90 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^scalar.ph" --version 52; RUN: opt -S -passes=loop-vectorize -enable-vplan-native-path -verify-loop-info -verify-dom-info < %s | FileCheck %s3 4; extern int arr[8][8];5; extern int arr2[8];6;7; void foo(int n)8; {9; int i1, i2;10;11; #pragma clang loop vectorize(enable) vectorize_width(4)12; for (i1 = 0; i1 < 8; i1++) {13; arr2[i1] = i1;14; for (i2 = 0; i2 < 8; i2++)15; arr[i2][i1] = i1 + n;16; }17; }18;19 20@arr2 = external global [8 x i32], align 1621@arr = external global [8 x [8 x i32]], align 1622 23; Function Attrs: norecurse nounwind uwtable24define void @foo(i32 %n) {25; CHECK-LABEL: define void @foo(26; CHECK-SAME: i32 [[N:%.*]]) {27; CHECK-NEXT: [[ENTRY:.*:]]28; CHECK-NEXT: br label %[[VECTOR_PH:.*]]29; CHECK: [[VECTOR_PH]]:30; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[N]], i64 031; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer32; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]33; CHECK: [[VECTOR_BODY]]:34; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_LATCH:.*]] ]35; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_LATCH]] ]36; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds [8 x i32], ptr @arr2, i64 0, <4 x i64> [[VEC_IND]]37; CHECK-NEXT: [[TMP1:%.*]] = trunc <4 x i64> [[VEC_IND]] to <4 x i32>38; CHECK-NEXT: call void @llvm.masked.scatter.v4i32.v4p0(<4 x i32> [[TMP1]], <4 x ptr> align 4 [[TMP0]], <4 x i1> splat (i1 true))39; CHECK-NEXT: [[TMP2:%.*]] = add nsw <4 x i32> [[TMP1]], [[BROADCAST_SPLAT]]40; CHECK-NEXT: br label %[[FOR_BODY31:.*]]41; CHECK: [[FOR_BODY31]]:42; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_BODY]] ], [ [[TMP4:%.*]], %[[FOR_BODY31]] ]43; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds [8 x [8 x i32]], ptr @arr, i64 0, <4 x i64> [[VEC_PHI]], <4 x i64> [[VEC_IND]]44; CHECK-NEXT: call void @llvm.masked.scatter.v4i32.v4p0(<4 x i32> [[TMP2]], <4 x ptr> align 4 [[TMP3]], <4 x i1> splat (i1 true))45; CHECK-NEXT: [[TMP4]] = add nuw nsw <4 x i64> [[VEC_PHI]], splat (i64 1)46; CHECK-NEXT: [[TMP5:%.*]] = icmp eq <4 x i64> [[TMP4]], splat (i64 8)47; CHECK-NEXT: [[TMP6:%.*]] = extractelement <4 x i1> [[TMP5]], i32 048; CHECK-NEXT: br i1 [[TMP6]], label %[[VECTOR_LATCH]], label %[[FOR_BODY31]]49; CHECK: [[VECTOR_LATCH]]:50; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 451; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 4)52; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], 853; CHECK-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]54; CHECK: [[MIDDLE_BLOCK]]:55; CHECK-NEXT: br i1 true, [[FOR_END10:label %.*]], label %[[SCALAR_PH:.*]]56; CHECK: [[SCALAR_PH]]:57;58entry:59 br label %for.body60 61for.body: ; preds = %for.inc8, %entry62 %indvars.iv21 = phi i64 [ 0, %entry ], [ %indvars.iv.next22, %for.inc8 ]63 %arrayidx = getelementptr inbounds [8 x i32], ptr @arr2, i64 0, i64 %indvars.iv2164 %0 = trunc i64 %indvars.iv21 to i3265 store i32 %0, ptr %arrayidx, align 466 %1 = trunc i64 %indvars.iv21 to i3267 %add = add nsw i32 %1, %n68 br label %for.body369 70for.body3: ; preds = %for.body3, %for.body71 %indvars.iv = phi i64 [ 0, %for.body ], [ %indvars.iv.next, %for.body3 ]72 %arrayidx7 = getelementptr inbounds [8 x [8 x i32]], ptr @arr, i64 0, i64 %indvars.iv, i64 %indvars.iv2173 store i32 %add, ptr %arrayidx7, align 474 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 175 %exitcond = icmp eq i64 %indvars.iv.next, 876 br i1 %exitcond, label %for.inc8, label %for.body377 78for.inc8: ; preds = %for.body379 %indvars.iv.next22 = add nuw nsw i64 %indvars.iv21, 180 %exitcond23 = icmp eq i64 %indvars.iv.next22, 881 br i1 %exitcond23, label %for.end10, label %for.body, !llvm.loop !182 83for.end10: ; preds = %for.inc884 ret void85}86 87!1 = distinct !{!1, !2, !3}88!2 = !{!"llvm.loop.vectorize.width", i32 4}89!3 = !{!"llvm.loop.vectorize.enable", i1 true}90