1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -S -mtriple=x86_64-pc_linux -passes=loop-vectorize,instcombine < %s | FileCheck %s --check-prefix=SSE 3; RUN: opt -S -mtriple=x86_64-pc_linux -passes=loop-vectorize,instcombine -mcpu=sandybridge < %s | FileCheck %s --check-prefix=AVX1 4; RUN: opt -S -mtriple=x86_64-pc_linux -passes=loop-vectorize,instcombine -mcpu=haswell < %s | FileCheck %s --check-prefix=AVX2 5; RUN: opt -S -mtriple=x86_64-pc_linux -passes=loop-vectorize,instcombine -mcpu=slm < %s | FileCheck %s --check-prefix=SSE 6; RUN: opt -S -mtriple=x86_64-pc_linux -passes=loop-vectorize,instcombine -mcpu=atom < %s | FileCheck %s --check-prefix=ATOM 7 8define void @foo(ptr noalias nocapture %a, ptr noalias nocapture readonly %b) { 9; SSE-LABEL: @foo( 10; SSE-NEXT: entry: 11; SSE-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 12; SSE: vector.ph: 13; SSE-NEXT: br label [[VECTOR_BODY:%.*]] 14; SSE: vector.body: 15; SSE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 16; SSE-NEXT: [[TMP0:%.*]] = shl i64 [[INDEX]], 1 17; SSE-NEXT: [[TMP1:%.*]] = or disjoint i64 [[TMP0]], 8 18; SSE-NEXT: [[DOTIDX:%.*]] = shl nsw i64 [[INDEX]], 3 19; SSE-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[B:%.*]], i64 [[DOTIDX]] 20; SSE-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP1]] 21; SSE-NEXT: [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP2]], align 4 22; SSE-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6> 23; SSE-NEXT: [[STRIDED_VEC1:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7> 24; SSE-NEXT: [[WIDE_VEC2:%.*]] = load <8 x i32>, ptr [[TMP3]], align 4 25; SSE-NEXT: [[STRIDED_VEC3:%.*]] = shufflevector <8 x i32> [[WIDE_VEC2]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6> 26; SSE-NEXT: [[STRIDED_VEC4:%.*]] = shufflevector <8 x i32> [[WIDE_VEC2]], <8 x i32> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7> 27; SSE-NEXT: [[TMP4:%.*]] = add nsw <4 x i32> [[STRIDED_VEC1]], [[STRIDED_VEC]] 28; SSE-NEXT: [[TMP5:%.*]] = add nsw <4 x i32> [[STRIDED_VEC4]], [[STRIDED_VEC3]] 29; SSE-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]] 30; SSE-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP6]], i64 16 31; SSE-NEXT: store <4 x i32> [[TMP4]], ptr [[TMP6]], align 4 32; SSE-NEXT: store <4 x i32> [[TMP5]], ptr [[TMP7]], align 4 33; SSE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 34; SSE-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 35; SSE-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 36; SSE: middle.block: 37; SSE-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]] 38; SSE: scalar.ph: 39; SSE-NEXT: br label [[FOR_BODY:%.*]] 40; SSE: for.cond.cleanup: 41; SSE-NEXT: ret void 42; SSE: for.body: 43; SSE-NEXT: br i1 poison, label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 44; 45; AVX1-LABEL: @foo( 46; AVX1-NEXT: entry: 47; AVX1-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 48; AVX1: vector.ph: 49; AVX1-NEXT: br label [[VECTOR_BODY:%.*]] 50; AVX1: vector.body: 51; AVX1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 52; AVX1-NEXT: [[TMP0:%.*]] = shl i64 [[INDEX]], 1 53; AVX1-NEXT: [[TMP1:%.*]] = or disjoint i64 [[TMP0]], 8 54; AVX1-NEXT: [[TMP2:%.*]] = shl i64 [[INDEX]], 1 55; AVX1-NEXT: [[TMP3:%.*]] = or disjoint i64 [[TMP2]], 16 56; AVX1-NEXT: [[TMP4:%.*]] = shl i64 [[INDEX]], 1 57; AVX1-NEXT: [[TMP5:%.*]] = or disjoint i64 [[TMP4]], 24 58; AVX1-NEXT: [[DOTIDX:%.*]] = shl nsw i64 [[INDEX]], 3 59; AVX1-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[B:%.*]], i64 [[DOTIDX]] 60; AVX1-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP1]] 61; AVX1-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP3]] 62; AVX1-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP5]] 63; AVX1-NEXT: [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP6]], align 4 64; AVX1-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6> 65; AVX1-NEXT: [[STRIDED_VEC1:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7> 66; AVX1-NEXT: [[WIDE_VEC2:%.*]] = load <8 x i32>, ptr [[TMP7]], align 4 67; AVX1-NEXT: [[STRIDED_VEC3:%.*]] = shufflevector <8 x i32> [[WIDE_VEC2]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6> 68; AVX1-NEXT: [[STRIDED_VEC4:%.*]] = shufflevector <8 x i32> [[WIDE_VEC2]], <8 x i32> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7> 69; AVX1-NEXT: [[WIDE_VEC5:%.*]] = load <8 x i32>, ptr [[TMP8]], align 4 70; AVX1-NEXT: [[STRIDED_VEC6:%.*]] = shufflevector <8 x i32> [[WIDE_VEC5]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6> 71; AVX1-NEXT: [[STRIDED_VEC7:%.*]] = shufflevector <8 x i32> [[WIDE_VEC5]], <8 x i32> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7> 72; AVX1-NEXT: [[WIDE_VEC8:%.*]] = load <8 x i32>, ptr [[TMP9]], align 4 73; AVX1-NEXT: [[STRIDED_VEC9:%.*]] = shufflevector <8 x i32> [[WIDE_VEC8]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6> 74; AVX1-NEXT: [[STRIDED_VEC10:%.*]] = shufflevector <8 x i32> [[WIDE_VEC8]], <8 x i32> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7> 75; AVX1-NEXT: [[TMP10:%.*]] = add nsw <4 x i32> [[STRIDED_VEC1]], [[STRIDED_VEC]] 76; AVX1-NEXT: [[TMP11:%.*]] = add nsw <4 x i32> [[STRIDED_VEC4]], [[STRIDED_VEC3]] 77; AVX1-NEXT: [[TMP12:%.*]] = add nsw <4 x i32> [[STRIDED_VEC7]], [[STRIDED_VEC6]] 78; AVX1-NEXT: [[TMP13:%.*]] = add nsw <4 x i32> [[STRIDED_VEC10]], [[STRIDED_VEC9]] 79; AVX1-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]] 80; AVX1-NEXT: [[TMP15:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP14]], i64 16 81; AVX1-NEXT: [[TMP16:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP14]], i64 32 82; AVX1-NEXT: [[TMP17:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP14]], i64 48 83; AVX1-NEXT: store <4 x i32> [[TMP10]], ptr [[TMP14]], align 4 84; AVX1-NEXT: store <4 x i32> [[TMP11]], ptr [[TMP15]], align 4 85; AVX1-NEXT: store <4 x i32> [[TMP12]], ptr [[TMP16]], align 4 86; AVX1-NEXT: store <4 x i32> [[TMP13]], ptr [[TMP17]], align 4 87; AVX1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 88; AVX1-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 89; AVX1-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 90; AVX1: middle.block: 91; AVX1-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]] 92; AVX1: scalar.ph: 93; AVX1-NEXT: br label [[FOR_BODY:%.*]] 94; AVX1: for.cond.cleanup: 95; AVX1-NEXT: ret void 96; AVX1: for.body: 97; AVX1-NEXT: br i1 poison, label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 98; 99; AVX2-LABEL: @foo( 100; AVX2-NEXT: entry: 101; AVX2-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 102; AVX2: vector.ph: 103; AVX2-NEXT: br label [[VECTOR_BODY:%.*]] 104; AVX2: vector.body: 105; AVX2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 106; AVX2-NEXT: [[TMP0:%.*]] = shl i64 [[INDEX]], 1 107; AVX2-NEXT: [[TMP1:%.*]] = or disjoint i64 [[TMP0]], 16 108; AVX2-NEXT: [[TMP2:%.*]] = shl i64 [[INDEX]], 1 109; AVX2-NEXT: [[TMP3:%.*]] = or disjoint i64 [[TMP2]], 32 110; AVX2-NEXT: [[TMP4:%.*]] = shl i64 [[INDEX]], 1 111; AVX2-NEXT: [[TMP5:%.*]] = or disjoint i64 [[TMP4]], 48 112; AVX2-NEXT: [[DOTIDX:%.*]] = shl nsw i64 [[INDEX]], 3 113; AVX2-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[B:%.*]], i64 [[DOTIDX]] 114; AVX2-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP1]] 115; AVX2-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP3]] 116; AVX2-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP5]] 117; AVX2-NEXT: [[WIDE_VEC:%.*]] = load <16 x i32>, ptr [[TMP6]], align 4 118; AVX2-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <16 x i32> [[WIDE_VEC]], <16 x i32> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14> 119; AVX2-NEXT: [[STRIDED_VEC1:%.*]] = shufflevector <16 x i32> [[WIDE_VEC]], <16 x i32> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15> 120; AVX2-NEXT: [[WIDE_VEC2:%.*]] = load <16 x i32>, ptr [[TMP7]], align 4 121; AVX2-NEXT: [[STRIDED_VEC3:%.*]] = shufflevector <16 x i32> [[WIDE_VEC2]], <16 x i32> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14> 122; AVX2-NEXT: [[STRIDED_VEC4:%.*]] = shufflevector <16 x i32> [[WIDE_VEC2]], <16 x i32> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15> 123; AVX2-NEXT: [[WIDE_VEC5:%.*]] = load <16 x i32>, ptr [[TMP8]], align 4 124; AVX2-NEXT: [[STRIDED_VEC6:%.*]] = shufflevector <16 x i32> [[WIDE_VEC5]], <16 x i32> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14> 125; AVX2-NEXT: [[STRIDED_VEC7:%.*]] = shufflevector <16 x i32> [[WIDE_VEC5]], <16 x i32> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15> 126; AVX2-NEXT: [[WIDE_VEC8:%.*]] = load <16 x i32>, ptr [[TMP9]], align 4 127; AVX2-NEXT: [[STRIDED_VEC9:%.*]] = shufflevector <16 x i32> [[WIDE_VEC8]], <16 x i32> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14> 128; AVX2-NEXT: [[STRIDED_VEC10:%.*]] = shufflevector <16 x i32> [[WIDE_VEC8]], <16 x i32> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15> 129; AVX2-NEXT: [[TMP10:%.*]] = add nsw <8 x i32> [[STRIDED_VEC1]], [[STRIDED_VEC]] 130; AVX2-NEXT: [[TMP11:%.*]] = add nsw <8 x i32> [[STRIDED_VEC4]], [[STRIDED_VEC3]] 131; AVX2-NEXT: [[TMP12:%.*]] = add nsw <8 x i32> [[STRIDED_VEC7]], [[STRIDED_VEC6]] 132; AVX2-NEXT: [[TMP13:%.*]] = add nsw <8 x i32> [[STRIDED_VEC10]], [[STRIDED_VEC9]] 133; AVX2-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]] 134; AVX2-NEXT: [[TMP15:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP14]], i64 32 135; AVX2-NEXT: [[TMP16:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP14]], i64 64 136; AVX2-NEXT: [[TMP17:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP14]], i64 96 137; AVX2-NEXT: store <8 x i32> [[TMP10]], ptr [[TMP14]], align 4 138; AVX2-NEXT: store <8 x i32> [[TMP11]], ptr [[TMP15]], align 4 139; AVX2-NEXT: store <8 x i32> [[TMP12]], ptr [[TMP16]], align 4 140; AVX2-NEXT: store <8 x i32> [[TMP13]], ptr [[TMP17]], align 4 141; AVX2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32 142; AVX2-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 143; AVX2-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 144; AVX2: middle.block: 145; AVX2-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]] 146; AVX2: scalar.ph: 147; AVX2-NEXT: br label [[FOR_BODY:%.*]] 148; AVX2: for.cond.cleanup: 149; AVX2-NEXT: ret void 150; AVX2: for.body: 151; AVX2-NEXT: br i1 poison, label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 152; 153; ATOM-LABEL: @foo( 154; ATOM-NEXT: entry: 155; ATOM-NEXT: br label [[FOR_BODY:%.*]] 156; ATOM: for.cond.cleanup: 157; ATOM-NEXT: ret void 158; ATOM: for.body: 159; ATOM-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] 160; ATOM-NEXT: [[TMP0:%.*]] = shl nuw nsw i64 [[INDVARS_IV]], 1 161; ATOM-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[B:%.*]], i64 [[TMP0]] 162; ATOM-NEXT: [[TMP1:%.*]] = load i32, ptr [[ARRAYIDX]], align 4 163; ATOM-NEXT: [[TMP2:%.*]] = or disjoint i64 [[TMP0]], 1 164; ATOM-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds nuw i32, ptr [[B]], i64 [[TMP2]] 165; ATOM-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX3]], align 4 166; ATOM-NEXT: [[ADD4:%.*]] = add nsw i32 [[TMP3]], [[TMP1]] 167; ATOM-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds nuw i32, ptr [[A:%.*]], i64 [[INDVARS_IV]] 168; ATOM-NEXT: store i32 [[ADD4]], ptr [[ARRAYIDX6]], align 4 169; ATOM-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 170; ATOM-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 1024 171; ATOM-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY]] 172; 173entry: 174 br label %for.body 175 176for.cond.cleanup: ; preds = %for.body 177 ret void 178 179for.body: ; preds = %for.body, %entry 180 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 181 %0 = shl nsw i64 %indvars.iv, 1 182 %arrayidx = getelementptr inbounds i32, ptr %b, i64 %0 183 %1 = load i32, ptr %arrayidx, align 4 184 %2 = or disjoint i64 %0, 1 185 %arrayidx3 = getelementptr inbounds i32, ptr %b, i64 %2 186 %3 = load i32, ptr %arrayidx3, align 4 187 %add4 = add nsw i32 %3, %1 188 %arrayidx6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv 189 store i32 %add4, ptr %arrayidx6, align 4 190 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 191 %exitcond = icmp eq i64 %indvars.iv.next, 1024 192 br i1 %exitcond, label %for.cond.cleanup, label %for.body 193} 194