Mesa (master): radeonsi: CP DMA clear supports unaligned destination addresses
Nicolai Hähnle
nh at kemper.freedesktop.org
Tue Mar 28 08:22:40 UTC 2017
Module: Mesa
Branch: master
Commit: f0d9af772e0fbb5854fc8293186a70ea3721748a
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=f0d9af772e0fbb5854fc8293186a70ea3721748a
Author: Nicolai Hähnle <nicolai.haehnle at amd.com>
Date: Mon Feb 13 13:08:52 2017 +0100
radeonsi: CP DMA clear supports unaligned destination addresses
Reviewed-by: Marek Olšák <marek.olsak at amd.com>
---
src/gallium/drivers/radeonsi/si_cp_dma.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/src/gallium/drivers/radeonsi/si_cp_dma.c b/src/gallium/drivers/radeonsi/si_cp_dma.c
index b40f5cc587..0cf7b3b3cb 100644
--- a/src/gallium/drivers/radeonsi/si_cp_dma.c
+++ b/src/gallium/drivers/radeonsi/si_cp_dma.c
@@ -197,7 +197,7 @@ static void si_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
offset + size);
/* Fallback for unaligned clears. */
- if (offset % 4 != 0 || size % 4 != 0) {
+ if (size % 4 != 0) {
uint8_t *map = r600_buffer_map_sync_with_rings(&sctx->b, rdst,
PIPE_TRANSFER_WRITE);
map += offset;
@@ -211,6 +211,7 @@ static void si_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
/* dma_clear_buffer can use clear_buffer on failure. Make sure that
* doesn't happen. We don't want an infinite recursion: */
if (sctx->b.dma.cs &&
+ (offset % 4 == 0) &&
/* CP DMA is very slow. Always use SDMA for big clears. This
* alone improves DeusEx:MD performance by 70%. */
(size > 128 * 1024 ||
More information about the mesa-commit
mailing list