diff --git a/lib/Transforms/Utils/InlineFunction.cpp b/lib/Transforms/Utils/InlineFunction.cpp index 4fa27673ca7..f6289b5c891 100644 --- a/lib/Transforms/Utils/InlineFunction.cpp +++ b/lib/Transforms/Utils/InlineFunction.cpp @@ -449,17 +449,28 @@ static void AddAliasScopeMetadata(CallSite CS, ValueToValueMapTy &VMap, PtrArgs.push_back(CXI->getPointerOperand()); else if (const AtomicRMWInst *RMWI = dyn_cast(I)) PtrArgs.push_back(RMWI->getPointerOperand()); - else if (const MemIntrinsic *MI = dyn_cast(I)) { - PtrArgs.push_back(MI->getRawDest()); - if (const MemTransferInst *MTI = dyn_cast(MI)) - PtrArgs.push_back(MTI->getRawSource()); + else if (ImmutableCallSite ICS = ImmutableCallSite(I)) { + // If we know that the call does not access memory, then we'll still + // know that about the inlined clone of this call site, and we don't + // need to add metadata. + if (ICS.doesNotAccessMemory()) + continue; + + for (ImmutableCallSite::arg_iterator AI = ICS.arg_begin(), + AE = ICS.arg_end(); AI != AE; ++AI) + // We need to check the underlying objects of all arguments, not just + // the pointer arguments, because we might be passing pointers as + // integers, etc. + // FIXME: If we know that the call only accesses pointer arguments, + // then we only need to check the pointer arguments. + PtrArgs.push_back(*AI); } // If we found no pointers, then this instruction is not suitable for // pairing with an instruction to receive aliasing metadata. - // Simplification during cloning could make this happen, and skip these - // cases for now. - if (PtrArgs.empty()) + // However, if this is a call, this we might just alias with none of the + // noalias arguments. + if (PtrArgs.empty() && !isa(I) && !isa(I)) continue; // It is possible that there is only one underlying object, but you diff --git a/test/Transforms/Inline/noalias-calls.ll b/test/Transforms/Inline/noalias-calls.ll new file mode 100644 index 00000000000..df21fc880f7 --- /dev/null +++ b/test/Transforms/Inline/noalias-calls.ll @@ -0,0 +1,41 @@ +; RUN: opt -inline -enable-noalias-to-md-conversion -S < %s | FileCheck %s +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-S128" +target triple = "x86_64-unknown-linux-gnu" + +declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture readonly, i64, i32, i1) #0 +declare void @hey() #0 + +define void @hello(i8* noalias nocapture %a, i8* noalias nocapture readonly %c, i8* nocapture %b) #1 { +entry: + call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* %b, i64 16, i32 16, i1 0) + call void @llvm.memcpy.p0i8.p0i8.i64(i8* %b, i8* %c, i64 16, i32 16, i1 0) + call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* %c, i64 16, i32 16, i1 0) + call void @hey() + ret void +} + +define void @foo(i8* nocapture %a, i8* nocapture readonly %c, i8* nocapture %b) #1 { +entry: + tail call void @hello(i8* %a, i8* %c, i8* %b) + ret void +} + +; CHECK: define void @foo(i8* nocapture %a, i8* nocapture readonly %c, i8* nocapture %b) #1 { +; CHECK: entry: +; CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* %b, i64 16, i32 16, i1 false) #0, !alias.scope !0, !noalias !3 +; CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %b, i8* %c, i64 16, i32 16, i1 false) #0, !alias.scope !3, !noalias !0 +; CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* %c, i64 16, i32 16, i1 false) #0, !alias.scope !5 +; CHECK: call void @hey() #0, !noalias !5 +; CHECK: ret void +; CHECK: } + +attributes #0 = { nounwind } +attributes #1 = { nounwind uwtable } + +; CHECK: !0 = metadata !{metadata !1} +; CHECK: !1 = metadata !{metadata !1, metadata !2, metadata !"hello: %a"} +; CHECK: !2 = metadata !{metadata !2, metadata !"hello"} +; CHECK: !3 = metadata !{metadata !4} +; CHECK: !4 = metadata !{metadata !4, metadata !2, metadata !"hello: %c"} +; CHECK: !5 = metadata !{metadata !1, metadata !4} +