Okay, I'll drop the patch. It's not too important anyway. Marek
On Tue, Feb 25, 2014 at 3:54 AM, Michel Dänzer <mic...@daenzer.net> wrote: > On Die, 2014-02-25 at 00:48 +0100, Marek Olšák wrote: >> >> diff --git a/src/gallium/drivers/radeon/r600_texture.c >> b/src/gallium/drivers/radeon/r600_texture.c >> index 35ad97b..1966251 100644 >> --- a/src/gallium/drivers/radeon/r600_texture.c >> +++ b/src/gallium/drivers/radeon/r600_texture.c >> @@ -910,7 +910,8 @@ static void *r600_texture_transfer_map(struct >> pipe_context *ctx, >> >> /* Untiled buffers in VRAM, which is slow for CPU reads and writes */ >> if (!(usage & PIPE_TRANSFER_MAP_DIRECTLY) && >> - (rtex->resource.domains == RADEON_DOMAIN_VRAM)) { >> + (rctx->ws->buffer_get_current_domain(rtex->resource.cs_buf) == >> + RADEON_DOMAIN_VRAM)) { >> use_staging_texture = TRUE; >> } > > If rtex is already referenced by the command stream being prepared, the > decision needs to be based on that instead. The ioctl just returns where > the BO happens to be when it's called, which is not necessarily where it > will be when this command stream is executed. > > > -- > Earthling Michel Dänzer | http://www.amd.com > Libre software enthusiast | Mesa and X developer > _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev