermode */
diff --git a/mesalib/src/mesa/state_tracker/st_draw.c b/mesalib/src/mesa/state_tracker/st_draw.c
index 0a35ab2be..edab76bf5 100644
--- a/mesalib/src/mesa/state_tracker/st_draw.c
+++ b/mesalib/src/mesa/state_tracker/st_draw.c
@@ -795,7 +795,8 @@ find_sub_primitives(const void *elements, unsigned element_size,
  * sub-primitives.
  */
 static void
-handle_fallback_primitive_restart(struct pipe_context *pipe,
+handle_fallback_primitive_restart(struct cso_context *cso,
+                                  struct pipe_context *pipe,
                                   const struct _mesa_index_buffer *ib,
                                   struct pipe_index_buffer *ibuffer,
                                   struct pipe_draw_info *orig_info)
@@ -851,7 +852,7 @@ handle_fallback_primitive_restart(struct pipe_context *pipe,
          info.start = sub_prims[i].start;
          info.count = sub_prims[i].count;
          if (u_trim_pipe_prim(info.mode, &info.count)) {
-            pipe->draw_vbo(pipe, &info);
+            cso_draw_vbo(cso, &info);
          }
       }
    }
@@ -1075,7 +1076,7 @@ st_draw_vbo(struct gl_context *ctx,
    }
 
    setup_index_buffer(ctx, ib, &ibuffer);
-   pipe->set_index_buffer(pipe, &ibuffer);
+   cso_set_index_buffer(st->cso_context, &ibuffer);
 
    util_draw_init_info(&info);
    if (ib) {
@@ -1110,20 +1111,21 @@ st_draw_vbo(struct gl_context *ctx,
       }
 
       if (info.count_from_stream_output) {
-         pipe->draw_vbo(pipe, &info);
+         cso_draw_vbo(st->cso_context, &info);
       }
       else if (info.primitive_restart) {
          if (st->sw_primitive_restart) {
             /* Handle primitive restart for drivers that doesn't support it */
-            handle_fallback_primitive_restart(pipe, ib, &ibuffer, &info);
+            handle_fallback_primitive_restart(st->cso_context, pipe, ib,
+                                              &ibuffer, &info);
          }
          else {
             /* don't trim, restarts might be inside index list */
-            pipe->draw_vbo(pipe, &info);
+            cso_draw_vbo(st->cso_context, &info);
          }
       }
       else if (u_trim_pipe_prim(info.mode, &info.count))
-         pipe->draw_vbo(pipe, &info);
+         cso_draw_vbo(st->cso_context, &info);
    }
 
    pipe_resource_reference(&ibuffer.buffer, NULL);
diff --git a/mesalib/src/mesa/state_tracker/st_extensions.c b/mesalib/src/mesa/state_tracker/st_extensions.c
index 34e0329be..1b4bca681 100644
--- a/mesalib/src/mesa/state_tracker/st_extensions.c
+++ b/mesalib/src/mesa/state_tracker/st_extensions.c
@@ -459,12 +459,6 @@ void st_init_extensions(struct st_context *st)
 
    /* Required: vertex fetch support. */
    static const struct st_extension_format_mapping vertex_mapping[] = {
-      { { o(ARB_ES2_compatibility) },
-        { PIPE_FORMAT_R32G32B32A32_FIXED } },
-
-      { { o(ARB_half_float_vertex) },
-        { PIPE_FORMAT_R16G16B16A16_FLOAT } },
-
       { { o(ARB_vertex_type_2_10_10_10_rev) },
         { PIPE_FORMAT_R10G10B10A2_UNORM,
           PIPE_FORMAT_B10G10R10A2_UNORM,
@@ -479,6 +473,7 @@ void st_init_extensions(struct st_context *st)
    /*
     * Extensions that are supported by all Gallium drivers:
     */
+   ctx->Extensions.ARB_ES2_compatibility = GL_TRUE;
    ctx->Extensions.ARB_copy_buffer = GL_TRUE;
    ctx->Extensions.ARB_draw_elements_base_vertex = GL_TRUE;
    ctx->Extensions.ARB_explicit_attrib_location = GL_TRUE;
@@ -486,6 +481,7 @@ void st_init_extensions(struct st_context *st)
    ctx->Extensions.ARB_fragment_program = GL_TRUE;
    ctx->Extensions.ARB_fragment_shader = GL_TRUE;
    ctx->Extensions.ARB_half_float_pixel = GL_TRUE;
+   ctx->Extensions.ARB_half_float_vertex = GL_TRUE;
    ctx->Extensions.ARB_map_buffer_range = GL_TRUE;
    ctx->Extensions.ARB_sampler_objects = GL_TRUE;
    ctx->Extensions.ARB_shader_objects = GL_TRUE;
diff --git a/mesalib/src/mesa/vbo/vbo_exec.h b/mesalib/src/mesa/vbo/vbo_exec.h
index 5cdf5ced9..be9f3d78d 100644
--- a/mesalib/src/mesa/vbo/vbo_exec.h
+++ b/mesalib/src/mesa/vbo/vbo_exec.h
@@ -189,7 +189,8 @@ static inline void
 vbo_draw_method(struct vbo_exec_context *exec, enum draw_method method)
 {
    if (exec->last_draw_method != method) {
-      exec->ctx->NewState |= _NEW_ARRAY;
+      struct gl_context *ctx = exec->ctx;
+      ctx->Driver.UpdateState(ctx, _NEW_ARRAY);
       exec->last_draw_method = method;
    }
 }
diff --git a/mesalib/src/mesa/vbo/vbo_exec_array.c b/mesalib/src/mesa/vbo/vbo_exec_array.c
index f95e062fb..2dcfb8e5b 100644
--- a/mesalib/src/mesa/vbo/vbo_exec_array.c
+++ b/mesalib/src/mesa/vbo/vbo_exec_array.c
@@ -438,14 +438,6 @@ recalculate_input_bindings(struct gl_context *ctx)
 	 inputs[VERT_ATTRIB_GENERIC(i)] = &vbo->currval[VBO_ATTRIB_GENERIC0+i];
          const_inputs |= VERT_BIT_GENERIC(i);
       }
-
-      /* There is no need to make _NEW_ARRAY dirty here for the TnL program,
-       * because it already takes care of invalidating the state necessary
-       * to revalidate vertex arrays. Not marking the state as dirty also
-       * improves performance (quite significantly in some apps).
-       */
-      if (!ctx->VertexProgram._MaintainTnlProgram)
-         ctx->NewState |= _NEW_ARRAY;
       break;
 
    case VP_NV:
@@ -472,8 +464,6 @@ recalculate_input_bindings(struct gl_context *ctx)
 	 inputs[VERT_ATTRIB_GENERIC(i)] = &vbo->currval[VBO_ATTRIB_GENERIC0+i];
          const_inputs |= VERT_BIT_GENERIC(i);
       }
-
-      ctx->NewState |= _NEW_ARRAY;
       break;
 
    case VP_ARB:
@@ -512,11 +502,11 @@ recalculate_input_bindings(struct gl_context *ctx)
       }
 
       inputs[VERT_ATTRIB_GENERIC0] = inputs[0];
-      ctx->NewState |= _NEW_ARRAY;
       break;
    }
 
    _mesa_set_varying_vp_inputs( ctx, VERT_BIT_ALL & (~const_inputs) );
+   ctx->Driver.UpdateState(ctx, _NEW_ARRAY);
 }
 
 
@@ -640,11 +630,11 @@ vbo_exec_DrawArrays(GLenum mode, GLint start, GLsizei count)
       _mesa_debug(ctx, "glDrawArrays(%s, %d, %d)\n",
                   _mesa_lookup_enum_by_nr(mode), start, count);
 
+   FLUSH_CURRENT(ctx, 0);
+
    if (!_mesa_validate_DrawArrays( ctx, mode, start, count ))
       return;
 
-   FLUSH_CURRENT( ctx, 0 );
-
    if (0)
       check_draw_arrays_data(ctx, start, count);
 
@@ -669,11 +659,11 @@ vbo_exec_DrawArraysInstanced(GLenum mode, GLint start, GLsizei count,
       _mesa_debug(ctx, "glDrawArraysInstanced(%s, %d, %d, %d)\n",
                   _mesa_lookup_enum_by_nr(mode), start, count, numInstances);
 
+   FLUSH_CURRENT(ctx, 0);
+
    if (!_mesa_validate_DrawArraysInstanced(ctx, mode, start, count, numInstances))
       return;
 
-   FLUSH_CURRENT( ctx, 0 );
-
    if (0)
       check_draw_arrays_data(ctx, start, count);
 
@@ -761,8 +751,6 @@ vbo_validated_drawrangeelements(struct gl_context *ctx, GLenum mode,
    struct _mesa_index_buffer ib;
    struct _mesa_prim prim[1];
 
-   FLUSH_CURRENT( ctx, 0 );
-
    vbo_bind_arrays(ctx);
 
    ib.count = count;
@@ -838,6 +826,8 @@ vbo_exec_DrawRangeElementsBaseVertex(GLenum mode,
                 _mesa_lookup_enum_by_nr(mode), start, end, count,
                 _mesa_lookup_enum_by_nr(type), indices, basevertex);
 
+   FLUSH_CURRENT(ctx, 0);
+
    if (!_mesa_validate_DrawRangeElements( ctx, mode, start, end, count,
                                           type, indices, basevertex ))
       return;
@@ -936,6 +926,8 @@ vbo_exec_DrawElements(GLenum mode, GLsizei count, GLenum type,
                   _mesa_lookup_enum_by_nr(mode), count,
                   _mesa_lookup_enum_by_nr(type), indices);
 
+   FLUSH_CURRENT(ctx, 0);
+
    if (!_mesa_validate_DrawElements( ctx, mode, count, type, indices, 0 ))
       return;
 
@@ -958,6 +950,8 @@ vbo_exec_DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type,
                   _mesa_lookup_enum_by_nr(mode), count,
                   _mesa_lookup_enum_by_nr(type), indices, basevertex);
 
+   FLUSH_CURRENT(ctx, 0);
+
    if (!_mesa_validate_DrawElements( ctx, mode, count, type, indices,
 				     basevertex ))
       return;
@@ -981,6 +975,8 @@ vbo_exec_DrawElementsInstanced(GLenum mode, GLsizei count, GLenum type,
                   _mesa_lookup_enum_by_nr(mode), count,
                   _mesa_lookup_enum_by_nr(type), indices, numInstances);
 
+   FLUSH_CURRENT(ctx, 0);
+
    if (!_mesa_validate_DrawElementsInstanced(ctx, mode, count, type, indices,
                                              numInstances, 0))
       return;
@@ -1005,6 +1001,8 @@ vbo_exec_DrawElementsInstancedBaseVertex(GLenum mode, GLsizei count, GLenum type
                   _mesa_lookup_enum_by_nr(type), indices,
                   numInstances, basevertex);
 
+   FLUSH_CURRENT(ctx, 0);
+
    if (!_mesa_validate_DrawElementsInstanced(ctx, mode, count, type, indices,
                                              numInstances, basevertex))
       return;
@@ -1037,8 +1035,6 @@ vbo_validated_multidrawelements(struct gl_context *ctx, GLenum mode,
    if (primcount == 0)
       return;
 
-   FLUSH_CURRENT( ctx, 0 );
-
    prim = calloc(1, primcount * sizeof(*prim));
    if (prim == NULL) {
       _mesa_error(ctx, GL_OUT_OF_MEMORY, "glMultiDrawElements");
@@ -1226,12 +1222,12 @@ vbo_exec_DrawTransformFeedback(GLenum mode, GLuint name)
       _mesa_debug(ctx, "glDrawTransformFeedback(%s, %d)\n",
                   _mesa_lookup_enum_by_nr(mode), name);
 
+   FLUSH_CURRENT(ctx, 0);
+
    if (!_mesa_validate_DrawTransformFeedback(ctx, mode, obj)) {
       return;
    }
 
-   FLUSH_CURRENT(ctx, 0);
-
    vbo_draw_transform_feedback(ctx, mode, obj, 1);
 }
 
diff --git a/mesalib/src/mesa/vbo/vbo_exec_draw.c b/mesalib/src/mesa/vbo/vbo_exec_draw.c
index 5cc6586b3..da5ca695e 100644
--- a/mesalib/src/mesa/vbo/vbo_exec_draw.c
+++ b/mesalib/src/mesa/vbo/vbo_exec_draw.c
@@ -253,11 +253,11 @@ vbo_exec_bind_arrays( struct gl_context *ctx )
 	 arrays[attr]._MaxElement = count; /* ??? */
 
          varying_inputs |= VERT_BIT(attr);
-         ctx->NewState |= _NEW_ARRAY;
       }
    }
 
    _mesa_set_varying_vp_inputs( ctx, varying_inputs );
+   ctx->Driver.UpdateState(ctx, _NEW_ARRAY);
 }
 
 
diff --git a/mesalib/src/mesa/vbo/vbo_save_draw.c b/mesalib/src/mesa/vbo/vbo_save_draw.c
index 57186de64..88a9a7e34 100644
--- a/mesalib/src/mesa/vbo/vbo_save_draw.c
+++ b/mesalib/src/mesa/vbo/vbo_save_draw.c
@@ -209,11 +209,11 @@ static void vbo_bind_vertex_list(struct gl_context *ctx,
 
 	 buffer_offset += node_attrsz[src] * sizeof(GLfloat);
          varying_inputs |= VERT_BIT(attr);
-         ctx->NewState |= _NEW_ARRAY;
       }
    }
 
    _mesa_set_varying_vp_inputs( ctx, varying_inputs );
+   ctx->Driver.UpdateState(ctx, _NEW_ARRAY);
 }