[PATCH 3/4 v2] amd/amdgpu: add sched array to IPs with multiple run-queues
This sched array can be passed on to entity creation routine
instead of manually creating such sched array on every context creation.
Signed-off-by: Nirmoy Das
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c| 113 +
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h| 3 +
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h| 4 +
drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h| 9 +-
7 files changed, 89 insertions(+), 46 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 1d6850af9908..165d1a397266 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -74,7 +74,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
struct amdgpu_ctx *ctx)
{
unsigned num_entities = amdgpu_ctx_total_num_entities();
- unsigned i, j, k;
+ unsigned i, j;
int r;
if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX)
@@ -121,73 +121,56 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
ctx->override_priority = DRM_SCHED_PRIORITY_UNSET;
for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
- struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
- struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
- unsigned num_rings = 0;
- unsigned num_rqs = 0;
+ struct drm_gpu_scheduler **scheds;
+ struct drm_gpu_scheduler *sched;
+ unsigned num_scheds = 0;
switch (i) {
case AMDGPU_HW_IP_GFX:
- rings[0] = &adev->gfx.gfx_ring[0];
- num_rings = 1;
+ scheds = adev->gfx.gfx_sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_COMPUTE:
- for (j = 0; j < adev->gfx.num_compute_rings; ++j)
- rings[j] = &adev->gfx.compute_ring[j];
- num_rings = adev->gfx.num_compute_rings;
+ scheds = adev->gfx.compute_sched;
+ num_scheds = adev->gfx.num_compute_sched;
break;
case AMDGPU_HW_IP_DMA:
- for (j = 0; j < adev->sdma.num_instances; ++j)
- rings[j] = &adev->sdma.instance[j].ring;
- num_rings = adev->sdma.num_instances;
+ scheds = adev->sdma.sdma_sched;
+ num_scheds = adev->sdma.num_sdma_sched;
break;
case AMDGPU_HW_IP_UVD:
- rings[0] = &adev->uvd.inst[0].ring;
- num_rings = 1;
+ sched = &adev->uvd.inst[0].ring.sched;
+ scheds = &sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_VCE:
- rings[0] = &adev->vce.ring[0];
- num_rings = 1;
+ sched = &adev->vce.ring[0].sched;
+ scheds = &sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_UVD_ENC:
- rings[0] = &adev->uvd.inst[0].ring_enc[0];
- num_rings = 1;
+ sched = &adev->uvd.inst[0].ring_enc[0].sched;
+ scheds = &sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_VCN_DEC:
- for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
- if (adev->vcn.harvest_config & (1 << j))
- continue;
- rings[num_rings++] =
&adev->vcn.inst[j].ring_dec;
- }
+ scheds = adev->vcn.vcn_dec_sched;
+ num_scheds = adev->vcn.num_vcn_dec_sched;
break;
case AMDGPU_HW_IP_VCN_ENC:
- for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
- if (adev->vcn.harvest_config & (1 << j))
- continue;
- for (k = 0; k < adev->vcn.num_enc_rings; ++k)
- rings[num_rings++] =
&adev->vcn.inst[j].ring_enc[k];
- }
+ scheds = adev->vcn.vcn_enc_sched;
+ num_scheds = adev->vcn.num_vcn_enc_sched;
break;
case AMDGPU_HW_IP_VCN_JPEG:
- for (j = 0; j < adev->jpeg.num_jpeg_inst; ++j) {
- if (adev->vcn.harvest_c
[PATCH 3/4 v2] amd/amdgpu: add sched array to IPs with multiple run-queues
This sched array can be passed on to entity creation routine
instead of manually creating such sched array on every context creation.
Signed-off-by: Nirmoy Das
Reviewed-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c| 113 +
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h| 3 +
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h| 4 +
drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h| 9 +-
7 files changed, 89 insertions(+), 46 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 1d6850af9908..165d1a397266 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -74,7 +74,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
struct amdgpu_ctx *ctx)
{
unsigned num_entities = amdgpu_ctx_total_num_entities();
- unsigned i, j, k;
+ unsigned i, j;
int r;
if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX)
@@ -121,73 +121,56 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
ctx->override_priority = DRM_SCHED_PRIORITY_UNSET;
for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
- struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
- struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
- unsigned num_rings = 0;
- unsigned num_rqs = 0;
+ struct drm_gpu_scheduler **scheds;
+ struct drm_gpu_scheduler *sched;
+ unsigned num_scheds = 0;
switch (i) {
case AMDGPU_HW_IP_GFX:
- rings[0] = &adev->gfx.gfx_ring[0];
- num_rings = 1;
+ scheds = adev->gfx.gfx_sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_COMPUTE:
- for (j = 0; j < adev->gfx.num_compute_rings; ++j)
- rings[j] = &adev->gfx.compute_ring[j];
- num_rings = adev->gfx.num_compute_rings;
+ scheds = adev->gfx.compute_sched;
+ num_scheds = adev->gfx.num_compute_sched;
break;
case AMDGPU_HW_IP_DMA:
- for (j = 0; j < adev->sdma.num_instances; ++j)
- rings[j] = &adev->sdma.instance[j].ring;
- num_rings = adev->sdma.num_instances;
+ scheds = adev->sdma.sdma_sched;
+ num_scheds = adev->sdma.num_sdma_sched;
break;
case AMDGPU_HW_IP_UVD:
- rings[0] = &adev->uvd.inst[0].ring;
- num_rings = 1;
+ sched = &adev->uvd.inst[0].ring.sched;
+ scheds = &sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_VCE:
- rings[0] = &adev->vce.ring[0];
- num_rings = 1;
+ sched = &adev->vce.ring[0].sched;
+ scheds = &sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_UVD_ENC:
- rings[0] = &adev->uvd.inst[0].ring_enc[0];
- num_rings = 1;
+ sched = &adev->uvd.inst[0].ring_enc[0].sched;
+ scheds = &sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_VCN_DEC:
- for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
- if (adev->vcn.harvest_config & (1 << j))
- continue;
- rings[num_rings++] =
&adev->vcn.inst[j].ring_dec;
- }
+ scheds = adev->vcn.vcn_dec_sched;
+ num_scheds = adev->vcn.num_vcn_dec_sched;
break;
case AMDGPU_HW_IP_VCN_ENC:
- for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
- if (adev->vcn.harvest_config & (1 << j))
- continue;
- for (k = 0; k < adev->vcn.num_enc_rings; ++k)
- rings[num_rings++] =
&adev->vcn.inst[j].ring_enc[k];
- }
+ scheds = adev->vcn.vcn_enc_sched;
+ num_scheds = adev->vcn.num_vcn_enc_sched;
break;
case AMDGPU_HW_IP_VCN_JPEG:
- for (j = 0; j < adev->jpeg.num_jpeg_inst; ++j) {
-
[PATCH 3/4 v2] amd/amdgpu: add sched array to IPs with multiple run-queues
This sched array can be passed on to entity creation routine
instead of manually creating such sched array on every context creation.
Signed-off-by: Nirmoy Das
Reviewed-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c| 113 +
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h| 3 +
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h| 4 +
drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h| 9 +-
7 files changed, 89 insertions(+), 46 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 1d6850af9908..165d1a397266 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -74,7 +74,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
struct amdgpu_ctx *ctx)
{
unsigned num_entities = amdgpu_ctx_total_num_entities();
- unsigned i, j, k;
+ unsigned i, j;
int r;
if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX)
@@ -121,73 +121,56 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
ctx->override_priority = DRM_SCHED_PRIORITY_UNSET;
for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
- struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
- struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
- unsigned num_rings = 0;
- unsigned num_rqs = 0;
+ struct drm_gpu_scheduler **scheds;
+ struct drm_gpu_scheduler *sched;
+ unsigned num_scheds = 0;
switch (i) {
case AMDGPU_HW_IP_GFX:
- rings[0] = &adev->gfx.gfx_ring[0];
- num_rings = 1;
+ scheds = adev->gfx.gfx_sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_COMPUTE:
- for (j = 0; j < adev->gfx.num_compute_rings; ++j)
- rings[j] = &adev->gfx.compute_ring[j];
- num_rings = adev->gfx.num_compute_rings;
+ scheds = adev->gfx.compute_sched;
+ num_scheds = adev->gfx.num_compute_sched;
break;
case AMDGPU_HW_IP_DMA:
- for (j = 0; j < adev->sdma.num_instances; ++j)
- rings[j] = &adev->sdma.instance[j].ring;
- num_rings = adev->sdma.num_instances;
+ scheds = adev->sdma.sdma_sched;
+ num_scheds = adev->sdma.num_sdma_sched;
break;
case AMDGPU_HW_IP_UVD:
- rings[0] = &adev->uvd.inst[0].ring;
- num_rings = 1;
+ sched = &adev->uvd.inst[0].ring.sched;
+ scheds = &sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_VCE:
- rings[0] = &adev->vce.ring[0];
- num_rings = 1;
+ sched = &adev->vce.ring[0].sched;
+ scheds = &sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_UVD_ENC:
- rings[0] = &adev->uvd.inst[0].ring_enc[0];
- num_rings = 1;
+ sched = &adev->uvd.inst[0].ring_enc[0].sched;
+ scheds = &sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_VCN_DEC:
- for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
- if (adev->vcn.harvest_config & (1 << j))
- continue;
- rings[num_rings++] =
&adev->vcn.inst[j].ring_dec;
- }
+ scheds = adev->vcn.vcn_dec_sched;
+ num_scheds = adev->vcn.num_vcn_dec_sched;
break;
case AMDGPU_HW_IP_VCN_ENC:
- for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
- if (adev->vcn.harvest_config & (1 << j))
- continue;
- for (k = 0; k < adev->vcn.num_enc_rings; ++k)
- rings[num_rings++] =
&adev->vcn.inst[j].ring_enc[k];
- }
+ scheds = adev->vcn.vcn_enc_sched;
+ num_scheds = adev->vcn.num_vcn_enc_sched;
break;
case AMDGPU_HW_IP_VCN_JPEG:
- for (j = 0; j < adev->jpeg.num_jpeg_inst; ++j) {
-
Re: [PATCH 3/4 v2] amd/amdgpu: add sched array to IPs with multiple run-queues
Am 10.12.19 um 13:52 schrieb Nirmoy Das:
This sched array can be passed on to entity creation routine
instead of manually creating such sched array on every context creation.
Signed-off-by: Nirmoy Das
Reviewed-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c| 113 +
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h| 3 +
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h| 4 +
drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h| 9 +-
7 files changed, 89 insertions(+), 46 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 1d6850af9908..165d1a397266 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -74,7 +74,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
struct amdgpu_ctx *ctx)
{
unsigned num_entities = amdgpu_ctx_total_num_entities();
- unsigned i, j, k;
+ unsigned i, j;
int r;
if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX)
@@ -121,73 +121,56 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
ctx->override_priority = DRM_SCHED_PRIORITY_UNSET;
for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
- struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
- struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
- unsigned num_rings = 0;
- unsigned num_rqs = 0;
+ struct drm_gpu_scheduler **scheds;
+ struct drm_gpu_scheduler *sched;
+ unsigned num_scheds = 0;
switch (i) {
case AMDGPU_HW_IP_GFX:
- rings[0] = &adev->gfx.gfx_ring[0];
- num_rings = 1;
+ scheds = adev->gfx.gfx_sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_COMPUTE:
- for (j = 0; j < adev->gfx.num_compute_rings; ++j)
- rings[j] = &adev->gfx.compute_ring[j];
- num_rings = adev->gfx.num_compute_rings;
+ scheds = adev->gfx.compute_sched;
+ num_scheds = adev->gfx.num_compute_sched;
break;
case AMDGPU_HW_IP_DMA:
- for (j = 0; j < adev->sdma.num_instances; ++j)
- rings[j] = &adev->sdma.instance[j].ring;
- num_rings = adev->sdma.num_instances;
+ scheds = adev->sdma.sdma_sched;
+ num_scheds = adev->sdma.num_sdma_sched;
break;
case AMDGPU_HW_IP_UVD:
- rings[0] = &adev->uvd.inst[0].ring;
- num_rings = 1;
+ sched = &adev->uvd.inst[0].ring.sched;
+ scheds = &sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_VCE:
- rings[0] = &adev->vce.ring[0];
- num_rings = 1;
+ sched = &adev->vce.ring[0].sched;
+ scheds = &sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_UVD_ENC:
- rings[0] = &adev->uvd.inst[0].ring_enc[0];
- num_rings = 1;
+ sched = &adev->uvd.inst[0].ring_enc[0].sched;
+ scheds = &sched;
+ num_scheds = 1;
break;
case AMDGPU_HW_IP_VCN_DEC:
- for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
- if (adev->vcn.harvest_config & (1 << j))
- continue;
- rings[num_rings++] =
&adev->vcn.inst[j].ring_dec;
- }
+ scheds = adev->vcn.vcn_dec_sched;
+ num_scheds = adev->vcn.num_vcn_dec_sched;
break;
case AMDGPU_HW_IP_VCN_ENC:
- for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
- if (adev->vcn.harvest_config & (1 << j))
- continue;
- for (k = 0; k < adev->vcn.num_enc_rings; ++k)
- rings[num_rings++] =
&adev->vcn.inst[j].ring_enc[k];
- }
+ scheds = adev->vcn.vcn_enc_sched;
+ num_scheds = adev->vcn.num_vcn_enc_sched;
break;
case AMDGPU_HW_IP_VCN_JPEG:
- for (j = 0; j < ad
