brintos

brintos / llvm-project-archived public Read only

0
0
Text · 18.7 KiB · 85549b8 Raw
220 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py2; RUN: llc -mtriple=amdgcn -mcpu=gfx908 -stop-after=amdgpu-isel -amdgpu-atomic-optimizer-strategy=DPP < %s | FileCheck -check-prefixes=GFX908_GFX11_GFX12,GFX908 %s3; RUN: llc -mtriple=amdgcn -mcpu=gfx90a -stop-after=amdgpu-isel -amdgpu-atomic-optimizer-strategy=DPP < %s | FileCheck -check-prefixes=GFX90A_GFX942 %s4; RUN: llc -mtriple=amdgcn -mcpu=gfx942 -stop-after=amdgpu-isel -amdgpu-atomic-optimizer-strategy=DPP < %s | FileCheck -check-prefixes=GFX90A_GFX942 %s5; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -stop-after=amdgpu-isel -amdgpu-atomic-optimizer-strategy=DPP < %s | FileCheck -check-prefixes=GFX908_GFX11_GFX12,GFX11_GFX12 %s6; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -stop-after=amdgpu-isel -amdgpu-atomic-optimizer-strategy=DPP < %s | FileCheck -check-prefixes=GFX908_GFX11_GFX12,GFX11_GFX12 %s7 8define amdgpu_ps void @global_atomic_fadd_f32_no_rtn_atomicrmw(ptr addrspace(1) %ptr, float %data) {9  ; GFX908_GFX11_GFX12-LABEL: name: global_atomic_fadd_f32_no_rtn_atomicrmw10  ; GFX908_GFX11_GFX12: bb.0 (%ir-block.0):11  ; GFX908_GFX11_GFX12-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr212  ; GFX908_GFX11_GFX12-NEXT: {{  $}}13  ; GFX908_GFX11_GFX12-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr214  ; GFX908_GFX11_GFX12-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr115  ; GFX908_GFX11_GFX12-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr016  ; GFX908_GFX11_GFX12-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub117  ; GFX908_GFX11_GFX12-NEXT:   [[COPY3:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]18  ; GFX908_GFX11_GFX12-NEXT:   GLOBAL_ATOMIC_ADD_F32 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)19  ; GFX908_GFX11_GFX12-NEXT:   S_ENDPGM 020  ;21  ; GFX90A_GFX942-LABEL: name: global_atomic_fadd_f32_no_rtn_atomicrmw22  ; GFX90A_GFX942: bb.0 (%ir-block.0):23  ; GFX90A_GFX942-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr224  ; GFX90A_GFX942-NEXT: {{  $}}25  ; GFX90A_GFX942-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr226  ; GFX90A_GFX942-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr127  ; GFX90A_GFX942-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr028  ; GFX90A_GFX942-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub129  ; GFX90A_GFX942-NEXT:   [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]30  ; GFX90A_GFX942-NEXT:   GLOBAL_ATOMIC_ADD_F32 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)31  ; GFX90A_GFX942-NEXT:   S_ENDPGM 032  %ret = atomicrmw fadd ptr addrspace(1) %ptr, float %data syncscope("wavefront") monotonic, !amdgpu.no.fine.grained.memory !0, !amdgpu.ignore.denormal.mode !033  ret void34}35 36define amdgpu_ps void @global_atomic_fadd_f32_saddr_no_rtn_atomicrmw(ptr addrspace(1) inreg %ptr, float %data) {37  ; GFX908-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_atomicrmw38  ; GFX908: bb.0 (%ir-block.0):39  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.4(0x40000000)40  ; GFX908-NEXT:   liveins: $sgpr0, $sgpr1, $vgpr041  ; GFX908-NEXT: {{  $}}42  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr043  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr144  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr045  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub146  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:sreg_64_xexec_xnull = COPY [[REG_SEQUENCE]]47  ; GFX908-NEXT:   [[SI_PS_LIVE:%[0-9]+]]:sreg_64 = SI_PS_LIVE48  ; GFX908-NEXT:   [[SI_IF:%[0-9]+]]:sreg_64 = SI_IF killed [[SI_PS_LIVE]], %bb.4, implicit-def dead $exec, implicit-def dead $scc, implicit $exec49  ; GFX908-NEXT:   S_BRANCH %bb.150  ; GFX908-NEXT: {{  $}}51  ; GFX908-NEXT: bb.1 (%ir-block.5):52  ; GFX908-NEXT:   successors: %bb.2(0x40000000), %bb.3(0x40000000)53  ; GFX908-NEXT: {{  $}}54  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:sreg_64 = COPY $exec55  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY [[COPY4]].sub156  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY [[COPY4]].sub057  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 058  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]59  ; GFX908-NEXT:   [[V_MBCNT_LO_U32_B32_e64_:%[0-9]+]]:vgpr_32 = V_MBCNT_LO_U32_B32_e64 killed [[COPY6]], [[COPY7]], implicit $exec60  ; GFX908-NEXT:   [[V_MBCNT_HI_U32_B32_e64_:%[0-9]+]]:vgpr_32 = V_MBCNT_HI_U32_B32_e64 killed [[COPY5]], killed [[V_MBCNT_LO_U32_B32_e64_]], implicit $exec61  ; GFX908-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -2147483648, implicit $exec62  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF63  ; GFX908-NEXT:   [[V_SET_INACTIVE_B32_:%[0-9]+]]:vgpr_32 = V_SET_INACTIVE_B32 0, [[COPY]], 0, [[V_MOV_B32_e32_]], killed [[DEF]], implicit $exec64  ; GFX908-NEXT:   [[V_MOV_B32_dpp:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_SET_INACTIVE_B32_]], 273, 15, 15, 0, implicit $exec65  ; GFX908-NEXT:   [[V_ADD_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_SET_INACTIVE_B32_]], 0, killed [[V_MOV_B32_dpp]], 0, 0, implicit $mode, implicit $exec66  ; GFX908-NEXT:   [[V_MOV_B32_dpp1:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_]], 274, 15, 15, 0, implicit $exec67  ; GFX908-NEXT:   [[V_ADD_F32_e64_1:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_]], 0, killed [[V_MOV_B32_dpp1]], 0, 0, implicit $mode, implicit $exec68  ; GFX908-NEXT:   [[V_MOV_B32_dpp2:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_1]], 276, 15, 15, 0, implicit $exec69  ; GFX908-NEXT:   [[V_ADD_F32_e64_2:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_1]], 0, killed [[V_MOV_B32_dpp2]], 0, 0, implicit $mode, implicit $exec70  ; GFX908-NEXT:   [[V_MOV_B32_dpp3:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_2]], 280, 15, 15, 0, implicit $exec71  ; GFX908-NEXT:   [[V_ADD_F32_e64_3:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_2]], 0, killed [[V_MOV_B32_dpp3]], 0, 0, implicit $mode, implicit $exec72  ; GFX908-NEXT:   [[V_MOV_B32_dpp4:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_3]], 322, 10, 15, 0, implicit $exec73  ; GFX908-NEXT:   [[V_ADD_F32_e64_4:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_3]], 0, killed [[V_MOV_B32_dpp4]], 0, 0, implicit $mode, implicit $exec74  ; GFX908-NEXT:   [[V_MOV_B32_dpp5:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_4]], 323, 12, 15, 0, implicit $exec75  ; GFX908-NEXT:   [[V_ADD_F32_e64_5:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_4]], 0, killed [[V_MOV_B32_dpp5]], 0, 0, implicit $mode, implicit $exec76  ; GFX908-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 6377  ; GFX908-NEXT:   [[V_READLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READLANE_B32 killed [[V_ADD_F32_e64_5]], killed [[S_MOV_B32_1]]78  ; GFX908-NEXT:   early-clobber %1:sgpr_32 = STRICT_WWM killed [[V_READLANE_B32_]], implicit $exec79  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 killed [[V_MBCNT_HI_U32_B32_e64_]], [[S_MOV_B32_]], implicit $exec80  ; GFX908-NEXT:   [[SI_IF1:%[0-9]+]]:sreg_64 = SI_IF killed [[V_CMP_EQ_U32_e64_]], %bb.3, implicit-def dead $exec, implicit-def dead $scc, implicit $exec81  ; GFX908-NEXT:   S_BRANCH %bb.282  ; GFX908-NEXT: {{  $}}83  ; GFX908-NEXT: bb.2 (%ir-block.31):84  ; GFX908-NEXT:   successors: %bb.3(0x80000000)85  ; GFX908-NEXT: {{  $}}86  ; GFX908-NEXT:   [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec87  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY %188  ; GFX908-NEXT:   GLOBAL_ATOMIC_ADD_F32_SADDR killed [[V_MOV_B32_e32_1]], [[COPY8]], [[COPY3]], 0, 0, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)89  ; GFX908-NEXT: {{  $}}90  ; GFX908-NEXT: bb.3.Flow:91  ; GFX908-NEXT:   successors: %bb.4(0x80000000)92  ; GFX908-NEXT: {{  $}}93  ; GFX908-NEXT:   SI_END_CF [[SI_IF1]], implicit-def dead $exec, implicit-def dead $scc, implicit $exec94  ; GFX908-NEXT: {{  $}}95  ; GFX908-NEXT: bb.4 (%ir-block.33):96  ; GFX908-NEXT:   SI_END_CF [[SI_IF]], implicit-def dead $exec, implicit-def dead $scc, implicit $exec97  ; GFX908-NEXT:   S_ENDPGM 098  ;99  ; GFX90A_GFX942-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_atomicrmw100  ; GFX90A_GFX942: bb.0 (%ir-block.0):101  ; GFX90A_GFX942-NEXT:   successors: %bb.1(0x40000000), %bb.4(0x40000000)102  ; GFX90A_GFX942-NEXT:   liveins: $sgpr0, $sgpr1, $vgpr0103  ; GFX90A_GFX942-NEXT: {{  $}}104  ; GFX90A_GFX942-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0105  ; GFX90A_GFX942-NEXT:   [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1106  ; GFX90A_GFX942-NEXT:   [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0107  ; GFX90A_GFX942-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1108  ; GFX90A_GFX942-NEXT:   [[COPY3:%[0-9]+]]:sreg_64_xexec_xnull = COPY [[REG_SEQUENCE]]109  ; GFX90A_GFX942-NEXT:   [[SI_PS_LIVE:%[0-9]+]]:sreg_64 = SI_PS_LIVE110  ; GFX90A_GFX942-NEXT:   [[SI_IF:%[0-9]+]]:sreg_64 = SI_IF killed [[SI_PS_LIVE]], %bb.4, implicit-def dead $exec, implicit-def dead $scc, implicit $exec111  ; GFX90A_GFX942-NEXT:   S_BRANCH %bb.1112  ; GFX90A_GFX942-NEXT: {{  $}}113  ; GFX90A_GFX942-NEXT: bb.1 (%ir-block.5):114  ; GFX90A_GFX942-NEXT:   successors: %bb.2(0x40000000), %bb.3(0x40000000)115  ; GFX90A_GFX942-NEXT: {{  $}}116  ; GFX90A_GFX942-NEXT:   [[COPY4:%[0-9]+]]:sreg_64 = COPY $exec117  ; GFX90A_GFX942-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY [[COPY4]].sub1118  ; GFX90A_GFX942-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY [[COPY4]].sub0119  ; GFX90A_GFX942-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0120  ; GFX90A_GFX942-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]121  ; GFX90A_GFX942-NEXT:   [[V_MBCNT_LO_U32_B32_e64_:%[0-9]+]]:vgpr_32 = V_MBCNT_LO_U32_B32_e64 killed [[COPY6]], [[COPY7]], implicit $exec122  ; GFX90A_GFX942-NEXT:   [[V_MBCNT_HI_U32_B32_e64_:%[0-9]+]]:vgpr_32 = V_MBCNT_HI_U32_B32_e64 killed [[COPY5]], killed [[V_MBCNT_LO_U32_B32_e64_]], implicit $exec123  ; GFX90A_GFX942-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -2147483648, implicit $exec124  ; GFX90A_GFX942-NEXT:   [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF125  ; GFX90A_GFX942-NEXT:   [[V_SET_INACTIVE_B32_:%[0-9]+]]:vgpr_32 = V_SET_INACTIVE_B32 0, [[COPY]], 0, [[V_MOV_B32_e32_]], killed [[DEF]], implicit $exec126  ; GFX90A_GFX942-NEXT:   [[V_MOV_B32_dpp:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_SET_INACTIVE_B32_]], 273, 15, 15, 0, implicit $exec127  ; GFX90A_GFX942-NEXT:   [[V_ADD_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_SET_INACTIVE_B32_]], 0, killed [[V_MOV_B32_dpp]], 0, 0, implicit $mode, implicit $exec128  ; GFX90A_GFX942-NEXT:   [[V_MOV_B32_dpp1:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_]], 274, 15, 15, 0, implicit $exec129  ; GFX90A_GFX942-NEXT:   [[V_ADD_F32_e64_1:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_]], 0, killed [[V_MOV_B32_dpp1]], 0, 0, implicit $mode, implicit $exec130  ; GFX90A_GFX942-NEXT:   [[V_MOV_B32_dpp2:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_1]], 276, 15, 15, 0, implicit $exec131  ; GFX90A_GFX942-NEXT:   [[V_ADD_F32_e64_2:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_1]], 0, killed [[V_MOV_B32_dpp2]], 0, 0, implicit $mode, implicit $exec132  ; GFX90A_GFX942-NEXT:   [[V_MOV_B32_dpp3:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_2]], 280, 15, 15, 0, implicit $exec133  ; GFX90A_GFX942-NEXT:   [[V_ADD_F32_e64_3:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_2]], 0, killed [[V_MOV_B32_dpp3]], 0, 0, implicit $mode, implicit $exec134  ; GFX90A_GFX942-NEXT:   [[V_MOV_B32_dpp4:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_3]], 322, 10, 15, 0, implicit $exec135  ; GFX90A_GFX942-NEXT:   [[V_ADD_F32_e64_4:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_3]], 0, killed [[V_MOV_B32_dpp4]], 0, 0, implicit $mode, implicit $exec136  ; GFX90A_GFX942-NEXT:   [[V_MOV_B32_dpp5:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_4]], 323, 12, 15, 0, implicit $exec137  ; GFX90A_GFX942-NEXT:   [[V_ADD_F32_e64_5:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_4]], 0, killed [[V_MOV_B32_dpp5]], 0, 0, implicit $mode, implicit $exec138  ; GFX90A_GFX942-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 63139  ; GFX90A_GFX942-NEXT:   [[V_READLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READLANE_B32 killed [[V_ADD_F32_e64_5]], killed [[S_MOV_B32_1]]140  ; GFX90A_GFX942-NEXT:   early-clobber %1:sgpr_32 = STRICT_WWM killed [[V_READLANE_B32_]], implicit $exec141  ; GFX90A_GFX942-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 killed [[V_MBCNT_HI_U32_B32_e64_]], [[S_MOV_B32_]], implicit $exec142  ; GFX90A_GFX942-NEXT:   [[SI_IF1:%[0-9]+]]:sreg_64 = SI_IF killed [[V_CMP_EQ_U32_e64_]], %bb.3, implicit-def dead $exec, implicit-def dead $scc, implicit $exec143  ; GFX90A_GFX942-NEXT:   S_BRANCH %bb.2144  ; GFX90A_GFX942-NEXT: {{  $}}145  ; GFX90A_GFX942-NEXT: bb.2 (%ir-block.31):146  ; GFX90A_GFX942-NEXT:   successors: %bb.3(0x80000000)147  ; GFX90A_GFX942-NEXT: {{  $}}148  ; GFX90A_GFX942-NEXT:   [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec149  ; GFX90A_GFX942-NEXT:   [[COPY8:%[0-9]+]]:av_32 = COPY %1150  ; GFX90A_GFX942-NEXT:   GLOBAL_ATOMIC_ADD_F32_SADDR killed [[V_MOV_B32_e32_1]], [[COPY8]], [[COPY3]], 0, 0, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)151  ; GFX90A_GFX942-NEXT: {{  $}}152  ; GFX90A_GFX942-NEXT: bb.3.Flow:153  ; GFX90A_GFX942-NEXT:   successors: %bb.4(0x80000000)154  ; GFX90A_GFX942-NEXT: {{  $}}155  ; GFX90A_GFX942-NEXT:   SI_END_CF [[SI_IF1]], implicit-def dead $exec, implicit-def dead $scc, implicit $exec156  ; GFX90A_GFX942-NEXT: {{  $}}157  ; GFX90A_GFX942-NEXT: bb.4 (%ir-block.33):158  ; GFX90A_GFX942-NEXT:   SI_END_CF [[SI_IF]], implicit-def dead $exec, implicit-def dead $scc, implicit $exec159  ; GFX90A_GFX942-NEXT:   S_ENDPGM 0160  ;161  ; GFX11_GFX12-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_atomicrmw162  ; GFX11_GFX12: bb.0 (%ir-block.0):163  ; GFX11_GFX12-NEXT:   successors: %bb.1(0x40000000), %bb.4(0x40000000)164  ; GFX11_GFX12-NEXT:   liveins: $sgpr0, $sgpr1, $vgpr0165  ; GFX11_GFX12-NEXT: {{  $}}166  ; GFX11_GFX12-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0167  ; GFX11_GFX12-NEXT:   [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1168  ; GFX11_GFX12-NEXT:   [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0169  ; GFX11_GFX12-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1170  ; GFX11_GFX12-NEXT:   [[COPY3:%[0-9]+]]:sreg_64_xexec_xnull = COPY [[REG_SEQUENCE]]171  ; GFX11_GFX12-NEXT:   [[SI_PS_LIVE:%[0-9]+]]:sreg_32 = SI_PS_LIVE172  ; GFX11_GFX12-NEXT:   [[SI_IF:%[0-9]+]]:sreg_32 = SI_IF killed [[SI_PS_LIVE]], %bb.4, implicit-def dead $exec, implicit-def dead $scc, implicit $exec173  ; GFX11_GFX12-NEXT:   S_BRANCH %bb.1174  ; GFX11_GFX12-NEXT: {{  $}}175  ; GFX11_GFX12-NEXT: bb.1 (%ir-block.5):176  ; GFX11_GFX12-NEXT:   successors: %bb.2(0x40000000), %bb.3(0x40000000)177  ; GFX11_GFX12-NEXT: {{  $}}178  ; GFX11_GFX12-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $exec_lo179  ; GFX11_GFX12-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0180  ; GFX11_GFX12-NEXT:   [[V_MBCNT_LO_U32_B32_e64_:%[0-9]+]]:vgpr_32 = V_MBCNT_LO_U32_B32_e64 [[COPY4]], [[S_MOV_B32_]], implicit $exec181  ; GFX11_GFX12-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -2147483648, implicit $exec182  ; GFX11_GFX12-NEXT:   [[DEF:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF183  ; GFX11_GFX12-NEXT:   [[V_SET_INACTIVE_B32_:%[0-9]+]]:vgpr_32 = V_SET_INACTIVE_B32 0, [[COPY]], 0, [[V_MOV_B32_e32_]], killed [[DEF]], implicit $exec184  ; GFX11_GFX12-NEXT:   [[V_MOV_B32_dpp:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_SET_INACTIVE_B32_]], 353, 15, 15, 0, implicit $exec185  ; GFX11_GFX12-NEXT:   [[V_ADD_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_SET_INACTIVE_B32_]], 0, killed [[V_MOV_B32_dpp]], 0, 0, implicit $mode, implicit $exec186  ; GFX11_GFX12-NEXT:   [[V_MOV_B32_dpp1:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_]], 354, 15, 15, 0, implicit $exec187  ; GFX11_GFX12-NEXT:   [[V_ADD_F32_e64_1:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_]], 0, killed [[V_MOV_B32_dpp1]], 0, 0, implicit $mode, implicit $exec188  ; GFX11_GFX12-NEXT:   [[V_MOV_B32_dpp2:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_1]], 356, 15, 15, 0, implicit $exec189  ; GFX11_GFX12-NEXT:   [[V_ADD_F32_e64_2:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_1]], 0, killed [[V_MOV_B32_dpp2]], 0, 0, implicit $mode, implicit $exec190  ; GFX11_GFX12-NEXT:   [[V_MOV_B32_dpp3:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[V_MOV_B32_e32_]], [[V_ADD_F32_e64_2]], 360, 15, 15, 0, implicit $exec191  ; GFX11_GFX12-NEXT:   [[V_ADD_F32_e64_3:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_2]], 0, killed [[V_MOV_B32_dpp3]], 0, 0, implicit $mode, implicit $exec192  ; GFX11_GFX12-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF193  ; GFX11_GFX12-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[DEF1]]194  ; GFX11_GFX12-NEXT:   [[V_PERMLANEX16_B32_e64_:%[0-9]+]]:vgpr_32 = V_PERMLANEX16_B32_e64 0, [[V_ADD_F32_e64_3]], 0, [[S_MOV_B32_]], 0, [[S_MOV_B32_]], [[COPY5]], 0, implicit $exec195  ; GFX11_GFX12-NEXT:   [[V_ADD_F32_e64_4:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_3]], 0, killed [[V_PERMLANEX16_B32_e64_]], 0, 0, implicit $mode, implicit $exec196  ; GFX11_GFX12-NEXT:   early-clobber %1:vgpr_32 = STRICT_WWM killed [[V_ADD_F32_e64_4]], implicit $exec197  ; GFX11_GFX12-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32 = V_CMP_EQ_U32_e64 killed [[V_MBCNT_LO_U32_B32_e64_]], [[S_MOV_B32_]], implicit $exec198  ; GFX11_GFX12-NEXT:   [[SI_IF1:%[0-9]+]]:sreg_32 = SI_IF killed [[V_CMP_EQ_U32_e64_]], %bb.3, implicit-def dead $exec, implicit-def dead $scc, implicit $exec199  ; GFX11_GFX12-NEXT:   S_BRANCH %bb.2200  ; GFX11_GFX12-NEXT: {{  $}}201  ; GFX11_GFX12-NEXT: bb.2 (%ir-block.24):202  ; GFX11_GFX12-NEXT:   successors: %bb.3(0x80000000)203  ; GFX11_GFX12-NEXT: {{  $}}204  ; GFX11_GFX12-NEXT:   [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec205  ; GFX11_GFX12-NEXT:   GLOBAL_ATOMIC_ADD_F32_SADDR killed [[V_MOV_B32_e32_1]], %1, [[COPY3]], 0, 0, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)206  ; GFX11_GFX12-NEXT: {{  $}}207  ; GFX11_GFX12-NEXT: bb.3.Flow:208  ; GFX11_GFX12-NEXT:   successors: %bb.4(0x80000000)209  ; GFX11_GFX12-NEXT: {{  $}}210  ; GFX11_GFX12-NEXT:   SI_END_CF [[SI_IF1]], implicit-def dead $exec, implicit-def dead $scc, implicit $exec211  ; GFX11_GFX12-NEXT: {{  $}}212  ; GFX11_GFX12-NEXT: bb.4 (%ir-block.26):213  ; GFX11_GFX12-NEXT:   SI_END_CF [[SI_IF]], implicit-def dead $exec, implicit-def dead $scc, implicit $exec214  ; GFX11_GFX12-NEXT:   S_ENDPGM 0215  %ret = atomicrmw fadd ptr addrspace(1) %ptr, float %data syncscope("wavefront") monotonic, !amdgpu.no.fine.grained.memory !0, !amdgpu.ignore.denormal.mode !0216  ret void217}218 219!0 = !{}220