Skip to contentSkip to conversation

drm/amdgpu/gfx9: allocate queues horizontally across pipes

PATCH In review Alex Deucher 4 participants 6 messages Jun 5, 2017 – Jun 6, 2017 3 messages not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 drm/amdgpu/gfx9: allocate queues horizontally across pipes

Pipes provide better concurrency than queues, therefore we want to make sure that apps use queues from different pipes whenever possible.

Optimize for the trivial case where an app will consume rings in order, therefore we don't want adjacent rings to belong to the same pipe.

gfx9 was missed when these patches were rebased.

---
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 115 +++++++++++++++-------------------
 1 file changed, 52 insertions(+), 63 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +52 -63
index 9502353..0c48f6c 100644
@@ -1483,9 +1483,44 @@ static int gfx_v9_0_ngg_en(struct amdgpu_device *adev)
 	return 0;
 }
 
+static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
+				      int mec, int pipe, int queue)
+{
+	int r;
+	unsigned irq_type;
+	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
+
+	ring = &adev->gfx.compute_ring[ring_id];
+
+	/* mec0 is me1 */
+	ring->me = mec + 1;
+	ring->pipe = pipe;
+	ring->queue = queue;
+
+	ring->ring_obj = NULL;
+	ring->use_doorbell = true;
+	ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + ring_id;
+	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
+				+ (ring_id * GFX9_MEC_HPD_SIZE);
+	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
+
+	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
+		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
+		+ ring->pipe;
+
+	/* type-2 packets are deprecated on MEC, use type-3 instead */
+	r = amdgpu_ring_init(adev, ring, 1024,
+			     &adev->gfx.eop_irq, irq_type);
+	if (r)
+		return r;
+
+
+	return 0;
+}
+
 static int gfx_v9_0_sw_init(void *handle)
 {
-	int i, r, ring_id;
+	int i, j, k, r, ring_id;
 	struct amdgpu_ring *ring;
 	struct amdgpu_kiq *kiq;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
@@ -1547,69 +1582,23 @@ static int gfx_v9_0_sw_init(void *handle)
 			return r;
 	}
 
-	/* set up the compute queues */
-	for (i = 0, ring_id = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; i++) {
-		unsigned irq_type;
-
-		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
-			continue;
-
-		if (WARN_ON(ring_id >= AMDGPU_MAX_COMPUTE_RINGS))
-			break;
-
-		ring = &adev->gfx.compute_ring[ring_id];
-
-		/* mec0 is me1 */
-		ring->me = ((i / adev->gfx.mec.num_queue_per_pipe)
-				/ adev->gfx.mec.num_pipe_per_mec)
-				+ 1;
-		ring->pipe = (i / adev->gfx.mec.num_queue_per_pipe)
-				% adev->gfx.mec.num_pipe_per_mec;
-		ring->queue = i % adev->gfx.mec.num_queue_per_pipe;
-
-		ring->ring_obj = NULL;
-		ring->use_doorbell = true;
-		ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (ring_id * GFX9_MEC_HPD_SIZE);
-		ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + ring_id;
-		sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
-
-		irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
-			+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
-			+ ring->pipe;
-
-		/* type-2 packets are deprecated on MEC, use type-3 instead */
-		r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
-				     irq_type);
-		if (r)
-			return r;
-
-		ring_id++;
-	}
-
-	/* set up the compute queues */
-	for (i = 0, ring_id = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; i++) {
-		unsigned irq_type;
-
-		/* max 32 queues per MEC */
-		if ((i >= 32) || (i >= AMDGPU_MAX_COMPUTE_RINGS)) {
-			DRM_ERROR("Too many (%d) compute rings!\n", i);
-			break;
+	/* set up the compute queues - allocate horizontally across pipes */
+	ring_id = 0;
+	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
+		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
+			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
+				if (!amdgpu_is_mec_queue_enabled(adev, i, k, j))
+					continue;
+
+				r = gfx_v9_0_compute_ring_init(adev,
+							       ring_id,
+							       i, k, j);
+				if (r)
+					return r;
+
+				ring_id++;
+			}
 		}
-		ring = &adev->gfx.compute_ring[i];
-		ring->ring_obj = NULL;
-		ring->use_doorbell = true;
-		ring->doorbell_index = (AMDGPU_DOORBELL64_MEC_RING0 + i) << 1;
-		ring->me = 1; /* first MEC */
-		ring->pipe = i / 8;
-		ring->queue = i % 8;
-		ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (i * GFX9_MEC_HPD_SIZE);
-		sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
-		irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP + ring->pipe;
-		/* type-2 packets are deprecated on MEC, use type-3 instead */
-		r = amdgpu_ring_init(adev, ring, 1024,
-				     &adev->gfx.eop_irq, irq_type);
-		if (r)
-			return r;
 	}
 
 	r = gfx_v9_0_kiq_init(adev);
Signature · 6 lines
2.5.5

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 1496675193-4963-1-git-send-email-alexander.deucher-5C7GfCeVMHo@public.gmane.org

#2 lore raw

PATCH 2/2 drm/amdgpu/gfx9: new queue policy, take first 2 queues of each pipe

Instead of taking the first pipe and giving the rest to kfd, take the first 2 queues of each pipe.

Effectively, amdgpu and amdkfd own the same number of queues. But because the queues are spread over multiple pipes the hardware will be able to better handle concurrent compute workloads.

amdgpu goes from 1 pipe to 4 pipes, i.e. from 1 compute threads to 4 amdkfd goes from 3 pipe to 4 pipes, i.e. from 3 compute threads to 4

gfx9 was missed when this patch set was rebased to include gfx9.

---
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +2 -2
index 0c48f6c..276dc06 100644
@@ -873,8 +873,8 @@ static void gfx_v9_0_compute_queue_acquire(struct amdgpu_device *adev)
 		if (mec >= adev->gfx.mec.num_mec)
 			break;
 
-		/* policy: amdgpu owns all queues in the first pipe */
-		if (mec == 0 && pipe == 0)
+		/* policy: amdgpu owns the first two queues of the first MEC */
+		if (mec == 0 && queue < 2)
 			set_bit(i, adev->gfx.mec.queue_bitmap);
 	}
 
Signature · 6 lines
2.5.5

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 1496675193-4963-2-git-send-email-alexander.deucher-5C7GfCeVMHo@public.gmane.org

Reviewed-by
#4 lore raw

Re: [PATCH 2/2] drm/amdgpu/gfx9: new queue policy, take first 2 queues of each pipe

-Andres

Quoted 23 lines · Alex Deucher

Instead of taking the first pipe and giving the rest to kfd, take the first 2 queues of each pipe.

Effectively, amdgpu and amdkfd own the same number of queues. But because the queues are spread over multiple pipes the hardware will be able to better handle concurrent compute workloads.

amdgpu goes from 1 pipe to 4 pipes, i.e. from 1 compute threads to 4 amdkfd goes from 3 pipe to 4 pipes, i.e. from 3 compute threads to 4

gfx9 was missed when this patch set was rebased to include gfx9.

---
  drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +0 -0
index 0c48f6c..276dc06 100644

drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c static void gfx_v9_0_compute_queue_acquire(struct amdgpu_device *adev) · line 880

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +2 -2
@@ -873,8 +873,8 @@ static void gfx_v9_0_compute_queue_acquire(struct amdgpu_device *adev)
  		if (mec >= adev->gfx.mec.num_mec)
  			break;
 
-		/* policy: amdgpu owns all queues in the first pipe */
-		if (mec == 0 && pipe == 0)
+		/* policy: amdgpu owns the first two queues of the first MEC */
+		if (mec == 0 && queue < 2)
  			set_bit(i, adev->gfx.mec.queue_bitmap);
  	}
 

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Reviewed-by
#5 lore raw
In reply to a message not in the archive

Thanks for the gfx9 fixes, and sorry for the trouble. Hopefully a gfx9 card will be out soon so that I can start testing on it :)

-Andres

Quoted 140 lines · Alex Deucher

Pipes provide better concurrency than queues, therefore we want to make sure that apps use queues from different pipes whenever possible.

Optimize for the trivial case where an app will consume rings in order, therefore we don't want adjacent rings to belong to the same pipe.

gfx9 was missed when these patches were rebased.

---
  drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 115 +++++++++++++++-------------------
  1 file changed, 52 insertions(+), 63 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +52 -56
index 9502353..0c48f6c 100644
@@ -1483,9 +1483,44 @@ static int gfx_v9_0_ngg_en(struct amdgpu_device *adev)
  	return 0;
  }
 
+static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
+				      int mec, int pipe, int queue)
+{
+	int r;
+	unsigned irq_type;
+	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
+
+	ring = &adev->gfx.compute_ring[ring_id];
+
+	/* mec0 is me1 */
+	ring->me = mec + 1;
+	ring->pipe = pipe;
+	ring->queue = queue;
+
+	ring->ring_obj = NULL;
+	ring->use_doorbell = true;
+	ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + ring_id;
+	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
+				+ (ring_id * GFX9_MEC_HPD_SIZE);
+	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
+
+	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
+		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
+		+ ring->pipe;
+
+	/* type-2 packets are deprecated on MEC, use type-3 instead */
+	r = amdgpu_ring_init(adev, ring, 1024,
+			     &adev->gfx.eop_irq, irq_type);
+	if (r)
+		return r;
+
+
+	return 0;
+}
+
  static int gfx_v9_0_sw_init(void *handle)
  {
-	int i, r, ring_id;
+	int i, j, k, r, ring_id;
  	struct amdgpu_ring *ring;
  	struct amdgpu_kiq *kiq;
  	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
@@ -1547,69 +1582,23 @@ static int gfx_v9_0_sw_init(void *handle)
  			return r;
  	}
 
-	/* set up the compute queues */
-	for (i = 0, ring_id = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; i++) {
-		unsigned irq_type;
-
-		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
-			continue;
-
-		if (WARN_ON(ring_id >= AMDGPU_MAX_COMPUTE_RINGS))
-			break;
-
-		ring = &adev->gfx.compute_ring[ring_id];
-
-		/* mec0 is me1 */
-		ring->me = ((i / adev->gfx.mec.num_queue_per_pipe)
-				/ adev->gfx.mec.num_pipe_per_mec)
-				+ 1;
-		ring->pipe = (i / adev->gfx.mec.num_queue_per_pipe)
-				% adev->gfx.mec.num_pipe_per_mec;
-		ring->queue = i % adev->gfx.mec.num_queue_per_pipe;
-
-		ring->ring_obj = NULL;
-		ring->use_doorbell = true;
-		ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (ring_id * GFX9_MEC_HPD_SIZE);
-		ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + ring_id;
-		sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
-
-		irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
-			+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
-			+ ring->pipe;
-
-		/* type-2 packets are deprecated on MEC, use type-3 instead */
-		r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
-				     irq_type);
-		if (r)
-			return r;
-
-		ring_id++;
-	}
-
-	/* set up the compute queues */
-	for (i = 0, ring_id = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; i++) {
-		unsigned irq_type;
-
-		/* max 32 queues per MEC */
-		if ((i >= 32) || (i >= AMDGPU_MAX_COMPUTE_RINGS)) {
-			DRM_ERROR("Too many (%d) compute rings!\n", i);
-			break;
+	/* set up the compute queues - allocate horizontally across pipes */
+	ring_id = 0;
+	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
+		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
+			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
+				if (!amdgpu_is_mec_queue_enabled(adev, i, k, j))
+					continue;
+
+				r = gfx_v9_0_compute_ring_init(adev,
+							       ring_id,
+							       i, k, j);
+				if (r)
+					return r;
+
+				ring_id++;
+			}
  		}
-		ring = &adev->gfx.compute_ring[i];
-		ring->ring_obj = NULL;
-		ring->use_doorbell = true;
-		ring->doorbell_index = (AMDGPU_DOORBELL64_MEC_RING0 + i) << 1;
-		ring->me = 1; /* first MEC */
-		ring->pipe = i / 8;
-		ring->queue = i % 8;
-		ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (i * GFX9_MEC_HPD_SIZE);

drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c static int gfx_v9_0_sw_init(void *handle) · line 1604

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +0 -7
-		sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
-		irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP + ring->pipe;
-		/* type-2 packets are deprecated on MEC, use type-3 instead */
-		r = amdgpu_ring_init(adev, ring, 1024,
-				     &adev->gfx.eop_irq, irq_type);
-		if (r)
-			return r;
  	}
 
  	r = gfx_v9_0_kiq_init(adev);

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#6 lore raw
In reply to a message not in the archive

First is Reviewed-by: Tom St Denis <[email protected]> and second is Acked-by.

Cheers,
Tom

Quoted 140 lines · Alex Deucher

Pipes provide better concurrency than queues, therefore we want to make sure that apps use queues from different pipes whenever possible.

Optimize for the trivial case where an app will consume rings in order, therefore we don't want adjacent rings to belong to the same pipe.

gfx9 was missed when these patches were rebased.

---
  drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 115 +++++++++++++++-------------------
  1 file changed, 52 insertions(+), 63 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +52 -56
index 9502353..0c48f6c 100644
@@ -1483,9 +1483,44 @@ static int gfx_v9_0_ngg_en(struct amdgpu_device *adev)
  	return 0;
  }
 
+static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
+				      int mec, int pipe, int queue)
+{
+	int r;
+	unsigned irq_type;
+	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
+
+	ring = &adev->gfx.compute_ring[ring_id];
+
+	/* mec0 is me1 */
+	ring->me = mec + 1;
+	ring->pipe = pipe;
+	ring->queue = queue;
+
+	ring->ring_obj = NULL;
+	ring->use_doorbell = true;
+	ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + ring_id;
+	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
+				+ (ring_id * GFX9_MEC_HPD_SIZE);
+	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
+
+	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
+		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
+		+ ring->pipe;
+
+	/* type-2 packets are deprecated on MEC, use type-3 instead */
+	r = amdgpu_ring_init(adev, ring, 1024,
+			     &adev->gfx.eop_irq, irq_type);
+	if (r)
+		return r;
+
+
+	return 0;
+}
+
  static int gfx_v9_0_sw_init(void *handle)
  {
-	int i, r, ring_id;
+	int i, j, k, r, ring_id;
  	struct amdgpu_ring *ring;
  	struct amdgpu_kiq *kiq;
  	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
@@ -1547,69 +1582,23 @@ static int gfx_v9_0_sw_init(void *handle)
  			return r;
  	}
 
-	/* set up the compute queues */
-	for (i = 0, ring_id = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; i++) {
-		unsigned irq_type;
-
-		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
-			continue;
-
-		if (WARN_ON(ring_id >= AMDGPU_MAX_COMPUTE_RINGS))
-			break;
-
-		ring = &adev->gfx.compute_ring[ring_id];
-
-		/* mec0 is me1 */
-		ring->me = ((i / adev->gfx.mec.num_queue_per_pipe)
-				/ adev->gfx.mec.num_pipe_per_mec)
-				+ 1;
-		ring->pipe = (i / adev->gfx.mec.num_queue_per_pipe)
-				% adev->gfx.mec.num_pipe_per_mec;
-		ring->queue = i % adev->gfx.mec.num_queue_per_pipe;
-
-		ring->ring_obj = NULL;
-		ring->use_doorbell = true;
-		ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (ring_id * GFX9_MEC_HPD_SIZE);
-		ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + ring_id;
-		sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
-
-		irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
-			+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
-			+ ring->pipe;
-
-		/* type-2 packets are deprecated on MEC, use type-3 instead */
-		r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
-				     irq_type);
-		if (r)
-			return r;
-
-		ring_id++;
-	}
-
-	/* set up the compute queues */
-	for (i = 0, ring_id = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; i++) {
-		unsigned irq_type;
-
-		/* max 32 queues per MEC */
-		if ((i >= 32) || (i >= AMDGPU_MAX_COMPUTE_RINGS)) {
-			DRM_ERROR("Too many (%d) compute rings!\n", i);
-			break;
+	/* set up the compute queues - allocate horizontally across pipes */
+	ring_id = 0;
+	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
+		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
+			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
+				if (!amdgpu_is_mec_queue_enabled(adev, i, k, j))
+					continue;
+
+				r = gfx_v9_0_compute_ring_init(adev,
+							       ring_id,
+							       i, k, j);
+				if (r)
+					return r;
+
+				ring_id++;
+			}
  		}
-		ring = &adev->gfx.compute_ring[i];
-		ring->ring_obj = NULL;
-		ring->use_doorbell = true;
-		ring->doorbell_index = (AMDGPU_DOORBELL64_MEC_RING0 + i) << 1;
-		ring->me = 1; /* first MEC */
-		ring->pipe = i / 8;
-		ring->queue = i % 8;
-		ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (i * GFX9_MEC_HPD_SIZE);

drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c static int gfx_v9_0_sw_init(void *handle) · line 1604

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +0 -7
-		sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
-		irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP + ring->pipe;
-		/* type-2 packets are deprecated on MEC, use type-3 instead */
-		r = amdgpu_ring_init(adev, ring, 1024,
-				     &adev->gfx.eop_irq, irq_type);
-		if (r)
-			return r;
  	}
 
  	r = gfx_v9_0_kiq_init(adev);

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 692d3fb2-98de-0e11-d8bc-e255cc07b605-5C7GfCeVMHo@public.gmane.org

#8 lore raw

Fell free to add an Acked-by: Christian König <[email protected]> on both.

Christian.

Am 06.06.2017 um 13:43 schrieb Tom St Denis:

Quoted 159 lines

First is Reviewed-by: Tom St Denis <[email protected]> and second is Acked-by.

Cheers,
Tom

Quoted 153 lines · Alex Deucher

Pipes provide better concurrency than queues, therefore we want to make sure that apps use queues from different pipes whenever possible.

Optimize for the trivial case where an app will consume rings in order, therefore we don't want adjacent rings to belong to the same pipe.

gfx9 was missed when these patches were rebased.

---
  drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 115 
+++++++++++++++-------------------
  1 file changed, 52 insertions(+), 63 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +36 -47
index 9502353..0c48f6c 100644
@@ -1483,9 +1483,44 @@ static int gfx_v9_0_ngg_en(struct 
 amdgpu_device *adev)
      return 0;
  }
  +static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, 
 int ring_id,
+                      int mec, int pipe, int queue)
+{
+    int r;
+    unsigned irq_type;
+    struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
+
+    ring = &adev->gfx.compute_ring[ring_id];
+
+    /* mec0 is me1 */
+    ring->me = mec + 1;
+    ring->pipe = pipe;
+    ring->queue = queue;
+
+    ring->ring_obj = NULL;
+    ring->use_doorbell = true;
+    ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + ring_id;
+    ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
+                + (ring_id * GFX9_MEC_HPD_SIZE);
+    sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, 
 ring->queue);
+
+    irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
+        + ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
+        + ring->pipe;
+
+    /* type-2 packets are deprecated on MEC, use type-3 instead */
+    r = amdgpu_ring_init(adev, ring, 1024,
+                 &adev->gfx.eop_irq, irq_type);
+    if (r)
+        return r;
+
+
+    return 0;
+}
+
  static int gfx_v9_0_sw_init(void *handle)
  {
-    int i, r, ring_id;
+    int i, j, k, r, ring_id;
      struct amdgpu_ring *ring;
      struct amdgpu_kiq *kiq;
      struct amdgpu_device *adev = (struct amdgpu_device *)handle;
@@ -1547,69 +1582,23 @@ static int gfx_v9_0_sw_init(void *handle)
              return r;
      }
  -    /* set up the compute queues */
-    for (i = 0, ring_id = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; i++) {
-        unsigned irq_type;
-
-        if (!test_bit(i, adev->gfx.mec.queue_bitmap))
-            continue;
-
-        if (WARN_ON(ring_id >= AMDGPU_MAX_COMPUTE_RINGS))
-            break;
-
-        ring = &adev->gfx.compute_ring[ring_id];
-
-        /* mec0 is me1 */
-        ring->me = ((i / adev->gfx.mec.num_queue_per_pipe)
-                / adev->gfx.mec.num_pipe_per_mec)
-                + 1;
-        ring->pipe = (i / adev->gfx.mec.num_queue_per_pipe)
-                % adev->gfx.mec.num_pipe_per_mec;
-        ring->queue = i % adev->gfx.mec.num_queue_per_pipe;
-
-        ring->ring_obj = NULL;
-        ring->use_doorbell = true;
-        ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + 
 (ring_id * GFX9_MEC_HPD_SIZE);
-        ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + ring_id;
-        sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, 
 ring->queue);
-
-        irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
-            + ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
-            + ring->pipe;
-
-        /* type-2 packets are deprecated on MEC, use type-3 instead */
-        r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
-                     irq_type);
-        if (r)
-            return r;
-
-        ring_id++;
-    }
-
-    /* set up the compute queues */
-    for (i = 0, ring_id = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; i++) {
-        unsigned irq_type;
-
-        /* max 32 queues per MEC */
-        if ((i >= 32) || (i >= AMDGPU_MAX_COMPUTE_RINGS)) {
-            DRM_ERROR("Too many (%d) compute rings!\n", i);
-            break;
+    /* set up the compute queues - allocate horizontally across 

pipes */

drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +15 -8
+    ring_id = 0;
+    for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
+        for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
+            for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
+                if (!amdgpu_is_mec_queue_enabled(adev, i, k, j))
+                    continue;
+
+                r = gfx_v9_0_compute_ring_init(adev,
+                                   ring_id,
+                                   i, k, j);
+                if (r)
+                    return r;
+
+                ring_id++;
+            }
          }
-        ring = &adev->gfx.compute_ring[i];
-        ring->ring_obj = NULL;
-        ring->use_doorbell = true;
-        ring->doorbell_index = (AMDGPU_DOORBELL64_MEC_RING0 + i) << 1;
-        ring->me = 1; /* first MEC */
-        ring->pipe = i / 8;
-        ring->queue = i % 8;
-        ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (i * 

GFX9_MEC_HPD_SIZE);
- sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe,
ring->queue);

drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +0 -6
-        irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP + ring->pipe;
-        /* type-2 packets are deprecated on MEC, use type-3 instead */
-        r = amdgpu_ring_init(adev, ring, 1024,
-                     &adev->gfx.eop_irq, irq_type);
-        if (r)
-            return r;
Quoted 8 lines
Quoted 3 lines
      }
        r = gfx_v9_0_kiq_init(adev);

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx