Re: [PATCH v7 1/4] drm/amdgpu: set compute queue priority at mqd_init
Am 09.03.20 um 13:59 schrieb Nirmoy Das:
We were changing compute ring priority while rings were being used
before every job submission which is not recommended. This patch
sets compute queue priority at mqd initialization for gfx8, gfx9 and
gfx10.
Policy: make queue 0 of each pipe as high priority compute queue
High/normal priority compute sched lists are generated from set of high/normal
priority compute queues. At context creation, entity of compute queue
get a sched list from high or normal priority depending on ctx->priority
Signed-off-by: Nirmoy Das
Acked-by: Alex Deucher
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 4 --
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 53
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 8
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h | 16 ++-
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 6 ---
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 1 +
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 19 +
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c| 23 --
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c| 20 +
9 files changed, 129 insertions(+), 21 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index f397ff97b4e4..8304d0c87899 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -1205,7 +1205,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
struct drm_sched_entity *entity = p->entity;
enum drm_sched_priority priority;
- struct amdgpu_ring *ring;
struct amdgpu_bo_list_entry *e;
struct amdgpu_job *job;
uint64_t seq;
@@ -1258,9 +1257,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
priority = job->base.s_priority;
drm_sched_entity_push_job(&job->base, entity);
- ring = to_amdgpu_ring(entity->rq->sched);
- amdgpu_ring_priority_get(ring, priority);
-
amdgpu_vm_move_to_lru_tail(p->adev, &fpriv->vm);
ttm_eu_fence_buffer_objects(&p->ticket, &p->validated, p->fence);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 94a6c42f29ea..1d05eb64b885 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -61,12 +61,24 @@ static int amdgpu_ctx_priority_permit(struct drm_file *filp,
return -EACCES;
}
+static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_compute_prio(enum
drm_sched_priority prio)
+{
+ switch (prio) {
+ case DRM_SCHED_PRIORITY_HIGH_HW:
+ case DRM_SCHED_PRIORITY_KERNEL:
+ return AMDGPU_GFX_PIPE_PRIO_HIGH;
+ default:
+ return AMDGPU_GFX_PIPE_PRIO_NORMAL;
+ }
+}
+
static int amdgpu_ctx_init_entity(struct amdgpu_ctx *ctx, const u32 hw_ip,
const u32 ring)
{
struct amdgpu_device *adev = ctx->adev;
struct amdgpu_ctx_entity *entity;
struct drm_gpu_scheduler **scheds = NULL, *sched = NULL;
unsigned num_scheds = 0;
+ enum gfx_pipe_priority hw_prio;
enum drm_sched_priority priority;
int r;
@@ -85,8 +97,9 @@ static int amdgpu_ctx_init_entity(struct amdgpu_ctx *ctx,
const u32 hw_ip, const
num_scheds = 1;
break;
case AMDGPU_HW_IP_COMPUTE:
- scheds = adev->gfx.compute_sched;
- num_scheds = adev->gfx.num_compute_sched;
+ hw_prio =
amdgpu_ctx_sched_prio_to_compute_prio(priority);
+ scheds = adev->gfx.compute_prio_sched[hw_prio];
+ num_scheds = adev->gfx.num_compute_sched[hw_prio];
break;
case AMDGPU_HW_IP_DMA:
scheds = adev->sdma.sdma_sched;
@@ -628,20 +641,46 @@ void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr)
mutex_destroy(&mgr->lock);
}
+
+static void amdgpu_ctx_init_compute_sched(struct amdgpu_device *adev)
+{
+ int num_compute_sched_normal = 0;
+ int num_compute_sched_high = AMDGPU_MAX_COMPUTE_RINGS - 1;
+ int i;
+
+ /* use one drm sched array, gfx.compute_sched to store both high and
+* normal priority drm compute schedulers */
+ for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+ if (!adev->gfx.compute_ring[i].has_high_prio)
+ adev->gfx.compute_sched[num_compute_sched_normal++] =
+ &adev->gfx.compute_ring[i].sched;
+ else
+ adev->gfx.compute_sched[num_compute_sched_high--] =
+ &adev->gfx.compute_ring[i].sched;
+ }
+
+ /* compute ring only has two priority for now */
+ i = AMDGPU_GFX_PIPE_PRIO_NORMAL;
+ adev->gfx.compute_prio_sched[i] = &adev->gfx.compute_sched[0];
+ adev->gfx.num_compute
[PATCH v7 1/4] drm/amdgpu: set compute queue priority at mqd_init
We were changing compute ring priority while rings were being used
before every job submission which is not recommended. This patch
sets compute queue priority at mqd initialization for gfx8, gfx9 and
gfx10.
Policy: make queue 0 of each pipe as high priority compute queue
High/normal priority compute sched lists are generated from set of high/normal
priority compute queues. At context creation, entity of compute queue
get a sched list from high or normal priority depending on ctx->priority
Signed-off-by: Nirmoy Das
Acked-by: Alex Deucher
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 4 --
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 53
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 8
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h | 16 ++-
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 6 ---
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 1 +
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 19 +
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c| 23 --
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c| 20 +
9 files changed, 129 insertions(+), 21 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index f397ff97b4e4..8304d0c87899 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -1205,7 +1205,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
struct drm_sched_entity *entity = p->entity;
enum drm_sched_priority priority;
- struct amdgpu_ring *ring;
struct amdgpu_bo_list_entry *e;
struct amdgpu_job *job;
uint64_t seq;
@@ -1258,9 +1257,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
priority = job->base.s_priority;
drm_sched_entity_push_job(&job->base, entity);
- ring = to_amdgpu_ring(entity->rq->sched);
- amdgpu_ring_priority_get(ring, priority);
-
amdgpu_vm_move_to_lru_tail(p->adev, &fpriv->vm);
ttm_eu_fence_buffer_objects(&p->ticket, &p->validated, p->fence);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 94a6c42f29ea..1d05eb64b885 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -61,12 +61,24 @@ static int amdgpu_ctx_priority_permit(struct drm_file *filp,
return -EACCES;
}
+static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_compute_prio(enum
drm_sched_priority prio)
+{
+ switch (prio) {
+ case DRM_SCHED_PRIORITY_HIGH_HW:
+ case DRM_SCHED_PRIORITY_KERNEL:
+ return AMDGPU_GFX_PIPE_PRIO_HIGH;
+ default:
+ return AMDGPU_GFX_PIPE_PRIO_NORMAL;
+ }
+}
+
static int amdgpu_ctx_init_entity(struct amdgpu_ctx *ctx, const u32 hw_ip,
const u32 ring)
{
struct amdgpu_device *adev = ctx->adev;
struct amdgpu_ctx_entity *entity;
struct drm_gpu_scheduler **scheds = NULL, *sched = NULL;
unsigned num_scheds = 0;
+ enum gfx_pipe_priority hw_prio;
enum drm_sched_priority priority;
int r;
@@ -85,8 +97,9 @@ static int amdgpu_ctx_init_entity(struct amdgpu_ctx *ctx,
const u32 hw_ip, const
num_scheds = 1;
break;
case AMDGPU_HW_IP_COMPUTE:
- scheds = adev->gfx.compute_sched;
- num_scheds = adev->gfx.num_compute_sched;
+ hw_prio =
amdgpu_ctx_sched_prio_to_compute_prio(priority);
+ scheds = adev->gfx.compute_prio_sched[hw_prio];
+ num_scheds = adev->gfx.num_compute_sched[hw_prio];
break;
case AMDGPU_HW_IP_DMA:
scheds = adev->sdma.sdma_sched;
@@ -628,20 +641,46 @@ void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr)
mutex_destroy(&mgr->lock);
}
+
+static void amdgpu_ctx_init_compute_sched(struct amdgpu_device *adev)
+{
+ int num_compute_sched_normal = 0;
+ int num_compute_sched_high = AMDGPU_MAX_COMPUTE_RINGS - 1;
+ int i;
+
+ /* use one drm sched array, gfx.compute_sched to store both high and
+* normal priority drm compute schedulers */
+ for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+ if (!adev->gfx.compute_ring[i].has_high_prio)
+ adev->gfx.compute_sched[num_compute_sched_normal++] =
+ &adev->gfx.compute_ring[i].sched;
+ else
+ adev->gfx.compute_sched[num_compute_sched_high--] =
+ &adev->gfx.compute_ring[i].sched;
+ }
+
+ /* compute ring only has two priority for now */
+ i = AMDGPU_GFX_PIPE_PRIO_NORMAL;
+ adev->gfx.compute_prio_sched[i] = &adev->gfx.compute_sched[0];
+ adev->gfx.num_compute_sched[i] = num_compute_sched_normal;
+
+ i = AMDGP
