vbo: cache last dlist vertex in malloced memory

Avoids repeated mapping of the VBO buffer on display list replay.  We
need access to the final vertex in order to update the GL current attrib
values.

3 files changed, 53 insertions, 10 deletions

diff --git a/src/mesa/vbo/vbo_save.h b/src/mesa/vbo/vbo_save.h
index 9558f838837..86bbd24f7b1 100644
--- a/src/mesa/vbo/vbo_save.h
+++ b/src/mesa/vbo/vbo_save.h
@@ -64,6 +64,13 @@ struct vbo_save_vertex_list {
    GLubyte attrsz[VBO_ATTRIB_MAX];
    GLuint vertex_size;
 
+   /* Copy of the final vertex from node->vertex_store->bufferobj.
+    * Keep this in regular (non-VBO) memory to avoid repeated
+    * map/unmap of the VBO when updating GL current data.
+    */
+   GLfloat *current_data;
+   GLuint current_size;
+
    GLuint buffer_offset;
    GLuint count;
    GLuint wrap_count;		/* number of copied vertices at start */
diff --git a/src/mesa/vbo/vbo_save_api.c b/src/mesa/vbo/vbo_save_api.c
index 52b6f1884ee..868226075ac 100644
--- a/src/mesa/vbo/vbo_save_api.c
+++ b/src/mesa/vbo/vbo_save_api.c
@@ -289,6 +289,31 @@ static void _save_compile_vertex_list( GLcontext *ctx )
    node->vertex_store->refcount++;
    node->prim_store->refcount++;
 
+
+   node->current_size = node->vertex_size - node->attrsz[0];
+   node->current_data = NULL;
+
+   if (node->current_size) {
+      /* If the malloc fails, we just pull the data out of the VBO
+       * later instead.
+       */
+      node->current_data = MALLOC( node->current_size * sizeof(GLfloat) );
+      if (node->current_data) {
+         const char *buffer = (const char *)save->vertex_store->buffer;
+         unsigned attr_offset = node->attrsz[0] * sizeof(GLfloat);
+         unsigned vertex_offset = 0;
+
+         if (node->count)
+            vertex_offset = (node->count-1) * node->vertex_size * sizeof(GLfloat);
+
+         memcpy( node->current_data,
+                 buffer + node->buffer_offset + vertex_offset + attr_offset,
+                 node->current_size * sizeof(GLfloat) );
+      }
+   }
+
+
+
    assert(node->attrsz[VBO_ATTRIB_POS] != 0 ||
 	  node->count == 0);
 
diff --git a/src/mesa/vbo/vbo_save_draw.c b/src/mesa/vbo/vbo_save_draw.c
index f59e1036d01..5110648c282 100644
--- a/src/mesa/vbo/vbo_save_draw.c
+++ b/src/mesa/vbo/vbo_save_draw.c
@@ -46,20 +46,31 @@ static void _playback_copy_to_current( GLcontext *ctx,
 				       const struct vbo_save_vertex_list *node )
 {
    struct vbo_context *vbo = vbo_context(ctx);
-   GLfloat vertex[VBO_ATTRIB_MAX * 4], *data = vertex;
+   GLfloat vertex[VBO_ATTRIB_MAX * 4];
+   GLfloat *data;
    GLuint i, offset;
 
-   if (node->count)
-      offset = (node->buffer_offset + 
-		(node->count-1) * node->vertex_size * sizeof(GLfloat));
-   else
-      offset = node->buffer_offset;
+   if (node->current_size == 0)
+      return;
 
-   ctx->Driver.GetBufferSubData( ctx, 0, offset, 
-				 node->vertex_size * sizeof(GLfloat), 
-				 data, node->vertex_store->bufferobj );
+   if (node->current_data) {
+      data = node->current_data;
+   }
+   else {
+      data = vertex;
+
+      if (node->count)
+         offset = (node->buffer_offset + 
+                   (node->count-1) * node->vertex_size * sizeof(GLfloat));
+      else
+         offset = node->buffer_offset;
 
-   data += node->attrsz[0]; /* skip vertex position */
+      ctx->Driver.GetBufferSubData( ctx, 0, offset, 
+                                    node->vertex_size * sizeof(GLfloat), 
+                                    data, node->vertex_store->bufferobj );
+
+      data += node->attrsz[0]; /* skip vertex position */
+   }
 
    for (i = VBO_ATTRIB_POS+1 ; i < VBO_ATTRIB_MAX ; i++) {
       if (node->attrsz[i]) {