Index: llvm/lib/Transforms/IPO/Attributor.cpp =================================================================== --- llvm/lib/Transforms/IPO/Attributor.cpp +++ llvm/lib/Transforms/IPO/Attributor.cpp @@ -2458,15 +2458,52 @@ << " is assumed NoAlias in the definition\n"); // (ii) Check whether the value is captured in the scope using AANoCapture. - // FIXME: This is conservative though, it is better to look at CFG and - // check only uses possibly executed before this callsite. + // Look at CFG and check only uses possibly executed before this + // callsite. + + auto UsePred = [&](const Use &U, bool &Follow) -> bool { + Instruction *UserI = cast(U.getUser()); + const Function *ScopeFn = IRP.getAnchorScope(); + if (ScopeFn) { + const auto &ReachabilityAA = + A.getAAFor(*this, IRPosition::function(*ScopeFn)); + + if (auto *CB = dyn_cast(UserI)) { + if (CB->isArgOperand(&U)) { + unsigned ArgNo = CB->getArgOperandNo(&U); + + const auto &NoCaptureAA = A.getAAFor( + *this, IRPosition::callsite_argument(*CB, ArgNo)); + + // Doesn't have nocapture but reachable. + if (!NoCaptureAA.isAssumedNoCapture() && + ReachabilityAA.isAssumedReachable(UserI, getCtxI())) + return true; + } + } + } + + // For cases which can potentially have more users + if (isa(U) || isa(U) || isa(U) || + isa(U)) { + Follow = true; + return true; + } + + LLVM_DEBUG(dbgs() << "[Attributor][AANoAliasCSArg] Unknown user: " << *U + << "\n"); + return false; + }; auto &NoCaptureAA = A.getAAFor(*this, IRP); if (!NoCaptureAA.isAssumedNoCaptureMaybeReturned()) { - LLVM_DEBUG( - dbgs() << "[Attributor][AANoAliasCSArg] " << V - << " cannot be noalias as it is potentially captured\n"); - return indicatePessimisticFixpoint(); + // Check all of V's uses. + if (!A.checkForAllUses(UsePred, *this, V)) { + LLVM_DEBUG( + dbgs() << "[Attributor][AANoAliasCSArg] " << V + << " cannot be noalias as it is potentially captured\n"); + return indicatePessimisticFixpoint(); + } } // (iii) Check there is no other pointer argument which could alias with the Index: llvm/test/Transforms/Attributor/IPConstantProp/pthreads.ll =================================================================== --- llvm/test/Transforms/Attributor/IPConstantProp/pthreads.ll +++ llvm/test/Transforms/Attributor/IPConstantProp/pthreads.ll @@ -33,10 +33,10 @@ ; CHECK-NEXT: [[ALLOC1:%.*]] = alloca i8, align 8 ; CHECK-NEXT: [[ALLOC2:%.*]] = alloca i8, align 8 ; CHECK-NEXT: [[THREAD:%.*]] = alloca i64, align 8 -; CHECK-NEXT: [[CALL:%.*]] = call i32 @pthread_create(i64* nonnull align 8 dereferenceable(8) [[THREAD]], %union.pthread_attr_t* noalias null, i8* (i8*)* nonnull @foo, i8* noalias nofree readnone null) -; CHECK-NEXT: [[CALL1:%.*]] = call i32 @pthread_create(i64* nonnull align 8 dereferenceable(8) [[THREAD]], %union.pthread_attr_t* noalias null, i8* (i8*)* nonnull @bar, i8* nofree nonnull readnone align 8 dereferenceable(8) bitcast (i8** @GlobalVPtr to i8*)) -; CHECK-NEXT: [[CALL2:%.*]] = call i32 @pthread_create(i64* nonnull align 8 dereferenceable(8) [[THREAD]], %union.pthread_attr_t* noalias null, i8* (i8*)* nonnull @baz, i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(1) [[ALLOC1]]) -; CHECK-NEXT: [[CALL3:%.*]] = call i32 @pthread_create(i64* nonnull align 8 dereferenceable(8) [[THREAD]], %union.pthread_attr_t* noalias null, i8* (i8*)* nonnull @buz, i8* nofree nonnull readnone align 8 dereferenceable(1) [[ALLOC2]]) +; CHECK-NEXT: [[CALL:%.*]] = call i32 @pthread_create(i64* noalias nonnull align 8 dereferenceable(8) [[THREAD]], %union.pthread_attr_t* noalias null, i8* (i8*)* nonnull @foo, i8* noalias nofree readnone null) +; CHECK-NEXT: [[CALL1:%.*]] = call i32 @pthread_create(i64* noalias nonnull align 8 dereferenceable(8) [[THREAD]], %union.pthread_attr_t* noalias null, i8* (i8*)* nonnull @bar, i8* nofree nonnull readnone align 8 dereferenceable(8) bitcast (i8** @GlobalVPtr to i8*)) +; CHECK-NEXT: [[CALL2:%.*]] = call i32 @pthread_create(i64* noalias nonnull align 8 dereferenceable(8) [[THREAD]], %union.pthread_attr_t* noalias null, i8* (i8*)* nonnull @baz, i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(1) [[ALLOC1]]) +; CHECK-NEXT: [[CALL3:%.*]] = call i32 @pthread_create(i64* noalias nonnull align 8 dereferenceable(8) [[THREAD]], %union.pthread_attr_t* noalias null, i8* (i8*)* nonnull @buz, i8* noalias nofree nonnull readnone align 8 dereferenceable(1) [[ALLOC2]]) ; CHECK-NEXT: ret i32 0 ; entry: @@ -84,7 +84,7 @@ define internal i8* @buz(i8* %arg) { ; CHECK-LABEL: define {{[^@]+}}@buz -; CHECK-SAME: (i8* nofree nonnull readnone returned align 8 dereferenceable(1) [[ARG:%.*]]) +; CHECK-SAME: (i8* noalias nofree nonnull readnone returned align 8 dereferenceable(1) [[ARG:%.*]]) ; CHECK-NEXT: entry: ; CHECK-NEXT: ret i8* [[ARG]] ; Index: llvm/test/Transforms/Attributor/callbacks.ll =================================================================== --- llvm/test/Transforms/Attributor/callbacks.ll +++ llvm/test/Transforms/Attributor/callbacks.ll @@ -23,7 +23,7 @@ ; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[B]] to i8* ; CHECK-NEXT: store i32 42, i32* [[B]], align 32 ; CHECK-NEXT: store i32* [[B]], i32** [[C]], align 64 -; CHECK-NEXT: call void (i32*, i32*, void (i32*, i32*, ...)*, ...) @t0_callback_broker(i32* noalias null, i32* nonnull align 128 dereferenceable(4) [[PTR]], void (i32*, i32*, ...)* nonnull bitcast (void (i32*, i32*, i32*, i64, i32**)* @t0_callback_callee to void (i32*, i32*, ...)*), i32* align 256 [[A:%.*]], i64 99, i32** noalias nocapture nonnull readonly align 64 dereferenceable(8) [[C]]) +; CHECK-NEXT: call void (i32*, i32*, void (i32*, i32*, ...)*, ...) @t0_callback_broker(i32* noalias null, i32* noalias nonnull align 128 dereferenceable(4) [[PTR]], void (i32*, i32*, ...)* nonnull bitcast (void (i32*, i32*, i32*, i64, i32**)* @t0_callback_callee to void (i32*, i32*, ...)*), i32* align 256 [[A:%.*]], i64 99, i32** noalias nocapture nonnull readonly align 64 dereferenceable(8) [[C]]) ; CHECK-NEXT: ret void ; Index: llvm/test/Transforms/Attributor/noalias.ll =================================================================== --- llvm/test/Transforms/Attributor/noalias.ll +++ llvm/test/Transforms/Attributor/noalias.ll @@ -182,14 +182,14 @@ declare void @test10_helper_1(i8* %a) define void @test10_helper_2(i8* noalias %a) { -; CHECK: tail call void @test10_helper_1(i8* %a) +; CHECK: tail call void @test10_helper_1(i8* noalias %a) tail call void @test10_helper_1(i8* %a) ret void } define void @test10(i8* noalias %a) { ; CHECK: define void @test10(i8* noalias %a) ; FIXME: missing noalias -; CHECK-NEXT: tail call void @test10_helper_1(i8* %a) +; CHECK-NEXT: tail call void @test10_helper_1(i8* noalias %a) tail call void @test10_helper_1(i8* %a) ; CHECK-NEXT: tail call void @test10_helper_2(i8* noalias %a)