1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -basic-aa -loop-distribute -enable-loop-distribute \ 3; RUN: -verify-loop-info -verify-dom-info -S < %s | FileCheck %s 4 5; Derived from crash-in-memcheck-generation.ll 6 7; Make sure the loop is distributed even with a convergent 8; op. LoopAccessAnalysis says that runtime checks are necessary, but 9; none are cross partition, so none are truly needed. 10 11define void @f(i32* %a, i32* %b, i32* noalias %c, i32* noalias %d, i32* noalias %e) #1 { 12; CHECK-LABEL: @f( 13; CHECK-NEXT: entry: 14; CHECK-NEXT: br label [[ENTRY_SPLIT_LDIST1:%.*]] 15; CHECK: entry.split.ldist1: 16; CHECK-NEXT: br label [[FOR_BODY_LDIST1:%.*]] 17; CHECK: for.body.ldist1: 18; CHECK-NEXT: [[IND_LDIST1:%.*]] = phi i64 [ 0, [[ENTRY_SPLIT_LDIST1]] ], [ [[ADD_LDIST1:%.*]], [[FOR_BODY_LDIST1]] ] 19; CHECK-NEXT: [[ARRAYIDXA_LDIST1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[IND_LDIST1]] 20; CHECK-NEXT: [[LOADA_LDIST1:%.*]] = load i32, i32* [[ARRAYIDXA_LDIST1]], align 4 21; CHECK-NEXT: [[ARRAYIDXB_LDIST1:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[IND_LDIST1]] 22; CHECK-NEXT: [[LOADB_LDIST1:%.*]] = load i32, i32* [[ARRAYIDXB_LDIST1]], align 4 23; CHECK-NEXT: [[MULA_LDIST1:%.*]] = mul i32 [[LOADB_LDIST1]], [[LOADA_LDIST1]] 24; CHECK-NEXT: [[ADD_LDIST1]] = add nuw nsw i64 [[IND_LDIST1]], 1 25; CHECK-NEXT: [[ARRAYIDXA_PLUS_4_LDIST1:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[ADD_LDIST1]] 26; CHECK-NEXT: store i32 [[MULA_LDIST1]], i32* [[ARRAYIDXA_PLUS_4_LDIST1]], align 4 27; CHECK-NEXT: [[EXITCOND_LDIST1:%.*]] = icmp eq i64 [[ADD_LDIST1]], 20 28; CHECK-NEXT: br i1 [[EXITCOND_LDIST1]], label [[ENTRY_SPLIT:%.*]], label [[FOR_BODY_LDIST1]] 29; CHECK: entry.split: 30; CHECK-NEXT: br label [[FOR_BODY:%.*]] 31; CHECK: for.body: 32; CHECK-NEXT: [[IND:%.*]] = phi i64 [ 0, [[ENTRY_SPLIT]] ], [ [[ADD:%.*]], [[FOR_BODY]] ] 33; CHECK-NEXT: [[ADD]] = add nuw nsw i64 [[IND]], 1 34; CHECK-NEXT: [[ARRAYIDXD:%.*]] = getelementptr inbounds i32, i32* [[D:%.*]], i64 [[IND]] 35; CHECK-NEXT: [[LOADD:%.*]] = load i32, i32* [[ARRAYIDXD]], align 4 36; CHECK-NEXT: [[CONVERGENTD:%.*]] = call i32 @llvm.convergent(i32 [[LOADD]]) 37; CHECK-NEXT: [[ARRAYIDXE:%.*]] = getelementptr inbounds i32, i32* [[E:%.*]], i64 [[IND]] 38; CHECK-NEXT: [[LOADE:%.*]] = load i32, i32* [[ARRAYIDXE]], align 4 39; CHECK-NEXT: [[MULC:%.*]] = mul i32 [[CONVERGENTD]], [[LOADE]] 40; CHECK-NEXT: [[ARRAYIDXC:%.*]] = getelementptr inbounds i32, i32* [[C:%.*]], i64 [[IND]] 41; CHECK-NEXT: store i32 [[MULC]], i32* [[ARRAYIDXC]], align 4 42; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[ADD]], 20 43; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]] 44; CHECK: for.end: 45; CHECK-NEXT: ret void 46; 47entry: 48 br label %for.body 49 50for.body: ; preds = %for.body, %entry 51 %ind = phi i64 [ 0, %entry ], [ %add, %for.body ] 52 53 %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind 54 %loadA = load i32, i32* %arrayidxA, align 4 55 56 %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind 57 %loadB = load i32, i32* %arrayidxB, align 4 58 59 %mulA = mul i32 %loadB, %loadA 60 61 %add = add nuw nsw i64 %ind, 1 62 %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add 63 store i32 %mulA, i32* %arrayidxA_plus_4, align 4 64 65 %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %ind 66 %loadD = load i32, i32* %arrayidxD, align 4 67 %convergentD = call i32 @llvm.convergent(i32 %loadD) 68 69 %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %ind 70 %loadE = load i32, i32* %arrayidxE, align 4 71 72 %mulC = mul i32 %convergentD, %loadE 73 74 %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind 75 store i32 %mulC, i32* %arrayidxC, align 4 76 77 %exitcond = icmp eq i64 %add, 20 78 br i1 %exitcond, label %for.end, label %for.body 79 80for.end: ; preds = %for.body 81 ret void 82} 83 84declare i32 @llvm.convergent(i32) #0 85 86attributes #0 = { nounwind readnone convergent } 87attributes #1 = { nounwind convergent } 88