diff options
| author | Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl> | 2020-10-12 04:06:13 +0200 |
|---|---|---|
| committer | Marge Bot <eric+marge@anholt.net> | 2021-01-13 16:27:19 +0000 |
| commit | 0af86341a20259932955f0386ca2d865928ea409 (patch) | |
| tree | acf6dfdef43c5162ec389e8e04702e3dd87e996b /src | |
| parent | 8f8d72af555c2f4d7a44107bf52602571fc0c4d2 (diff) | |
radv: Use L2 for CP DMA on GFX9+.
This enables assuming that the L2 is always up to date for barriers.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7202>
Diffstat (limited to 'src')
| -rw-r--r-- | src/amd/vulkan/si_cmd_buffer.c | 17 |
1 files changed, 13 insertions, 4 deletions
diff --git a/src/amd/vulkan/si_cmd_buffer.c b/src/amd/vulkan/si_cmd_buffer.c index 97636976b1a..28d4a01a996 100644 --- a/src/amd/vulkan/si_cmd_buffer.c +++ b/src/amd/vulkan/si_cmd_buffer.c @@ -1766,10 +1766,17 @@ void si_cp_dma_buffer_copy(struct radv_cmd_buffer *cmd_buffer, unsigned dma_flags = 0; unsigned byte_count = MIN2(size, cp_dma_max_byte_count(cmd_buffer)); - if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX10) { + if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) { /* DMA operations via L2 are coherent and faster. - * TODO: GFX7-GFX9 should also support this but it + * TODO: GFX7-GFX8 should also support this but it * requires tests/benchmarks. + * + * Also enable on GFX9 so we can use L2 at rest on GFX9+. On Raven + * this didn't seem to be worse. + * + * Note that we only use CP DMA for sizes < RADV_BUFFER_OPS_CS_THRESHOLD, + * which is 4k at the moment, so this is really unlikely to cause + * significant thrashing. */ dma_flags |= CP_DMA_USE_L2; } @@ -1818,10 +1825,12 @@ void si_cp_dma_clear_buffer(struct radv_cmd_buffer *cmd_buffer, uint64_t va, unsigned byte_count = MIN2(size, cp_dma_max_byte_count(cmd_buffer)); unsigned dma_flags = CP_DMA_CLEAR; - if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX10) { + if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) { /* DMA operations via L2 are coherent and faster. - * TODO: GFX7-GFX9 should also support this but it + * TODO: GFX7-GFX8 should also support this but it * requires tests/benchmarks. + * + * Also enable on GFX9 so we can use L2 at rest on GFX9+. */ dma_flags |= CP_DMA_USE_L2; } |