radv: allocate the SQTT BO in GTT for faster readback

Reading VRAM from CPU is very slow.

This is similar to the SPM BO, and generating RGP captures is now
way faster.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38551>
This commit is contained in:
Samuel Pitoiset 2025-11-20 12:51:00 +01:00 committed by Marge Bot
parent d2ebe7719c
commit da07f1ef3f

View file

@ -281,7 +281,7 @@ radv_sqtt_init_bo(struct radv_device *device)
size += device->sqtt.buffer_size * (uint64_t)max_se;
struct radeon_winsys_bo *bo = NULL;
result = radv_bo_create(device, NULL, size, 4096, RADEON_DOMAIN_VRAM,
result = radv_bo_create(device, NULL, size, 4096, RADEON_DOMAIN_GTT,
RADEON_FLAG_CPU_ACCESS | RADEON_FLAG_NO_INTERPROCESS_SHARING | RADEON_FLAG_ZERO_VRAM,
RADV_BO_PRIORITY_SCRATCH, 0, true, &bo);
device->sqtt.bo = bo;