Mesa (master): radeonsi: CP DMA clear supports unaligned destination addresses

[Nicolai Hähnle]

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 ||