208 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-vectorize < %s -S -o - | FileCheck %s3 4target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-n32:64-S128"5target triple = "aarch64-unknown-linux-gnu"6 7define void @firstorderrec(ptr nocapture noundef readonly %x, ptr noalias nocapture noundef writeonly %y, i32 noundef %n) {8; CHECK-LABEL: @firstorderrec(9; CHECK-NEXT: entry:10; CHECK-NEXT: [[CMP18:%.*]] = icmp sgt i32 [[N:%.*]], 111; CHECK-NEXT: br i1 [[CMP18]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]12; CHECK: for.body.preheader:13; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i6414; CHECK-NEXT: [[DOTPRE:%.*]] = load i8, ptr [[X:%.*]], align 115; CHECK-NEXT: [[TMP0:%.*]] = add nsw i64 [[WIDE_TRIP_COUNT]], -116; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 3217; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]18; CHECK: vector.ph:19; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 3220; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]21; CHECK-NEXT: [[IND_END:%.*]] = add i64 1, [[N_VEC]]22; CHECK-NEXT: [[VECTOR_RECUR_INIT:%.*]] = insertelement <16 x i8> poison, i8 [[DOTPRE]], i32 1523; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]24; CHECK: vector.body:25; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]26; CHECK-NEXT: [[VECTOR_RECUR:%.*]] = phi <16 x i8> [ [[VECTOR_RECUR_INIT]], [[VECTOR_PH]] ], [ [[WIDE_LOAD1:%.*]], [[VECTOR_BODY]] ]27; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 1, [[INDEX]]28; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[X]], i64 [[OFFSET_IDX]]29; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[TMP3]], i64 1630; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP3]], align 131; CHECK-NEXT: [[WIDE_LOAD1]] = load <16 x i8>, ptr [[TMP6]], align 132; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <16 x i8> [[VECTOR_RECUR]], <16 x i8> [[WIDE_LOAD]], <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>33; CHECK-NEXT: [[TMP8:%.*]] = shufflevector <16 x i8> [[WIDE_LOAD]], <16 x i8> [[WIDE_LOAD1]], <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>34; CHECK-NEXT: [[TMP9:%.*]] = add <16 x i8> [[WIDE_LOAD]], [[TMP7]]35; CHECK-NEXT: [[TMP10:%.*]] = add <16 x i8> [[WIDE_LOAD1]], [[TMP8]]36; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[Y:%.*]], i64 [[OFFSET_IDX]]37; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[TMP11]], i64 1638; CHECK-NEXT: store <16 x i8> [[TMP9]], ptr [[TMP11]], align 139; CHECK-NEXT: store <16 x i8> [[TMP10]], ptr [[TMP14]], align 140; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 3241; CHECK-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]42; CHECK-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]43; CHECK: middle.block:44; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <16 x i8> [[WIDE_LOAD1]], i32 1545; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]46; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]47; CHECK: scalar.ph:48; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i8 [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ], [ [[DOTPRE]], [[FOR_BODY_PREHEADER]] ]49; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 1, [[FOR_BODY_PREHEADER]] ]50; CHECK-NEXT: br label [[FOR_BODY:%.*]]51; CHECK: for.cond.cleanup.loopexit:52; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]53; CHECK: for.cond.cleanup:54; CHECK-NEXT: ret void55; CHECK: for.body:56; CHECK-NEXT: [[SCALAR_RECUR:%.*]] = phi i8 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[TMP16:%.*]], [[FOR_BODY]] ]57; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]58; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i8, ptr [[X]], i64 [[INDVARS_IV]]59; CHECK-NEXT: [[TMP16]] = load i8, ptr [[ARRAYIDX4]], align 160; CHECK-NEXT: [[ADD7:%.*]] = add i8 [[TMP16]], [[SCALAR_RECUR]]61; CHECK-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds i8, ptr [[Y]], i64 [[INDVARS_IV]]62; CHECK-NEXT: store i8 [[ADD7]], ptr [[ARRAYIDX10]], align 163; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 164; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]65; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]66;67entry:68 %cmp18 = icmp sgt i32 %n, 169 br i1 %cmp18, label %for.body.preheader, label %for.cond.cleanup70 71for.body.preheader: ; preds = %entry72 %wide.trip.count = zext i32 %n to i6473 %.pre = load i8, ptr %x, align 174 br label %for.body75 76for.cond.cleanup: ; preds = %for.body, %entry77 ret void78 79for.body: ; preds = %for.body.preheader, %for.body80 %0 = phi i8 [ %.pre, %for.body.preheader ], [ %1, %for.body ]81 %indvars.iv = phi i64 [ 1, %for.body.preheader ], [ %indvars.iv.next, %for.body ]82 %arrayidx4 = getelementptr inbounds i8, ptr %x, i64 %indvars.iv83 %1 = load i8, ptr %arrayidx4, align 184 %add7 = add i8 %1, %085 %arrayidx10 = getelementptr inbounds i8, ptr %y, i64 %indvars.iv86 store i8 %add7, ptr %arrayidx10, align 187 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 188 %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count89 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body90}91 92define void @thirdorderrec(ptr nocapture noundef readonly %x, ptr noalias nocapture noundef writeonly %y, i32 noundef %n) #0 {93; CHECK-LABEL: @thirdorderrec(94; CHECK-NEXT: entry:95; CHECK-NEXT: [[CMP38:%.*]] = icmp sgt i32 [[N:%.*]], 396; CHECK-NEXT: br i1 [[CMP38]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]97; CHECK: for.body.preheader:98; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i6499; CHECK-NEXT: [[DOTPRE:%.*]] = load i8, ptr [[X:%.*]], align 1100; CHECK-NEXT: [[ARRAYIDX5_PHI_TRANS_INSERT:%.*]] = getelementptr inbounds i8, ptr [[X]], i64 1101; CHECK-NEXT: [[DOTPRE44:%.*]] = load i8, ptr [[ARRAYIDX5_PHI_TRANS_INSERT]], align 1102; CHECK-NEXT: [[ARRAYIDX12_PHI_TRANS_INSERT:%.*]] = getelementptr inbounds i8, ptr [[X]], i64 2103; CHECK-NEXT: [[DOTPRE45:%.*]] = load i8, ptr [[ARRAYIDX12_PHI_TRANS_INSERT]], align 1104; CHECK-NEXT: [[TMP0:%.*]] = add nsw i64 [[WIDE_TRIP_COUNT]], -3105; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 32106; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]107; CHECK: vector.ph:108; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 32109; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]110; CHECK-NEXT: [[IND_END:%.*]] = add i64 3, [[N_VEC]]111; CHECK-NEXT: [[VECTOR_RECUR_INIT:%.*]] = insertelement <16 x i8> poison, i8 [[DOTPRE45]], i32 15112; CHECK-NEXT: [[VECTOR_RECUR_INIT1:%.*]] = insertelement <16 x i8> poison, i8 [[DOTPRE44]], i32 15113; CHECK-NEXT: [[VECTOR_RECUR_INIT3:%.*]] = insertelement <16 x i8> poison, i8 [[DOTPRE]], i32 15114; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]115; CHECK: vector.body:116; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]117; CHECK-NEXT: [[VECTOR_RECUR:%.*]] = phi <16 x i8> [ [[VECTOR_RECUR_INIT]], [[VECTOR_PH]] ], [ [[WIDE_LOAD5:%.*]], [[VECTOR_BODY]] ]118; CHECK-NEXT: [[VECTOR_RECUR2:%.*]] = phi <16 x i8> [ [[VECTOR_RECUR_INIT1]], [[VECTOR_PH]] ], [ [[TMP8:%.*]], [[VECTOR_BODY]] ]119; CHECK-NEXT: [[VECTOR_RECUR4:%.*]] = phi <16 x i8> [ [[VECTOR_RECUR_INIT3]], [[VECTOR_PH]] ], [ [[TMP10:%.*]], [[VECTOR_BODY]] ]120; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 3, [[INDEX]]121; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[X]], i64 [[OFFSET_IDX]]122; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[TMP3]], i64 16123; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1124; CHECK-NEXT: [[WIDE_LOAD5]] = load <16 x i8>, ptr [[TMP6]], align 1125; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <16 x i8> [[VECTOR_RECUR]], <16 x i8> [[WIDE_LOAD]], <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>126; CHECK-NEXT: [[TMP8]] = shufflevector <16 x i8> [[WIDE_LOAD]], <16 x i8> [[WIDE_LOAD5]], <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>127; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <16 x i8> [[VECTOR_RECUR2]], <16 x i8> [[TMP7]], <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>128; CHECK-NEXT: [[TMP10]] = shufflevector <16 x i8> [[TMP7]], <16 x i8> [[TMP8]], <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>129; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <16 x i8> [[VECTOR_RECUR4]], <16 x i8> [[TMP9]], <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>130; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <16 x i8> [[TMP9]], <16 x i8> [[TMP10]], <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>131; CHECK-NEXT: [[TMP13:%.*]] = add <16 x i8> [[TMP9]], [[TMP11]]132; CHECK-NEXT: [[TMP14:%.*]] = add <16 x i8> [[TMP10]], [[TMP12]]133; CHECK-NEXT: [[TMP15:%.*]] = add <16 x i8> [[TMP13]], [[TMP7]]134; CHECK-NEXT: [[TMP16:%.*]] = add <16 x i8> [[TMP14]], [[TMP8]]135; CHECK-NEXT: [[TMP17:%.*]] = add <16 x i8> [[TMP15]], [[WIDE_LOAD]]136; CHECK-NEXT: [[TMP18:%.*]] = add <16 x i8> [[TMP16]], [[WIDE_LOAD5]]137; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i8, ptr [[Y:%.*]], i64 [[OFFSET_IDX]]138; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds i8, ptr [[TMP19]], i64 16139; CHECK-NEXT: store <16 x i8> [[TMP17]], ptr [[TMP19]], align 1140; CHECK-NEXT: store <16 x i8> [[TMP18]], ptr [[TMP22]], align 1141; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32142; CHECK-NEXT: [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]143; CHECK-NEXT: br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]144; CHECK: middle.block:145; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <16 x i8> [[WIDE_LOAD5]], i32 15146; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT6:%.*]] = extractelement <16 x i8> [[TMP8]], i32 15147; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT9:%.*]] = extractelement <16 x i8> [[TMP10]], i32 15148; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]149; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]150; CHECK: scalar.ph:151; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i8 [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ], [ [[DOTPRE45]], [[FOR_BODY_PREHEADER]] ]152; CHECK-NEXT: [[SCALAR_RECUR_INIT7:%.*]] = phi i8 [ [[VECTOR_RECUR_EXTRACT6]], [[MIDDLE_BLOCK]] ], [ [[DOTPRE44]], [[FOR_BODY_PREHEADER]] ]153; CHECK-NEXT: [[SCALAR_RECUR_INIT10:%.*]] = phi i8 [ [[VECTOR_RECUR_EXTRACT9]], [[MIDDLE_BLOCK]] ], [ [[DOTPRE]], [[FOR_BODY_PREHEADER]] ]154; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 3, [[FOR_BODY_PREHEADER]] ]155; CHECK-NEXT: br label [[FOR_BODY:%.*]]156; CHECK: for.cond.cleanup.loopexit:157; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]158; CHECK: for.cond.cleanup:159; CHECK-NEXT: ret void160; CHECK: for.body:161; CHECK-NEXT: [[SCALAR_RECUR:%.*]] = phi i8 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[TMP24:%.*]], [[FOR_BODY]] ]162; CHECK-NEXT: [[SCALAR_RECUR8:%.*]] = phi i8 [ [[SCALAR_RECUR_INIT7]], [[SCALAR_PH]] ], [ [[SCALAR_RECUR]], [[FOR_BODY]] ]163; CHECK-NEXT: [[SCALAR_RECUR11:%.*]] = phi i8 [ [[SCALAR_RECUR_INIT10]], [[SCALAR_PH]] ], [ [[SCALAR_RECUR8]], [[FOR_BODY]] ]164; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]165; CHECK-NEXT: [[ADD8:%.*]] = add i8 [[SCALAR_RECUR8]], [[SCALAR_RECUR11]]166; CHECK-NEXT: [[ADD15:%.*]] = add i8 [[ADD8]], [[SCALAR_RECUR]]167; CHECK-NEXT: [[ARRAYIDX18:%.*]] = getelementptr inbounds i8, ptr [[X]], i64 [[INDVARS_IV]]168; CHECK-NEXT: [[TMP24]] = load i8, ptr [[ARRAYIDX18]], align 1169; CHECK-NEXT: [[ADD21:%.*]] = add i8 [[ADD15]], [[TMP24]]170; CHECK-NEXT: [[ARRAYIDX24:%.*]] = getelementptr inbounds i8, ptr [[Y]], i64 [[INDVARS_IV]]171; CHECK-NEXT: store i8 [[ADD21]], ptr [[ARRAYIDX24]], align 1172; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1173; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]174; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]175;176entry:177 %cmp38 = icmp sgt i32 %n, 3178 br i1 %cmp38, label %for.body.preheader, label %for.cond.cleanup179 180for.body.preheader: ; preds = %entry181 %wide.trip.count = zext i32 %n to i64182 %.pre = load i8, ptr %x, align 1183 %arrayidx5.phi.trans.insert = getelementptr inbounds i8, ptr %x, i64 1184 %.pre44 = load i8, ptr %arrayidx5.phi.trans.insert, align 1185 %arrayidx12.phi.trans.insert = getelementptr inbounds i8, ptr %x, i64 2186 %.pre45 = load i8, ptr %arrayidx12.phi.trans.insert, align 1187 br label %for.body188 189for.cond.cleanup: ; preds = %for.body, %entry190 ret void191 192for.body: ; preds = %for.body.preheader, %for.body193 %0 = phi i8 [ %.pre45, %for.body.preheader ], [ %3, %for.body ]194 %1 = phi i8 [ %.pre44, %for.body.preheader ], [ %0, %for.body ]195 %2 = phi i8 [ %.pre, %for.body.preheader ], [ %1, %for.body ]196 %indvars.iv = phi i64 [ 3, %for.body.preheader ], [ %indvars.iv.next, %for.body ]197 %add8 = add i8 %1, %2198 %add15 = add i8 %add8, %0199 %arrayidx18 = getelementptr inbounds i8, ptr %x, i64 %indvars.iv200 %3 = load i8, ptr %arrayidx18, align 1201 %add21 = add i8 %add15, %3202 %arrayidx24 = getelementptr inbounds i8, ptr %y, i64 %indvars.iv203 store i8 %add21, ptr %arrayidx24, align 1204 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1205 %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count206 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body207}208