Merge commit 'origin/gallium-0.1'

Conflicts:

	scons/gallium.py
	src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
	src/gallium/include/pipe/p_defines.h
	src/mesa/vbo/vbo_exec_api.c
	src/mesa/vbo/vbo_exec_draw.c

11 files changed, 303 insertions, 72 deletions

diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
index 533c0a15f3..a0a483ecaa 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
@@ -284,16 +284,26 @@ fenced_buffer_map(struct pb_buffer *buf,
                   unsigned flags)
 {
    struct fenced_buffer *fenced_buf = fenced_buffer(buf);
+   struct fenced_buffer_list *fenced_list = fenced_buf->list;
+   struct pipe_winsys *winsys = fenced_list->winsys;
    void *map;
 
-   assert(flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE);
-   assert(!(flags & ~PIPE_BUFFER_USAGE_CPU_READ_WRITE));
-   flags &= PIPE_BUFFER_USAGE_CPU_READ_WRITE;
-   
-   /* Check for GPU read/write access */
-   if(fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_WRITE) {
-      /* Wait for the GPU to finish writing */
-      _fenced_buffer_finish(fenced_buf);
+   assert(!(flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE));
+   
+   /* Serialize writes */
+   if((fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_WRITE) ||
+      ((fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_READ) && (flags & PIPE_BUFFER_USAGE_CPU_WRITE))) {
+      if(flags & PIPE_BUFFER_USAGE_DONTBLOCK) {
+         /* Don't wait for the GPU to finish writing */
+         if(winsys->fence_finish(winsys, fenced_buf->fence, 0) == 0)
+            _fenced_buffer_remove(fenced_list, fenced_buf);
+         else
+            return NULL;
+      }
+      else {
+         /* Wait for the GPU to finish writing */
+         _fenced_buffer_finish(fenced_buf);
+      }
    }
 
 #if 0
@@ -307,7 +317,7 @@ fenced_buffer_map(struct pb_buffer *buf,
    map = pb_map(fenced_buf->buffer, flags);
    if(map) {
       ++fenced_buf->mapcount;
-      fenced_buf->flags |= flags;
+      fenced_buf->flags |= flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE;
    }
 
    return map;
@@ -471,6 +481,51 @@ fenced_buffer_create(struct fenced_buffer_list *fenced_list,
 }
 
 
+<<<<<<< HEAD:src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
+=======
+void
+buffer_fence(struct pb_buffer *buf,
+             struct pipe_fence_handle *fence)
+{
+   struct fenced_buffer *fenced_buf;
+   struct fenced_buffer_list *fenced_list;
+   struct pipe_winsys *winsys;
+   /* FIXME: receive this as a parameter */
+   unsigned flags = fence ? PIPE_BUFFER_USAGE_GPU_READ_WRITE : 0;
+
+   /* This is a public function, so be extra cautious with the buffer passed, 
+    * as happens frequently to receive null buffers, or pointer to buffers 
+    * other than fenced buffers. */
+   assert(buf);
+   if(!buf)
+      return;
+   assert(buf->vtbl == &fenced_buffer_vtbl);
+   if(buf->vtbl != &fenced_buffer_vtbl)
+      return;
+   
+   if(!fence)
+      return;
+
+   fenced_buf = fenced_buffer(buf);
+   fenced_list = fenced_buf->list;
+   winsys = fenced_list->winsys;
+   
+   if(fence == fenced_buf->fence) {
+      fenced_buf->flags |= flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE;
+      return;
+   }
+   
+   pipe_mutex_lock(fenced_list->mutex);
+   if (fenced_buf->fence)
+      _fenced_buffer_remove(fenced_list, fenced_buf);
+   winsys->fence_reference(winsys, &fenced_buf->fence, fence);
+   fenced_buf->flags |= flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE;
+   _fenced_buffer_add(fenced_buf);
+   pipe_mutex_unlock(fenced_list->mutex);
+}
+
+
+>>>>>>> origin/gallium-0.1:src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
 struct fenced_buffer_list *
 fenced_buffer_list_create(struct pb_fence_ops *ops) 
 {
diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h
index 3cbc93d12b..52d443970b 100644
--- a/src/gallium/include/pipe/p_defines.h
+++ b/src/gallium/include/pipe/p_defines.h
@@ -215,6 +215,7 @@ enum pipe_transfer_usage {
 #define PIPE_BUFFER_USAGE_INDEX     (1 << 6)
 #define PIPE_BUFFER_USAGE_CONSTANT  (1 << 7)
 #define PIPE_BUFFER_USAGE_DISCARD   (1 << 8)
+#define PIPE_BUFFER_USAGE_DONTBLOCK (1 << 9)
 /** Pipe driver custom usage flags should be greater or equal to this value */
 #define PIPE_BUFFER_USAGE_CUSTOM    (1 << 16)
 
diff --git a/src/mesa/main/dd.h b/src/mesa/main/dd.h
index 411b6a7b21..a335f77479 100644
--- a/src/mesa/main/dd.h
+++ b/src/mesa/main/dd.h
@@ -36,6 +36,22 @@
 struct gl_pixelstore_attrib;
 struct gl_display_list;
 
+#if FEATURE_ARB_vertex_buffer_object
+/* Modifies GL_MAP_UNSYNCHRONIZED_BIT to allow driver to fail (return
+ * NULL) if buffer is unavailable for immediate mapping.
+ *
+ * Does GL_MAP_INVALIDATE_RANGE_BIT do this?  It seems so, but it
+ * would require more book-keeping in the driver than seems necessary
+ * at this point.
+ *
+ * Does GL_MAP_INVALDIATE_BUFFER_BIT do this?  Not really -- we don't
+ * want to provoke the driver to throw away the old storage, we will
+ * respect the contents of already referenced data.
+ */
+#define MESA_MAP_NOWAIT_BIT       0x0040
+#endif
+
+
 /**
  * Device driver function table.
  * Core Mesa uses these function pointers to call into device drivers.
@@ -785,6 +801,16 @@ struct dd_function_table {
    void * (*MapBuffer)( GLcontext *ctx, GLenum target, GLenum access,
 			struct gl_buffer_object *obj );
 
+   /* May return NULL if MESA_MAP_NOWAIT_BIT is set in access:
+    */
+   void * (*MapBufferRange)( GLcontext *ctx, GLenum target,
+                             GLintptr offset, GLsizeiptr length, GLbitfield access,
+                             struct gl_buffer_object *obj);
+
+   void (*FlushMappedBufferRange) (GLcontext *ctx, GLenum target, 
+                                   GLintptr offset, GLsizeiptr length,
+                                   struct gl_buffer_object *obj);
+
    GLboolean (*UnmapBuffer)( GLcontext *ctx, GLenum target,
 			     struct gl_buffer_object *obj );
    /*@}*/
@@ -954,6 +980,12 @@ struct dd_function_table {
    GLuint NeedFlush;
    GLuint SaveNeedFlush;
 
+
+   /* Called prior to any of the GLvertexformat functions being
+    * called.  Paired with Driver.FlushVertices().
+    */
+   void (*BeginVertices)( GLcontext *ctx );
+
    /**
     * If inside glBegin()/glEnd(), it should ASSERT(0).  Otherwise, if
     * FLUSH_STORED_VERTICES bit in \p flags is set flushes any buffered
diff --git a/src/mesa/main/vtxfmt.c b/src/mesa/main/vtxfmt.c
index 0204979003..1f807dc3dc 100644
--- a/src/mesa/main/vtxfmt.c
+++ b/src/mesa/main/vtxfmt.c
@@ -54,9 +54,12 @@
    ASSERT( tnl->Current );						\
    ASSERT( tnl->SwapCount < NUM_VERTEX_FORMAT_ENTRIES );		\
    ASSERT( tmp_offset >= 0 );						\
-									\
-   /* Save the swapped function's dispatch entry so it can be */	\
-   /* restored later. */						\
+                                                                        \
+   if (tnl->SwapCount == 0)                                             \
+      ctx->Driver.BeginVertices( ctx );                                 \
+                                                                        \
+   /* Save the swapped function's dispatch entry so it can be */        \
+   /* restored later. */                                                \
    tnl->Swapped[tnl->SwapCount].location = & (((_glapi_proc *)ctx->Exec)[tmp_offset]); \
    tnl->Swapped[tnl->SwapCount].function = (_glapi_proc)TAG(FUNC);	\
    tnl->SwapCount++;							\
diff --git a/src/mesa/state_tracker/st_cb_bufferobjects.c b/src/mesa/state_tracker/st_cb_bufferobjects.c
index 562ac6c65c..28e387c399 100644
--- a/src/mesa/state_tracker/st_cb_bufferobjects.c
+++ b/src/mesa/state_tracker/st_cb_bufferobjects.c
@@ -206,8 +206,40 @@ st_bufferobj_map(GLcontext *ctx, GLenum target, GLenum access,
 }
 
 
+
 /**
- * Called via glMapBufferARB().
+ * Called via glMapBufferRange().
+ */
+static void *
+st_bufferobj_map_range(GLcontext *ctx, GLenum target, 
+                       GLintptr offset, GLsizeiptr length, GLbitfield access,
+                       struct gl_buffer_object *obj)
+{
+   struct pipe_context *pipe = st_context(ctx)->pipe;
+   struct st_buffer_object *st_obj = st_buffer_object(obj);
+   GLuint flags = 0;
+
+   if (access & GL_MAP_WRITE_BIT)
+      flags |= PIPE_BUFFER_USAGE_CPU_WRITE;
+
+   if (access & GL_MAP_READ_BIT)
+      flags |= PIPE_BUFFER_USAGE_CPU_READ;
+
+   /* ... other flags ...
+    */
+
+   if (access & MESA_MAP_NOWAIT_BIT)
+      flags |= PIPE_BUFFER_USAGE_DONTBLOCK;
+
+   obj->Pointer = pipe_buffer_map(pipe->screen, st_obj->buffer, flags);
+   return obj->Pointer;
+}
+
+
+
+
+/**
+ * Called via glUnmapBufferARB().
  */
 static GLboolean
 st_bufferobj_unmap(GLcontext *ctx, GLenum target, struct gl_buffer_object *obj)
@@ -230,5 +262,6 @@ st_init_bufferobject_functions(struct dd_function_table *functions)
    functions->BufferSubData = st_bufferobj_subdata;
    functions->GetBufferSubData = st_bufferobj_get_subdata;
    functions->MapBuffer = st_bufferobj_map;
+   functions->MapBufferRange = st_bufferobj_map_range;
    functions->UnmapBuffer = st_bufferobj_unmap;
 }
diff --git a/src/mesa/vbo/vbo_exec.c b/src/mesa/vbo/vbo_exec.c
index 635f239acc..e168a89ea5 100644
--- a/src/mesa/vbo/vbo_exec.c
+++ b/src/mesa/vbo/vbo_exec.c
@@ -57,6 +57,7 @@ void vbo_exec_init( GLcontext *ctx )
 
    ctx->Driver.NeedFlush = 0;
    ctx->Driver.CurrentExecPrimitive = PRIM_OUTSIDE_BEGIN_END;
+   ctx->Driver.BeginVertices = vbo_exec_BeginVertices;
    ctx->Driver.FlushVertices = vbo_exec_FlushVertices;
 
    vbo_exec_invalidate_state( ctx, ~0 );
diff --git a/src/mesa/vbo/vbo_exec.h b/src/mesa/vbo/vbo_exec.h
index ddbcbe1181..100bb8a5de 100644
--- a/src/mesa/vbo/vbo_exec.h
+++ b/src/mesa/vbo/vbo_exec.h
@@ -43,7 +43,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 /* Wierd implementation stuff:
  */
-#define VBO_VERT_BUFFER_SIZE (1024*16)	/* dwords == 64k */
+#define VBO_VERT_BUFFER_SIZE (1024*64)	/* bytes */
 #define VBO_MAX_ATTR_CODEGEN 16 
 #define ERROR_ATTRIB 16
 
@@ -78,14 +78,15 @@ struct vbo_exec_context
 
    struct {
       struct gl_buffer_object *bufferobj;
-      GLubyte *buffer_map;
 
-      GLuint vertex_size;
+      GLuint vertex_size;       /* in dwords */
 
       struct _mesa_prim prim[VBO_MAX_PRIM];
       GLuint prim_count;
 
-      GLfloat *vbptr;		     /* cursor, points into buffer */
+      GLfloat *buffer_map;
+      GLfloat *buffer_ptr;              /* cursor, points into buffer */
+      GLuint   buffer_used;             /* in bytes */
       GLfloat vertex[VBO_ATTRIB_MAX*4]; /* current vertex */
 
       GLuint vert_count;
@@ -140,6 +141,9 @@ struct vbo_exec_context
 void vbo_exec_init( GLcontext *ctx );
 void vbo_exec_destroy( GLcontext *ctx );
 void vbo_exec_invalidate_state( GLcontext *ctx, GLuint new_state );
+void vbo_exec_FlushVertices_internal( GLcontext *ctx, GLboolean unmap );
+
+void vbo_exec_BeginVertices( GLcontext *ctx );
 void vbo_exec_FlushVertices( GLcontext *ctx, GLuint flags );
 
 
@@ -151,7 +155,8 @@ void vbo_exec_array_destroy( struct vbo_exec_context *exec );
 
 void vbo_exec_vtx_init( struct vbo_exec_context *exec );
 void vbo_exec_vtx_destroy( struct vbo_exec_context *exec );
-void vbo_exec_vtx_flush( struct vbo_exec_context *exec );
+void vbo_exec_vtx_flush( struct vbo_exec_context *exec, GLboolean unmap );
+void vbo_exec_vtx_map( struct vbo_exec_context *exec );
 void vbo_exec_vtx_wrap( struct vbo_exec_context *exec );
 
 void vbo_exec_eval_update( struct vbo_exec_context *exec );
diff --git a/src/mesa/vbo/vbo_exec_api.c b/src/mesa/vbo/vbo_exec_api.c
index 2743bf6b55..c0ffdb55e4 100644
--- a/src/mesa/vbo/vbo_exec_api.c
+++ b/src/mesa/vbo/vbo_exec_api.c
@@ -62,7 +62,7 @@ static void vbo_exec_wrap_buffers( struct vbo_exec_context *exec )
    if (exec->vtx.prim_count == 0) {
       exec->vtx.copied.nr = 0;
       exec->vtx.vert_count = 0;
-      exec->vtx.vbptr = (GLfloat *)exec->vtx.buffer_map;
+      exec->vtx.buffer_ptr = exec->vtx.buffer_map;
    }
    else {
       GLuint last_begin = exec->vtx.prim[exec->vtx.prim_count-1].begin;
@@ -80,7 +80,7 @@ static void vbo_exec_wrap_buffers( struct vbo_exec_context *exec )
       /* Execute the buffer and save copied vertices.
        */
       if (exec->vtx.vert_count)
-	 vbo_exec_vtx_flush( exec );
+	 vbo_exec_vtx_flush( exec, GL_FALSE );
       else {
 	 exec->vtx.prim_count = 0;
 	 exec->vtx.copied.nr = 0;
@@ -121,9 +121,9 @@ void vbo_exec_vtx_wrap( struct vbo_exec_context *exec )
    assert(exec->vtx.max_vert - exec->vtx.vert_count > exec->vtx.copied.nr);
 
    for (i = 0 ; i < exec->vtx.copied.nr ; i++) {
-      _mesa_memcpy( exec->vtx.vbptr, data, 
+      _mesa_memcpy( exec->vtx.buffer_ptr, data, 
 		    exec->vtx.vertex_size * sizeof(GLfloat));
-      exec->vtx.vbptr += exec->vtx.vertex_size;
+      exec->vtx.buffer_ptr += exec->vtx.vertex_size;
       data += exec->vtx.vertex_size;
       exec->vtx.vert_count++;
    }
@@ -251,9 +251,10 @@ static void vbo_exec_wrap_upgrade_vertex( struct vbo_exec_context *exec,
    exec->vtx.attrsz[attr] = newsz;
 
    exec->vtx.vertex_size += newsz - oldsz;
-   exec->vtx.max_vert = VBO_VERT_BUFFER_SIZE / exec->vtx.vertex_size;
+   exec->vtx.max_vert = ((VBO_VERT_BUFFER_SIZE - exec->vtx.buffer_used) / 
+                         (exec->vtx.vertex_size * sizeof(GLfloat)));
    exec->vtx.vert_count = 0;
-   exec->vtx.vbptr = (GLfloat *)exec->vtx.buffer_map;
+   exec->vtx.buffer_ptr = exec->vtx.buffer_map;
    
 
    /* Recalculate all the attrptr[] values
@@ -279,10 +280,10 @@ static void vbo_exec_wrap_upgrade_vertex( struct vbo_exec_context *exec,
    if (exec->vtx.copied.nr)
    {
       GLfloat *data = exec->vtx.copied.buffer;
-      GLfloat *dest = exec->vtx.vbptr;
+      GLfloat *dest = exec->vtx.buffer_ptr;
       GLuint j;
 
-      assert(exec->vtx.vbptr == (GLfloat *)exec->vtx.buffer_map);
+      assert(exec->vtx.buffer_ptr == exec->vtx.buffer_map);
       
       for (i = 0 ; i < exec->vtx.copied.nr ; i++) {
 	 for (j = 0 ; j < VBO_ATTRIB_MAX ; j++) {
@@ -308,7 +309,7 @@ static void vbo_exec_wrap_upgrade_vertex( struct vbo_exec_context *exec,
 	 }
       }
 
-      exec->vtx.vbptr = dest;
+      exec->vtx.buffer_ptr = dest;
       exec->vtx.vert_count += exec->vtx.copied.nr;
       exec->vtx.copied.nr = 0;
    }
@@ -373,9 +374,9 @@ do {								\
       GLuint i;							\
 								\
       for (i = 0; i < exec->vtx.vertex_size; i++)		\
-	 exec->vtx.vbptr[i] = exec->vtx.vertex[i];		\
+	 exec->vtx.buffer_ptr[i] = exec->vtx.vertex[i];		\
 								\
-      exec->vtx.vbptr += exec->vtx.vertex_size;			\
+      exec->vtx.buffer_ptr += exec->vtx.vertex_size;			\
       exec->ctx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;	\
 								\
       if (++exec->vtx.vert_count >= exec->vtx.max_vert)		\
@@ -532,7 +533,7 @@ static void GLAPIENTRY vbo_exec_Begin( GLenum mode )
        * begin/end pairs.
        */
       if (exec->vtx.vertex_size && !exec->vtx.attrsz[0]) 
-	 vbo_exec_FlushVertices( ctx, ~0 );
+	 vbo_exec_FlushVertices_internal( ctx, GL_FALSE );
 
       i = exec->vtx.prim_count++;
       exec->vtx.prim[i].mode = mode;
@@ -566,7 +567,7 @@ static void GLAPIENTRY vbo_exec_End( void )
       ctx->Driver.CurrentExecPrimitive = PRIM_OUTSIDE_BEGIN_END;
 
       if (exec->vtx.prim_count == VBO_MAX_PRIM)
-	 vbo_exec_vtx_flush( exec );	
+	 vbo_exec_vtx_flush( exec, GL_FALSE );
    }
    else 
       _mesa_error( ctx, GL_INVALID_OPERATION, "glEnd" );
@@ -672,23 +673,19 @@ void vbo_use_buffer_objects(GLcontext *ctx)
     */
    GLuint bufName = 0xaabbccdd;
    GLenum target = GL_ARRAY_BUFFER_ARB;
-   GLenum access = GL_READ_WRITE_ARB;
    GLenum usage = GL_STREAM_DRAW_ARB;
-   GLsizei size = VBO_VERT_BUFFER_SIZE * sizeof(GLfloat);
+   GLsizei size = VBO_VERT_BUFFER_SIZE;
 
    /* Make sure this func is only used once */
    assert(exec->vtx.bufferobj == ctx->Array.NullBufferObj);
    if (exec->vtx.buffer_map) {
       _mesa_align_free(exec->vtx.buffer_map);
+      exec->vtx.buffer_map = NULL;
    }
 
    /* Allocate a real buffer object now */
    exec->vtx.bufferobj = ctx->Driver.NewBufferObject(ctx, bufName, target);
    ctx->Driver.BufferData(ctx, target, size, NULL, usage, exec->vtx.bufferobj);
-
-   /* and map it */
-   exec->vtx.buffer_map
-      = ctx->Driver.MapBuffer(ctx, target, access, exec->vtx.bufferobj);
 }
 
 
@@ -708,7 +705,8 @@ void vbo_exec_vtx_init( struct vbo_exec_context *exec )
                                  ctx->Array.NullBufferObj);
 
    ASSERT(!exec->vtx.buffer_map);
-   exec->vtx.buffer_map = ALIGN_MALLOC(VBO_VERT_BUFFER_SIZE * sizeof(GLfloat), 64);
+   exec->vtx.buffer_map = (GLfloat *)ALIGN_MALLOC(VBO_VERT_BUFFER_SIZE, 64);
+
    vbo_exec_vtxfmt_init( exec );
 
    /* Hook our functions into the dispatch table.
@@ -747,22 +745,45 @@ void vbo_exec_vtx_destroy( struct vbo_exec_context *exec )
    }
 }
 
-
-void vbo_exec_FlushVertices( GLcontext *ctx, GLuint flags )
+void vbo_exec_BeginVertices( GLcontext *ctx )
 {
    struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+   if (0) _mesa_printf("%s\n", __FUNCTION__);
+   vbo_exec_vtx_map( exec );
+}
 
-   if (exec->ctx->Driver.CurrentExecPrimitive != PRIM_OUTSIDE_BEGIN_END)
-      return;
+void vbo_exec_FlushVertices_internal( GLcontext *ctx, GLboolean unmap )
+{
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
 
-   if (exec->vtx.vert_count) {
-      vbo_exec_vtx_flush( exec );
+   if (exec->vtx.vert_count || unmap) {
+      vbo_exec_vtx_flush( exec, unmap );
    }
 
    if (exec->vtx.vertex_size) {
       vbo_exec_copy_to_current( exec );
       reset_attrfv( exec );
    }
+}
+
+
+
+void vbo_exec_FlushVertices( GLcontext *ctx, GLuint flags )
+{
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+
+   if (0) _mesa_printf("%s\n", __FUNCTION__);
+
+   if (exec->ctx->Driver.CurrentExecPrimitive != PRIM_OUTSIDE_BEGIN_END) {
+      if (0) _mesa_printf("%s - inside begin/end\n", __FUNCTION__);
+      return;
+   }
+
+   vbo_exec_FlushVertices_internal( ctx, GL_TRUE );
+
+   /* Need to do this to ensure BeginVertices gets called again:
+    */
+   _mesa_restore_exec_vtxfmt( ctx );
 
    exec->ctx->Driver.NeedFlush = 0;
 }
diff --git a/src/mesa/vbo/vbo_exec_draw.c b/src/mesa/vbo/vbo_exec_draw.c
index 45133fa4f3..38b6c56f47 100644
--- a/src/mesa/vbo/vbo_exec_draw.c
+++ b/src/mesa/vbo/vbo_exec_draw.c
@@ -70,7 +70,7 @@ static GLuint vbo_copy_vertices( struct vbo_exec_context *exec )
    GLuint ovf, i;
    GLuint sz = exec->vtx.vertex_size;
    GLfloat *dst = exec->vtx.copied.buffer;
-   GLfloat *src = ((GLfloat *)exec->vtx.buffer_map + 
+   GLfloat *src = (exec->vtx.buffer_map + 
 		   exec->vtx.prim[exec->vtx.prim_count-1].start * 
 		   exec->vtx.vertex_size);
 
@@ -147,7 +147,7 @@ static void vbo_exec_bind_arrays( GLcontext *ctx )
    struct vbo_exec_context *exec = &vbo->exec;
    struct gl_client_array *arrays = exec->vtx.arrays;
    GLuint count = exec->vtx.vert_count;
-   GLubyte *data = exec->vtx.buffer_map;
+   GLubyte *data = (GLubyte *)exec->vtx.buffer_map;
    const GLuint *map;
    GLuint attr;
    GLbitfield varying_inputs = 0x0;
@@ -234,10 +234,78 @@ static void vbo_exec_bind_arrays( GLcontext *ctx )
 }
 
 
+static void vbo_exec_vtx_unmap( struct vbo_exec_context *exec )
+{
+   GLenum target = GL_ARRAY_BUFFER_ARB;
+
+   if (exec->vtx.bufferobj->Name) {
+      GLcontext *ctx = exec->ctx;
+
+      exec->vtx.buffer_used += (exec->vtx.buffer_ptr -
+                                exec->vtx.buffer_map) * sizeof(float);
+            
+      ctx->Driver.UnmapBuffer(ctx, target, exec->vtx.bufferobj);
+      exec->vtx.buffer_map = NULL;
+      exec->vtx.buffer_ptr = NULL;
+      exec->vtx.max_vert = 0;
+   }
+}
+
+void vbo_exec_vtx_map( struct vbo_exec_context *exec )
+{
+   GLcontext *ctx = exec->ctx;
+   GLenum target = GL_ARRAY_BUFFER_ARB;
+   GLenum access = GL_READ_WRITE_ARB;
+   GLenum usage = GL_STREAM_DRAW_ARB;
+   
+   if (exec->vtx.bufferobj->Name == 0)
+      return;
+
+   if (exec->vtx.buffer_map != NULL) {
+      assert(0);
+      exec->vtx.buffer_map = NULL;
+   }
+
+   if (VBO_VERT_BUFFER_SIZE > exec->vtx.buffer_used + 1024 &&
+       ctx->Driver.MapBufferRange)
+   {
+      exec->vtx.buffer_map = 
+         (GLfloat *)ctx->Driver.MapBufferRange(ctx, 
+                                               target, 
+                                               exec->vtx.buffer_used,
+                                               (VBO_VERT_BUFFER_SIZE - 
+                                                exec->vtx.buffer_used),
+                                               (GL_MAP_WRITE_BIT |
+                                                GL_MAP_INVALIDATE_RANGE_BIT | 
+                                                GL_MAP_UNSYNCHRONIZED_BIT | 
+                                                MESA_MAP_NOWAIT_BIT),
+                                               exec->vtx.bufferobj);
+   }
+
+   if (exec->vtx.buffer_map) {
+      exec->vtx.buffer_map += exec->vtx.buffer_used / sizeof(float);
+   }
+   else {
+      exec->vtx.buffer_used = 0;
+
+      ctx->Driver.BufferData(ctx, target, 
+                             VBO_VERT_BUFFER_SIZE, 
+                             NULL, usage, exec->vtx.bufferobj);
+
+      exec->vtx.buffer_map = 
+         (GLfloat *)ctx->Driver.MapBuffer(ctx, target, access, exec->vtx.bufferobj);
+   }
+
+   if (0) _mesa_printf("map %d..\n", exec->vtx.buffer_used);
+}
+
+
+
 /**
  * Execute the buffer and save copied verts.
  */
-void vbo_exec_vtx_flush( struct vbo_exec_context *exec )
+void vbo_exec_vtx_flush( struct vbo_exec_context *exec,
+                         GLboolean unmap )
 {
    if (0)
       vbo_exec_debug_verts( exec );
@@ -250,25 +318,22 @@ void vbo_exec_vtx_flush( struct vbo_exec_context *exec )
 
       if (exec->vtx.copied.nr != exec->vtx.vert_count) {
 	 GLcontext *ctx = exec->ctx;
-
-	 GLenum target = GL_ARRAY_BUFFER_ARB;
-	 GLenum access = GL_READ_WRITE_ARB;
-	 GLenum usage = GL_STREAM_DRAW_ARB;
-	 GLsizei size = VBO_VERT_BUFFER_SIZE * sizeof(GLfloat);
 	 
-	 /* Before the unmap (why?)
+	 /* Before the update_state() as this may raise _NEW_ARRAY
+          * from _mesa_set_varying_vp_inputs().
 	  */
 	 vbo_exec_bind_arrays( ctx );
 
          if (ctx->NewState)
             _mesa_update_state( ctx );
 
-         /* if using a real VBO, unmap it before drawing */
          if (exec->vtx.bufferobj->Name) {
-            ctx->Driver.UnmapBuffer(ctx, target, exec->vtx.bufferobj);
-            exec->vtx.buffer_map = NULL;
+            vbo_exec_vtx_unmap( exec );
          }
 
+         if (0) _mesa_printf("%s %d %d\n", __FUNCTION__, exec->vtx.prim_count,
+                      exec->vtx.vert_count);
+
 	 vbo_context(ctx)->draw_prims( ctx, 
 				       exec->vtx.inputs, 
 				       exec->vtx.prim, 
@@ -277,16 +342,31 @@ void vbo_exec_vtx_flush( struct vbo_exec_context *exec )
 				       0,
 				       exec->vtx.vert_count - 1);
 
-	 /* If using a real VBO, get new storage */
-         if (exec->vtx.bufferobj->Name) {
-            ctx->Driver.BufferData(ctx, target, size, NULL, usage, exec->vtx.bufferobj);
-            exec->vtx.buffer_map = 
-               ctx->Driver.MapBuffer(ctx, target, access, exec->vtx.bufferobj);
-         }
+	 /* If using a real VBO, get new storage -- unless asked not to.
+          */
+         if (exec->vtx.bufferobj->Name && !unmap) {
+            vbo_exec_vtx_map( exec );
+          }
       }
    }
 
+   /* May have to unmap explicitly if we didn't draw:
+    */
+   if (unmap && 
+       exec->vtx.bufferobj->Name &&
+       exec->vtx.buffer_map) {
+      vbo_exec_vtx_unmap( exec );
+   }
+
+
+   if (unmap) 
+      exec->vtx.max_vert = 0;
+   else
+      exec->vtx.max_vert = ((VBO_VERT_BUFFER_SIZE - exec->vtx.buffer_used) / 
+                            (exec->vtx.vertex_size * sizeof(GLfloat)));
+
+
+   exec->vtx.buffer_ptr = exec->vtx.buffer_map;
    exec->vtx.prim_count = 0;
    exec->vtx.vert_count = 0;
-   exec->vtx.vbptr = (GLfloat *)exec->vtx.buffer_map;
 }
diff --git a/src/mesa/vbo/vbo_save.h b/src/mesa/vbo/vbo_save.h
index 7dda54af48..9558f83883 100644
--- a/src/mesa/vbo/vbo_save.h
+++ b/src/mesa/vbo/vbo_save.h
@@ -130,7 +130,7 @@ struct vbo_save_context {
    struct vbo_save_vertex_store *vertex_store;
    struct vbo_save_primitive_store *prim_store;
 
-   GLfloat *vbptr;		   /* cursor, points into buffer */
+   GLfloat *buffer_ptr;		   /* cursor, points into buffer */
    GLfloat vertex[VBO_ATTRIB_MAX*4];	   /* current values */
    GLfloat *attrptr[VBO_ATTRIB_MAX];
    GLuint vert_count;
diff --git a/src/mesa/vbo/vbo_save_api.c b/src/mesa/vbo/vbo_save_api.c
index ddfd276577..52b6f1884e 100644
--- a/src/mesa/vbo/vbo_save_api.c
+++ b/src/mesa/vbo/vbo_save_api.c
@@ -241,7 +241,7 @@ static void _save_reset_counters( GLcontext *ctx )
    save->buffer = (save->vertex_store->buffer + 
 		   save->vertex_store->used);
 
-   assert(save->buffer == save->vbptr);
+   assert(save->buffer == save->buffer_ptr);
 
    if (save->vertex_size)
       save->max_vert = ((VBO_SAVE_BUFFER_SIZE - save->vertex_store->used) / 
@@ -343,7 +343,7 @@ static void _save_compile_vertex_list( GLcontext *ctx )
       /* Allocate and map new store:
        */
       save->vertex_store = alloc_vertex_store( ctx );
-      save->vbptr = map_vertex_store( ctx, save->vertex_store );
+      save->buffer_ptr = map_vertex_store( ctx, save->vertex_store );
    } 
 
    if (save->prim_store->used > VBO_SAVE_PRIM_SIZE - 6) {
@@ -414,9 +414,9 @@ static void _save_wrap_filled_vertex( GLcontext *ctx )
    assert(save->max_vert - save->vert_count > save->copied.nr);
 
    for (i = 0 ; i < save->copied.nr ; i++) {
-      _mesa_memcpy( save->vbptr, data, save->vertex_size * sizeof(GLfloat));
+      _mesa_memcpy( save->buffer_ptr, data, save->vertex_size * sizeof(GLfloat));
       data += save->vertex_size;
-      save->vbptr += save->vertex_size;
+      save->buffer_ptr += save->vertex_size;
       save->vert_count++;
    }
 }
@@ -550,7 +550,7 @@ static void _save_upgrade_vertex( GLcontext *ctx,
 	 }
       }
 
-      save->vbptr = dest;
+      save->buffer_ptr = dest;
       save->vert_count += save->copied.nr;
    }
 }
@@ -622,9 +622,9 @@ do {								\
       GLuint i;							\
 								\
       for (i = 0; i < save->vertex_size; i++)			\
-	 save->vbptr[i] = save->vertex[i];			\
+	 save->buffer_ptr[i] = save->vertex[i];			\
 								\
-      save->vbptr += save->vertex_size;				\
+      save->buffer_ptr += save->vertex_size;				\
 								\
       if (++save->vert_count >= save->max_vert)			\
 	 _save_wrap_filled_vertex( ctx );			\
@@ -1035,7 +1035,7 @@ void vbo_save_NewList( GLcontext *ctx, GLuint list, GLenum mode )
    if (!save->vertex_store) 
       save->vertex_store = alloc_vertex_store( ctx );
       
-   save->vbptr = map_vertex_store( ctx, save->vertex_store );
+   save->buffer_ptr = map_vertex_store( ctx, save->vertex_store );
    
    _save_reset_vertex( ctx );
    _save_reset_counters( ctx );