1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 2; RUN: opt -passes=loop-distribute -enable-loop-distribute -S -verify-loop-info -verify-dom-info < %s \ 3; RUN: | FileCheck %s 4 5; Distributing this loop to avoid the dependence cycle would require to 6; reorder S1 and S2 to form the two partitions: {S2} | {S1, S3}. The analysis 7; provided by LoopAccessAnalysis does not allow us to reorder memory 8; operations so make sure we bail on this loop. 9; 10; for (i = 0; i < n; i++) { 11; S1: d = D[i]; 12; S2: A[i + 1] = A[i] * B[i]; 13; S3: C[i] = d * E[i]; 14; } 15 16define void @f(ptr noalias %a, ptr noalias %b, ptr noalias %c, ptr noalias %d, ptr noalias %e) { 17; CHECK-LABEL: define void @f( 18; CHECK-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], ptr noalias [[C:%.*]], ptr noalias [[D:%.*]], ptr noalias [[E:%.*]]) { 19; CHECK-NEXT: [[ENTRY:.*]]: 20; CHECK-NEXT: br label %[[FOR_BODY:.*]] 21; CHECK: [[FOR_BODY]]: 22; CHECK-NEXT: [[IND:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[ADD:%.*]], %[[FOR_BODY]] ] 23; CHECK-NEXT: [[ARRAYIDXA:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IND]] 24; CHECK-NEXT: [[LOADA:%.*]] = load i32, ptr [[ARRAYIDXA]], align 4 25; CHECK-NEXT: [[ARRAYIDXB:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IND]] 26; CHECK-NEXT: [[LOADB:%.*]] = load i32, ptr [[ARRAYIDXB]], align 4 27; CHECK-NEXT: [[MULA:%.*]] = mul i32 [[LOADB]], [[LOADA]] 28; CHECK-NEXT: [[ARRAYIDXD:%.*]] = getelementptr inbounds i32, ptr [[D]], i64 [[IND]] 29; CHECK-NEXT: [[LOADD:%.*]] = load i32, ptr [[ARRAYIDXD]], align 4 30; CHECK-NEXT: [[ADD]] = add nuw nsw i64 [[IND]], 1 31; CHECK-NEXT: [[ARRAYIDXA_PLUS_4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[ADD]] 32; CHECK-NEXT: store i32 [[MULA]], ptr [[ARRAYIDXA_PLUS_4]], align 4 33; CHECK-NEXT: [[ARRAYIDXC:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[IND]] 34; CHECK-NEXT: [[ARRAYIDXE:%.*]] = getelementptr inbounds i32, ptr [[E]], i64 [[IND]] 35; CHECK-NEXT: [[LOADE:%.*]] = load i32, ptr [[ARRAYIDXE]], align 4 36; CHECK-NEXT: [[MULC:%.*]] = mul i32 [[LOADD]], [[LOADE]] 37; CHECK-NEXT: store i32 [[MULC]], ptr [[ARRAYIDXC]], align 4 38; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[ADD]], 20 39; CHECK-NEXT: br i1 [[EXITCOND]], label %[[FOR_END:.*]], label %[[FOR_BODY]] 40; CHECK: [[FOR_END]]: 41; CHECK-NEXT: ret void 42; 43entry: 44 br label %for.body 45 46for.body: ; preds = %for.body, %entry 47 %ind = phi i64 [ 0, %entry ], [ %add, %for.body ] 48 49 %arrayidxA = getelementptr inbounds i32, ptr %a, i64 %ind 50 %loadA = load i32, ptr %arrayidxA, align 4 51 52 %arrayidxB = getelementptr inbounds i32, ptr %b, i64 %ind 53 %loadB = load i32, ptr %arrayidxB, align 4 54 55 %mulA = mul i32 %loadB, %loadA 56 57 %arrayidxD = getelementptr inbounds i32, ptr %d, i64 %ind 58 %loadD = load i32, ptr %arrayidxD, align 4 59 60 %add = add nuw nsw i64 %ind, 1 61 %arrayidxA_plus_4 = getelementptr inbounds i32, ptr %a, i64 %add 62 store i32 %mulA, ptr %arrayidxA_plus_4, align 4 63 64 %arrayidxC = getelementptr inbounds i32, ptr %c, i64 %ind 65 66 %arrayidxE = getelementptr inbounds i32, ptr %e, i64 %ind 67 %loadE = load i32, ptr %arrayidxE, align 4 68 69 %mulC = mul i32 %loadD, %loadE 70 71 store i32 %mulC, ptr %arrayidxC, align 4 72 73 %exitcond = icmp eq i64 %add, 20 74 br i1 %exitcond, label %for.end, label %for.body 75 76for.end: ; preds = %for.body 77 ret void 78} 79