mesa: replace ctx->Const.{Vertex,Fragment,Geomtery}Program with an array.
[mesa.git] / src / mesa / state_tracker / st_draw.c
index 20ba993800a635b2eeeb011b6efcd258c3112e8d..ef0a60732bcc59fb2e1bf4ef9d6137bc7ef41a23 100644 (file)
@@ -1,8 +1,8 @@
 /**************************************************************************
- * 
+ *
  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
  * All Rights Reserved.
- * 
+ *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the
  * "Software"), to deal in the Software without restriction, including
  * distribute, sub license, and/or sell copies of the Software, and to
  * permit persons to whom the Software is furnished to do so, subject to
  * the following conditions:
- * 
+ *
  * The above copyright notice and this permission notice (including the
  * next paragraph) shall be included in all copies or substantial portions
  * of the Software.
- * 
+ *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
@@ -22,7 +22,7 @@
  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
+ *
  **************************************************************************/
 
 /*
  * this function whether the user called glBegin/End, glDrawArrays,
  * glDrawElements, glEvalMesh, or glCalList, etc.
  *
- * We basically convert the VBO's vertex attribute/array information into
- * Gallium vertex state, bind the vertex buffer objects and call
- * pipe->draw_vbo().
- *
  * Authors:
  *   Keith Whitwell <keith@tungstengraphics.com>
  */
 #include "main/image.h"
 #include "main/bufferobj.h"
 #include "main/macros.h"
-#include "main/mfeatures.h"
-#include "program/prog_uniform.h"
 
 #include "vbo/vbo.h"
 
 #include "st_context.h"
 #include "st_atom.h"
 #include "st_cb_bufferobjects.h"
+#include "st_cb_xformfb.h"
+#include "st_debug.h"
 #include "st_draw.h"
 #include "st_program.h"
 
 #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"
 
-
-static GLuint double_types[4] = {
-   PIPE_FORMAT_R64_FLOAT,
-   PIPE_FORMAT_R64G64_FLOAT,
-   PIPE_FORMAT_R64G64B64_FLOAT,
-   PIPE_FORMAT_R64G64B64A64_FLOAT
-};
-
-static GLuint float_types[4] = {
-   PIPE_FORMAT_R32_FLOAT,
-   PIPE_FORMAT_R32G32_FLOAT,
-   PIPE_FORMAT_R32G32B32_FLOAT,
-   PIPE_FORMAT_R32G32B32A32_FLOAT
-};
-
-static GLuint half_float_types[4] = {
-   PIPE_FORMAT_R16_FLOAT,
-   PIPE_FORMAT_R16G16_FLOAT,
-   PIPE_FORMAT_R16G16B16_FLOAT,
-   PIPE_FORMAT_R16G16B16A16_FLOAT
-};
-
-static GLuint uint_types_norm[4] = {
-   PIPE_FORMAT_R32_UNORM,
-   PIPE_FORMAT_R32G32_UNORM,
-   PIPE_FORMAT_R32G32B32_UNORM,
-   PIPE_FORMAT_R32G32B32A32_UNORM
-};
-
-static GLuint uint_types_scale[4] = {
-   PIPE_FORMAT_R32_USCALED,
-   PIPE_FORMAT_R32G32_USCALED,
-   PIPE_FORMAT_R32G32B32_USCALED,
-   PIPE_FORMAT_R32G32B32A32_USCALED
-};
-
-static GLuint int_types_norm[4] = {
-   PIPE_FORMAT_R32_SNORM,
-   PIPE_FORMAT_R32G32_SNORM,
-   PIPE_FORMAT_R32G32B32_SNORM,
-   PIPE_FORMAT_R32G32B32A32_SNORM
-};
-
-static GLuint int_types_scale[4] = {
-   PIPE_FORMAT_R32_SSCALED,
-   PIPE_FORMAT_R32G32_SSCALED,
-   PIPE_FORMAT_R32G32B32_SSCALED,
-   PIPE_FORMAT_R32G32B32A32_SSCALED
-};
-
-static GLuint ushort_types_norm[4] = {
-   PIPE_FORMAT_R16_UNORM,
-   PIPE_FORMAT_R16G16_UNORM,
-   PIPE_FORMAT_R16G16B16_UNORM,
-   PIPE_FORMAT_R16G16B16A16_UNORM
-};
-
-static GLuint ushort_types_scale[4] = {
-   PIPE_FORMAT_R16_USCALED,
-   PIPE_FORMAT_R16G16_USCALED,
-   PIPE_FORMAT_R16G16B16_USCALED,
-   PIPE_FORMAT_R16G16B16A16_USCALED
-};
-
-static GLuint short_types_norm[4] = {
-   PIPE_FORMAT_R16_SNORM,
-   PIPE_FORMAT_R16G16_SNORM,
-   PIPE_FORMAT_R16G16B16_SNORM,
-   PIPE_FORMAT_R16G16B16A16_SNORM
-};
-
-static GLuint short_types_scale[4] = {
-   PIPE_FORMAT_R16_SSCALED,
-   PIPE_FORMAT_R16G16_SSCALED,
-   PIPE_FORMAT_R16G16B16_SSCALED,
-   PIPE_FORMAT_R16G16B16A16_SSCALED
-};
-
-static GLuint ubyte_types_norm[4] = {
-   PIPE_FORMAT_R8_UNORM,
-   PIPE_FORMAT_R8G8_UNORM,
-   PIPE_FORMAT_R8G8B8_UNORM,
-   PIPE_FORMAT_R8G8B8A8_UNORM
-};
-
-static GLuint ubyte_types_scale[4] = {
-   PIPE_FORMAT_R8_USCALED,
-   PIPE_FORMAT_R8G8_USCALED,
-   PIPE_FORMAT_R8G8B8_USCALED,
-   PIPE_FORMAT_R8G8B8A8_USCALED
-};
-
-static GLuint byte_types_norm[4] = {
-   PIPE_FORMAT_R8_SNORM,
-   PIPE_FORMAT_R8G8_SNORM,
-   PIPE_FORMAT_R8G8B8_SNORM,
-   PIPE_FORMAT_R8G8B8A8_SNORM
-};
-
-static GLuint byte_types_scale[4] = {
-   PIPE_FORMAT_R8_SSCALED,
-   PIPE_FORMAT_R8G8_SSCALED,
-   PIPE_FORMAT_R8G8B8_SSCALED,
-   PIPE_FORMAT_R8G8B8A8_SSCALED
-};
-
-static GLuint fixed_types[4] = {
-   PIPE_FORMAT_R32_FIXED,
-   PIPE_FORMAT_R32G32_FIXED,
-   PIPE_FORMAT_R32G32B32_FIXED,
-   PIPE_FORMAT_R32G32B32A32_FIXED
-};
-
-
-
-/**
- * Return a PIPE_FORMAT_x for the given GL datatype and size.
- */
-enum pipe_format
-st_pipe_vertex_format(GLenum type, GLuint size, GLenum format,
-                      GLboolean normalized)
-{
-   assert((type >= GL_BYTE && type <= GL_DOUBLE) ||
-          type == GL_FIXED || type == GL_HALF_FLOAT ||
-          type == GL_INT_2_10_10_10_REV ||
-          type == GL_UNSIGNED_INT_2_10_10_10_REV);
-   assert(size >= 1);
-   assert(size <= 4);
-   assert(format == GL_RGBA || format == GL_BGRA);
-
-   if (type == GL_INT_2_10_10_10_REV ||
-       type == GL_UNSIGNED_INT_2_10_10_10_REV) {
-      assert(size == 4);
-
-      if (format == GL_BGRA) {
-         if (type == GL_INT_2_10_10_10_REV) {
-            if (normalized)
-               return PIPE_FORMAT_B10G10R10A2_SNORM;
-            else
-               return PIPE_FORMAT_B10G10R10A2_SSCALED;
-         } else {
-            if (normalized)
-               return PIPE_FORMAT_B10G10R10A2_UNORM;
-            else
-               return PIPE_FORMAT_B10G10R10A2_USCALED;
-         }
-      } else {
-         if (type == GL_INT_2_10_10_10_REV) {
-            if (normalized)
-               return PIPE_FORMAT_R10G10B10A2_SNORM;
-            else
-               return PIPE_FORMAT_R10G10B10A2_SSCALED;
-         } else {
-            if (normalized)
-               return PIPE_FORMAT_R10G10B10A2_UNORM;
-            else
-               return PIPE_FORMAT_R10G10B10A2_USCALED;
-         }
-      }
-   }
-
-   if (format == GL_BGRA) {
-      /* this is an odd-ball case */
-      assert(type == GL_UNSIGNED_BYTE);
-      assert(normalized);
-      return PIPE_FORMAT_B8G8R8A8_UNORM;
-   }
-
-   if (normalized) {
-      switch (type) {
-      case GL_DOUBLE: return double_types[size-1];
-      case GL_FLOAT: return float_types[size-1];
-      case GL_HALF_FLOAT: return half_float_types[size-1];
-      case GL_INT: return int_types_norm[size-1];
-      case GL_SHORT: return short_types_norm[size-1];
-      case GL_BYTE: return byte_types_norm[size-1];
-      case GL_UNSIGNED_INT: return uint_types_norm[size-1];
-      case GL_UNSIGNED_SHORT: return ushort_types_norm[size-1];
-      case GL_UNSIGNED_BYTE: return ubyte_types_norm[size-1];
-      case GL_FIXED: return fixed_types[size-1];
-      default: assert(0); return 0;
-      }
-   }
-   else {
-      switch (type) {
-      case GL_DOUBLE: return double_types[size-1];
-      case GL_FLOAT: return float_types[size-1];
-      case GL_HALF_FLOAT: return half_float_types[size-1];
-      case GL_INT: return int_types_scale[size-1];
-      case GL_SHORT: return short_types_scale[size-1];
-      case GL_BYTE: return byte_types_scale[size-1];
-      case GL_UNSIGNED_INT: return uint_types_scale[size-1];
-      case GL_UNSIGNED_SHORT: return ushort_types_scale[size-1];
-      case GL_UNSIGNED_BYTE: return ubyte_types_scale[size-1];
-      case GL_FIXED: return fixed_types[size-1];
-      default: assert(0); return 0;
-      }
-   }
-   return PIPE_FORMAT_NONE; /* silence compiler warning */
-}
+#include "../glsl/ir_uniform.h"
 
 
 /**
@@ -274,7 +72,7 @@ static GLboolean
 all_varyings_in_vbos(const struct gl_client_array *arrays[])
 {
    GLuint i;
-   
+
    for (i = 0; i < VERT_ATTRIB_MAX; i++)
       if (arrays[i]->StrideB &&
           !arrays[i]->InstanceDivisor &&
@@ -286,267 +84,42 @@ all_varyings_in_vbos(const struct gl_client_array *arrays[])
 
 
 /**
- * Examine the active arrays to determine if we have interleaved
- * vertex arrays all living in one VBO, or all living in user space.
+ * Basically, translate Mesa's index buffer information into
+ * a pipe_index_buffer object.
+ * \return TRUE or FALSE for success/failure
  */
-static GLboolean
-is_interleaved_arrays(const struct st_vertex_program *vp,
-                      const struct st_vp_variant *vpv,
-                      const struct gl_client_array **arrays)
+static boolean
+setup_index_buffer(struct st_context *st,
+                   const struct _mesa_index_buffer *ib,
+                   struct pipe_index_buffer *ibuffer)
 {
-   GLuint attr;
-   const struct gl_buffer_object *firstBufObj = NULL;
-   GLint firstStride = -1;
-   const GLubyte *firstPtr = NULL;
-   GLboolean userSpaceBuffer = GL_FALSE;
-
-   for (attr = 0; attr < vpv->num_inputs; attr++) {
-      const GLuint mesaAttr = vp->index_to_input[attr];
-      const struct gl_client_array *array = arrays[mesaAttr];
-      const struct gl_buffer_object *bufObj = array->BufferObj;
-      const GLsizei stride = array->StrideB; /* in bytes */
-
-      if (attr == 0) {
-         /* save info about the first array */
-         firstStride = stride;
-         firstPtr = array->Ptr;         
-         firstBufObj = bufObj;
-         userSpaceBuffer = !bufObj || !bufObj->Name;
-      }
-      else {
-         /* check if other arrays interleave with the first, in same buffer */
-         if (stride != firstStride)
-            return GL_FALSE; /* strides don't match */
-
-         if (bufObj != firstBufObj)
-            return GL_FALSE; /* arrays in different VBOs */
+   struct gl_buffer_object *bufobj = ib->obj;
 
-         if (abs(array->Ptr - firstPtr) > firstStride)
-            return GL_FALSE; /* arrays start too far apart */
+   ibuffer->index_size = vbo_sizeof_ib_type(ib->type);
 
-         if ((!bufObj || !_mesa_is_bufferobj(bufObj)) != userSpaceBuffer)
-            return GL_FALSE; /* mix of VBO and user-space arrays */
-      }
+   /* 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);
    }
-
-   return GL_TRUE;
-}
-
-
-/**
- * Set up for drawing interleaved arrays that all live in one VBO
- * or all live in user space.
- * \param vbuffer  returns vertex buffer info
- * \param velements  returns vertex element info
- */
-static void
-setup_interleaved_attribs(struct gl_context *ctx,
-                          const struct st_vertex_program *vp,
-                          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 st_context *st = st_context(ctx);
-   struct pipe_context *pipe = st->pipe;
-   GLuint attr;
-   const GLubyte *low_addr = NULL;
-
-   /* Find the lowest address of the arrays we're drawing */
-   if (vpv->num_inputs) {
-      low_addr = arrays[vp->index_to_input[0]]->Ptr;
-
-      for (attr = 1; attr < vpv->num_inputs; attr++) {
-         const GLubyte *start = arrays[vp->index_to_input[attr]]->Ptr;
-         low_addr = MIN2(low_addr, start);
+   else if (st->indexbuf_uploader) {
+      /* upload indexes from user memory into a real buffer */
+      if (u_upload_data(st->indexbuf_uploader, 0,
+                        ib->count * ibuffer->index_size, ib->ptr,
+                        &ibuffer->offset, &ibuffer->buffer) != PIPE_OK) {
+         /* out of memory */
+         return FALSE;
       }
+      u_upload_unmap(st->indexbuf_uploader);
    }
-
-   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;
-      struct st_buffer_object *stobj = st_buffer_object(bufobj);
-      unsigned src_offset = (unsigned) (array->Ptr - low_addr);
-      GLuint element_size = array->_ElementSize;
-      GLsizei stride = array->StrideB;
-
-      assert(element_size == array->Size * _mesa_sizeof_type(array->Type));
-
-      if (attr == 0) {
-         if (bufobj && _mesa_is_bufferobj(bufobj)) {
-            vbuffer->buffer = NULL;
-            pipe_resource_reference(&vbuffer->buffer, stobj->buffer);
-            vbuffer->buffer_offset = pointer_to_offset(low_addr);
-         }
-         else {
-            uint divisor = array->InstanceDivisor;
-            uint last_index = divisor ? num_instances / divisor : max_index;
-            uint bytes = src_offset + stride * last_index + element_size;
-
-            vbuffer->buffer = pipe_user_buffer_create(pipe->screen,
-                                                      (void*) low_addr,
-                                                      bytes,
-                                                      PIPE_BIND_VERTEX_BUFFER);
-            vbuffer->buffer_offset = 0;
-
-            /* Track user vertex buffers. */
-            pipe_resource_reference(&st->user_attrib[0].buffer, vbuffer->buffer);
-            st->user_attrib[0].element_size = element_size;
-            st->user_attrib[0].stride = stride;
-            st->num_user_attribs = 1;
-         }
-         vbuffer->stride = stride; /* in bytes */
-      }
-
-      velements[attr].src_offset = src_offset;
-      velements[attr].instance_divisor = array->InstanceDivisor;
-      velements[attr].vertex_buffer_index = 0;
-      velements[attr].src_format = st_pipe_vertex_format(array->Type,
-                                                         array->Size,
-                                                         array->Format,
-                                                         array->Normalized);
-      assert(velements[attr].src_format);
+   else {
+      /* indices are in user space memory */
+      ibuffer->user_buffer = ib->ptr;
    }
-}
-
-
-/**
- * Set up a separate pipe_vertex_buffer and pipe_vertex_element for each
- * vertex attribute.
- * \param vbuffer  returns vertex buffer info
- * \param velements  returns vertex element info
- */
-static void
-setup_non_interleaved_attribs(struct gl_context *ctx,
-                              const struct st_vertex_program *vp,
-                              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 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));
-
-      if (bufobj && _mesa_is_bufferobj(bufobj)) {
-         /* Attribute data is in a VBO.
-          * Recall that for VBOs, the gl_client_array->Ptr field is
-          * really an offset from the start of the VBO, not a pointer.
-          */
-         struct st_buffer_object *stobj = st_buffer_object(bufobj);
-         assert(stobj->buffer);
-
-         vbuffer[attr].buffer = NULL;
-         pipe_resource_reference(&vbuffer[attr].buffer, stobj->buffer);
-         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_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);
-      }
-
-      /* common-case setup */
-      vbuffer[attr].stride = stride; /* in bytes */
-
-      velements[attr].src_offset = 0;
-      velements[attr].instance_divisor = array->InstanceDivisor;
-      velements[attr].vertex_buffer_index = attr;
-      velements[attr].src_format = st_pipe_vertex_format(array->Type,
-                                                         array->Size,
-                                                         array->Format,
-                                                         array->Normalized);
-      assert(velements[attr].src_format);
-   }
-}
-
-
-static void
-setup_index_buffer(struct gl_context *ctx,
-                   const struct _mesa_index_buffer *ib,
-                   struct pipe_index_buffer *ibuffer)
-{
-   struct st_context *st = st_context(ctx);
-   struct pipe_context *pipe = st->pipe;
-
-   memset(ibuffer, 0, sizeof(*ibuffer));
-   if (ib) {
-      struct gl_buffer_object *bufobj = ib->obj;
-
-      switch (ib->type) {
-      case GL_UNSIGNED_INT:
-         ibuffer->index_size = 4;
-         break;
-      case GL_UNSIGNED_SHORT:
-         ibuffer->index_size = 2;
-         break;
-      case GL_UNSIGNED_BYTE:
-         ibuffer->index_size = 1;
-         break;
-      default:
-         assert(0);
-        return;
-      }
-
-      /* get/create the index buffer object */
-      if (bufobj && _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);
-      }
-   }
+   cso_set_index_buffer(st->cso_context, ibuffer);
+   return TRUE;
 }
 
 
@@ -571,134 +144,15 @@ check_uniforms(struct gl_context *ctx)
       if (shProg[j] == NULL || !shProg[j]->LinkStatus)
         continue;
 
-      for (i = 0; i < shProg[j]->Uniforms->NumUniforms; i++) {
-         const struct gl_uniform *u = &shProg[j]->Uniforms->Uniforms[i];
-         if (!u->Initialized) {
+      for (i = 0; i < shProg[j]->NumUserUniformStorage; i++) {
+         const struct gl_uniform_storage *u = &shProg[j]->UniformStorage[i];
+         if (!u->initialized) {
             _mesa_warning(ctx,
                           "Using shader with uninitialized uniform: %s",
-                          u->Name);
-         }
-      }
-   }
-}
-
-/** Helper code for primitive restart fallback */
-#define DO_DRAW(pipe, cur_start, cur_count) \
-   do { \
-      info.start = cur_start; \
-      info.count = cur_count; \
-      if (u_trim_pipe_prim(info.mode, &info.count)) { \
-         if (transfer) \
-            pipe_buffer_unmap(pipe, transfer); \
-         pipe->draw_vbo(pipe, &info); \
-         if (transfer) { \
-            ptr = pipe_buffer_map(pipe, ibuffer->buffer, PIPE_TRANSFER_READ, &transfer); \
-            assert(ptr != NULL); \
-            ptr = ADD_POINTERS(ptr, ibuffer->offset); \
-         } \
-      } \
-   } while(0)
-      
-/** More helper code for primitive restart fallback */
-#define PRIM_RESTART_LOOP(elements) \
-   do { \
-      for (i = start; i < end; i++) { \
-         if (elements[i] == info.restart_index) { \
-            if (cur_count > 0) { \
-               /* draw elts up to prev pos */ \
-               DO_DRAW(pipe, cur_start, cur_count); \
-            } \
-            /* begin new prim at next elt */ \
-            cur_start = i + 1; \
-            cur_count = 0; \
-         } \
-         else { \
-            cur_count++; \
-         } \
-      } \
-      if (cur_count > 0) { \
-         DO_DRAW(pipe, cur_start, cur_count); \
-      } \
-   } while (0)
-
-static void
-handle_fallback_primitive_restart(struct pipe_context *pipe,
-                                  const struct _mesa_index_buffer *ib,
-                                  struct pipe_index_buffer *ibuffer,
-                                  struct pipe_draw_info *orig_info)
-{
-   const unsigned start = orig_info->start;
-   const unsigned count = orig_info->count;
-   const unsigned end = start + count;
-   struct pipe_draw_info info = *orig_info;
-   struct pipe_transfer *transfer = NULL;
-   unsigned instance, i, cur_start, cur_count;
-   const void *ptr;
-
-   info.primitive_restart = FALSE;
-
-   if (!info.indexed) {
-      /* Splitting the draw arrays call is handled by the VBO module */
-      if (u_trim_pipe_prim(info.mode, &info.count))
-         pipe->draw_vbo(pipe, &info);
-
-      return;
-   }
-
-   /* info.indexed == TRUE */
-   assert(ibuffer);
-   assert(ibuffer->buffer);
-
-   if (ib) {
-      struct gl_buffer_object *bufobj = ib->obj;
-      if (bufobj && bufobj->Name) {
-         ptr = NULL;
-      }
-      else {
-         ptr = ib->ptr;
-      }
-   }
-
-   if (!ptr)
-      ptr = pipe_buffer_map(pipe, ibuffer->buffer, PIPE_TRANSFER_READ, &transfer);
-
-   if (!ptr)
-     return;
-   ptr = ADD_POINTERS(ptr, ibuffer->offset);
-
-   /* Need to loop over instances as well to preserve draw order */
-   for (instance = 0; instance < orig_info->instance_count; instance++) {
-      info.start_instance = instance + orig_info->start_instance;
-      info.instance_count = 1;
-      cur_start = start;
-      cur_count = 0;
-
-      switch (ibuffer->index_size) {
-      case 1:
-         {
-            const ubyte *elt_ub = (const ubyte *)ptr; 
-            PRIM_RESTART_LOOP(elt_ub);
-         }
-         break;
-      case 2:
-         {
-            const ushort *elt_us = (const ushort *)ptr;
-            PRIM_RESTART_LOOP(elt_us);
-         }
-         break;
-      case 4:
-         {
-            const uint *elt_ui = (const uint *)ptr;
-            PRIM_RESTART_LOOP(elt_ui);
+                          u->name);
          }
-         break;
-      default:
-         assert(0 && "bad index_size in handle_fallback_primitive_restart()");
       }
    }
-
-   if (transfer)
-      pipe_buffer_unmap(pipe, transfer);
 }
 
 
@@ -710,9 +164,9 @@ static unsigned
 translate_prim(const struct gl_context *ctx, unsigned prim)
 {
    /* GL prims should match Gallium prims, spot-check a few */
-   assert(GL_POINTS == PIPE_PRIM_POINTS);
-   assert(GL_QUADS == PIPE_PRIM_QUADS);
-   assert(GL_TRIANGLE_STRIP_ADJACENCY == PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY);
+   STATIC_ASSERT(GL_POINTS == PIPE_PRIM_POINTS);
+   STATIC_ASSERT(GL_QUADS == PIPE_PRIM_QUADS);
+   STATIC_ASSERT(GL_TRIANGLE_STRIP_ADJACENCY == PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY);
 
    /* Avoid quadstrips if it's easy to do so:
     * Note: it's important to do the correct trimming if we change the
@@ -728,66 +182,6 @@ translate_prim(const struct gl_context *ctx, unsigned prim)
 }
 
 
-static void
-st_validate_varrays(struct gl_context *ctx,
-                    const struct gl_client_array **arrays,
-                    unsigned max_index,
-                    unsigned num_instances)
-{
-   struct st_context *st = st_context(ctx);
-   const struct st_vertex_program *vp;
-   const struct st_vp_variant *vpv;
-   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;
-   vpv = st->vp_variant;
-
-   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)) {
-      setup_interleaved_attribs(ctx, vp, vpv, arrays, vbuffer, velements,
-                                max_index, num_instances);
-
-      num_vbuffers = 1;
-      num_velements = vpv->num_inputs;
-      if (num_velements == 0)
-         num_vbuffers = 0;
-   }
-   else {
-      setup_non_interleaved_attribs(ctx, vp, vpv, arrays,
-                                    vbuffer, velements, max_index,
-                                    num_instances);
-      num_vbuffers = vpv->num_inputs;
-      num_velements = vpv->num_inputs;
-   }
-
-   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);
-   }
-}
-
-
 /**
  * This function gets plugged into the VBO module and is called when
  * we have something to render.
@@ -795,68 +189,28 @@ st_validate_varrays(struct gl_context *ctx,
  */
 void
 st_draw_vbo(struct gl_context *ctx,
-            const struct gl_client_array **arrays,
             const struct _mesa_prim *prims,
             GLuint nr_prims,
             const struct _mesa_index_buffer *ib,
            GLboolean index_bounds_valid,
             GLuint min_index,
-            GLuint max_index)
+            GLuint max_index,
+            struct gl_transform_feedback_object *tfb_vertcount,
+            struct gl_buffer_object *indirect)
 {
    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;
-   unsigned i, num_instances = 1;
-   GLboolean new_array =
-      st->dirty.st &&
-      (st->dirty.mesa & (_NEW_ARRAY | _NEW_PROGRAM | _NEW_BUFFER_OBJECT)) != 0;
+   const struct gl_client_array **arrays = ctx->Array._DrawArrays;
+   unsigned i;
 
    /* Mesa core state should have been validated already */
    assert(ctx->NewState == 0x0);
 
-   if (ib) {
-      /* Gallium probably doesn't want this in some cases. */
-      if (!index_bounds_valid)
-         if (!all_varyings_in_vbos(arrays))
-            vbo_get_minmax_index(ctx, prims, ib, &min_index, &max_index);
-
-      for (i = 0; i < nr_prims; i++) {
-         num_instances = MAX2(num_instances, prims[i].num_instances);
-      }
-   }
-   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);
-      }
-   }
-
    /* Validate state. */
-   if (st->dirty.st) {
-      GLboolean vertDataEdgeFlags;
-
-      /* sanity check for pointer arithmetic below */
-      assert(sizeof(arrays[0]->Ptr[0]) == 1);
-
-      vertDataEdgeFlags = arrays[VERT_ATTRIB_EDGEFLAG]->BufferObj &&
-                          arrays[VERT_ATTRIB_EDGEFLAG]->BufferObj->Name;
-      if (vertDataEdgeFlags != st->vertdata_edgeflags) {
-         st->vertdata_edgeflags = vertDataEdgeFlags;
-         st->dirty.st |= ST_NEW_EDGEFLAGS_DATA;
-      }
-
+   if (st->dirty.st || ctx->NewDriverState) {
       st_validate_state(st);
 
-      if (new_array) {
-         st_validate_varrays(ctx, arrays, max_index, num_instances);
-      }
-
 #if 0
       if (MESA_VERBOSE & VERBOSE_GLSL) {
          check_uniforms(ctx);
@@ -866,46 +220,50 @@ 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 * stride + element_size;
-
-            assert(max_offset > min_offset);
-
-            pipe->redefine_user_buffer(pipe, st->user_attrib[i].buffer,
-                                       min_offset,
-                                       max_offset - min_offset);
-         }
-      }
+   if (st->vertex_array_out_of_memory) {
+      return;
    }
 
-   setup_index_buffer(ctx, ib, &ibuffer);
-   pipe->set_index_buffer(pipe, &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);
+
+      if (!setup_index_buffer(st, ib, &ibuffer)) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glBegin/DrawElements/DrawArray");
+         return;
+      }
+
       info.indexed = TRUE;
       if (min_index != ~0 && max_index != ~0) {
          info.min_index = min_index;
          info.max_index = max_index;
       }
-   }
 
-   info.primitive_restart = ctx->Array.PrimitiveRestart;
-   info.restart_index = ctx->Array.RestartIndex;
+      /* The VBO module handles restart for the non-indexed GLDrawArrays
+       * so we only set these fields for indexed drawing:
+       */
+      info.primitive_restart = ctx->Array._PrimitiveRestart;
+      info.restart_index = ctx->Array.RestartIndex;
+   }
+   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 */
    for (i = 0; i < nr_prims; i++) {
-      info.mode = translate_prim( ctx, prims[i].mode );
+      info.mode = translate_prim(ctx, prims[i].mode);
       info.start = prims[i].start;
       info.count = prims[i].count;
+      info.start_instance = prims[i].base_instance;
       info.instance_count = prims[i].num_instances;
       info.index_bias = prims[i].basevertex;
       if (!ib) {
@@ -913,26 +271,29 @@ st_draw_vbo(struct gl_context *ctx,
          info.max_index = info.start + info.count - 1;
       }
 
-      if (info.primitive_restart) {
-         /*
-          * Handle primitive restart for drivers that doesn't support it.
-          *
-          * The VBO module handles restart inside of draw_arrays for us,
-          * but we should still remove the primitive_restart flag on the
-          * info struct, the fallback function does this for us. Just
-          * remove the flag for all drivers in this case as well.
-          */
-         if (st->sw_primitive_restart || !info.indexed)
-            handle_fallback_primitive_restart(pipe, ib, &ibuffer, &info);
-         else
-            /* don't trim, restarts might be inside index list */
-            pipe->draw_vbo(pipe, &info);
+      if (ST_DEBUG & DEBUG_DRAW) {
+         debug_printf("st/draw: mode %s  start %u  count %u  indexed %d\n",
+                      u_prim_name(info.mode),
+                      info.start,
+                      info.count,
+                      info.indexed);
+      }
+
+      if (info.count_from_stream_output) {
+         cso_draw_vbo(st->cso_context, &info);
+      }
+      else if (info.primitive_restart) {
+         /* don't trim, restarts might be inside index list */
+         cso_draw_vbo(st->cso_context, &info);
+      }
+      else if (u_trim_pipe_prim(prims[i].mode, &info.count)) {
+         cso_draw_vbo(st->cso_context, &info);
       }
-      else if (u_trim_pipe_prim(info.mode, &info.count))
-         pipe->draw_vbo(pipe, &info);
    }
 
-   pipe_resource_reference(&ibuffer.buffer, NULL);
+   if (ib && st->indexbuf_uploader && !_mesa_is_bufferobj(ib->obj)) {
+      pipe_resource_reference(&ibuffer.buffer, NULL);
+   }
 }
 
 
@@ -943,7 +304,6 @@ st_init_draw(struct st_context *st)
 
    vbo_set_draw_func(ctx, st_draw_vbo);
 
-#if FEATURE_feedback || FEATURE_rastpos
    st->draw = draw_create(st->pipe); /* for selection/feedback */
 
    /* Disable draw options that might convert points/lines to tris, etc.
@@ -953,14 +313,11 @@ st_init_draw(struct st_context *st)
    draw_wide_point_threshold(st->draw, 1000.0f);
    draw_enable_line_stipple(st->draw, FALSE);
    draw_enable_point_sprites(st->draw, FALSE);
-#endif
 }
 
 
 void
 st_destroy_draw(struct st_context *st)
 {
-#if FEATURE_feedback || FEATURE_rastpos
    draw_destroy(st->draw);
-#endif
 }