derek/TheRock
1
0 Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity
Files
1a182e2dac58879f215c0fad0f5f78c76dcafe56
TheRock/compiler/amd-llvm/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw-fmin-fmax.ll
T
derek.holloway 7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
Details
CodeQL / Analyze (python) (push) Canceled after 0s
Details
pre-commit / pre-commit (push) Canceled after 0s
Details
Multi-Arch CI / setup (push) Canceled after 0s
Details
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Details
Setup (Multi-Arch) / setup (push) Canceled after 0s
Details
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Details
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
Details
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Details
Multi-Arch CI / CI Summary (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Details
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Details
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
Details
Flatten subprojects into a single base monorepo
2026-09-02 16:34:47 -07:00

597 lines
38 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -stop-after=instruction-select < %s | FileCheck %s --check-prefixes=GFX10
; --- atomicrmw fmax: global f32/f64, flat f32/f64, local f32/f64 ---
define void @atomicrmw_fmax_global_f32_vv_noret(ptr addrspace(1) %ptr, float %val) {
; GFX10-LABEL: name: atomicrmw_fmax_global_f32_vv_noret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: GLOBAL_ATOMIC_FMAX [[REG_SEQUENCE]], [[COPY2]], 0, 0, implicit $exec :: (load store seq_cst (s32) on %ir.ptr, addrspace 1)
; GFX10-NEXT: SI_RETURN
%old = atomicrmw fmax ptr addrspace(1) %ptr, float %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret void
}
define float @atomicrmw_fmax_global_f32_vv_ret(ptr addrspace(1) %ptr, float %val) {
; GFX10-LABEL: name: atomicrmw_fmax_global_f32_vv_ret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[GLOBAL_ATOMIC_FMAX_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_FMAX_RTN [[REG_SEQUENCE]], [[COPY2]], 0, 1, implicit $exec :: (load store seq_cst (s32) on %ir.ptr, addrspace 1)
; GFX10-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_FMAX_RTN]]
; GFX10-NEXT: SI_RETURN implicit $vgpr0
%old = atomicrmw fmax ptr addrspace(1) %ptr, float %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret float %old
}
define void @atomicrmw_fmax_global_f64_vv_noret(ptr addrspace(1) %ptr, double %val) {
; GFX10-LABEL: name: atomicrmw_fmax_global_f64_vv_noret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
; GFX10-NEXT: GLOBAL_ATOMIC_MAX_F64 [[REG_SEQUENCE]], [[REG_SEQUENCE1]], 0, 0, implicit $exec :: (load store seq_cst (s64) on %ir.ptr, addrspace 1)
; GFX10-NEXT: SI_RETURN
%old = atomicrmw fmax ptr addrspace(1) %ptr, double %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret void
}
define double @atomicrmw_fmax_global_f64_vv_ret(ptr addrspace(1) %ptr, double %val) {
; GFX10-LABEL: name: atomicrmw_fmax_global_f64_vv_ret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
; GFX10-NEXT: [[GLOBAL_ATOMIC_MAX_F64_RTN:%[0-9]+]]:vreg_64 = GLOBAL_ATOMIC_MAX_F64_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE1]], 0, 1, implicit $exec :: (load store seq_cst (s64) on %ir.ptr, addrspace 1)
; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_ATOMIC_MAX_F64_RTN]].sub0
; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_ATOMIC_MAX_F64_RTN]].sub1
; GFX10-NEXT: $vgpr0 = COPY [[COPY4]]
; GFX10-NEXT: $vgpr1 = COPY [[COPY5]]
; GFX10-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1
%old = atomicrmw fmax ptr addrspace(1) %ptr, double %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret double %old
}
define void @atomicrmw_fmax_flat_f32_vv_noret(ptr %ptr, float %val) {
; GFX10-LABEL: name: atomicrmw_fmax_flat_f32_vv_noret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: FLAT_ATOMIC_FMAX [[REG_SEQUENCE]], [[COPY2]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32) on %ir.ptr)
; GFX10-NEXT: SI_RETURN
%old = atomicrmw fmax ptr %ptr, float %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret void
}
define float @atomicrmw_fmax_flat_f32_vv_ret(ptr %ptr, float %val) {
; GFX10-LABEL: name: atomicrmw_fmax_flat_f32_vv_ret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[FLAT_ATOMIC_FMAX_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_FMAX_RTN [[REG_SEQUENCE]], [[COPY2]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32) on %ir.ptr)
; GFX10-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_FMAX_RTN]]
; GFX10-NEXT: SI_RETURN implicit $vgpr0
%old = atomicrmw fmax ptr %ptr, float %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret float %old
}
define void @atomicrmw_fmax_flat_f64_vv_noret(ptr %ptr, double %val) {
; GFX10-LABEL: name: atomicrmw_fmax_flat_f64_vv_noret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: successors: %bb.4(0x40000000), %bb.2(0x40000000)
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
; GFX10-NEXT: [[COPY4:%[0-9]+]]:sreg_64 = COPY $src_private_base
; GFX10-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[COPY4]].sub0
; GFX10-NEXT: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[COPY4]].sub1
; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY6]]
; GFX10-NEXT: [[V_CMP_NE_U32_e64_:%[0-9]+]]:sreg_32 = V_CMP_NE_U32_e64 [[COPY1]], [[COPY7]], implicit $exec
; GFX10-NEXT: [[COPY8:%[0-9]+]]:sreg_32_xm0_xexec = COPY [[V_CMP_NE_U32_e64_]]
; GFX10-NEXT: [[SI_IF:%[0-9]+]]:sreg_32_xm0_xexec = SI_IF [[COPY8]], %bb.2, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX10-NEXT: S_BRANCH %bb.4
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.2.Flow:
; GFX10-NEXT: successors: %bb.3(0x40000000), %bb.5(0x40000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[SI_ELSE:%[0-9]+]]:sreg_32_xm0_xexec = SI_ELSE [[SI_IF]], %bb.5, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX10-NEXT: S_BRANCH %bb.3
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.3.atomicrmw.private:
; GFX10-NEXT: successors: %bb.5(0x80000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
; GFX10-NEXT: [[S_MOV_B:%[0-9]+]]:sreg_64 = S_MOV_B64_IMM_PSEUDO 0
; GFX10-NEXT: [[COPY9:%[0-9]+]]:vreg_64 = COPY [[S_MOV_B]]
; GFX10-NEXT: [[V_CMP_NE_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_NE_U64_e64 [[REG_SEQUENCE]], [[COPY9]], implicit $exec
; GFX10-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
; GFX10-NEXT: [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, [[COPY10]], 0, [[COPY]], [[V_CMP_NE_U64_e64_]], implicit $exec
; GFX10-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (load (s32) from %ir.7, align 8, addrspace 5)
; GFX10-NEXT: [[BUFFER_LOAD_DWORD_OFFEN1:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (load (s32) from %ir.7 + 4, basealign 8, addrspace 5)
; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[BUFFER_LOAD_DWORD_OFFEN]], %subreg.sub0, [[BUFFER_LOAD_DWORD_OFFEN1]], %subreg.sub1
; GFX10-NEXT: [[V_MAX_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[REG_SEQUENCE2]], 0, [[REG_SEQUENCE2]], 0, 0, implicit $mode, implicit $exec
; GFX10-NEXT: [[V_MAX_F64_e64_1:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[REG_SEQUENCE1]], 0, [[REG_SEQUENCE1]], 0, 0, implicit $mode, implicit $exec
; GFX10-NEXT: [[V_MAX_F64_e64_2:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[V_MAX_F64_e64_]], 0, [[V_MAX_F64_e64_1]], 0, 0, implicit $mode, implicit $exec
; GFX10-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[V_MAX_F64_e64_2]].sub0
; GFX10-NEXT: [[COPY12:%[0-9]+]]:vgpr_32 = COPY [[V_MAX_F64_e64_2]].sub1
; GFX10-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY11]], [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into %ir.7, align 8, addrspace 5)
; GFX10-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY12]], [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into %ir.7 + 4, basealign 8, addrspace 5)
; GFX10-NEXT: S_BRANCH %bb.5
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.4.atomicrmw.global:
; GFX10-NEXT: successors: %bb.2(0x80000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: FLAT_ATOMIC_MAX_F64 [[REG_SEQUENCE]], [[REG_SEQUENCE1]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64) on %ir.ptr, !noalias.addrspace !1)
; GFX10-NEXT: S_BRANCH %bb.2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.5.atomicrmw.phi:
; GFX10-NEXT: SI_END_CF [[SI_ELSE]], implicit-def $exec, implicit-def $scc, implicit $exec
; GFX10-NEXT: SI_RETURN
%old = atomicrmw fmax ptr %ptr, double %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret void
}
define double @atomicrmw_fmax_flat_f64_vv_ret(ptr %ptr, double %val) {
; GFX10-LABEL: name: atomicrmw_fmax_flat_f64_vv_ret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: successors: %bb.4(0x40000000), %bb.2(0x40000000)
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
; GFX10-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; GFX10-NEXT: [[COPY4:%[0-9]+]]:sreg_64 = COPY $src_private_base
; GFX10-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[COPY4]].sub0
; GFX10-NEXT: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[COPY4]].sub1
; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY6]]
; GFX10-NEXT: [[V_CMP_NE_U32_e64_:%[0-9]+]]:sreg_32 = V_CMP_NE_U32_e64 [[COPY1]], [[COPY7]], implicit $exec
; GFX10-NEXT: [[COPY8:%[0-9]+]]:sreg_32_xm0_xexec = COPY [[V_CMP_NE_U32_e64_]]
; GFX10-NEXT: [[SI_IF:%[0-9]+]]:sreg_32_xm0_xexec = SI_IF [[COPY8]], %bb.2, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX10-NEXT: S_BRANCH %bb.4
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.2.Flow:
; GFX10-NEXT: successors: %bb.3(0x40000000), %bb.5(0x40000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[PHI:%[0-9]+]]:vreg_64 = PHI %19, %bb.4, [[DEF]], %bb.1
; GFX10-NEXT: [[SI_ELSE:%[0-9]+]]:sreg_32_xm0_xexec = SI_ELSE [[SI_IF]], %bb.5, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX10-NEXT: S_BRANCH %bb.3
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.3.atomicrmw.private:
; GFX10-NEXT: successors: %bb.5(0x80000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
; GFX10-NEXT: [[S_MOV_B:%[0-9]+]]:sreg_64 = S_MOV_B64_IMM_PSEUDO 0
; GFX10-NEXT: [[COPY9:%[0-9]+]]:vreg_64 = COPY [[S_MOV_B]]
; GFX10-NEXT: [[V_CMP_NE_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_NE_U64_e64 [[REG_SEQUENCE]], [[COPY9]], implicit $exec
; GFX10-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
; GFX10-NEXT: [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, [[COPY10]], 0, [[COPY]], [[V_CMP_NE_U64_e64_]], implicit $exec
; GFX10-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (load (s32) from %ir.8, align 8, addrspace 5)
; GFX10-NEXT: [[BUFFER_LOAD_DWORD_OFFEN1:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (load (s32) from %ir.8 + 4, basealign 8, addrspace 5)
; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[BUFFER_LOAD_DWORD_OFFEN]], %subreg.sub0, [[BUFFER_LOAD_DWORD_OFFEN1]], %subreg.sub1
; GFX10-NEXT: [[V_MAX_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[REG_SEQUENCE2]], 0, [[REG_SEQUENCE2]], 0, 0, implicit $mode, implicit $exec
; GFX10-NEXT: [[V_MAX_F64_e64_1:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[REG_SEQUENCE1]], 0, [[REG_SEQUENCE1]], 0, 0, implicit $mode, implicit $exec
; GFX10-NEXT: [[V_MAX_F64_e64_2:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[V_MAX_F64_e64_]], 0, [[V_MAX_F64_e64_1]], 0, 0, implicit $mode, implicit $exec
; GFX10-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[V_MAX_F64_e64_2]].sub0
; GFX10-NEXT: [[COPY12:%[0-9]+]]:vgpr_32 = COPY [[V_MAX_F64_e64_2]].sub1
; GFX10-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY11]], [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into %ir.8, align 8, addrspace 5)
; GFX10-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY12]], [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into %ir.8 + 4, basealign 8, addrspace 5)
; GFX10-NEXT: S_BRANCH %bb.5
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.4.atomicrmw.global:
; GFX10-NEXT: successors: %bb.2(0x80000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[FLAT_ATOMIC_MAX_F64_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_MAX_F64_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE1]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64) on %ir.ptr, !noalias.addrspace !1)
; GFX10-NEXT: S_BRANCH %bb.2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.5.atomicrmw.phi:
; GFX10-NEXT: successors: %bb.6(0x80000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[PHI1:%[0-9]+]]:vreg_64 = PHI [[PHI]], %bb.2, [[REG_SEQUENCE2]], %bb.3
; GFX10-NEXT: SI_END_CF [[SI_ELSE]], implicit-def $exec, implicit-def $scc, implicit $exec
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.6.atomicrmw.end:
; GFX10-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY [[PHI1]].sub0
; GFX10-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[PHI1]].sub1
; GFX10-NEXT: $vgpr0 = COPY [[COPY13]]
; GFX10-NEXT: $vgpr1 = COPY [[COPY14]]
; GFX10-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1
%old = atomicrmw fmax ptr %ptr, double %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret double %old
}
define void @atomicrmw_fmax_local_f32_vv_noret(ptr addrspace(3) %ptr, float %val) {
; GFX10-LABEL: name: atomicrmw_fmax_local_f32_vv_noret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: DS_MAX_F32_gfx9 [[COPY]], [[COPY1]], 0, 0, implicit $exec :: (load store seq_cst (s32) on %ir.ptr, addrspace 3)
; GFX10-NEXT: SI_RETURN
%old = atomicrmw fmax ptr addrspace(3) %ptr, float %val seq_cst
ret void
}
define float @atomicrmw_fmax_local_f32_vv_ret(ptr addrspace(3) %ptr, float %val) {
; GFX10-LABEL: name: atomicrmw_fmax_local_f32_vv_ret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[DS_MAX_RTN_F32_gfx9_:%[0-9]+]]:vgpr_32 = DS_MAX_RTN_F32_gfx9 [[COPY]], [[COPY1]], 0, 0, implicit $exec :: (load store seq_cst (s32) on %ir.ptr, addrspace 3)
; GFX10-NEXT: $vgpr0 = COPY [[DS_MAX_RTN_F32_gfx9_]]
; GFX10-NEXT: SI_RETURN implicit $vgpr0
%old = atomicrmw fmax ptr addrspace(3) %ptr, float %val seq_cst
ret float %old
}
define void @atomicrmw_fmax_local_f64_vv_noret(ptr addrspace(3) %ptr, double %val) {
; GFX10-LABEL: name: atomicrmw_fmax_local_f64_vv_noret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX10-NEXT: DS_MAX_F64_gfx9 [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst (s64) on %ir.ptr, addrspace 3)
; GFX10-NEXT: SI_RETURN
%old = atomicrmw fmax ptr addrspace(3) %ptr, double %val seq_cst
ret void
}
define double @atomicrmw_fmax_local_f64_vv_ret(ptr addrspace(3) %ptr, double %val) {
; GFX10-LABEL: name: atomicrmw_fmax_local_f64_vv_ret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX10-NEXT: [[DS_MAX_RTN_F64_gfx9_:%[0-9]+]]:vreg_64 = DS_MAX_RTN_F64_gfx9 [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst (s64) on %ir.ptr, addrspace 3)
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[DS_MAX_RTN_F64_gfx9_]].sub0
; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[DS_MAX_RTN_F64_gfx9_]].sub1
; GFX10-NEXT: $vgpr0 = COPY [[COPY3]]
; GFX10-NEXT: $vgpr1 = COPY [[COPY4]]
; GFX10-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1
%old = atomicrmw fmax ptr addrspace(3) %ptr, double %val seq_cst
ret double %old
}
; --- atomicrmw fmin: global f32/f64, flat f32/f64, local f32/f64 ---
define void @atomicrmw_fmin_global_f32_vv_noret(ptr addrspace(1) %ptr, float %val) {
; GFX10-LABEL: name: atomicrmw_fmin_global_f32_vv_noret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: GLOBAL_ATOMIC_FMIN [[REG_SEQUENCE]], [[COPY2]], 0, 0, implicit $exec :: (load store seq_cst (s32) on %ir.ptr, addrspace 1)
; GFX10-NEXT: SI_RETURN
%old = atomicrmw fmin ptr addrspace(1) %ptr, float %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret void
}
define float @atomicrmw_fmin_global_f32_vv_ret(ptr addrspace(1) %ptr, float %val) {
; GFX10-LABEL: name: atomicrmw_fmin_global_f32_vv_ret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[GLOBAL_ATOMIC_FMIN_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_FMIN_RTN [[REG_SEQUENCE]], [[COPY2]], 0, 1, implicit $exec :: (load store seq_cst (s32) on %ir.ptr, addrspace 1)
; GFX10-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_FMIN_RTN]]
; GFX10-NEXT: SI_RETURN implicit $vgpr0
%old = atomicrmw fmin ptr addrspace(1) %ptr, float %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret float %old
}
define void @atomicrmw_fmin_global_f64_vv_noret(ptr addrspace(1) %ptr, double %val) {
; GFX10-LABEL: name: atomicrmw_fmin_global_f64_vv_noret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
; GFX10-NEXT: GLOBAL_ATOMIC_MIN_F64 [[REG_SEQUENCE]], [[REG_SEQUENCE1]], 0, 0, implicit $exec :: (load store seq_cst (s64) on %ir.ptr, addrspace 1)
; GFX10-NEXT: SI_RETURN
%old = atomicrmw fmin ptr addrspace(1) %ptr, double %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret void
}
define double @atomicrmw_fmin_global_f64_vv_ret(ptr addrspace(1) %ptr, double %val) {
; GFX10-LABEL: name: atomicrmw_fmin_global_f64_vv_ret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
; GFX10-NEXT: [[GLOBAL_ATOMIC_MIN_F64_RTN:%[0-9]+]]:vreg_64 = GLOBAL_ATOMIC_MIN_F64_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE1]], 0, 1, implicit $exec :: (load store seq_cst (s64) on %ir.ptr, addrspace 1)
; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_ATOMIC_MIN_F64_RTN]].sub0
; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_ATOMIC_MIN_F64_RTN]].sub1
; GFX10-NEXT: $vgpr0 = COPY [[COPY4]]
; GFX10-NEXT: $vgpr1 = COPY [[COPY5]]
; GFX10-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1
%old = atomicrmw fmin ptr addrspace(1) %ptr, double %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret double %old
}
define void @atomicrmw_fmin_flat_f32_vv_noret(ptr %ptr, float %val) {
; GFX10-LABEL: name: atomicrmw_fmin_flat_f32_vv_noret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: FLAT_ATOMIC_FMIN [[REG_SEQUENCE]], [[COPY2]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32) on %ir.ptr)
; GFX10-NEXT: SI_RETURN
%old = atomicrmw fmin ptr %ptr, float %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret void
}
define float @atomicrmw_fmin_flat_f32_vv_ret(ptr %ptr, float %val) {
; GFX10-LABEL: name: atomicrmw_fmin_flat_f32_vv_ret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[FLAT_ATOMIC_FMIN_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_FMIN_RTN [[REG_SEQUENCE]], [[COPY2]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32) on %ir.ptr)
; GFX10-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_FMIN_RTN]]
; GFX10-NEXT: SI_RETURN implicit $vgpr0
%old = atomicrmw fmin ptr %ptr, float %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret float %old
}
define void @atomicrmw_fmin_flat_f64_vv_noret(ptr %ptr, double %val) {
; GFX10-LABEL: name: atomicrmw_fmin_flat_f64_vv_noret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: successors: %bb.4(0x40000000), %bb.2(0x40000000)
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
; GFX10-NEXT: [[COPY4:%[0-9]+]]:sreg_64 = COPY $src_private_base
; GFX10-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[COPY4]].sub0
; GFX10-NEXT: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[COPY4]].sub1
; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY6]]
; GFX10-NEXT: [[V_CMP_NE_U32_e64_:%[0-9]+]]:sreg_32 = V_CMP_NE_U32_e64 [[COPY1]], [[COPY7]], implicit $exec
; GFX10-NEXT: [[COPY8:%[0-9]+]]:sreg_32_xm0_xexec = COPY [[V_CMP_NE_U32_e64_]]
; GFX10-NEXT: [[SI_IF:%[0-9]+]]:sreg_32_xm0_xexec = SI_IF [[COPY8]], %bb.2, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX10-NEXT: S_BRANCH %bb.4
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.2.Flow:
; GFX10-NEXT: successors: %bb.3(0x40000000), %bb.5(0x40000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[SI_ELSE:%[0-9]+]]:sreg_32_xm0_xexec = SI_ELSE [[SI_IF]], %bb.5, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX10-NEXT: S_BRANCH %bb.3
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.3.atomicrmw.private:
; GFX10-NEXT: successors: %bb.5(0x80000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
; GFX10-NEXT: [[S_MOV_B:%[0-9]+]]:sreg_64 = S_MOV_B64_IMM_PSEUDO 0
; GFX10-NEXT: [[COPY9:%[0-9]+]]:vreg_64 = COPY [[S_MOV_B]]
; GFX10-NEXT: [[V_CMP_NE_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_NE_U64_e64 [[REG_SEQUENCE]], [[COPY9]], implicit $exec
; GFX10-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
; GFX10-NEXT: [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, [[COPY10]], 0, [[COPY]], [[V_CMP_NE_U64_e64_]], implicit $exec
; GFX10-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (load (s32) from %ir.7, align 8, addrspace 5)
; GFX10-NEXT: [[BUFFER_LOAD_DWORD_OFFEN1:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (load (s32) from %ir.7 + 4, basealign 8, addrspace 5)
; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[BUFFER_LOAD_DWORD_OFFEN]], %subreg.sub0, [[BUFFER_LOAD_DWORD_OFFEN1]], %subreg.sub1
; GFX10-NEXT: [[V_MAX_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[REG_SEQUENCE2]], 0, [[REG_SEQUENCE2]], 0, 0, implicit $mode, implicit $exec
; GFX10-NEXT: [[V_MAX_F64_e64_1:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[REG_SEQUENCE1]], 0, [[REG_SEQUENCE1]], 0, 0, implicit $mode, implicit $exec
; GFX10-NEXT: [[V_MIN_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MIN_F64_e64 0, [[V_MAX_F64_e64_]], 0, [[V_MAX_F64_e64_1]], 0, 0, implicit $mode, implicit $exec
; GFX10-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[V_MIN_F64_e64_]].sub0
; GFX10-NEXT: [[COPY12:%[0-9]+]]:vgpr_32 = COPY [[V_MIN_F64_e64_]].sub1
; GFX10-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY11]], [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into %ir.7, align 8, addrspace 5)
; GFX10-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY12]], [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into %ir.7 + 4, basealign 8, addrspace 5)
; GFX10-NEXT: S_BRANCH %bb.5
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.4.atomicrmw.global:
; GFX10-NEXT: successors: %bb.2(0x80000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: FLAT_ATOMIC_MIN_F64 [[REG_SEQUENCE]], [[REG_SEQUENCE1]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64) on %ir.ptr, !noalias.addrspace !1)
; GFX10-NEXT: S_BRANCH %bb.2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.5.atomicrmw.phi:
; GFX10-NEXT: SI_END_CF [[SI_ELSE]], implicit-def $exec, implicit-def $scc, implicit $exec
; GFX10-NEXT: SI_RETURN
%old = atomicrmw fmin ptr %ptr, double %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret void
}
define double @atomicrmw_fmin_flat_f64_vv_ret(ptr %ptr, double %val) {
; GFX10-LABEL: name: atomicrmw_fmin_flat_f64_vv_ret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: successors: %bb.4(0x40000000), %bb.2(0x40000000)
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
; GFX10-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; GFX10-NEXT: [[COPY4:%[0-9]+]]:sreg_64 = COPY $src_private_base
; GFX10-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[COPY4]].sub0
; GFX10-NEXT: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[COPY4]].sub1
; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY6]]
; GFX10-NEXT: [[V_CMP_NE_U32_e64_:%[0-9]+]]:sreg_32 = V_CMP_NE_U32_e64 [[COPY1]], [[COPY7]], implicit $exec
; GFX10-NEXT: [[COPY8:%[0-9]+]]:sreg_32_xm0_xexec = COPY [[V_CMP_NE_U32_e64_]]
; GFX10-NEXT: [[SI_IF:%[0-9]+]]:sreg_32_xm0_xexec = SI_IF [[COPY8]], %bb.2, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX10-NEXT: S_BRANCH %bb.4
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.2.Flow:
; GFX10-NEXT: successors: %bb.3(0x40000000), %bb.5(0x40000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[PHI:%[0-9]+]]:vreg_64 = PHI %19, %bb.4, [[DEF]], %bb.1
; GFX10-NEXT: [[SI_ELSE:%[0-9]+]]:sreg_32_xm0_xexec = SI_ELSE [[SI_IF]], %bb.5, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX10-NEXT: S_BRANCH %bb.3
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.3.atomicrmw.private:
; GFX10-NEXT: successors: %bb.5(0x80000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
; GFX10-NEXT: [[S_MOV_B:%[0-9]+]]:sreg_64 = S_MOV_B64_IMM_PSEUDO 0
; GFX10-NEXT: [[COPY9:%[0-9]+]]:vreg_64 = COPY [[S_MOV_B]]
; GFX10-NEXT: [[V_CMP_NE_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_NE_U64_e64 [[REG_SEQUENCE]], [[COPY9]], implicit $exec
; GFX10-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
; GFX10-NEXT: [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, [[COPY10]], 0, [[COPY]], [[V_CMP_NE_U64_e64_]], implicit $exec
; GFX10-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (load (s32) from %ir.8, align 8, addrspace 5)
; GFX10-NEXT: [[BUFFER_LOAD_DWORD_OFFEN1:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (load (s32) from %ir.8 + 4, basealign 8, addrspace 5)
; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[BUFFER_LOAD_DWORD_OFFEN]], %subreg.sub0, [[BUFFER_LOAD_DWORD_OFFEN1]], %subreg.sub1
; GFX10-NEXT: [[V_MAX_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[REG_SEQUENCE2]], 0, [[REG_SEQUENCE2]], 0, 0, implicit $mode, implicit $exec
; GFX10-NEXT: [[V_MAX_F64_e64_1:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[REG_SEQUENCE1]], 0, [[REG_SEQUENCE1]], 0, 0, implicit $mode, implicit $exec
; GFX10-NEXT: [[V_MIN_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MIN_F64_e64 0, [[V_MAX_F64_e64_]], 0, [[V_MAX_F64_e64_1]], 0, 0, implicit $mode, implicit $exec
; GFX10-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[V_MIN_F64_e64_]].sub0
; GFX10-NEXT: [[COPY12:%[0-9]+]]:vgpr_32 = COPY [[V_MIN_F64_e64_]].sub1
; GFX10-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY11]], [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into %ir.8, align 8, addrspace 5)
; GFX10-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY12]], [[V_CNDMASK_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into %ir.8 + 4, basealign 8, addrspace 5)
; GFX10-NEXT: S_BRANCH %bb.5
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.4.atomicrmw.global:
; GFX10-NEXT: successors: %bb.2(0x80000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[FLAT_ATOMIC_MIN_F64_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_MIN_F64_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE1]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64) on %ir.ptr, !noalias.addrspace !1)
; GFX10-NEXT: S_BRANCH %bb.2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.5.atomicrmw.phi:
; GFX10-NEXT: successors: %bb.6(0x80000000)
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[PHI1:%[0-9]+]]:vreg_64 = PHI [[PHI]], %bb.2, [[REG_SEQUENCE2]], %bb.3
; GFX10-NEXT: SI_END_CF [[SI_ELSE]], implicit-def $exec, implicit-def $scc, implicit $exec
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: bb.6.atomicrmw.end:
; GFX10-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY [[PHI1]].sub0
; GFX10-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[PHI1]].sub1
; GFX10-NEXT: $vgpr0 = COPY [[COPY13]]
; GFX10-NEXT: $vgpr1 = COPY [[COPY14]]
; GFX10-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1
%old = atomicrmw fmin ptr %ptr, double %val seq_cst, !amdgpu.no.fine.grained.memory !0
ret double %old
}
define void @atomicrmw_fmin_local_f32_vv_noret(ptr addrspace(3) %ptr, float %val) {
; GFX10-LABEL: name: atomicrmw_fmin_local_f32_vv_noret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: DS_MIN_F32_gfx9 [[COPY]], [[COPY1]], 0, 0, implicit $exec :: (load store seq_cst (s32) on %ir.ptr, addrspace 3)
; GFX10-NEXT: SI_RETURN
%old = atomicrmw fmin ptr addrspace(3) %ptr, float %val seq_cst
ret void
}
define float @atomicrmw_fmin_local_f32_vv_ret(ptr addrspace(3) %ptr, float %val) {
; GFX10-LABEL: name: atomicrmw_fmin_local_f32_vv_ret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[DS_MIN_RTN_F32_gfx9_:%[0-9]+]]:vgpr_32 = DS_MIN_RTN_F32_gfx9 [[COPY]], [[COPY1]], 0, 0, implicit $exec :: (load store seq_cst (s32) on %ir.ptr, addrspace 3)
; GFX10-NEXT: $vgpr0 = COPY [[DS_MIN_RTN_F32_gfx9_]]
; GFX10-NEXT: SI_RETURN implicit $vgpr0
%old = atomicrmw fmin ptr addrspace(3) %ptr, float %val seq_cst
ret float %old
}
define void @atomicrmw_fmin_local_f64_vv_noret(ptr addrspace(3) %ptr, double %val) {
; GFX10-LABEL: name: atomicrmw_fmin_local_f64_vv_noret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX10-NEXT: DS_MIN_F64_gfx9 [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst (s64) on %ir.ptr, addrspace 3)
; GFX10-NEXT: SI_RETURN
%old = atomicrmw fmin ptr addrspace(3) %ptr, double %val seq_cst
ret void
}
define double @atomicrmw_fmin_local_f64_vv_ret(ptr addrspace(3) %ptr, double %val) {
; GFX10-LABEL: name: atomicrmw_fmin_local_f64_vv_ret
; GFX10: bb.1 (%ir-block.0):
; GFX10-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX10-NEXT: [[DS_MIN_RTN_F64_gfx9_:%[0-9]+]]:vreg_64 = DS_MIN_RTN_F64_gfx9 [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst (s64) on %ir.ptr, addrspace 3)
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[DS_MIN_RTN_F64_gfx9_]].sub0
; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[DS_MIN_RTN_F64_gfx9_]].sub1
; GFX10-NEXT: $vgpr0 = COPY [[COPY3]]
; GFX10-NEXT: $vgpr1 = COPY [[COPY4]]
; GFX10-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1
%old = atomicrmw fmin ptr addrspace(3) %ptr, double %val seq_cst
ret double %old
}
!0 = !{}
Reference in new issue View Git Blame Copy Permalink