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
1055 lines
41 KiB
YAML
1055 lines
41 KiB
YAML
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
|
|
# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs -run-pass post-RA-hazard-rec,amdgpu-wait-sgpr-hazards -amdgpu-sgpr-hazard-boundary-cull=0 -o - %s | FileCheck -check-prefixes=GCN,NOBC,NOMEMC %s
|
|
# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs -run-pass post-RA-hazard-rec,amdgpu-wait-sgpr-hazards -amdgpu-sgpr-hazard-boundary-cull=1 -o - %s | FileCheck -check-prefixes=GCN,BC,NOMEMC %s
|
|
# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs -run-pass post-RA-hazard-rec,amdgpu-wait-sgpr-hazards -amdgpu-sgpr-hazard-boundary-cull=0 -amdgpu-sgpr-hazard-mem-wait-cull=1 -amdgpu-sgpr-hazard-mem-wait-cull-threshold=1 -o - %s | FileCheck -check-prefixes=GCN,NOBC,MEMC %s
|
|
|
|
--- |
|
|
@mem = internal unnamed_addr addrspace(4) constant [4 x <4 x i32>] [<4 x i32> <i32 0, i32 0, i32 0, i32 0>, <4 x i32> <i32 0, i32 0, i32 0, i32 0>, <4 x i32> <i32 0, i32 0, i32 0, i32 0>, <4 x i32> <i32 0, i32 0, i32 0, i32 0>]
|
|
|
|
define amdgpu_gs void @hazard_getpc1() { ret void }
|
|
define amdgpu_gs void @hazard_getpc2() { ret void }
|
|
define amdgpu_gs void @hazard_getpc3() { ret void }
|
|
define amdgpu_gs void @hazard_getpc4() { ret void }
|
|
define amdgpu_gs void @hazard_vcc1() { ret void }
|
|
define amdgpu_gs void @hazard_vcc2() { ret void }
|
|
define amdgpu_gs void @hazard_vcc3() { ret void }
|
|
define amdgpu_gs void @hazard_merge_vcc() { ret void }
|
|
define amdgpu_gs void @hazard_addc1() { ret void }
|
|
define amdgpu_gs void @hazard_addc2() { ret void }
|
|
define amdgpu_gs void @hazard_addc3() { ret void }
|
|
define amdgpu_gs void @hazard_addc4() { ret void }
|
|
define amdgpu_gs void @hazard_addc5() { ret void }
|
|
define amdgpu_gs void @hazard_addc6() { ret void }
|
|
define amdgpu_gs void @hazard_vaddc1() { ret void }
|
|
define amdgpu_gs void @hazard_gap1() { ret void }
|
|
define amdgpu_gs void @hazard_gap2() { ret void }
|
|
define amdgpu_gs void @hazard_gap3() { ret void }
|
|
define amdgpu_gs void @hazard_gap4_no_hazard() { ret void }
|
|
define amdgpu_gs void @hazard_valu_write1_no_hazard() { ret void }
|
|
define amdgpu_gs void @hazard_post_order1() { ret void }
|
|
define amdgpu_gs void @hazard_post_order2() { ret void }
|
|
define amdgpu_gs void @hazard_post_order_cycle() { ret void }
|
|
define amdgpu_cs void @hazard_calls() { ret void }
|
|
define void @hazard_callee1() { ret void }
|
|
define void @hazard_callee2() { ret void }
|
|
define amdgpu_cs void @hazard_carry_vcc() { ret void }
|
|
define amdgpu_cs void @hazard_carry_vcc_no_hazard() { ret void }
|
|
define amdgpu_cs void @hazard_carry_sgpr() { ret void }
|
|
define amdgpu_cs void @hazard_carry_sgpr_no_hazard1() { ret void }
|
|
define amdgpu_cs void @hazard_carry_sgpr_no_hazard2() { ret void }
|
|
define amdgpu_cs void @hazard_carry_sgpr_no_hazard3() { ret void }
|
|
define amdgpu_cs void @hazard_cull_vmem() { ret void }
|
|
define amdgpu_cs void @hazard_cull_sample() { ret void }
|
|
define amdgpu_cs void @hazard_cull_bvh() { ret void }
|
|
define amdgpu_cs void @hazard_nocull_scratch() { ret void }
|
|
define amdgpu_cs void @hazard_cull_global() { ret void }
|
|
define amdgpu_cs void @hazard_nocull_flat() { ret void }
|
|
define amdgpu_cs void @hazard_existing_cull() { ret void }
|
|
...
|
|
|
|
---
|
|
name: hazard_getpc1
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_getpc1
|
|
; GCN: $vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr0, implicit $exec
|
|
; GCN-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr0, implicit $exec
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_getpc2
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_getpc2
|
|
; GCN: $vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr1, implicit $exec
|
|
; GCN-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr1, implicit $exec
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_getpc3
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_getpc3
|
|
; GCN: $vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr0, implicit $exec
|
|
; GCN-NEXT: BUNDLE implicit-def $sgpr0_sgpr1 {
|
|
; GCN-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr0 = S_ADD_U32 $sgpr0, target-flags(amdgpu-rel32-lo) @mem + 8, implicit-def $scc
|
|
; GCN-NEXT: $sgpr1 = S_ADDC_U32 $sgpr1, target-flags(amdgpu-rel32-hi) @mem + 16, implicit-def $scc, implicit $scc
|
|
; GCN-NEXT: }
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr0, implicit $exec
|
|
BUNDLE implicit-def $sgpr0_sgpr1 {
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr0 = S_ADD_U32 $sgpr0, target-flags(amdgpu-rel32-lo) @mem + 4, implicit-def $scc
|
|
$sgpr1 = S_ADDC_U32 $sgpr1, target-flags(amdgpu-rel32-hi) @mem + 12, implicit-def $scc, implicit $scc
|
|
}
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_getpc4
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_getpc4
|
|
; GCN: $vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr0, implicit $exec
|
|
; GCN-NEXT: BUNDLE implicit-def $sgpr0_sgpr1 {
|
|
; GCN-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr1 = S_SEXT_I32_I16 $sgpr1
|
|
; GCN-NEXT: $sgpr0 = S_ADD_U32 $sgpr0, target-flags(amdgpu-rel32-lo) @mem + 12, implicit-def $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr1 = S_ADDC_U32 $sgpr1, target-flags(amdgpu-rel32-hi) @mem + 24, implicit-def $scc, implicit $scc
|
|
; GCN-NEXT: }
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr0, implicit $exec
|
|
BUNDLE implicit-def $sgpr0_sgpr1 {
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr1 = S_SEXT_I32_I16 $sgpr1
|
|
$sgpr0 = S_ADD_U32 $sgpr0, target-flags(amdgpu-rel32-lo) @mem + 8, implicit-def $scc
|
|
$sgpr1 = S_ADDC_U32 $sgpr1, target-flags(amdgpu-rel32-hi) @mem + 16, implicit-def $scc, implicit $scc
|
|
}
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_vcc1
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_vcc1
|
|
; GCN: $vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr2, implicit $exec
|
|
; GCN-NEXT: $sgpr3 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr4 = S_ADD_U32 $sgpr3, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr2, implicit $exec
|
|
$sgpr3 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr4 = S_ADD_U32 $sgpr3, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_vcc2
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_vcc2
|
|
; GCN: $vgpr1 = V_CNDMASK_B32_e32 $vgpr1, $vgpr2, implicit $vcc_lo, implicit $exec
|
|
; GCN-NEXT: $vcc_lo = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr4 = S_ADD_U32 $vcc_lo, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1 = V_CNDMASK_B32_e32 $vgpr1, $vgpr2, implicit $vcc, implicit $exec
|
|
$vcc_lo = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr4 = S_ADD_U32 $vcc_lo, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_vcc3
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_vcc3
|
|
; GCN: $vgpr1 = V_CNDMASK_B32_e32 $vgpr1, $vgpr2, implicit $vcc_lo, implicit $exec
|
|
; GCN-NEXT: $vcc_lo = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $vgpr3 = V_CNDMASK_B32_e32 $vgpr4, $vgpr5, implicit $vcc_lo, implicit $exec
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1 = V_CNDMASK_B32_e32 $vgpr1, $vgpr2, implicit $vcc, implicit $exec
|
|
$vcc_lo = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$vgpr3 = V_CNDMASK_B32_e32 $vgpr4, $vgpr5, implicit $vcc, implicit $exec
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_merge_vcc
|
|
body: |
|
|
; GCN-LABEL: name: hazard_merge_vcc
|
|
; GCN: bb.0:
|
|
; GCN-NEXT: successors: %bb.1(0x40000000), %bb.2(0x40000000)
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: $vgpr1 = V_CNDMASK_B32_e32 $vgpr1, $vgpr2, implicit $vcc_lo, implicit $exec
|
|
; GCN-NEXT: S_CBRANCH_SCC0 %bb.1, implicit $scc
|
|
; GCN-NEXT: S_BRANCH %bb.2
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: bb.1:
|
|
; GCN-NEXT: successors: %bb.3(0x80000000)
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: $vcc_lo = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_BRANCH %bb.3
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: bb.2:
|
|
; GCN-NEXT: successors: %bb.3(0x80000000)
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: $vcc_lo = V_CMP_EQ_F32_e64 0, $vgpr0, 0, $vgpr1, 1, implicit $mode, implicit $exec
|
|
; GCN-NEXT: S_BRANCH %bb.3
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: bb.3:
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVcc_0_SaSdst_0
|
|
; GCN-NEXT: $vgpr3 = V_CNDMASK_B32_e32 $vgpr4, $vgpr5, implicit $vcc_lo, implicit $exec
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
bb.0:
|
|
$vgpr1 = V_CNDMASK_B32_e32 $vgpr1, $vgpr2, implicit $vcc, implicit $exec
|
|
S_CBRANCH_SCC0 %bb.1, implicit $scc
|
|
S_BRANCH %bb.2
|
|
bb.1:
|
|
$vcc_lo = S_CSELECT_B32 -1, 0, implicit $scc
|
|
S_BRANCH %bb.3
|
|
bb.2:
|
|
$vcc_lo = V_CMP_EQ_F32_e64 0, $vgpr0, 0, $vgpr1, 1, implicit $mode, implicit $exec
|
|
S_BRANCH %bb.3
|
|
bb.3:
|
|
$vgpr3 = V_CNDMASK_B32_e32 $vgpr4, $vgpr5, implicit $vcc, implicit $exec
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_addc1
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_addc1
|
|
; GCN: $vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $sgpr0 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr2 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$sgpr0 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr2 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_addc2
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_addc2
|
|
; GCN: $vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr1, 0, implicit $exec
|
|
; GCN-NEXT: $sgpr0 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr2 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr1, 0, implicit $exec
|
|
$sgpr0 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr2 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_addc3
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_addc3
|
|
; GCN: $vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_addc4
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_addc4
|
|
; GCN: $vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr3, 0, implicit $exec
|
|
; GCN-NEXT: $sgpr3 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr2 = S_ADD_U32 $sgpr3, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr3, 0, implicit $exec
|
|
$sgpr3 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr2 = S_ADD_U32 $sgpr3, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_addc5
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_addc5
|
|
; GCN: $vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $sgpr16 = S_MOV_B32 0
|
|
; GCN-NEXT: $sgpr32 = S_MOV_B32 0
|
|
; GCN-NEXT: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$sgpr16 = S_MOV_B32 0
|
|
$sgpr32 = S_MOV_B32 0
|
|
$sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_addc6
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_addc6
|
|
; GCN: $vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $sgpr16 = S_MOV_B32 0
|
|
; GCN-NEXT: $sgpr32 = S_MOV_B32 0
|
|
; GCN-NEXT: $sgpr48 = S_MOV_B32 0
|
|
; GCN-NEXT: $sgpr80 = S_MOV_B32 0
|
|
; GCN-NEXT: $sgpr96 = S_MOV_B32 0
|
|
; GCN-NEXT: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$sgpr16 = S_MOV_B32 0
|
|
$sgpr32 = S_MOV_B32 0
|
|
$sgpr48 = S_MOV_B32 0
|
|
$sgpr80 = S_MOV_B32 0
|
|
$sgpr96 = S_MOV_B32 0
|
|
$sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_vaddc1
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_vaddc1
|
|
; GCN: $vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $vgpr2, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr1, 0, implicit $exec
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$vgpr2, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr1, 0, implicit $exec
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_gap1
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_gap1
|
|
; GCN: $vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
S_NOP 0
|
|
S_NOP 0
|
|
S_NOP 0
|
|
S_NOP 0
|
|
S_NOP 0
|
|
S_NOP 0
|
|
S_NOP 0
|
|
S_NOP 0
|
|
$sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_gap2
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_gap2
|
|
; GCN: $vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
S_NOP 0
|
|
S_NOP 0
|
|
S_NOP 0
|
|
S_NOP 0
|
|
S_NOP 0
|
|
S_NOP 0
|
|
S_NOP 0
|
|
S_NOP 0
|
|
$sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_gap3
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_gap3
|
|
; GCN: $vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: $sgpr4 = S_ADD_U32 $sgpr3, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr6 = S_ADD_U32 $sgpr5, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr8 = S_ADD_U32 $sgpr7, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr10 = S_ADD_U32 $sgpr9, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr12 = S_ADD_U32 $sgpr11, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr14 = S_ADD_U32 $sgpr13, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr16 = S_ADD_U32 $sgpr15, 0, implicit-def $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr4 = S_ADD_U32 $sgpr3, 0, implicit-def $scc
|
|
$sgpr6 = S_ADD_U32 $sgpr5, 0, implicit-def $scc
|
|
$sgpr8 = S_ADD_U32 $sgpr7, 0, implicit-def $scc
|
|
$sgpr10 = S_ADD_U32 $sgpr9, 0, implicit-def $scc
|
|
$sgpr12 = S_ADD_U32 $sgpr11, 0, implicit-def $scc
|
|
$sgpr14 = S_ADD_U32 $sgpr13, 0, implicit-def $scc
|
|
$sgpr16 = S_ADD_U32 $sgpr15, 0, implicit-def $scc
|
|
$sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_gap4_no_hazard
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_gap4_no_hazard
|
|
; GCN: $vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: $sgpr4 = S_ADD_U32 $sgpr3, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr6 = S_ADD_U32 $sgpr5, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr8 = S_ADD_U32 $sgpr7, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr10 = S_ADD_U32 $sgpr9, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr12 = S_ADD_U32 $sgpr11, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr14 = S_ADD_U32 $sgpr13, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr16 = S_ADD_U32 $sgpr15, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr18 = S_ADD_U32 $sgpr17, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr20 = S_ADD_U32 $sgpr19, 0, implicit-def $scc
|
|
; GCN-NEXT: $sgpr22 = S_ADD_U32 $sgpr21, 0, implicit-def $scc
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $vcc_lo = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr4 = S_ADD_U32 $sgpr3, 0, implicit-def $scc
|
|
$sgpr6 = S_ADD_U32 $sgpr5, 0, implicit-def $scc
|
|
$sgpr8 = S_ADD_U32 $sgpr7, 0, implicit-def $scc
|
|
$sgpr10 = S_ADD_U32 $sgpr9, 0, implicit-def $scc
|
|
$sgpr12 = S_ADD_U32 $sgpr11, 0, implicit-def $scc
|
|
$sgpr14 = S_ADD_U32 $sgpr13, 0, implicit-def $scc
|
|
$sgpr16 = S_ADD_U32 $sgpr15, 0, implicit-def $scc
|
|
$sgpr18 = S_ADD_U32 $sgpr17, 0, implicit-def $scc
|
|
$sgpr20 = S_ADD_U32 $sgpr19, 0, implicit-def $scc
|
|
$sgpr22 = S_ADD_U32 $sgpr21, 0, implicit-def $scc
|
|
$sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_valu_write1_no_hazard
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_valu_write1_no_hazard
|
|
; GCN: $sgpr1 = V_CMP_EQ_F32_e64 0, $vgpr0, 0, $vgpr1, 1, implicit $mode, implicit $exec
|
|
; GCN-NEXT: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; GCN-NEXT: $sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$sgpr1 = V_CMP_EQ_F32_e64 0, $vgpr0, 0, $vgpr1, 1, implicit $mode, implicit $exec
|
|
$sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_post_order1
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_post_order1
|
|
; GCN: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; GCN-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; GCN-NEXT: $vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr0, implicit $exec
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
$vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr0, implicit $exec
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_post_order2
|
|
body: |
|
|
; GCN-LABEL: name: hazard_post_order2
|
|
; GCN: bb.0:
|
|
; GCN-NEXT: successors: %bb.1(0x80000000)
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; GCN-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; GCN-NEXT: S_BRANCH %bb.1
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: bb.1:
|
|
; GCN-NEXT: $vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr0, implicit $exec
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
bb.0:
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
S_BRANCH %bb.1
|
|
|
|
bb.1:
|
|
$vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr0, implicit $exec
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_post_order_cycle
|
|
body: |
|
|
; GCN-LABEL: name: hazard_post_order_cycle
|
|
; GCN: bb.0:
|
|
; GCN-NEXT: successors: %bb.1(0x80000000)
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: S_NOP 0
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: bb.1:
|
|
; GCN-NEXT: successors: %bb.2(0x80000000)
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: bb.2:
|
|
; GCN-NEXT: successors: %bb.1(0x40000000), %bb.3(0x40000000)
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: $vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr0, implicit $exec
|
|
; GCN-NEXT: S_CBRANCH_SCC0 %bb.1, implicit $scc
|
|
; GCN-NEXT: {{ $}}
|
|
; GCN-NEXT: bb.3:
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
bb.0:
|
|
S_NOP 0
|
|
|
|
bb.1:
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
|
|
bb.2:
|
|
$vgpr1 = V_CNDMASK_B32_e64 0, $vgpr1, 0, $vgpr2, $sgpr0, implicit $exec
|
|
S_CBRANCH_SCC0 %bb.1, implicit $scc
|
|
|
|
bb.3:
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_calls
|
|
frameInfo:
|
|
hasCalls: true
|
|
body: |
|
|
; NOBC-LABEL: name: hazard_calls
|
|
; NOBC: bb.0:
|
|
; NOBC-NEXT: successors: %bb.2(0x40000000), %bb.1(0x40000000)
|
|
; NOBC-NEXT: {{ $}}
|
|
; NOBC-NEXT: $vgpr0 = V_WRITELANE_B32 0, $sgpr4, $vgpr0
|
|
; NOBC-NEXT: $vgpr0 = V_WRITELANE_B32 1, $sgpr8, $vgpr0
|
|
; NOBC-NEXT: $vgpr0 = V_WRITELANE_B32 2, $sgpr16, $vgpr0
|
|
; NOBC-NEXT: $vgpr0 = V_WRITELANE_B32 3, $sgpr18, $vgpr0
|
|
; NOBC-NEXT: $vgpr0 = V_WRITELANE_B32 4, $sgpr20, $vgpr0
|
|
; NOBC-NEXT: $vgpr0 = V_WRITELANE_B32 5, $sgpr22, $vgpr0
|
|
; NOBC-NEXT: S_CBRANCH_SCC0 %bb.2, implicit $scc
|
|
; NOBC-NEXT: S_BRANCH %bb.1
|
|
; NOBC-NEXT: {{ $}}
|
|
; NOBC-NEXT: bb.1:
|
|
; NOBC-NEXT: successors: %bb.3(0x40000000), %bb.4(0x40000000)
|
|
; NOBC-NEXT: {{ $}}
|
|
; NOBC-NEXT: S_CBRANCH_SCC0 %bb.3, implicit $scc
|
|
; NOBC-NEXT: S_BRANCH %bb.4
|
|
; NOBC-NEXT: {{ $}}
|
|
; NOBC-NEXT: bb.2:
|
|
; NOBC-NEXT: $sgpr16 = S_MOV_B32 0
|
|
; NOBC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOBC-NEXT: S_SETPC_B64 $sgpr0_sgpr1
|
|
; NOBC-NEXT: {{ $}}
|
|
; NOBC-NEXT: bb.3:
|
|
; NOBC-NEXT: $sgpr18 = S_MOV_B32 0
|
|
; NOBC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOBC-NEXT: S_SETPC_B64_return $sgpr0_sgpr1
|
|
; NOBC-NEXT: {{ $}}
|
|
; NOBC-NEXT: bb.4:
|
|
; NOBC-NEXT: successors: %bb.5(0x80000000)
|
|
; NOBC-NEXT: {{ $}}
|
|
; NOBC-NEXT: $vcc_lo = S_MOV_B32 0
|
|
; NOBC-NEXT: $sgpr20 = S_MOV_B32 0
|
|
; NOBC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOBC-NEXT: $sgpr4_sgpr5 = S_SWAPPC_B64 $sgpr2_sgpr3
|
|
; NOBC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOBC-NEXT: $sgpr4 = S_ADD_U32 $sgpr4, 0, implicit-def $scc
|
|
; NOBC-NEXT: {{ $}}
|
|
; NOBC-NEXT: bb.5:
|
|
; NOBC-NEXT: successors: %bb.6(0x80000000)
|
|
; NOBC-NEXT: {{ $}}
|
|
; NOBC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOBC-NEXT: $sgpr8_sgpr9 = S_CALL_B64 0
|
|
; NOBC-NEXT: {{ $}}
|
|
; NOBC-NEXT: bb.6:
|
|
; NOBC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOBC-NEXT: $sgpr22 = S_MOV_B32 $sgpr8
|
|
; NOBC-NEXT: S_ENDPGM 0
|
|
;
|
|
; BC-LABEL: name: hazard_calls
|
|
; BC: bb.0:
|
|
; BC-NEXT: successors: %bb.2(0x40000000), %bb.1(0x40000000)
|
|
; BC-NEXT: {{ $}}
|
|
; BC-NEXT: $vgpr0 = V_WRITELANE_B32 0, $sgpr4, $vgpr0
|
|
; BC-NEXT: $vgpr0 = V_WRITELANE_B32 1, $sgpr8, $vgpr0
|
|
; BC-NEXT: $vgpr0 = V_WRITELANE_B32 2, $sgpr16, $vgpr0
|
|
; BC-NEXT: $vgpr0 = V_WRITELANE_B32 3, $sgpr18, $vgpr0
|
|
; BC-NEXT: $vgpr0 = V_WRITELANE_B32 4, $sgpr20, $vgpr0
|
|
; BC-NEXT: $vgpr0 = V_WRITELANE_B32 5, $sgpr22, $vgpr0
|
|
; BC-NEXT: S_CBRANCH_SCC0 %bb.2, implicit $scc
|
|
; BC-NEXT: S_BRANCH %bb.1
|
|
; BC-NEXT: {{ $}}
|
|
; BC-NEXT: bb.1:
|
|
; BC-NEXT: successors: %bb.3(0x40000000), %bb.4(0x40000000)
|
|
; BC-NEXT: {{ $}}
|
|
; BC-NEXT: S_CBRANCH_SCC0 %bb.3, implicit $scc
|
|
; BC-NEXT: S_BRANCH %bb.4
|
|
; BC-NEXT: {{ $}}
|
|
; BC-NEXT: bb.2:
|
|
; BC-NEXT: $sgpr16 = S_MOV_B32 0
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; BC-NEXT: S_SETPC_B64 $sgpr0_sgpr1
|
|
; BC-NEXT: {{ $}}
|
|
; BC-NEXT: bb.3:
|
|
; BC-NEXT: $sgpr18 = S_MOV_B32 0
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; BC-NEXT: S_SETPC_B64_return $sgpr0_sgpr1
|
|
; BC-NEXT: {{ $}}
|
|
; BC-NEXT: bb.4:
|
|
; BC-NEXT: successors: %bb.5(0x80000000)
|
|
; BC-NEXT: {{ $}}
|
|
; BC-NEXT: $vcc_lo = S_MOV_B32 0
|
|
; BC-NEXT: $sgpr20 = S_MOV_B32 0
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; BC-NEXT: $sgpr4_sgpr5 = S_SWAPPC_B64 $sgpr2_sgpr3
|
|
; BC-NEXT: $sgpr4 = S_ADD_U32 $sgpr4, 0, implicit-def $scc
|
|
; BC-NEXT: {{ $}}
|
|
; BC-NEXT: bb.5:
|
|
; BC-NEXT: successors: %bb.6(0x80000000)
|
|
; BC-NEXT: {{ $}}
|
|
; BC-NEXT: $sgpr8_sgpr9 = S_CALL_B64 0
|
|
; BC-NEXT: {{ $}}
|
|
; BC-NEXT: bb.6:
|
|
; BC-NEXT: $sgpr22 = S_MOV_B32 $sgpr8
|
|
; BC-NEXT: S_ENDPGM 0
|
|
bb.0:
|
|
$vgpr0 = V_WRITELANE_B32 0, $sgpr4, $vgpr0
|
|
$vgpr0 = V_WRITELANE_B32 1, $sgpr8, $vgpr0
|
|
$vgpr0 = V_WRITELANE_B32 2, $sgpr16, $vgpr0
|
|
$vgpr0 = V_WRITELANE_B32 3, $sgpr18, $vgpr0
|
|
$vgpr0 = V_WRITELANE_B32 4, $sgpr20, $vgpr0
|
|
$vgpr0 = V_WRITELANE_B32 5, $sgpr22, $vgpr0
|
|
S_CBRANCH_SCC0 %bb.2, implicit $scc
|
|
S_BRANCH %bb.1
|
|
|
|
bb.1:
|
|
S_CBRANCH_SCC0 %bb.3, implicit $scc
|
|
S_BRANCH %bb.4
|
|
|
|
bb.2:
|
|
$sgpr16 = S_MOV_B32 0
|
|
S_SETPC_B64 $sgpr0_sgpr1
|
|
|
|
bb.3:
|
|
$sgpr18 = S_MOV_B32 0
|
|
S_SETPC_B64_return $sgpr0_sgpr1
|
|
|
|
bb.4:
|
|
$vcc_lo = S_MOV_B32 0
|
|
$sgpr20 = S_MOV_B32 0
|
|
$sgpr4_sgpr5 = S_SWAPPC_B64 $sgpr2_sgpr3
|
|
$sgpr4 = S_ADD_U32 $sgpr4, 0, implicit-def $scc
|
|
|
|
bb.5:
|
|
$sgpr8_sgpr9 = S_CALL_B64 0
|
|
|
|
bb.6:
|
|
$sgpr22 = S_MOV_B32 $sgpr8
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_callee1
|
|
body: |
|
|
bb.0:
|
|
; NOBC-LABEL: name: hazard_callee1
|
|
; NOBC: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; NOBC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOBC-NEXT: $sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
; NOBC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOBC-NEXT: S_SETPC_B64_return $sgpr30_sgpr31
|
|
;
|
|
; BC-LABEL: name: hazard_callee1
|
|
; BC: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; BC-NEXT: $sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
; BC-NEXT: S_SETPC_B64_return $sgpr30_sgpr31
|
|
$sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
S_SETPC_B64_return $sgpr30_sgpr31
|
|
...
|
|
|
|
---
|
|
name: hazard_callee2
|
|
body: |
|
|
bb.0:
|
|
; NOBC-LABEL: name: hazard_callee2
|
|
; NOBC: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; NOBC-NEXT: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; NOBC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOBC-NEXT: $sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
; NOBC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOBC-NEXT: S_SETPC_B64_return $sgpr30_sgpr31
|
|
;
|
|
; BC-LABEL: name: hazard_callee2
|
|
; BC: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; BC-NEXT: $sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
; BC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; BC-NEXT: $sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; BC-NEXT: S_SETPC_B64_return $sgpr30_sgpr31
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$sgpr1 = S_CSELECT_B32 -1, 0, implicit $scc
|
|
$sgpr2 = S_ADD_U32 $sgpr1, 0, implicit-def $scc
|
|
S_SETPC_B64_return $sgpr30_sgpr31
|
|
...
|
|
|
|
---
|
|
name: hazard_carry_vcc
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_carry_vcc
|
|
; GCN: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $vcc_lo, 0, implicit $exec
|
|
; GCN-NEXT: $vgpr0 = V_ADD_CO_U32_e32 $vgpr0, $vgpr1, implicit-def $vcc_lo, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaVcc_0
|
|
; GCN-NEXT: $vgpr1 = V_ADDC_U32_e32 $vgpr2, $vgpr3, implicit-def $vcc_lo, implicit $vcc_lo, implicit $exec
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $vcc_lo, 0, implicit $exec
|
|
$vgpr0 = V_ADD_CO_U32_e32 $vgpr0, $vgpr1, implicit-def $vcc, implicit $exec
|
|
$vgpr1 = V_ADDC_U32_e32 $vgpr2, $vgpr3, implicit-def $vcc, implicit $vcc, implicit $exec
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_carry_vcc_no_hazard
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_carry_vcc_no_hazard
|
|
; GCN: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $vcc_lo, 0, implicit $exec
|
|
; GCN-NEXT: $vgpr0 = V_ADD_CO_U32_e32 $vgpr0, $vgpr1, implicit-def $vcc_lo, implicit $exec
|
|
; GCN-NEXT: $sgpr8 = S_MOV_B32 $vcc_lo
|
|
; GCN-NEXT: $vgpr1 = V_ADDC_U32_e32 $vgpr2, $vgpr3, implicit-def $vcc_lo, implicit $vcc_lo, implicit $exec
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $vcc_lo, 0, implicit $exec
|
|
$vgpr0 = V_ADD_CO_U32_e32 $vgpr0, $vgpr1, implicit-def $vcc, implicit $exec
|
|
$sgpr8 = S_MOV_B32 $vcc_lo
|
|
$vgpr1 = V_ADDC_U32_e32 $vgpr2, $vgpr3, implicit-def $vcc, implicit $vcc, implicit $exec
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_carry_sgpr
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_carry_sgpr
|
|
; GCN: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $vgpr0, $sgpr0 = V_ADD_CO_U32_e64 $vgpr0, $vgpr1, 0, implicit $exec
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .VaSdst_0
|
|
; GCN-NEXT: $vgpr1, $sgpr1 = V_ADDC_U32_e64 $vgpr2, $vgpr3, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$vgpr0, $sgpr0 = V_ADD_CO_U32_e64 $vgpr0, $vgpr1, 0, implicit $exec
|
|
$vgpr1, $sgpr1 = V_ADDC_U32_e64 $vgpr2, $vgpr3, $sgpr0, 0, implicit $exec
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_carry_sgpr_no_hazard1
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_carry_sgpr_no_hazard1
|
|
; GCN: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $vgpr0, $sgpr0 = V_ADD_CO_U32_e64 $vgpr0, $vgpr1, 0, implicit $exec
|
|
; GCN-NEXT: $sgpr8 = S_MOV_B32 $sgpr0
|
|
; GCN-NEXT: $vgpr1, $sgpr1 = V_ADDC_U32_e64 $vgpr2, $vgpr3, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$vgpr0, $sgpr0 = V_ADD_CO_U32_e64 $vgpr0, $vgpr1, 0, implicit $exec
|
|
$sgpr8 = S_MOV_B32 $sgpr0
|
|
$vgpr1, $sgpr1 = V_ADDC_U32_e64 $vgpr2, $vgpr3, $sgpr0, 0, implicit $exec
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_carry_sgpr_no_hazard2
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_carry_sgpr_no_hazard2
|
|
; GCN: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $vgpr0, $sgpr0 = V_ADD_CO_U32_e64 $vgpr0, $vgpr1, 0, implicit $exec
|
|
; GCN-NEXT: $vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8, 0, 0, 0, implicit $exec
|
|
; GCN-NEXT: $vgpr1, $sgpr1 = V_ADDC_U32_e64 $vgpr2, $vgpr3, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$vgpr0, $sgpr0 = V_ADD_CO_U32_e64 $vgpr0, $vgpr1, 0, implicit $exec
|
|
$vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8, 0, 0, 0, implicit $exec
|
|
$vgpr1, $sgpr1 = V_ADDC_U32_e64 $vgpr2, $vgpr3, $sgpr0, 0, implicit $exec
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_carry_sgpr_no_hazard3
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_carry_sgpr_no_hazard3
|
|
; GCN: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $vgpr0, $sgpr0 = V_ADD_CO_U32_e64 $vgpr0, $vgpr1, 0, implicit $exec
|
|
; GCN-NEXT: $sgpr8 = S_LOAD_DWORD_IMM $sgpr6_sgpr7, 0, 0
|
|
; GCN-NEXT: $vgpr1, $sgpr1 = V_ADDC_U32_e64 $vgpr2, $vgpr3, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$vgpr0, $sgpr0 = V_ADD_CO_U32_e64 $vgpr0, $vgpr1, 0, implicit $exec
|
|
$sgpr8 = S_LOAD_DWORD_IMM $sgpr6_sgpr7, 0, 0
|
|
$vgpr1, $sgpr1 = V_ADDC_U32_e64 $vgpr2, $vgpr3, $sgpr0, 0, implicit $exec
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_cull_vmem
|
|
body: |
|
|
bb.0:
|
|
; NOMEMC-LABEL: name: hazard_cull_vmem
|
|
; NOMEMC: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; NOMEMC-NEXT: $vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8, 0, 0, 0, implicit $exec
|
|
; NOMEMC-NEXT: S_WAIT_LOADCNT 0
|
|
; NOMEMC-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; NOMEMC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOMEMC-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; NOMEMC-NEXT: S_ENDPGM 0
|
|
;
|
|
; MEMC-LABEL: name: hazard_cull_vmem
|
|
; MEMC: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; MEMC-NEXT: $vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8, 0, 0, 0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: S_WAIT_LOADCNT 0
|
|
; MEMC-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; MEMC-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; MEMC-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8, 0, 0, 0, implicit $exec
|
|
S_WAIT_LOADCNT 0
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_cull_sample
|
|
body: |
|
|
bb.0:
|
|
; NOMEMC-LABEL: name: hazard_cull_sample
|
|
; NOMEMC: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; NOMEMC-NEXT: $vgpr10 = IMAGE_SAMPLE_LZ_V1_V2_gfx12 $vgpr3, $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, 1, 1, -1, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))
|
|
; NOMEMC-NEXT: S_WAIT_SAMPLECNT 0
|
|
; NOMEMC-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; NOMEMC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOMEMC-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; NOMEMC-NEXT: S_ENDPGM 0
|
|
;
|
|
; MEMC-LABEL: name: hazard_cull_sample
|
|
; MEMC: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; MEMC-NEXT: $vgpr10 = IMAGE_SAMPLE_LZ_V1_V2_gfx12 $vgpr3, $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, 1, 1, -1, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: S_WAIT_SAMPLECNT 0
|
|
; MEMC-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; MEMC-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; MEMC-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$vgpr10 = IMAGE_SAMPLE_LZ_V1_V2_gfx12 $vgpr3, $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, 1, 1, -1, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))
|
|
S_WAIT_SAMPLECNT 0
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_cull_bvh
|
|
body: |
|
|
bb.0:
|
|
; NOMEMC-LABEL: name: hazard_cull_bvh
|
|
; NOMEMC: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; NOMEMC-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = IMAGE_BVH_INTERSECT_RAY_sa_gfx11 $vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14, renamable $sgpr0_sgpr1_sgpr2_sgpr3, 0, implicit $exec :: (dereferenceable load (s128), addrspace 7)
|
|
; NOMEMC-NEXT: S_WAIT_BVHCNT 0
|
|
; NOMEMC-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; NOMEMC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOMEMC-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; NOMEMC-NEXT: S_ENDPGM 0
|
|
;
|
|
; MEMC-LABEL: name: hazard_cull_bvh
|
|
; MEMC: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; MEMC-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = IMAGE_BVH_INTERSECT_RAY_sa_gfx11 $vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14, renamable $sgpr0_sgpr1_sgpr2_sgpr3, 0, implicit $exec :: (dereferenceable load (s128), addrspace 7)
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: S_WAIT_BVHCNT 0
|
|
; MEMC-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; MEMC-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; MEMC-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$vgpr0_vgpr1_vgpr2_vgpr3 = IMAGE_BVH_INTERSECT_RAY_sa_gfx11 $vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14, renamable $sgpr0_sgpr1_sgpr2_sgpr3, 0, implicit $exec :: (dereferenceable load (s128), addrspace 7)
|
|
S_WAIT_BVHCNT 0
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_nocull_scratch
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_nocull_scratch
|
|
; GCN: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD $vgpr0, 0, 0, implicit $exec, implicit $flat_scr
|
|
; GCN-NEXT: S_WAIT_LOADCNT 0
|
|
; GCN-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$vgpr0 = SCRATCH_LOAD_DWORD $vgpr0, 0, 0, implicit $exec, implicit $flat_scr
|
|
S_WAIT_LOADCNT 0
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_cull_global
|
|
body: |
|
|
bb.0:
|
|
; NOMEMC-LABEL: name: hazard_cull_global
|
|
; NOMEMC: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; NOMEMC-NEXT: $vgpr0 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec
|
|
; NOMEMC-NEXT: S_WAIT_LOADCNT 0
|
|
; NOMEMC-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; NOMEMC-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; NOMEMC-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; NOMEMC-NEXT: S_ENDPGM 0
|
|
;
|
|
; MEMC-LABEL: name: hazard_cull_global
|
|
; MEMC: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; MEMC-NEXT: $vgpr0 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; MEMC-NEXT: S_WAIT_LOADCNT 0
|
|
; MEMC-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; MEMC-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; MEMC-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$vgpr0 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec
|
|
S_WAIT_LOADCNT 0
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_nocull_flat
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_nocull_flat
|
|
; GCN: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
|
|
; GCN-NEXT: S_WAIT_LOADCNT 0
|
|
; GCN-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; GCN-NEXT: S_WAITCNT_DEPCTR .SaSdst_0
|
|
; GCN-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
$vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
|
|
S_WAIT_LOADCNT 0
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|
|
|
|
---
|
|
name: hazard_existing_cull
|
|
body: |
|
|
bb.0:
|
|
; GCN-LABEL: name: hazard_existing_cull
|
|
; GCN: $vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
; GCN-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; GCN-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; GCN-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; GCN-NEXT: DS_NOP implicit $m0, implicit $exec
|
|
; GCN-NEXT: $sgpr0_sgpr1 = S_GETPC_B64
|
|
; GCN-NEXT: $sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
; GCN-NEXT: S_ENDPGM 0
|
|
$vgpr1, $sgpr0 = V_ADDC_U32_e64 0, $vgpr1, $sgpr0, 0, implicit $exec
|
|
DS_NOP implicit $m0, implicit $exec
|
|
DS_NOP implicit $m0, implicit $exec
|
|
DS_NOP implicit $m0, implicit $exec
|
|
DS_NOP implicit $m0, implicit $exec
|
|
$sgpr0_sgpr1 = S_GETPC_B64
|
|
$sgpr3 = S_ADD_U32 $sgpr0, 0, implicit-def $scc
|
|
S_ENDPGM 0
|
|
...
|