drm/amdgpu/vcn:enable priority queues for encoder
Subsystems (2)
PATCH 1/1 drm/amdgpu/vcn:enable priority queues for encoder
VCN and VCE support multiple queues with different priority. Use differnt encoder queue based on the priority set by UMD.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 35 +++++++++++++++++++++-- drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c | 14 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h | 9 ++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 14 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 8 ++++++ drivers/gpu/drm/amd/amdgpu/vce_v2_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vce_v4_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c | 5 ++-- 13 files changed, 99 insertions(+), 12 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +32 -3
@@ -73,15 +73,44 @@ static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_compute_prio(enum drm_sch } } +static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_vcn_prio (enum drm_sched_priority prio) +{ +switch (prio) { +case DRM_SCHED_PRIORITY_HIGH : +return AMDGPU_VCN_ENC_PRIO_HIGH ; +case DRM_SCHED_PRIORITY_KERNEL : +return AMDGPU_VCN_ENC_PRIO_VERY_HIGH ; +default : +return AMDGPU_VCN_ENC_PRIO_NORMAL ; + } +} + +static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_vce_prio (enum drm_sched_priority prio) +{ +switch (prio) { +case DRM_SCHED_PRIORITY_HIGH : +return AMDGPU_VCE_ENC_PRIO_HIGH ; +case DRM_SCHED_PRIORITY_KERNEL : +return AMDGPU_VCE_ENC_PRIO_VERY_HIGH ; +default : +return AMDGPU_VCE_ENC_PRIO_NORMAL ; + } +} +static unsigned int amdgpu_ctx_prio_sched_to_hw (struct amdgpu_device *adev,enum drm_sched_priority prio,u32 hw_ip) {unsigned int hw_prio; - hw_prio = (hw_ip ==AMDGPU_HW_IP_COMPUTE ) ? -amdgpu_ctx_sched_prio_to_compute_prio (prio) : -AMDGPU_RING_PRIO_DEFAULT ; +if (hw_ip ==AMDGPU_HW_IP_COMPUTE ) + hw_prio =amdgpu_ctx_sched_prio_to_compute_prio (prio); +else if (hw_ip ==AMDGPU_HW_IP_VCN_ENC ) + hw_prio =amdgpu_ctx_sched_prio_to_vcn_prio (prio); +else if (hw_ip ==AMDGPU_HW_IP_VCE ) + hw_prio =amdgpu_ctx_sched_prio_to_vce_prio (prio); +else + hw_prio =AMDGPU_RING_PRIO_DEFAULT ; hw_ip =array_index_nospec (hw_ip,AMDGPU_HW_IP_NUM );if (adev->gpu_sched[hw_ip][hw_prio].num_scheds ==0 ) hw_prio =AMDGPU_RING_PRIO_DEFAULT ;
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +1 -1
@@ -37,7 +37,7 @@ int amdgpu_to_sched_priority(int amdgpu_priority, {switch (amdgpu_priority) {case AMDGPU_CTX_PRIORITY_VERY_HIGH : - *prio =DRM_SCHED_PRIORITY_HIGH ; + *prio =DRM_SCHED_PRIORITY_KERNEL ;break ;case AMDGPU_CTX_PRIORITY_HIGH : *prio =DRM_SCHED_PRIORITY_HIGH ;
drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +14 -0
@@ -1168,3 +1168,17 @@ int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring, long timeout)amdgpu_bo_free_kernel (&bo,NULL ,NULL );return r; } + +enum vce_enc_ring_priority get_vce_ring_prio (int index) +{ +switch (index) { +case 0 : +return AMDGPU_VCE_ENC_PRIO_NORMAL ; +case 1 : +return AMDGPU_VCE_ENC_PRIO_HIGH ; +case 2 : +return AMDGPU_VCE_ENC_PRIO_VERY_HIGH ; +default : +return AMDGPU_VCE_ENC_PRIO_NORMAL ; + } +}
drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h +9 -0
@@ -32,6 +32,13 @@#define AMDGPU_VCE_FW_53_45 ((53 << 24) | (45 << 16)) +enum vce_enc_ring_priority { +AMDGPU_VCE_ENC_PRIO_NORMAL =1 , +AMDGPU_VCE_ENC_PRIO_HIGH , +AMDGPU_VCE_ENC_PRIO_VERY_HIGH , +AMDGPU_VCE_ENC_PRIO_MAX +}; +struct amdgpu_vce {struct amdgpu_bo *vcpu_bo;uint64_t gpu_addr;
@@ -72,4 +79,6 @@ void amdgpu_vce_ring_end_use(struct amdgpu_ring *ring);unsigned amdgpu_vce_ring_get_emit_ib_size (struct amdgpu_ring *ring);unsigned amdgpu_vce_ring_get_dma_frame_size (struct amdgpu_ring *ring); +enum vce_enc_ring_priority get_vce_ring_prio (int index); +#endif
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +14 -0
@@ -951,3 +951,17 @@ int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)return r; } + +enum vcn_enc_ring_priority get_vcn_enc_ring_prio (int index) +{ +switch (index) { +case 0 : +return AMDGPU_VCN_ENC_PRIO_NORMAL ; +case 1 : +return AMDGPU_VCN_ENC_PRIO_HIGH ; +case 2 : +return AMDGPU_VCN_ENC_PRIO_VERY_HIGH ; +default : +return AMDGPU_VCN_ENC_PRIO_NORMAL ; + } +}
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +8 -0
@@ -290,6 +290,13 @@ enum vcn_ring_type { VCN_UNIFIED_RING, }; +enum vcn_enc_ring_priority { +AMDGPU_VCN_ENC_PRIO_NORMAL =1 , +AMDGPU_VCN_ENC_PRIO_HIGH , +AMDGPU_VCN_ENC_PRIO_VERY_HIGH , +AMDGPU_VCN_ENC_PRIO_MAX +}; +int amdgpu_vcn_sw_init (struct amdgpu_device *adev);int amdgpu_vcn_sw_fini (struct amdgpu_device *adev);int amdgpu_vcn_suspend (struct amdgpu_device *adev);
@@ -307,5 +314,6 @@ int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout);int amdgpu_vcn_enc_ring_test_ring (struct amdgpu_ring *ring);int amdgpu_vcn_enc_ring_test_ib (struct amdgpu_ring *ring,long timeout); +enum vcn_enc_ring_priority get_vcn_enc_ring_prio (int index);#endif
drivers/gpu/drm/amd/amdgpu/vce_v2_0.c +3 -1
@@ -431,10 +431,12 @@ static int vce_v2_0_sw_init(void *handle)return r;for (i =0 ; i < adev->vce.num_rings; i++) { +unsigned int hw_prio =get_vce_ring_prio (i); + ring = &adev->vce.ring[i];sprintf (ring->name,"vce%d" , i); r =amdgpu_ring_init (adev, ring,512 , &adev->vce.irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }
drivers/gpu/drm/amd/amdgpu/vce_v3_0.c +3 -1
@@ -440,10 +440,12 @@ static int vce_v3_0_sw_init(void *handle)return r;for (i =0 ; i < adev->vce.num_rings; i++) { +unsigned int hw_prio =get_vce_ring_prio (i); + ring = &adev->vce.ring[i];sprintf (ring->name,"vce%d" , i); r =amdgpu_ring_init (adev, ring,512 , &adev->vce.irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }
drivers/gpu/drm/amd/amdgpu/vce_v4_0.c +3 -1
@@ -463,6 +463,8 @@ static int vce_v4_0_sw_init(void *handle) }for (i =0 ; i < adev->vce.num_rings; i++) { +unsigned int hw_prio =get_vce_ring_prio (i); + ring = &adev->vce.ring[i];sprintf (ring->name,"vce%d" , i);if (amdgpu_sriov_vf (adev)) {
@@ -478,7 +480,7 @@ static int vce_v4_0_sw_init(void *handle) ring->doorbell_index = adev->doorbell_index.uvd_vce.vce_ring2_3 *2 +1 ; } r =amdgpu_ring_init (adev, ring,512 , &adev->vce.irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }
drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c +3 -1
@@ -145,10 +145,12 @@ static int vcn_v1_0_sw_init(void *handle)SOC15_REG_OFFSET (UVD ,0 , mmUVD_NO_OP);for (i =0 ; i < adev->vcn.num_enc_rings; ++i) { +unsigned int hw_prio =get_vcn_enc_ring_prio (i); + ring = &adev->vcn.inst->ring_enc[i];sprintf (ring->name,"vcn_enc%d" , i); r =amdgpu_ring_init (adev, ring,512 , &adev->vcn.inst->irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }
drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +3 -1
@@ -159,6 +159,8 @@ static int vcn_v2_0_sw_init(void *handle) adev->vcn.inst->external.nop =SOC15_REG_OFFSET (UVD ,0 , mmUVD_NO_OP);for (i =0 ; i < adev->vcn.num_enc_rings; ++i) { +unsigned int hw_prio =get_vcn_enc_ring_prio (i); + ring = &adev->vcn.inst->ring_enc[i]; ring->use_doorbell = true;if (!amdgpu_sriov_vf (adev))
@@ -167,7 +169,7 @@ static int vcn_v2_0_sw_init(void *handle) ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 <<1 ) +1 + i;sprintf (ring->name,"vcn_enc%d" , i); r =amdgpu_ring_init (adev, ring,512 , &adev->vcn.inst->irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +3 -1
@@ -194,6 +194,8 @@ static int vcn_v2_5_sw_init(void *handle)return r;for (i =0 ; i < adev->vcn.num_enc_rings; ++i) { +unsigned int hw_prio =get_vcn_enc_ring_prio (i); + ring = &adev->vcn.inst[j].ring_enc[i]; ring->use_doorbell = true;
@@ -203,7 +205,7 @@ static int vcn_v2_5_sw_init(void *handle)sprintf (ring->name,"vcn_enc_%d.%d" , j, i); r =amdgpu_ring_init (adev, ring,512 , &adev->vcn.inst[j].irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }
drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c +3 -2
@@ -224,6 +224,8 @@ static int vcn_v3_0_sw_init(void *handle)return r;for (j =0 ; j < adev->vcn.num_enc_rings; ++j) { +unsigned int hw_prio =get_vcn_enc_ring_prio (j); +/* VCN ENC TRAP */ r =amdgpu_irq_add_id (adev, amdgpu_ih_clientid_vcns[i], j +VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE , &adev->vcn.inst[i].irq);
@@ -239,8 +241,7 @@ static int vcn_v3_0_sw_init(void *handle) }sprintf (ring->name,"vcn_enc_%d.%d" , i, j); r =amdgpu_ring_init (adev, ring,512 , &adev->vcn.inst[i].irq,0 , -AMDGPU_RING_PRIO_DEFAULT , - &adev->vcn.inst[i].sched_score); + hw_prio, &adev->vcn.inst[i].sched_score);if (r)return r; }
Signature · 1 line
2.25.1
Quoted 77 lines · Satyajit Sahu
VCN and VCE support multiple queues with different priority. Use differnt encoder queue based on the priority set by UMD.
Signed-off-by Satyajit Sahu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 35 +++++++++++++++++++++-- drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c | 14 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h | 9 ++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 14 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 8 ++++++ drivers/gpu/drm/amd/amdgpu/vce_v2_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vce_v4_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c | 5 ++-- 13 files changed, 99 insertions(+), 12 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +32
-3@@ -73,15 +73,44 @@ static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_compute_prio(enum drm_sch } } +static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_vcn_prio (enum drm_sched_priority prio) +{ +switch (prio) { +case DRM_SCHED_PRIORITY_HIGH : +return AMDGPU_VCN_ENC_PRIO_HIGH ; +case DRM_SCHED_PRIORITY_KERNEL : +return AMDGPU_VCN_ENC_PRIO_VERY_HIGH ; +default : +return AMDGPU_VCN_ENC_PRIO_NORMAL ; + } +} + +static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_vce_prio (enum drm_sched_priority prio) +{ +switch (prio) { +case DRM_SCHED_PRIORITY_HIGH : +return AMDGPU_VCE_ENC_PRIO_HIGH ; +case DRM_SCHED_PRIORITY_KERNEL : +return AMDGPU_VCE_ENC_PRIO_VERY_HIGH ; +default : +return AMDGPU_VCE_ENC_PRIO_NORMAL ; + } +} +static unsigned int amdgpu_ctx_prio_sched_to_hw (struct amdgpu_device *adev,enum drm_sched_priority prio,u32 hw_ip) {unsigned int hw_prio; - hw_prio = (hw_ip ==AMDGPU_HW_IP_COMPUTE ) ? -amdgpu_ctx_sched_prio_to_compute_prio (prio) : -AMDGPU_RING_PRIO_DEFAULT ; +if (hw_ip ==AMDGPU_HW_IP_COMPUTE ) + hw_prio =amdgpu_ctx_sched_prio_to_compute_prio (prio); +else if (hw_ip ==AMDGPU_HW_IP_VCN_ENC ) + hw_prio =amdgpu_ctx_sched_prio_to_vcn_prio (prio); +else if (hw_ip ==AMDGPU_HW_IP_VCE ) + hw_prio =amdgpu_ctx_sched_prio_to_vce_prio (prio); +else + hw_prio =AMDGPU_RING_PRIO_DEFAULT ; hw_ip =array_index_nospec (hw_ip,AMDGPU_HW_IP_NUM );if (adev->gpu_sched[hw_ip][hw_prio].num_scheds ==0 ) hw_prio =AMDGPU_RING_PRIO_DEFAULT ;drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +0
-0
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c int amdgpu_to_sched_priority(int amdgpu_priority, · line 40
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +1
-1@@ -37,7 +37,7 @@ int amdgpu_to_sched_priority(int amdgpu_priority, {switch (amdgpu_priority) {case AMDGPU_CTX_PRIORITY_VERY_HIGH : - *prio =DRM_SCHED_PRIORITY_HIGH ; + *prio =DRM_SCHED_PRIORITY_KERNEL ;
This change looks unrelated to the patch. AFAIK KERNEL priority is reserved for kernel tasks, userspace shouldn't be able to set this privilege priority.
Regards,
Nirmoy
Quoted 240 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +0
-0break ;case AMDGPU_CTX_PRIORITY_HIGH : *prio =DRM_SCHED_PRIORITY_HIGH ;drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +14
-0@@ -1168,3 +1168,17 @@ int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring, long timeout)amdgpu_bo_free_kernel (&bo,NULL ,NULL );return r; } + +enum vce_enc_ring_priority get_vce_ring_prio (int index) +{ +switch (index) { +case 0 : +return AMDGPU_VCE_ENC_PRIO_NORMAL ; +case 1 : +return AMDGPU_VCE_ENC_PRIO_HIGH ; +case 2 : +return AMDGPU_VCE_ENC_PRIO_VERY_HIGH ; +default : +return AMDGPU_VCE_ENC_PRIO_NORMAL ; + } +}drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h +9
-0@@ -32,6 +32,13 @@#define AMDGPU_VCE_FW_53_45 ((53 << 24) | (45 << 16)) +enum vce_enc_ring_priority { +AMDGPU_VCE_ENC_PRIO_NORMAL =1 , +AMDGPU_VCE_ENC_PRIO_HIGH , +AMDGPU_VCE_ENC_PRIO_VERY_HIGH , +AMDGPU_VCE_ENC_PRIO_MAX +}; +struct amdgpu_vce {struct amdgpu_bo *vcpu_bo;uint64_t gpu_addr;@@ -72,4 +79,6 @@ void amdgpu_vce_ring_end_use(struct amdgpu_ring *ring);unsigned amdgpu_vce_ring_get_emit_ib_size (struct amdgpu_ring *ring);unsigned amdgpu_vce_ring_get_dma_frame_size (struct amdgpu_ring *ring); +enum vce_enc_ring_priority get_vce_ring_prio (int index); +#endif drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +14
-0@@ -951,3 +951,17 @@ int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)return r; } + +enum vcn_enc_ring_priority get_vcn_enc_ring_prio (int index) +{ +switch (index) { +case 0 : +return AMDGPU_VCN_ENC_PRIO_NORMAL ; +case 1 : +return AMDGPU_VCN_ENC_PRIO_HIGH ; +case 2 : +return AMDGPU_VCN_ENC_PRIO_VERY_HIGH ; +default : +return AMDGPU_VCN_ENC_PRIO_NORMAL ; + } +}drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +8
-0@@ -290,6 +290,13 @@ enum vcn_ring_type { VCN_UNIFIED_RING, }; +enum vcn_enc_ring_priority { +AMDGPU_VCN_ENC_PRIO_NORMAL =1 , +AMDGPU_VCN_ENC_PRIO_HIGH , +AMDGPU_VCN_ENC_PRIO_VERY_HIGH , +AMDGPU_VCN_ENC_PRIO_MAX +}; +int amdgpu_vcn_sw_init (struct amdgpu_device *adev);int amdgpu_vcn_sw_fini (struct amdgpu_device *adev);int amdgpu_vcn_suspend (struct amdgpu_device *adev);@@ -307,5 +314,6 @@ int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout);int amdgpu_vcn_enc_ring_test_ring (struct amdgpu_ring *ring);int amdgpu_vcn_enc_ring_test_ib (struct amdgpu_ring *ring,long timeout); +enum vcn_enc_ring_priority get_vcn_enc_ring_prio (int index);#endif drivers/gpu/drm/amd/amdgpu/vce_v2_0.c +3
-1@@ -431,10 +431,12 @@ static int vce_v2_0_sw_init(void *handle)return r;for (i =0 ; i < adev->vce.num_rings; i++) { +unsigned int hw_prio =get_vce_ring_prio (i); + ring = &adev->vce.ring[i];sprintf (ring->name,"vce%d" , i); r =amdgpu_ring_init (adev, ring,512 , &adev->vce.irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }drivers/gpu/drm/amd/amdgpu/vce_v3_0.c +3
-1@@ -440,10 +440,12 @@ static int vce_v3_0_sw_init(void *handle)return r;for (i =0 ; i < adev->vce.num_rings; i++) { +unsigned int hw_prio =get_vce_ring_prio (i); + ring = &adev->vce.ring[i];sprintf (ring->name,"vce%d" , i); r =amdgpu_ring_init (adev, ring,512 , &adev->vce.irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }drivers/gpu/drm/amd/amdgpu/vce_v4_0.c +3
-1@@ -463,6 +463,8 @@ static int vce_v4_0_sw_init(void *handle) }for (i =0 ; i < adev->vce.num_rings; i++) { +unsigned int hw_prio =get_vce_ring_prio (i); + ring = &adev->vce.ring[i];sprintf (ring->name,"vce%d" , i);if (amdgpu_sriov_vf (adev)) {@@ -478,7 +480,7 @@ static int vce_v4_0_sw_init(void *handle) ring->doorbell_index = adev->doorbell_index.uvd_vce.vce_ring2_3 *2 +1 ; } r =amdgpu_ring_init (adev, ring,512 , &adev->vce.irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c +3
-1@@ -145,10 +145,12 @@ static int vcn_v1_0_sw_init(void *handle)SOC15_REG_OFFSET (UVD ,0 , mmUVD_NO_OP);for (i =0 ; i < adev->vcn.num_enc_rings; ++i) { +unsigned int hw_prio =get_vcn_enc_ring_prio (i); + ring = &adev->vcn.inst->ring_enc[i];sprintf (ring->name,"vcn_enc%d" , i); r =amdgpu_ring_init (adev, ring,512 , &adev->vcn.inst->irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +3
-1@@ -159,6 +159,8 @@ static int vcn_v2_0_sw_init(void *handle) adev->vcn.inst->external.nop =SOC15_REG_OFFSET (UVD ,0 , mmUVD_NO_OP);for (i =0 ; i < adev->vcn.num_enc_rings; ++i) { +unsigned int hw_prio =get_vcn_enc_ring_prio (i); + ring = &adev->vcn.inst->ring_enc[i]; ring->use_doorbell = true;if (!amdgpu_sriov_vf (adev))@@ -167,7 +169,7 @@ static int vcn_v2_0_sw_init(void *handle) ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 <<1 ) +1 + i;sprintf (ring->name,"vcn_enc%d" , i); r =amdgpu_ring_init (adev, ring,512 , &adev->vcn.inst->irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +3
-1@@ -194,6 +194,8 @@ static int vcn_v2_5_sw_init(void *handle)return r;for (i =0 ; i < adev->vcn.num_enc_rings; ++i) { +unsigned int hw_prio =get_vcn_enc_ring_prio (i); + ring = &adev->vcn.inst[j].ring_enc[i]; ring->use_doorbell = true;@@ -203,7 +205,7 @@ static int vcn_v2_5_sw_init(void *handle)sprintf (ring->name,"vcn_enc_%d.%d" , j, i); r =amdgpu_ring_init (adev, ring,512 , &adev->vcn.inst[j].irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c +3
-2@@ -224,6 +224,8 @@ static int vcn_v3_0_sw_init(void *handle)return r;for (j =0 ; j < adev->vcn.num_enc_rings; ++j) { +unsigned int hw_prio =get_vcn_enc_ring_prio (j); +/* VCN ENC TRAP */ r =amdgpu_irq_add_id (adev, amdgpu_ih_clientid_vcns[i], j +VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE , &adev->vcn.inst[i].irq);@@ -239,8 +241,7 @@ static int vcn_v3_0_sw_init(void *handle) }sprintf (ring->name,"vcn_enc_%d.%d" , i, j); r =amdgpu_ring_init (adev, ring,512 , &adev->vcn.inst[i].irq,0 , -AMDGPU_RING_PRIO_DEFAULT , - &adev->vcn.inst[i].sched_score); + hw_prio, &adev->vcn.inst[i].sched_score);if (r)return r; }
Quoted 89 lines · Nirmoy
Quoted 79 lines · Satyajit Sahu
VCN and VCE support multiple queues with different priority. Use differnt encoder queue based on the priority set by UMD.
Signed-off-by Satyajit Sahu <[email protected]>---
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 35 +++++++++++++++++++++--drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c| 2 +-drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c | 14 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h | 9 ++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 14 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 8 ++++++ drivers/gpu/drm/amd/amdgpu/vce_v2_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vce_v4_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c | 5 ++-- 13 files changed, 99 insertions(+), 12 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0
-0@@ -73,15 +73,44 @@ static enum gfx_pipe_priorityamdgpu_ctx_sched_prio_to_compute_prio(enum
drm_sch
}
}
+static enumgfx_pipe_priorityamdgpu_ctx_sched_prio_to_vcn_prio(enum drm_sched_priority prio)drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +12
-0+{ + switch (prio) { + case DRM_SCHED_PRIORITY_HIGH: + return AMDGPU_VCN_ENC_PRIO_HIGH; + case DRM_SCHED_PRIORITY_KERNEL: + return AMDGPU_VCN_ENC_PRIO_VERY_HIGH; + default: + return AMDGPU_VCN_ENC_PRIO_NORMAL; + } +} + +static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_vce_prio(enum
drm_sched_priorityprio)drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +11
-0+{ + switch (prio) { + case DRM_SCHED_PRIORITY_HIGH: + return AMDGPU_VCE_ENC_PRIO_HIGH; + case DRM_SCHED_PRIORITY_KERNEL: + return AMDGPU_VCE_ENC_PRIO_VERY_HIGH; + default: + return AMDGPU_VCE_ENC_PRIO_NORMAL; + } +} +static unsigned int amdgpu_ctx_prio_sched_to_hw(struct
amdgpu_device*adev, enumdrm_sched_priorityprio,
u32hw_ip)
{
unsigned inthw_prio;
-hw_prio= (hw_ip==AMDGPU_HW_IP_COMPUTE) ?
-amdgpu_ctx_sched_prio_to_compute_prio(prio):
-AMDGPU_RING_PRIO_DEFAULT;
+ if (hw_ip==AMDGPU_HW_IP_COMPUTE)
+hw_prio=amdgpu_ctx_sched_prio_to_compute_prio(prio);
+ else if (hw_ip==AMDGPU_HW_IP_VCN_ENC)
+hw_prio=amdgpu_ctx_sched_prio_to_vcn_prio(prio);
+ else if (hw_ip==AMDGPU_HW_IP_VCE)
+hw_prio=amdgpu_ctx_sched_prio_to_vce_prio(prio);
+ else
+hw_prio=AMDGPU_RING_PRIO_DEFAULT;
hw_ip=array_index_nospec(hw_ip, AMDGPU_HW_IP_NUM);
if (adev->gpu_sched[hw_ip][hw_prio].num_scheds == 0)hw_prio=AMDGPU_RING_PRIO_DEFAULT;drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +0
-0@@ -37,7 +37,7 @@ int amdgpu_to_sched_priority(int amdgpu_priority, { switch (amdgpu_priority) {Quoted 3 lines
case
AMDGPU_CTX_PRIORITY_VERY_HIGH:
- *prio =DRM_SCHED_PRIORITY_HIGH;
+ *prio =DRM_SCHED_PRIORITY_KERNEL;This change looks unrelated to the patch. AFAIK KERNEL priority is reserved for kernel tasks, userspace shouldn't be able to set this privilege priority.
Quoted 3 lines
Regards,
Nirmoy
VCE has 3 rings (General purpose, Low latency and Real time). From the user side when priority VERY_HIGH(1023) is set, real time should be chosen, if available. That is the reason i have mapped AMDGPU_CTX_PRIORITY_VERY_HIGH to DRM_SCHED_PRIORITY_KERNEL.
regards,
Satyajit
Quoted 230 lines
Quoted 230 lines
break;
caseAMDGPU_CTX_PRIORITY_HIGH:
*prio =DRM_SCHED_PRIORITY_HIGH;drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +0
-0@@ -1168,3 +1168,17 @@ int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring, long timeout) amdgpu_bo_free_kernel(&bo, NULL, NULL); return r;}
drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +14
-0+ +enum vce_enc_ring_priority get_vce_ring_prio (int index) +{ + switch(index) { + case 0: + return AMDGPU_VCE_ENC_PRIO_NORMAL; + case 1: + return AMDGPU_VCE_ENC_PRIO_HIGH; + case 2: + return AMDGPU_VCE_ENC_PRIO_VERY_HIGH; + default: + return AMDGPU_VCE_ENC_PRIO_NORMAL; + } +}drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h +7
-0@@ -32,6 +32,13 @@ #define AMDGPU_VCE_FW_53_45 ((53 << 24) | (45 << 16)) +enum vce_enc_ring_priority { + AMDGPU_VCE_ENC_PRIO_NORMAL = 1, + AMDGPU_VCE_ENC_PRIO_HIGH, + AMDGPU_VCE_ENC_PRIO_VERY_HIGH, + AMDGPU_VCE_ENC_PRIO_MAX +}; + struct amdgpu_vce { struct amdgpu_bo *vcpu_bo; uint64_t gpu_addr;@@ -72,4 +79,6 @@ void amdgpu_vce_ring_end_use(struct amdgpu_ring *ring); unsigned amdgpu_vce_ring_get_emit_ib_size(struct amdgpu_ring *ring); unsigned amdgpu_vce_ring_get_dma_frame_size(struct amdgpu_ring *ring); +enum vce_enc_ring_priority get_vce_ring_prio(int index); +#endif
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +14
-0@@ -951,3 +951,17 @@ int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout) return r; } + +enum vcn_enc_ring_priority get_vcn_enc_ring_prio (int index) +{ + switch(index) { + case 0: + return AMDGPU_VCN_ENC_PRIO_NORMAL; + case 1: + return AMDGPU_VCN_ENC_PRIO_HIGH; + case 2: + return AMDGPU_VCN_ENC_PRIO_VERY_HIGH; + default: + return AMDGPU_VCN_ENC_PRIO_NORMAL; + } +}drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +7
-0@@ -290,6 +290,13 @@ enum vcn_ring_type { VCN_UNIFIED_RING, }; +enum vcn_enc_ring_priority { + AMDGPU_VCN_ENC_PRIO_NORMAL = 1, + AMDGPU_VCN_ENC_PRIO_HIGH, + AMDGPU_VCN_ENC_PRIO_VERY_HIGH, + AMDGPU_VCN_ENC_PRIO_MAX +}; + int amdgpu_vcn_sw_init(struct amdgpu_device *adev); int amdgpu_vcn_sw_fini(struct amdgpu_device *adev); int amdgpu_vcn_suspend(struct amdgpu_device *adev);@@ -307,5 +314,6 @@ int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout); int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring); int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout); +enum vcn_enc_ring_priority get_vcn_enc_ring_prio (int index); #endifdrivers/gpu/drm/amd/amdgpu/vce_v2_0.c +3
-1@@ -431,10 +431,12 @@ static int vce_v2_0_sw_init(void *handle) return r; for (i = 0; i < adev->vce.num_rings; i++) { + unsigned int hw_prio = get_vce_ring_prio(i); + ring = &adev->vce.ring[i]; sprintf(ring->name, "vce%d", i); r = amdgpu_ring_init(adev, ring, 512, &adev->vce.irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vce_v3_0.c +3
-1@@ -440,10 +440,12 @@ static int vce_v3_0_sw_init(void *handle) return r; for (i = 0; i < adev->vce.num_rings; i++) { + unsigned int hw_prio = get_vce_ring_prio(i); + ring = &adev->vce.ring[i]; sprintf(ring->name, "vce%d", i); r = amdgpu_ring_init(adev, ring, 512, &adev->vce.irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vce_v4_0.c +3
-1@@ -463,6 +463,8 @@ static int vce_v4_0_sw_init(void *handle) } for (i = 0; i < adev->vce.num_rings; i++) { + unsigned int hw_prio = get_vce_ring_prio(i); + ring = &adev->vce.ring[i]; sprintf(ring->name, "vce%d", i); if (amdgpu_sriov_vf(adev)) {@@ -478,7 +480,7 @@ static int vce_v4_0_sw_init(void *handle) ring->doorbell_index = adev->doorbell_index.uvd_vce.vce_ring2_3 * 2 + 1; } r = amdgpu_ring_init(adev, ring, 512, &adev->vce.irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c +3
-1@@ -145,10 +145,12 @@ static int vcn_v1_0_sw_init(void *handle) SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP); for (i = 0; i < adev->vcn.num_enc_rings; ++i) { + unsigned int hw_prio = get_vcn_enc_ring_prio(i); + ring = &adev->vcn.inst->ring_enc[i]; sprintf(ring->name, "vcn_enc%d", i); r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +3
-1@@ -159,6 +159,8 @@ static int vcn_v2_0_sw_init(void *handle) adev->vcn.inst->external.nop = SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP); for (i = 0; i < adev->vcn.num_enc_rings; ++i) { + unsigned int hw_prio = get_vcn_enc_ring_prio(i); + ring = &adev->vcn.inst->ring_enc[i]; ring->use_doorbell = true; if (!amdgpu_sriov_vf(adev))@@ -167,7 +169,7 @@ static int vcn_v2_0_sw_init(void *handle) ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 1 + i; sprintf(ring->name, "vcn_enc%d", i); r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +2
-0@@ -194,6 +194,8 @@ static int vcn_v2_5_sw_init(void *handle) return r; for (i = 0; i < adev->vcn.num_enc_rings; ++i) { + unsigned int hw_prio = get_vcn_enc_ring_prio(i); + ring = &adev->vcn.inst[j].ring_enc[i]; ring->use_doorbell = true; @@ -203,7 +205,7 @@ static int vcn_v2_5_sw_init(void *handle) sprintf(ring->name, "vcn_enc_%d.%d", j, i);r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq, 0, -
AMDGPU_RING_PRIO_DEFAULT, NULL); +hw_prio, NULL);
if (r)
return r;
}drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c +3
-2@@ -224,6 +224,8 @@ static int vcn_v3_0_sw_init(void *handle) return r; for (j = 0; j < adev->vcn.num_enc_rings; ++j) { + unsigned int hw_prio = get_vcn_enc_ring_prio(j); + /* VCN ENC TRAP */ r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i], j + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);@@ -239,8 +241,7 @@ static int vcn_v3_0_sw_init(void *handle) } sprintf(ring->name, "vcn_enc_%d.%d", i, j); r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0, - AMDGPU_RING_PRIO_DEFAULT, - &adev->vcn.inst[i].sched_score); + hw_prio, &adev->vcn.inst[i].sched_score); if (r) return r; }
Quoted 92 lines · Satyajit
Quoted 90 lines · Nirmoy
Quoted 80 lines · Satyajit Sahu
VCN and VCE support multiple queues with different priority. Use differnt encoder queue based on the priority set by UMD.
Signed-off-by Satyajit Sahu <[email protected]>---
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 35
+++++++++++++++++++++--
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c| 2 +-drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c | 14 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h | 9 ++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 14 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 8 ++++++ drivers/gpu/drm/amd/amdgpu/vce_v2_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vce_v4_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c | 5 ++-- 13 files changed, 99 insertions(+), 12 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0
-0@@ -73,15 +73,44 @@ static enum gfx_pipe_priorityamdgpu_ctx_sched_prio_to_compute_prio(enum
drm_sch
}
}
+static enumgfx_pipe_priorityamdgpu_ctx_sched_prio_to_vcn_prio(enum drm_sched_priority prio)drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +12
-0+{ + switch (prio) { + case DRM_SCHED_PRIORITY_HIGH: + return AMDGPU_VCN_ENC_PRIO_HIGH; + case DRM_SCHED_PRIORITY_KERNEL: + return AMDGPU_VCN_ENC_PRIO_VERY_HIGH; + default: + return AMDGPU_VCN_ENC_PRIO_NORMAL; + } +} + +static enum gfx_pipe_priority
amdgpu_ctx_sched_prio_to_vce_prio(enum drm_sched_priority prio)drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +11
-0+{ + switch (prio) { + case DRM_SCHED_PRIORITY_HIGH: + return AMDGPU_VCE_ENC_PRIO_HIGH; + case DRM_SCHED_PRIORITY_KERNEL: + return AMDGPU_VCE_ENC_PRIO_VERY_HIGH; + default: + return AMDGPU_VCE_ENC_PRIO_NORMAL; + } +} +static unsigned int amdgpu_ctx_prio_sched_to_hw(struct
amdgpu_device*adev, enumdrm_sched_priorityprio,
u32hw_ip)
{
unsigned inthw_prio;
-hw_prio= (hw_ip==AMDGPU_HW_IP_COMPUTE) ?
-amdgpu_ctx_sched_prio_to_compute_prio(prio):
-AMDGPU_RING_PRIO_DEFAULT;
+ if (hw_ip==AMDGPU_HW_IP_COMPUTE)
+hw_prio=amdgpu_ctx_sched_prio_to_compute_prio(prio);
+ else if (hw_ip==AMDGPU_HW_IP_VCN_ENC)
+hw_prio=amdgpu_ctx_sched_prio_to_vcn_prio(prio);
+ else if (hw_ip==AMDGPU_HW_IP_VCE)
+hw_prio=amdgpu_ctx_sched_prio_to_vce_prio(prio);
+ else
+hw_prio=AMDGPU_RING_PRIO_DEFAULT;
hw_ip=array_index_nospec(hw_ip, AMDGPU_HW_IP_NUM);
if (adev->gpu_sched[hw_ip][hw_prio].num_scheds == 0)hw_prio=AMDGPU_RING_PRIO_DEFAULT;drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +0
-0@@ -37,7 +37,7 @@ int amdgpu_to_sched_priority(int amdgpu_priority, { switch (amdgpu_priority) {Quoted 3 lines
case
AMDGPU_CTX_PRIORITY_VERY_HIGH:
- *prio =DRM_SCHED_PRIORITY_HIGH;
+ *prio =DRM_SCHED_PRIORITY_KERNEL;This change looks unrelated to the patch. AFAIK KERNEL priority is reserved for kernel tasks, userspace shouldn't be able to set this privilege priority.
Quoted 5 lines
Quoted 3 lines
Regards,
Nirmoy
VCE has 3 rings (General purpose, Low latency and Real time). From the user side when priority
VERY_HIGH(1023)is set, real time should be chosen, if available. That is the reason i have mappedAMDGPU_CTX_PRIORITY_VERY_HIGHtoDRM_SCHED_PRIORITY_KERNEL.
How are low latency and real-time ring deffer ? Do we have usecase for all three types of ring ?
Apart from the priority name there is another issue with using DRM_SCHED_PRIORITY_KERNEL, jobs with kernel priority will not be punished[1]
[1] https://elixir.bootlin.com/linux/v5.13.9/source/drivers/gpu/drm/scheduler/sched_main.c#L939
May be we should have DRM_SCHED_PRIORITY_VERY_HIGH ?
Regards,
Nirmoy
Quoted 235 lines
regards,
Satyajit
Quoted 230 lines
Quoted 230 lines
break;
caseAMDGPU_CTX_PRIORITY_HIGH:
*prio =DRM_SCHED_PRIORITY_HIGH;drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +0
-0@@ -1168,3 +1168,17 @@ int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring, long timeout) amdgpu_bo_free_kernel(&bo, NULL, NULL); return r;}
drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +14
-0+ +enum vce_enc_ring_priority get_vce_ring_prio (int index) +{ + switch(index) { + case 0: + return AMDGPU_VCE_ENC_PRIO_NORMAL; + case 1: + return AMDGPU_VCE_ENC_PRIO_HIGH; + case 2: + return AMDGPU_VCE_ENC_PRIO_VERY_HIGH; + default: + return AMDGPU_VCE_ENC_PRIO_NORMAL; + } +}drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h +7
-0@@ -32,6 +32,13 @@ #define AMDGPU_VCE_FW_53_45 ((53 << 24) | (45 << 16)) +enum vce_enc_ring_priority { + AMDGPU_VCE_ENC_PRIO_NORMAL = 1, + AMDGPU_VCE_ENC_PRIO_HIGH, + AMDGPU_VCE_ENC_PRIO_VERY_HIGH, + AMDGPU_VCE_ENC_PRIO_MAX +}; + struct amdgpu_vce { struct amdgpu_bo *vcpu_bo; uint64_t gpu_addr;@@ -72,4 +79,6 @@ void amdgpu_vce_ring_end_use(struct amdgpu_ring *ring); unsigned amdgpu_vce_ring_get_emit_ib_size(struct amdgpu_ring *ring); unsigned amdgpu_vce_ring_get_dma_frame_size(struct amdgpu_ring *ring); +enum vce_enc_ring_priority get_vce_ring_prio(int index); +#endif
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +14
-0@@ -951,3 +951,17 @@ int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout) return r; } + +enum vcn_enc_ring_priority get_vcn_enc_ring_prio (int index) +{ + switch(index) { + case 0: + return AMDGPU_VCN_ENC_PRIO_NORMAL; + case 1: + return AMDGPU_VCN_ENC_PRIO_HIGH; + case 2: + return AMDGPU_VCN_ENC_PRIO_VERY_HIGH; + default: + return AMDGPU_VCN_ENC_PRIO_NORMAL; + } +}drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +7
-0@@ -290,6 +290,13 @@ enum vcn_ring_type { VCN_UNIFIED_RING, }; +enum vcn_enc_ring_priority { + AMDGPU_VCN_ENC_PRIO_NORMAL = 1, + AMDGPU_VCN_ENC_PRIO_HIGH, + AMDGPU_VCN_ENC_PRIO_VERY_HIGH, + AMDGPU_VCN_ENC_PRIO_MAX +}; + int amdgpu_vcn_sw_init(struct amdgpu_device *adev); int amdgpu_vcn_sw_fini(struct amdgpu_device *adev); int amdgpu_vcn_suspend(struct amdgpu_device *adev);@@ -307,5 +314,6 @@ int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout); int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring); int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout); +enum vcn_enc_ring_priority get_vcn_enc_ring_prio (int index); #endifdrivers/gpu/drm/amd/amdgpu/vce_v2_0.c +3
-1@@ -431,10 +431,12 @@ static int vce_v2_0_sw_init(void *handle) return r; for (i = 0; i < adev->vce.num_rings; i++) { + unsigned int hw_prio = get_vce_ring_prio(i); + ring = &adev->vce.ring[i]; sprintf(ring->name, "vce%d", i); r = amdgpu_ring_init(adev, ring, 512, &adev->vce.irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vce_v3_0.c +3
-1@@ -440,10 +440,12 @@ static int vce_v3_0_sw_init(void *handle) return r; for (i = 0; i < adev->vce.num_rings; i++) { + unsigned int hw_prio = get_vce_ring_prio(i); + ring = &adev->vce.ring[i]; sprintf(ring->name, "vce%d", i); r = amdgpu_ring_init(adev, ring, 512, &adev->vce.irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vce_v4_0.c +3
-1@@ -463,6 +463,8 @@ static int vce_v4_0_sw_init(void *handle) } for (i = 0; i < adev->vce.num_rings; i++) { + unsigned int hw_prio = get_vce_ring_prio(i); + ring = &adev->vce.ring[i]; sprintf(ring->name, "vce%d", i); if (amdgpu_sriov_vf(adev)) {@@ -478,7 +480,7 @@ static int vce_v4_0_sw_init(void *handle) ring->doorbell_index = adev->doorbell_index.uvd_vce.vce_ring2_3 * 2 + 1; } r = amdgpu_ring_init(adev, ring, 512, &adev->vce.irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c +3
-1@@ -145,10 +145,12 @@ static int vcn_v1_0_sw_init(void *handle) SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP); for (i = 0; i < adev->vcn.num_enc_rings; ++i) { + unsigned int hw_prio = get_vcn_enc_ring_prio(i); + ring = &adev->vcn.inst->ring_enc[i]; sprintf(ring->name, "vcn_enc%d", i); r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +3
-1@@ -159,6 +159,8 @@ static int vcn_v2_0_sw_init(void *handle) adev->vcn.inst->external.nop = SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP); for (i = 0; i < adev->vcn.num_enc_rings; ++i) { + unsigned int hw_prio = get_vcn_enc_ring_prio(i); + ring = &adev->vcn.inst->ring_enc[i]; ring->use_doorbell = true; if (!amdgpu_sriov_vf(adev))@@ -167,7 +169,7 @@ static int vcn_v2_0_sw_init(void *handle) ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 1 + i; sprintf(ring->name, "vcn_enc%d", i); r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +2
-0@@ -194,6 +194,8 @@ static int vcn_v2_5_sw_init(void *handle) return r; for (i = 0; i < adev->vcn.num_enc_rings; ++i) { + unsigned int hw_prio = get_vcn_enc_ring_prio(i); + ring = &adev->vcn.inst[j].ring_enc[i]; ring->use_doorbell = true; @@ -203,7 +205,7 @@ static int vcn_v2_5_sw_init(void *handle) sprintf(ring->name, "vcn_enc_%d.%d", j, i);r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq, 0, -
AMDGPU_RING_PRIO_DEFAULT, NULL); +hw_prio, NULL);
if (r)
return r;
}drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c +3
-2@@ -224,6 +224,8 @@ static int vcn_v3_0_sw_init(void *handle) return r; for (j = 0; j < adev->vcn.num_enc_rings; ++j) { + unsigned int hw_prio = get_vcn_enc_ring_prio(j); + /* VCN ENC TRAP */ r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i], j + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);@@ -239,8 +241,7 @@ static int vcn_v3_0_sw_init(void *handle) } sprintf(ring->name, "vcn_enc_%d.%d", i, j); r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0, - AMDGPU_RING_PRIO_DEFAULT, - &adev->vcn.inst[i].sched_score); + hw_prio, &adev->vcn.inst[i].sched_score); if (r) return r; }
Quoted 110 lines · Nirmoy
Quoted 92 lines · Satyajit
Quoted 90 lines · Nirmoy
Quoted 80 lines · Satyajit Sahu
VCN and VCE support multiple queues with different priority. Use differnt encoder queue based on the priority set by UMD.
Signed-off-by Satyajit Sahu <[email protected]>---
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 35
+++++++++++++++++++++--
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c| 2 +-drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c | 14 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h | 9 ++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 14 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 8 ++++++ drivers/gpu/drm/amd/amdgpu/vce_v2_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vce_v4_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c | 5 ++-- 13 files changed, 99 insertions(+), 12 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0
-0@@ -73,15 +73,44 @@ static enum gfx_pipe_priorityamdgpu_ctx_sched_prio_to_compute_prio(enum
drm_sch
}
}
+static enumgfx_pipe_priorityamdgpu_ctx_sched_prio_to_vcn_prio(enum drm_sched_priority prio)drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +12
-0+{ + switch (prio) { + case DRM_SCHED_PRIORITY_HIGH: + return AMDGPU_VCN_ENC_PRIO_HIGH; + case DRM_SCHED_PRIORITY_KERNEL: + return AMDGPU_VCN_ENC_PRIO_VERY_HIGH; + default: + return AMDGPU_VCN_ENC_PRIO_NORMAL; + } +} + +static enum gfx_pipe_priority
amdgpu_ctx_sched_prio_to_vce_prio(enum drm_sched_priority prio)drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +11
-0+{ + switch (prio) { + case DRM_SCHED_PRIORITY_HIGH: + return AMDGPU_VCE_ENC_PRIO_HIGH; + case DRM_SCHED_PRIORITY_KERNEL: + return AMDGPU_VCE_ENC_PRIO_VERY_HIGH; + default: + return AMDGPU_VCE_ENC_PRIO_NORMAL; + } +} +static unsigned int amdgpu_ctx_prio_sched_to_hw(struct
amdgpu_device*adev, enumdrm_sched_priorityprio,
u32hw_ip)
{
unsigned inthw_prio;
-hw_prio= (hw_ip==AMDGPU_HW_IP_COMPUTE) ?
-amdgpu_ctx_sched_prio_to_compute_prio(prio):
-AMDGPU_RING_PRIO_DEFAULT;
+ if (hw_ip==AMDGPU_HW_IP_COMPUTE)
+hw_prio=amdgpu_ctx_sched_prio_to_compute_prio(prio);
+ else if (hw_ip==AMDGPU_HW_IP_VCN_ENC)
+hw_prio=amdgpu_ctx_sched_prio_to_vcn_prio(prio);
+ else if (hw_ip==AMDGPU_HW_IP_VCE)
+hw_prio=amdgpu_ctx_sched_prio_to_vce_prio(prio);
+ else
+hw_prio=AMDGPU_RING_PRIO_DEFAULT;
hw_ip=array_index_nospec(hw_ip, AMDGPU_HW_IP_NUM);
if (adev->gpu_sched[hw_ip][hw_prio].num_scheds == 0)hw_prio=AMDGPU_RING_PRIO_DEFAULT;drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +0
-0@@ -37,7 +37,7 @@ int amdgpu_to_sched_priority(int amdgpu_priority, { switch (amdgpu_priority) {Quoted 3 lines
case
AMDGPU_CTX_PRIORITY_VERY_HIGH:
- *prio =DRM_SCHED_PRIORITY_HIGH;
+ *prio =DRM_SCHED_PRIORITY_KERNEL;This change looks unrelated to the patch. AFAIK KERNEL priority is reserved for kernel tasks, userspace shouldn't be able to set this privilege priority.
Quoted 5 lines
Quoted 3 lines
Regards,
Nirmoy
VCE has 3 rings (General purpose, Low latency and Real time). From the user side when priority
VERY_HIGH(1023)is set, real time should be chosen, if available. That is the reason i have mappedAMDGPU_CTX_PRIORITY_VERY_HIGHtoDRM_SCHED_PRIORITY_KERNEL.How are low latency and real-time ring deffer ? Do we have usecase for all three types of ring ?
Apart from the priority name there is another issue with using
DRM_SCHED_PRIORITY_KERNEL, jobs with kernel priority will not be punished[1][1] https://elixir.bootlin.com/linux/v5.13.9/source/drivers/gpu/drm/scheduler/sched_main.c#L939
Quoted 6 lines
May be we should have
DRM_SCHED_PRIORITY_VERY_HIGH?Regards,
Nirmoy
Low latency and real time rings are different rings with different priorities. Adding DRM_SCHED_PRIORITY_VERY_HIGH sounds good. I will send the patch again with that modification.
regards,
Satyajit
Quoted 237 lines
Quoted 237 lines
regards,
Satyajit
Quoted 230 lines
Quoted 230 lines
break;
caseAMDGPU_CTX_PRIORITY_HIGH:
*prio =DRM_SCHED_PRIORITY_HIGH;drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +0
-0@@ -1168,3 +1168,17 @@ int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring, long timeout) amdgpu_bo_free_kernel(&bo, NULL, NULL); return r;}
drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +14
-0+ +enum vce_enc_ring_priority get_vce_ring_prio (int index) +{ + switch(index) { + case 0: + return AMDGPU_VCE_ENC_PRIO_NORMAL; + case 1: + return AMDGPU_VCE_ENC_PRIO_HIGH; + case 2: + return AMDGPU_VCE_ENC_PRIO_VERY_HIGH; + default: + return AMDGPU_VCE_ENC_PRIO_NORMAL; + } +}drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h +7
-0@@ -32,6 +32,13 @@ #define AMDGPU_VCE_FW_53_45 ((53 << 24) | (45 << 16)) +enum vce_enc_ring_priority { + AMDGPU_VCE_ENC_PRIO_NORMAL = 1, + AMDGPU_VCE_ENC_PRIO_HIGH, + AMDGPU_VCE_ENC_PRIO_VERY_HIGH, + AMDGPU_VCE_ENC_PRIO_MAX +}; + struct amdgpu_vce { struct amdgpu_bo *vcpu_bo; uint64_t gpu_addr;@@ -72,4 +79,6 @@ void amdgpu_vce_ring_end_use(struct amdgpu_ring *ring); unsigned amdgpu_vce_ring_get_emit_ib_size(struct amdgpu_ring *ring); unsigned amdgpu_vce_ring_get_dma_frame_size(struct amdgpu_ring *ring); +enum vce_enc_ring_priority get_vce_ring_prio(int index); +#endif
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +14
-0@@ -951,3 +951,17 @@ int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout) return r; } + +enum vcn_enc_ring_priority get_vcn_enc_ring_prio (int index) +{ + switch(index) { + case 0: + return AMDGPU_VCN_ENC_PRIO_NORMAL; + case 1: + return AMDGPU_VCN_ENC_PRIO_HIGH; + case 2: + return AMDGPU_VCN_ENC_PRIO_VERY_HIGH; + default: + return AMDGPU_VCN_ENC_PRIO_NORMAL; + } +}drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +7
-0@@ -290,6 +290,13 @@ enum vcn_ring_type { VCN_UNIFIED_RING, }; +enum vcn_enc_ring_priority { + AMDGPU_VCN_ENC_PRIO_NORMAL = 1, + AMDGPU_VCN_ENC_PRIO_HIGH, + AMDGPU_VCN_ENC_PRIO_VERY_HIGH, + AMDGPU_VCN_ENC_PRIO_MAX +}; + int amdgpu_vcn_sw_init(struct amdgpu_device *adev); int amdgpu_vcn_sw_fini(struct amdgpu_device *adev); int amdgpu_vcn_suspend(struct amdgpu_device *adev);@@ -307,5 +314,6 @@ int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout); int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring); int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout); +enum vcn_enc_ring_priority get_vcn_enc_ring_prio (int index); #endifdrivers/gpu/drm/amd/amdgpu/vce_v2_0.c +3
-1@@ -431,10 +431,12 @@ static int vce_v2_0_sw_init(void *handle) return r; for (i = 0; i < adev->vce.num_rings; i++) { + unsigned int hw_prio = get_vce_ring_prio(i); + ring = &adev->vce.ring[i]; sprintf(ring->name, "vce%d", i); r = amdgpu_ring_init(adev, ring, 512, &adev->vce.irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vce_v3_0.c +3
-1@@ -440,10 +440,12 @@ static int vce_v3_0_sw_init(void *handle) return r; for (i = 0; i < adev->vce.num_rings; i++) { + unsigned int hw_prio = get_vce_ring_prio(i); + ring = &adev->vce.ring[i]; sprintf(ring->name, "vce%d", i); r = amdgpu_ring_init(adev, ring, 512, &adev->vce.irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vce_v4_0.c +3
-1@@ -463,6 +463,8 @@ static int vce_v4_0_sw_init(void *handle) } for (i = 0; i < adev->vce.num_rings; i++) { + unsigned int hw_prio = get_vce_ring_prio(i); + ring = &adev->vce.ring[i]; sprintf(ring->name, "vce%d", i); if (amdgpu_sriov_vf(adev)) {@@ -478,7 +480,7 @@ static int vce_v4_0_sw_init(void *handle) ring->doorbell_index = adev->doorbell_index.uvd_vce.vce_ring2_3 * 2 + 1; } r = amdgpu_ring_init(adev, ring, 512, &adev->vce.irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c +3
-1@@ -145,10 +145,12 @@ static int vcn_v1_0_sw_init(void *handle) SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP); for (i = 0; i < adev->vcn.num_enc_rings; ++i) { + unsigned int hw_prio = get_vcn_enc_ring_prio(i); + ring = &adev->vcn.inst->ring_enc[i]; sprintf(ring->name, "vcn_enc%d", i); r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +3
-1@@ -159,6 +159,8 @@ static int vcn_v2_0_sw_init(void *handle) adev->vcn.inst->external.nop = SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP); for (i = 0; i < adev->vcn.num_enc_rings; ++i) { + unsigned int hw_prio = get_vcn_enc_ring_prio(i); + ring = &adev->vcn.inst->ring_enc[i]; ring->use_doorbell = true; if (!amdgpu_sriov_vf(adev))@@ -167,7 +169,7 @@ static int vcn_v2_0_sw_init(void *handle) ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 1 + i; sprintf(ring->name, "vcn_enc%d", i); r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); + hw_prio, NULL); if (r) return r; }drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +2
-0@@ -194,6 +194,8 @@ static int vcn_v2_5_sw_init(void *handle) return r; for (i = 0; i < adev->vcn.num_enc_rings; ++i) { + unsigned int hw_prio = get_vcn_enc_ring_prio(i); + ring = &adev->vcn.inst[j].ring_enc[i]; ring->use_doorbell = true; @@ -203,7 +205,7 @@ static int vcn_v2_5_sw_init(void *handle) sprintf(ring->name, "vcn_enc_%d.%d", j, i);r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq, 0, -
AMDGPU_RING_PRIO_DEFAULT, NULL); +hw_prio, NULL);
if (r)
return r;
}drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c +3
-2@@ -224,6 +224,8 @@ static int vcn_v3_0_sw_init(void *handle) return r; for (j = 0; j < adev->vcn.num_enc_rings; ++j) { + unsigned int hw_prio = get_vcn_enc_ring_prio(j); + /* VCN ENC TRAP */ r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i], j + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);@@ -239,8 +241,7 @@ static int vcn_v3_0_sw_init(void *handle) } sprintf(ring->name, "vcn_enc_%d.%d", i, j); r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0, - AMDGPU_RING_PRIO_DEFAULT, - &adev->vcn.inst[i].sched_score); + hw_prio, &adev->vcn.inst[i].sched_score); if (r) return r; }
Hi Satyajit,
Quoted 42 lines · Satyajit Sahu
VCN and VCE support multiple queues with different priority. Use differnt encoder queue based on the priority set by UMD.
Signed-off-by Satyajit Sahu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 35 +++++++++++++++++++++-- drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c | 14 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h | 9 ++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 14 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 8 ++++++ drivers/gpu/drm/amd/amdgpu/vce_v2_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vce_v4_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 4 ++- drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c | 5 ++-- 13 files changed, 99 insertions(+), 12 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +14
-0@@ -73,15 +73,44 @@ static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_compute_prio(enum drm_sch } } +static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_vcn_prio (enum drm_sched_priority prio) +{ +switch (prio) { +case DRM_SCHED_PRIORITY_HIGH : +return AMDGPU_VCN_ENC_PRIO_HIGH ; +case DRM_SCHED_PRIORITY_KERNEL : +return AMDGPU_VCN_ENC_PRIO_VERY_HIGH ; +default : +return AMDGPU_VCN_ENC_PRIO_NORMAL ; + } +} + +static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_vce_prio (enum drm_sched_priority prio) +{
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_compute_prio(enum drm_sch · line 99
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +10
-0+switch (prio) { +case DRM_SCHED_PRIORITY_HIGH : +return AMDGPU_VCE_ENC_PRIO_HIGH ; +case DRM_SCHED_PRIORITY_KERNEL : +return AMDGPU_VCE_ENC_PRIO_VERY_HIGH ; +default : +return AMDGPU_VCE_ENC_PRIO_NORMAL ; + } +} +
As it is also mentioned by Nirmoy, we can't use kernel priority for VCE/VCN ENC work due to the delay/stale punishment.
But adding DRM_SCHED_PRIORITY_VERY_HIGH also might not be a good way as other drivers might not be very interested in that (or are they ?)
I can think of two ways to go ahead from here:
Mrthod 1:
=========
- Introduce DRM_SCHED_PRIORITY_VERY_HIGH in the DRM scheduler layer, and demo a usecase of the same in our high priority VCN queues (in fact this should work for high priority gfx queues as well)
- Check the feedback from the community
If there are not many consumer or there is not much interest around:
Mehod 2:
========
- Add a AMDGPU specific priority flag while submitting the work
- Use this flag to keep the DRM scheduler level: DRM_SCHED_PRIORITY_HIGH but the AMD scheduler level: AMDGPU_VCE_ENC_PRIO_VERY_HIGH
- Pass this flag parameter to amdgpu_ctx_sched_prio_to_vcn_prio/amdgpu_ctx_sched_prio_to_vce_prio functions and set the proper scheduler level which will allow AMDGPU driver to schedule it in the correct ring
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0
-1static unsigned int amdgpu_ctx_prio_sched_to_hw (struct amdgpu_device *adev,enum drm_sched_priority prio,u32 hw_ip) {unsigned int hw_prio; - hw_prio = (hw_ip == AMDGPU_HW_IP_COMPUTE) ?
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_compute_prio(enum drm_sch · line 113
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +8
-2- amdgpu_ctx_sched_prio_to_compute_prio(prio) : -AMDGPU_RING_PRIO_DEFAULT ; +if (hw_ip ==AMDGPU_HW_IP_COMPUTE ) + hw_prio =amdgpu_ctx_sched_prio_to_compute_prio (prio); +else if (hw_ip ==AMDGPU_HW_IP_VCN_ENC ) + hw_prio =amdgpu_ctx_sched_prio_to_vcn_prio (prio); +else if (hw_ip ==AMDGPU_HW_IP_VCE ) + hw_prio =amdgpu_ctx_sched_prio_to_vce_prio (prio); +else + hw_prio =AMDGPU_RING_PRIO_DEFAULT ;
Move to switch(case) probably for the better appearance ?
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c int amdgpu_to_sched_priority(int amdgpu_priority, · line 40
Quoted 13 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0
-0hw_ip =array_index_nospec (hw_ip,AMDGPU_HW_IP_NUM );if (adev->gpu_sched[hw_ip][hw_prio].num_scheds ==0 ) hw_prio =AMDGPU_RING_PRIO_DEFAULT ;drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +1
-1@@ -37,7 +37,7 @@ int amdgpu_to_sched_priority(int amdgpu_priority, {switch (amdgpu_priority) {case AMDGPU_CTX_PRIORITY_VERY_HIGH : - *prio =DRM_SCHED_PRIORITY_HIGH ; + *prio =DRM_SCHED_PRIORITY_KERNEL ;
as mentioned above
Quoted 11 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +0
-0break ;case AMDGPU_CTX_PRIORITY_HIGH : *prio =DRM_SCHED_PRIORITY_HIGH ;drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +0
-0@@ -1168,3 +1168,17 @@ int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring, long timeout)amdgpu_bo_free_kernel (&bo,NULL ,NULL );return r; }
drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring, long timeout) · line 1180
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +10
-0+ +enum vce_enc_ring_priority get_vce_ring_prio (int index) +{ +switch (index) { +case 0 : +return AMDGPU_VCE_ENC_PRIO_NORMAL ; +case 1 : +return AMDGPU_VCE_ENC_PRIO_HIGH ; +case 2 : +return AMDGPU_VCE_ENC_PRIO_VERY_HIGH ;
Please use AMDGPU_VCE_ENC_PRIO_NORMAL/HIGH/VERY_HIGH instead of 0/1/2 for better readability.
Quoted 37 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +4
-0+default : +return AMDGPU_VCE_ENC_PRIO_NORMAL ; + } +}drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h +9
-0@@ -32,6 +32,13 @@#define AMDGPU_VCE_FW_53_45 ((53 << 24) | (45 << 16)) +enum vce_enc_ring_priority { +AMDGPU_VCE_ENC_PRIO_NORMAL =1 , +AMDGPU_VCE_ENC_PRIO_HIGH , +AMDGPU_VCE_ENC_PRIO_VERY_HIGH , +AMDGPU_VCE_ENC_PRIO_MAX +}; +struct amdgpu_vce {struct amdgpu_bo *vcpu_bo;uint64_t gpu_addr;@@ -72,4 +79,6 @@ void amdgpu_vce_ring_end_use(struct amdgpu_ring *ring);unsigned amdgpu_vce_ring_get_emit_ib_size (struct amdgpu_ring *ring);unsigned amdgpu_vce_ring_get_dma_frame_size (struct amdgpu_ring *ring); +enum vce_enc_ring_priority get_vce_ring_prio (int index); +#endif drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +0
-0@@ -951,3 +951,17 @@ int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)return r; }
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout) · line 963
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +10
-0+ +enum vcn_enc_ring_priority get_vcn_enc_ring_prio (int index) +{ +switch (index) { +case 0 : +return AMDGPU_VCN_ENC_PRIO_NORMAL ; +case 1 : +return AMDGPU_VCN_ENC_PRIO_HIGH ; +case 2 : +return AMDGPU_VCN_ENC_PRIO_VERY_HIGH ;
Same here
Quoted 164 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +4
-0+default : +return AMDGPU_VCN_ENC_PRIO_NORMAL ; + } +}drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +8
-0@@ -290,6 +290,13 @@ enum vcn_ring_type { VCN_UNIFIED_RING, }; +enum vcn_enc_ring_priority { +AMDGPU_VCN_ENC_PRIO_NORMAL =1 , +AMDGPU_VCN_ENC_PRIO_HIGH , +AMDGPU_VCN_ENC_PRIO_VERY_HIGH , +AMDGPU_VCN_ENC_PRIO_MAX +}; +int amdgpu_vcn_sw_init (struct amdgpu_device *adev);int amdgpu_vcn_sw_fini (struct amdgpu_device *adev);int amdgpu_vcn_suspend (struct amdgpu_device *adev);@@ -307,5 +314,6 @@ int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout);int amdgpu_vcn_enc_ring_test_ring (struct amdgpu_ring *ring);int amdgpu_vcn_enc_ring_test_ib (struct amdgpu_ring *ring,long timeout); +enum vcn_enc_ring_priority get_vcn_enc_ring_prio (int index);#endif drivers/gpu/drm/amd/amdgpu/vce_v2_0.c +3
-1@@ -431,10 +431,12 @@ static int vce_v2_0_sw_init(void *handle)return r;for (i =0 ; i < adev->vce.num_rings; i++) { +unsigned int hw_prio =get_vce_ring_prio (i); + ring = &adev->vce.ring[i];sprintf (ring->name,"vce%d" , i); r =amdgpu_ring_init (adev, ring,512 , &adev->vce.irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }drivers/gpu/drm/amd/amdgpu/vce_v3_0.c +3
-1@@ -440,10 +440,12 @@ static int vce_v3_0_sw_init(void *handle)return r;for (i =0 ; i < adev->vce.num_rings; i++) { +unsigned int hw_prio =get_vce_ring_prio (i); + ring = &adev->vce.ring[i];sprintf (ring->name,"vce%d" , i); r =amdgpu_ring_init (adev, ring,512 , &adev->vce.irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }drivers/gpu/drm/amd/amdgpu/vce_v4_0.c +3
-1@@ -463,6 +463,8 @@ static int vce_v4_0_sw_init(void *handle) }for (i =0 ; i < adev->vce.num_rings; i++) { +unsigned int hw_prio =get_vce_ring_prio (i); + ring = &adev->vce.ring[i];sprintf (ring->name,"vce%d" , i);if (amdgpu_sriov_vf (adev)) {@@ -478,7 +480,7 @@ static int vce_v4_0_sw_init(void *handle) ring->doorbell_index = adev->doorbell_index.uvd_vce.vce_ring2_3 *2 +1 ; } r =amdgpu_ring_init (adev, ring,512 , &adev->vce.irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c +3
-1@@ -145,10 +145,12 @@ static int vcn_v1_0_sw_init(void *handle)SOC15_REG_OFFSET (UVD ,0 , mmUVD_NO_OP);for (i =0 ; i < adev->vcn.num_enc_rings; ++i) { +unsigned int hw_prio =get_vcn_enc_ring_prio (i); + ring = &adev->vcn.inst->ring_enc[i];sprintf (ring->name,"vcn_enc%d" , i); r =amdgpu_ring_init (adev, ring,512 , &adev->vcn.inst->irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +3
-1@@ -159,6 +159,8 @@ static int vcn_v2_0_sw_init(void *handle) adev->vcn.inst->external.nop =SOC15_REG_OFFSET (UVD ,0 , mmUVD_NO_OP);for (i =0 ; i < adev->vcn.num_enc_rings; ++i) { +unsigned int hw_prio =get_vcn_enc_ring_prio (i); + ring = &adev->vcn.inst->ring_enc[i]; ring->use_doorbell = true;if (!amdgpu_sriov_vf (adev))@@ -167,7 +169,7 @@ static int vcn_v2_0_sw_init(void *handle) ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 <<1 ) +1 + i;sprintf (ring->name,"vcn_enc%d" , i); r =amdgpu_ring_init (adev, ring,512 , &adev->vcn.inst->irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +3
-1@@ -194,6 +194,8 @@ static int vcn_v2_5_sw_init(void *handle)return r;for (i =0 ; i < adev->vcn.num_enc_rings; ++i) { +unsigned int hw_prio =get_vcn_enc_ring_prio (i); + ring = &adev->vcn.inst[j].ring_enc[i]; ring->use_doorbell = true;@@ -203,7 +205,7 @@ static int vcn_v2_5_sw_init(void *handle)sprintf (ring->name,"vcn_enc_%d.%d" , j, i); r =amdgpu_ring_init (adev, ring,512 , &adev->vcn.inst[j].irq,0 , -AMDGPU_RING_PRIO_DEFAULT ,NULL ); + hw_prio,NULL );if (r)return r; }drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c +2
-0@@ -224,6 +224,8 @@ static int vcn_v3_0_sw_init(void *handle)return r;for (j =0 ; j < adev->vcn.num_enc_rings; ++j) { +unsigned int hw_prio =get_vcn_enc_ring_prio (j); +/* VCN ENC TRAP */ r =amdgpu_irq_add_id (adev, amdgpu_ih_clientid_vcns[i], j +VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE , &adev->vcn.inst[i].irq);
drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c static int vcn_v3_0_sw_init(void *handle) · line 247
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c +1
-2@@ -239,8 +241,7 @@ static int vcn_v3_0_sw_init(void *handle) }sprintf (ring->name,"vcn_enc_%d.%d" , i, j); r =amdgpu_ring_init (adev, ring,512 , &adev->vcn.inst[i].irq,0 , -AMDGPU_RING_PRIO_DEFAULT , - &adev->vcn.inst[i].sched_score); + hw_prio, &adev->vcn.inst[i].sched_score);if (r)return r; }
I think this patch can be split into 2,
patch 1: introduce the priority core logic and functions patch 2: consume priority logic across various vcn generations sw_init
functions (vcn_v1_sw_init, vcn_v2_sw_init, vcn_v2_5_sw_init and so on)
- Shashank