Index: include/llvm/Analysis/AliasSetTracker.h =================================================================== --- include/llvm/Analysis/AliasSetTracker.h +++ include/llvm/Analysis/AliasSetTracker.h @@ -69,10 +69,21 @@ if (AAInfo == DenseMapInfo::getEmptyKey()) // We don't have a AAInfo yet. Set it to NewAAInfo. AAInfo = NewAAInfo; - else if (AAInfo != NewAAInfo) - // NewAAInfo conflicts with AAInfo. - AAInfo = DenseMapInfo::getTombstoneKey(); - + else { + if (AAInfo.TBAA != NewAAInfo.TBAA) { + // NewAAInfo conflicts with AAInfo. + AAInfo = DenseMapInfo::getTombstoneKey(); + return SizeChanged; + } + // Could intersect for these two, but that might insert new MDNodes, + // which is probably not worth it. + if (AAInfo.Scope != NewAAInfo.Scope) { + AAInfo.Scope = nullptr; + } + if (AAInfo.NoAlias != NewAAInfo.NoAlias) { + AAInfo.NoAlias = nullptr; + } + } return SizeChanged; } Index: test/Transforms/LICM/dropped-tbaa.ll =================================================================== --- /dev/null +++ test/Transforms/LICM/dropped-tbaa.ll @@ -0,0 +1,90 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py +; RUN: opt < %s -scoped-noalias -tbaa -licm -S | FileCheck %s + +; This test case case is generated from the following C code with -fstrict-aliasing, +; and after passing through -inline -mem2reg -loop-rotate -instcombine +; void add(double *restrict data, int *restrict addend) { +; *data += *addend; +; } +; +; void foo(double *data, int *addend) { +; for (int i = 0; i < 1000; ++i) { +; *data += *addend; +; add(data, addend); +; } +; } +; We want to make sure the load of addend gets hoisted, independent of the second load +; load having different noalias metadata. + +define void @foo(double* %data, i32* %addend) #0 { +; CHECK: for.body.lr.ph: +; CHECK-NEXT: [[TMP1:%.*]] = load i32, i32* [[ADDEND:%.*]], align 4, !tbaa !1 +; CHECK-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP1]] to double +; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* [[ADDEND]], align 4, !tbaa !1, !alias.scope !5, !noalias !8 +; CHECK-NEXT: [[CONV_I:%.*]] = sitofp i32 [[TMP2]] to double +entry: + %i = alloca i32, align 4 + %0 = bitcast i32* %i to i8* + call void @llvm.lifetime.start.p0i8(i64 4, i8* %0) #2 + store i32 0, i32* %i, align 4, !tbaa !1 + br i1 true, label %for.body.lr.ph, label %for.cond.cleanup + +for.body.lr.ph: ; preds = %entry + br label %for.body + +for.cond.for.cond.cleanup_crit_edge: ; preds = %for.inc + br label %for.cond.cleanup + +for.cond.cleanup: ; preds = %for.cond.for.cond.cleanup_crit_edge, %entry + %1 = bitcast i32* %i to i8* + call void @llvm.lifetime.end.p0i8(i64 4, i8* %1) #2 + br label %for.end + +for.body: ; preds = %for.body.lr.ph, %for.inc + %2 = load i32, i32* %addend, align 4, !tbaa !1 + %conv = sitofp i32 %2 to double + %3 = load i32, i32* %i, align 4, !tbaa !1 + %idxprom = sext i32 %3 to i64 + %arrayidx = getelementptr inbounds double, double* %data, i64 %idxprom + %4 = load double, double* %arrayidx, align 8, !tbaa !5 + %add = fadd double %4, %conv + store double %add, double* %arrayidx, align 8, !tbaa !5 + %idxprom1 = sext i32 %3 to i64 + %arrayidx2 = getelementptr inbounds double, double* %data, i64 %idxprom1 + %5 = load i32, i32* %addend, align 4, !tbaa !1, !alias.scope !7, !noalias !10 + %conv.i = sitofp i32 %5 to double + %6 = load double, double* %arrayidx2, align 8, !tbaa !5, !alias.scope !10, !noalias !7 + %add.i = fadd double %6, %conv.i + store double %add.i, double* %arrayidx2, align 8, !tbaa !5, !alias.scope !10, !noalias !7 + br label %for.inc + +for.inc: ; preds = %for.body + %7 = load i32, i32* %i, align 4, !tbaa !1 + %inc = add nsw i32 %7, 1 + store i32 %inc, i32* %i, align 4, !tbaa !1 + %cmp = icmp slt i32 %inc, 1000 + br i1 %cmp, label %for.body, label %for.cond.for.cond.cleanup_crit_edge + +for.end: ; preds = %for.cond.cleanup + ret void +} + +declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #0 +declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #0 + +attributes #0 = { argmemonly nounwind } + +!llvm.ident = !{!0} + +!0 = !{!"clang version 5.0.0 (llvm/trunk 299971)"} +!1 = !{!2, !2, i64 0} +!2 = !{!"int", !3, i64 0} +!3 = !{!"omnipotent char", !4, i64 0} +!4 = !{!"Simple C/C++ TBAA"} +!5 = !{!6, !6, i64 0} +!6 = !{!"double", !3, i64 0} +!7 = !{!8} +!8 = distinct !{!8, !9, !"add: %addend"} +!9 = distinct !{!9, !"add"} +!10 = !{!11} +!11 = distinct !{!11, !9, !"add: %data"}