182 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -S -mtriple=x86_64-pc_linux -passes=loop-vectorize,instcombine < %s | FileCheck %s --check-prefix=SSE3; RUN: opt -S -mtriple=x86_64-pc_linux -passes=loop-vectorize,instcombine -mcpu=sandybridge < %s | FileCheck %s --check-prefix=AVX14; RUN: opt -S -mtriple=x86_64-pc_linux -passes=loop-vectorize,instcombine -mcpu=haswell < %s | FileCheck %s --check-prefix=AVX25; RUN: opt -S -mtriple=x86_64-pc_linux -passes=loop-vectorize,instcombine -mcpu=slm < %s | FileCheck %s --check-prefix=SSE6; RUN: opt -S -mtriple=x86_64-pc_linux -passes=loop-vectorize,instcombine -mcpu=atom < %s | FileCheck %s --check-prefix=ATOM7 8define void @foo(ptr noalias nocapture %a, ptr noalias nocapture readonly %b) {9; SSE-LABEL: @foo(10; SSE-NEXT: entry:11; SSE-NEXT: br 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: [[DOTIDX:%.*]] = shl nsw i64 [[INDEX]], 317; SSE-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[B:%.*]], i64 [[DOTIDX]]18; SSE-NEXT: [[DOTIDX5:%.*]] = shl i64 [[INDEX]], 319; SSE-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[DOTIDX5]]20; SSE-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP1]], i64 3221; SSE-NEXT: [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP2]], align 422; 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 425; 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 1631; SSE-NEXT: store <4 x i32> [[TMP4]], ptr [[TMP6]], align 432; SSE-NEXT: store <4 x i32> [[TMP5]], ptr [[TMP7]], align 433; SSE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 834; SSE-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 102435; SSE-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]36; SSE: middle.block:37; SSE-NEXT: br label [[FOR_BODY:%.*]]38; SSE: for.cond.cleanup:39; SSE-NEXT: ret void40;41; AVX1-LABEL: @foo(42; AVX1-NEXT: entry:43; AVX1-NEXT: br label [[VECTOR_PH:%.*]]44; AVX1: vector.ph:45; AVX1-NEXT: br label [[VECTOR_BODY:%.*]]46; AVX1: vector.body:47; AVX1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]48; AVX1-NEXT: [[DOTIDX:%.*]] = shl nsw i64 [[INDEX]], 349; AVX1-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[B:%.*]], i64 [[DOTIDX]]50; AVX1-NEXT: [[DOTIDX11:%.*]] = shl i64 [[INDEX]], 351; AVX1-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[DOTIDX11]]52; AVX1-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP1]], i64 3253; AVX1-NEXT: [[DOTIDX12:%.*]] = shl i64 [[INDEX]], 354; AVX1-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[B]], i64 [[DOTIDX12]]55; AVX1-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[TMP3]], i64 6456; AVX1-NEXT: [[DOTIDX13:%.*]] = shl i64 [[INDEX]], 357; AVX1-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[DOTIDX13]]58; AVX1-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[TMP5]], i64 9659; AVX1-NEXT: [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP6]], align 460; 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>61; 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>62; AVX1-NEXT: [[WIDE_VEC2:%.*]] = load <8 x i32>, ptr [[TMP7]], align 463; 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>64; 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>65; AVX1-NEXT: [[WIDE_VEC5:%.*]] = load <8 x i32>, ptr [[TMP8]], align 466; 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>67; 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>68; AVX1-NEXT: [[WIDE_VEC8:%.*]] = load <8 x i32>, ptr [[TMP9]], align 469; 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>70; 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>71; AVX1-NEXT: [[TMP10:%.*]] = add nsw <4 x i32> [[STRIDED_VEC1]], [[STRIDED_VEC]]72; AVX1-NEXT: [[TMP11:%.*]] = add nsw <4 x i32> [[STRIDED_VEC4]], [[STRIDED_VEC3]]73; AVX1-NEXT: [[TMP12:%.*]] = add nsw <4 x i32> [[STRIDED_VEC7]], [[STRIDED_VEC6]]74; AVX1-NEXT: [[TMP13:%.*]] = add nsw <4 x i32> [[STRIDED_VEC10]], [[STRIDED_VEC9]]75; AVX1-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]]76; AVX1-NEXT: [[TMP15:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP14]], i64 1677; AVX1-NEXT: [[TMP16:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP14]], i64 3278; AVX1-NEXT: [[TMP17:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP14]], i64 4879; AVX1-NEXT: store <4 x i32> [[TMP10]], ptr [[TMP14]], align 480; AVX1-NEXT: store <4 x i32> [[TMP11]], ptr [[TMP15]], align 481; AVX1-NEXT: store <4 x i32> [[TMP12]], ptr [[TMP16]], align 482; AVX1-NEXT: store <4 x i32> [[TMP13]], ptr [[TMP17]], align 483; AVX1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 1684; AVX1-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], 102485; AVX1-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]86; AVX1: middle.block:87; AVX1-NEXT: br label [[FOR_BODY:%.*]]88; AVX1: for.cond.cleanup:89; AVX1-NEXT: ret void90;91; AVX2-LABEL: @foo(92; AVX2-NEXT: entry:93; AVX2-NEXT: br label [[VECTOR_PH:%.*]]94; AVX2: vector.ph:95; AVX2-NEXT: br label [[VECTOR_BODY:%.*]]96; AVX2: vector.body:97; AVX2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]98; AVX2-NEXT: [[DOTIDX:%.*]] = shl nsw i64 [[INDEX]], 399; AVX2-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[B:%.*]], i64 [[DOTIDX]]100; AVX2-NEXT: [[DOTIDX11:%.*]] = shl i64 [[INDEX]], 3101; AVX2-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[DOTIDX11]]102; AVX2-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP1]], i64 64103; AVX2-NEXT: [[DOTIDX12:%.*]] = shl i64 [[INDEX]], 3104; AVX2-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[B]], i64 [[DOTIDX12]]105; AVX2-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[TMP3]], i64 128106; AVX2-NEXT: [[DOTIDX13:%.*]] = shl i64 [[INDEX]], 3107; AVX2-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[DOTIDX13]]108; AVX2-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[TMP5]], i64 192109; AVX2-NEXT: [[WIDE_VEC:%.*]] = load <16 x i32>, ptr [[TMP6]], align 4110; 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>111; 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>112; AVX2-NEXT: [[WIDE_VEC2:%.*]] = load <16 x i32>, ptr [[TMP7]], align 4113; 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>114; 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>115; AVX2-NEXT: [[WIDE_VEC5:%.*]] = load <16 x i32>, ptr [[TMP8]], align 4116; 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>117; 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>118; AVX2-NEXT: [[WIDE_VEC8:%.*]] = load <16 x i32>, ptr [[TMP9]], align 4119; 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>120; 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>121; AVX2-NEXT: [[TMP10:%.*]] = add nsw <8 x i32> [[STRIDED_VEC1]], [[STRIDED_VEC]]122; AVX2-NEXT: [[TMP11:%.*]] = add nsw <8 x i32> [[STRIDED_VEC4]], [[STRIDED_VEC3]]123; AVX2-NEXT: [[TMP12:%.*]] = add nsw <8 x i32> [[STRIDED_VEC7]], [[STRIDED_VEC6]]124; AVX2-NEXT: [[TMP13:%.*]] = add nsw <8 x i32> [[STRIDED_VEC10]], [[STRIDED_VEC9]]125; AVX2-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]]126; AVX2-NEXT: [[TMP15:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP14]], i64 32127; AVX2-NEXT: [[TMP16:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP14]], i64 64128; AVX2-NEXT: [[TMP17:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP14]], i64 96129; AVX2-NEXT: store <8 x i32> [[TMP10]], ptr [[TMP14]], align 4130; AVX2-NEXT: store <8 x i32> [[TMP11]], ptr [[TMP15]], align 4131; AVX2-NEXT: store <8 x i32> [[TMP12]], ptr [[TMP16]], align 4132; AVX2-NEXT: store <8 x i32> [[TMP13]], ptr [[TMP17]], align 4133; AVX2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32134; AVX2-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024135; AVX2-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]136; AVX2: middle.block:137; AVX2-NEXT: br label [[FOR_BODY:%.*]]138; AVX2: for.cond.cleanup:139; AVX2-NEXT: ret void140;141; ATOM-LABEL: @foo(142; ATOM-NEXT: entry:143; ATOM-NEXT: br label [[FOR_BODY:%.*]]144; ATOM: for.cond.cleanup:145; ATOM-NEXT: ret void146; ATOM: for.body:147; ATOM-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]148; ATOM-NEXT: [[TMP0:%.*]] = shl nuw nsw i64 [[INDVARS_IV]], 1149; ATOM-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[B:%.*]], i64 [[TMP0]]150; ATOM-NEXT: [[TMP1:%.*]] = load i32, ptr [[ARRAYIDX]], align 4151; ATOM-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw i32, ptr [[B]], i64 [[TMP0]]152; ATOM-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP2]], i64 4153; ATOM-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX3]], align 4154; ATOM-NEXT: [[ADD4:%.*]] = add nsw i32 [[TMP3]], [[TMP1]]155; ATOM-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds nuw i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]156; ATOM-NEXT: store i32 [[ADD4]], ptr [[ARRAYIDX6]], align 4157; ATOM-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1158; ATOM-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 1024159; ATOM-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY]]160;161entry:162 br label %for.body163 164for.cond.cleanup: ; preds = %for.body165 ret void166 167for.body: ; preds = %for.body, %entry168 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]169 %0 = shl nsw i64 %indvars.iv, 1170 %arrayidx = getelementptr inbounds i32, ptr %b, i64 %0171 %1 = load i32, ptr %arrayidx, align 4172 %2 = or disjoint i64 %0, 1173 %arrayidx3 = getelementptr inbounds i32, ptr %b, i64 %2174 %3 = load i32, ptr %arrayidx3, align 4175 %add4 = add nsw i32 %3, %1176 %arrayidx6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv177 store i32 %add4, ptr %arrayidx6, align 4178 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1179 %exitcond = icmp eq i64 %indvars.iv.next, 1024180 br i1 %exitcond, label %for.cond.cleanup, label %for.body181}182