Merge remote branch 'origin/master' into gallium_draw_llvm

9 files changed, 152 insertions, 190 deletions

diff --git a/src/gallium/auxiliary/draw/draw_pipe_aaline.c b/src/gallium/auxiliary/draw/draw_pipe_aaline.c
index 1c07ab13654..f4615064e65 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_aaline.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_aaline.c
@@ -40,6 +40,7 @@
 #include "util/u_format.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
+#include "util/u_sampler.h"
 
 #include "tgsi/tgsi_transform.h"
 #include "tgsi/tgsi_dump.h"
@@ -88,8 +89,9 @@ struct aaline_stage
 
    void *sampler_cso;
    struct pipe_texture *texture;
+   struct pipe_sampler_view *sampler_view;
    uint num_samplers;
-   uint num_textures;
+   uint num_sampler_views;
 
 
    /*
@@ -98,7 +100,7 @@ struct aaline_stage
    struct aaline_fragment_shader *fs;
    struct {
       void *sampler[PIPE_MAX_SAMPLERS];
-      struct pipe_texture *texture[PIPE_MAX_SAMPLERS];
+      struct pipe_sampler_view *sampler_views[PIPE_MAX_SAMPLERS];
    } state;
 
    /*
@@ -111,8 +113,9 @@ struct aaline_stage
 
    void (*driver_bind_sampler_states)(struct pipe_context *, unsigned,
                                       void **);
-   void (*driver_set_sampler_textures)(struct pipe_context *, unsigned,
-                                       struct pipe_texture **);
+   void (*driver_set_sampler_views)(struct pipe_context *,
+                                    unsigned,
+                                    struct pipe_sampler_view **);
 
    struct pipe_context *pipe;
 };
@@ -394,6 +397,7 @@ aaline_create_texture(struct aaline_stage *aaline)
    struct pipe_context *pipe = aaline->pipe;
    struct pipe_screen *screen = pipe->screen;
    struct pipe_texture texTemp;
+   struct pipe_sampler_view viewTempl;
    uint level;
 
    memset(&texTemp, 0, sizeof(texTemp));
@@ -408,6 +412,16 @@ aaline_create_texture(struct aaline_stage *aaline)
    if (!aaline->texture)
       return FALSE;
 
+   u_sampler_view_default_template(&viewTempl,
+                                   aaline->texture,
+                                   aaline->texture->format);
+   aaline->sampler_view = pipe->create_sampler_view(pipe,
+                                                    aaline->texture,
+                                                    &viewTempl);
+   if (!aaline->sampler_view) {
+      return FALSE;
+   }
+
    /* Fill in mipmap images.
     * Basically each level is solid opaque, except for the outermost
     * texels which are zero.  Special case the 1x1 and 2x2 levels.
@@ -669,16 +683,16 @@ aaline_first_line(struct draw_stage *stage, struct prim_header *header)
 
    /* how many samplers? */
    /* we'll use sampler/texture[pstip->sampler_unit] for the stipple */
-   num_samplers = MAX2(aaline->num_textures, aaline->num_samplers);
+   num_samplers = MAX2(aaline->num_sampler_views, aaline->num_samplers);
    num_samplers = MAX2(num_samplers, aaline->fs->sampler_unit + 1);
 
    aaline->state.sampler[aaline->fs->sampler_unit] = aaline->sampler_cso;
-   pipe_texture_reference(&aaline->state.texture[aaline->fs->sampler_unit],
-                          aaline->texture);
+   pipe_sampler_view_reference(&aaline->state.sampler_views[aaline->fs->sampler_unit],
+                               aaline->sampler_view);
 
    draw->suspend_flushing = TRUE;
    aaline->driver_bind_sampler_states(pipe, num_samplers, aaline->state.sampler);
-   aaline->driver_set_sampler_textures(pipe, num_samplers, aaline->state.texture);
+   aaline->driver_set_sampler_views(pipe, num_samplers, aaline->state.sampler_views);
    draw->suspend_flushing = FALSE;
 
    /* now really draw first line */
@@ -702,8 +716,9 @@ aaline_flush(struct draw_stage *stage, unsigned flags)
    aaline->driver_bind_fs_state(pipe, aaline->fs->driver_fs);
    aaline->driver_bind_sampler_states(pipe, aaline->num_samplers,
                                       aaline->state.sampler);
-   aaline->driver_set_sampler_textures(pipe, aaline->num_textures,
-                                       aaline->state.texture);
+   aaline->driver_set_sampler_views(pipe,
+                                    aaline->num_sampler_views,
+                                    aaline->state.sampler_views);
    draw->suspend_flushing = FALSE;
 
    draw->extra_shader_outputs.slot = 0;
@@ -724,7 +739,7 @@ aaline_destroy(struct draw_stage *stage)
    uint i;
 
    for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
-      pipe_texture_reference(&aaline->state.texture[i], NULL);
+      pipe_sampler_view_reference(&aaline->state.sampler_views[i], NULL);
    }
 
    if (aaline->sampler_cso)
@@ -733,6 +748,10 @@ aaline_destroy(struct draw_stage *stage)
    if (aaline->texture)
       pipe_texture_reference(&aaline->texture, NULL);
 
+   if (aaline->sampler_view) {
+      pipe_sampler_view_reference(&aaline->sampler_view, NULL);
+   }
+
    draw_free_temp_verts( stage );
 
    FREE( stage );
@@ -844,23 +863,24 @@ aaline_bind_sampler_states(struct pipe_context *pipe,
 
 
 static void
-aaline_set_sampler_textures(struct pipe_context *pipe,
-                            unsigned num, struct pipe_texture **texture)
+aaline_set_sampler_views(struct pipe_context *pipe,
+                         unsigned num,
+                         struct pipe_sampler_view **views)
 {
    struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
    uint i;
 
    /* save current */
    for (i = 0; i < num; i++) {
-      pipe_texture_reference(&aaline->state.texture[i], texture[i]);
+      pipe_sampler_view_reference(&aaline->state.sampler_views[i], views[i]);
    }
    for ( ; i < PIPE_MAX_SAMPLERS; i++) {
-      pipe_texture_reference(&aaline->state.texture[i], NULL);
+      pipe_sampler_view_reference(&aaline->state.sampler_views[i], NULL);
    }
-   aaline->num_textures = num;
+   aaline->num_sampler_views = num;
 
    /* pass-through */
-   aaline->driver_set_sampler_textures(aaline->pipe, num, texture);
+   aaline->driver_set_sampler_views(aaline->pipe, num, views);
 }
 
 
@@ -898,7 +918,7 @@ draw_install_aaline_stage(struct draw_context *draw, struct pipe_context *pipe)
    aaline->driver_delete_fs_state = pipe->delete_fs_state;
 
    aaline->driver_bind_sampler_states = pipe->bind_fragment_sampler_states;
-   aaline->driver_set_sampler_textures = pipe->set_fragment_sampler_textures;
+   aaline->driver_set_sampler_views = pipe->set_fragment_sampler_views;
 
    /* override the driver's functions */
    pipe->create_fs_state = aaline_create_fs_state;
@@ -906,7 +926,7 @@ draw_install_aaline_stage(struct draw_context *draw, struct pipe_context *pipe)
    pipe->delete_fs_state = aaline_delete_fs_state;
 
    pipe->bind_fragment_sampler_states = aaline_bind_sampler_states;
-   pipe->set_fragment_sampler_textures = aaline_set_sampler_textures;
+   pipe->set_fragment_sampler_views = aaline_set_sampler_views;
    
    /* Install once everything is known to be OK:
     */
diff --git a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c
index 38c22bf4e94..794fd81d70f 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c
@@ -42,6 +42,7 @@
 #include "util/u_format.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
+#include "util/u_sampler.h"
 
 #include "tgsi/tgsi_transform.h"
 #include "tgsi/tgsi_dump.h"
@@ -75,8 +76,9 @@ struct pstip_stage
 
    void *sampler_cso;
    struct pipe_texture *texture;
+   struct pipe_sampler_view *sampler_view;
    uint num_samplers;
-   uint num_textures;
+   uint num_sampler_views;
 
    /*
     * Currently bound state
@@ -84,7 +86,7 @@ struct pstip_stage
    struct pstip_fragment_shader *fs;
    struct {
       void *samplers[PIPE_MAX_SAMPLERS];
-      struct pipe_texture *textures[PIPE_MAX_SAMPLERS];
+      struct pipe_sampler_view *sampler_views[PIPE_MAX_SAMPLERS];
       const struct pipe_poly_stipple *stipple;
    } state;
 
@@ -98,8 +100,9 @@ struct pstip_stage
 
    void (*driver_bind_sampler_states)(struct pipe_context *, unsigned, void **);
 
-   void (*driver_set_sampler_textures)(struct pipe_context *, unsigned,
-                                       struct pipe_texture **);
+   void (*driver_set_sampler_views)(struct pipe_context *,
+                                    unsigned,
+                                    struct pipe_sampler_view **);
 
    void (*driver_set_polygon_stipple)(struct pipe_context *,
                                       const struct pipe_poly_stipple *);
@@ -424,6 +427,7 @@ pstip_create_texture(struct pstip_stage *pstip)
    struct pipe_context *pipe = pstip->pipe;
    struct pipe_screen *screen = pipe->screen;
    struct pipe_texture texTemp;
+   struct pipe_sampler_view viewTempl;
 
    memset(&texTemp, 0, sizeof(texTemp));
    texTemp.target = PIPE_TEXTURE_2D;
@@ -437,6 +441,16 @@ pstip_create_texture(struct pstip_stage *pstip)
    if (pstip->texture == NULL)
       return FALSE;
 
+   u_sampler_view_default_template(&viewTempl,
+                                   pstip->texture,
+                                   pstip->texture->format);
+   pstip->sampler_view = pipe->create_sampler_view(pipe,
+                                                   pstip->texture,
+                                                   &viewTempl);
+   if (!pstip->sampler_view) {
+      return FALSE;
+   }
+
    return TRUE;
 }
 
@@ -515,19 +529,19 @@ pstip_first_tri(struct draw_stage *stage, struct prim_header *header)
 
    /* how many samplers? */
    /* we'll use sampler/texture[pstip->sampler_unit] for the stipple */
-   num_samplers = MAX2(pstip->num_textures, pstip->num_samplers);
+   num_samplers = MAX2(pstip->num_sampler_views, pstip->num_samplers);
    num_samplers = MAX2(num_samplers, pstip->fs->sampler_unit + 1);
 
    /* plug in our sampler, texture */
    pstip->state.samplers[pstip->fs->sampler_unit] = pstip->sampler_cso;
-   pipe_texture_reference(&pstip->state.textures[pstip->fs->sampler_unit],
-                          pstip->texture);
+   pipe_sampler_view_reference(&pstip->state.sampler_views[pstip->fs->sampler_unit],
+                               pstip->sampler_view);
 
    assert(num_samplers <= PIPE_MAX_SAMPLERS);
 
    draw->suspend_flushing = TRUE;
    pstip->driver_bind_sampler_states(pipe, num_samplers, pstip->state.samplers);
-   pstip->driver_set_sampler_textures(pipe, num_samplers, pstip->state.textures);
+   pstip->driver_set_sampler_views(pipe, num_samplers, pstip->state.sampler_views);
    draw->suspend_flushing = FALSE;
 
    /* now really draw first triangle */
@@ -551,8 +565,9 @@ pstip_flush(struct draw_stage *stage, unsigned flags)
    pstip->driver_bind_fs_state(pipe, pstip->fs->driver_fs);
    pstip->driver_bind_sampler_states(pipe, pstip->num_samplers,
                                      pstip->state.samplers);
-   pstip->driver_set_sampler_textures(pipe, pstip->num_textures,
-                                      pstip->state.textures);
+   pstip->driver_set_sampler_views(pipe,
+                                   pstip->num_sampler_views,
+                                   pstip->state.sampler_views);
    draw->suspend_flushing = FALSE;
 }
 
@@ -571,13 +586,17 @@ pstip_destroy(struct draw_stage *stage)
    uint i;
 
    for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
-      pipe_texture_reference(&pstip->state.textures[i], NULL);
+      pipe_sampler_view_reference(&pstip->state.sampler_views[i], NULL);
    }
 
    pstip->pipe->delete_sampler_state(pstip->pipe, pstip->sampler_cso);
 
    pipe_texture_reference(&pstip->texture, NULL);
 
+   if (pstip->sampler_view) {
+      pipe_sampler_view_reference(&pstip->sampler_view, NULL);
+   }
+
    draw_free_temp_verts( stage );
    FREE( stage );
 }
@@ -682,24 +701,25 @@ pstip_bind_sampler_states(struct pipe_context *pipe,
 
 
 static void
-pstip_set_sampler_textures(struct pipe_context *pipe,
-                           unsigned num, struct pipe_texture **texture)
+pstip_set_sampler_views(struct pipe_context *pipe,
+                        unsigned num,
+                        struct pipe_sampler_view **views)
 {
    struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
    uint i;
 
    /* save current */
    for (i = 0; i < num; i++) {
-      pipe_texture_reference(&pstip->state.textures[i], texture[i]);
+      pipe_sampler_view_reference(&pstip->state.sampler_views[i], views[i]);
    }
    for (; i < PIPE_MAX_SAMPLERS; i++) {
-      pipe_texture_reference(&pstip->state.textures[i], NULL);
+      pipe_sampler_view_reference(&pstip->state.sampler_views[i], NULL);
    }
 
-   pstip->num_textures = num;
+   pstip->num_sampler_views = num;
 
    /* pass-through */
-   pstip->driver_set_sampler_textures(pstip->pipe, num, texture);
+   pstip->driver_set_sampler_views(pstip->pipe, num, views);
 }
 
 
@@ -756,7 +776,7 @@ draw_install_pstipple_stage(struct draw_context *draw,
    pstip->driver_delete_fs_state = pipe->delete_fs_state;
 
    pstip->driver_bind_sampler_states = pipe->bind_fragment_sampler_states;
-   pstip->driver_set_sampler_textures = pipe->set_fragment_sampler_textures;
+   pstip->driver_set_sampler_views = pipe->set_fragment_sampler_views;
    pstip->driver_set_polygon_stipple = pipe->set_polygon_stipple;
 
    /* override the driver's functions */
@@ -765,7 +785,7 @@ draw_install_pstipple_stage(struct draw_context *draw,
    pipe->delete_fs_state = pstip_delete_fs_state;
 
    pipe->bind_fragment_sampler_states = pstip_bind_sampler_states;
-   pipe->set_fragment_sampler_textures = pstip_set_sampler_textures;
+   pipe->set_fragment_sampler_views = pstip_set_sampler_views;
    pipe->set_polygon_stipple = pstip_set_polygon_stipple;
 
    return TRUE;
diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
index 27099579618..1c7db0005a9 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
@@ -238,38 +238,15 @@ vbuf_start_prim( struct vbuf_stage *vbuf, uint prim )
       unsigned output_format;
       unsigned src_offset = (vbuf->vinfo->attrib[i].src_index * 4 * sizeof(float) );
 
-      switch (vbuf->vinfo->attrib[i].emit) {
-      case EMIT_4F:
-	 output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
-	 emit_sz = 4 * sizeof(float);
-	 break;
-      case EMIT_3F:
-	 output_format = PIPE_FORMAT_R32G32B32_FLOAT;
-	 emit_sz = 3 * sizeof(float);
-	 break;
-      case EMIT_2F:
-	 output_format = PIPE_FORMAT_R32G32_FLOAT;
-	 emit_sz = 2 * sizeof(float);
-	 break;
-      case EMIT_1F:
-	 output_format = PIPE_FORMAT_R32_FLOAT;
-	 emit_sz = 1 * sizeof(float);
-	 break;
-      case EMIT_1F_PSIZE:
-	 output_format = PIPE_FORMAT_R32_FLOAT;
-	 emit_sz = 1 * sizeof(float);
+      output_format = draw_translate_vinfo_format(vbuf->vinfo->attrib[i].emit);
+      emit_sz = draw_translate_vinfo_size(vbuf->vinfo->attrib[i].emit);
+
+      /* doesn't handle EMIT_OMIT */
+      assert(emit_sz != 0);
+
+      if (vbuf->vinfo->attrib[i].emit == EMIT_1F_PSIZE) {
 	 src_buffer = 1;
 	 src_offset = 0;
-	 break;
-      case EMIT_4UB:
-	 output_format = PIPE_FORMAT_A8R8G8B8_UNORM;
-	 emit_sz = 4 * sizeof(ubyte);
-         break;
-      default:
-	 assert(0);
-	 output_format = PIPE_FORMAT_NONE;
-	 emit_sz = 0;
-	 break;
       }
 
       hw_key.element[i].type = TRANSLATE_ELEMENT_NORMAL;
diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c
index ae357b51226..a7917f54b04 100644
--- a/src/gallium/auxiliary/draw/draw_pt_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_emit.c
@@ -86,40 +86,15 @@ void draw_pt_emit_prepare( struct pt_emit *emit,
       unsigned output_format;
       unsigned src_offset = (vinfo->attrib[i].src_index * 4 * sizeof(float) );
 
+      output_format = draw_translate_vinfo_format(vinfo->attrib[i].emit);
+      emit_sz = draw_translate_vinfo_size(vinfo->attrib[i].emit);
 
-         
-      switch (vinfo->attrib[i].emit) {
-      case EMIT_4F:
-	 output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
-	 emit_sz = 4 * sizeof(float);
-	 break;
-      case EMIT_3F:
-	 output_format = PIPE_FORMAT_R32G32B32_FLOAT;
-	 emit_sz = 3 * sizeof(float);
-	 break;
-      case EMIT_2F:
-	 output_format = PIPE_FORMAT_R32G32_FLOAT;
-	 emit_sz = 2 * sizeof(float);
-	 break;
-      case EMIT_1F:
-	 output_format = PIPE_FORMAT_R32_FLOAT;
-	 emit_sz = 1 * sizeof(float);
-	 break;
-      case EMIT_1F_PSIZE:
-	 output_format = PIPE_FORMAT_R32_FLOAT;
-	 emit_sz = 1 * sizeof(float);
+      /* doesn't handle EMIT_OMIT */
+      assert(emit_sz != 0);
+
+      if (vinfo->attrib[i].emit == EMIT_1F_PSIZE) {
 	 src_buffer = 1;
 	 src_offset = 0;
-	 break;
-      case EMIT_4UB:
-	 output_format = PIPE_FORMAT_A8R8G8B8_UNORM;
-	 emit_sz = 4 * sizeof(ubyte);
-         break;
-      default:
-	 assert(0);
-	 output_format = PIPE_FORMAT_NONE;
-	 emit_sz = 0;
-	 break;
       }
 
       hw_key.element[i].type = TRANSLATE_ELEMENT_NORMAL;
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
index 2a604470e9a..1994ddf2bcc 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
@@ -129,41 +129,16 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle,
       unsigned input_offset = src->src_offset;
       unsigned output_format;
 
-      switch (vinfo->attrib[i].emit) {
-      case EMIT_4UB:
-	 output_format = PIPE_FORMAT_R8G8B8A8_UNORM;
-	 emit_sz = 4 * sizeof(unsigned char);
-         break;
-      case EMIT_4F:
-	 output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
-	 emit_sz = 4 * sizeof(float);
-         break;
-      case EMIT_3F:
-	 output_format = PIPE_FORMAT_R32G32B32_FLOAT;
-	 emit_sz = 3 * sizeof(float);
-         break;
-      case EMIT_2F:
-	 output_format = PIPE_FORMAT_R32G32_FLOAT;
-	 emit_sz = 2 * sizeof(float);
-         break;
-      case EMIT_1F:
-	 output_format = PIPE_FORMAT_R32_FLOAT;
-	 emit_sz = 1 * sizeof(float);
-         break;
-      case EMIT_1F_PSIZE:
+      output_format = draw_translate_vinfo_format(vinfo->attrib[i].emit);
+      emit_sz = draw_translate_vinfo_size(vinfo->attrib[i].emit);
+
+      if (vinfo->attrib[i].emit == EMIT_OMIT)
+	 continue;
+
+      if (vinfo->attrib[i].emit == EMIT_1F_PSIZE) {
 	 input_format = PIPE_FORMAT_R32_FLOAT;
 	 input_buffer = draw->pt.nr_vertex_buffers;
 	 input_offset = 0;
-	 output_format = PIPE_FORMAT_R32_FLOAT;
-	 emit_sz = 1 * sizeof(float);
-         break;
-      case EMIT_OMIT:
-         continue;
-      default:
-         assert(0);
-	 output_format = PIPE_FORMAT_NONE;
-	 emit_sz = 0;
-	 continue;
       }
 
       key.element[i].type = TRANSLATE_ELEMENT_NORMAL;
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
index 1aecb510777..389e2b105e5 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
@@ -130,31 +130,10 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
       unsigned dst_offset = 0;
 
       for (i = 0; i < vinfo->num_attribs; i++) {
-         unsigned emit_sz = 0;
-
-         switch (vinfo->attrib[i].emit) {
-         case EMIT_4F:
-            emit_sz = 4 * sizeof(float);
-            break;
-         case EMIT_3F:
-            emit_sz = 3 * sizeof(float);
-            break;
-         case EMIT_2F:
-            emit_sz = 2 * sizeof(float);
-            break;
-         case EMIT_1F:
-            emit_sz = 1 * sizeof(float);
-            break;
-         case EMIT_1F_PSIZE:
-            emit_sz = 1 * sizeof(float);
-            break;
-         case EMIT_4UB:
-            emit_sz = 4 * sizeof(ubyte);
-            break;
-         default:
-            assert(0);
-            break;
-         }
+         unsigned emit_sz = draw_translate_vinfo_size(vinfo->attrib[i].emit);
+
+         /* doesn't handle EMIT_OMIT */
+         assert(emit_sz != 0);
 
          /* The elements in the key correspond to vertex shader output
           * numbers, not to positions in the hw vertex description --
diff --git a/src/gallium/auxiliary/draw/draw_vertex.c b/src/gallium/auxiliary/draw/draw_vertex.c
index 3214213e445..a4f5e882c0a 100644
--- a/src/gallium/auxiliary/draw/draw_vertex.c
+++ b/src/gallium/auxiliary/draw/draw_vertex.c
@@ -48,30 +48,12 @@ draw_compute_vertex_size(struct vertex_info *vinfo)
    uint i;
 
    vinfo->size = 0;
-   for (i = 0; i < vinfo->num_attribs; i++) {
-      switch (vinfo->attrib[i].emit) {
-      case EMIT_OMIT:
-         break;
-      case EMIT_4UB:
-         /* fall-through */
-      case EMIT_1F_PSIZE:
-         /* fall-through */
-      case EMIT_1F:
-         vinfo->size += 1;
-         break;
-      case EMIT_2F:
-         vinfo->size += 2;
-         break;
-      case EMIT_3F:
-         vinfo->size += 3;
-         break;
-      case EMIT_4F:
-         vinfo->size += 4;
-         break;
-      default:
-         assert(0);
-      }
-   }
+   for (i = 0; i < vinfo->num_attribs; i++)
+      vinfo->size += draw_translate_vinfo_size(vinfo->attrib[i].emit);
+
+   assert(vinfo->size % 4 == 0);
+   /* in dwords */
+   vinfo->size /= 4;
 }
 
 
@@ -120,6 +102,13 @@ draw_dump_emitted_vertex(const struct vertex_info *vinfo, const uint8_t *data)
          debug_printf("%u ", *data++);
          debug_printf("%u ", *data++);
          break;
+      case EMIT_4UB_BGRA:
+         debug_printf("EMIT_4UB_BGRA:\t");
+         debug_printf("%u ", *data++);
+         debug_printf("%u ", *data++);
+         debug_printf("%u ", *data++);
+         debug_printf("%u ", *data++);
+         break;
       default:
          assert(0);
       }
diff --git a/src/gallium/auxiliary/draw/draw_vertex.h b/src/gallium/auxiliary/draw/draw_vertex.h
index 8c3c7befbc7..ca272371267 100644
--- a/src/gallium/auxiliary/draw/draw_vertex.h
+++ b/src/gallium/auxiliary/draw/draw_vertex.h
@@ -54,7 +54,8 @@ enum attrib_emit {
    EMIT_2F,
    EMIT_3F,
    EMIT_4F,
-   EMIT_4UB  /**< XXX may need variations for RGBA vs BGRA, etc */
+   EMIT_4UB, /**< is RGBA like the rest */
+   EMIT_4UB_BGRA
 };
 
 
@@ -141,9 +142,11 @@ void draw_dump_emitted_vertex(const struct vertex_info *vinfo,
                               const uint8_t *data);
 
 
-static INLINE unsigned draw_translate_vinfo_format(unsigned format )
+static INLINE unsigned draw_translate_vinfo_format(enum attrib_emit emit)
 {
-   switch (format) {
+   switch (emit) {
+   case EMIT_OMIT:
+      return PIPE_FORMAT_NONE;
    case EMIT_1F:
    case EMIT_1F_PSIZE:
       return PIPE_FORMAT_R32_FLOAT;
@@ -155,10 +158,36 @@ static INLINE unsigned draw_translate_vinfo_format(unsigned format )
       return PIPE_FORMAT_R32G32B32A32_FLOAT;
    case EMIT_4UB:
       return PIPE_FORMAT_R8G8B8A8_UNORM;
+   case EMIT_4UB_BGRA:
+      return PIPE_FORMAT_B8G8R8A8_UNORM;
    default:
+      assert(!"unexpected format");
       return PIPE_FORMAT_NONE;
    }
 }
 
+static INLINE unsigned draw_translate_vinfo_size(enum attrib_emit emit)
+{
+   switch (emit) {
+   case EMIT_OMIT:
+      return 0;
+   case EMIT_1F:
+   case EMIT_1F_PSIZE:
+      return 1 * sizeof(float);
+   case EMIT_2F:
+      return 2 * sizeof(float);
+   case EMIT_3F:
+      return 3 * sizeof(float);
+   case EMIT_4F:
+      return 4 * sizeof(float);
+   case EMIT_4UB:
+      return 4 * sizeof(unsigned char);
+   case EMIT_4UB_BGRA:
+      return 4 * sizeof(unsigned char);
+   default:
+      assert(!"unexpected format");
+      return 0;
+   }
+}
 
 #endif /* DRAW_VERTEX_H */
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
index ece1ddde0cb..8f8bbe7cb88 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
@@ -401,13 +401,11 @@ static boolean emit_output( struct aos_compilation *cp,
       emit_store_R32G32B32A32(cp, ptr, dataXMM);
       break;
    case EMIT_4UB:
-      if (1) {
-         emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W));
-         emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
-      }
-      else {
-         emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
-      }
+      emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
+      break;
+   case EMIT_4UB_BGRA:
+      emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W));
+      emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
       break;
    default:
       AOS_ERROR(cp, "unhandled output format");