1 # RUN: llc -march=amdgcn -mcpu=gfx803 -run-pass si-memory-legalizer %s -o - | FileCheck %s 2 3 --- | 4 ; ModuleID = '<stdin>' 5 source_filename = "<stdin>" 6 target datalayout = "e-p:32:32-p1:64:64-p2:64:64-p3:32:32-p4:64:64-p5:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64" 7 8 ; Function Attrs: nounwind readnone 9 declare i32 @llvm.amdgcn.workitem.id.x() #0 10 11 ; Function Attrs: nounwind 12 define amdgpu_kernel void @atomic_max_i32_noret( 13 i32 addrspace(1)* %out, 14 i32 addrspace(1)* addrspace(1)* %in, 15 i32 addrspace(1)* %x, 16 i32 %y) #1 { 17 %tid = call i32 @llvm.amdgcn.workitem.id.x() 18 %idxprom = sext i32 %tid to i64 19 %tid.gep = getelementptr i32 addrspace(1)*, i32 addrspace(1)* addrspace(1)* %in, i64 %idxprom 20 %ptr = load volatile i32 addrspace(1)*, i32 addrspace(1)* addrspace(1)* %tid.gep 21 %xor = xor i32 %tid, 1 22 %cmp = icmp ne i32 %xor, 0 23 %1 = call { i1, i64 } @llvm.amdgcn.if(i1 %cmp) 24 %2 = extractvalue { i1, i64 } %1, 0 25 %3 = extractvalue { i1, i64 } %1, 1 26 br i1 %2, label %atomic, label %exit 27 28 atomic: ; preds = %0 29 %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 100 30 %ret = atomicrmw max i32 addrspace(1)* %gep, i32 %y seq_cst 31 br label %exit 32 33 exit: ; preds = %atomic, %0 34 call void @llvm.amdgcn.end.cf(i64 %3) 35 ret void 36 } 37 38 declare { i1, i64 } @llvm.amdgcn.if(i1) 39 40 declare void @llvm.amdgcn.end.cf(i64) 41 42 ; Function Attrs: nounwind 43 declare void @llvm.stackprotector(i8*, i8**) #3 44 45 attributes #0 = { nounwind readnone "target-cpu"="tahiti" } 46 attributes #1 = { nounwind "target-cpu"="tahiti" } 47 attributes #2 = { readnone } 48 attributes #3 = { nounwind } 49 50 ... 51 --- 52 53 # CHECK-LABEL: name: atomic_max_i32_noret 54 55 # CHECK-LABEL: bb.1.atomic: 56 # CHECK: BUFFER_ATOMIC_SMAX_ADDR64 57 # CHECK-NEXT: S_WAITCNT 3952 58 # CHECK-NEXT: BUFFER_WBINVL1_VOL 59 60 name: atomic_max_i32_noret 61 alignment: 0 62 exposesReturnsTwice: false 63 legalized: false 64 regBankSelected: false 65 selected: false 66 tracksRegLiveness: true 67 liveins: 68 - { reg: '$sgpr0_sgpr1' } 69 - { reg: '$vgpr0' } 70 frameInfo: 71 isFrameAddressTaken: false 72 isReturnAddressTaken: false 73 hasStackMap: false 74 hasPatchPoint: false 75 stackSize: 0 76 offsetAdjustment: 0 77 maxAlignment: 0 78 adjustsStack: false 79 hasCalls: false 80 maxCallFrameSize: 0 81 hasOpaqueSPAdjustment: false 82 hasVAStart: false 83 hasMustTailInVarArgFunc: false 84 body: | 85 bb.0 (%ir-block.0): 86 successors: %bb.1.atomic(0x40000000), %bb.2.exit(0x40000000) 87 liveins: $vgpr0, $sgpr0_sgpr1 88 89 $sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM $sgpr0_sgpr1, 11, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(2)* undef`) 90 $vgpr1 = V_ASHRREV_I32_e32 31, $vgpr0, implicit $exec 91 $vgpr1_vgpr2 = V_LSHL_B64 $vgpr0_vgpr1, 3, implicit $exec 92 $sgpr7 = S_MOV_B32 61440 93 $sgpr6 = S_MOV_B32 0 94 S_WAITCNT 127 95 $vgpr1_vgpr2 = BUFFER_LOAD_DWORDX2_ADDR64 killed $vgpr1_vgpr2, $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, implicit $exec :: (volatile load 8 from %ir.tid.gep) 96 $vgpr0 = V_XOR_B32_e32 1, killed $vgpr0, implicit $exec 97 V_CMP_NE_U32_e32 0, killed $vgpr0, implicit-def $vcc, implicit $exec 98 $sgpr2_sgpr3 = S_AND_SAVEEXEC_B64 killed $vcc, implicit-def $exec, implicit-def $scc, implicit $exec 99 $sgpr2_sgpr3 = S_XOR_B64 $exec, killed $sgpr2_sgpr3, implicit-def dead $scc 100 SI_MASK_BRANCH %bb.2.exit, implicit $exec 101 102 bb.1.atomic: 103 successors: %bb.2.exit(0x80000000) 104 liveins: $sgpr4_sgpr5_sgpr6_sgpr7:0x0000000C, $sgpr0_sgpr1, $sgpr2_sgpr3, $vgpr1_vgpr2_vgpr3_vgpr4:0x00000003 105 106 $sgpr0 = S_LOAD_DWORD_IMM killed $sgpr0_sgpr1, 15, 0 :: (non-temporal dereferenceable invariant load 4 from `i32 addrspace(2)* undef`) 107 dead $vgpr0 = V_MOV_B32_e32 -1, implicit $exec 108 dead $vgpr0 = V_MOV_B32_e32 61440, implicit $exec 109 $sgpr4_sgpr5 = S_MOV_B64 0 110 S_WAITCNT 127 111 $vgpr0 = V_MOV_B32_e32 killed $sgpr0, implicit $exec, implicit $exec 112 S_WAITCNT 3952 113 BUFFER_ATOMIC_SMAX_ADDR64 killed $vgpr0, killed $vgpr1_vgpr2, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 400, 0, implicit $exec :: (volatile load seq_cst 4 from %ir.gep) 114 115 bb.2.exit: 116 liveins: $sgpr2_sgpr3 117 118 $exec = S_OR_B64 $exec, killed $sgpr2_sgpr3, implicit-def $scc 119 S_ENDPGM 120 121 ... 122 123
