Commit e3de58f8 authored by Rajneesh Bhardwaj's avatar Rajneesh Bhardwaj Committed by Alex Deucher

drm/amdkfd: update SIMD distribution algo for GFXIP 9.4.2 onwards

In certain cooperative group dispatch scenarios the default SPI resource
allocation may cause reduced per-CU workgroup occupancy. Set
COMPUTE_RESOURCE_LIMITS.FORCE_SIMD_DIST=1 to mitigate soft hang
scenarions.
Reviewed-by: default avatarFelix Kuehling <felix.kuehling@amd.com>
Suggested-by: default avatarJoseph Greathouse <Joseph.Greathouse@amd.com>
Signed-off-by: default avatarRajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>
Signed-off-by: default avatarAlex Deucher <alexander.deucher@amd.com>
parent 0d555e48
...@@ -303,6 +303,15 @@ static void update_mqd(struct mqd_manager *mm, void *mqd, ...@@ -303,6 +303,15 @@ static void update_mqd(struct mqd_manager *mm, void *mqd,
update_cu_mask(mm, mqd, minfo, 0); update_cu_mask(mm, mqd, minfo, 0);
set_priority(m, q); set_priority(m, q);
if (minfo && KFD_GC_VERSION(mm->dev) >= IP_VERSION(9, 4, 2)) {
if (minfo->update_flag & UPDATE_FLAG_IS_GWS)
m->compute_resource_limits |=
COMPUTE_RESOURCE_LIMITS__FORCE_SIMD_DIST_MASK;
else
m->compute_resource_limits &=
~COMPUTE_RESOURCE_LIMITS__FORCE_SIMD_DIST_MASK;
}
q->is_active = QUEUE_IS_ACTIVE(*q); q->is_active = QUEUE_IS_ACTIVE(*q);
} }
......
...@@ -532,6 +532,7 @@ struct queue_properties { ...@@ -532,6 +532,7 @@ struct queue_properties {
enum mqd_update_flag { enum mqd_update_flag {
UPDATE_FLAG_DBG_WA_ENABLE = 1, UPDATE_FLAG_DBG_WA_ENABLE = 1,
UPDATE_FLAG_DBG_WA_DISABLE = 2, UPDATE_FLAG_DBG_WA_DISABLE = 2,
UPDATE_FLAG_IS_GWS = 4, /* quirk for gfx9 IP */
}; };
struct mqd_update_info { struct mqd_update_info {
......
...@@ -95,6 +95,7 @@ void kfd_process_dequeue_from_device(struct kfd_process_device *pdd) ...@@ -95,6 +95,7 @@ void kfd_process_dequeue_from_device(struct kfd_process_device *pdd)
int pqm_set_gws(struct process_queue_manager *pqm, unsigned int qid, int pqm_set_gws(struct process_queue_manager *pqm, unsigned int qid,
void *gws) void *gws)
{ {
struct mqd_update_info minfo = {0};
struct kfd_node *dev = NULL; struct kfd_node *dev = NULL;
struct process_queue_node *pqn; struct process_queue_node *pqn;
struct kfd_process_device *pdd; struct kfd_process_device *pdd;
...@@ -146,9 +147,10 @@ int pqm_set_gws(struct process_queue_manager *pqm, unsigned int qid, ...@@ -146,9 +147,10 @@ int pqm_set_gws(struct process_queue_manager *pqm, unsigned int qid,
} }
pdd->qpd.num_gws = gws ? dev->adev->gds.gws_size : 0; pdd->qpd.num_gws = gws ? dev->adev->gds.gws_size : 0;
minfo.update_flag = gws ? UPDATE_FLAG_IS_GWS : 0;
return pqn->q->device->dqm->ops.update_queue(pqn->q->device->dqm, return pqn->q->device->dqm->ops.update_queue(pqn->q->device->dqm,
pqn->q, NULL); pqn->q, &minfo);
} }
void kfd_process_dequeue_from_all_devices(struct kfd_process *p) void kfd_process_dequeue_from_all_devices(struct kfd_process *p)
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment