vbo: Avoid extra validation of DrawElements.

This saves mapping the index buffer to get a bounds on the indices that
drivers just drop on the floor in the VBO case (cache win), saves a bonus
walk of the indices in the CheckArrayBounds case, and other miscellaneous
validation.  On intel it's a particularly a large win (50-100% in my app)
because even though we let the indices stay in both CPU and GPU caches, we
still end up waiting for the GPU to be done with the buffer before reading
from it.

Drivers that want the min/max_index fields must now check index_bounds_valid
and use vbo_get_minmax_index before using them.

17 files changed, 156 insertions, 117 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_draw.c b/src/mesa/drivers/dri/i965/brw_draw.c
index 5152c3f3a5f..8c94c904c1f 100644
--- a/src/mesa/drivers/dri/i965/brw_draw.c
+++ b/src/mesa/drivers/dri/i965/brw_draw.c
@@ -422,54 +422,31 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx,
    return retval;
 }
 
-static GLboolean brw_need_rebase( GLcontext *ctx,
-				  const struct gl_client_array *arrays[],
-				  const struct _mesa_index_buffer *ib,
-				  GLuint min_index )
-{
-   if (min_index == 0) 
-      return GL_FALSE;
-
-   if (ib) {
-      if (!vbo_all_varyings_in_vbos(arrays))
-	 return GL_TRUE;
-      else
-	 return GL_FALSE;
-   }
-   else {
-      /* Hmm.  This isn't quite what I wanted.  BRW can actually
-       * handle the mixed case well enough that we shouldn't need to
-       * rebase.  However, it's probably not very common, nor hugely
-       * expensive to do it this way:
-       */
-      if (!vbo_all_varyings_in_vbos(arrays))
-	 return GL_TRUE;
-      else
-	 return GL_FALSE;
-   }
-}
-				  
-
 void brw_draw_prims( GLcontext *ctx,
 		     const struct gl_client_array *arrays[],
 		     const struct _mesa_prim *prim,
 		     GLuint nr_prims,
 		     const struct _mesa_index_buffer *ib,
+		     GLboolean index_bounds_valid,
 		     GLuint min_index,
 		     GLuint max_index )
 {
    GLboolean retval;
 
-   /* Decide if we want to rebase.  If so we end up recursing once
-    * only into this function.
-    */
-   if (brw_need_rebase( ctx, arrays, ib, min_index )) {
-      vbo_rebase_prims( ctx, arrays, 
-			prim, nr_prims, 
-			ib, min_index, max_index, 
-			brw_draw_prims );
-      
-      return;
+   if (!vbo_all_varyings_in_vbos(arrays)) {
+      if (!index_bounds_valid)
+	 vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
+
+      /* Decide if we want to rebase.  If so we end up recursing once
+       * only into this function.
+       */
+      if (min_index != 0) {
+	 vbo_rebase_prims(ctx, arrays,
+			  prim, nr_prims,
+			  ib, min_index, max_index,
+			  brw_draw_prims );
+	 return;
+      }
    }
 
    /* Make a first attempt at drawing:
diff --git a/src/mesa/drivers/dri/i965/brw_draw.h b/src/mesa/drivers/dri/i965/brw_draw.h
index 9aebbdb1b86..2a14db217fc 100644
--- a/src/mesa/drivers/dri/i965/brw_draw.h
+++ b/src/mesa/drivers/dri/i965/brw_draw.h
@@ -39,6 +39,7 @@ void brw_draw_prims( GLcontext *ctx,
 		     const struct _mesa_prim *prims,
 		     GLuint nr_prims,
 		     const struct _mesa_index_buffer *ib,
+		     GLboolean index_bounds_valid,
 		     GLuint min_index,
 		     GLuint max_index );
 
diff --git a/src/mesa/drivers/dri/r300/r300_draw.c b/src/mesa/drivers/dri/r300/r300_draw.c
index fcfd3099332..aedc6cfb2a1 100644
--- a/src/mesa/drivers/dri/r300/r300_draw.c
+++ b/src/mesa/drivers/dri/r300/r300_draw.c
@@ -462,6 +462,7 @@ static void r300DrawPrims(GLcontext *ctx,
 			 const struct _mesa_prim *prim,
 			 GLuint nr_prims,
 			 const struct _mesa_index_buffer *ib,
+			 GLboolean index_bounds_valid,
 			 GLuint min_index,
 			 GLuint max_index)
 {
@@ -476,6 +477,12 @@ static void r300DrawPrims(GLcontext *ctx,
 	limits.max_indices = 65535;
 	limits.max_vb_size = 1024*1024;
 
+	/* This check should get folded into just the places that
+	 * min/max index are really needed.
+	 */
+	if (!index_bounds_valid)
+	   vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
+
 	if (min_index) {
 		vbo_rebase_prims( ctx, arrays, prim, nr_prims, ib, min_index, max_index, r300DrawPrims );
 		return;
diff --git a/src/mesa/state_tracker/st_cb_rasterpos.c b/src/mesa/state_tracker/st_cb_rasterpos.c
index 3bcccd0df46..d82b2a2035f 100644
--- a/src/mesa/state_tracker/st_cb_rasterpos.c
+++ b/src/mesa/state_tracker/st_cb_rasterpos.c
@@ -251,7 +251,7 @@ st_RasterPos(GLcontext *ctx, const GLfloat v[4])
    rs->array[0].Ptr = (GLubyte *) v;
 
    /* draw the point */
-   st_feedback_draw_vbo(ctx, rs->arrays, &rs->prim, 1, NULL, 0, 1);
+   st_feedback_draw_vbo(ctx, rs->arrays, &rs->prim, 1, NULL, GL_TRUE, 0, 1);
 }
 
 
diff --git a/src/mesa/state_tracker/st_draw.c b/src/mesa/state_tracker/st_draw.c
index 914a507bef6..503a5f34a3f 100644
--- a/src/mesa/state_tracker/st_draw.c
+++ b/src/mesa/state_tracker/st_draw.c
@@ -533,6 +533,7 @@ st_draw_vbo(GLcontext *ctx,
             const struct _mesa_prim *prims,
             GLuint nr_prims,
             const struct _mesa_index_buffer *ib,
+	    GLboolean index_bounds_valid,
             GLuint min_index,
             GLuint max_index)
 {
@@ -545,6 +546,10 @@ st_draw_vbo(GLcontext *ctx,
    unsigned num_vbuffers, num_velements;
    GLboolean userSpace;
 
+   /* Gallium probably doesn't want this in some cases. */
+   if (!index_bounds_valid)
+      vbo_get_minmax_index(ctx, prims, ib, &min_index, &max_index);
+
    /* sanity check for pointer arithmetic below */
    assert(sizeof(arrays[0]->Ptr[0]) == 1);
 
diff --git a/src/mesa/state_tracker/st_draw.h b/src/mesa/state_tracker/st_draw.h
index dcfe7e15361..3e0face656b 100644
--- a/src/mesa/state_tracker/st_draw.h
+++ b/src/mesa/state_tracker/st_draw.h
@@ -47,6 +47,7 @@ st_draw_vbo(GLcontext *ctx,
             const struct _mesa_prim *prims,
             GLuint nr_prims,
             const struct _mesa_index_buffer *ib,
+	    GLboolean index_bounds_valid,
             GLuint min_index,
             GLuint max_index);
 
@@ -56,6 +57,7 @@ st_feedback_draw_vbo(GLcontext *ctx,
                      const struct _mesa_prim *prims,
                      GLuint nr_prims,
                      const struct _mesa_index_buffer *ib,
+		     GLboolean index_bounds_valid,
                      GLuint min_index,
                      GLuint max_index);
 
diff --git a/src/mesa/state_tracker/st_draw_feedback.c b/src/mesa/state_tracker/st_draw_feedback.c
index 2712c131c0d..b2d682ef640 100644
--- a/src/mesa/state_tracker/st_draw_feedback.c
+++ b/src/mesa/state_tracker/st_draw_feedback.c
@@ -96,6 +96,7 @@ st_feedback_draw_vbo(GLcontext *ctx,
                      const struct _mesa_prim *prims,
                      GLuint nr_prims,
                      const struct _mesa_index_buffer *ib,
+		     GLboolean index_bounds_valid,
                      GLuint min_index,
                      GLuint max_index)
 {
@@ -114,6 +115,9 @@ st_feedback_draw_vbo(GLcontext *ctx,
 
    st_validate_state(ctx->st);
 
+   if (!index_bounds_valid)
+      vbo_get_minmax_index(ctx, prims, ib, &min_index, &max_index);
+
    /* must get these after state validation! */
    vp = ctx->st->vp;
    vs = &st->vp->state;
diff --git a/src/mesa/tnl/t_context.c b/src/mesa/tnl/t_context.c
index f69b1220461..f2771cde095 100644
--- a/src/mesa/tnl/t_context.c
+++ b/src/mesa/tnl/t_context.c
@@ -81,7 +81,7 @@ _tnl_CreateContext( GLcontext *ctx )
    tnl->nr_blocks = 0;
 
    /* plug in the VBO drawing function */
-   vbo_set_draw_func(ctx, _tnl_draw_prims);
+   vbo_set_draw_func(ctx, _tnl_vbo_draw_prims);
 
    _math_init_transformation();
    _math_init_translate();
diff --git a/src/mesa/tnl/t_draw.c b/src/mesa/tnl/t_draw.c
index 2ec65d53233..c64c2c20778 100644
--- a/src/mesa/tnl/t_draw.c
+++ b/src/mesa/tnl/t_draw.c
@@ -360,6 +360,20 @@ static void unmap_vbos( GLcontext *ctx,
 }
 
 
+void _tnl_vbo_draw_prims(GLcontext *ctx,
+			 const struct gl_client_array *arrays[],
+			 const struct _mesa_prim *prim,
+			 GLuint nr_prims,
+			 const struct _mesa_index_buffer *ib,
+			 GLboolean index_bounds_valid,
+			 GLuint min_index,
+			 GLuint max_index)
+{
+   if (!index_bounds_valid)
+      vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
+
+   _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
+}
 
 /* This is the main entrypoint into the slimmed-down software tnl
  * module.  In a regular swtnl driver, this can be plugged straight
@@ -393,7 +407,7 @@ void _tnl_draw_prims( GLcontext *ctx,
        */
       vbo_rebase_prims( ctx, arrays, prim, nr_prims, ib, 
 			min_index, max_index,
-			_tnl_draw_prims );
+			_tnl_vbo_draw_prims );
       return;
    }
    else if (max_index > max) {
@@ -411,7 +425,7 @@ void _tnl_draw_prims( GLcontext *ctx,
        */
       vbo_split_prims( ctx, arrays, prim, nr_prims, ib, 
 		       0, max_index,
-		       _tnl_draw_prims,
+		       _tnl_vbo_draw_prims,
 		       &limits );
    }
    else {
diff --git a/src/mesa/tnl/tnl.h b/src/mesa/tnl/tnl.h
index 4d628aa9a60..9c66d3b0192 100644
--- a/src/mesa/tnl/tnl.h
+++ b/src/mesa/tnl/tnl.h
@@ -81,6 +81,16 @@ _tnl_draw_prims( GLcontext *ctx,
 		 GLuint min_index,
 		 GLuint max_index);
 
+void
+_tnl_vbo_draw_prims( GLcontext *ctx,
+		     const struct gl_client_array *arrays[],
+		     const struct _mesa_prim *prim,
+		     GLuint nr_prims,
+		     const struct _mesa_index_buffer *ib,
+		     GLboolean index_bounds_valid,
+		     GLuint min_index,
+		     GLuint max_index);
+
 extern void
 _mesa_load_tracked_matrices(GLcontext *ctx);
 
diff --git a/src/mesa/vbo/vbo.h b/src/mesa/vbo/vbo.h
index 5362226c2f9..5986e93576c 100644
--- a/src/mesa/vbo/vbo.h
+++ b/src/mesa/vbo/vbo.h
@@ -69,6 +69,7 @@ typedef void (*vbo_draw_func)( GLcontext *ctx,
 			       const struct _mesa_prim *prims,
 			       GLuint nr_prims,
 			       const struct _mesa_index_buffer *ib,
+			       GLboolean index_bounds_valid,
 			       GLuint min_index,
 			       GLuint max_index );
 
@@ -112,7 +113,10 @@ void vbo_rebase_prims( GLcontext *ctx,
 		       GLuint min_index,
 		       GLuint max_index,
 		       vbo_draw_func draw );
-
+void
+vbo_get_minmax_index(GLcontext *ctx, const struct _mesa_prim *prim,
+		     const struct _mesa_index_buffer *ib,
+		     GLuint *min_index, GLuint *max_index);
 
 void vbo_use_buffer_objects(GLcontext *ctx);
 
diff --git a/src/mesa/vbo/vbo_exec_array.c b/src/mesa/vbo/vbo_exec_array.c
index f4b9b2f7443..4fb4845f6b6 100644
--- a/src/mesa/vbo/vbo_exec_array.c
+++ b/src/mesa/vbo/vbo_exec_array.c
@@ -41,15 +41,29 @@
 /**
  * Compute min and max elements for glDraw[Range]Elements() calls.
  */
-static void
-get_minmax_index(GLuint count, GLuint type, const GLvoid *indices,
-                 GLuint *min_index, GLuint *max_index)
+void
+vbo_get_minmax_index(GLcontext *ctx,
+		     const struct _mesa_prim *prim,
+		     const struct _mesa_index_buffer *ib,
+		     GLuint *min_index, GLuint *max_index)
 {
    GLuint i;
+   GLsizei count = prim->count;
+   const void *indices;
+
+   if (ib->obj->Name) {
+      const GLvoid *map = ctx->Driver.MapBuffer(ctx,
+                                                GL_ELEMENT_ARRAY_BUFFER_ARB,
+                                                GL_READ_ONLY,
+                                                ib->obj);
+      indices = ADD_POINTERS(map, ib->ptr);
+   } else {
+      indices = ib->ptr;
+   }
 
-   switch(type) {
+   switch (ib->type) {
    case GL_UNSIGNED_INT: {
-      const GLuint *ui_indices = (const GLuint *)indices;
+      const GLuint *ui_indices = (const GLuint *)ib->ptr;
       GLuint max_ui = ui_indices[count-1];
       GLuint min_ui = ui_indices[0];
       for (i = 0; i < count; i++) {
@@ -88,6 +102,12 @@ get_minmax_index(GLuint count, GLuint type, const GLvoid *indices,
       assert(0);
       break;
    }
+
+   if (ib->obj->Name != 0) {
+      ctx->Driver.UnmapBuffer(ctx,
+			      GL_ELEMENT_ARRAY_BUFFER_ARB,
+			      ib->obj);
+   }
 }
 
 
@@ -500,7 +520,7 @@ vbo_exec_DrawArrays(GLenum mode, GLint start, GLsizei count)
    prim[0].indexed = 0;
 
    vbo->draw_prims( ctx, exec->array.inputs, prim, 1, NULL,
-                    start, start + count - 1 );
+                    GL_TRUE, start, start + count - 1 );
 
 #if 0
    print_draw_arrays(ctx, exec, mode, start, count);
@@ -566,53 +586,19 @@ dump_element_buffer(GLcontext *ctx, GLenum type)
                            ctx->Array.ElementArrayBufferObj);
 }
 
-
-static void GLAPIENTRY
-vbo_exec_DrawRangeElements(GLenum mode,
-			   GLuint start, GLuint end,
-			   GLsizei count, GLenum type, const GLvoid *indices)
+/* Inner support for both _mesa_DrawElements and _mesa_DrawRangeElements */
+static void
+vbo_validated_drawrangeelements(GLcontext *ctx, GLenum mode,
+				GLboolean index_bounds_valid,
+				GLuint start, GLuint end,
+				GLsizei count, GLenum type,
+				const GLvoid *indices)
 {
-   GET_CURRENT_CONTEXT(ctx);
    struct vbo_context *vbo = vbo_context(ctx);
    struct vbo_exec_context *exec = &vbo->exec;
    struct _mesa_index_buffer ib;
    struct _mesa_prim prim[1];
 
-   if (!_mesa_validate_DrawRangeElements( ctx, mode, start, end, count,
-                                          type, indices ))
-      return;
-
-   if (end >= ctx->Array.ArrayObj->_MaxElement) {
-      /* the max element is out of bounds of one or more enabled arrays */
-      _mesa_warning(ctx, "glDraw[Range]Elements(start %u, end %u, count %d, "
-                    "type 0x%x, indices=%p)\n"
-                    "\tindex=%u is out of bounds (max=%u)  "
-                    "Element Buffer %u (size %d)",
-                    start, end, count, type, indices, end,
-                    ctx->Array.ArrayObj->_MaxElement - 1,
-                    ctx->Array.ElementArrayBufferObj->Name,
-                    ctx->Array.ElementArrayBufferObj->Size);
-
-      if (0)
-         dump_element_buffer(ctx, type);
-
-      if (0)
-         _mesa_print_arrays(ctx);
-      return;
-   }
-   else if (0) {
-      _mesa_printf("glDraw[Range]Elements"
-                   "(start %u, end %u, type 0x%x, count %d) ElemBuf %u\n",
-                   start, end, type, count,
-                   ctx->Array.ElementArrayBufferObj->Name);
-   }
-
-#if 0
-   check_draw_elements_data(ctx, count, type, indices);
-#else
-   (void) check_draw_elements_data;
-#endif
-
    FLUSH_CURRENT( ctx, 0 );
 
    if (ctx->NewState)
@@ -623,13 +609,13 @@ vbo_exec_DrawRangeElements(GLenum mode,
       return;
    }
 
-   bind_arrays( ctx );
-
    if (ctx->NewState)
       _mesa_update_state( ctx );
 
+   bind_arrays( ctx );
+
    ib.count = count;
-   ib.type = type; 
+   ib.type = type;
    ib.obj = ctx->Array.ElementArrayBufferObj;
    ib.ptr = indices;
 
@@ -673,44 +659,68 @@ vbo_exec_DrawRangeElements(GLenum mode,
     * for the latter case elsewhere.
     */
 
-   vbo->draw_prims( ctx, exec->array.inputs, prim, 1, &ib, start, end );
+   vbo->draw_prims( ctx, exec->array.inputs, prim, 1, &ib,
+		    index_bounds_valid, start, end );
 }
 
-
 static void GLAPIENTRY
-vbo_exec_DrawElements(GLenum mode, GLsizei count, GLenum type,
-                      const GLvoid *indices)
+vbo_exec_DrawRangeElements(GLenum mode,
+			   GLuint start, GLuint end,
+			   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 ))
+   if (!_mesa_validate_DrawRangeElements( ctx, mode, start, end, count,
+                                          type, indices ))
       return;
 
-   if (!vbo_validate_shaders(ctx)) {
-      _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawElements(bad shader)");
+   if (end >= ctx->Array.ArrayObj->_MaxElement) {
+      /* the max element is out of bounds of one or more enabled arrays */
+      _mesa_warning(ctx, "glDraw[Range]Elements(start %u, end %u, count %d, "
+                    "type 0x%x, indices=%p)\n"
+                    "\tindex=%u is out of bounds (max=%u)  "
+                    "Element Buffer %u (size %d)",
+                    start, end, count, type, indices, end,
+                    ctx->Array.ArrayObj->_MaxElement - 1,
+                    ctx->Array.ElementArrayBufferObj->Name,
+                    ctx->Array.ElementArrayBufferObj->Size);
+
+      if (0)
+         dump_element_buffer(ctx, type);
+
+      if (0)
+         _mesa_print_arrays(ctx);
       return;
    }
+   else if (0) {
+      _mesa_printf("glDraw[Range]Elements"
+                   "(start %u, end %u, type 0x%x, count %d) ElemBuf %u\n",
+                   start, end, type, count,
+                   ctx->Array.ElementArrayBufferObj->Name);
+   }
 
-   if (ctx->Array.ElementArrayBufferObj->Name) {
-      const GLvoid *map = ctx->Driver.MapBuffer(ctx,
-                                                GL_ELEMENT_ARRAY_BUFFER_ARB,
-                                                GL_READ_ONLY,
-                                                ctx->Array.ElementArrayBufferObj);
+#if 0
+   check_draw_elements_data(ctx, count, type, indices);
+#else
+   (void) check_draw_elements_data;
+#endif
+
+   vbo_validated_drawrangeelements(ctx, mode, GL_TRUE, start, end,
+				   count, type, indices);
+}
 
-      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);
-   }
+static void GLAPIENTRY
+vbo_exec_DrawElements(GLenum mode, GLsizei count, GLenum type,
+                      const GLvoid *indices)
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if (!_mesa_validate_DrawElements( ctx, mode, count, type, indices ))
+      return;
 
-   vbo_exec_DrawRangeElements(mode, min_index, max_index, count, type, indices);
+   vbo_validated_drawrangeelements(ctx, mode, GL_FALSE, ~0, ~0,
+				   count, type, indices);
 }
 
 
diff --git a/src/mesa/vbo/vbo_exec_draw.c b/src/mesa/vbo/vbo_exec_draw.c
index 18419928b26..a988424d212 100644
--- a/src/mesa/vbo/vbo_exec_draw.c
+++ b/src/mesa/vbo/vbo_exec_draw.c
@@ -378,6 +378,7 @@ vbo_exec_vtx_flush( struct vbo_exec_context *exec, GLboolean unmap )
 				       exec->vtx.prim, 
 				       exec->vtx.prim_count,
 				       NULL,
+				       GL_TRUE,
 				       0,
 				       exec->vtx.vert_count - 1);
 
diff --git a/src/mesa/vbo/vbo_rebase.c b/src/mesa/vbo/vbo_rebase.c
index ea87dede646..3bf7ef580fc 100644
--- a/src/mesa/vbo/vbo_rebase.c
+++ b/src/mesa/vbo/vbo_rebase.c
@@ -208,6 +208,7 @@ void vbo_rebase_prims( GLcontext *ctx,
 	 prim, 
 	 nr_prims, 
 	 ib, 
+	 GL_TRUE,
 	 0, 
 	 max_index - min_index );
    
diff --git a/src/mesa/vbo/vbo_save_draw.c b/src/mesa/vbo/vbo_save_draw.c
index 5110648c282..d834fa1f2e7 100644
--- a/src/mesa/vbo/vbo_save_draw.c
+++ b/src/mesa/vbo/vbo_save_draw.c
@@ -279,6 +279,7 @@ void vbo_save_playback_vertex_list( GLcontext *ctx, void *data )
 				    node->prim, 
 				    node->prim_count,
 				    NULL,
+				    GL_TRUE,
 				    0,	/* Node is a VBO, so this is ok */
 				    node->count - 1);
    }
diff --git a/src/mesa/vbo/vbo_split_copy.c b/src/mesa/vbo/vbo_split_copy.c
index d7ffebf6076..3f8a222805d 100644
--- a/src/mesa/vbo/vbo_split_copy.c
+++ b/src/mesa/vbo/vbo_split_copy.c
@@ -194,6 +194,7 @@ flush( struct copy_context *copy )
 	       copy->dstprim,
 	       copy->dstprim_nr,
 	       &copy->dstib,
+	       GL_TRUE,
 	       0,
 	       copy->dstbuf_nr );
 
diff --git a/src/mesa/vbo/vbo_split_inplace.c b/src/mesa/vbo/vbo_split_inplace.c
index 9628227e7c4..da84eaa6ead 100644
--- a/src/mesa/vbo/vbo_split_inplace.c
+++ b/src/mesa/vbo/vbo_split_inplace.c
@@ -85,6 +85,7 @@ static void flush_vertex( struct split_context *split )
 		split->dstprim,
 		split->dstprim_nr,
 		NULL,
+		GL_TRUE,
 		min_index,
 		max_index);