186 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=memcpyopt -S < %s -verify-memoryssa | FileCheck %s3 4target datalayout = "e-i64:64-f80:128-n8:16:32:64"5target triple = "x86_64-unknown-linux-gnu"6 7%S = type { ptr, i8, i32 }8 9define void @copy(ptr %src, ptr %dst) {10; CHECK-LABEL: @copy(11; CHECK-NEXT: call void @llvm.memmove.p0.p0.i64(ptr align 8 [[DST:%.*]], ptr align 8 [[SRC:%.*]], i64 16, i1 false)12; CHECK-NEXT: ret void13;14 %1 = load %S, ptr %src15 store %S %1, ptr %dst16 ret void17}18 19define void @noaliassrc(ptr noalias %src, ptr %dst) {20; CHECK-LABEL: @noaliassrc(21; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DST:%.*]], ptr align 8 [[SRC:%.*]], i64 16, i1 false)22; CHECK-NEXT: ret void23;24 %1 = load %S, ptr %src25 store %S %1, ptr %dst26 ret void27}28 29define void @noaliasdst(ptr %src, ptr noalias %dst) {30; CHECK-LABEL: @noaliasdst(31; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DST:%.*]], ptr align 8 [[SRC:%.*]], i64 16, i1 false)32; CHECK-NEXT: ret void33;34 %1 = load %S, ptr %src35 store %S %1, ptr %dst36 ret void37}38 39define void @destroysrc(ptr %src, ptr %dst) {40; CHECK-LABEL: @destroysrc(41; CHECK-NEXT: [[TMP1:%.*]] = load [[S:%.*]], ptr [[SRC:%.*]], align 842; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 8 [[SRC]], i8 0, i64 16, i1 false)43; CHECK-NEXT: store [[S]] [[TMP1]], ptr [[DST:%.*]], align 844; CHECK-NEXT: ret void45;46 %1 = load %S, ptr %src47 store %S zeroinitializer, ptr %src48 store %S %1, ptr %dst49 ret void50}51 52define void @destroynoaliassrc(ptr noalias %src, ptr %dst) {53; CHECK-LABEL: @destroynoaliassrc(54; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DST:%.*]], ptr align 8 [[SRC:%.*]], i64 16, i1 false)55; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 8 [[SRC]], i8 0, i64 16, i1 false)56; CHECK-NEXT: ret void57;58 %1 = load %S, ptr %src59 store %S zeroinitializer, ptr %src60 store %S %1, ptr %dst61 ret void62}63 64define void @copyalias(ptr %src, ptr %dst) {65; CHECK-LABEL: @copyalias(66; CHECK-NEXT: [[TMP1:%.*]] = load [[S:%.*]], ptr [[SRC:%.*]], align 867; CHECK-NEXT: call void @llvm.memmove.p0.p0.i64(ptr align 8 [[DST:%.*]], ptr align 8 [[SRC]], i64 16, i1 false)68; CHECK-NEXT: store [[S]] [[TMP1]], ptr [[DST]], align 869; CHECK-NEXT: ret void70;71 %1 = load %S, ptr %src72 %2 = load %S, ptr %src73 store %S %1, ptr %dst74 store %S %2, ptr %dst75 ret void76}77 78; If the store address is computed in a complex manner, make79; sure we lift the computation as well if needed and possible.80define void @addrproducer(ptr %src, ptr %dst) {81; CHECK-LABEL: @addrproducer(82; CHECK-NEXT: [[DST2:%.*]] = getelementptr [[S:%.*]], ptr [[DST:%.*]], i64 183; CHECK-NEXT: call void @llvm.memmove.p0.p0.i64(ptr align 8 [[DST2]], ptr align 8 [[SRC:%.*]], i64 16, i1 false)84; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 8 [[DST]], i8 undef, i64 16, i1 false)85; CHECK-NEXT: ret void86;87 %1 = load %S, ptr %src88 store %S undef, ptr %dst89 %dst2 = getelementptr %S , ptr %dst, i64 190 store %S %1, ptr %dst291 ret void92}93 94define void @aliasaddrproducer(ptr %src, ptr %dst, ptr %dstidptr) {95; CHECK-LABEL: @aliasaddrproducer(96; CHECK-NEXT: [[TMP1:%.*]] = load [[S:%.*]], ptr [[SRC:%.*]], align 897; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 8 [[DST:%.*]], i8 undef, i64 16, i1 false)98; CHECK-NEXT: [[DSTINDEX:%.*]] = load i32, ptr [[DSTIDPTR:%.*]], align 499; CHECK-NEXT: [[DST2:%.*]] = getelementptr [[S]], ptr [[DST]], i32 [[DSTINDEX]]100; CHECK-NEXT: store [[S]] [[TMP1]], ptr [[DST2]], align 8101; CHECK-NEXT: ret void102;103 %1 = load %S, ptr %src104 store %S undef, ptr %dst105 %dstindex = load i32, ptr %dstidptr106 %dst2 = getelementptr %S , ptr %dst, i32 %dstindex107 store %S %1, ptr %dst2108 ret void109}110 111define void @noaliasaddrproducer(ptr %src, ptr noalias %dst, ptr noalias %dstidptr) {112; CHECK-LABEL: @noaliasaddrproducer(113; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr [[DSTIDPTR:%.*]], align 4114; CHECK-NEXT: [[DSTINDEX:%.*]] = or i32 [[TMP2]], 1115; CHECK-NEXT: [[DST2:%.*]] = getelementptr [[S:%.*]], ptr [[DST:%.*]], i32 [[DSTINDEX]]116; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DST2]], ptr align 8 [[SRC:%.*]], i64 16, i1 false)117; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 8 [[SRC]], i8 undef, i64 16, i1 false)118; CHECK-NEXT: ret void119;120 %1 = load %S, ptr %src121 store %S undef, ptr %src122 %2 = load i32, ptr %dstidptr123 %dstindex = or i32 %2, 1124 %dst2 = getelementptr %S , ptr %dst, i32 %dstindex125 store %S %1, ptr %dst2126 ret void127}128 129define void @throwing_call(ptr noalias %src, ptr %dst) {130; CHECK-LABEL: @throwing_call(131; CHECK-NEXT: [[TMP1:%.*]] = load [[S:%.*]], ptr [[SRC:%.*]], align 8132; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 8 [[SRC]], i8 0, i64 16, i1 false)133; CHECK-NEXT: call void @call() #[[ATTR2:[0-9]+]]134; CHECK-NEXT: store [[S]] [[TMP1]], ptr [[DST:%.*]], align 8135; CHECK-NEXT: ret void136;137 %1 = load %S, ptr %src138 store %S zeroinitializer, ptr %src139 call void @call() readnone140 store %S %1, ptr %dst141 ret void142}143 144define void @loop_memoryphi(ptr %a, ptr %b) {145; CHECK-LABEL: @loop_memoryphi(146; CHECK-NEXT: br label [[LOOP:%.*]]147; CHECK: loop:148; CHECK-NEXT: call void @llvm.memmove.p0.p0.i64(ptr align 8 [[B:%.*]], ptr align 8 [[A:%.*]], i64 16, i1 false)149; CHECK-NEXT: br label [[LOOP]]150;151 br label %loop152 153loop:154 %v = load { i64, i64 }, ptr %a155 store { i64, i64 } %v, ptr %b156 br label %loop157}158 159; There are multiple instructions that can clobber the source memory here.160; We can move the dest write past the store to %ptr.24, but not the memcpy.161; Make sure we don't perform fca2memcpy conversion in this case.162define void @multiple_clobbering(ptr %ptr, ptr %ptr.copy) {163; CHECK-LABEL: @multiple_clobbering(164; CHECK-NEXT: [[PTR_8:%.*]] = getelementptr inbounds nuw i8, ptr [[PTR:%.*]], i64 8165; CHECK-NEXT: [[PTR_24:%.*]] = getelementptr inbounds nuw i8, ptr [[PTR]], i64 24166; CHECK-NEXT: [[PTR_32:%.*]] = getelementptr inbounds nuw i8, ptr [[PTR]], i64 32167; CHECK-NEXT: [[PTR_COPY_8:%.*]] = getelementptr inbounds nuw i8, ptr [[PTR_COPY:%.*]], i64 8168; CHECK-NEXT: [[STRUCT:%.*]] = load { i32, i64 }, ptr [[PTR_COPY_8]], align 8169; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[PTR_8]], ptr [[PTR_32]], i64 12, i1 false)170; CHECK-NEXT: store i64 1, ptr [[PTR_24]], align 8171; CHECK-NEXT: store { i32, i64 } [[STRUCT]], ptr [[PTR_32]], align 8172; CHECK-NEXT: ret void173;174 %ptr.8 = getelementptr inbounds nuw i8, ptr %ptr, i64 8175 %ptr.24 = getelementptr inbounds nuw i8, ptr %ptr, i64 24176 %ptr.32 = getelementptr inbounds nuw i8, ptr %ptr, i64 32177 %ptr.copy.8 = getelementptr inbounds nuw i8, ptr %ptr.copy, i64 8178 %struct = load { i32, i64 }, ptr %ptr.copy.8, align 8179 call void @llvm.memcpy.p0.p0.i64(ptr %ptr.8, ptr %ptr.32, i64 12, i1 false)180 store i64 1, ptr %ptr.24, align 8181 store { i32, i64 } %struct, ptr %ptr.32, align 8182 ret void183}184 185declare void @call()186