From 424b7771711d8be54f8ac65a24d1870d931c18ee Mon Sep 17 00:00:00 2001 From: Dan Gohman Date: Sat, 21 May 2011 01:05:08 +0000 Subject: [PATCH] When BasicAA can determine that two pointers have the same base but differ by a dynamic offset, return PartialAlias instead of MayAlias. See the comment in the code for details. This fixes PR9971. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131781 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Analysis/BasicAliasAnalysis.cpp | 12 +- .../2010-09-15-GEP-SignedArithmetic.ll | 2 +- .../TypeBasedAliasAnalysis/dynamic-indices.ll | 131 ++++++++++++++++++ 3 files changed, 143 insertions(+), 2 deletions(-) create mode 100644 test/Analysis/TypeBasedAliasAnalysis/dynamic-indices.ll diff --git a/lib/Analysis/BasicAliasAnalysis.cpp b/lib/Analysis/BasicAliasAnalysis.cpp index f1bb8a38f09..f50660844fe 100644 --- a/lib/Analysis/BasicAliasAnalysis.cpp +++ b/lib/Analysis/BasicAliasAnalysis.cpp @@ -940,7 +940,17 @@ BasicAliasAnalysis::aliasGEP(const GEPOperator *GEP1, uint64_t V1Size, return NoAlias; } - return MayAlias; + // Statically, we can see that the base objects are the same, but the + // pointers have dynamic offsets which we can't resolve. And none of our + // little tricks above worked. + // + // TODO: Returning PartialAlias instead of MayAlias is a mild hack; the + // practical effect of this is protecting TBAA in the case of dynamic + // indices into arrays of unions. An alternative way to solve this would + // be to have clang emit extra metadata for unions and/or union accesses. + // A union-specific solution wouldn't handle the problem for malloc'd + // memory however. + return PartialAlias; } /// aliasSelect - Provide a bunch of ad-hoc rules to disambiguate a Select diff --git a/test/Analysis/BasicAA/2010-09-15-GEP-SignedArithmetic.ll b/test/Analysis/BasicAA/2010-09-15-GEP-SignedArithmetic.ll index 2b0cd78fece..7b5584e0bd8 100644 --- a/test/Analysis/BasicAA/2010-09-15-GEP-SignedArithmetic.ll +++ b/test/Analysis/BasicAA/2010-09-15-GEP-SignedArithmetic.ll @@ -1,4 +1,4 @@ -; RUN: opt < %s -basicaa -aa-eval -print-all-alias-modref-info -disable-output |& grep {1 may alias} +; RUN: opt < %s -basicaa -aa-eval -print-all-alias-modref-info -disable-output |& grep {1 partial alias} ; PR7959 target datalayout = "e-p:32:32:32" diff --git a/test/Analysis/TypeBasedAliasAnalysis/dynamic-indices.ll b/test/Analysis/TypeBasedAliasAnalysis/dynamic-indices.ll new file mode 100644 index 00000000000..52e394b2d09 --- /dev/null +++ b/test/Analysis/TypeBasedAliasAnalysis/dynamic-indices.ll @@ -0,0 +1,131 @@ +; RUN: opt -tbaa -basicaa -gvn -S < %s | FileCheck %s +; PR9971 + +target 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-s0:64:64-f80:128:128-n8:16:32:64" + +%struct.X = type { i32, float } +%union.vector_t = type { [2 x i64] } + +; Don't delete the load after the loop, because it loads values stored +; inside the loop. + +; CHECK: define void @vrlh( + +; CHECK: for.end: +; CHECK: %arrayidx31 = getelementptr inbounds %union.vector_t* %t, i64 0, i32 0, i64 1 +; CHECK: %tmp32 = load i64* %arrayidx31, align 8, !tbaa !3 + +define void @vrlh(%union.vector_t* %va, %union.vector_t* %vb, %union.vector_t* %vd) nounwind { +entry: + %t = alloca %union.vector_t, align 8 + br label %for.body + +for.body: ; preds = %entry, %for.body + %i.01 = phi i32 [ 0, %entry ], [ %inc, %for.body ] + %sub = sub nsw i32 7, %i.01 + %idxprom = sext i32 %sub to i64 + %half = bitcast %union.vector_t* %vb to [8 x i16]* + %arrayidx = getelementptr inbounds [8 x i16]* %half, i64 0, i64 %idxprom + %tmp4 = load i16* %arrayidx, align 2, !tbaa !0 + %conv = zext i16 %tmp4 to i32 + %and = and i32 %conv, 15 + %sub6 = sub nsw i32 7, %i.01 + %idxprom7 = sext i32 %sub6 to i64 + %half9 = bitcast %union.vector_t* %va to [8 x i16]* + %arrayidx10 = getelementptr inbounds [8 x i16]* %half9, i64 0, i64 %idxprom7 + %tmp11 = load i16* %arrayidx10, align 2, !tbaa !0 + %conv12 = zext i16 %tmp11 to i32 + %shl = shl i32 %conv12, %and + %sub15 = sub nsw i32 7, %i.01 + %idxprom16 = sext i32 %sub15 to i64 + %half18 = bitcast %union.vector_t* %va to [8 x i16]* + %arrayidx19 = getelementptr inbounds [8 x i16]* %half18, i64 0, i64 %idxprom16 + %tmp20 = load i16* %arrayidx19, align 2, !tbaa !0 + %conv21 = zext i16 %tmp20 to i32 + %sub23 = sub nsw i32 16, %and + %shr = lshr i32 %conv21, %sub23 + %or = or i32 %shl, %shr + %conv24 = trunc i32 %or to i16 + %sub26 = sub nsw i32 7, %i.01 + %idxprom27 = sext i32 %sub26 to i64 + %half28 = bitcast %union.vector_t* %t to [8 x i16]* + %arrayidx29 = getelementptr inbounds [8 x i16]* %half28, i64 0, i64 %idxprom27 + store i16 %conv24, i16* %arrayidx29, align 2, !tbaa !0 + %inc = add nsw i32 %i.01, 1 + %cmp = icmp slt i32 %inc, 8 + br i1 %cmp, label %for.body, label %for.end + +for.end: ; preds = %for.body + %arrayidx31 = getelementptr inbounds %union.vector_t* %t, i64 0, i32 0, i64 1 + %tmp32 = load i64* %arrayidx31, align 8, !tbaa !3 + %arrayidx35 = getelementptr inbounds %union.vector_t* %vd, i64 0, i32 0, i64 1 + store i64 %tmp32, i64* %arrayidx35, align 8, !tbaa !3 + %arrayidx37 = getelementptr inbounds %union.vector_t* %t, i64 0, i32 0, i64 0 + %tmp38 = load i64* %arrayidx37, align 8, !tbaa !3 + %arrayidx41 = getelementptr inbounds %union.vector_t* %vd, i64 0, i32 0, i64 0 + store i64 %tmp38, i64* %arrayidx41, align 8, !tbaa !3 + ret void +} + +; Do delete the load after the loop. + +; CHECK: define i32 @test0( + +; CHECK: ret i32 0 + +define i32 @test0(%struct.X* %a) nounwind { +entry: + %i = getelementptr inbounds %struct.X* %a, i64 0, i32 0 + store i32 0, i32* %i, align 4, !tbaa !4 + br label %for.body + +for.body: ; preds = %entry, %for.body + %i2.01 = phi i64 [ 0, %entry ], [ %inc, %for.body ] + %f = getelementptr inbounds %struct.X* %a, i64 %i2.01, i32 1 + %tmp6 = load float* %f, align 4, !tbaa !5 + %mul = fmul float %tmp6, 0x40019999A0000000 + store float %mul, float* %f, align 4, !tbaa !5 + %inc = add nsw i64 %i2.01, 1 + %cmp = icmp slt i64 %inc, 10000 + br i1 %cmp, label %for.body, label %for.end + +for.end: ; preds = %for.body + %i9 = getelementptr inbounds %struct.X* %a, i64 0, i32 0 + %tmp10 = load i32* %i9, align 4, !tbaa !4 + ret i32 %tmp10 +} + +; Do delete the load after the loop. + +; CHECK: define float @test1( + +; CHECK: ret float 0x3FD3333340000000 + +define float @test1(%struct.X* %a) nounwind { +entry: + %f = getelementptr inbounds %struct.X* %a, i64 0, i32 1 + store float 0x3FD3333340000000, float* %f, align 4, !tbaa !5 + br label %for.body + +for.body: ; preds = %entry, %for.body + %i.01 = phi i64 [ 0, %entry ], [ %inc, %for.body ] + %i5 = getelementptr inbounds %struct.X* %a, i64 %i.01, i32 0 + %tmp6 = load i32* %i5, align 4, !tbaa !4 + %mul = mul nsw i32 %tmp6, 3 + store i32 %mul, i32* %i5, align 4, !tbaa !4 + %inc = add nsw i64 %i.01, 1 + %cmp = icmp slt i64 %inc, 10000 + br i1 %cmp, label %for.body, label %for.end + +for.end: ; preds = %for.body + %f9 = getelementptr inbounds %struct.X* %a, i64 0, i32 1 + %tmp10 = load float* %f9, align 4, !tbaa !5 + ret float %tmp10 +} + +!0 = metadata !{metadata !"short", metadata !1} +!1 = metadata !{metadata !"omnipotent char", metadata !2} +!2 = metadata !{metadata !"Simple C/C++ TBAA", null} +!3 = metadata !{metadata !"long long", metadata !1} +!4 = metadata !{metadata !"int", metadata !1} +!5 = metadata !{metadata !"float", metadata !1} -- 2.34.1