514 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -S < %s -passes=loop-vectorize -force-vector-interleave=4 2>&1 | FileCheck %s3 4target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"5target triple = "aarch64"6 7define void @add_i8(ptr noalias nocapture noundef writeonly %A, ptr nocapture noundef readonly %B, ptr nocapture noundef readonly %C, i64 noundef %Iterations) {8; CHECK-LABEL: @add_i8(9; CHECK-NEXT: iter.check:10; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[ITERATIONS:%.*]], 811; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]12; CHECK: vector.main.loop.iter.check:13; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[ITERATIONS]], 6414; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]15; CHECK: vector.ph:16; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[ITERATIONS]], 6417; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[ITERATIONS]], [[N_MOD_VF]]18; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]19; CHECK: vector.body:20; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]21; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[B:%.*]], i64 [[INDEX]]22; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i64 1623; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i64 3224; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i64 4825; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 126; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 127; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP4]], align 128; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP5]], align 129; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[C:%.*]], i64 [[INDEX]]30; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i8, ptr [[TMP6]], i64 1631; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i8, ptr [[TMP6]], i64 3232; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[TMP6]], i64 4833; CHECK-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP6]], align 134; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP8]], align 135; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <16 x i8>, ptr [[TMP9]], align 136; CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <16 x i8>, ptr [[TMP10]], align 137; CHECK-NEXT: [[TMP11:%.*]] = add <16 x i8> [[WIDE_LOAD5]], [[WIDE_LOAD]]38; CHECK-NEXT: [[TMP12:%.*]] = add <16 x i8> [[WIDE_LOAD6]], [[WIDE_LOAD2]]39; CHECK-NEXT: [[TMP13:%.*]] = add <16 x i8> [[WIDE_LOAD7]], [[WIDE_LOAD3]]40; CHECK-NEXT: [[TMP14:%.*]] = add <16 x i8> [[WIDE_LOAD8]], [[WIDE_LOAD4]]41; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i8, ptr [[A:%.*]], i64 [[INDEX]]42; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds i8, ptr [[TMP15]], i64 1643; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i8, ptr [[TMP15]], i64 3244; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i8, ptr [[TMP15]], i64 4845; CHECK-NEXT: store <16 x i8> [[TMP11]], ptr [[TMP15]], align 146; CHECK-NEXT: store <16 x i8> [[TMP12]], ptr [[TMP17]], align 147; CHECK-NEXT: store <16 x i8> [[TMP13]], ptr [[TMP18]], align 148; CHECK-NEXT: store <16 x i8> [[TMP14]], ptr [[TMP19]], align 149; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 6450; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]51; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]52; CHECK: middle.block:53; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[ITERATIONS]], [[N_VEC]]54; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]55; CHECK: vec.epilog.iter.check:56; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 857; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]]58; CHECK: vec.epilog.ph:59; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]60; CHECK-NEXT: [[N_MOD_VF9:%.*]] = urem i64 [[ITERATIONS]], 861; CHECK-NEXT: [[N_VEC10:%.*]] = sub i64 [[ITERATIONS]], [[N_MOD_VF9]]62; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]63; CHECK: vec.epilog.vector.body:64; CHECK-NEXT: [[INDEX11:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT14:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]65; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDEX11]]66; CHECK-NEXT: [[WIDE_LOAD12:%.*]] = load <8 x i8>, ptr [[TMP22]], align 167; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds i8, ptr [[C]], i64 [[INDEX11]]68; CHECK-NEXT: [[WIDE_LOAD13:%.*]] = load <8 x i8>, ptr [[TMP24]], align 169; CHECK-NEXT: [[TMP26:%.*]] = add <8 x i8> [[WIDE_LOAD13]], [[WIDE_LOAD12]]70; CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX11]]71; CHECK-NEXT: store <8 x i8> [[TMP26]], ptr [[TMP27]], align 172; CHECK-NEXT: [[INDEX_NEXT14]] = add nuw i64 [[INDEX11]], 873; CHECK-NEXT: [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT14]], [[N_VEC10]]74; CHECK-NEXT: br i1 [[TMP29]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]75; CHECK: vec.epilog.middle.block:76; CHECK-NEXT: [[CMP_N15:%.*]] = icmp eq i64 [[ITERATIONS]], [[N_VEC10]]77; CHECK-NEXT: br i1 [[CMP_N15]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]78; CHECK: vec.epilog.scalar.ph:79; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC10]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]80; CHECK-NEXT: br label [[FOR_BODY:%.*]]81; CHECK: for.body:82; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]83; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[IV]]84; CHECK-NEXT: [[TMP30:%.*]] = load i8, ptr [[ARRAYIDX]], align 185; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i8, ptr [[C]], i64 [[IV]]86; CHECK-NEXT: [[TMP31:%.*]] = load i8, ptr [[ARRAYIDX2]], align 187; CHECK-NEXT: [[ADD:%.*]] = add i8 [[TMP31]], [[TMP30]]88; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[IV]]89; CHECK-NEXT: store i8 [[ADD]], ptr [[ARRAYIDX6]], align 190; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 191; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[ITERATIONS]]92; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]93; CHECK: exit:94; CHECK-NEXT: ret void95;96entry:97 br label %for.body98 99for.body:100 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]101 %arrayidx = getelementptr inbounds i8, ptr %B, i64 %iv102 %0 = load i8, ptr %arrayidx, align 1103 %arrayidx2 = getelementptr inbounds i8, ptr %C, i64 %iv104 %1 = load i8, ptr %arrayidx2, align 1105 %add = add i8 %1, %0106 %arrayidx6 = getelementptr inbounds i8, ptr %A, i64 %iv107 store i8 %add, ptr %arrayidx6, align 1108 %iv.next = add nuw nsw i64 %iv, 1109 %exitcond.not = icmp eq i64 %iv.next, %Iterations110 br i1 %exitcond.not, label %exit, label %for.body111 112exit:113 ret void114}115 116define void @add_i16(ptr noalias nocapture noundef writeonly %A, ptr nocapture noundef readonly %B, ptr nocapture noundef readonly %C, i64 noundef %Iterations) {117; CHECK-LABEL: @add_i16(118; CHECK-NEXT: iter.check:119; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[ITERATIONS:%.*]], 4120; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]121; CHECK: vector.main.loop.iter.check:122; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[ITERATIONS]], 32123; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]124; CHECK: vector.ph:125; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[ITERATIONS]], 32126; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[ITERATIONS]], [[N_MOD_VF]]127; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]128; CHECK: vector.body:129; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]130; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i16, ptr [[B:%.*]], i64 [[INDEX]]131; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i16, ptr [[TMP1]], i64 8132; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i16, ptr [[TMP1]], i64 16133; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i16, ptr [[TMP1]], i64 24134; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[TMP1]], align 1135; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <8 x i16>, ptr [[TMP3]], align 1136; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <8 x i16>, ptr [[TMP4]], align 1137; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <8 x i16>, ptr [[TMP5]], align 1138; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i16, ptr [[C:%.*]], i64 [[INDEX]]139; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i16, ptr [[TMP6]], i64 8140; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i16, ptr [[TMP6]], i64 16141; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i16, ptr [[TMP6]], i64 24142; CHECK-NEXT: [[WIDE_LOAD5:%.*]] = load <8 x i16>, ptr [[TMP6]], align 1143; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <8 x i16>, ptr [[TMP8]], align 1144; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <8 x i16>, ptr [[TMP9]], align 1145; CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <8 x i16>, ptr [[TMP10]], align 1146; CHECK-NEXT: [[TMP11:%.*]] = add <8 x i16> [[WIDE_LOAD5]], [[WIDE_LOAD]]147; CHECK-NEXT: [[TMP12:%.*]] = add <8 x i16> [[WIDE_LOAD6]], [[WIDE_LOAD2]]148; CHECK-NEXT: [[TMP13:%.*]] = add <8 x i16> [[WIDE_LOAD7]], [[WIDE_LOAD3]]149; CHECK-NEXT: [[TMP14:%.*]] = add <8 x i16> [[WIDE_LOAD8]], [[WIDE_LOAD4]]150; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i16, ptr [[A:%.*]], i64 [[INDEX]]151; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds i16, ptr [[TMP15]], i64 8152; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i16, ptr [[TMP15]], i64 16153; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i16, ptr [[TMP15]], i64 24154; CHECK-NEXT: store <8 x i16> [[TMP11]], ptr [[TMP15]], align 1155; CHECK-NEXT: store <8 x i16> [[TMP12]], ptr [[TMP17]], align 1156; CHECK-NEXT: store <8 x i16> [[TMP13]], ptr [[TMP18]], align 1157; CHECK-NEXT: store <8 x i16> [[TMP14]], ptr [[TMP19]], align 1158; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32159; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]160; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]161; CHECK: middle.block:162; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[ITERATIONS]], [[N_VEC]]163; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]164; CHECK: vec.epilog.iter.check:165; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 4166; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF7:![0-9]+]]167; CHECK: vec.epilog.ph:168; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]169; CHECK-NEXT: [[N_MOD_VF9:%.*]] = urem i64 [[ITERATIONS]], 4170; CHECK-NEXT: [[N_VEC10:%.*]] = sub i64 [[ITERATIONS]], [[N_MOD_VF9]]171; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]172; CHECK: vec.epilog.vector.body:173; CHECK-NEXT: [[INDEX11:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT14:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]174; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds i16, ptr [[B]], i64 [[INDEX11]]175; CHECK-NEXT: [[WIDE_LOAD12:%.*]] = load <4 x i16>, ptr [[TMP22]], align 1176; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds i16, ptr [[C]], i64 [[INDEX11]]177; CHECK-NEXT: [[WIDE_LOAD13:%.*]] = load <4 x i16>, ptr [[TMP24]], align 1178; CHECK-NEXT: [[TMP26:%.*]] = add <4 x i16> [[WIDE_LOAD13]], [[WIDE_LOAD12]]179; CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds i16, ptr [[A]], i64 [[INDEX11]]180; CHECK-NEXT: store <4 x i16> [[TMP26]], ptr [[TMP27]], align 1181; CHECK-NEXT: [[INDEX_NEXT14]] = add nuw i64 [[INDEX11]], 4182; CHECK-NEXT: [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT14]], [[N_VEC10]]183; CHECK-NEXT: br i1 [[TMP29]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]184; CHECK: vec.epilog.middle.block:185; CHECK-NEXT: [[CMP_N15:%.*]] = icmp eq i64 [[ITERATIONS]], [[N_VEC10]]186; CHECK-NEXT: br i1 [[CMP_N15]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]187; CHECK: vec.epilog.scalar.ph:188; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC10]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]189; CHECK-NEXT: br label [[FOR_BODY:%.*]]190; CHECK: for.body:191; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]192; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[B]], i64 [[IV]]193; CHECK-NEXT: [[TMP30:%.*]] = load i16, ptr [[ARRAYIDX]], align 1194; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[C]], i64 [[IV]]195; CHECK-NEXT: [[TMP31:%.*]] = load i16, ptr [[ARRAYIDX2]], align 1196; CHECK-NEXT: [[ADD:%.*]] = add i16 [[TMP31]], [[TMP30]]197; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds i16, ptr [[A]], i64 [[IV]]198; CHECK-NEXT: store i16 [[ADD]], ptr [[ARRAYIDX6]], align 1199; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1200; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[ITERATIONS]]201; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]202; CHECK: exit:203; CHECK-NEXT: ret void204;205entry:206 br label %for.body207 208for.body:209 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]210 %arrayidx = getelementptr inbounds i16, ptr %B, i64 %iv211 %0 = load i16, ptr %arrayidx, align 1212 %arrayidx2 = getelementptr inbounds i16, ptr %C, i64 %iv213 %1 = load i16, ptr %arrayidx2, align 1214 %add = add i16 %1, %0215 %arrayidx6 = getelementptr inbounds i16, ptr %A, i64 %iv216 store i16 %add, ptr %arrayidx6, align 1217 %iv.next = add nuw nsw i64 %iv, 1218 %exitcond.not = icmp eq i64 %iv.next, %Iterations219 br i1 %exitcond.not, label %exit, label %for.body220 221exit:222 ret void223}224 225define void @add_i32(ptr noalias nocapture noundef writeonly %A, ptr nocapture noundef readonly %B, ptr nocapture noundef readonly %C, i64 noundef %Iterations) {226; CHECK-LABEL: @add_i32(227; CHECK-NEXT: iter.check:228; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[ITERATIONS:%.*]], 4229; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]230; CHECK: vector.main.loop.iter.check:231; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[ITERATIONS]], 16232; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]233; CHECK: vector.ph:234; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[ITERATIONS]], 16235; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[ITERATIONS]], [[N_MOD_VF]]236; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]237; CHECK: vector.body:238; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]239; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[INDEX]]240; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 4241; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 8242; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 12243; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP1]], align 1244; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <4 x i32>, ptr [[TMP3]], align 1245; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <4 x i32>, ptr [[TMP4]], align 1246; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <4 x i32>, ptr [[TMP5]], align 1247; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[C:%.*]], i64 [[INDEX]]248; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i64 4249; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i64 8250; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i64 12251; CHECK-NEXT: [[WIDE_LOAD5:%.*]] = load <4 x i32>, ptr [[TMP6]], align 1252; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <4 x i32>, ptr [[TMP8]], align 1253; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <4 x i32>, ptr [[TMP9]], align 1254; CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <4 x i32>, ptr [[TMP10]], align 1255; CHECK-NEXT: [[TMP11:%.*]] = add <4 x i32> [[WIDE_LOAD5]], [[WIDE_LOAD]]256; CHECK-NEXT: [[TMP12:%.*]] = add <4 x i32> [[WIDE_LOAD6]], [[WIDE_LOAD2]]257; CHECK-NEXT: [[TMP13:%.*]] = add <4 x i32> [[WIDE_LOAD7]], [[WIDE_LOAD3]]258; CHECK-NEXT: [[TMP14:%.*]] = add <4 x i32> [[WIDE_LOAD8]], [[WIDE_LOAD4]]259; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]]260; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds i32, ptr [[TMP15]], i64 4261; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i32, ptr [[TMP15]], i64 8262; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i32, ptr [[TMP15]], i64 12263; CHECK-NEXT: store <4 x i32> [[TMP11]], ptr [[TMP15]], align 1264; CHECK-NEXT: store <4 x i32> [[TMP12]], ptr [[TMP17]], align 1265; CHECK-NEXT: store <4 x i32> [[TMP13]], ptr [[TMP18]], align 1266; CHECK-NEXT: store <4 x i32> [[TMP14]], ptr [[TMP19]], align 1267; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16268; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]269; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]270; CHECK: middle.block:271; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[ITERATIONS]], [[N_VEC]]272; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]273; CHECK: vec.epilog.iter.check:274; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 4275; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF11:![0-9]+]]276; CHECK: vec.epilog.ph:277; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]278; CHECK-NEXT: [[N_MOD_VF9:%.*]] = urem i64 [[ITERATIONS]], 4279; CHECK-NEXT: [[N_VEC10:%.*]] = sub i64 [[ITERATIONS]], [[N_MOD_VF9]]280; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]281; CHECK: vec.epilog.vector.body:282; CHECK-NEXT: [[INDEX11:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT14:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]283; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDEX11]]284; CHECK-NEXT: [[WIDE_LOAD12:%.*]] = load <4 x i32>, ptr [[TMP22]], align 1285; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[INDEX11]]286; CHECK-NEXT: [[WIDE_LOAD13:%.*]] = load <4 x i32>, ptr [[TMP24]], align 1287; CHECK-NEXT: [[TMP26:%.*]] = add <4 x i32> [[WIDE_LOAD13]], [[WIDE_LOAD12]]288; CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDEX11]]289; CHECK-NEXT: store <4 x i32> [[TMP26]], ptr [[TMP27]], align 1290; CHECK-NEXT: [[INDEX_NEXT14]] = add nuw i64 [[INDEX11]], 4291; CHECK-NEXT: [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT14]], [[N_VEC10]]292; CHECK-NEXT: br i1 [[TMP29]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]293; CHECK: vec.epilog.middle.block:294; CHECK-NEXT: [[CMP_N15:%.*]] = icmp eq i64 [[ITERATIONS]], [[N_VEC10]]295; CHECK-NEXT: br i1 [[CMP_N15]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]296; CHECK: vec.epilog.scalar.ph:297; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC10]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]298; CHECK-NEXT: br label [[FOR_BODY:%.*]]299; CHECK: for.body:300; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]301; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IV]]302; CHECK-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARRAYIDX]], align 1303; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[IV]]304; CHECK-NEXT: [[TMP31:%.*]] = load i32, ptr [[ARRAYIDX2]], align 1305; CHECK-NEXT: [[ADD:%.*]] = add i32 [[TMP31]], [[TMP30]]306; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV]]307; CHECK-NEXT: store i32 [[ADD]], ptr [[ARRAYIDX6]], align 1308; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1309; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[ITERATIONS]]310; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]311; CHECK: exit:312; CHECK-NEXT: ret void313;314entry:315 br label %for.body316 317for.body:318 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]319 %arrayidx = getelementptr inbounds i32, ptr %B, i64 %iv320 %0 = load i32, ptr %arrayidx, align 1321 %arrayidx2 = getelementptr inbounds i32, ptr %C, i64 %iv322 %1 = load i32, ptr %arrayidx2, align 1323 %add = add i32 %1, %0324 %arrayidx6 = getelementptr inbounds i32, ptr %A, i64 %iv325 store i32 %add, ptr %arrayidx6, align 1326 %iv.next = add nuw nsw i64 %iv, 1327 %exitcond.not = icmp eq i64 %iv.next, %Iterations328 br i1 %exitcond.not, label %exit, label %for.body329 330exit:331 ret void332}333 334; TODO: Choose smaller VF * UF for main loop, so we do not create a dead vector loop.335define void @small_trip_count_loop(ptr %arg, ptr %arg2) {336; CHECK-LABEL: @small_trip_count_loop(337; CHECK-NEXT: iter.check:338; CHECK-NEXT: [[ARG3:%.*]] = ptrtoint ptr [[ARG:%.*]] to i64339; CHECK-NEXT: [[ARG21:%.*]] = ptrtoint ptr [[ARG2:%.*]] to i64340; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]341; CHECK: vector.memcheck:342; CHECK-NEXT: [[TMP0:%.*]] = sub i64 [[ARG21]], [[ARG3]]343; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP0]], 64344; CHECK-NEXT: br i1 [[DIFF_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]345; CHECK: vector.main.loop.iter.check:346; CHECK-NEXT: br i1 true, label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]347; CHECK: vector.ph:348; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]349; CHECK: vector.body:350; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[ARG]], i64 16351; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[ARG]], i64 32352; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[ARG]], i64 48353; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[ARG]], align 1354; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1355; CHECK-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1356; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1357; CHECK-NEXT: [[TMP4:%.*]] = add <16 x i8> [[WIDE_LOAD]], splat (i8 10)358; CHECK-NEXT: [[TMP5:%.*]] = add <16 x i8> [[WIDE_LOAD4]], splat (i8 10)359; CHECK-NEXT: [[TMP6:%.*]] = add <16 x i8> [[WIDE_LOAD5]], splat (i8 10)360; CHECK-NEXT: [[TMP7:%.*]] = add <16 x i8> [[WIDE_LOAD6]], splat (i8 10)361; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i8, ptr [[ARG2]], i64 16362; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i8, ptr [[ARG2]], i64 32363; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[ARG2]], i64 48364; CHECK-NEXT: store <16 x i8> [[TMP4]], ptr [[ARG2]], align 1365; CHECK-NEXT: store <16 x i8> [[TMP5]], ptr [[TMP8]], align 1366; CHECK-NEXT: store <16 x i8> [[TMP6]], ptr [[TMP9]], align 1367; CHECK-NEXT: store <16 x i8> [[TMP7]], ptr [[TMP10]], align 1368; CHECK-NEXT: br label [[MIDDLE_BLOCK:%.*]]369; CHECK: middle.block:370; CHECK-NEXT: br i1 false, label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]371; CHECK: vec.epilog.iter.check:372; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF14:![0-9]+]]373; CHECK: vec.epilog.ph:374; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i32 [ 0, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]375; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]376; CHECK: vec.epilog.vector.body:377; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]378; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[ARG]], i32 [[INDEX]]379; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <16 x i8>, ptr [[TMP11]], align 1380; CHECK-NEXT: [[TMP12:%.*]] = add <16 x i8> [[WIDE_LOAD7]], splat (i8 10)381; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[ARG2]], i32 [[INDEX]]382; CHECK-NEXT: store <16 x i8> [[TMP12]], ptr [[TMP13]], align 1383; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 16384; CHECK-NEXT: [[TMP14:%.*]] = icmp eq i32 [[INDEX_NEXT]], 16385; CHECK-NEXT: br i1 [[TMP14]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]386; CHECK: vec.epilog.middle.block:387; CHECK-NEXT: br i1 false, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]388; CHECK: vec.epilog.scalar.ph:389; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 16, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 0, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]390; CHECK-NEXT: br label [[LOOP:%.*]]391; CHECK: loop:392; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]393; CHECK-NEXT: [[GEP_A:%.*]] = getelementptr inbounds i8, ptr [[ARG]], i32 [[IV]]394; CHECK-NEXT: [[LOAD:%.*]] = load i8, ptr [[GEP_A]], align 1395; CHECK-NEXT: [[SELECT:%.*]] = add i8 [[LOAD]], 10396; CHECK-NEXT: [[GEP_B:%.*]] = getelementptr inbounds i8, ptr [[ARG2]], i32 [[IV]]397; CHECK-NEXT: store i8 [[SELECT]], ptr [[GEP_B]], align 1398; CHECK-NEXT: [[IV_NEXT]] = add i32 [[IV]], 1399; CHECK-NEXT: [[EC:%.*]] = icmp eq i32 [[IV]], 20400; CHECK-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP16:![0-9]+]]401; CHECK: exit:402; CHECK-NEXT: ret void403;404entry:405 br label %loop406 407loop:408 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]409 %gep.A = getelementptr inbounds i8, ptr %arg, i32 %iv410 %load = load i8, ptr %gep.A, align 1411 %select = add i8 %load, 10412 %gep.B = getelementptr inbounds i8, ptr %arg2, i32 %iv413 store i8 %select, ptr %gep.B, align 1414 %iv.next = add i32 %iv, 1415 %ec = icmp eq i32 %iv, 20416 br i1 %ec, label %exit, label %loop417 418exit:419 ret void420}421 422; TODO: Should not try to create dead main vector loop.423define void @trip_count_based_on_ptrtoint(i64 %x) "target-cpu"="apple-m1" {424; CHECK-LABEL: @trip_count_based_on_ptrtoint(425; CHECK-NEXT: iter.check:426; CHECK-NEXT: [[PTR_START:%.*]] = inttoptr i64 [[X:%.*]] to ptr427; CHECK-NEXT: [[ADD:%.*]] = add i64 [[X]], 40428; CHECK-NEXT: [[PTR_END:%.*]] = inttoptr i64 [[ADD]] to ptr429; CHECK-NEXT: [[TMP0:%.*]] = sub i64 [[ADD]], [[X]]430; CHECK-NEXT: [[TMP1:%.*]] = lshr i64 [[TMP0]], 2431; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1432; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]433; CHECK: vector.scevcheck:434; CHECK-NEXT: [[TMP3:%.*]] = trunc i64 [[ADD]] to i2435; CHECK-NEXT: [[TMP4:%.*]] = trunc i64 [[X]] to i2436; CHECK-NEXT: [[TMP5:%.*]] = sub i2 [[TMP3]], [[TMP4]]437; CHECK-NEXT: [[TMP6:%.*]] = zext i2 [[TMP5]] to i64438; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[TMP6]], 0439; CHECK-NEXT: br i1 [[IDENT_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]440; CHECK: vector.main.loop.iter.check:441; CHECK-NEXT: br i1 true, label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]442; CHECK: vector.ph:443; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 16444; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]445; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]446; CHECK: vector.body:447; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]448; CHECK-NEXT: [[TMP7:%.*]] = mul i64 [[INDEX]], 4449; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PTR_START]], i64 [[TMP7]]450; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i32, ptr [[NEXT_GEP]], i64 4451; CHECK-NEXT: [[TMP9:%.*]] = getelementptr i32, ptr [[NEXT_GEP]], i64 8452; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i32, ptr [[NEXT_GEP]], i64 12453; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[NEXT_GEP]], align 4454; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[TMP8]], align 4455; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[TMP9]], align 4456; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[TMP10]], align 4457; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16458; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]459; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP17:![0-9]+]]460; CHECK: middle.block:461; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]462; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]463; CHECK: vec.epilog.iter.check:464; CHECK-NEXT: [[TMP12:%.*]] = mul i64 [[N_VEC]], 4465; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[PTR_START]], i64 [[TMP12]]466; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 4467; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF11]]468; CHECK: vec.epilog.ph:469; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]470; CHECK-NEXT: [[N_MOD_VF1:%.*]] = urem i64 [[TMP2]], 4471; CHECK-NEXT: [[N_VEC2:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF1]]472; CHECK-NEXT: [[TMP13:%.*]] = mul i64 [[N_VEC2]], 4473; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[PTR_START]], i64 [[TMP13]]474; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]475; CHECK: vec.epilog.vector.body:476; CHECK-NEXT: [[INDEX3:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT5:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]477; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX3]], 4478; CHECK-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[PTR_START]], i64 [[OFFSET_IDX]]479; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[NEXT_GEP4]], align 4480; CHECK-NEXT: [[INDEX_NEXT5]] = add nuw i64 [[INDEX3]], 4481; CHECK-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT5]], [[N_VEC2]]482; CHECK-NEXT: br i1 [[TMP15]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]]483; CHECK: vec.epilog.middle.block:484; CHECK-NEXT: [[CMP_N6:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC2]]485; CHECK-NEXT: br i1 [[CMP_N6]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]486; CHECK: vec.epilog.scalar.ph:487; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[TMP14]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PTR_START]], [[VECTOR_SCEVCHECK]] ], [ [[PTR_START]], [[ITER_CHECK:%.*]] ]488; CHECK-NEXT: br label [[LOOP:%.*]]489; CHECK: loop:490; CHECK-NEXT: [[IV:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]491; CHECK-NEXT: store i32 0, ptr [[IV]], align 4492; CHECK-NEXT: [[IV_NEXT]] = getelementptr i8, ptr [[IV]], i64 4493; CHECK-NEXT: [[EC:%.*]] = icmp eq ptr [[IV]], [[PTR_END]]494; CHECK-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP19:![0-9]+]]495; CHECK: exit:496; CHECK-NEXT: ret void497;498entry:499 %ptr.start = inttoptr i64 %x to ptr500 %add = add i64 %x, 40501 %ptr.end = inttoptr i64 %add to ptr502 br label %loop503 504loop:505 %iv = phi ptr [ %ptr.start, %entry ], [ %iv.next, %loop ]506 store i32 0, ptr %iv, align 4507 %iv.next = getelementptr i8, ptr %iv, i64 4508 %ec = icmp eq ptr %iv, %ptr.end509 br i1 %ec, label %exit, label %loop510 511exit:512 ret void513}514