160 lines · plain
1# RUN: llc -mtriple=amdgcn -mcpu=gfx803 -run-pass si-memory-legalizer %s -o - | FileCheck %s2 3--- |4 define amdgpu_kernel void @multiple_mem_operands(ptr addrspace(1) %out, i32 %cond, i32 %if_offset, i32 %else_offset) #0 {5 entry:6 %scratch0 = alloca [8192 x i32], addrspace(5)7 %scratch1 = alloca [8192 x i32], addrspace(5)8 %scratchptr01 = bitcast ptr addrspace(5) %scratch0 to ptr addrspace(5)9 store i32 1, ptr addrspace(5) %scratchptr0110 %scratchptr12 = bitcast ptr addrspace(5) %scratch1 to ptr addrspace(5)11 store i32 2, ptr addrspace(5) %scratchptr1212 %cmp = icmp eq i32 %cond, 013 br i1 %cmp, label %if, label %else, !structurizecfg.uniform !0, !amdgpu.uniform !014 15 if: ; preds = %entry16 %if_ptr = getelementptr [8192 x i32], ptr addrspace(5) %scratch0, i32 0, i32 %if_offset, !amdgpu.uniform !017 %if_value = load i32, ptr addrspace(5) %if_ptr, align 4, !nontemporal !118 br label %done, !structurizecfg.uniform !019 20 else: ; preds = %entry21 %else_ptr = getelementptr [8192 x i32], ptr addrspace(5) %scratch1, i32 0, i32 %else_offset, !amdgpu.uniform !022 %else_value = load i32, ptr addrspace(5) %else_ptr, align 4, !nontemporal !123 br label %done, !structurizecfg.uniform !024 25 done: ; preds = %else, %if26 %value = phi i32 [ %if_value, %if ], [ %else_value, %else ]27 store i32 %value, ptr addrspace(1) %out28 ret void29 }30 31 ; Function Attrs: convergent nounwind32 declare { i1, i64 } @llvm.amdgcn.if(i1) #133 34 ; Function Attrs: convergent nounwind35 declare { i1, i64 } @llvm.amdgcn.else(i64) #136 37 ; Function Attrs: convergent nounwind readnone38 declare i64 @llvm.amdgcn.break(i64) #239 40 ; Function Attrs: convergent nounwind readnone41 declare i64 @llvm.amdgcn.if.break(i1, i64) #242 43 ; Function Attrs: convergent nounwind readnone44 declare i64 @llvm.amdgcn.else.break(i64, i64) #245 46 ; Function Attrs: convergent nounwind47 declare i1 @llvm.amdgcn.loop(i64) #148 49 ; Function Attrs: convergent nounwind50 declare void @llvm.amdgcn.end.cf(i64) #151 52 attributes #0 = { "target-cpu"="gfx803" }53 attributes #1 = { convergent nounwind }54 attributes #2 = { convergent nounwind readnone }55 56 !0 = !{}57 !1 = !{i32 1}58 59...60---61 62# CHECK-LABEL: name: multiple_mem_operands63 64# CHECK-LABEL: bb.3.done:65# CHECK: BUFFER_LOAD_DWORD_OFFEN killed $vgpr0, killed $sgpr8_sgpr9_sgpr10_sgpr11, $sgpr3, 0, 366 67name: multiple_mem_operands68alignment: 169exposesReturnsTwice: false70legalized: false71regBankSelected: false72selected: false73tracksRegLiveness: true74registers:75liveins:76 - { reg: '$sgpr0_sgpr1', virtual-reg: '' }77 - { reg: '$sgpr3', virtual-reg: '' }78frameInfo:79 isFrameAddressTaken: false80 isReturnAddressTaken: false81 hasStackMap: false82 hasPatchPoint: false83 stackSize: 6554084 offsetAdjustment: 085 maxAlignment: 486 adjustsStack: false87 hasCalls: false88 stackProtector: ''89 maxCallFrameSize: 090 hasOpaqueSPAdjustment: false91 hasVAStart: false92 hasMustTailInVarArgFunc: false93 savePoint: []94 restorePoint: []95fixedStack:96 - { id: 0, type: default, offset: 0, size: 4, alignment: 4, stack-id: default,97 isImmutable: false, isAliased: false, callee-saved-register: '' }98stack:99 - { id: 0, name: scratch0, type: default, offset: 4, size: 32768, alignment: 4,100 stack-id: default, callee-saved-register: '', local-offset: 0,101 debug-info-variable: '', debug-info-expression: '',102 debug-info-location: '' }103 - { id: 1, name: scratch1, type: default, offset: 32772, size: 32768,104 alignment: 4, stack-id: default, callee-saved-register: '', local-offset: 32768,105 debug-info-variable: '', debug-info-expression: '',106 debug-info-location: '' }107constants:108body: |109 bb.0.entry:110 successors: %bb.1.if(0x30000000), %bb.2.else(0x50000000)111 liveins: $sgpr0_sgpr1, $sgpr3112 113 $sgpr2 = S_LOAD_DWORD_IMM $sgpr0_sgpr1, 44, 0 :: (non-temporal dereferenceable invariant load (s32) from `ptr addrspace(4) poison`)114 $sgpr8 = S_MOV_B32 &SCRATCH_RSRC_DWORD0, implicit-def $sgpr8_sgpr9_sgpr10_sgpr11115 $sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM $sgpr0_sgpr1, 36, 0 :: (non-temporal dereferenceable invariant load (s64) from `ptr addrspace(4) poison`)116 $sgpr9 = S_MOV_B32 &SCRATCH_RSRC_DWORD1, implicit-def $sgpr8_sgpr9_sgpr10_sgpr11117 $sgpr10 = S_MOV_B32 4294967295, implicit-def $sgpr8_sgpr9_sgpr10_sgpr11118 $sgpr11 = S_MOV_B32 15204352, implicit-def $sgpr8_sgpr9_sgpr10_sgpr11119 $vgpr0 = V_MOV_B32_e32 1, implicit $exec120 BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr8_sgpr9_sgpr10_sgpr11, $sgpr3, 4, 0, 0, implicit $exec :: (store (s32) into %ir.scratchptr01)121 S_WAITCNT 127122 S_CMP_LG_U32 killed $sgpr2, 0, implicit-def $scc123 S_WAITCNT 3855124 $vgpr0 = V_MOV_B32_e32 2, implicit $exec125 $vgpr1 = V_MOV_B32_e32 32772, implicit $exec126 BUFFER_STORE_DWORD_OFFEN killed $vgpr0, killed $vgpr1, $sgpr8_sgpr9_sgpr10_sgpr11, $sgpr3, 0, 0, 0, implicit $exec :: (store (s32) into %ir.scratchptr12)127 S_CBRANCH_SCC0 %bb.1.if, implicit killed $scc128 129 bb.2.else:130 successors: %bb.3.done(0x80000000)131 liveins: $sgpr0_sgpr1, $sgpr4_sgpr5, $sgpr3, $sgpr8_sgpr9_sgpr10_sgpr11132 133 $sgpr0 = S_LOAD_DWORD_IMM killed $sgpr0_sgpr1, 52, 0 :: (non-temporal dereferenceable invariant load (s32) from `ptr addrspace(4) poison`)134 S_WAITCNT 3855135 $vgpr0 = V_MOV_B32_e32 32772, implicit $exec136 S_BRANCH %bb.3.done137 138 bb.1.if:139 successors: %bb.3.done(0x80000000)140 liveins: $sgpr0_sgpr1, $sgpr4_sgpr5, $sgpr3, $sgpr8_sgpr9_sgpr10_sgpr11141 142 $sgpr0 = S_LOAD_DWORD_IMM killed $sgpr0_sgpr1, 48, 0 :: (non-temporal dereferenceable invariant load (s32) from `ptr addrspace(4) poison`)143 S_WAITCNT 3855144 $vgpr0 = V_MOV_B32_e32 4, implicit $exec145 146 bb.3.done:147 liveins: $sgpr3, $sgpr4_sgpr5, $sgpr8_sgpr9_sgpr10_sgpr11, $vgpr0, $sgpr0148 149 S_WAITCNT 127150 $sgpr0 = S_LSHL_B32 killed $sgpr0, 2, implicit-def dead $scc151 $vgpr0 = V_ADD_CO_U32_e32 killed $sgpr0, killed $vgpr0, implicit-def dead $vcc, implicit $exec152 $vgpr0 = BUFFER_LOAD_DWORD_OFFEN killed $vgpr0, killed $sgpr8_sgpr9_sgpr10_sgpr11, $sgpr3, 0, 0, 0, implicit $exec :: (non-temporal load (s32) from %ir.else_ptr), (non-temporal load (s32) from %ir.if_ptr)153 $vgpr1 = V_MOV_B32_e32 $sgpr4, implicit $exec, implicit-def $vgpr1_vgpr2, implicit $sgpr4_sgpr5154 $vgpr2 = V_MOV_B32_e32 killed $sgpr5, implicit $exec, implicit $sgpr4_sgpr5, implicit $exec155 S_WAITCNT 3952156 FLAT_STORE_DWORD killed $vgpr1_vgpr2, killed $vgpr0, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %ir.out)157 S_ENDPGM 0158 159...160