Merge branch 'gallium-userbuf'

Conflicts:
	src/gallium/docs/source/screen.rst
	src/gallium/drivers/nv50/nv50_state.c
	src/gallium/include/pipe/p_defines.h
	src/mesa/state_tracker/st_draw.c

5 files changed, 111 insertions, 257 deletions

diff --git a/src/mesa/state_tracker/st_atom_constbuf.c b/src/mesa/state_tracker/st_atom_constbuf.c
index 05667a74305..fd9228583d2 100644
--- a/src/mesa/state_tracker/st_atom_constbuf.c
+++ b/src/mesa/state_tracker/st_atom_constbuf.c
@@ -38,6 +38,7 @@
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
 #include "util/u_inlines.h"
+#include "util/u_upload_mgr.h"
 
 #include "st_debug.h"
 #include "st_context.h"
@@ -55,15 +56,13 @@ void st_upload_constants( struct st_context *st,
                           struct gl_program_parameter_list *params,
                           unsigned shader_type)
 {
-   struct pipe_context *pipe = st->pipe;
-
    assert(shader_type == PIPE_SHADER_VERTEX ||
           shader_type == PIPE_SHADER_FRAGMENT ||
           shader_type == PIPE_SHADER_GEOMETRY);
 
    /* update constants */
    if (params && params->NumParameters) {
-      struct pipe_resource *cbuf;
+      struct pipe_constant_buffer cb;
       const uint paramBytes = params->NumParameters * sizeof(GLfloat) * 4;
 
       /* Update the constants which come from fixed-function state, such as
@@ -77,20 +76,28 @@ void st_upload_constants( struct st_context *st,
        * avoid gratuitous rendering synchronization.
        * Let's use a user buffer to avoid an unnecessary copy.
        */
-      cbuf = pipe_user_buffer_create(pipe->screen,
-                                     params->ParameterValues,
-                                     paramBytes,
-                                     PIPE_BIND_CONSTANT_BUFFER);
+      if (st->constbuf_uploader) {
+         cb.buffer = NULL;
+         cb.user_buffer = NULL;
+         u_upload_data(st->constbuf_uploader, 0, paramBytes,
+                       params->ParameterValues, &cb.buffer_offset, &cb.buffer);
+         u_upload_unmap(st->constbuf_uploader);
+      } else {
+         cb.buffer = NULL;
+         cb.user_buffer = params->ParameterValues;
+         cb.buffer_offset = 0;
+      }
+      cb.buffer_size = paramBytes;
 
       if (ST_DEBUG & DEBUG_CONSTANTS) {
-	 debug_printf("%s(shader=%d, numParams=%d, stateFlags=0x%x)\n", 
+         debug_printf("%s(shader=%d, numParams=%d, stateFlags=0x%x)\n",
                       __FUNCTION__, shader_type, params->NumParameters,
                       params->StateFlags);
          _mesa_print_parameter_list(params);
       }
 
-      st->pipe->set_constant_buffer(st->pipe, shader_type, 0, cbuf);
-      pipe_resource_reference(&cbuf, NULL);
+      st->pipe->set_constant_buffer(st->pipe, shader_type, 0, &cb);
+      pipe_resource_reference(&cb.buffer, NULL);
 
       st->state.constants[shader_type].ptr = params->ParameterValues;
       st->state.constants[shader_type].size = paramBytes;
diff --git a/src/mesa/state_tracker/st_context.c b/src/mesa/state_tracker/st_context.c
index ce7dbb3f39a..b4497652539 100644
--- a/src/mesa/state_tracker/st_context.c
+++ b/src/mesa/state_tracker/st_context.c
@@ -65,7 +65,6 @@
 #include "util/u_inlines.h"
 #include "util/u_upload_mgr.h"
 #include "cso_cache/cso_context.h"
-#include "util/u_vbuf.h"
 
 
 DEBUG_GET_ONCE_BOOL_OPTION(mesa_mvp_dp4, "MESA_MVP_DP4", FALSE)
@@ -112,32 +111,12 @@ st_get_msaa(void)
 }
 
 
-static void st_init_vbuf(struct st_context *st)
-{
-   struct u_vbuf_caps caps;
-
-   u_vbuf_get_caps(st->pipe->screen, &caps);
-
-   /* Create u_vbuf if there is anything unsupported. */
-   if (!caps.buffer_offset_unaligned ||
-       !caps.buffer_stride_unaligned ||
-       !caps.velem_src_offset_unaligned ||
-       !caps.format_fixed32 ||
-       !caps.format_float16 ||
-       !caps.format_float64 ||
-       !caps.format_norm32 ||
-       !caps.format_scaled32 ||
-       !caps.user_vertex_buffers) {
-      /* XXX user vertex buffers are always uploaded regardless of the CAP. */
-      st->vbuf = u_vbuf_create(st->pipe, &caps);
-      cso_install_vbuf(st->cso_context, st->vbuf);
-   }
-}
 
 
 static struct st_context *
 st_create_context_priv( struct gl_context *ctx, struct pipe_context *pipe )
 {
+   struct pipe_screen *screen = pipe->screen;
    uint i;
    struct st_context *st = ST_CALLOC_STRUCT( st_context );
    
@@ -156,9 +135,22 @@ st_create_context_priv( struct gl_context *ctx, struct pipe_context *pipe )
    st->dirty.st = ~0;
 
    st->uploader = u_upload_create(st->pipe, 65536, 4, PIPE_BIND_VERTEX_BUFFER);
+
+   if (!screen->get_param(screen, PIPE_CAP_USER_INDEX_BUFFERS)) {
+      st->indexbuf_uploader = u_upload_create(st->pipe, 128 * 1024, 4,
+                                              PIPE_BIND_INDEX_BUFFER);
+   }
+
+   if (!screen->get_param(screen, PIPE_CAP_USER_CONSTANT_BUFFERS)) {
+      unsigned alignment =
+         screen->get_param(screen, PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT);
+
+      st->constbuf_uploader = u_upload_create(pipe, 128 * 1024, alignment,
+                                              PIPE_BIND_CONSTANT_BUFFER);
+   }
+
    st->cso_context = cso_create_context(pipe);
 
-   st_init_vbuf(st);
    st_init_atoms( st );
    st_init_bitmap(st);
    st_init_clear(st);
@@ -254,11 +246,6 @@ static void st_destroy_context_priv( struct st_context *st )
    st_destroy_drawpix(st);
    st_destroy_drawtex(st);
 
-   /* Unreference any user vertex buffers. */
-   for (i = 0; i < st->num_user_attribs; i++) {
-      pipe_resource_reference(&st->user_attrib[i].buffer, NULL);
-   }
-
    for (i = 0; i < Elements(st->state.sampler_views); i++) {
       pipe_sampler_view_reference(&st->state.sampler_views[i], NULL);
    }
@@ -269,6 +256,12 @@ static void st_destroy_context_priv( struct st_context *st )
    }
 
    u_upload_destroy(st->uploader);
+   if (st->indexbuf_uploader) {
+      u_upload_destroy(st->indexbuf_uploader);
+   }
+   if (st->constbuf_uploader) {
+      u_upload_destroy(st->constbuf_uploader);
+   }
    free( st );
 }
 
@@ -276,7 +269,6 @@ static void st_destroy_context_priv( struct st_context *st )
 void st_destroy_context( struct st_context *st )
 {
    struct pipe_context *pipe = st->pipe;
-   struct u_vbuf *vbuf = st->vbuf;
    struct cso_context *cso = st->cso_context;
    struct gl_context *ctx = st->ctx;
    GLuint i;
@@ -312,9 +304,6 @@ void st_destroy_context( struct st_context *st )
    st_destroy_context_priv(st);
    st = NULL;
 
-   if (vbuf)
-      u_vbuf_destroy(vbuf);
-
    cso_destroy_context(cso);
 
    pipe->destroy( pipe );
diff --git a/src/mesa/state_tracker/st_context.h b/src/mesa/state_tracker/st_context.h
index 4786ed22fe9..00a405b69f9 100644
--- a/src/mesa/state_tracker/st_context.h
+++ b/src/mesa/state_tracker/st_context.h
@@ -41,7 +41,6 @@ struct gen_mipmap_state;
 struct st_context;
 struct st_fragment_program;
 struct u_upload_mgr;
-struct u_vbuf;
 
 
 #define ST_NEW_MESA                    (1 << 0) /* Mesa state has changed */
@@ -74,8 +73,7 @@ struct st_context
 
    struct pipe_context *pipe;
 
-   struct u_upload_mgr *uploader;
-   struct u_vbuf *vbuf;
+   struct u_upload_mgr *uploader, *indexbuf_uploader, *constbuf_uploader;
 
    struct draw_context *draw;  /**< For selection/feedback/rastpos only */
    struct draw_stage *feedback_stage;  /**< For GL_FEEDBACK rendermode */
@@ -191,18 +189,6 @@ struct st_context
    int force_msaa;
    void *winsys_drawable_handle;
 
-   /* User vertex buffers. */
-   struct {
-      struct pipe_resource *buffer;
-
-      /** Element size */
-      GLuint element_size;
-
-      /** Attribute stride */
-      GLsizei stride;
-   } user_attrib[PIPE_MAX_ATTRIBS];
-   unsigned num_user_attribs;
-
    /* Active render condition. */
    struct pipe_query *render_condition;
    unsigned condition_mode;
diff --git a/src/mesa/state_tracker/st_draw.c b/src/mesa/state_tracker/st_draw.c
index 42dc3757615..a8c20f45acd 100644
--- a/src/mesa/state_tracker/st_draw.c
+++ b/src/mesa/state_tracker/st_draw.c
@@ -61,6 +61,7 @@
 #include "util/u_format.h"
 #include "util/u_prim.h"
 #include "util/u_draw_quad.h"
+#include "util/u_upload_mgr.h"
 #include "draw/draw_context.h"
 #include "cso_cache/cso_context.h"
 
@@ -402,18 +403,12 @@ setup_interleaved_attribs(struct gl_context *ctx,
                           const struct st_vp_variant *vpv,
                           const struct gl_client_array **arrays,
                           struct pipe_vertex_buffer *vbuffer,
-                          struct pipe_vertex_element velements[],
-                          unsigned max_index,
-                          unsigned num_instances)
+                          struct pipe_vertex_element velements[])
 {
-   struct st_context *st = st_context(ctx);
-   struct pipe_context *pipe = st->pipe;
    GLuint attr;
    const GLubyte *low_addr = NULL;
    GLboolean usingVBO;      /* all arrays in a VBO? */
    struct gl_buffer_object *bufobj;
-   GLuint user_buffer_size = 0;
-   GLuint vertex_size = 0;  /* bytes per vertex, in bytes */
    GLsizei stride;
 
    /* Find the lowest address of the arrays we're drawing,
@@ -464,18 +459,6 @@ setup_interleaved_attribs(struct gl_context *ctx,
                                                          array->Normalized,
                                                          array->Integer);
       assert(velements[attr].src_format);
-
-      if (!usingVBO) {
-         /* how many bytes referenced by this attribute array? */
-         uint divisor = array->InstanceDivisor;
-         uint last_index = divisor ? num_instances / divisor : max_index;
-         uint bytes = src_offset + stride * last_index + element_size;
-
-         user_buffer_size = MAX2(user_buffer_size, bytes);
-
-         /* update vertex size */
-         vertex_size = MAX2(vertex_size, src_offset + element_size);
-      }
    }
 
    /*
@@ -484,9 +467,9 @@ setup_interleaved_attribs(struct gl_context *ctx,
    if (vpv->num_inputs == 0) {
       /* just defensive coding here */
       vbuffer->buffer = NULL;
+      vbuffer->user_buffer = NULL;
       vbuffer->buffer_offset = 0;
       vbuffer->stride = 0;
-      st->num_user_attribs = 0;
    }
    else if (usingVBO) {
       /* all interleaved arrays in a VBO */
@@ -497,26 +480,17 @@ setup_interleaved_attribs(struct gl_context *ctx,
          return GL_FALSE;
       }
 
-      vbuffer->buffer = NULL;
-      pipe_resource_reference(&vbuffer->buffer, stobj->buffer);
+      vbuffer->buffer = stobj->buffer;
+      vbuffer->user_buffer = NULL;
       vbuffer->buffer_offset = pointer_to_offset(low_addr);
       vbuffer->stride = stride;
-      st->num_user_attribs = 0;
    }
    else {
       /* all interleaved arrays in user memory */
-      vbuffer->buffer = pipe_user_buffer_create(pipe->screen,
-                                                (void*) low_addr,
-                                                user_buffer_size,
-                                                PIPE_BIND_VERTEX_BUFFER);
+      vbuffer->buffer = NULL;
+      vbuffer->user_buffer = low_addr;
       vbuffer->buffer_offset = 0;
       vbuffer->stride = stride;
-
-      /* Track user vertex buffers. */
-      pipe_resource_reference(&st->user_attrib[0].buffer, vbuffer->buffer);
-      st->user_attrib[0].element_size = vertex_size;
-      st->user_attrib[0].stride = stride;
-      st->num_user_attribs = 1;
    }
 
    return GL_TRUE;
@@ -536,22 +510,17 @@ setup_non_interleaved_attribs(struct gl_context *ctx,
                               const struct st_vp_variant *vpv,
                               const struct gl_client_array **arrays,
                               struct pipe_vertex_buffer vbuffer[],
-                              struct pipe_vertex_element velements[],
-                              unsigned max_index,
-                              unsigned num_instances)
+                              struct pipe_vertex_element velements[])
 {
-   struct st_context *st = st_context(ctx);
-   struct pipe_context *pipe = st->pipe;
    GLuint attr;
 
    for (attr = 0; attr < vpv->num_inputs; attr++) {
       const GLuint mesaAttr = vp->index_to_input[attr];
       const struct gl_client_array *array = arrays[mesaAttr];
       struct gl_buffer_object *bufobj = array->BufferObj;
-      GLuint element_size = array->_ElementSize;
       GLsizei stride = array->StrideB;
 
-      assert(element_size == array->Size * _mesa_sizeof_type(array->Type));
+      assert(array->_ElementSize == array->Size * _mesa_sizeof_type(array->Type));
 
       if (_mesa_is_bufferobj(bufobj)) {
          /* Attribute data is in a VBO.
@@ -565,49 +534,28 @@ setup_non_interleaved_attribs(struct gl_context *ctx,
             return GL_FALSE;
          }
 
-         vbuffer[attr].buffer = NULL;
-         pipe_resource_reference(&vbuffer[attr].buffer, stobj->buffer);
+         vbuffer[attr].buffer = stobj->buffer;
+         vbuffer[attr].user_buffer = NULL;
          vbuffer[attr].buffer_offset = pointer_to_offset(array->Ptr);
       }
       else {
          /* wrap user data */
-         uint bytes;
          void *ptr;
 
          if (array->Ptr) {
-            uint divisor = array->InstanceDivisor;
-            uint last_index = divisor ? num_instances / divisor : max_index;
-
-            bytes = stride * last_index + element_size;
-
             ptr = (void *) array->Ptr;
          }
          else {
             /* no array, use ctx->Current.Attrib[] value */
-            bytes = element_size = sizeof(ctx->Current.Attrib[0]);
             ptr = (void *) ctx->Current.Attrib[mesaAttr];
             stride = 0;
          }
 
          assert(ptr);
-         assert(bytes);
-
-         vbuffer[attr].buffer =
-            pipe_user_buffer_create(pipe->screen, ptr, bytes,
-                                    PIPE_BIND_VERTEX_BUFFER);
 
+         vbuffer[attr].buffer = NULL;
+         vbuffer[attr].user_buffer = ptr;
          vbuffer[attr].buffer_offset = 0;
-
-         /* Track user vertex buffers. */
-         pipe_resource_reference(&st->user_attrib[attr].buffer, vbuffer[attr].buffer);
-         st->user_attrib[attr].element_size = element_size;
-         st->user_attrib[attr].stride = stride;
-         st->num_user_attribs = MAX2(st->num_user_attribs, attr + 1);
-
-         if (!vbuffer[attr].buffer) {
-            /* probably ran out of memory */
-            return GL_FALSE;
-         }
       }
 
       /* common-case setup */
@@ -629,34 +577,31 @@ setup_non_interleaved_attribs(struct gl_context *ctx,
 
 
 static void
-setup_index_buffer(struct gl_context *ctx,
+setup_index_buffer(struct st_context *st,
                    const struct _mesa_index_buffer *ib,
                    struct pipe_index_buffer *ibuffer)
 {
-   struct st_context *st = st_context(ctx);
-   struct pipe_context *pipe = st->pipe;
+   struct gl_buffer_object *bufobj = ib->obj;
 
-   memset(ibuffer, 0, sizeof(*ibuffer));
-   if (ib) {
-      struct gl_buffer_object *bufobj = ib->obj;
-
-      ibuffer->index_size = vbo_sizeof_ib_type(ib->type);
+   ibuffer->index_size = vbo_sizeof_ib_type(ib->type);
 
-      /* get/create the index buffer object */
-      if (_mesa_is_bufferobj(bufobj)) {
-         /* elements/indexes are in a real VBO */
-         struct st_buffer_object *stobj = st_buffer_object(bufobj);
-         pipe_resource_reference(&ibuffer->buffer, stobj->buffer);
-         ibuffer->offset = pointer_to_offset(ib->ptr);
-      }
-      else {
-         /* element/indicies are in user space memory */
-         ibuffer->buffer =
-            pipe_user_buffer_create(pipe->screen, (void *) ib->ptr,
-                                    ib->count * ibuffer->index_size,
-                                    PIPE_BIND_INDEX_BUFFER);
-      }
+   /* get/create the index buffer object */
+   if (_mesa_is_bufferobj(bufobj)) {
+      /* indices are in a real VBO */
+      ibuffer->buffer = st_buffer_object(bufobj)->buffer;
+      ibuffer->offset = pointer_to_offset(ib->ptr);
    }
+   else if (st->indexbuf_uploader) {
+      u_upload_data(st->indexbuf_uploader, 0, ib->count * ibuffer->index_size,
+                    ib->ptr, &ibuffer->offset, &ibuffer->buffer);
+      u_upload_unmap(st->indexbuf_uploader);
+   }
+   else {
+      /* indices are in user space memory */
+      ibuffer->user_buffer = ib->ptr;
+   }
+
+   cso_set_index_buffer(st->cso_context, ibuffer);
 }
 
 
@@ -811,10 +756,10 @@ handle_fallback_primitive_restart(struct cso_context *cso,
    unsigned num_sub_prims;
 
    assert(info.indexed);
-   assert(ibuffer->buffer);
+   assert(ibuffer->buffer || ibuffer->user_buffer);
    assert(ib);
 
-   if (!ibuffer->buffer || !ib)
+   if (!ibuffer->buffer || !ibuffer->user_buffer || !ib)
       return;
 
    info.primitive_restart = FALSE;
@@ -894,9 +839,7 @@ translate_prim(const struct gl_context *ctx, unsigned prim)
  */
 static GLboolean
 st_validate_varrays(struct gl_context *ctx,
-                    const struct gl_client_array **arrays,
-                    unsigned max_index,
-                    unsigned num_instances)
+                    const struct gl_client_array **arrays)
 {
    struct st_context *st = st_context(ctx);
    const struct st_vertex_program *vp;
@@ -904,8 +847,6 @@ st_validate_varrays(struct gl_context *ctx,
    struct pipe_vertex_buffer vbuffer[PIPE_MAX_SHADER_INPUTS];
    struct pipe_vertex_element velements[PIPE_MAX_ATTRIBS];
    unsigned num_vbuffers, num_velements;
-   GLuint attr;
-   unsigned i;
 
    /* must get these after state validation! */
    vp = st->vp;
@@ -913,18 +854,12 @@ st_validate_varrays(struct gl_context *ctx,
 
    memset(velements, 0, sizeof(struct pipe_vertex_element) * vpv->num_inputs);
 
-   /* Unreference any user vertex buffers. */
-   for (i = 0; i < st->num_user_attribs; i++) {
-      pipe_resource_reference(&st->user_attrib[i].buffer, NULL);
-   }
-   st->num_user_attribs = 0;
-
    /*
     * Setup the vbuffer[] and velements[] arrays.
     */
    if (is_interleaved_arrays(vp, vpv, arrays)) {
-      if (!setup_interleaved_attribs(ctx, vp, vpv, arrays, vbuffer, velements,
-                                     max_index, num_instances)) {
+      if (!setup_interleaved_attribs(ctx, vp, vpv, arrays, vbuffer,
+                                     velements)) {
          return GL_FALSE;
       }
 
@@ -935,8 +870,7 @@ st_validate_varrays(struct gl_context *ctx,
    }
    else {
       if (!setup_non_interleaved_attribs(ctx, vp, vpv, arrays,
-                                         vbuffer, velements, max_index,
-                                         num_instances)) {
+                                         vbuffer, velements)) {
          return GL_FALSE;
       }
 
@@ -947,14 +881,6 @@ st_validate_varrays(struct gl_context *ctx,
    cso_set_vertex_buffers(st->cso_context, num_vbuffers, vbuffer);
    cso_set_vertex_elements(st->cso_context, num_velements, velements);
 
-   /* unreference buffers (frees wrapped user-space buffer objects)
-    * This is OK, because the pipe driver should reference buffers by itself
-    * in set_vertex_buffers. */
-   for (attr = 0; attr < num_vbuffers; attr++) {
-      pipe_resource_reference(&vbuffer[attr].buffer, NULL);
-      assert(!vbuffer[attr].buffer);
-   }
-
    return GL_TRUE;
 }
 
@@ -976,11 +902,10 @@ st_draw_vbo(struct gl_context *ctx,
 {
    struct st_context *st = st_context(ctx);
    struct pipe_context *pipe = st->pipe;
-   struct pipe_index_buffer ibuffer;
+   struct pipe_index_buffer ibuffer = {0};
    struct pipe_draw_info info;
    const struct gl_client_array **arrays = ctx->Array._DrawArrays;
-   unsigned i, num_instances = 1;
-   unsigned max_index_plus_base;
+   unsigned i;
    GLboolean new_array;
 
    /* Mesa core state should have been validated already */
@@ -994,43 +919,6 @@ st_draw_vbo(struct gl_context *ctx,
       (st->dirty.st & (ST_NEW_VERTEX_ARRAYS | ST_NEW_VERTEX_PROGRAM)) ||
       (st->dirty.mesa & (_NEW_PROGRAM | _NEW_BUFFER_OBJECT)) != 0;
 
-   if (ib) {
-      int max_base_vertex = 0;
-
-      /* Gallium probably doesn't want this in some cases. */
-      if (!index_bounds_valid)
-         if (!all_varyings_in_vbos(arrays))
-            vbo_get_minmax_indices(ctx, prims, ib, &min_index, &max_index,
-                                   nr_prims);
-
-      for (i = 0; i < nr_prims; i++) {
-         num_instances = MAX2(num_instances, prims[i].num_instances);
-         max_base_vertex = MAX2(max_base_vertex, prims[i].basevertex);
-      }
-
-      /* Compute the sum of max_index and max_base_vertex.  That's the value
-       * we need to use when creating buffers.
-       */
-      if (max_index == ~0)
-         max_index_plus_base = max_index;
-      else
-         max_index_plus_base = max_index + max_base_vertex;
-   }
-   else {
-      /* Get min/max index for non-indexed drawing. */
-      min_index = ~0;
-      max_index = 0;
-
-      for (i = 0; i < nr_prims; i++) {
-         min_index = MIN2(min_index, prims[i].start);
-         max_index = MAX2(max_index, prims[i].start + prims[i].count - 1);
-         num_instances = MAX2(num_instances, prims[i].num_instances);
-      }
-
-      /* The base vertex offset only applies to indexed drawing */
-      max_index_plus_base = max_index;
-   }
-
    /* Validate state. */
    if (st->dirty.st) {
       GLboolean vertDataEdgeFlags;
@@ -1045,8 +933,7 @@ st_draw_vbo(struct gl_context *ctx,
       st_validate_state(st);
 
       if (new_array) {
-         if (!st_validate_varrays(ctx, arrays, max_index_plus_base,
-                                  num_instances)) {
+         if (!st_validate_varrays(ctx, arrays)) {
             /* probably out of memory, no-op the draw call */
             return;
          }
@@ -1061,31 +948,16 @@ st_draw_vbo(struct gl_context *ctx,
 #endif
    }
 
-   /* Notify the driver that the content of user buffers may have been
-    * changed. */
-   assert(max_index >= min_index);
-   if (!new_array && st->num_user_attribs) {
-      for (i = 0; i < st->num_user_attribs; i++) {
-         if (st->user_attrib[i].buffer) {
-            unsigned element_size = st->user_attrib[i].element_size;
-            unsigned stride = st->user_attrib[i].stride;
-            unsigned min_offset = min_index * stride;
-            unsigned max_offset = max_index_plus_base * stride + element_size;
-
-            assert(max_offset > min_offset);
-
-            pipe->redefine_user_buffer(pipe, st->user_attrib[i].buffer,
-                                       min_offset,
-                                       max_offset - min_offset);
-         }
-      }
-   }
-
-   setup_index_buffer(ctx, ib, &ibuffer);
-   cso_set_index_buffer(st->cso_context, &ibuffer);
-
    util_draw_init_info(&info);
    if (ib) {
+      /* Get index bounds for user buffers. */
+      if (!index_bounds_valid)
+         if (!all_varyings_in_vbos(arrays))
+            vbo_get_minmax_indices(ctx, prims, ib, &min_index, &max_index,
+                                   nr_prims);
+
+      setup_index_buffer(st, ib, &ibuffer);
+
       info.indexed = TRUE;
       if (min_index != ~0 && max_index != ~0) {
          info.min_index = min_index;
@@ -1098,10 +970,12 @@ st_draw_vbo(struct gl_context *ctx,
       info.primitive_restart = ctx->Array.PrimitiveRestart;
       info.restart_index = ctx->Array.RestartIndex;
    }
-
-   /* Set info.count_from_stream_output. */
-   if (tfb_vertcount) {
-      st_transform_feedback_draw_init(tfb_vertcount, &info);
+   else {
+      /* Transform feedback drawing is always non-indexed. */
+      /* Set info.count_from_stream_output. */
+      if (tfb_vertcount) {
+         st_transform_feedback_draw_init(tfb_vertcount, &info);
+      }
    }
 
    /* do actual drawing */
@@ -1134,7 +1008,9 @@ st_draw_vbo(struct gl_context *ctx,
          cso_draw_vbo(st->cso_context, &info);
    }
 
-   pipe_resource_reference(&ibuffer.buffer, NULL);
+   if (ib && st->indexbuf_uploader && !_mesa_is_bufferobj(ib->obj)) {
+      pipe_resource_reference(&ibuffer.buffer, NULL);
+   }
 }
 
 
diff --git a/src/mesa/state_tracker/st_draw_feedback.c b/src/mesa/state_tracker/st_draw_feedback.c
index 257618aa992..4209fb214f8 100644
--- a/src/mesa/state_tracker/st_draw_feedback.c
+++ b/src/mesa/state_tracker/st_draw_feedback.c
@@ -107,7 +107,7 @@ st_feedback_draw_vbo(struct gl_context *ctx,
    struct pipe_vertex_buffer vbuffers[PIPE_MAX_SHADER_INPUTS];
    struct pipe_vertex_element velements[PIPE_MAX_ATTRIBS];
    struct pipe_index_buffer ibuffer;
-   struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
+   struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS] = {NULL};
    struct pipe_transfer *ib_transfer = NULL;
    const struct gl_client_array **arrays = ctx->Array._DrawArrays;
    GLuint attr, i;
@@ -169,23 +169,24 @@ st_feedback_draw_vbo(struct gl_context *ctx,
          assert(stobj->buffer);
 
          vbuffers[attr].buffer = NULL;
+         vbuffers[attr].user_buffer = NULL;
          pipe_resource_reference(&vbuffers[attr].buffer, stobj->buffer);
          vbuffers[attr].buffer_offset = pointer_to_offset(low_addr);
          velements[attr].src_offset = arrays[mesaAttr]->Ptr - low_addr;
+
+         /* map the attrib buffer */
+         map = pipe_buffer_map(pipe, vbuffers[attr].buffer,
+                               PIPE_TRANSFER_READ,
+                               &vb_transfer[attr]);
+         draw_set_mapped_vertex_buffer(draw, attr, map);
       }
       else {
-         /* attribute data is in user-space memory, not a VBO */
-         uint bytes = (arrays[mesaAttr]->Size
-                       * _mesa_sizeof_type(arrays[mesaAttr]->Type)
-                       * (max_index + 1));
-
-         /* wrap user data */
-         vbuffers[attr].buffer
-            = pipe_user_buffer_create(pipe->screen, (void *) arrays[mesaAttr]->Ptr,
-                                      bytes,
-				      PIPE_BIND_VERTEX_BUFFER);
+         vbuffers[attr].buffer = NULL;
+         vbuffers[attr].user_buffer = arrays[mesaAttr]->Ptr;
          vbuffers[attr].buffer_offset = 0;
          velements[attr].src_offset = 0;
+
+         draw_set_mapped_vertex_buffer(draw, attr, vbuffers[attr].user_buffer);
       }
 
       /* common-case setup */
@@ -204,12 +205,6 @@ st_feedback_draw_vbo(struct gl_context *ctx,
 #if 0
       draw_set_vertex_buffer(draw, attr, &vbuffer[attr]);
 #endif
-
-      /* map the attrib buffer */
-      map = pipe_buffer_map(pipe, vbuffers[attr].buffer,
-                            PIPE_TRANSFER_READ,
-			    &vb_transfer[attr]);
-      draw_set_mapped_vertex_buffer(draw, attr, map);
    }
 
    draw_set_vertex_buffers(draw, vp->num_inputs, vbuffers);
@@ -267,7 +262,8 @@ st_feedback_draw_vbo(struct gl_context *ctx,
 
  out_unref_vertex:
    for (attr = 0; attr < vp->num_inputs; attr++) {
-      pipe_buffer_unmap(pipe, vb_transfer[attr]);
+      if (vb_transfer[attr])
+         pipe_buffer_unmap(pipe, vb_transfer[attr]);
       draw_set_mapped_vertex_buffer(draw, attr, NULL);
       pipe_resource_reference(&vbuffers[attr].buffer, NULL);
    }