gallium: add start_slot parameter to set_vertex_buffers

This allows updating only a subrange of buffer bindings.

set_vertex_buffers(pipe, start_slot, count, NULL) unbinds buffers in that
range. Binding NULL resources unbinds buffers too (both buffer and user_buffer
must be NULL).

The meta ops are adapted to only save, change, and restore the single slot
they use. The cso_context can save and restore only one vertex buffer slot.
The clients can query which one it is using cso_get_aux_vertex_buffer_slot.
It's currently set to 0. (the Draw module breaks if it's set to non-zero)

It should decrease the CPU overhead when using a lot of meta ops, but
the drivers must be able to treat each vertex buffer slot as a separate
state (only r600g does so at the moment).

I can imagine this also being useful for optimizing some OpenGL use cases.

Reviewed-by: Brian Paul <[email protected]>

30 files changed, 158 insertions, 183 deletions

diff --git a/src/gallium/drivers/galahad/glhd_context.c b/src/gallium/drivers/galahad/glhd_context.c
index 3400a4d213a..ecdfe068f79 100644
--- a/src/gallium/drivers/galahad/glhd_context.c
+++ b/src/gallium/drivers/galahad/glhd_context.c
@@ -621,7 +621,7 @@ galahad_context_set_fragment_sampler_views(struct pipe_context *_pipe,
 
 static void
 galahad_context_set_vertex_buffers(struct pipe_context *_pipe,
-                            unsigned num_buffers,
+                            unsigned start_slot, unsigned num_buffers,
                             const struct pipe_vertex_buffer *_buffers)
 {
    struct galahad_context *glhd_pipe = galahad_context(_pipe);
@@ -638,7 +638,7 @@ galahad_context_set_vertex_buffers(struct pipe_context *_pipe,
    }
 
    pipe->set_vertex_buffers(pipe,
-                            num_buffers,
+                            start_slot, num_buffers,
                             buffers);
 }
 
diff --git a/src/gallium/drivers/i915/i915_state.c b/src/gallium/drivers/i915/i915_state.c
index d96ac7ab938..7fa648fc4d9 100644
--- a/src/gallium/drivers/i915/i915_state.c
+++ b/src/gallium/drivers/i915/i915_state.c
@@ -30,6 +30,7 @@
 
 
 #include "draw/draw_context.h"
+#include "util/u_helpers.h"
 #include "util/u_inlines.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
@@ -926,19 +927,18 @@ static void i915_delete_rasterizer_state(struct pipe_context *pipe,
 }
 
 static void i915_set_vertex_buffers(struct pipe_context *pipe,
-                                    unsigned count,
+                                    unsigned start_slot, unsigned count,
                                     const struct pipe_vertex_buffer *buffers)
 {
    struct i915_context *i915 = i915_context(pipe);
    struct draw_context *draw = i915->draw;
 
-   util_copy_vertex_buffers(i915->vertex_buffers,
-                            &i915->nr_vertex_buffers,
-                            buffers, count);
+   util_set_vertex_buffers_count(i915->vertex_buffers,
+                                 &i915->nr_vertex_buffers,
+                                 buffers, start_slot, count);
 
    /* pass-through to draw module */
-   draw_set_vertex_buffers(draw, count, buffers);
-
+   draw_set_vertex_buffers(draw, start_slot, count, buffers);
 }
 
 static void *
diff --git a/src/gallium/drivers/i915/i915_surface.c b/src/gallium/drivers/i915/i915_surface.c
index 331d6e230d0..08a6f343922 100644
--- a/src/gallium/drivers/i915/i915_surface.c
+++ b/src/gallium/drivers/i915/i915_surface.c
@@ -55,7 +55,7 @@ i915_util_blitter_save_states(struct i915_context *i915)
    util_blitter_save_viewport(i915->blitter, &i915->viewport);
    util_blitter_save_scissor(i915->blitter, &i915->scissor);
    util_blitter_save_vertex_elements(i915->blitter, i915->velems);
-   util_blitter_save_vertex_buffers(i915->blitter, i915->nr_vertex_buffers,
+   util_blitter_save_vertex_buffer_slot(i915->blitter,
                                     i915->vertex_buffers);
 
    util_blitter_save_framebuffer(i915->blitter, &i915->framebuffer);
diff --git a/src/gallium/drivers/identity/id_context.c b/src/gallium/drivers/identity/id_context.c
index fb2f78764a4..0325b151f83 100644
--- a/src/gallium/drivers/identity/id_context.c
+++ b/src/gallium/drivers/identity/id_context.c
@@ -565,7 +565,7 @@ identity_set_vertex_sampler_views(struct pipe_context *_pipe,
 
 static void
 identity_set_vertex_buffers(struct pipe_context *_pipe,
-                            unsigned num_buffers,
+                            unsigned start_slot, unsigned num_buffers,
                             const struct pipe_vertex_buffer *_buffers)
 {
    struct identity_context *id_pipe = identity_context(_pipe);
@@ -582,7 +582,7 @@ identity_set_vertex_buffers(struct pipe_context *_pipe,
    }
 
    pipe->set_vertex_buffers(pipe,
-                            num_buffers,
+                            start_slot, num_buffers,
                             buffers);
 }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_draw_arrays.c b/src/gallium/drivers/llvmpipe/lp_draw_arrays.c
index f37b3ca1cea..3497edf9d1c 100644
--- a/src/gallium/drivers/llvmpipe/lp_draw_arrays.c
+++ b/src/gallium/drivers/llvmpipe/lp_draw_arrays.c
@@ -68,8 +68,12 @@ llvmpipe_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
     */
    for (i = 0; i < lp->num_vertex_buffers; i++) {
       const void *buf = lp->vertex_buffer[i].user_buffer;
-      if (!buf)
+      if (!buf) {
+         if (!lp->vertex_buffer[i].buffer) {
+            continue;
+         }
          buf = llvmpipe_resource_data(lp->vertex_buffer[i].buffer);
+      }
       draw_set_mapped_vertex_buffer(draw, i, buf);
    }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state_vertex.c b/src/gallium/drivers/llvmpipe/lp_state_vertex.c
index 4a74cd7c9d4..7594d028b71 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_vertex.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_vertex.c
@@ -33,6 +33,7 @@
 #include "lp_state.h"
 
 #include "draw/draw_context.h"
+#include "util/u_helpers.h"
 #include "util/u_inlines.h"
 #include "util/u_transfer.h"
 
@@ -75,20 +76,20 @@ llvmpipe_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
 
 static void
 llvmpipe_set_vertex_buffers(struct pipe_context *pipe,
-                            unsigned count,
+                            unsigned start_slot, unsigned count,
                             const struct pipe_vertex_buffer *buffers)
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
 
    assert(count <= PIPE_MAX_ATTRIBS);
 
-   util_copy_vertex_buffers(llvmpipe->vertex_buffer,
-                            &llvmpipe->num_vertex_buffers,
-                            buffers, count);
+   util_set_vertex_buffers_count(llvmpipe->vertex_buffer,
+                                 &llvmpipe->num_vertex_buffers,
+                                 buffers, start_slot, count);
 
    llvmpipe->dirty |= LP_NEW_VERTEX;
 
-   draw_set_vertex_buffers(llvmpipe->draw, count, buffers);
+   draw_set_vertex_buffers(llvmpipe->draw, start_slot, count, buffers);
 }
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_surface.c b/src/gallium/drivers/llvmpipe/lp_surface.c
index a4789b4f5c3..e42f4c838e9 100644
--- a/src/gallium/drivers/llvmpipe/lp_surface.c
+++ b/src/gallium/drivers/llvmpipe/lp_surface.c
@@ -204,8 +204,7 @@ static void lp_blit(struct pipe_context *pipe,
 
    /* XXX turn off occlusion and streamout queries */
 
-   util_blitter_save_vertex_buffers(lp->blitter, lp->num_vertex_buffers,
-                                    lp->vertex_buffer);
+   util_blitter_save_vertex_buffer_slot(lp->blitter, lp->vertex_buffer);
    util_blitter_save_vertex_elements(lp->blitter, (void*)lp->velems);
    util_blitter_save_vertex_shader(lp->blitter, (void*)lp->vs);
    util_blitter_save_geometry_shader(lp->blitter, (void*)lp->gs);
diff --git a/src/gallium/drivers/noop/noop_state.c b/src/gallium/drivers/noop/noop_state.c
index dbfe35b668c..16ce9b457d9 100644
--- a/src/gallium/drivers/noop/noop_state.c
+++ b/src/gallium/drivers/noop/noop_state.c
@@ -215,8 +215,9 @@ static void noop_set_index_buffer(struct pipe_context *ctx,
 {
 }
 
-static void noop_set_vertex_buffers(struct pipe_context *ctx, unsigned count,
-					const struct pipe_vertex_buffer *buffers)
+static void noop_set_vertex_buffers(struct pipe_context *ctx,
+				    unsigned start_slot, unsigned count,
+				    const struct pipe_vertex_buffer *buffers)
 {
 }
 
diff --git a/src/gallium/drivers/nv30/nv30_draw.c b/src/gallium/drivers/nv30/nv30_draw.c
index 46e85362815..51fe4f86ed1 100644
--- a/src/gallium/drivers/nv30/nv30_draw.c
+++ b/src/gallium/drivers/nv30/nv30_draw.c
@@ -379,7 +379,7 @@ nv30_render_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
    if (nv30->draw_dirty & NV30_NEW_CLIP)
       draw_set_clip_state(draw, &nv30->clip);
    if (nv30->draw_dirty & NV30_NEW_ARRAYS) {
-      draw_set_vertex_buffers(draw, nv30->num_vtxbufs, nv30->vtxbuf);
+      draw_set_vertex_buffers(draw, 0, nv30->num_vtxbufs, nv30->vtxbuf);
       draw_set_vertex_elements(draw, nv30->vertex->num_elements, nv30->vertex->pipe);
    }
    if (nv30->draw_dirty & NV30_NEW_FRAGPROG) {
@@ -404,10 +404,14 @@ nv30_render_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
 
    for (i = 0; i < nv30->num_vtxbufs; i++) {
       const void *map = nv30->vtxbuf[i].user_buffer;
-      if (!map)
+      if (!map) {
+         if (!nv30->vtxbuf[i].buffer) {
+            continue;
+         }
          map = pipe_buffer_map(pipe, nv30->vtxbuf[i].buffer,
                                   PIPE_TRANSFER_UNSYNCHRONIZED |
                                   PIPE_TRANSFER_READ, &transfer[i]);
+      }
       draw_set_mapped_vertex_buffer(draw, i, map);
    }
 
diff --git a/src/gallium/drivers/nv30/nv30_miptree.c b/src/gallium/drivers/nv30/nv30_miptree.c
index e89725feb2f..5d941ab771e 100644
--- a/src/gallium/drivers/nv30/nv30_miptree.c
+++ b/src/gallium/drivers/nv30/nv30_miptree.c
@@ -193,9 +193,7 @@ nv30_blit(struct pipe_context *pipe,
 
    /* XXX turn off occlusion queries */
 
-   util_blitter_save_vertex_buffers(nv30->blitter,
-                                    nv30->num_vtxbufs,
-                                    nv30->vtxbuf);
+   util_blitter_save_vertex_buffer_slot(nv30->blitter, nv30->vtxbuf);
    util_blitter_save_vertex_elements(nv30->blitter, nv30->vertex);
    util_blitter_save_vertex_shader(nv30->blitter, nv30->vertprog.program);
    util_blitter_save_rasterizer(nv30->blitter, nv30->rast);
diff --git a/src/gallium/drivers/nv30/nv30_push.c b/src/gallium/drivers/nv30/nv30_push.c
index c5867bfa11a..5ec2d462a68 100644
--- a/src/gallium/drivers/nv30/nv30_push.c
+++ b/src/gallium/drivers/nv30/nv30_push.c
@@ -211,6 +211,10 @@ nv30_push_vbo(struct nv30_context *nv30, const struct pipe_draw_info *info)
       struct pipe_vertex_buffer *vb = &nv30->vtxbuf[i];
       struct nv04_resource *res = nv04_resource(vb->buffer);
 
+      if (!vb->buffer && !vb->user_buffer) {
+         continue;
+      }
+
       data = nouveau_resource_map_offset(&nv30->base, res,
                                          vb->buffer_offset, NOUVEAU_BO_RD);
 
@@ -276,8 +280,11 @@ nv30_push_vbo(struct nv30_context *nv30, const struct pipe_draw_info *info)
    if (info->indexed)
       nouveau_resource_unmap(nv04_resource(nv30->idxbuf.buffer));
 
-   for (i = 0; i < nv30->num_vtxbufs; ++i)
-      nouveau_resource_unmap(nv04_resource(nv30->vtxbuf[i].buffer));
+   for (i = 0; i < nv30->num_vtxbufs; ++i) {
+      if (nv30->vtxbuf[i].buffer) {
+         nouveau_resource_unmap(nv04_resource(nv30->vtxbuf[i].buffer));
+      }
+   }
 
    nv30_state_release(nv30);
 }
diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c
index 046ba723dae..af8bb449fce 100644
--- a/src/gallium/drivers/nv30/nv30_state.c
+++ b/src/gallium/drivers/nv30/nv30_state.c
@@ -23,6 +23,7 @@
  *
  */
 
+#include "util/u_helpers.h"
 #include "util/u_inlines.h"
 
 #include "nouveau/nouveau_gldefs.h"
@@ -393,21 +394,15 @@ nv30_set_viewport_state(struct pipe_context *pipe,
 
 static void
 nv30_set_vertex_buffers(struct pipe_context *pipe,
-                        unsigned count,
+                        unsigned start_slot, unsigned count,
                         const struct pipe_vertex_buffer *vb)
 {
     struct nv30_context *nv30 = nv30_context(pipe);
-    unsigned i;
 
     nouveau_bufctx_reset(nv30->bufctx, BUFCTX_VTXBUF);
 
-    for (i = 0; i < count; ++i)
-       pipe_resource_reference(&nv30->vtxbuf[i].buffer, vb[i].buffer);
-    for (; i < nv30->num_vtxbufs; ++i)
-       pipe_resource_reference(&nv30->vtxbuf[i].buffer, NULL);
-
-    memcpy(nv30->vtxbuf, vb, sizeof(*vb) * count);
-    nv30->num_vtxbufs = count;
+    util_set_vertex_buffers_count(nv30->vtxbuf, &nv30->num_vtxbufs,
+                                  vb, start_slot, count);
 
     nv30->dirty |= NV30_NEW_ARRAYS;
 }
diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c
index 2052f914258..819b0a2e2bf 100644
--- a/src/gallium/drivers/nv50/nv50_state.c
+++ b/src/gallium/drivers/nv50/nv50_state.c
@@ -21,6 +21,7 @@
  */
 
 #include "pipe/p_defines.h"
+#include "util/u_helpers.h"
 #include "util/u_inlines.h"
 #include "util/u_transfer.h"
 
@@ -883,30 +884,35 @@ nv50_set_viewport_state(struct pipe_context *pipe,
 
 static void
 nv50_set_vertex_buffers(struct pipe_context *pipe,
-                        unsigned count,
+                        unsigned start_slot, unsigned count,
                         const struct pipe_vertex_buffer *vb)
 {
    struct nv50_context *nv50 = nv50_context(pipe);
    unsigned i;
 
-   nv50->vbo_user = nv50->vbo_constant = 0;
+   util_set_vertex_buffers_count(nv50->vtxbuf, &nv50->num_vtxbufs, vb,
+                                 start_slot, count);
+
+   if (!vb) {
+      nv50->vbo_user &= ~(((1ull << count) - 1) << start_slot);
+      nv50->vbo_constant &= ~(((1ull << count) - 1) << start_slot);
+      return;
+   }
 
    for (i = 0; i < count; ++i) {
-      nv50->vtxbuf[i].stride = vb[i].stride;
-      pipe_resource_reference(&nv50->vtxbuf[i].buffer, vb[i].buffer);
+      unsigned dst_index = start_slot + i;
+
       if (!vb[i].buffer && vb[i].user_buffer) {
-         nv50->vtxbuf[i].user_buffer = vb[i].user_buffer;
-         nv50->vbo_user |= 1 << i;
+         nv50->vbo_user |= 1 << dst_index;
          if (!vb[i].stride)
-            nv50->vbo_constant |= 1 << i;
+            nv50->vbo_constant |= 1 << dst_index;
+         else
+            nv50->vbo_constant &= ~(1 << dst_index);
       } else {
-         nv50->vtxbuf[i].buffer_offset = vb[i].buffer_offset;
+         nv50->vbo_user &= ~(1 << dst_index);
+         nv50->vbo_constant &= ~(1 << dst_index);
       }
    }
-   for (; i < nv50->num_vtxbufs; ++i)
-      pipe_resource_reference(&nv50->vtxbuf[i].buffer, NULL);
-
-   nv50->num_vtxbufs = count;
 
    nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_VERTEX);
 
diff --git a/src/gallium/drivers/nvc0/nvc0_state.c b/src/gallium/drivers/nvc0/nvc0_state.c
index 1bd54342e81..30011df4dc1 100644
--- a/src/gallium/drivers/nvc0/nvc0_state.c
+++ b/src/gallium/drivers/nvc0/nvc0_state.c
@@ -21,6 +21,7 @@
  */
 
 #include "pipe/p_defines.h"
+#include "util/u_helpers.h"
 #include "util/u_inlines.h"
 #include "util/u_transfer.h"
 
@@ -778,59 +779,38 @@ nvc0_set_viewport_state(struct pipe_context *pipe,
 
 static void
 nvc0_set_vertex_buffers(struct pipe_context *pipe,
-                        unsigned count,
+                        unsigned start_slot, unsigned count,
                         const struct pipe_vertex_buffer *vb)
 {
     struct nvc0_context *nvc0 = nvc0_context(pipe);
-    uint32_t constant_vbos = 0;
     unsigned i;
 
-    nvc0->vbo_user = 0;
-
-    if (count != nvc0->num_vtxbufs) {
-       for (i = 0; i < count; ++i) {
-          pipe_resource_reference(&nvc0->vtxbuf[i].buffer, vb[i].buffer);
-          if (vb[i].user_buffer) {
-             nvc0->vbo_user |= 1 << i;
-             nvc0->vtxbuf[i].user_buffer = vb[i].user_buffer;
-             if (!vb[i].stride)
-                constant_vbos |= 1 << i;
-          } else {
-             nvc0->vtxbuf[i].buffer_offset = vb[i].buffer_offset;
-          }
-          nvc0->vtxbuf[i].stride = vb[i].stride;
-       }
-       for (; i < nvc0->num_vtxbufs; ++i)
-          pipe_resource_reference(&nvc0->vtxbuf[i].buffer, NULL);
+    util_set_vertex_buffers_count(nvc0->vtxbuf, &nvc0->num_vtxbufs, vb,
+                                  start_slot, count);
 
-       nvc0->num_vtxbufs = count;
-       nvc0->dirty |= NVC0_NEW_ARRAYS;
-    } else {
-       for (i = 0; i < count; ++i) {
-          if (vb[i].user_buffer) {
-             nvc0->vtxbuf[i].user_buffer = vb[i].user_buffer;
-             nvc0->vbo_user |= 1 << i;
-             if (!vb[i].stride)
-                constant_vbos |= 1 << i;
-             assert(!vb[i].buffer);
-          }
-          if (nvc0->vtxbuf[i].buffer == vb[i].buffer &&
-              nvc0->vtxbuf[i].buffer_offset == vb[i].buffer_offset &&
-              nvc0->vtxbuf[i].stride == vb[i].stride)
-             continue;
-          pipe_resource_reference(&nvc0->vtxbuf[i].buffer, vb[i].buffer);
-          nvc0->vtxbuf[i].buffer_offset = vb[i].buffer_offset;
-          nvc0->vtxbuf[i].stride = vb[i].stride;
-          nvc0->dirty |= NVC0_NEW_ARRAYS;
-       }
+    if (!vb) {
+       nvc0->vbo_user &= ~(((1ull << count) - 1) << start_slot);
+       nvc0->constant_vbos &= ~(((1ull << count) - 1) << start_slot);
+       return;
     }
-    if (constant_vbos != nvc0->constant_vbos) {
-       nvc0->constant_vbos = constant_vbos;
-       nvc0->dirty |= NVC0_NEW_ARRAYS;
+
+    for (i = 0; i < count; ++i) {
+       unsigned dst_index = start_slot + i;
+
+       if (vb[i].user_buffer) {
+          nvc0->vbo_user |= 1 << dst_index;
+          if (!vb[i].stride)
+             nvc0->constant_vbos |= 1 << dst_index;
+          else
+             nvc0->constant_vbos &= ~(1 << dst_index);
+       } else {
+          nvc0->vbo_user &= ~(1 << dst_index);
+          nvc0->constant_vbos &= ~(1 << dst_index);
+       }
     }
 
-    if (nvc0->dirty & NVC0_NEW_ARRAYS)
-       nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_VTX);
+    nvc0->dirty |= NVC0_NEW_ARRAYS;
+    nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_VTX);
 }
 
 static void
diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c
index 96c96c367a2..651ea8529e8 100644
--- a/src/gallium/drivers/r300/r300_blit.c
+++ b/src/gallium/drivers/r300/r300_blit.c
@@ -66,8 +66,7 @@ static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op o
     util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
     util_blitter_save_viewport(r300->blitter, &r300->viewport);
     util_blitter_save_scissor(r300->blitter, r300->scissor_state.state);
-    util_blitter_save_vertex_buffers(r300->blitter, r300->nr_vertex_buffers,
-                                     r300->vertex_buffer);
+    util_blitter_save_vertex_buffer_slot(r300->blitter, r300->vertex_buffer);
     util_blitter_save_vertex_elements(r300->blitter, r300->velems);
 
     if (op & R300_SAVE_FRAMEBUFFER) {
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index 198ab73d257..46d7ea1748b 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -1191,8 +1191,6 @@ void r300_blitter_draw_rectangle(struct blitter_context *blitter,
     if (r300->skip_rendering)
         return;
 
-    r300->context.set_vertex_buffers(&r300->context, 0, NULL);
-
     if (type == UTIL_BLITTER_ATTRIB_TEXCOORD)
         r300->sprite_coord_enable = 1;
 
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index b43f330af98..85d05bd2ce3 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -25,6 +25,7 @@
 
 #include "util/u_framebuffer.h"
 #include "util/u_half.h"
+#include "util/u_helpers.h"
 #include "util/u_math.h"
 #include "util/u_mm.h"
 #include "util/u_memory.h"
@@ -1626,7 +1627,7 @@ static void r300_set_viewport_state(struct pipe_context* pipe,
 }
 
 static void r300_set_vertex_buffers_hwtcl(struct pipe_context* pipe,
-                                    unsigned count,
+                                    unsigned start_slot, unsigned count,
                                     const struct pipe_vertex_buffer* buffers)
 {
     struct r300_context* r300 = r300_context(pipe);
@@ -1637,31 +1638,31 @@ static void r300_set_vertex_buffers_hwtcl(struct pipe_context* pipe,
         count = 1;
     }
 
-    util_copy_vertex_buffers(r300->vertex_buffer,
-                             &r300->nr_vertex_buffers,
-                             buffers, count);
+    util_set_vertex_buffers_count(r300->vertex_buffer,
+                                  &r300->nr_vertex_buffers,
+                                  buffers, start_slot, count);
 
     r300->vertex_arrays_dirty = TRUE;
 }
 
 static void r300_set_vertex_buffers_swtcl(struct pipe_context* pipe,
-                                    unsigned count,
+                                    unsigned start_slot, unsigned count,
                                     const struct pipe_vertex_buffer* buffers)
 {
     struct r300_context* r300 = r300_context(pipe);
     unsigned i;
 
-    util_copy_vertex_buffers(r300->vertex_buffer,
-                             &r300->nr_vertex_buffers,
-                             buffers, count);
-    draw_set_vertex_buffers(r300->draw, count, buffers);
+    util_set_vertex_buffers_count(r300->vertex_buffer,
+                                  &r300->nr_vertex_buffers,
+                                  buffers, start_slot, count);
+    draw_set_vertex_buffers(r300->draw, start_slot, count, buffers);
 
     for (i = 0; i < count; i++) {
         if (buffers[i].user_buffer) {
-            draw_set_mapped_vertex_buffer(r300->draw, i,
+            draw_set_mapped_vertex_buffer(r300->draw, start_slot + i,
                                           buffers[i].user_buffer);
         } else if (buffers[i].buffer) {
-            draw_set_mapped_vertex_buffer(r300->draw, i,
+            draw_set_mapped_vertex_buffer(r300->draw, start_slot + i,
                 r300_resource(buffers[i].buffer)->malloced_buffer);
         }
     }
diff --git a/src/gallium/drivers/r600/r600_blit.c b/src/gallium/drivers/r600/r600_blit.c
index a19248da3a2..085108d3709 100644
--- a/src/gallium/drivers/r600/r600_blit.c
+++ b/src/gallium/drivers/r600/r600_blit.c
@@ -55,9 +55,7 @@ static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op
 
 	r600_suspend_nontimer_queries(rctx);
 
-	util_blitter_save_vertex_buffers(rctx->blitter,
-					 util_last_bit(rctx->vertex_buffer_state.enabled_mask),
-					 rctx->vertex_buffer_state.vb);
+	util_blitter_save_vertex_buffer_slot(rctx->blitter, rctx->vertex_buffer_state.vb);
 	util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_fetch_shader.cso);
 	util_blitter_save_vertex_shader(rctx->blitter, rctx->vs_shader);
 	util_blitter_save_so_targets(rctx->blitter, rctx->num_so_targets,
diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index a4d3e461ef1..3503c3964a6 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -495,41 +495,42 @@ void r600_vertex_buffers_dirty(struct r600_context *rctx)
 	}
 }
 
-static void r600_set_vertex_buffers(struct pipe_context *ctx, unsigned count,
-			     const struct pipe_vertex_buffer *input)
+static void r600_set_vertex_buffers(struct pipe_context *ctx,
+				    unsigned start_slot, unsigned count,
+				    const struct pipe_vertex_buffer *input)
 {
 	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct r600_vertexbuf_state *state = &rctx->vertex_buffer_state;
-	struct pipe_vertex_buffer *vb = state->vb;
+	struct pipe_vertex_buffer *vb = state->vb + start_slot;
 	unsigned i;
-	/* This sets 1-bit for buffers with index >= count. */
-	uint32_t disable_mask = ~((1ull << count) - 1);
+	uint32_t disable_mask = 0;
 	/* These are the new buffers set by this function. */
 	uint32_t new_buffer_mask = 0;
 
-	/* Set buffers with index >= count to NULL. */
-	uint32_t remaining_buffers_mask =
-		rctx->vertex_buffer_state.enabled_mask & disable_mask;
-
-	while (remaining_buffers_mask) {
-		i = u_bit_scan(&remaining_buffers_mask);
-		pipe_resource_reference(&vb[i].buffer, NULL);
-	}
-
 	/* Set vertex buffers. */
-	for (i = 0; i < count; i++) {
-		if (memcmp(&input[i], &vb[i], sizeof(struct pipe_vertex_buffer))) {
-			if (input[i].buffer) {
-				vb[i].stride = input[i].stride;
-				vb[i].buffer_offset = input[i].buffer_offset;
-				pipe_resource_reference(&vb[i].buffer, input[i].buffer);
-				new_buffer_mask |= 1 << i;
-			} else {
-				pipe_resource_reference(&vb[i].buffer, NULL);
-				disable_mask |= 1 << i;
+	if (input) {
+		for (i = 0; i < count; i++) {
+			if (memcmp(&input[i], &vb[i], sizeof(struct pipe_vertex_buffer))) {
+				if (input[i].buffer) {
+					vb[i].stride = input[i].stride;
+					vb[i].buffer_offset = input[i].buffer_offset;
+					pipe_resource_reference(&vb[i].buffer, input[i].buffer);
+					new_buffer_mask |= 1 << i;
+				} else {
+					pipe_resource_reference(&vb[i].buffer, NULL);
+					disable_mask |= 1 << i;
+				}
 			}
 		}
-        }
+	} else {
+		for (i = 0; i < count; i++) {
+			pipe_resource_reference(&vb[i].buffer, NULL);
+		}
+		disable_mask = ((1ull << count) - 1);
+	}
+
+	disable_mask <<= start_slot;
+	new_buffer_mask <<= start_slot;
 
 	rctx->vertex_buffer_state.enabled_mask &= ~disable_mask;
 	rctx->vertex_buffer_state.dirty_mask &= rctx->vertex_buffer_state.enabled_mask;
@@ -1371,7 +1372,7 @@ void r600_draw_rectangle(struct blitter_context *blitter,
 	}
 
 	/* draw */
-	util_draw_vertex_buffer(&rctx->context, NULL, buf, offset,
+	util_draw_vertex_buffer(&rctx->context, NULL, buf, rctx->blitter->vb_slot, offset,
 				R600_PRIM_RECTANGLE_LIST, 3, 2);
 	pipe_resource_reference(&buf, NULL);
 }
diff --git a/src/gallium/drivers/radeonsi/r600_blit.c b/src/gallium/drivers/radeonsi/r600_blit.c
index cea7e128018..30736fcf8a7 100644
--- a/src/gallium/drivers/radeonsi/r600_blit.c
+++ b/src/gallium/drivers/radeonsi/r600_blit.c
@@ -61,9 +61,7 @@ static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op
 	if (rctx->queued.named.viewport) {
 		util_blitter_save_viewport(rctx->blitter, &rctx->queued.named.viewport->viewport);
 	}
-	util_blitter_save_vertex_buffers(rctx->blitter,
-					 rctx->nr_vertex_buffers,
-					 rctx->vertex_buffer);
+	util_blitter_save_vertex_buffer_slot(rctx->blitter, rctx->vertex_buffer);
 	util_blitter_save_so_targets(rctx->blitter, rctx->num_so_targets,
 				     (struct pipe_stream_output_target**)rctx->so_targets);
 
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index 66f0bd886fe..bd44d732935 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -27,6 +27,7 @@
 #include "util/u_memory.h"
 #include "util/u_framebuffer.h"
 #include "util/u_blitter.h"
+#include "util/u_helpers.h"
 #include "util/u_math.h"
 #include "util/u_pack_color.h"
 #include "tgsi/tgsi_parse.h"
@@ -2488,12 +2489,12 @@ static void si_delete_vertex_element(struct pipe_context *ctx, void *state)
 	FREE(state);
 }
 
-static void si_set_vertex_buffers(struct pipe_context *ctx, unsigned count,
+static void si_set_vertex_buffers(struct pipe_context *ctx, unsigned start_slot, unsigned count,
 				  const struct pipe_vertex_buffer *buffers)
 {
 	struct r600_context *rctx = (struct r600_context *)ctx;
 
-	util_copy_vertex_buffers(rctx->vertex_buffer, &rctx->nr_vertex_buffers, buffers, count);
+	util_set_vertex_buffers_count(rctx->vertex_buffer, &rctx->nr_vertex_buffers, buffers, start_slot, count);
 }
 
 static void si_set_index_buffer(struct pipe_context *ctx,
diff --git a/src/gallium/drivers/rbug/rbug_context.c b/src/gallium/drivers/rbug/rbug_context.c
index 0b59a0121c2..28f300877cf 100644
--- a/src/gallium/drivers/rbug/rbug_context.c
+++ b/src/gallium/drivers/rbug/rbug_context.c
@@ -783,7 +783,7 @@ rbug_set_fragment_sampler_views(struct pipe_context *_pipe,
 
 static void
 rbug_set_vertex_buffers(struct pipe_context *_pipe,
-                        unsigned num_buffers,
+                        unsigned start_slot, unsigned num_buffers,
                         const struct pipe_vertex_buffer *_buffers)
 {
    struct rbug_context *rb_pipe = rbug_context(_pipe);
@@ -801,7 +801,7 @@ rbug_set_vertex_buffers(struct pipe_context *_pipe,
       buffers = unwrapped_buffers;
    }
 
-   pipe->set_vertex_buffers(pipe,
+   pipe->set_vertex_buffers(pipe, start_slot,
                             num_buffers,
                             buffers);
 
diff --git a/src/gallium/drivers/softpipe/sp_draw_arrays.c b/src/gallium/drivers/softpipe/sp_draw_arrays.c
index 59226fb9c15..ba9e7817ceb 100644
--- a/src/gallium/drivers/softpipe/sp_draw_arrays.c
+++ b/src/gallium/drivers/softpipe/sp_draw_arrays.c
@@ -76,8 +76,12 @@ softpipe_draw_vbo(struct pipe_context *pipe,
    /* Map vertex buffers */
    for (i = 0; i < sp->num_vertex_buffers; i++) {
       const void *buf = sp->vertex_buffer[i].user_buffer;
-      if (!buf)
+      if (!buf) {
+         if (!sp->vertex_buffer[i].buffer) {
+            continue;
+         }
          buf = softpipe_resource(sp->vertex_buffer[i].buffer)->data;
+      }
       draw_set_mapped_vertex_buffer(draw, i, buf);
    }
 
diff --git a/src/gallium/drivers/softpipe/sp_state_vertex.c b/src/gallium/drivers/softpipe/sp_state_vertex.c
index 95acecd3be4..2aee700fec2 100644
--- a/src/gallium/drivers/softpipe/sp_state_vertex.c
+++ b/src/gallium/drivers/softpipe/sp_state_vertex.c
@@ -33,6 +33,7 @@
 #include "sp_state.h"
 
 #include "util/u_memory.h"
+#include "util/u_helpers.h"
 #include "util/u_inlines.h"
 #include "util/u_transfer.h"
 #include "draw/draw_context.h"
@@ -79,20 +80,20 @@ softpipe_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
 
 static void
 softpipe_set_vertex_buffers(struct pipe_context *pipe,
-                            unsigned count,
+                            unsigned start_slot, unsigned count,
                             const struct pipe_vertex_buffer *buffers)
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
    assert(count <= PIPE_MAX_ATTRIBS);
 
-   util_copy_vertex_buffers(softpipe->vertex_buffer,
-                            &softpipe->num_vertex_buffers,
-                            buffers, count);
+   util_set_vertex_buffers_count(softpipe->vertex_buffer,
+                                 &softpipe->num_vertex_buffers,
+                                 buffers, start_slot, count);
 
    softpipe->dirty |= SP_NEW_VERTEX;
 
-   draw_set_vertex_buffers(softpipe->draw, count, buffers);
+   draw_set_vertex_buffers(softpipe->draw, start_slot, count, buffers);
 }
 
 
diff --git a/src/gallium/drivers/softpipe/sp_surface.c b/src/gallium/drivers/softpipe/sp_surface.c
index bb143f4cb3a..911c34df1b8 100644
--- a/src/gallium/drivers/softpipe/sp_surface.c
+++ b/src/gallium/drivers/softpipe/sp_surface.c
@@ -56,8 +56,7 @@ static void sp_blit(struct pipe_context *pipe,
 
    /* XXX turn off occlusion and streamout queries */
 
-   util_blitter_save_vertex_buffers(sp->blitter, sp->num_vertex_buffers,
-                                    sp->vertex_buffer);
+   util_blitter_save_vertex_buffer_slot(sp->blitter, sp->vertex_buffer);
    util_blitter_save_vertex_elements(sp->blitter, sp->velems);
    util_blitter_save_vertex_shader(sp->blitter, sp->vs);
    util_blitter_save_geometry_shader(sp->blitter, sp->gs);
diff --git a/src/gallium/drivers/svga/svga_context.h b/src/gallium/drivers/svga/svga_context.h
index f2f26da1cd5..32671ecafa7 100644
--- a/src/gallium/drivers/svga/svga_context.h
+++ b/src/gallium/drivers/svga/svga_context.h
@@ -234,8 +234,6 @@ struct svga_state
       unsigned flag_1d;
       unsigned flag_srgb;
    } tex_flags;
-
-   boolean any_user_vertex_buffers;
 };
 
 struct svga_prescale {
diff --git a/src/gallium/drivers/svga/svga_pipe_blit.c b/src/gallium/drivers/svga/svga_pipe_blit.c
index 34ac995ef60..a44ed12adf7 100644
--- a/src/gallium/drivers/svga/svga_pipe_blit.c
+++ b/src/gallium/drivers/svga/svga_pipe_blit.c
@@ -183,9 +183,7 @@ static void svga_blit(struct pipe_context *pipe,
 
    /* XXX turn off occlusion and streamout queries */
 
-   util_blitter_save_vertex_buffers(svga->blitter,
-                                    svga->curr.num_vertex_buffers,
-                                    svga->curr.vb);
+   util_blitter_save_vertex_buffer_slot(svga->blitter, svga->curr.vb);
    util_blitter_save_vertex_elements(svga->blitter, (void*)svga->curr.velems);
    util_blitter_save_vertex_shader(svga->blitter, svga->curr.vs);
    /*util_blitter_save_geometry_shader(svga->blitter, svga->curr.gs);*/
diff --git a/src/gallium/drivers/svga/svga_pipe_vertex.c b/src/gallium/drivers/svga/svga_pipe_vertex.c
index 58469910732..82bd1b33efe 100644
--- a/src/gallium/drivers/svga/svga_pipe_vertex.c
+++ b/src/gallium/drivers/svga/svga_pipe_vertex.c
@@ -23,6 +23,7 @@
  *
  **********************************************************/
 
+#include "util/u_helpers.h"
 #include "util/u_inlines.h"
 #include "pipe/p_defines.h"
 #include "util/u_math.h"
@@ -36,32 +37,14 @@
 
 
 static void svga_set_vertex_buffers(struct pipe_context *pipe,
-                                    unsigned count,
+                                    unsigned start_slot, unsigned count,
                                     const struct pipe_vertex_buffer *buffers)
 {
    struct svga_context *svga = svga_context(pipe);
-   unsigned i;
-   boolean any_user_buffer = FALSE;
-
-   /* Check for no change */
-   if (count == svga->curr.num_vertex_buffers &&
-       memcmp(svga->curr.vb, buffers, count * sizeof buffers[0]) == 0)
-      return;
-
-   /* Adjust refcounts */
-   for (i = 0; i < count; i++) {
-      pipe_resource_reference(&svga->curr.vb[i].buffer, buffers[i].buffer);
-      if (svga_buffer_is_user_buffer(buffers[i].buffer))
-         any_user_buffer = TRUE;
-   }
-
-   for ( ; i < svga->curr.num_vertex_buffers; i++)
-      pipe_resource_reference(&svga->curr.vb[i].buffer, NULL);
 
-   /* Copy remaining data */
-   memcpy(svga->curr.vb, buffers, count * sizeof buffers[0]);
-   svga->curr.num_vertex_buffers = count;
-   svga->curr.any_user_vertex_buffers = any_user_buffer;
+   util_set_vertex_buffers_count(svga->curr.vb,
+                                 &svga->curr.num_vertex_buffers,
+                                 buffers, start_slot, count);
 
    svga->dirty |= SVGA_NEW_VBUFFER;
 }
diff --git a/src/gallium/drivers/svga/svga_swtnl_state.c b/src/gallium/drivers/svga/svga_swtnl_state.c
index e9239bafef9..dea3a26b6f2 100644
--- a/src/gallium/drivers/svga/svga_swtnl_state.c
+++ b/src/gallium/drivers/svga/svga_swtnl_state.c
@@ -98,7 +98,7 @@ update_swtnl_draw( struct svga_context *svga,
                                 svga->curr.fs->draw_shader);
 
    if (dirty & SVGA_NEW_VBUFFER)
-      draw_set_vertex_buffers(svga->swtnl.draw, 
+      draw_set_vertex_buffers(svga->swtnl.draw, 0,
                               svga->curr.num_vertex_buffers, 
                               svga->curr.vb);
 
diff --git a/src/gallium/drivers/trace/tr_context.c b/src/gallium/drivers/trace/tr_context.c
index a5110bec487..ba1882ea652 100644
--- a/src/gallium/drivers/trace/tr_context.c
+++ b/src/gallium/drivers/trace/tr_context.c
@@ -1060,7 +1060,7 @@ trace_context_set_vertex_sampler_views(struct pipe_context *_pipe,
 
 static INLINE void
 trace_context_set_vertex_buffers(struct pipe_context *_pipe,
-                                 unsigned num_buffers,
+                                 unsigned start_slot, unsigned num_buffers,
                                  const struct pipe_vertex_buffer *buffers)
 {
    struct trace_context *tr_ctx = trace_context(_pipe);
@@ -1070,6 +1070,7 @@ trace_context_set_vertex_buffers(struct pipe_context *_pipe,
    trace_dump_call_begin("pipe_context", "set_vertex_buffers");
 
    trace_dump_arg(ptr, pipe);
+   trace_dump_arg(uint, start_slot);
    trace_dump_arg(uint, num_buffers);
 
    trace_dump_arg_begin("buffers");
@@ -1081,10 +1082,10 @@ trace_context_set_vertex_buffers(struct pipe_context *_pipe,
       memcpy(_buffers, buffers, num_buffers * sizeof(*_buffers));
       for (i = 0; i < num_buffers; i++)
          _buffers[i].buffer = trace_resource_unwrap(tr_ctx, buffers[i].buffer);
-      pipe->set_vertex_buffers(pipe, num_buffers, _buffers);
+      pipe->set_vertex_buffers(pipe, start_slot, num_buffers, _buffers);
       FREE(_buffers);
    } else {
-      pipe->set_vertex_buffers(pipe, num_buffers, NULL);
+      pipe->set_vertex_buffers(pipe, start_slot, num_buffers, NULL);
    }
 
    trace_dump_call_end();