CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
687 lines
40 KiB
YAML
687 lines
40 KiB
YAML
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
|
|
# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs -amdgpu-expert-scheduling-mode -run-pass si-insert-waitcnts %s -o - | FileCheck -check-prefix=GCN %s
|
|
# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs -amdgpu-waitcnt-forcezero -run-pass si-insert-waitcnts %s -o - | FileCheck %s
|
|
|
|
---
|
|
# Make sure we don't output expert mode waitcnts with amdgpu-waitcnt-forcezero flag if hazard mode isn't enabled,
|
|
# otherwise it will cause an assert.
|
|
# CHECK-LABEL: name: raw_exp
|
|
|
|
name: raw_exp
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: raw_exp
|
|
; GCN: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr0 = V_MOV_B32_e32 1082130432, implicit $exec
|
|
; GCN-NEXT: $vgpr1 = V_MOV_B32_e32 1073741824, implicit $exec
|
|
; GCN-NEXT: $vgpr2 = V_MOV_B32_e32 1065353216, implicit $exec
|
|
; GCN-NEXT: $vgpr3 = V_MOV_B32_e32 1056964608, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: EXP_DONE 15, $vgpr3, $vgpr2, $vgpr1, $vgpr0, 0, 0, 1, implicit $exec
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr0 = V_MOV_B32_e32 1082130432, implicit $exec
|
|
$vgpr1 = V_MOV_B32_e32 1073741824, implicit $exec
|
|
$vgpr2 = V_MOV_B32_e32 1065353216, implicit $exec
|
|
$vgpr3 = V_MOV_B32_e32 1056964608, implicit $exec
|
|
EXP_DONE 15, $vgpr3, $vgpr2, $vgpr1, $vgpr0, 0, 0, 1, implicit $exec
|
|
S_ENDPGM 0
|
|
|
|
...
|
|
|
|
---
|
|
name: raw_vmem
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0
|
|
|
|
; GCN-LABEL: name: raw_vmem
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr4 = V_MOV_B32_e32 $vgpr0, implicit $exec, implicit $exec
|
|
; GCN-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = IMAGE_LOAD_V4_V1_gfx12 $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
; GCN-NEXT: S_WAIT_LOADCNT 0
|
|
; GCN-NEXT: $vgpr3 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr3, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr2 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr2, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr1 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr1, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr0 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr0, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 0, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: SI_RETURN_TO_EPILOG $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
$vgpr4 = V_MOV_B32_e32 $vgpr0, implicit $exec, implicit $exec
|
|
$vgpr0_vgpr1_vgpr2_vgpr3 = IMAGE_LOAD_V4_V1_gfx12 $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
$vgpr3 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr3, implicit $mode, implicit $exec
|
|
$vgpr2 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr2, implicit $mode, implicit $exec
|
|
$vgpr1 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr1, implicit $mode, implicit $exec
|
|
$vgpr0 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr0, implicit $mode, implicit $exec
|
|
IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
SI_RETURN_TO_EPILOG $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
|
|
...
|
|
|
|
---
|
|
name: war_lds
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1
|
|
|
|
; GCN-LABEL: name: war_lds
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr5_vgpr6_vgpr7_vgpr8 = IMAGE_LOAD_V4_V1_gfx12 $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = DS_READ_B128_gfx9 $vgpr1, 0, 0, implicit $exec :: (load (s128), addrspace 3)
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: $vgpr0 = nofpexcept V_ADD_F32_e32 $vgpr5, $vgpr0, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr1 = nofpexcept V_ADD_F32_e32 $vgpr6, $vgpr1, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr2 = nofpexcept V_ADD_F32_e32 $vgpr7, $vgpr2, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr3 = nofpexcept V_ADD_F32_e32 $vgpr8, $vgpr3, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 0, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: SI_RETURN_TO_EPILOG $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
$vgpr5_vgpr6_vgpr7_vgpr8 = IMAGE_LOAD_V4_V1_gfx12 $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
$vgpr0_vgpr1_vgpr2_vgpr3 = DS_READ_B128_gfx9 $vgpr1, 0, 0, implicit $exec :: (load (s128), addrspace 3)
|
|
$vgpr0 = nofpexcept V_ADD_F32_e32 $vgpr5, $vgpr0, implicit $mode, implicit $exec
|
|
$vgpr1 = nofpexcept V_ADD_F32_e32 $vgpr6, $vgpr1, implicit $mode, implicit $exec
|
|
$vgpr2 = nofpexcept V_ADD_F32_e32 $vgpr7, $vgpr2, implicit $mode, implicit $exec
|
|
$vgpr3 = nofpexcept V_ADD_F32_e32 $vgpr8, $vgpr3, implicit $mode, implicit $exec
|
|
SI_RETURN_TO_EPILOG $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
|
|
...
|
|
|
|
---
|
|
name: war_tex
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2
|
|
|
|
; GCN-LABEL: name: war_tex
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr4 = V_MOV_B32_e32 $vgpr0, implicit $exec, implicit $exec
|
|
; GCN-NEXT: $vgpr5 = V_MOV_B32_e32 $vgpr2, implicit $exec, implicit $exec
|
|
; GCN-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = DS_READ_B128_gfx9 $vgpr1, 0, 0, implicit $exec :: (load (s128), addrspace 3)
|
|
; GCN-NEXT: S_WAIT_DSCNT 0
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: DS_WRITE_B128_gfx9 $vgpr5, $vgpr0_vgpr1_vgpr2_vgpr3, 0, 0, implicit $exec :: (store (s128), addrspace 3)
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = IMAGE_LOAD_V4_V1_gfx12 $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 0, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: SI_RETURN_TO_EPILOG $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
$vgpr4 = V_MOV_B32_e32 $vgpr0, implicit $exec, implicit $exec
|
|
$vgpr5 = V_MOV_B32_e32 $vgpr2, implicit $exec, implicit $exec
|
|
$vgpr0_vgpr1_vgpr2_vgpr3 = DS_READ_B128_gfx9 $vgpr1, 0, 0, implicit $exec :: (load (s128), addrspace 3)
|
|
DS_WRITE_B128_gfx9 $vgpr5, $vgpr0_vgpr1_vgpr2_vgpr3, 0, 0, implicit $exec :: (store (s128), addrspace 3)
|
|
$vgpr0_vgpr1_vgpr2_vgpr3 = IMAGE_LOAD_V4_V1_gfx12 $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
SI_RETURN_TO_EPILOG $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
|
|
...
|
|
|
|
---
|
|
name: war_valu
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0
|
|
|
|
; GCN-LABEL: name: war_valu
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr5_vgpr6_vgpr7_vgpr8 = IMAGE_LOAD_V4_V1_gfx12 $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
; GCN-NEXT: S_WAIT_LOADCNT 0
|
|
; GCN-NEXT: $vgpr4 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr5, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr1 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr6, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr2 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr7, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr3 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr8, implicit $mode, implicit $exec
|
|
; GCN-NEXT: IMAGE_STORE_V4_V1_gfx12 $vgpr5_vgpr6_vgpr7_vgpr8, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr0 = V_MOV_B32_e32 $vgpr4, implicit $exec, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 0, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: SI_RETURN_TO_EPILOG $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
$vgpr5_vgpr6_vgpr7_vgpr8 = IMAGE_LOAD_V4_V1_gfx12 $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
$vgpr4 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr5, implicit $mode, implicit $exec
|
|
$vgpr1 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr6, implicit $mode, implicit $exec
|
|
$vgpr2 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr7, implicit $mode, implicit $exec
|
|
$vgpr3 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr8, implicit $mode, implicit $exec
|
|
IMAGE_STORE_V4_V1_gfx12 $vgpr5_vgpr6_vgpr7_vgpr8, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
$vgpr0 = V_MOV_B32_e32 $vgpr4, implicit $exec, implicit $exec
|
|
SI_RETURN_TO_EPILOG $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
|
|
...
|
|
|
|
---
|
|
name: war_vmem
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0
|
|
|
|
; GCN-LABEL: name: war_vmem
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr4 = V_MOV_B32_e32 $vgpr0, implicit $exec, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = IMAGE_LOAD_V4_V1_gfx12 $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
; GCN-NEXT: S_WAIT_LOADCNT 0
|
|
; GCN-NEXT: $vgpr3 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr3, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr2 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr2, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr1 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr1, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr0 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr0, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 0, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: SI_RETURN_TO_EPILOG $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
$vgpr4 = V_MOV_B32_e32 $vgpr0, implicit $exec, implicit $exec
|
|
$vgpr0_vgpr1_vgpr2_vgpr3 = IMAGE_LOAD_V4_V1_gfx12 $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
$vgpr3 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr3, implicit $mode, implicit $exec
|
|
$vgpr2 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr2, implicit $mode, implicit $exec
|
|
$vgpr1 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr1, implicit $mode, implicit $exec
|
|
$vgpr0 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr0, implicit $mode, implicit $exec
|
|
IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
SI_RETURN_TO_EPILOG $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
|
|
...
|
|
|
|
---
|
|
name: waw_vmem
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0
|
|
|
|
; GCN-LABEL: name: waw_vmem
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr4 = V_MOV_B32_e32 $vgpr0, implicit $exec, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: $vgpr4_vgpr5_vgpr6_vgpr7 = IMAGE_LOAD_V4_V1_gfx12 $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
; GCN-NEXT: S_WAIT_LOADCNT 0
|
|
; GCN-NEXT: $vgpr3 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr7, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr2 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr6, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr1 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr5, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr0 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr4, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 0, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: SI_RETURN_TO_EPILOG $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
$vgpr4 = V_MOV_B32_e32 $vgpr0, implicit $exec, implicit $exec
|
|
$vgpr4_vgpr5_vgpr6_vgpr7 = IMAGE_LOAD_V4_V1_gfx12 $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
$vgpr3 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr7, implicit $mode, implicit $exec
|
|
$vgpr2 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr6, implicit $mode, implicit $exec
|
|
$vgpr1 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr5, implicit $mode, implicit $exec
|
|
$vgpr0 = nofpexcept V_ADD_F32_e32 1065353216, $vgpr4, implicit $mode, implicit $exec
|
|
SI_RETURN_TO_EPILOG $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
|
|
...
|
|
|
|
---
|
|
name: raw_valu_scratch
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: raw_valu_scratch
|
|
; GCN: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr1 = V_MOV_B32_e32 $vgpr0, implicit $exec
|
|
; GCN-NEXT: $vgpr2 = V_MOV_B32_e32 $vgpr0, implicit $exec
|
|
; GCN-NEXT: $vgpr3 = V_MOV_B32_e32 $vgpr1, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_1
|
|
; GCN-NEXT: $vgpr4 = SCRATCH_LOAD_UBYTE_SVS $vgpr2, $sgpr0, 0, 0, implicit $exec, implicit $flat_scr
|
|
$vgpr1 = V_MOV_B32_e32 $vgpr0, implicit $exec
|
|
$vgpr2 = V_MOV_B32_e32 $vgpr0, implicit $exec
|
|
$vgpr3 = V_MOV_B32_e32 $vgpr1, implicit $exec
|
|
$vgpr4 = SCRATCH_LOAD_UBYTE_SVS $vgpr2, $sgpr0, 0, 0, implicit $exec, implicit $flat_scr
|
|
...
|
|
|
|
---
|
|
# This pre-existing S_WAITCNT_DEPCTR should not be updated.
|
|
name: valu_depctr_valu
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: valu_depctr_valu
|
|
; GCN: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr0 = V_MOV_B32_e32 $vgpr0, implicit $exec
|
|
; GCN-NEXT: $vgpr1 = V_MOV_B32_e32 $vgpr0, implicit $exec
|
|
$vgpr0 = V_MOV_B32_e32 $vgpr0, implicit $exec
|
|
S_WAITCNT_DEPCTR .VmVsrc_1_VaVdst_1
|
|
$vgpr1 = V_MOV_B32_e32 $vgpr0, implicit $exec
|
|
...
|
|
|
|
---
|
|
name: test_xdl_csmacc_ooo_completion
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18
|
|
|
|
; GCN-LABEL: name: test_xdl_csmacc_ooo_completion
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: early-clobber $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = V_WMMA_F32_16X16X16_F16_w32_twoaddr 8, killed $vgpr10_vgpr11_vgpr12_vgpr13, 8, killed $vgpr14_vgpr15_vgpr16_vgpr17, 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, implicit $exec
|
|
; GCN-NEXT: $vgpr8 = V_ADD_F32_e32 244, $vgpr11, implicit $exec, implicit $mode
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
$vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = V_WMMA_F32_16X16X16_F16_w32_twoaddr 8, killed $vgpr10_vgpr11_vgpr12_vgpr13, 8, killed $vgpr14_vgpr15_vgpr16_vgpr17, 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, implicit $exec
|
|
$vgpr8 = V_ADD_F32_e32 244, $vgpr11, implicit $exec, implicit $mode
|
|
IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
...
|
|
|
|
---
|
|
name: test_dpmacc_csmacc_ooo_completion
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr11, $vgpr18
|
|
|
|
; GCN-LABEL: name: test_dpmacc_csmacc_ooo_completion
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr11, $vgpr18
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr0_vgpr1 = V_FRACT_F64_e32 $vgpr0_vgpr1, implicit $exec, implicit $mode
|
|
; GCN-NEXT: $vgpr8 = V_ADD_F32_e32 244, $vgpr11, implicit $exec, implicit $mode
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
$vgpr0_vgpr1 = V_FRACT_F64_e32 $vgpr0_vgpr1, implicit $exec, implicit $mode
|
|
$vgpr8 = V_ADD_F32_e32 244, $vgpr11, implicit $exec, implicit $mode
|
|
IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
...
|
|
|
|
---
|
|
name: test_trans_csmacc_ooo_completion
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr11, $vgpr18
|
|
|
|
; GCN-LABEL: name: test_trans_csmacc_ooo_completion
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr11, $vgpr18
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr0 = V_SQRT_F32_e32 $vgpr0, implicit $exec, implicit $mode
|
|
; GCN-NEXT: $vgpr8 = V_ADD_F32_e32 244, $vgpr11, implicit $exec, implicit $mode
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
$vgpr0 = V_SQRT_F32_e32 $vgpr0, implicit $exec, implicit $mode
|
|
$vgpr8 = V_ADD_F32_e32 244, $vgpr11, implicit $exec, implicit $mode
|
|
IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
...
|
|
|
|
---
|
|
name: test_csmacc_csmacc_io_completion
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr11, $vgpr18
|
|
|
|
; GCN-LABEL: name: test_csmacc_csmacc_io_completion
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr11, $vgpr18
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr0 = V_ADD_F32_e32 244, $vgpr0, implicit $exec, implicit $mode
|
|
; GCN-NEXT: $vgpr8 = V_ADD_F32_e32 244, $vgpr11, implicit $exec, implicit $mode
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_1
|
|
; GCN-NEXT: IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
$vgpr0 = V_ADD_F32_e32 244, $vgpr0, implicit $exec, implicit $mode
|
|
$vgpr8 = V_ADD_F32_e32 244, $vgpr11, implicit $exec, implicit $mode
|
|
IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
...
|
|
|
|
---
|
|
name: test_dpmacc_dpmacc_io_completion
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr18
|
|
|
|
; GCN-LABEL: name: test_dpmacc_dpmacc_io_completion
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr18
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr0_vgpr1 = V_FRACT_F64_e32 $vgpr0_vgpr1, implicit $exec, implicit $mode
|
|
; GCN-NEXT: $vgpr4_vgpr5 = V_TRUNC_F64_e32 $vgpr2_vgpr3, implicit $exec, implicit $mode
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_1
|
|
; GCN-NEXT: IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
$vgpr0_vgpr1 = V_FRACT_F64_e32 $vgpr0_vgpr1, implicit $exec, implicit $mode
|
|
$vgpr4_vgpr5 = V_TRUNC_F64_e32 $vgpr2_vgpr3, implicit $exec, implicit $mode
|
|
IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
...
|
|
|
|
---
|
|
name: test_trans_trans_io_completion
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr11, $vgpr18
|
|
|
|
; GCN-LABEL: name: test_trans_trans_io_completion
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr11, $vgpr18
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr0 = V_SQRT_F32_e32 $vgpr0, implicit $exec, implicit $mode
|
|
; GCN-NEXT: $vgpr8 = V_LOG_F32_e32 $vgpr11, implicit $exec, implicit $mode
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_1
|
|
; GCN-NEXT: IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
$vgpr0 = V_SQRT_F32_e32 $vgpr0, implicit $exec, implicit $mode
|
|
$vgpr8 = V_LOG_F32_e32 $vgpr11, implicit $exec, implicit $mode
|
|
IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
...
|
|
|
|
---
|
|
name: test_xdl_xdl_io_completion
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26
|
|
|
|
; GCN-LABEL: name: test_xdl_xdl_io_completion
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: early-clobber $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = V_WMMA_F32_16X16X16_F16_w32_twoaddr 8, $vgpr10_vgpr11_vgpr12_vgpr13, 8, $vgpr14_vgpr15_vgpr16_vgpr17, 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, implicit $exec
|
|
; GCN-NEXT: early-clobber $vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26 = V_WMMA_F32_16X16X16_F16_w32_twoaddr 8, killed $vgpr10_vgpr11_vgpr12_vgpr13, 8, killed $vgpr14_vgpr15_vgpr16_vgpr17, 8, killed $vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26, 0, 0, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_1
|
|
; GCN-NEXT: IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
$vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = V_WMMA_F32_16X16X16_F16_w32_twoaddr 8, $vgpr10_vgpr11_vgpr12_vgpr13, 8, $vgpr14_vgpr15_vgpr16_vgpr17, 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, implicit $exec
|
|
$vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26 = V_WMMA_F32_16X16X16_F16_w32_twoaddr 8, killed $vgpr10_vgpr11_vgpr12_vgpr13, 8, killed $vgpr14_vgpr15_vgpr16_vgpr17, 8, killed $vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26, 0, 0, implicit $exec
|
|
IMAGE_STORE_V4_V1_gfx12 $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr18, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), addrspace 8)
|
|
...
|
|
|
|
---
|
|
name: test_war_vmvsrc_vmem_vmem_io
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1
|
|
|
|
; GCN-LABEL: name: test_war_vmvsrc_vmem_vmem_io
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr8_vgpr9_vgpr10_vgpr11 = IMAGE_LOAD_V4_V1_gfx12 $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
; GCN-NEXT: $vgpr4_vgpr5_vgpr6_vgpr7 = IMAGE_LOAD_V4_V1_gfx12 $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_1
|
|
; GCN-NEXT: $vgpr0 = nofpexcept V_ADD_F32_e32 $vgpr0, $vgpr0, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr1 = nofpexcept V_ADD_F32_e32 $vgpr1, $vgpr1, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAIT_LOADCNT 0
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 0, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: SI_RETURN_TO_EPILOG $vgpr0, $vgpr1
|
|
$vgpr8_vgpr9_vgpr10_vgpr11 = IMAGE_LOAD_V4_V1_gfx12 $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
$vgpr4_vgpr5_vgpr6_vgpr7 = IMAGE_LOAD_V4_V1_gfx12 $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
$vgpr0 = nofpexcept V_ADD_F32_e32 $vgpr0, $vgpr0, implicit $mode, implicit $exec
|
|
$vgpr1 = nofpexcept V_ADD_F32_e32 $vgpr1, $vgpr1, implicit $mode, implicit $exec
|
|
SI_RETURN_TO_EPILOG $vgpr0, $vgpr1
|
|
|
|
...
|
|
|
|
---
|
|
name: test_war_vmvsrc_flat_flat_io
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
|
|
; GCN-LABEL: name: test_war_vmvsrc_flat_flat_io
|
|
; GCN: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr4 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
|
|
; GCN-NEXT: $vgpr5 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_1
|
|
; GCN-NEXT: $vgpr0 = nofpexcept V_ADD_F32_e32 $vgpr0, $vgpr1, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr2 = nofpexcept V_ADD_F32_e32 $vgpr2, $vgpr3, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 0, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: SI_RETURN_TO_EPILOG $vgpr6, $vgpr7
|
|
$vgpr4 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
|
|
$vgpr5 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr
|
|
$vgpr0 = nofpexcept V_ADD_F32_e32 $vgpr0, $vgpr1, implicit $mode, implicit $exec
|
|
$vgpr2 = nofpexcept V_ADD_F32_e32 $vgpr2, $vgpr3, implicit $mode, implicit $exec
|
|
SI_RETURN_TO_EPILOG $vgpr6, $vgpr7
|
|
|
|
...
|
|
|
|
---
|
|
name: test_war_vmvsrc_lds_lds_io
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0, $vgpr1
|
|
|
|
; GCN-LABEL: name: test_war_vmvsrc_lds_lds_io
|
|
; GCN: liveins: $vgpr0, $vgpr1
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr8_vgpr9_vgpr10_vgpr11 = DS_READ_B128_gfx9 $vgpr0, 0, 0, implicit $exec :: (load (s128), addrspace 3)
|
|
; GCN-NEXT: $vgpr4_vgpr5_vgpr6_vgpr7 = DS_READ_B128_gfx9 $vgpr1, 0, 0, implicit $exec :: (load (s128), addrspace 3)
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_1
|
|
; GCN-NEXT: $vgpr0 = nofpexcept V_ADD_F32_e32 $vgpr0, $vgpr0, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr1 = nofpexcept V_ADD_F32_e32 $vgpr1, $vgpr1, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAIT_DSCNT 0
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 0, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: SI_RETURN_TO_EPILOG $vgpr0, $vgpr1
|
|
$vgpr8_vgpr9_vgpr10_vgpr11 = DS_READ_B128_gfx9 $vgpr0, 0, 0, implicit $exec :: (load (s128), addrspace 3)
|
|
$vgpr4_vgpr5_vgpr6_vgpr7 = DS_READ_B128_gfx9 $vgpr1, 0, 0, implicit $exec :: (load (s128), addrspace 3)
|
|
$vgpr0 = nofpexcept V_ADD_F32_e32 $vgpr0, $vgpr0, implicit $mode, implicit $exec
|
|
$vgpr1 = nofpexcept V_ADD_F32_e32 $vgpr1, $vgpr1, implicit $mode, implicit $exec
|
|
SI_RETURN_TO_EPILOG $vgpr0, $vgpr1
|
|
|
|
...
|
|
|
|
---
|
|
name: test_war_vmvsrc_vmem_lds_ooo
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1
|
|
|
|
; GCN-LABEL: name: test_war_vmvsrc_vmem_lds_ooo
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr8_vgpr9_vgpr10_vgpr11 = IMAGE_LOAD_V4_V1_gfx12 $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
; GCN-NEXT: $vgpr4_vgpr5_vgpr6_vgpr7 = DS_READ_B128_gfx9 $vgpr1, 0, 0, implicit $exec :: (load (s128), addrspace 3)
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr0 = nofpexcept V_ADD_F32_e32 $vgpr0, $vgpr0, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr1 = nofpexcept V_ADD_F32_e32 $vgpr1, $vgpr1, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 0, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: SI_RETURN_TO_EPILOG $vgpr0, $vgpr1
|
|
$vgpr8_vgpr9_vgpr10_vgpr11 = IMAGE_LOAD_V4_V1_gfx12 $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
$vgpr4_vgpr5_vgpr6_vgpr7 = DS_READ_B128_gfx9 $vgpr1, 0, 0, implicit $exec :: (load (s128), addrspace 3)
|
|
$vgpr0 = nofpexcept V_ADD_F32_e32 $vgpr0, $vgpr0, implicit $mode, implicit $exec
|
|
$vgpr1 = nofpexcept V_ADD_F32_e32 $vgpr1, $vgpr1, implicit $mode, implicit $exec
|
|
SI_RETURN_TO_EPILOG $vgpr0, $vgpr1
|
|
|
|
...
|
|
|
|
---
|
|
name: test_war_vmvsrc_flat_lds_ooo
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0, $vgpr1, $vgpr2
|
|
|
|
; GCN-LABEL: name: test_war_vmvsrc_flat_lds_ooo
|
|
; GCN: liveins: $vgpr0, $vgpr1, $vgpr2
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr8 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
|
|
; GCN-NEXT: $vgpr4_vgpr5_vgpr6_vgpr7 = DS_READ_B128_gfx9 $vgpr2, 0, 0, implicit $exec :: (load (s128), addrspace 3)
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr1 = nofpexcept V_ADD_F32_e32 $vgpr1, $vgpr1, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr2 = nofpexcept V_ADD_F32_e32 $vgpr2, $vgpr2, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 0, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: SI_RETURN_TO_EPILOG $vgpr1, $vgpr2
|
|
$vgpr8 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
|
|
$vgpr4_vgpr5_vgpr6_vgpr7 = DS_READ_B128_gfx9 $vgpr2, 0, 0, implicit $exec :: (load (s128), addrspace 3)
|
|
$vgpr1 = nofpexcept V_ADD_F32_e32 $vgpr1, $vgpr1, implicit $mode, implicit $exec
|
|
$vgpr2 = nofpexcept V_ADD_F32_e32 $vgpr2, $vgpr2, implicit $mode, implicit $exec
|
|
SI_RETURN_TO_EPILOG $vgpr1, $vgpr2
|
|
|
|
...
|
|
|
|
---
|
|
name: test_war_vmvsrc_flat_vmem_ooo
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2
|
|
|
|
; GCN-LABEL: name: test_war_vmvsrc_flat_vmem_ooo
|
|
; GCN: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr8 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
|
|
; GCN-NEXT: $vgpr4_vgpr5_vgpr6_vgpr7 = IMAGE_LOAD_V4_V1_gfx12 $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr1 = nofpexcept V_ADD_F32_e32 $vgpr1, $vgpr1, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $vgpr2 = nofpexcept V_ADD_F32_e32 $vgpr2, $vgpr2, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
|
|
; GCN-NEXT: S_SETREG_IMM32_B32 0, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: SI_RETURN_TO_EPILOG $vgpr1, $vgpr2
|
|
$vgpr8 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
|
|
$vgpr4_vgpr5_vgpr6_vgpr7 = IMAGE_LOAD_V4_V1_gfx12 $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 15, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), addrspace 8)
|
|
$vgpr1 = nofpexcept V_ADD_F32_e32 $vgpr1, $vgpr1, implicit $mode, implicit $exec
|
|
$vgpr2 = nofpexcept V_ADD_F32_e32 $vgpr2, $vgpr2, implicit $mode, implicit $exec
|
|
SI_RETURN_TO_EPILOG $vgpr1, $vgpr2
|
|
|
|
...
|
|
|
|
---
|
|
# Ensure there is no unnecessary wait on vm_vsrc when load has already
|
|
# completed.
|
|
name: test_war_completed_vmem
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: test_war_completed_vmem
|
|
; GCN: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
|
|
; GCN-NEXT: S_WAIT_LOADCNT_DSCNT .Loadcnt_0_Dscnt_0
|
|
; GCN-NEXT: S_WAIT_EXPCNT 0
|
|
; GCN-NEXT: S_WAIT_SAMPLECNT 0
|
|
; GCN-NEXT: S_WAIT_BVHCNT 0
|
|
; GCN-NEXT: S_WAIT_KMCNT 0
|
|
; GCN-NEXT: $vgpr2 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 100, 0, implicit $exec
|
|
; GCN-NEXT: S_WAIT_LOADCNT 0
|
|
; GCN-NEXT: $vgpr2 = V_MOV_B32_e32 $vgpr2, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VmVsrc_0
|
|
; GCN-NEXT: $vgpr0 = V_MOV_B32_e32 0, implicit $exec
|
|
$vgpr2 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 100, 0, implicit $exec
|
|
$vgpr2 = V_MOV_B32_e32 $vgpr2, implicit $exec
|
|
$vgpr0 = V_MOV_B32_e32 0, implicit $exec
|
|
...
|