Index: include/llvm/Transforms/Utils/CodeExtractor.h =================================================================== --- include/llvm/Transforms/Utils/CodeExtractor.h +++ include/llvm/Transforms/Utils/CodeExtractor.h @@ -106,6 +106,12 @@ /// significant impact on the cost however. void findInputsOutputs(ValueSet &Inputs, ValueSet &Outputs, const ValueSet &Allocas) const; + + /// Check if life time marker nodes can be hoisted/sinked into the outline + /// region. + /// + /// Returns true if it is safe to do the code motion. + bool isLegalToShrinkwrapLifetimeMarkers(Instruction *AllocaAddr) const; /// Find the set of allocas whose life ranges are contained within the /// outlined region. /// @@ -114,7 +120,18 @@ /// are used by the lifetime markers are also candidates for shrink- /// wrapping. The instructions that need to be sinked are collected in /// 'Allocas'. - void findAllocas(ValueSet &Allocas) const; + void findAllocas(ValueSet &SinkCands, ValueSet &HoistCands, + BasicBlock *&ExitBlock) const; + + /// Find or create a block within the outline region for placing hoisted + /// code. + /// + /// CommonExitBlock is block outside the outline region. It is the common + /// successor of blocks inside the region. If there exists a single block + /// inside the region that is the predecessor of CommonExitBlock, that block + /// will be returned. Otherwise CommonExitBlock will be split and the + /// original block will be added to the outline region. + BasicBlock *findOrCreateBlockForHoisting(BasicBlock *CommonExitBlock); private: void severSplitPHINodes(BasicBlock *&Header); Index: lib/Transforms/Utils/CodeExtractor.cpp =================================================================== --- lib/Transforms/Utils/CodeExtractor.cpp +++ lib/Transforms/Utils/CodeExtractor.cpp @@ -142,8 +142,134 @@ return false; } -void CodeExtractor::findAllocas(ValueSet &SinkCands) const { +static BasicBlock *getCommonExitBlock(const SetVector &Blocks) { + BasicBlock *CommonExitBlock = nullptr; + if (any_of(Blocks, [&](BasicBlock *Block) { + if (any_of(successors(Block), [&](BasicBlock *Succ) { + // Internal edges, ok. + if (Blocks.count(Succ)) + return false; + if (!CommonExitBlock) { + CommonExitBlock = Succ; + return false; + } + return CommonExitBlock != Succ; + })) + return true; + return false; + })) + return nullptr; + + return CommonExitBlock; +} + +bool CodeExtractor::isLegalToShrinkwrapLifetimeMarkers(Instruction *Addr) const { + AllocaInst *AI = cast(Addr->stripInBoundsConstantOffsets()); + Function *Func = (*Blocks.begin())->getParent(); + for (BasicBlock &BB : *Func) { + if (Blocks.count(&BB)) + continue; + for (Instruction &II : BB) { + + if (isa(II)) + continue; + + unsigned Opcode = II.getOpcode(); + switch (Opcode) { + case Instruction::BitCast: + case Instruction::GetElementPtr: + case Instruction::ICmp: + case Instruction::Br: + case Instruction::Ret: + case Instruction::Alloca: + break; + case Instruction::Load: { + LoadInst *LI = dyn_cast(&II); + Value *Mem = LI->getOperand(0); + // Global variable not can not be aliased with locals. + if (dyn_cast(Mem)) + break; + Value *Base = Mem->stripInBoundsConstantOffsets(); + if (!dyn_cast(Base) || Base == AI) + return false; + break; + } + default: { + IntrinsicInst *IntrInst = dyn_cast(&II); + if (IntrInst) { + if (IntrInst->getIntrinsicID() == Intrinsic::lifetime_start || + IntrInst->getIntrinsicID() == Intrinsic::lifetime_end) + break; + return false; + } + // Treat all the other cases conservatively: + return false; + } + } + } + } + + return true; +} + +BasicBlock * +CodeExtractor::findOrCreateBlockForHoisting(BasicBlock *CommonExitBlock) { + BasicBlock *SinglePredFromOutlineRegion = nullptr; + assert(!Blocks.count(CommonExitBlock) && + "Expect a block outside the region!"); + for (auto *Pred : predecessors(CommonExitBlock)) { + if (!Blocks.count(Pred)) + continue; + if (!SinglePredFromOutlineRegion) { + SinglePredFromOutlineRegion = Pred; + } else if (SinglePredFromOutlineRegion != Pred) { + SinglePredFromOutlineRegion = nullptr; + break; + } + } + + if (SinglePredFromOutlineRegion) + return SinglePredFromOutlineRegion; + +#ifndef NDEBUG + auto getFirstPHI = [](BasicBlock *BB) { + BasicBlock::iterator I = BB->begin(); + PHINode *FirstPhi = nullptr; + while (I != BB->end()) { + PHINode *Phi = dyn_cast(I); + if (!Phi) + break; + if (!FirstPhi) { + FirstPhi = Phi; + break; + } + } + return FirstPhi; + }; +#endif + + // If there are any phi nodes, the single pred either exists or has already + // be created before code extraction. + assert(!getFirstPHI(CommonExitBlock) && "Phi not expected"); + + BasicBlock *NewExitBlock = CommonExitBlock->splitBasicBlock( + CommonExitBlock->getFirstNonPHI()->getIterator()); + + for (auto *Pred : predecessors(CommonExitBlock)) { + if (Blocks.count(Pred)) + continue; + Pred->getTerminator()->replaceUsesOfWith(CommonExitBlock, NewExitBlock); + } + // Now add the old exit block to the outline region. + Blocks.insert(CommonExitBlock); + return CommonExitBlock; +} + +void CodeExtractor::findAllocas(ValueSet &SinkCands, ValueSet &HoistCands, + BasicBlock *&ExitBlock) const { Function *Func = (*Blocks.begin())->getParent(); + ExitBlock = getCommonExitBlock(Blocks); + for (BasicBlock &BB : *Func) { if (Blocks.count(&BB)) continue; @@ -152,49 +278,89 @@ if (!AI) continue; - // Returns true if matching life time markers are found within - // the outlined region. - auto GetLifeTimeMarkers = [&](Instruction *Addr) { + // Find the pair of life time markers for address 'Addr' that are either + // defined inside the outline region or can legally be shrinkwrapped into + // the outline region. If there are not other untracked uses of the + // address, return the pair of markers if found; otherwise return a pair + // of nullptr. + auto GetLifeTimeMarkers = + [&](Instruction *Addr, bool &SinkLifeStart, + bool &HoistLifeEnd) -> std::pair { Instruction *LifeStart = nullptr, *LifeEnd = nullptr; - for (User *U : Addr->users()) { - if (!definedInRegion(Blocks, U)) - return false; + for (User *U : Addr->users()) { IntrinsicInst *IntrInst = dyn_cast(U); if (IntrInst) { if (IntrInst->getIntrinsicID() == Intrinsic::lifetime_start) LifeStart = IntrInst; if (IntrInst->getIntrinsicID() == Intrinsic::lifetime_end) LifeEnd = IntrInst; + continue; } + // Find untracked uses of the address, bail. + if (!definedInRegion(Blocks, U)) + return std::make_pair(nullptr, nullptr); } - return LifeStart && LifeEnd; + + if (!LifeStart || !LifeEnd) + return std::make_pair(nullptr, nullptr); + + SinkLifeStart = !definedInRegion(Blocks, LifeStart); + HoistLifeEnd = !definedInRegion(Blocks, LifeEnd); + // Do legality Check. + if ((SinkLifeStart || HoistLifeEnd) && + !isLegalToShrinkwrapLifetimeMarkers(Addr)) + return std::make_pair(nullptr, nullptr); + + // Check to see if we have a place to do hoisting, if not, bail. + if (HoistLifeEnd && !ExitBlock) + return std::make_pair(nullptr, nullptr); + + return std::make_pair(LifeStart, LifeEnd); }; - if (GetLifeTimeMarkers(AI)) { + bool SinkLifeStart = false, HoistLifeEnd = false; + auto Markers = GetLifeTimeMarkers(AI, SinkLifeStart, HoistLifeEnd); + + if (Markers.first) { + if (SinkLifeStart) + SinkCands.insert(Markers.first); SinkCands.insert(AI); + if (HoistLifeEnd) + HoistCands.insert(Markers.second); continue; } - // Follow the bitcast: + // Follow the bitcast. Instruction *MarkerAddr = nullptr; for (User *U : AI->users()) { - if (U->stripPointerCasts() == AI) { + + if (U->stripInBoundsConstantOffsets() == AI) { + SinkLifeStart = false; + HoistLifeEnd = false; Instruction *Bitcast = cast(U); - if (GetLifeTimeMarkers(Bitcast)) { + Markers = GetLifeTimeMarkers(Bitcast, SinkLifeStart, HoistLifeEnd); + if (Markers.first) { MarkerAddr = Bitcast; continue; } } + + // Found unknown use of AI. if (!definedInRegion(Blocks, U)) { MarkerAddr = nullptr; break; } } + if (MarkerAddr) { + if (SinkLifeStart) + SinkCands.insert(Markers.first); if (!definedInRegion(Blocks, MarkerAddr)) SinkCands.insert(MarkerAddr); SinkCands.insert(AI); + if (HoistLifeEnd) + HoistCands.insert(Markers.second); } } } @@ -780,7 +946,8 @@ if (!isEligible()) return nullptr; - ValueSet inputs, outputs, SinkingCands; + ValueSet inputs, outputs, SinkingCands, HoistingCands; + BasicBlock *CommonExit = nullptr; // Assumption: this is a single-entry code region, and the header is the first // block in the region. @@ -819,7 +986,8 @@ "newFuncRoot"); newFuncRoot->getInstList().push_back(BranchInst::Create(header)); - findAllocas(SinkingCands); + findAllocas(SinkingCands, HoistingCands, CommonExit); + assert(HoistingCands.empty() || CommonExit); // Find inputs to, outputs from the code region. findInputsOutputs(inputs, outputs, SinkingCands); @@ -829,6 +997,13 @@ cast(II)->moveBefore(*newFuncRoot, newFuncRoot->getFirstInsertionPt()); + if (!HoistingCands.empty()) { + auto *HoistToBlock = findOrCreateBlockForHoisting(CommonExit); + Instruction *TI = HoistToBlock->getTerminator(); + for (auto *II : HoistingCands) + cast(II)->moveBefore(TI); + } + // Calculate the exit blocks for the extracted region and the total exit // weights for each of those blocks. DenseMap ExitWeights; Index: test/Transforms/CodeExtractor/live_shrink.ll =================================================================== --- test/Transforms/CodeExtractor/live_shrink.ll +++ test/Transforms/CodeExtractor/live_shrink.ll @@ -0,0 +1,82 @@ +; RUN: opt -partial-inliner < %s | | FileCheck %s +; RUN: opt -pass=partial-inliner < %s | | FileCheck %s + +%class.A = type { i32 } +@cond = local_unnamed_addr global i32 0, align 4 + +; Function Attrs: uwtable +define void @_Z3foov() local_unnamed_addr { +bb: + %tmp = alloca %class.A, align 4 + %tmp1 = bitcast %class.A* %tmp to i8* + call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %tmp1) + %tmp2 = load i32, i32* @cond, align 4, !tbaa !2 + %tmp3 = icmp eq i32 %tmp2, 0 + br i1 %tmp3, label %bb4, label %bb5 + +bb4: ; preds = %bb + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + br label %bb5 + +bb5: ; preds = %bb4, %bb + call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %tmp1) + ret void +} + +; Function Attrs: argmemonly nounwind +declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) + +declare void @_ZN1A7memfuncEv(%class.A*) local_unnamed_addr + +; Function Attrs: argmemonly nounwind +declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) + +; Function Attrs: uwtable +define void @_Z3goov() local_unnamed_addr { +; CHECK-LABEL: @_Z3goov() +bb: +; CHECK: bb: +; CHECK-NOT: alloca +; CHECK-NOT: bitcast +; CHECK-NOT: llvm.lifetime +; CHECK: br i1 +; CHECK: codeRepl.i: +; CHECK: call void @_Z3foov.1_ + + tail call void @_Z3foov() + ret void +} + +; CHECK-LABEL: define internal void @_Z3foov.1_ +; CHECK: newFuncRoot: +; CHECK-NEXT: %tmp = alloca %class.A +; CHECK-NEXT: %tmp1 = bitcast %class.A* %tmp to i8* +; CHECK-NEXT: call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %tmp1) +; CHECK: call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %tmp1) +; CHECK-NEXT: br label %bb5.exitStub + + +!llvm.module.flags = !{!0} +!llvm.ident = !{!1} + +!0 = !{i32 1, !"wchar_size", i32 4} +!1 = !{!"clang version 5.0.0 (trunk 304489)"} +!2 = !{!3, !3, i64 0} +!3 = !{!"int", !4, i64 0} +!4 = !{!"omnipotent char", !5, i64 0} +!5 = !{!"Simple C++ TBAA"} Index: test/Transforms/CodeExtractor/live_shrink_gep.ll =================================================================== --- test/Transforms/CodeExtractor/live_shrink_gep.ll +++ test/Transforms/CodeExtractor/live_shrink_gep.ll @@ -0,0 +1,81 @@ +; RUN: opt -partial-inliner < %s | | FileCheck %s +; RUN: opt -pass=partial-inliner < %s | | FileCheck %s + +%class.A = type { i8 } + +@cond = local_unnamed_addr global i32 0, align 4 + +; Function Attrs: uwtable +define void @_Z3foov() local_unnamed_addr { +bb: + %tmp = alloca %class.A, align 1 + %tmp1 = getelementptr inbounds %class.A, %class.A* %tmp, i64 0, i32 0 + call void @llvm.lifetime.start.p0i8(i64 1, i8* nonnull %tmp1) + %tmp2 = load i32, i32* @cond, align 4, !tbaa !2 + %tmp3 = icmp eq i32 %tmp2, 0 + br i1 %tmp3, label %bb4, label %bb5 + +bb4: ; preds = %bb + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + br label %bb5 + +bb5: ; preds = %bb4, %bb + call void @llvm.lifetime.end.p0i8(i64 1, i8* nonnull %tmp1) + ret void +} + +; Function Attrs: argmemonly nounwind +declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) + +declare void @_ZN1A7memfuncEv(%class.A*) local_unnamed_addr + +; Function Attrs: argmemonly nounwind +declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) + +; Function Attrs: uwtable +define void @_Z3goov() local_unnamed_addr { +; CHECK-LABEL: @_Z3goov() +bb: +; CHECK: bb: +; CHECK-NOT: alloca +; CHECK-NOT: getelementptr +; CHECK-NOT: llvm.lifetime +; CHECK: br i1 +; CHECK: codeRepl.i: +; CHECK: call void @_Z3foov.1_ + tail call void @_Z3foov() + ret void +} + +; CHECK-LABEL: define internal void @_Z3foov.1_ +; CHECK: newFuncRoot: +; CHECK-NEXT: %tmp = alloca %class.A +; CHECK-NEXT: %tmp1 = getelementptr +; CHECK-NEXT: call void @llvm.lifetime.start.p0i8 +; CHECK: call void @llvm.lifetime.end.p0i8 +; CHECK-NEXT: br label %bb5.exitStub + +!llvm.module.flags = !{!0} +!llvm.ident = !{!1} + +!0 = !{i32 1, !"wchar_size", i32 4} +!1 = !{!"clang version 5.0.0 (trunk 304489)"} +!2 = !{!3, !3, i64 0} +!3 = !{!"int", !4, i64 0} +!4 = !{!"omnipotent char", !5, i64 0} +!5 = !{!"Simple C++ TBAA"} Index: test/Transforms/CodeExtractor/live_shrink_hoist.ll =================================================================== --- test/Transforms/CodeExtractor/live_shrink_hoist.ll +++ test/Transforms/CodeExtractor/live_shrink_hoist.ll @@ -0,0 +1,80 @@ +; RUN: opt -partial-inliner -max-num-inline-blocks=2 < %s | | FileCheck %s +; RUN: opt -pass=partial-inliner -max-num-inline-blocks=2 < %s | | FileCheck %s + +%class.A = type { i32 } + +@cond = local_unnamed_addr global i32 0, align 4 + +; Function Attrs: uwtable +define void @_Z3foov() local_unnamed_addr { +bb: + %tmp = alloca %class.A, align 4 + %tmp1 = bitcast %class.A* %tmp to i8* + call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %tmp1) + %tmp2 = load i32, i32* @cond, align 4, !tbaa !2 + %tmp3 = icmp eq i32 %tmp2, 0 + br i1 %tmp3, label %bb4, label %bb9 + +bb4: ; preds = %bb + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + %tmp5 = getelementptr inbounds %class.A, %class.A* %tmp, i64 0, i32 0 + %tmp6 = load i32, i32* %tmp5, align 4, !tbaa !6 + %tmp7 = icmp sgt i32 %tmp6, 0 + br i1 %tmp7, label %bb9, label %bb8 + +bb8: ; preds = %bb4 + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + br label %bb9 + +bb9: ; preds = %bb8, %bb4, %bb + call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %tmp1) + ret void +} + +; Function Attrs: argmemonly nounwind +declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) + +declare void @_ZN1A7memfuncEv(%class.A*) local_unnamed_addr + +; Function Attrs: argmemonly nounwind +declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) + +; Function Attrs: uwtable +define void @_Z3goov() local_unnamed_addr { +bb: + tail call void @_Z3foov() + ret void +} + +; CHECK-LABEL: define internal void @_Z3foov.1_ +; CHECK: bb9: +; CHECK: call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %tmp1) +; CHECK: br label %.exitStub + + + +!llvm.module.flags = !{!0} +!llvm.ident = !{!1} + +!0 = !{i32 1, !"wchar_size", i32 4} +!1 = !{!"clang version 5.0.0 (trunk 304489)"} +!2 = !{!3, !3, i64 0} +!3 = !{!"int", !4, i64 0} +!4 = !{!"omnipotent char", !5, i64 0} +!5 = !{!"Simple C++ TBAA"} +!6 = !{!7, !3, i64 0} +!7 = !{!"_ZTS1A", !3, i64 0} Index: test/Transforms/CodeExtractor/live_shrink_multiple.ll =================================================================== --- test/Transforms/CodeExtractor/live_shrink_multiple.ll +++ test/Transforms/CodeExtractor/live_shrink_multiple.ll @@ -0,0 +1,81 @@ +; RUN: opt -partial-inliner < %s | | FileCheck %s +; RUN: opt -pass=partial-inliner < %s | | FileCheck %s + +%class.A = type { i32 } +@cond = local_unnamed_addr global i32 0, align 4 + +; Function Attrs: uwtable +define void @_Z3foov() local_unnamed_addr { +bb: + %tmp = alloca %class.A, align 4 + %tmp1 = alloca %class.A, align 4 + %tmp2 = bitcast %class.A* %tmp to i8* + call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %tmp2) + %tmp3 = bitcast %class.A* %tmp1 to i8* + call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %tmp3) + %tmp4 = load i32, i32* @cond, align 4, !tbaa !2 + %tmp5 = icmp eq i32 %tmp4, 0 + br i1 %tmp5, label %bb6, label %bb7 + +bb6: ; preds = %bb + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp1) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp1) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp1) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp1) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp1) + br label %bb7 + +bb7: ; preds = %bb6, %bb + call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %tmp3) + call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %tmp2) + ret void +} + +; Function Attrs: argmemonly nounwind +declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) + +declare void @_ZN1A7memfuncEv(%class.A*) local_unnamed_addr + +; Function Attrs: argmemonly nounwind +declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) + +; Function Attrs: uwtable +define void @_Z3goov() local_unnamed_addr { +bb: + tail call void @_Z3foov() + ret void +} + +; CHECK-LABEL: define internal void @_Z3foov.1_ +; CHECK: newFuncRoot: +; CHECK-NEXT: alloca +; CHECK-NEXT: bitcast +; CHECK-NEXT: call void @llvm.lifetime.start.p0i8 +; CHECK-NEXT: alloca +; CHECK-NEXT: bitcast +; CHECK-NEXT: call void @llvm.lifetime.start.p0i8 +; CHECK: call void @llvm.lifetime.end.p0i8 +; CHECK-NEXT: call void @llvm.lifetime.end.p0i8 +; CHECK-NEXT: br label {{.*}}exitStub + + +!llvm.module.flags = !{!0} +!llvm.ident = !{!1} + +!0 = !{i32 1, !"wchar_size", i32 4} +!1 = !{!"clang version 5.0.0 (trunk 304489)"} +!2 = !{!3, !3, i64 0} +!3 = !{!"int", !4, i64 0} +!4 = !{!"omnipotent char", !5, i64 0} +!5 = !{!"Simple C++ TBAA"} Index: test/Transforms/CodeExtractor/live_shrink_unsafe.ll =================================================================== --- test/Transforms/CodeExtractor/live_shrink_unsafe.ll +++ test/Transforms/CodeExtractor/live_shrink_unsafe.ll @@ -0,0 +1,104 @@ +; The expected behavior of this file is expected to change when partial +; inlining legality check is enhanced. + +; RUN: opt -partial-inliner -skip-partial-inlining-cost-analysis < %s | | FileCheck %s +; RUN: opt -pass=partial-inliner skip-partial-inlining-cost-analysis < %s | | FileCheck %s + +%class.A = type { i32 } + +@cond = local_unnamed_addr global i32 0, align 4 +@condptr = external local_unnamed_addr global i32*, align 8 + +; Function Attrs: uwtable +define void @_Z3foo_unknown_mem_accessv() local_unnamed_addr { +bb: + %tmp = alloca %class.A, align 4 + %tmp1 = alloca %class.A, align 4 + %tmp2 = bitcast %class.A* %tmp to i8* + call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %tmp2) + %tmp3 = bitcast %class.A* %tmp1 to i8* + call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %tmp3) + %tmp4 = load i32*, i32** @condptr, align 8, !tbaa !2 + %tmp5 = load i32, i32* %tmp4, align 4, !tbaa !6 + %tmp6 = icmp eq i32 %tmp5, 0 + br i1 %tmp6, label %bb7, label %bb8 + +bb7: ; preds = %bb + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp1) + br label %bb8 + +bb8: ; preds = %bb7, %bb + call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %tmp3) + call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %tmp2) + ret void +} + +declare void @_Z3barv() local_unnamed_addr +declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) +declare void @_ZN1A7memfuncEv(%class.A*) local_unnamed_addr +declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) + +define void @_Z3foo_unknown_calli(i32 %arg) local_unnamed_addr { +bb: + %tmp = alloca %class.A, align 4 + %tmp1 = bitcast %class.A* %tmp to i8* + call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %tmp1) + tail call void @_Z3barv() + %tmp2 = icmp eq i32 %arg, 0 + br i1 %tmp2, label %bb3, label %bb4 + +bb3: ; preds = %bb + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + call void @_ZN1A7memfuncEv(%class.A* nonnull %tmp) + br label %bb4 + +bb4: ; preds = %bb3, %bb + call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %tmp1) + ret void +} + +define void @_Z3goov() local_unnamed_addr { +; CHECK-LABEL: @_Z3goov +; CHECK-NEXT: bb: +; CHECK: alloca +; CHECK: lifetime +bb: + call void @_Z3foo_unknown_mem_accessv() + %tmp = load i32, i32* @cond, align 4, !tbaa !2 + tail call void @_Z3foo_unknown_calli(i32 %tmp) + ret void +} + +; CHECK-LABEL define internal void @_Z3foo_unknown_calli.1_bb3 +; CHECK: newFuncRoot: +; CHECK-NEXT: br label %bb3 + +; CHECK: bb4.exitStub: +; CHECK-NEXT: ret void + +; CHECK: bb3: +; CHECK-NOT: lifetime.ed +; CHECK: br label %bb4.exitStub + + + +!llvm.module.flags = !{!0} +!llvm.ident = !{!1} + +!0 = !{i32 1, !"wchar_size", i32 4} +!1 = !{!"clang version 5.0.0 (trunk 304489)"} +!2 = !{!3, !3, i64 0} +!3 = !{!"any pointer", !4, i64 0} +!4 = !{!"omnipotent char", !5, i64 0} +!5 = !{!"Simple C++ TBAA"} +!6 = !{!7, !7, i64 0} +!7 = !{!"int", !4, i64 0}