diff options
author | Marek Olšák <[email protected]> | 2017-02-09 03:14:22 +0100 |
---|---|---|
committer | Marek Olšák <[email protected]> | 2017-02-10 11:27:50 +0100 |
commit | d86099df0af7c22c8acfd48b38ad446d9c8df6bd (patch) | |
tree | 37150f84dbea42b5c86c32a5d1d007f6db8130bb /src/gallium/drivers | |
parent | 24e3b064088b837a930f6e62afa43b3fc3c10e13 (diff) |
gallium/radeon: fix performance of buffer readbacks
We want cached GTT for all non-persistent read mappings.
Set level = 0 on purpose.
Use dma_copy, because resource_copy_region causes a failure in the PBO
read of piglit/getteximage-luminance.
If Rocket League used the READ flag, it should get cached GTT.
v2: mask out UNSYNCHRONIZED
Cc: 13.0 17.0 <[email protected]>
Reviewed-by: Nicolai Hähnle <[email protected]>
Diffstat (limited to 'src/gallium/drivers')
-rw-r--r-- | src/gallium/drivers/radeon/r600_buffer_common.c | 17 |
1 files changed, 9 insertions, 8 deletions
diff --git a/src/gallium/drivers/radeon/r600_buffer_common.c b/src/gallium/drivers/radeon/r600_buffer_common.c index da6f0206d78..762181fb9d1 100644 --- a/src/gallium/drivers/radeon/r600_buffer_common.c +++ b/src/gallium/drivers/radeon/r600_buffer_common.c @@ -381,11 +381,11 @@ static void *r600_buffer_transfer_map(struct pipe_context *ctx, usage |= PIPE_TRANSFER_UNSYNCHRONIZED; } } - /* Using a staging buffer in GTT for larger reads is much faster. */ + /* Use a staging buffer in cached GTT for reads. */ else if ((usage & PIPE_TRANSFER_READ) && - !(usage & (PIPE_TRANSFER_WRITE | - PIPE_TRANSFER_PERSISTENT)) && - rbuffer->domains & RADEON_DOMAIN_VRAM && + !(usage & PIPE_TRANSFER_PERSISTENT) && + (rbuffer->domains & RADEON_DOMAIN_VRAM || + rbuffer->flags & RADEON_FLAG_GTT_WC) && r600_can_dma_copy_buffer(rctx, 0, box->x, box->width)) { struct r600_resource *staging; @@ -394,11 +394,12 @@ static void *r600_buffer_transfer_map(struct pipe_context *ctx, box->width + (box->x % R600_MAP_BUFFER_ALIGNMENT)); if (staging) { /* Copy the VRAM buffer to the staging buffer. */ - ctx->resource_copy_region(ctx, &staging->b.b, 0, - box->x % R600_MAP_BUFFER_ALIGNMENT, - 0, 0, resource, level, box); + rctx->dma_copy(ctx, &staging->b.b, 0, + box->x % R600_MAP_BUFFER_ALIGNMENT, + 0, 0, resource, 0, box); - data = r600_buffer_map_sync_with_rings(rctx, staging, PIPE_TRANSFER_READ); + data = r600_buffer_map_sync_with_rings(rctx, staging, + usage & ~PIPE_TRANSFER_UNSYNCHRONIZED); if (!data) { r600_resource_reference(&staging, NULL); return NULL; |