diff options
| author | Jonathan Kim <jonathan.kim@amd.com> | 2022-06-27 04:35:10 +0300 |
|---|---|---|
| committer | Alex Deucher <alexander.deucher@amd.com> | 2022-06-30 22:28:03 +0300 |
| commit | cff35798fad565b5f8bbf4ef84ac698270dcaf64 (patch) | |
| tree | f32c32d6091704d5d76b52fc83d129d271f72cbb /drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c | |
| parent | 6a4a1f6054318cd3590562668798304b4351ef36 (diff) | |
| download | linux-cff35798fad565b5f8bbf4ef84ac698270dcaf64.tar.xz | |
drm/amdkfd: fix cu mask for asics with wgps
GFX10 and up have work group processors (WGP) and WGP mode is the native
compile mode.
KFD and ROCr have no visibility into whether a dispatch is operating
in CU or WGP mode.
Enforce CU masking to be pairwise continguous in enablement and
round robin distribute CUs across the SEs in a pairwise manner to
assume WGP mode at all times.
Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c | 12 |
1 files changed, 7 insertions, 5 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c index 49a283be6b57..623ccd227b7d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c @@ -100,7 +100,9 @@ void mqd_symmetrically_map_cu_mask(struct mqd_manager *mm, { struct kfd_cu_info cu_info; uint32_t cu_per_sh[KFD_MAX_NUM_SE][KFD_MAX_NUM_SH_PER_SE] = {0}; - int i, se, sh, cu, cu_bitmap_sh_mul; + bool wgp_mode_req = KFD_GC_VERSION(mm->dev) >= IP_VERSION(10, 0, 0); + uint32_t en_mask = wgp_mode_req ? 0x3 : 0x1; + int i, se, sh, cu, cu_bitmap_sh_mul, inc = wgp_mode_req ? 2 : 1; amdgpu_amdkfd_get_cu_info(mm->dev->adev, &cu_info); @@ -167,13 +169,13 @@ void mqd_symmetrically_map_cu_mask(struct mqd_manager *mm, se_mask[i] = 0; i = 0; - for (cu = 0; cu < 16; cu++) { + for (cu = 0; cu < 16; cu += inc) { for (sh = 0; sh < cu_info.num_shader_arrays_per_engine; sh++) { for (se = 0; se < cu_info.num_shader_engines; se++) { if (cu_per_sh[se][sh] > cu) { - if (cu_mask[i / 32] & (1 << (i % 32))) - se_mask[se] |= 1 << (cu + sh * 16); - i++; + if (cu_mask[i / 32] & (en_mask << (i % 32))) + se_mask[se] |= en_mask << (cu + sh * 16); + i += inc; if (i == cu_mask_count) return; } |