OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

radv: prefer CP DMA for GTT buffer copies/clears on dGPUs due to slow PCIe

Browse Source 
The CP DMA bandwidth is always better than PCIe, so I think wasting
compute resources is not a good idea. This is only enabled on GFX10+
because untested on older gens and also because RadeonSI does that.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8763>
This commit is contained in:
Samuel Pitoiset
2021-01-28 16:26:47 +01:00
committed by Marge Bot
parent f5a02d16a6
commit e870796113
1 changed files with 26 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
src/amd/vulkan/radv_meta_buffer.c
View File

@@ -351,17 +351,38 @@ static void copy_buffer_shader(struct radv_cmd_buffer *cmd_buffer,
radv_meta_restore(&saved_state, cmd_buffer);
}
static bool
radv_prefer_compute_dma(const struct radv_device *device,
uint64_t size,
struct radeon_winsys_bo *src_bo,
struct radeon_winsys_bo *dst_bo)
{
bool use_compute = size >= RADV_BUFFER_OPS_CS_THRESHOLD;
if (device->physical_device->rad_info.chip_class >= GFX10 &&
device->physical_device->rad_info.has_dedicated_vram) {
if ((src_bo && !(src_bo->initial_domain & RADEON_DOMAIN_VRAM)) ||
!(dst_bo->initial_domain & RADEON_DOMAIN_VRAM)) {
/* Prefer CP DMA for GTT on dGPUS due to slow PCIe. */
use_compute = false;
}
}
return use_compute;
}
uint32_t radv_fill_buffer(struct radv_cmd_buffer *cmd_buffer,
const struct radv_image *image,
struct radeon_winsys_bo *bo,
uint64_t offset, uint64_t size, uint32_t value)
{
bool use_compute = radv_prefer_compute_dma(cmd_buffer->device, size, NULL, bo);
uint32_t flush_bits = 0;
assert(!(offset & 3));
assert(!(size & 3));
if (size >= RADV_BUFFER_OPS_CS_THRESHOLD) {
if (use_compute) {
cmd_buffer->state.flush_bits |=
radv_dst_access_flush(cmd_buffer, VK_ACCESS_SHADER_WRITE_BIT, image);
@@ -387,7 +408,10 @@ void radv_copy_buffer(struct radv_cmd_buffer *cmd_buffer,
uint64_t src_offset, uint64_t dst_offset,
uint64_t size)
{
if (size >= RADV_BUFFER_OPS_CS_THRESHOLD && !(size & 3) && !(src_offset & 3) && !(dst_offset & 3))
bool use_compute = !(size & 3) && !(src_offset & 3) && !(dst_offset & 3) &&
radv_prefer_compute_dma(cmd_buffer->device, size, src_bo, dst_bo);
if (use_compute)
copy_buffer_shader(cmd_buffer, src_bo, dst_bo,
src_offset, dst_offset, size);
else if (size) {