diff options
| author |   Felix Kuehling <Felix.Kuehling@amd.com> | 2018-07-14 19:05:59 -0400 |
|---|---|---|
| committer |   Oded Gabbay <oded.gabbay@gmail.com> | 2018-07-14 19:05:59 -0400 |
| commit | 39e7f331864d2b9e30d5f3fd2121e182b2c9c8a9 (patch) | |
| tree | 148394727c9e0b6eb318c0a3dff96e6992dffefb /drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c | |
| parent | drm/amd: Add CU-masking ioctl definition to kfd_ioctl.h (diff) | |
| download | linux-dev-39e7f331864d2b9e30d5f3fd2121e182b2c9c8a9.tar.xz linux-dev-39e7f331864d2b9e30d5f3fd2121e182b2c9c8a9.zip | |
drm/amdkfd: Add CU-masking ioctl to KFD
CU-masking allows a KFD client to control the set of CUs used by a
user mode queue for executing compute dispatches. This can be used
for optimizing the partitioning of the GPU and minimize conflicts
between concurrent tasks.
Signed-off-by: Flora Cui <flora.cui@amd.com>
Signed-off-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.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 | 41 |
1 files changed, 40 insertions, 1 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c index 4b8eb506642b..3bc25ab84f34 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c @@ -21,7 +21,7 @@ * */ -#include "kfd_priv.h" +#include "kfd_mqd_manager.h" struct mqd_manager *mqd_manager_init(enum KFD_MQD_TYPE type, struct kfd_dev *dev) @@ -48,3 +48,42 @@ struct mqd_manager *mqd_manager_init(enum KFD_MQD_TYPE type, return NULL; } + +void mqd_symmetrically_map_cu_mask(struct mqd_manager *mm, + const uint32_t *cu_mask, uint32_t cu_mask_count, + uint32_t *se_mask) +{ + struct kfd_cu_info cu_info; + uint32_t cu_per_sh[4] = {0}; + int i, se, cu = 0; + + mm->dev->kfd2kgd->get_cu_info(mm->dev->kgd, &cu_info); + + if (cu_mask_count > cu_info.cu_active_number) + cu_mask_count = cu_info.cu_active_number; + + for (se = 0; se < cu_info.num_shader_engines; se++) + for (i = 0; i < 4; i++) + cu_per_sh[se] += hweight32(cu_info.cu_bitmap[se][i]); + + /* Symmetrically map cu_mask to all SEs: + * cu_mask[0] bit0 -> se_mask[0] bit0; + * cu_mask[0] bit1 -> se_mask[1] bit0; + * ... (if # SE is 4) + * cu_mask[0] bit4 -> se_mask[0] bit1; + * ... + */ + se = 0; + for (i = 0; i < cu_mask_count; i++) { + if (cu_mask[i / 32] & (1 << (i % 32))) + se_mask[se] |= 1 << cu; + + do { + se++; + if (se == cu_info.num_shader_engines) { + se = 0; + cu++; + } + } while (cu >= cu_per_sh[se] && cu < 32); + } +} |