/**
 * Called via glArrayElement() and glDrawArrays().
 * Issue the glNormal, glVertex, glColor, glVertexAttrib, etc functions
 * for all enabled vertex arrays (for elt-th element).
 * Note: this may be called during display list construction.
 */
void GLAPIENTRY _ae_loopback_array_elt( GLint elt )
{
   GET_CURRENT_CONTEXT(ctx);
   const AEcontext *actx = AE_CONTEXT(ctx);
   const AEarray *aa;
   const AEattrib *at;
   const struct _glapi_table * const disp = GET_DISPATCH();


   if (actx->NewState)
      _ae_update_state( ctx );

   /* generic attributes */
   for (at = actx->attribs; at->func; at++) {
      const GLubyte *src
         = ADD_POINTERS(at->array->BufferObj->Data, at->array->Ptr)
         + elt * at->array->StrideB;
      at->func( at->index, src );
   }

   /* conventional arrays */
   for (aa = actx->arrays; aa->offset != -1 ; aa++) {
      const GLubyte *src
         = ADD_POINTERS(aa->array->BufferObj->Data, aa->array->Ptr)
         + elt * aa->array->StrideB;
      CALL_by_offset( disp, (array_func), aa->offset, 
		      ((const void *) src) );
   }
}

void GLAPIENTRY
_mesa_GetPixelMapfv( GLenum map, GLfloat *values )
{
   GET_CURRENT_CONTEXT(ctx);
   GLuint mapsize, i;
   const struct gl_pixelmap *pm;

   ASSERT_OUTSIDE_BEGIN_END(ctx);

   pm = get_pixelmap(ctx, map);
   if (!pm) {
      _mesa_error(ctx, GL_INVALID_ENUM, "glGetPixelMapfv(map)");
      return;
   }

   mapsize = pm->Size;

   if (ctx->Pack.BufferObj->Name) {
      /* pack pixelmap into PBO */
      GLubyte *buf;
      /* Note, need to use DefaultPacking and Pack's buffer object */
      ctx->DefaultPacking.BufferObj = ctx->Pack.BufferObj;
      if (!_mesa_validate_pbo_access(1, &ctx->DefaultPacking, mapsize, 1, 1,
                                     GL_INTENSITY, GL_FLOAT, values)) {
         _mesa_error(ctx, GL_INVALID_OPERATION,
                     "glGetPixelMapfv(invalid PBO access)");
         return;
      }
      /* restore */
      ctx->DefaultPacking.BufferObj = ctx->Array.NullBufferObj;
      buf = (GLubyte *) ctx->Driver.MapBuffer(ctx, GL_PIXEL_PACK_BUFFER_EXT,
                                              GL_WRITE_ONLY_ARB,
                                              ctx->Pack.BufferObj);
      if (!buf) {
         /* buffer is already mapped - that's an error */
         _mesa_error(ctx, GL_INVALID_OPERATION,
                     "glGetPixelMapfv(PBO is mapped)");
         return;
      }
      values = (GLfloat *) ADD_POINTERS(buf, values);
   }
   else if (!values) {
      return;
   }

   if (map == GL_PIXEL_MAP_S_TO_S) {
      /* special case */
      for (i = 0; i < mapsize; i++) {
         values[i] = (GLfloat) ctx->PixelMaps.StoS.Map[i];
      }
   }
   else {
      MEMCPY(values, pm->Map, mapsize * sizeof(GLfloat));
   }

   if (ctx->Pack.BufferObj->Name) {
      ctx->Driver.UnmapBuffer(ctx, GL_PIXEL_PACK_BUFFER_EXT,
                              ctx->Pack.BufferObj);
   }
}

/**
 * For commands that write to a PBO (glReadPixels, glGetColorTable, etc),
 * if we're writing to a PBO, map it write-only and return the pointer
 * into the PBO.  If we're not writing to a PBO, return \p dst as-is.
 * If non-null return, must call _mesa_unmap_pbo_dest() when done.
 *
 * \return NULL if error, else pointer to start of data
 */
void *
_mesa_map_pbo_dest(struct gl_context *ctx,
                   const struct gl_pixelstore_attrib *pack,
                   GLvoid *dest)
{
   void *buf;

   if (_mesa_is_bufferobj(pack->BufferObj)) {
      /* pack into PBO */
      buf = (GLubyte *) ctx->Driver.MapBufferRange(ctx, 0,
						   pack->BufferObj->Size,
						   GL_MAP_WRITE_BIT,
						   pack->BufferObj,
                                                   MAP_INTERNAL);
      if (!buf)
         return NULL;

      buf = ADD_POINTERS(buf, dest);
   }
   else {
      /* pack to normal memory */
      buf = dest;
   }

   return buf;
}

/**
 * Check if an unpack PBO is active prior to fetching a texture image.
 * If so, do bounds checking and map the buffer into main memory.
 * Any errors detected will be recorded.
 * The caller _must_ call _mesa_unmap_teximage_pbo() too!
 */
const GLvoid *
_mesa_validate_pbo_teximage(struct gl_context *ctx, GLuint dimensions,
			    GLsizei width, GLsizei height, GLsizei depth,
			    GLenum format, GLenum type, const GLvoid *pixels,
			    const struct gl_pixelstore_attrib *unpack,
			    const char *funcName)
{
   GLubyte *buf;

   if (!_mesa_is_bufferobj(unpack->BufferObj)) {
      /* no PBO */
      return pixels;
   }
   if (!_mesa_validate_pbo_access(dimensions, unpack, width, height, depth,
                                  format, type, INT_MAX, pixels)) {
      _mesa_error(ctx, GL_INVALID_OPERATION, "%s%uD(invalid PBO access)",
                  funcName, dimensions);
      return NULL;
   }

   buf = (GLubyte *) ctx->Driver.MapBufferRange(ctx, 0,
                                                unpack->BufferObj->Size,
						GL_MAP_READ_BIT,
						unpack->BufferObj,
                                                MAP_INTERNAL);
   if (!buf) {
      _mesa_error(ctx, GL_INVALID_OPERATION, "%s%uD(PBO is mapped)", funcName,
                  dimensions);
      return NULL;
   }

   return ADD_POINTERS(buf, pixels);
}

/**
 * Check if an unpack PBO is active prior to fetching a compressed texture
 * image.
 * If so, do bounds checking and map the buffer into main memory.
 * Any errors detected will be recorded.
 * The caller _must_ call _mesa_unmap_teximage_pbo() too!
 */
const GLvoid *
_mesa_validate_pbo_compressed_teximage(struct gl_context *ctx,
                                 GLsizei imageSize, const GLvoid *pixels,
                                 const struct gl_pixelstore_attrib *packing,
                                 const char *funcName)
{
   GLubyte *buf;

   if (!_mesa_is_bufferobj(packing->BufferObj)) {
      /* not using a PBO - return pointer unchanged */
      return pixels;
   }
   if ((const GLubyte *) pixels + imageSize >
       ((const GLubyte *) 0) + packing->BufferObj->Size) {
      /* out of bounds read! */
      _mesa_error(ctx, GL_INVALID_OPERATION, funcName, "(invalid PBO access)");
      return NULL;
   }

   buf = (GLubyte*) ctx->Driver.MapBufferRange(ctx, 0,
					       packing->BufferObj->Size,
					       GL_MAP_READ_BIT,
					       packing->BufferObj);
   if (!buf) {
      _mesa_error(ctx, GL_INVALID_OPERATION, funcName, "(PBO is mapped");
      return NULL;
   }

   return ADD_POINTERS(buf, pixels);
}

/* Unlike the other intel_pixel_* functions, the expectation here is
 * that the incoming data is not in a PBO.  With the XY_TEXT blit
 * method, there's no benefit haveing it in a PBO, but we could
 * implement a path based on XY_MONO_SRC_COPY_BLIT which might benefit
 * PBO bitmaps.  I think they are probably pretty rare though - I
 * wonder if Xgl uses them?
 */
static const GLubyte *map_pbo( struct gl_context *ctx,
			       GLsizei width, GLsizei height,
			       const struct gl_pixelstore_attrib *unpack,
			       const GLubyte *bitmap )
{
   GLubyte *buf;

   if (!_mesa_validate_pbo_access(2, unpack, width, height, 1,
				  GL_COLOR_INDEX, GL_BITMAP,
				  INT_MAX, (const GLvoid *) bitmap)) {
      _mesa_error(ctx, GL_INVALID_OPERATION,"glBitmap(invalid PBO access)");
      return NULL;
   }

   buf = (GLubyte *) ctx->Driver.MapBufferRange(ctx, 0, unpack->BufferObj->Size,
						GL_MAP_READ_BIT,
						unpack->BufferObj,
                                                MAP_INTERNAL);
   if (!buf) {
      _mesa_error(ctx, GL_INVALID_OPERATION, "glBitmap(PBO is mapped)");
      return NULL;
   }

   return ADD_POINTERS(buf, bitmap);
}

/**
 * Check if an unpack PBO is active prior to fetching a compressed texture
 * image.
 * If so, do bounds checking and map the buffer into main memory.
 * Any errors detected will be recorded.
 * The caller _must_ call _mesa_unmap_teximage_pbo() too!
 */
const GLvoid *
_mesa_validate_pbo_compressed_teximage(struct gl_context *ctx,
                                 GLuint dimensions, GLsizei imageSize,
                                 const GLvoid *pixels,
                                 const struct gl_pixelstore_attrib *packing,
                                 const char *funcName)
{
   GLubyte *buf;

   if (!_mesa_validate_pbo_source_compressed(ctx, dimensions, packing,
                                             imageSize, pixels, funcName)) {
     /* error is already set during validation */
      return NULL;
   }

   if (!_mesa_is_bufferobj(packing->BufferObj)) {
      /* not using a PBO - return pointer unchanged */
      return pixels;
   }

   buf = (GLubyte*) ctx->Driver.MapBufferRange(ctx, 0,
					       packing->BufferObj->Size,
					       GL_MAP_READ_BIT,
					       packing->BufferObj,
                                               MAP_INTERNAL);

   /* Validation above already checked that PBO is not mapped, so buffer
    * should not be null.
    */
   assert(buf);

   return ADD_POINTERS(buf, pixels);
}

/**
 * This is the software fallback for Driver.GetCompressedTexImage().
 * All error checking will have been done before this routine is called.
 */
void
_mesa_get_compressed_teximage(struct gl_context *ctx,
                              struct gl_texture_image *texImage,
                              GLvoid *img)
{
   const GLuint row_stride =
      _mesa_format_row_stride(texImage->TexFormat, texImage->Width);
   GLuint i;
   GLubyte *src;
   GLint srcRowStride;

   if (_mesa_is_bufferobj(ctx->Pack.BufferObj)) {
      /* pack texture image into a PBO */
      GLubyte *buf = (GLubyte *)
         ctx->Driver.MapBufferRange(ctx, 0, ctx->Pack.BufferObj->Size,
				    GL_MAP_WRITE_BIT, ctx->Pack.BufferObj);
      if (!buf) {
         /* out of memory or other unexpected error */
         _mesa_error(ctx, GL_OUT_OF_MEMORY,
                     "glGetCompresssedTexImage(map PBO failed)");
         return;
      }
      img = ADD_POINTERS(buf, img);
   }

   /* map src texture buffer */
   ctx->Driver.MapTextureImage(ctx, texImage, 0,
                               0, 0, texImage->Width, texImage->Height,
                               GL_MAP_READ_BIT, &src, &srcRowStride);

   if (src) {
      /* no pixelstore or pixel transfer, but respect stride */

      if (row_stride == srcRowStride) {
         const GLuint size = _mesa_format_image_size(texImage->TexFormat,
                                                     texImage->Width,
                                                     texImage->Height,
                                                     texImage->Depth);
         memcpy(img, src, size);
      }
      else {
         GLuint bw, bh;
         _mesa_get_format_block_size(texImage->TexFormat, &bw, &bh);
         for (i = 0; i < (texImage->Height + bh - 1) / bh; i++) {
            memcpy((GLubyte *)img + i * row_stride,
                   (GLubyte *)src + i * srcRowStride,
                   row_stride);
         }
      }

      ctx->Driver.UnmapTextureImage(ctx, texImage, 0);
   }
   else {
      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glGetCompresssedTexImage");
   }

   if (_mesa_is_bufferobj(ctx->Pack.BufferObj)) {
      ctx->Driver.UnmapBuffer(ctx, ctx->Pack.BufferObj);
   }
}

/**
 * This is the software fallback for Driver.GetTexImage().
 * All error checking will have been done before this routine is called.
 * We'll call ctx->Driver.MapTextureImage() to access the data, then
 * unmap with ctx->Driver.UnmapTextureImage().
 */
void
_mesa_get_teximage(struct gl_context *ctx,
                   GLenum format, GLenum type, GLvoid *pixels,
                   struct gl_texture_image *texImage)
{
   GLuint dimensions;

   switch (texImage->TexObject->Target) {
   case GL_TEXTURE_1D:
      dimensions = 1;
      break;
   case GL_TEXTURE_3D:
      dimensions = 3;
      break;
   default:
      dimensions = 2;
   }

   /* map dest buffer, if PBO */
   if (_mesa_is_bufferobj(ctx->Pack.BufferObj)) {
      /* Packing texture image into a PBO.
       * Map the (potentially) VRAM-based buffer into our process space so
       * we can write into it with the code below.
       * A hardware driver might use a sophisticated blit to move the
       * texture data to the PBO if the PBO is in VRAM along with the texture.
       */
      GLubyte *buf = (GLubyte *)
         ctx->Driver.MapBufferRange(ctx, 0, ctx->Pack.BufferObj->Size,
				    GL_MAP_WRITE_BIT, ctx->Pack.BufferObj);
      if (!buf) {
         /* out of memory or other unexpected error */
         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glGetTexImage(map PBO failed)");
         return;
      }
      /* <pixels> was an offset into the PBO.
       * Now make it a real, client-side pointer inside the mapped region.
       */
      pixels = ADD_POINTERS(buf, pixels);
   }

   if (get_tex_memcpy(ctx, format, type, pixels, texImage)) {
      /* all done */
   }
   else if (format == GL_DEPTH_COMPONENT) {
      get_tex_depth(ctx, dimensions, format, type, pixels, texImage);
   }
   else if (format == GL_DEPTH_STENCIL_EXT) {
      get_tex_depth_stencil(ctx, dimensions, format, type, pixels, texImage);
   }
   else if (format == GL_YCBCR_MESA) {
      get_tex_ycbcr(ctx, dimensions, format, type, pixels, texImage);
   }
   else {
      get_tex_rgba(ctx, dimensions, format, type, pixels, texImage);
   }

   if (_mesa_is_bufferobj(ctx->Pack.BufferObj)) {
      ctx->Driver.UnmapBuffer(ctx, ctx->Pack.BufferObj);
   }
}

static void GLAPIENTRY
vbo_exec_DrawElements(GLenum mode, GLsizei count, GLenum type, const GLvoid *indices)
{
   GET_CURRENT_CONTEXT(ctx);
   GLuint min_index = 0;
   GLuint max_index = 0;

   if (!_mesa_validate_DrawElements( ctx, mode, count, type, indices ))
      return;

   if (!vbo_validate_shaders(ctx)) {
      _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawElements(bad shader)");
      return;
   }

   if (ctx->Array.ElementArrayBufferObj->Name) {
      const GLvoid *map = ctx->Driver.MapBuffer(ctx,
						 GL_ELEMENT_ARRAY_BUFFER_ARB,
						 GL_READ_ONLY,
						 ctx->Array.ElementArrayBufferObj);

      get_minmax_index(count, type, ADD_POINTERS(map, indices), &min_index, &max_index);

      ctx->Driver.UnmapBuffer(ctx,
			      GL_ELEMENT_ARRAY_BUFFER_ARB,
			      ctx->Array.ElementArrayBufferObj);
   }
   else {
      get_minmax_index(count, type, indices, &min_index, &max_index);
   }

   vbo_exec_DrawRangeElements(mode, min_index, max_index, count, type, indices);
}

static void
vbo_bind_vertices(struct gl_context *ctx, const struct gl_client_array **arrays,
		  int base, unsigned min_index, unsigned max_index, int *pdelta)
{
	struct nouveau_render_state *render = to_render_state(ctx);
	struct nouveau_pushbuf *push = context_push(ctx);
	struct nouveau_bo *bo[NUM_VERTEX_ATTRS];
	unsigned offset[NUM_VERTEX_ATTRS];
	GLboolean dirty = GL_FALSE;
	int i, j, attr;
	RENDER_LOCALS(ctx);

	*pdelta = -1;

	FOR_EACH_BOUND_ATTR(render, i, attr) {
		const struct gl_client_array *array = arrays[attr];
		struct gl_buffer_object *obj = array->BufferObj;
		struct nouveau_array *a = &render->attrs[attr];
		unsigned delta = (base + min_index) * array->StrideB;

		bo[i] = NULL;

		if (nouveau_bufferobj_hw(obj)) {
			/* Array in a buffer obj. */
			nouveau_bo_ref(to_nouveau_bufferobj(obj)->bo, &bo[i]);
			offset[i] = delta + (intptr_t)array->Ptr;

		} else {
			int n = max_index - min_index + 1;
			char *sp = (char *)ADD_POINTERS(
				nouveau_bufferobj_sys(obj), array->Ptr) + delta;
			char *dp  = nouveau_get_scratch(ctx, n * a->stride,
							&bo[i], &offset[i]);

			/* Array in client memory, move it to a
			 * scratch buffer obj. */
			for (j = 0; j < n; j++)
				memcpy(dp + j * a->stride,
				       sp + j * array->StrideB,
				       a->stride);
		}

		dirty |= check_update_array(a, offset[i], bo[i], pdelta);
	}

	*pdelta -= min_index;

	if (dirty) {
		/* Buffers changed, update the attribute binding. */
		FOR_EACH_BOUND_ATTR(render, i, attr) {
			struct nouveau_array *a = &render->attrs[attr];

			nouveau_bo_ref(NULL, &a->bo);
			a->offset = offset[i];
			a->bo = bo[i];
		}

		TAG(render_release_vertices)(ctx);
		TAG(render_bind_vertices)(ctx);
	} else {

/**
 * For commands that read from a PBO (glDrawPixels, glTexImage,
 * glPolygonStipple, etc), if we're reading from a PBO, map it read-only
 * and return the pointer into the PBO.  If we're not reading from a
 * PBO, return \p src as-is.
 * If non-null return, must call _mesa_unmap_pbo_source() when done.
 *
 * \return NULL if error, else pointer to start of data
 */
const GLvoid *
_mesa_map_pbo_source(struct gl_context *ctx,
                     const struct gl_pixelstore_attrib *unpack,
                     const GLvoid *src)
{
   const GLubyte *buf;

   if (_mesa_is_bufferobj(unpack->BufferObj)) {
      /* unpack from PBO */
      buf = (GLubyte *) ctx->Driver.MapBufferRange(ctx, 0,
						   unpack->BufferObj->Size,
						   GL_MAP_READ_BIT,
						   unpack->BufferObj,
                                                   MAP_INTERNAL);
      if (!buf)
         return NULL;

      buf = ADD_POINTERS(buf, src);
   }
   else {
      /* unpack from normal memory */
      buf = src;
   }

   return buf;
}

void GLAPIENTRY
_mesa_PixelMapfv( GLenum map, GLsizei mapsize, const GLfloat *values )
{
   GET_CURRENT_CONTEXT(ctx);
   ASSERT_OUTSIDE_BEGIN_END(ctx);

   /* XXX someday, test against ctx->Const.MaxPixelMapTableSize */
   if (mapsize < 1 || mapsize > MAX_PIXEL_MAP_TABLE) {
      _mesa_error( ctx, GL_INVALID_VALUE, "glPixelMapfv(mapsize)" );
      return;
   }

   if (map >= GL_PIXEL_MAP_S_TO_S && map <= GL_PIXEL_MAP_I_TO_A) {
      /* test that mapsize is a power of two */
      if (_mesa_bitcount((GLuint) mapsize) != 1) {
	 _mesa_error( ctx, GL_INVALID_VALUE, "glPixelMapfv(mapsize)" );
         return;
      }
   }

   FLUSH_VERTICES(ctx, _NEW_PIXEL);

   if (ctx->Unpack.BufferObj->Name) {
      /* unpack pixelmap from PBO */
      GLubyte *buf;
      /* Note, need to use DefaultPacking and Unpack's buffer object */
      ctx->DefaultPacking.BufferObj = ctx->Unpack.BufferObj;
      if (!_mesa_validate_pbo_access(1, &ctx->DefaultPacking, mapsize, 1, 1,
                                     GL_INTENSITY, GL_FLOAT, values)) {
         _mesa_error(ctx, GL_INVALID_OPERATION,
                     "glPixelMapfv(invalid PBO access)");
         return;
      }
      /* restore */
      ctx->DefaultPacking.BufferObj = ctx->Array.NullBufferObj;
      buf = (GLubyte *) ctx->Driver.MapBuffer(ctx, GL_PIXEL_UNPACK_BUFFER_EXT,
                                              GL_READ_ONLY_ARB,
                                              ctx->Unpack.BufferObj);
      if (!buf) {
         /* buffer is already mapped - that's an error */
         _mesa_error(ctx, GL_INVALID_OPERATION,
                     "glPixelMapfv(PBO is mapped)");
         return;
      }
      values = (const GLfloat *) ADD_POINTERS(buf, values);
   }
   else if (!values) {
      return;
   }

   store_pixelmap(ctx, map, mapsize, values);

   if (ctx->Unpack.BufferObj->Name) {
      ctx->Driver.UnmapBuffer(ctx, GL_PIXEL_UNPACK_BUFFER_EXT,
                              ctx->Unpack.BufferObj);
   }
}

/* Translate indices to GLuints and store in VB->Elts.
 */
static void bind_indices( struct gl_context *ctx,
			  const struct _mesa_index_buffer *ib,
			  struct gl_buffer_object **bo,
			  GLuint *nr_bo)
{
   TNLcontext *tnl = TNL_CONTEXT(ctx);
   struct vertex_buffer *VB = &tnl->vb;
   GLuint i;
   const void *ptr;

   if (!ib) {
      VB->Elts = NULL;
      return;
   }

   if (_mesa_is_bufferobj(ib->obj) &&
       !_mesa_bufferobj_mapped(ib->obj, MAP_INTERNAL)) {
      /* if the buffer object isn't mapped yet, map it now */
      bo[*nr_bo] = ib->obj;
      (*nr_bo)++;
      ptr = ctx->Driver.MapBufferRange(ctx, (GLsizeiptr) ib->ptr,
                                       ib->count * vbo_sizeof_ib_type(ib->type),
				       GL_MAP_READ_BIT, ib->obj,
                                       MAP_INTERNAL);
      assert(ib->obj->Mappings[MAP_INTERNAL].Pointer);
   } else {
      /* user-space elements, or buffer already mapped */
      ptr = ADD_POINTERS(ib->obj->Mappings[MAP_INTERNAL].Pointer, ib->ptr);
   }

   if (ib->type == GL_UNSIGNED_INT && VB->Primitive[0].basevertex == 0) {
      VB->Elts = (GLuint *) ptr;
   }
   else {
      GLuint *elts = (GLuint *)get_space(ctx, ib->count * sizeof(GLuint));
      VB->Elts = elts;

      if (ib->type == GL_UNSIGNED_INT) {
	 const GLuint *in = (GLuint *)ptr;
	 for (i = 0; i < ib->count; i++)
	    *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
      }
      else if (ib->type == GL_UNSIGNED_SHORT) {
	 const GLushort *in = (GLushort *)ptr;
	 for (i = 0; i < ib->count; i++) 
	    *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
      }
      else {
	 const GLubyte *in = (GLubyte *)ptr;
	 for (i = 0; i < ib->count; i++) 
	    *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
      }
   }
}

/**
 * Called via glArrayElement() and glDrawArrays().
 * Issue the glNormal, glVertex, glColor, glVertexAttrib, etc functions
 * for all enabled vertex arrays (for elt-th element).
 * Note: this may be called during display list construction.
 */
void GLAPIENTRY _ae_ArrayElement( GLint elt )
{
   GET_CURRENT_CONTEXT(ctx);
   const AEcontext *actx = AE_CONTEXT(ctx);
   const AEarray *aa;
   const AEattrib *at;
   const struct _glapi_table * const disp = GET_DISPATCH();
   GLboolean do_map;

   if (actx->NewState) {
      assert(!actx->mapped_vbos);
      _ae_update_state( ctx );
   }

   /* Determine if w need to map/unmap VBOs */
   do_map = actx->nr_vbos && !actx->mapped_vbos;

   if (do_map)
      _ae_map_vbos(ctx);
   
   /* emit generic attribute elements */
   for (at = actx->attribs; at->func; at++) {
      const GLubyte *src
         = ADD_POINTERS(at->array->BufferObj->Pointer, at->array->Ptr)
         + elt * at->array->StrideB;
      at->func( at->index, src );
   }

   /* emit conventional arrays elements */
   for (aa = actx->arrays; aa->offset != -1 ; aa++) {
      const GLubyte *src
         = ADD_POINTERS(aa->array->BufferObj->Pointer, aa->array->Ptr)
         + elt * aa->array->StrideB;
      CALL_by_offset( disp, (array_func), aa->offset, 
		      ((const void *) src) );
   }

   if (do_map)
      _ae_unmap_vbos(ctx);
}

/* Translate indices to GLuints and store in VB->Elts.
 */
static void bind_indices( struct gl_context *ctx,
			  const struct _mesa_index_buffer *ib,
			  struct gl_buffer_object **bo,
			  GLuint *nr_bo)
{
   TNLcontext *tnl = TNL_CONTEXT(ctx);
   struct vertex_buffer *VB = &tnl->vb;
   GLuint i;
   void *ptr;

   if (!ib) {
      VB->Elts = NULL;
      return;
   }

   if (ib->obj->Name && !ib->obj->Pointer) {
      bo[*nr_bo] = ib->obj;
      (*nr_bo)++;
      ctx->Driver.MapBuffer(ctx, 
			    GL_ELEMENT_ARRAY_BUFFER,
			    GL_READ_ONLY_ARB,
			    ib->obj);

      assert(ib->obj->Pointer);
   }

   ptr = ADD_POINTERS(ib->obj->Pointer, ib->ptr);

   if (ib->type == GL_UNSIGNED_INT && VB->Primitive[0].basevertex == 0) {
      VB->Elts = (GLuint *) ptr;
   }
   else {
      GLuint *elts = (GLuint *)get_space(ctx, ib->count * sizeof(GLuint));
      VB->Elts = elts;

      if (ib->type == GL_UNSIGNED_INT) {
	 const GLuint *in = (GLuint *)ptr;
	 for (i = 0; i < ib->count; i++)
	    *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
      }
      else if (ib->type == GL_UNSIGNED_SHORT) {
	 const GLushort *in = (GLushort *)ptr;
	 for (i = 0; i < ib->count; i++) 
	    *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
      }
      else {
	 const GLubyte *in = (GLubyte *)ptr;
	 for (i = 0; i < ib->count; i++) 
	    *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
      }
   }
}

static void r300AlignDataToDword(GLcontext *ctx, const struct gl_client_array *input, int count, struct vertex_attribute *attr)
{
	r300ContextPtr r300 = R300_CONTEXT(ctx);
	const int dst_stride = (input->StrideB + 3) & ~3;
	const int size = getTypeSize(input->Type) * input->Size * count;
	GLboolean mapped_named_bo = GL_FALSE;

	radeonAllocDmaRegion(&r300->radeon, &attr->bo, &attr->bo_offset, size, 32);

	radeon_bo_map(attr->bo, 1);

	if (!input->BufferObj->Pointer) {
		ctx->Driver.MapBuffer(ctx, GL_ARRAY_BUFFER, GL_READ_ONLY_ARB, input->BufferObj);
		mapped_named_bo = GL_TRUE;
	}

	radeon_print(RADEON_FALLBACKS, RADEON_IMPORTANT, "%s. Vertex alignment doesn't match hw requirements.\n", __func__);

	{
		GLvoid *src_ptr = ADD_POINTERS(input->BufferObj->Pointer, input->Ptr);
		GLvoid *dst_ptr = ADD_POINTERS(attr->bo->ptr, attr->bo_offset);
		int i;

		for (i = 0; i < count; ++i) {
			memcpy(dst_ptr, src_ptr, input->StrideB);
			src_ptr += input->StrideB;
			dst_ptr += dst_stride;
		}
	}

	if (mapped_named_bo) {
		ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER, input->BufferObj);
	}

	radeon_bo_unmap(attr->bo);
	attr->stride = dst_stride;
}

/**
 * This is the software fallback for Driver.GetCompressedTexImage().
 * All error checking will have been done before this routine is called.
 */
void
_mesa_get_compressed_teximage(GLcontext *ctx, GLenum target, GLint level,
                              GLvoid *img,
                              struct gl_texture_object *texObj,
                              struct gl_texture_image *texImage)
{
   const GLuint row_stride = _mesa_format_row_stride(texImage->TexFormat,
                                                     texImage->Width);
   const GLuint row_stride_stored = _mesa_format_row_stride(texImage->TexFormat,
                                                            texImage->RowStride);
   GLuint i;

   if (_mesa_is_bufferobj(ctx->Pack.BufferObj)) {
      /* pack texture image into a PBO */
      GLubyte *buf = (GLubyte *)
         ctx->Driver.MapBuffer(ctx, GL_PIXEL_PACK_BUFFER_EXT,
                               GL_WRITE_ONLY_ARB, ctx->Pack.BufferObj);
      if (!buf) {
         /* out of memory or other unexpected error */
         _mesa_error(ctx, GL_OUT_OF_MEMORY,
                     "glGetCompresssedTexImage(map PBO failed)");
         return;
      }
      img = ADD_POINTERS(buf, img);
   }

   /* no pixelstore or pixel transfer, but respect stride */

   if (row_stride == row_stride_stored) {
      const GLuint size = _mesa_format_image_size(texImage->TexFormat,
                                                  texImage->Width,
                                                  texImage->Height,
                                                  texImage->Depth);
      _mesa_memcpy(img, texImage->Data, size);
   }
   else {
      GLuint bw, bh;
      _mesa_get_format_block_size(texImage->TexFormat, &bw, &bh);
      for (i = 0; i < (texImage->Height + bh - 1) / bh; i++) {
         memcpy((GLubyte *)img + i * row_stride,
                (GLubyte *)texImage->Data + i * row_stride_stored, row_stride);
      }
   }

   if (_mesa_is_bufferobj(ctx->Pack.BufferObj)) {
      ctx->Driver.UnmapBuffer(ctx, GL_PIXEL_PACK_BUFFER_EXT,
                              ctx->Pack.BufferObj);
   }
}

/**
 * This is the software fallback for Driver.GetCompressedTexImage().
 * All error checking will have been done before this routine is called.
 */
void
_mesa_get_compressed_teximage(GLcontext *ctx, GLenum target, GLint level,
                              GLvoid *img,
                              struct gl_texture_object *texObj,
                              struct gl_texture_image *texImage)
{
   GLuint size;

   if (ctx->Pack.BufferObj->Name) {
      /* pack texture image into a PBO */
      GLubyte *buf;
      if ((const GLubyte *) img + texImage->CompressedSize >
          (const GLubyte *) ctx->Pack.BufferObj->Size) {
         _mesa_error(ctx, GL_INVALID_OPERATION,
                     "glGetCompressedTexImage(invalid PBO access)");
         return;
      }
      buf = (GLubyte *) ctx->Driver.MapBuffer(ctx, GL_PIXEL_PACK_BUFFER_EXT,
                                              GL_WRITE_ONLY_ARB,
                                              ctx->Pack.BufferObj);
      if (!buf) {
         /* buffer is already mapped - that's an error */
         _mesa_error(ctx, GL_INVALID_OPERATION,
                     "glGetCompressedTexImage(PBO is mapped)");
         return;
      }
      img = ADD_POINTERS(buf, img);
   }
   else if (!img) {
      /* not an error */
      return;
   }

   /* don't use texImage->CompressedSize since that may be padded out */
   size = _mesa_compressed_texture_size(ctx, texImage->Width, texImage->Height,
                                        texImage->Depth,
                                        texImage->TexFormat->MesaFormat);

   /* just memcpy, no pixelstore or pixel transfer */
   _mesa_memcpy(img, texImage->Data, size);

   if (ctx->Pack.BufferObj->Name) {
      ctx->Driver.UnmapBuffer(ctx, GL_PIXEL_PACK_BUFFER_EXT,
                              ctx->Pack.BufferObj);
   }
}

/**
 * Find the max index in the given element/index buffer
 */
GLuint
_mesa_max_buffer_index(struct gl_context *ctx, GLuint count, GLenum type,
                       const void *indices,
                       struct gl_buffer_object *elementBuf)
{
   const GLubyte *map = NULL;
   GLuint max = 0;
   GLuint i;

   if (_mesa_is_bufferobj(elementBuf)) {
      /* elements are in a user-defined buffer object.  need to map it */
      map = ctx->Driver.MapBufferRange(ctx, 0, elementBuf->Size,
				       GL_MAP_READ_BIT, elementBuf);
      /* Actual address is the sum of pointers */
      indices = (const GLvoid *) ADD_POINTERS(map, (const GLubyte *) indices);
   }

   if (type == GL_UNSIGNED_INT) {
      for (i = 0; i < count; i++)
         if (((GLuint *) indices)[i] > max)
            max = ((GLuint *) indices)[i];
   }
   else if (type == GL_UNSIGNED_SHORT) {
      for (i = 0; i < count; i++)
         if (((GLushort *) indices)[i] > max)
            max = ((GLushort *) indices)[i];
   }
   else {
      ASSERT(type == GL_UNSIGNED_BYTE);
      for (i = 0; i < count; i++)
         if (((GLubyte *) indices)[i] > max)
            max = ((GLubyte *) indices)[i];
   }

   if (map) {
      ctx->Driver.UnmapBuffer(ctx, elementBuf);
   }

   return max;
}