Merge branch 'gallium-vertexelementcso'

Conflicts:
	src/gallium/auxiliary/cso_cache/cso_context.c
	src/gallium/auxiliary/util/u_blit.c
	src/gallium/auxiliary/util/u_gen_mipmap.c
	src/gallium/drivers/i915/i915_context.h
	src/gallium/drivers/i965/brw_draw_upload.c
	src/gallium/drivers/r300/r300_blit.c
	src/gallium/drivers/r300/r300_context.h
	src/gallium/drivers/r300/r300_render.c
	src/gallium/drivers/r300/r300_state_derived.c
	src/mesa/state_tracker/st_cb_bitmap.c
	src/mesa/state_tracker/st_cb_clear.c

16 files changed, 284 insertions, 100 deletions

diff --git a/src/gallium/auxiliary/cso_cache/cso_cache.c b/src/gallium/auxiliary/cso_cache/cso_cache.c
index a6a07e72c2f..900c64df4b9 100644
--- a/src/gallium/auxiliary/cso_cache/cso_cache.c
+++ b/src/gallium/auxiliary/cso_cache/cso_cache.c
@@ -43,6 +43,7 @@ struct cso_cache {
    struct cso_hash *vs_hash;
    struct cso_hash *rasterizer_hash;
    struct cso_hash *sampler_hash;
+   struct cso_hash *velements_hash;
    int    max_size;
 
    cso_sanitize_callback sanitize_cb;
@@ -108,6 +109,9 @@ static struct cso_hash *_cso_hash_for_type(struct cso_cache *sc, enum cso_cache_
    case CSO_VERTEX_SHADER:
       hash = sc->vs_hash;
       break;
+   case CSO_VELEMENTS:
+      hash = sc->velements_hash;
+      break;
    }
 
    return hash;
@@ -161,6 +165,13 @@ static void delete_vs_state(void *state, void *data)
    FREE(state);
 }
 
+static void delete_velements(void *state, void *data)
+{
+   struct cso_velements *cso = (struct cso_velements *)state;
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+}
 
 static INLINE void delete_cso(void *state, enum cso_cache_type type)
 {
@@ -183,6 +194,9 @@ static INLINE void delete_cso(void *state, enum cso_cache_type type)
    case CSO_VERTEX_SHADER:
       delete_vs_state(state, 0);
       break;
+   case CSO_VELEMENTS:
+      delete_velements(state, 0);
+      break;
    default:
       assert(0);
       FREE(state);
@@ -294,6 +308,7 @@ struct cso_cache *cso_cache_create(void)
    sc->rasterizer_hash    = cso_hash_create();
    sc->fs_hash            = cso_hash_create();
    sc->vs_hash            = cso_hash_create();
+   sc->velements_hash     = cso_hash_create();
    sc->sanitize_cb        = sanitize_cb;
    sc->sanitize_data      = 0;
 
@@ -325,6 +340,9 @@ void cso_for_each_state(struct cso_cache *sc, enum cso_cache_type type,
    case CSO_VERTEX_SHADER:
       hash = sc->vs_hash;
       break;
+   case CSO_VELEMENTS:
+      hash = sc->velements_hash;
+      break;
    }
 
    iter = cso_hash_first_node(hash);
@@ -351,6 +369,7 @@ void cso_cache_delete(struct cso_cache *sc)
    cso_for_each_state(sc, CSO_VERTEX_SHADER, delete_vs_state, 0);
    cso_for_each_state(sc, CSO_RASTERIZER, delete_rasterizer_state, 0);
    cso_for_each_state(sc, CSO_SAMPLER, delete_sampler_state, 0);
+   cso_for_each_state(sc, CSO_VELEMENTS, delete_velements, 0);
 
    cso_hash_delete(sc->blend_hash);
    cso_hash_delete(sc->sampler_hash);
@@ -358,6 +377,7 @@ void cso_cache_delete(struct cso_cache *sc)
    cso_hash_delete(sc->rasterizer_hash);
    cso_hash_delete(sc->fs_hash);
    cso_hash_delete(sc->vs_hash);
+   cso_hash_delete(sc->velements_hash);
    FREE(sc);
 }
 
@@ -372,6 +392,7 @@ void cso_set_maximum_cache_size(struct cso_cache *sc, int number)
    sanitize_hash(sc, sc->vs_hash, CSO_VERTEX_SHADER, sc->max_size);
    sanitize_hash(sc, sc->rasterizer_hash, CSO_RASTERIZER, sc->max_size);
    sanitize_hash(sc, sc->sampler_hash, CSO_SAMPLER, sc->max_size);
+   sanitize_hash(sc, sc->velements_hash, CSO_VELEMENTS, sc->max_size);
 }
 
 int cso_maximum_cache_size(const struct cso_cache *sc)
diff --git a/src/gallium/auxiliary/cso_cache/cso_cache.h b/src/gallium/auxiliary/cso_cache/cso_cache.h
index eea60b940bb..fb09b83c623 100644
--- a/src/gallium/auxiliary/cso_cache/cso_cache.h
+++ b/src/gallium/auxiliary/cso_cache/cso_cache.h
@@ -53,6 +53,7 @@
   * - rasterizer (old setup)
   * - sampler
   * - vertex shader
+  * - vertex elements
   *
   * Things that are not constant state objects include:
   * - blend_color
@@ -90,7 +91,8 @@ enum cso_cache_type {
    CSO_DEPTH_STENCIL_ALPHA,
    CSO_RASTERIZER,
    CSO_FRAGMENT_SHADER,
-   CSO_VERTEX_SHADER
+   CSO_VERTEX_SHADER,
+   CSO_VELEMENTS
 };
 
 typedef void (*cso_state_callback)(void *ctx, void *obj);
@@ -144,6 +146,18 @@ struct cso_sampler {
    struct pipe_context *context;
 };
 
+struct cso_velems_state {
+   unsigned count;
+   struct pipe_vertex_element velems[PIPE_MAX_ATTRIBS];
+};
+
+struct cso_velements {
+   struct cso_velems_state state;
+   void *data;
+   cso_state_callback delete_state;
+   struct pipe_context *context;
+};
+
 unsigned cso_construct_key(void *item, int item_size);
 
 struct cso_cache *cso_cache_create(void);
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
index a7335c340ca..292e489312b 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.c
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -89,6 +89,7 @@ struct cso_context {
    void *rasterizer, *rasterizer_saved;
    void *fragment_shader, *fragment_shader_saved, *geometry_shader;
    void *vertex_shader, *vertex_shader_saved, *geometry_shader_saved;
+   void *velements, *velements_saved;
 
    struct pipe_clip_state clip;
    struct pipe_clip_state clip_saved;
@@ -174,6 +175,20 @@ static boolean delete_vs_state(struct cso_context *ctx, void *state)
    return FALSE;
 }
 
+static boolean delete_vertex_elements(struct cso_context *ctx,
+                                      void *state)
+{
+   struct cso_velements *cso = (struct cso_velements *)state;
+
+   if (ctx->velements == cso->data)
+      return FALSE;
+
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+   return TRUE;
+}
+
 
 static INLINE boolean delete_cso(struct cso_context *ctx,
                                  void *state, enum cso_cache_type type)
@@ -197,6 +212,9 @@ static INLINE boolean delete_cso(struct cso_context *ctx,
    case CSO_VERTEX_SHADER:
       return delete_vs_state(ctx, state);
       break;
+   case CSO_VELEMENTS:
+      return delete_vertex_elements(ctx, state);
+      break;
    default:
       assert(0);
       FREE(state);
@@ -1130,7 +1148,6 @@ void cso_restore_geometry_shader(struct cso_context *ctx)
    ctx->geometry_shader_saved = NULL;
 }
 
-
 /* clip state */
 
 static INLINE void
@@ -1180,3 +1197,66 @@ cso_restore_clip(struct cso_context *ctx)
       ctx->pipe->set_clip_state(ctx->pipe, &ctx->clip_saved);
    }
 }
+
+enum pipe_error cso_set_vertex_elements(struct cso_context *ctx,
+                                        unsigned count,
+                                        const struct pipe_vertex_element *states)
+{
+   unsigned key_size, hash_key;
+   struct cso_hash_iter iter;
+   void *handle;
+   struct cso_velems_state velems_state;
+
+   /* need to include the count into the stored state data too.
+      Otherwise first few count pipe_vertex_elements could be identical even if count
+      is different, and there's no guarantee the hash would be different in that
+      case neither */
+   key_size = sizeof(struct pipe_vertex_element) * count + sizeof(unsigned);
+   velems_state.count = count;
+   memcpy(velems_state.velems, states, sizeof(struct pipe_vertex_element) * count);
+   hash_key = cso_construct_key((void*)&velems_state, key_size);
+   iter = cso_find_state_template(ctx->cache, hash_key, CSO_VELEMENTS, (void*)&velems_state, key_size);
+
+   if (cso_hash_iter_is_null(iter)) {
+      struct cso_velements *cso = MALLOC(sizeof(struct cso_velements));
+      if (!cso)
+         return PIPE_ERROR_OUT_OF_MEMORY;
+
+      memcpy(&cso->state, &velems_state, key_size);
+      cso->data = ctx->pipe->create_vertex_elements_state(ctx->pipe, count, &cso->state.velems[0]);
+      cso->delete_state = (cso_state_callback)ctx->pipe->delete_vertex_elements_state;
+      cso->context = ctx->pipe;
+
+      iter = cso_insert_state(ctx->cache, hash_key, CSO_VELEMENTS, cso);
+      if (cso_hash_iter_is_null(iter)) {
+         FREE(cso);
+         return PIPE_ERROR_OUT_OF_MEMORY;
+      }
+
+      handle = cso->data;
+   }
+   else {
+      handle = ((struct cso_velements *)cso_hash_iter_data(iter))->data;
+   }
+
+   if (ctx->velements != handle) {
+      ctx->velements = handle;
+      ctx->pipe->bind_vertex_elements_state(ctx->pipe, handle);
+   }
+   return PIPE_OK;
+}
+
+void cso_save_vertex_elements(struct cso_context *ctx)
+{
+   assert(!ctx->velements);
+   ctx->velements_saved = ctx->velements;
+}
+
+void cso_restore_vertex_elements(struct cso_context *ctx)
+{
+   if (ctx->velements != ctx->velements_saved) {
+      ctx->velements = ctx->velements_saved;
+      ctx->pipe->bind_vertex_elements_state(ctx->pipe, ctx->velements_saved);
+   }
+   ctx->velements_saved = NULL;
+}
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.h b/src/gallium/auxiliary/cso_cache/cso_context.h
index 251a9a644f8..9c16abd28dd 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.h
+++ b/src/gallium/auxiliary/cso_cache/cso_context.h
@@ -122,6 +122,12 @@ void
 cso_restore_vertex_sampler_textures(struct cso_context *cso);
 
 
+enum pipe_error cso_set_vertex_elements(struct cso_context *ctx,
+                                        unsigned count,
+                                        const struct pipe_vertex_element *states);
+void cso_save_vertex_elements(struct cso_context *ctx);
+void cso_restore_vertex_elements(struct cso_context *ctx);
+
 
 /* These aren't really sensible -- most of the time the api provides
  * object semantics for shaders anyway, and the cases where it doesn't
@@ -157,7 +163,6 @@ void cso_save_geometry_shader(struct cso_context *cso);
 void cso_restore_geometry_shader(struct cso_context *cso);
 
 
-
 enum pipe_error cso_set_framebuffer(struct cso_context *cso,
                                     const struct pipe_framebuffer_state *fb);
 void cso_save_framebuffer(struct cso_context *cso);
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index 6d90a6c42fd..a8cdc57ad96 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -307,9 +307,8 @@ draw_arrays_instanced(struct draw_context *draw,
       tgsi_dump(draw->vs.vertex_shader->state.tokens, 0);
       debug_printf("Elements:\n");
       for (i = 0; i < draw->pt.nr_vertex_elements; i++) {
-         debug_printf("  format=%s comps=%u\n",
-                      util_format_name(draw->pt.vertex_element[i].src_format),
-                      draw->pt.vertex_element[i].nr_components);
+         debug_printf("  format=%s\n",
+                      util_format_name(draw->pt.vertex_element[i].src_format));
       }
       debug_printf("Buffers:\n");
       for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
diff --git a/src/gallium/auxiliary/util/u_blit.c b/src/gallium/auxiliary/util/u_blit.c
index 0b263a9db5c..4d0737ccd3d 100644
--- a/src/gallium/auxiliary/util/u_blit.c
+++ b/src/gallium/auxiliary/util/u_blit.c
@@ -63,6 +63,7 @@ struct blit_state
    struct pipe_sampler_state sampler;
    struct pipe_viewport_state viewport;
    struct pipe_clip_state clip;
+   struct pipe_vertex_element velem[2];
 
    void *vs;
    void *fs[TGSI_WRITEMASK_XYZW + 1];
@@ -114,6 +115,15 @@ util_create_blit(struct pipe_context *pipe, struct cso_context *cso)
    ctx->sampler.mag_img_filter = 0; /* set later */
    ctx->sampler.normalized_coords = 1;
 
+   /* vertex elements state */
+   memset(&ctx->velem[0], 0, sizeof(ctx->velem[0]) * 2);
+   for (i = 0; i < 2; i++) {
+      ctx->velem[i].src_offset = i * 4 * sizeof(float);
+      ctx->velem[i].instance_divisor = 0;
+      ctx->velem[i].vertex_buffer_index = 0;
+      ctx->velem[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+   }
+
    /* vertex shader - still required to provide the linkage between
     * fragment shader input semantics and vertex_element/buffers.
     */
@@ -410,12 +420,14 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    cso_save_fragment_shader(ctx->cso);
    cso_save_vertex_shader(ctx->cso);
    cso_save_clip(ctx->cso);
+   cso_save_vertex_elements(ctx->cso);
 
    /* set misc state we care about */
    cso_set_blend(ctx->cso, &ctx->blend);
    cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
    cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
    cso_set_clip(ctx->cso, &ctx->clip);
+   cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
 
    /* sampler */
    ctx->sampler.min_img_filter = filter;
@@ -480,6 +492,7 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    cso_restore_fragment_shader(ctx->cso);
    cso_restore_vertex_shader(ctx->cso);
    cso_restore_clip(ctx->cso);
+   cso_restore_vertex_elements(ctx->cso);
 
    pipe_texture_reference(&tex, NULL);
 }
@@ -564,12 +577,14 @@ util_blit_pixels_tex(struct blit_state *ctx,
    cso_save_fragment_shader(ctx->cso);
    cso_save_vertex_shader(ctx->cso);
    cso_save_clip(ctx->cso);
+   cso_save_vertex_elements(ctx->cso);
 
    /* set misc state we care about */
    cso_set_blend(ctx->cso, &ctx->blend);
    cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
    cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
    cso_set_clip(ctx->cso, &ctx->clip);
+   cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
 
    /* sampler */
    ctx->sampler.min_img_filter = filter;
@@ -628,4 +643,5 @@ util_blit_pixels_tex(struct blit_state *ctx,
    cso_restore_fragment_shader(ctx->cso);
    cso_restore_vertex_shader(ctx->cso);
    cso_restore_clip(ctx->cso);
+   cso_restore_vertex_elements(ctx->cso);
 }
diff --git a/src/gallium/auxiliary/util/u_blitter.c b/src/gallium/auxiliary/util/u_blitter.c
index 0ba09d33bfc..33d09085f0b 100644
--- a/src/gallium/auxiliary/util/u_blitter.c
+++ b/src/gallium/auxiliary/util/u_blitter.c
@@ -88,6 +88,8 @@ struct blitter_context_priv
    void *dsa_write_depth_keep_stencil;
    void *dsa_keep_depth_stencil;
 
+   void *velem_state;
+
    /* Sampler state for clamping to a miplevel. */
    void *sampler_state[PIPE_MAX_TEXTURE_LEVELS];
 
@@ -108,6 +110,7 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
    struct pipe_depth_stencil_alpha_state dsa = { { 0 } };
    struct pipe_rasterizer_state rs_state = { 0 };
    struct pipe_sampler_state *sampler_state;
+   struct pipe_vertex_element velem[2];
    unsigned i;
 
    ctx = CALLOC_STRUCT(blitter_context_priv);
@@ -122,6 +125,7 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
    ctx->blitter.saved_rs_state = INVALID_PTR;
    ctx->blitter.saved_fs = INVALID_PTR;
    ctx->blitter.saved_vs = INVALID_PTR;
+   ctx->blitter.saved_velem_state = INVALID_PTR;
    ctx->blitter.saved_fb_state.nr_cbufs = ~0;
    ctx->blitter.saved_num_textures = ~0;
    ctx->blitter.saved_num_sampler_states = ~0;
@@ -170,6 +174,16 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
    rs_state.flatshade = 1;
    ctx->rs_state = pipe->create_rasterizer_state(pipe, &rs_state);
 
+   /* vertex elements state */
+   memset(&velem[0], 0, sizeof(velem[0]) * 2);
+   for (i = 0; i < 2; i++) {
+      velem[i].src_offset = i * 4 * sizeof(float);
+      velem[i].instance_divisor = 0;
+      velem[i].vertex_buffer_index = 0;
+      velem[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+   }
+   ctx->velem_state = pipe->create_vertex_elements_state(pipe, 2, &velem[0]);
+
    /* fragment shaders are created on-demand */
 
    /* vertex shaders */
@@ -219,6 +233,7 @@ void util_blitter_destroy(struct blitter_context *blitter)
    pipe->delete_rasterizer_state(pipe, ctx->rs_state);
    pipe->delete_vs_state(pipe, ctx->vs_col);
    pipe->delete_vs_state(pipe, ctx->vs_tex);
+   pipe->delete_vertex_elements_state(pipe, ctx->velem_state);
 
    for (i = 0; i < PIPE_MAX_TEXTURE_TYPES; i++) {
       if (ctx->fs_texfetch_col[i])
@@ -246,7 +261,8 @@ static void blitter_check_saved_CSOs(struct blitter_context_priv *ctx)
           ctx->blitter.saved_dsa_state != INVALID_PTR &&
           ctx->blitter.saved_rs_state != INVALID_PTR &&
           ctx->blitter.saved_fs != INVALID_PTR &&
-          ctx->blitter.saved_vs != INVALID_PTR);
+          ctx->blitter.saved_vs != INVALID_PTR &&
+          ctx->blitter.saved_velem_state != INVALID_PTR);
 }
 
 static void blitter_restore_CSOs(struct blitter_context_priv *ctx)
@@ -259,12 +275,14 @@ static void blitter_restore_CSOs(struct blitter_context_priv *ctx)
    pipe->bind_rasterizer_state(pipe, ctx->blitter.saved_rs_state);
    pipe->bind_fs_state(pipe, ctx->blitter.saved_fs);
    pipe->bind_vs_state(pipe, ctx->blitter.saved_vs);
+   pipe->bind_vertex_elements_state(pipe, ctx->blitter.saved_velem_state);
 
    ctx->blitter.saved_blend_state = INVALID_PTR;
    ctx->blitter.saved_dsa_state = INVALID_PTR;
    ctx->blitter.saved_rs_state = INVALID_PTR;
    ctx->blitter.saved_fs = INVALID_PTR;
    ctx->blitter.saved_vs = INVALID_PTR;
+   ctx->blitter.saved_velem_state = INVALID_PTR;
 
    pipe->set_stencil_ref(pipe, &ctx->blitter.saved_stencil_ref);
 
@@ -569,6 +587,7 @@ void util_blitter_clear(struct blitter_context *blitter,
       pipe->bind_depth_stencil_alpha_state(pipe, ctx->dsa_keep_depth_stencil);
 
    pipe->bind_rasterizer_state(pipe, ctx->rs_state);
+   pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
    pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, num_cbufs));
    pipe->bind_vs_state(pipe, ctx->vs_col);
 
@@ -634,6 +653,7 @@ static void util_blitter_do_copy(struct blitter_context *blitter,
    pipe->bind_vs_state(pipe, ctx->vs_tex);
    pipe->bind_fragment_sampler_states(pipe, 1,
       blitter_get_sampler_state(ctx, src->level));
+   pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
    pipe->set_fragment_sampler_textures(pipe, 1, &src->texture);
    pipe->set_framebuffer_state(pipe, &fb_state);
 
@@ -807,6 +827,7 @@ void util_blitter_fill(struct blitter_context *blitter,
    pipe->bind_rasterizer_state(pipe, ctx->rs_state);
    pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, 1));
    pipe->bind_vs_state(pipe, ctx->vs_col);
+   pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
 
    /* set a framebuffer state */
    fb_state.width = dst->width;
diff --git a/src/gallium/auxiliary/util/u_blitter.h b/src/gallium/auxiliary/util/u_blitter.h
index 92008fce992..ecafdabafae 100644
--- a/src/gallium/auxiliary/util/u_blitter.h
+++ b/src/gallium/auxiliary/util/u_blitter.h
@@ -43,6 +43,7 @@ struct blitter_context
    /* Private members, really. */
    void *saved_blend_state;   /**< blend state */
    void *saved_dsa_state;     /**< depth stencil alpha state */
+   void *saved_velem_state;   /**< vertex elements state */
    void *saved_rs_state;      /**< rasterizer state */
    void *saved_fs, *saved_vs; /**< fragment shader, vertex shader */
 
@@ -173,6 +174,13 @@ void util_blitter_save_depth_stencil_alpha(struct blitter_context *blitter,
 }
 
 static INLINE
+void util_blitter_save_vertex_elements(struct blitter_context *blitter,
+                                       void *state)
+{
+   blitter->saved_velem_state = state;
+}
+
+static INLINE
 void util_blitter_save_stencil_ref(struct blitter_context *blitter,
                                    const struct pipe_stencil_ref *state)
 {
diff --git a/src/gallium/auxiliary/util/u_draw_quad.c b/src/gallium/auxiliary/util/u_draw_quad.c
index 14506e84519..8c194102bfc 100644
--- a/src/gallium/auxiliary/util/u_draw_quad.c
+++ b/src/gallium/auxiliary/util/u_draw_quad.c
@@ -45,8 +45,6 @@ util_draw_vertex_buffer(struct pipe_context *pipe,
                         uint num_attribs)
 {
    struct pipe_vertex_buffer vbuffer;
-   struct pipe_vertex_element velements[PIPE_MAX_ATTRIBS];
-   uint i;
 
    assert(num_attribs <= PIPE_MAX_ATTRIBS);
 
@@ -58,15 +56,7 @@ util_draw_vertex_buffer(struct pipe_context *pipe,
    vbuffer.max_index = num_verts - 1;
    pipe->set_vertex_buffers(pipe, 1, &vbuffer);
 
-   /* tell pipe about the vertex attributes */
-   for (i = 0; i < num_attribs; i++) {
-      velements[i].src_offset = i * 4 * sizeof(float);
-      velements[i].instance_divisor = 0;
-      velements[i].vertex_buffer_index = 0;
-      velements[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
-      velements[i].nr_components = 4;
-   }
-   pipe->set_vertex_elements(pipe, num_attribs, velements);
+   /* note: vertex elements already set by caller */
 
    /* draw */
    pipe->draw_arrays(pipe, prim_type, 0, num_verts);
diff --git a/src/gallium/auxiliary/util/u_dump_state.c b/src/gallium/auxiliary/util/u_dump_state.c
index ae7afd7311e..52cf3ef4ce0 100644
--- a/src/gallium/auxiliary/util/u_dump_state.c
+++ b/src/gallium/auxiliary/util/u_dump_state.c
@@ -700,7 +700,6 @@ util_dump_vertex_element(struct os_stream *stream, const struct pipe_vertex_elem
    util_dump_member(stream, uint, state, src_offset);
 
    util_dump_member(stream, uint, state, vertex_buffer_index);
-   util_dump_member(stream, uint, state, nr_components);
 
    util_dump_member(stream, format, state, src_format);
 
diff --git a/src/gallium/auxiliary/util/u_format.h b/src/gallium/auxiliary/util/u_format.h
index b2aa5bfb188..20fdaca007f 100644
--- a/src/gallium/auxiliary/util/u_format.h
+++ b/src/gallium/auxiliary/util/u_format.h
@@ -415,6 +415,23 @@ util_format_has_alpha(enum pipe_format format)
    }
 }
 
+/**
+ * Return the number of components stored.
+ * Formats with block size != 1x1 will always have 1 component (the block).
+ */
+static INLINE unsigned
+util_format_get_nr_components(enum pipe_format format)
+{
+   const struct util_format_description *desc = util_format_description(format);
+   unsigned chan;
+   unsigned nr_comp = 0;
+   for (chan = 0; chan < 4; chan++) {
+      if (desc->channel[chan].size != 0) {
+         nr_comp++;
+      }
+   }
+   return nr_comp;
+}
 
 /*
  * Format access functions.
diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.c b/src/gallium/auxiliary/util/u_gen_mipmap.c
index fc027e48e4e..d421bee8efe 100644
--- a/src/gallium/auxiliary/util/u_gen_mipmap.c
+++ b/src/gallium/auxiliary/util/u_gen_mipmap.c
@@ -62,6 +62,7 @@ struct gen_mipmap_state
    struct pipe_rasterizer_state rasterizer;
    struct pipe_sampler_state sampler;
    struct pipe_clip_state clip;
+   struct pipe_vertex_element velem[2];
 
    void *vs;
    void *fs2d, *fsCube;
@@ -1307,6 +1308,15 @@ util_create_gen_mipmap(struct pipe_context *pipe,
    ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST;
    ctx->sampler.normalized_coords = 1;
 
+   /* vertex elements state */
+   memset(&ctx->velem[0], 0, sizeof(ctx->velem[0]) * 2);
+   for (i = 0; i < 2; i++) {
+      ctx->velem[i].src_offset = i * 4 * sizeof(float);
+      ctx->velem[i].instance_divisor = 0;
+      ctx->velem[i].vertex_buffer_index = 0;
+      ctx->velem[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+   }
+
    /* vertex shader - still needed to specify mapping from fragment
     * shader input semantics to vertex elements 
     */
@@ -1501,12 +1511,14 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
    cso_save_vertex_shader(ctx->cso);
    cso_save_viewport(ctx->cso);
    cso_save_clip(ctx->cso);
+   cso_save_vertex_elements(ctx->cso);
 
    /* bind our state */
    cso_set_blend(ctx->cso, &ctx->blend);
    cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
    cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
    cso_set_clip(ctx->cso, &ctx->clip);
+   cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
 
    cso_set_fragment_shader_handle(ctx->cso, fs);
    cso_set_vertex_shader_handle(ctx->cso, ctx->vs);
@@ -1593,4 +1605,5 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
    cso_restore_vertex_shader(ctx->cso);
    cso_restore_viewport(ctx->cso);
    cso_restore_clip(ctx->cso);
+   cso_restore_vertex_elements(ctx->cso);
 }
diff --git a/src/gallium/auxiliary/vl/vl_compositor.c b/src/gallium/auxiliary/vl/vl_compositor.c
index ba23435f698..6d461cb8800 100644
--- a/src/gallium/auxiliary/vl/vl_compositor.c
+++ b/src/gallium/auxiliary/vl/vl_compositor.c
@@ -230,6 +230,7 @@ static bool
 init_pipe_state(struct vl_compositor *c)
 {
    struct pipe_sampler_state sampler;
+   struct pipe_vertex_element vertex_elems[2];
 
    assert(c);
 
@@ -251,15 +252,27 @@ init_pipe_state(struct vl_compositor *c)
    /*sampler.border_color[i] = ;*/
    /*sampler.max_anisotropy = ;*/
    c->sampler = c->pipe->create_sampler_state(c->pipe, &sampler);
-	
+
+   vertex_elems[0].src_offset = 0;
+   vertex_elems[0].instance_divisor = 0;
+   vertex_elems[0].vertex_buffer_index = 0;
+   vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
+   vertex_elems[1].src_offset = 0;
+   vertex_elems[1].instance_divisor = 0;
+   vertex_elems[1].vertex_buffer_index = 1;
+   vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
+   c->vertex_elems = c->pipe->create_vertex_elements_state(c->pipe, 2, vertex_elems);
+
+
    return true;
 }
 
 static void cleanup_pipe_state(struct vl_compositor *c)
 {
    assert(c);
-	
+
    c->pipe->delete_sampler_state(c->pipe, c->sampler);
+   c->pipe->delete_vertex_elements_state(c->pipe, c->vertex_elems);
 }
 
 static bool
@@ -314,12 +327,6 @@ init_buffers(struct vl_compositor *c)
 
    pipe_buffer_unmap(c->pipe->screen, c->vertex_bufs[0].buffer);
 
-   c->vertex_elems[0].src_offset = 0;
-   c->vertex_elems[0].instance_divisor = 0;
-   c->vertex_elems[0].vertex_buffer_index = 0;
-   c->vertex_elems[0].nr_components = 2;
-   c->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
    /*
     * Create our texcoord buffer and texcoord buffer element
     * Texcoord buffer contains the TCs for mapping the rendered surface to the 4 vertices
@@ -344,12 +351,6 @@ init_buffers(struct vl_compositor *c)
 
    pipe_buffer_unmap(c->pipe->screen, c->vertex_bufs[1].buffer);
 
-   c->vertex_elems[1].src_offset = 0;
-   c->vertex_elems[1].instance_divisor = 0;
-   c->vertex_elems[1].vertex_buffer_index = 1;
-   c->vertex_elems[1].nr_components = 2;
-   c->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
    /*
     * Create our vertex shader's constant buffer
     * Const buffer contains scaling and translation vectors
@@ -483,7 +484,7 @@ void vl_compositor_render(struct vl_compositor          *compositor,
    compositor->pipe->bind_vs_state(compositor->pipe, compositor->vertex_shader);
    compositor->pipe->bind_fs_state(compositor->pipe, compositor->fragment_shader);
    compositor->pipe->set_vertex_buffers(compositor->pipe, 2, compositor->vertex_bufs);
-   compositor->pipe->set_vertex_elements(compositor->pipe, 2, compositor->vertex_elems);
+   compositor->pipe->bind_vertex_elements_state(compositor->pipe, compositor->vertex_elems);
    compositor->pipe->set_constant_buffer(compositor->pipe, PIPE_SHADER_VERTEX, 0, compositor->vs_const_buf);
    compositor->pipe->set_constant_buffer(compositor->pipe, PIPE_SHADER_FRAGMENT, 0, compositor->fs_const_buf);
 
diff --git a/src/gallium/auxiliary/vl/vl_compositor.h b/src/gallium/auxiliary/vl/vl_compositor.h
index 6a9a3fd7af1..51755554da1 100644
--- a/src/gallium/auxiliary/vl/vl_compositor.h
+++ b/src/gallium/auxiliary/vl/vl_compositor.h
@@ -43,10 +43,10 @@ struct vl_compositor
    void *sampler;
    void *vertex_shader;
    void *fragment_shader;
+   void *vertex_elems;
    struct pipe_viewport_state viewport;
    struct pipe_scissor_state scissor;
    struct pipe_vertex_buffer vertex_bufs[2];
-   struct pipe_vertex_element vertex_elems[2];
    struct pipe_buffer *vs_const_buf, *fs_const_buf;
 };
 
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
index f323de0ea55..0763b5bb0e4 100644
--- a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
@@ -708,6 +708,7 @@ static bool
 init_pipe_state(struct vl_mpeg12_mc_renderer *r)
 {
    struct pipe_sampler_state sampler;
+   struct pipe_vertex_element vertex_elems[8];
    unsigned filters[5];
    unsigned i;
 
@@ -771,6 +772,59 @@ init_pipe_state(struct vl_mpeg12_mc_renderer *r)
       r->samplers.all[i] = r->pipe->create_sampler_state(r->pipe, &sampler);
    }
 
+   /* Position element */
+   vertex_elems[0].src_offset = 0;
+   vertex_elems[0].instance_divisor = 0;
+   vertex_elems[0].vertex_buffer_index = 0;
+   vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Luma, texcoord element */
+   vertex_elems[1].src_offset = sizeof(struct vertex2f);
+   vertex_elems[1].instance_divisor = 0;
+   vertex_elems[1].vertex_buffer_index = 0;
+   vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Chroma Cr texcoord element */
+   vertex_elems[2].src_offset = sizeof(struct vertex2f) * 2;
+   vertex_elems[2].instance_divisor = 0;
+   vertex_elems[2].vertex_buffer_index = 0;
+   vertex_elems[2].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Chroma Cb texcoord element */
+   vertex_elems[3].src_offset = sizeof(struct vertex2f) * 3;
+   vertex_elems[3].instance_divisor = 0;
+   vertex_elems[3].vertex_buffer_index = 0;
+   vertex_elems[3].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* First ref surface top field texcoord element */
+   vertex_elems[4].src_offset = 0;
+   vertex_elems[4].instance_divisor = 0;
+   vertex_elems[4].vertex_buffer_index = 1;
+   vertex_elems[4].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* First ref surface bottom field texcoord element */
+   vertex_elems[5].src_offset = sizeof(struct vertex2f);
+   vertex_elems[5].instance_divisor = 0;
+   vertex_elems[5].vertex_buffer_index = 1;
+   vertex_elems[5].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Second ref surface top field texcoord element */
+   vertex_elems[6].src_offset = 0;
+   vertex_elems[6].instance_divisor = 0;
+   vertex_elems[6].vertex_buffer_index = 2;
+   vertex_elems[6].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Second ref surface bottom field texcoord element */
+   vertex_elems[7].src_offset = sizeof(struct vertex2f);
+   vertex_elems[7].instance_divisor = 0;
+   vertex_elems[7].vertex_buffer_index = 2;
+   vertex_elems[7].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* need versions with 4,6 and 8 vertex elems */
+   r->vertex_elems[0] = r->pipe->create_vertex_elements_state(r->pipe, 4, vertex_elems);
+   r->vertex_elems[1] = r->pipe->create_vertex_elements_state(r->pipe, 6, vertex_elems);
+   r->vertex_elems[2] = r->pipe->create_vertex_elements_state(r->pipe, 8, vertex_elems);
+
    return true;
 }
 
@@ -783,6 +837,8 @@ cleanup_pipe_state(struct vl_mpeg12_mc_renderer *r)
 
    for (i = 0; i < 5; ++i)
       r->pipe->delete_sampler_state(r->pipe, r->samplers.all[i]);
+   for (i = 0; i < 3; i++)
+      r->pipe->delete_vertex_elements_state(r->pipe, r->vertex_elems[i]);
 }
 
 static bool
@@ -888,62 +944,6 @@ init_buffers(struct vl_mpeg12_mc_renderer *r)
       );
    }
 
-   /* Position element */
-   r->vertex_elems[0].src_offset = 0;
-   r->vertex_elems[0].instance_divisor = 0;
-   r->vertex_elems[0].vertex_buffer_index = 0;
-   r->vertex_elems[0].nr_components = 2;
-   r->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* Luma, texcoord element */
-   r->vertex_elems[1].src_offset = sizeof(struct vertex2f);
-   r->vertex_elems[1].instance_divisor = 0;
-   r->vertex_elems[1].vertex_buffer_index = 0;
-   r->vertex_elems[1].nr_components = 2;
-   r->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* Chroma Cr texcoord element */
-   r->vertex_elems[2].src_offset = sizeof(struct vertex2f) * 2;
-   r->vertex_elems[2].instance_divisor = 0;
-   r->vertex_elems[2].vertex_buffer_index = 0;
-   r->vertex_elems[2].nr_components = 2;
-   r->vertex_elems[2].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* Chroma Cb texcoord element */
-   r->vertex_elems[3].src_offset = sizeof(struct vertex2f) * 3;
-   r->vertex_elems[3].instance_divisor = 0;
-   r->vertex_elems[3].vertex_buffer_index = 0;
-   r->vertex_elems[3].nr_components = 2;
-   r->vertex_elems[3].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* First ref surface top field texcoord element */
-   r->vertex_elems[4].src_offset = 0;
-   r->vertex_elems[4].instance_divisor = 0;
-   r->vertex_elems[4].vertex_buffer_index = 1;
-   r->vertex_elems[4].nr_components = 2;
-   r->vertex_elems[4].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* First ref surface bottom field texcoord element */
-   r->vertex_elems[5].src_offset = sizeof(struct vertex2f);
-   r->vertex_elems[5].instance_divisor = 0;
-   r->vertex_elems[5].vertex_buffer_index = 1;
-   r->vertex_elems[5].nr_components = 2;
-   r->vertex_elems[5].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* Second ref surface top field texcoord element */
-   r->vertex_elems[6].src_offset = 0;
-   r->vertex_elems[6].instance_divisor = 0;
-   r->vertex_elems[6].vertex_buffer_index = 2;
-   r->vertex_elems[6].nr_components = 2;
-   r->vertex_elems[6].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* Second ref surface bottom field texcoord element */
-   r->vertex_elems[7].src_offset = sizeof(struct vertex2f);
-   r->vertex_elems[7].instance_divisor = 0;
-   r->vertex_elems[7].vertex_buffer_index = 2;
-   r->vertex_elems[7].nr_components = 2;
-   r->vertex_elems[7].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
    r->vs_const_buf = pipe_buffer_create
    (
       r->pipe->screen,
@@ -1307,7 +1307,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (num_macroblocks[MACROBLOCK_TYPE_INTRA] > 0) {
       r->pipe->set_vertex_buffers(r->pipe, 1, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 4, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[0]);
       r->pipe->set_fragment_sampler_textures(r->pipe, 3, r->textures.all);
       r->pipe->bind_fragment_sampler_states(r->pipe, 3, r->samplers.all);
       r->pipe->bind_vs_state(r->pipe, r->i_vs);
@@ -1320,7 +1320,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] > 0) {
       r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[1]);
       r->textures.individual.ref[0] = r->past;
       r->pipe->set_fragment_sampler_textures(r->pipe, 4, r->textures.all);
       r->pipe->bind_fragment_sampler_states(r->pipe, 4, r->samplers.all);
@@ -1334,7 +1334,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (false /*num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] > 0 */ ) {
       r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[1]);
       r->textures.individual.ref[0] = r->past;
       r->pipe->set_fragment_sampler_textures(r->pipe, 4, r->textures.all);
       r->pipe->bind_fragment_sampler_states(r->pipe, 4, r->samplers.all);
@@ -1348,7 +1348,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] > 0) {
       r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[1]);
       r->textures.individual.ref[0] = r->future;
       r->pipe->set_fragment_sampler_textures(r->pipe, 4, r->textures.all);
       r->pipe->bind_fragment_sampler_states(r->pipe, 4, r->samplers.all);
@@ -1362,7 +1362,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (false /*num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] > 0 */ ) {
       r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[1]);
       r->textures.individual.ref[0] = r->future;
       r->pipe->set_fragment_sampler_textures(r->pipe, 4, r->textures.all);
       r->pipe->bind_fragment_sampler_states(r->pipe, 4, r->samplers.all);
@@ -1376,7 +1376,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] > 0) {
       r->pipe->set_vertex_buffers(r->pipe, 3, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 8, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[2]);
       r->textures.individual.ref[0] = r->past;
       r->textures.individual.ref[1] = r->future;
       r->pipe->set_fragment_sampler_textures(r->pipe, 5, r->textures.all);
@@ -1391,7 +1391,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (false /*num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] > 0 */ ) {
       r->pipe->set_vertex_buffers(r->pipe, 3, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 8, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[2]);
       r->textures.individual.ref[0] = r->past;
       r->textures.individual.ref[1] = r->future;
       r->pipe->set_fragment_sampler_textures(r->pipe, 5, r->textures.all);
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
index f00b8c7b8b1..a11a3e7307b 100644
--- a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
@@ -66,8 +66,8 @@ struct vl_mpeg12_mc_renderer
    struct pipe_buffer *vs_const_buf;
    struct pipe_buffer *fs_const_buf;
    struct pipe_framebuffer_state fb_state;
-   struct pipe_vertex_element vertex_elems[8];
-	
+   void *vertex_elems[3];
+
    union
    {
       void *all[5];