//===--- CodeGenFunction.cpp - Emit LLVM Code from ASTs for a Function ----===// // // The LLVM Compiler Infrastructure // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // // This coordinates the per-function state used while generating code. // //===----------------------------------------------------------------------===// #include "CodeGenFunction.h" #include "CodeGenModule.h" #include "CGDebugInfo.h" #include "CGException.h" #include "clang/Basic/TargetInfo.h" #include "clang/AST/APValue.h" #include "clang/AST/ASTContext.h" #include "clang/AST/Decl.h" #include "clang/AST/DeclCXX.h" #include "clang/AST/StmtCXX.h" #include "clang/Frontend/CodeGenOptions.h" #include "llvm/Target/TargetData.h" #include "llvm/Intrinsics.h" using namespace clang; using namespace CodeGen; CodeGenFunction::CodeGenFunction(CodeGenModule &cgm) : BlockFunction(cgm, *this, Builder), CGM(cgm), Target(CGM.getContext().Target), Builder(cgm.getModule().getContext()), ExceptionSlot(0), DebugInfo(0), IndirectBranch(0), SwitchInsn(0), CaseRangeBlock(0), InvokeDest(0), DidCallStackSave(false), UnreachableBlock(0), CXXThisDecl(0), CXXThisValue(0), CXXVTTDecl(0), CXXVTTValue(0), ConditionalBranchLevel(0), TerminateLandingPad(0), TerminateHandler(0), TrapBB(0) { // Get some frequently used types. LLVMPointerWidth = Target.getPointerWidth(0); llvm::LLVMContext &LLVMContext = CGM.getLLVMContext(); IntPtrTy = llvm::IntegerType::get(LLVMContext, LLVMPointerWidth); Int32Ty = llvm::Type::getInt32Ty(LLVMContext); Int64Ty = llvm::Type::getInt64Ty(LLVMContext); Exceptions = getContext().getLangOptions().Exceptions; CatchUndefined = getContext().getLangOptions().CatchUndefined; CGM.getMangleContext().startNewFunction(); } ASTContext &CodeGenFunction::getContext() const { return CGM.getContext(); } llvm::Value *CodeGenFunction::GetAddrOfLocalVar(const VarDecl *VD) { llvm::Value *Res = LocalDeclMap[VD]; assert(Res && "Invalid argument to GetAddrOfLocalVar(), no decl!"); return Res; } llvm::Constant * CodeGenFunction::GetAddrOfStaticLocalVar(const VarDecl *BVD) { return cast(GetAddrOfLocalVar(BVD)); } const llvm::Type *CodeGenFunction::ConvertTypeForMem(QualType T) { return CGM.getTypes().ConvertTypeForMem(T); } const llvm::Type *CodeGenFunction::ConvertType(QualType T) { return CGM.getTypes().ConvertType(T); } bool CodeGenFunction::hasAggregateLLVMType(QualType T) { return T->isRecordType() || T->isArrayType() || T->isAnyComplexType() || T->isMemberFunctionPointerType(); } void CodeGenFunction::EmitReturnBlock() { // For cleanliness, we try to avoid emitting the return block for // simple cases. llvm::BasicBlock *CurBB = Builder.GetInsertBlock(); if (CurBB) { assert(!CurBB->getTerminator() && "Unexpected terminated block."); // We have a valid insert point, reuse it if it is empty or there are no // explicit jumps to the return block. if (CurBB->empty() || ReturnBlock.Block->use_empty()) { ReturnBlock.Block->replaceAllUsesWith(CurBB); delete ReturnBlock.Block; } else EmitBlock(ReturnBlock.Block); return; } // Otherwise, if the return block is the target of a single direct // branch then we can just put the code in that block instead. This // cleans up functions which started with a unified return block. if (ReturnBlock.Block->hasOneUse()) { llvm::BranchInst *BI = dyn_cast(*ReturnBlock.Block->use_begin()); if (BI && BI->isUnconditional() && BI->getSuccessor(0) == ReturnBlock.Block) { // Reset insertion point and delete the branch. Builder.SetInsertPoint(BI->getParent()); BI->eraseFromParent(); delete ReturnBlock.Block; return; } } // FIXME: We are at an unreachable point, there is no reason to emit the block // unless it has uses. However, we still need a place to put the debug // region.end for now. EmitBlock(ReturnBlock.Block); } static void EmitIfUsed(CodeGenFunction &CGF, llvm::BasicBlock *BB) { if (!BB) return; if (!BB->use_empty()) return CGF.CurFn->getBasicBlockList().push_back(BB); delete BB; } void CodeGenFunction::FinishFunction(SourceLocation EndLoc) { assert(BreakContinueStack.empty() && "mismatched push/pop in break/continue stack!"); // Emit function epilog (to return). EmitReturnBlock(); EmitFunctionInstrumentation("__cyg_profile_func_exit"); // Emit debug descriptor for function end. if (CGDebugInfo *DI = getDebugInfo()) { DI->setLocation(EndLoc); DI->EmitRegionEnd(CurFn, Builder); } EmitFunctionEpilog(*CurFnInfo); EmitEndEHSpec(CurCodeDecl); assert(EHStack.empty() && "did not remove all scopes from cleanup stack!"); // If someone did an indirect goto, emit the indirect goto block at the end of // the function. if (IndirectBranch) { EmitBlock(IndirectBranch->getParent()); Builder.ClearInsertionPoint(); } // Remove the AllocaInsertPt instruction, which is just a convenience for us. llvm::Instruction *Ptr = AllocaInsertPt; AllocaInsertPt = 0; Ptr->eraseFromParent(); // If someone took the address of a label but never did an indirect goto, we // made a zero entry PHI node, which is illegal, zap it now. if (IndirectBranch) { llvm::PHINode *PN = cast(IndirectBranch->getAddress()); if (PN->getNumIncomingValues() == 0) { PN->replaceAllUsesWith(llvm::UndefValue::get(PN->getType())); PN->eraseFromParent(); } } EmitIfUsed(*this, TerminateLandingPad); EmitIfUsed(*this, TerminateHandler); EmitIfUsed(*this, UnreachableBlock); if (CGM.getCodeGenOpts().EmitDeclMetadata) EmitDeclMetadata(); } /// ShouldInstrumentFunction - Return true if the current function should be /// instrumented with __cyg_profile_func_* calls bool CodeGenFunction::ShouldInstrumentFunction() { if (!CGM.getCodeGenOpts().InstrumentFunctions) return false; if (CurFuncDecl->hasAttr()) return false; return true; } /// EmitFunctionInstrumentation - Emit LLVM code to call the specified /// instrumentation function with the current function and the call site, if /// function instrumentation is enabled. void CodeGenFunction::EmitFunctionInstrumentation(const char *Fn) { if (!ShouldInstrumentFunction()) return; const llvm::PointerType *PointerTy; const llvm::FunctionType *FunctionTy; std::vector ProfileFuncArgs; // void __cyg_profile_func_{enter,exit} (void *this_fn, void *call_site); PointerTy = llvm::Type::getInt8PtrTy(VMContext); ProfileFuncArgs.push_back(PointerTy); ProfileFuncArgs.push_back(PointerTy); FunctionTy = llvm::FunctionType::get( llvm::Type::getVoidTy(VMContext), ProfileFuncArgs, false); llvm::Constant *F = CGM.CreateRuntimeFunction(FunctionTy, Fn); llvm::CallInst *CallSite = Builder.CreateCall( CGM.getIntrinsic(llvm::Intrinsic::returnaddress, 0, 0), llvm::ConstantInt::get(Int32Ty, 0), "callsite"); Builder.CreateCall2(F, llvm::ConstantExpr::getBitCast(CurFn, PointerTy), CallSite); } void CodeGenFunction::StartFunction(GlobalDecl GD, QualType RetTy, llvm::Function *Fn, const FunctionArgList &Args, SourceLocation StartLoc) { const Decl *D = GD.getDecl(); DidCallStackSave = false; CurCodeDecl = CurFuncDecl = D; FnRetTy = RetTy; CurFn = Fn; assert(CurFn->isDeclaration() && "Function already has body?"); // Pass inline keyword to optimizer if it appears explicitly on any // declaration. if (const FunctionDecl *FD = dyn_cast_or_null(D)) for (FunctionDecl::redecl_iterator RI = FD->redecls_begin(), RE = FD->redecls_end(); RI != RE; ++RI) if (RI->isInlineSpecified()) { Fn->addFnAttr(llvm::Attribute::InlineHint); break; } llvm::BasicBlock *EntryBB = createBasicBlock("entry", CurFn); // Create a marker to make it easy to insert allocas into the entryblock // later. Don't create this with the builder, because we don't want it // folded. llvm::Value *Undef = llvm::UndefValue::get(Int32Ty); AllocaInsertPt = new llvm::BitCastInst(Undef, Int32Ty, "", EntryBB); if (Builder.isNamePreserving()) AllocaInsertPt->setName("allocapt"); ReturnBlock = getJumpDestInCurrentScope("return"); Builder.SetInsertPoint(EntryBB); QualType FnType = getContext().getFunctionType(RetTy, 0, 0, false, 0, false, false, 0, 0, /*FIXME?*/ FunctionType::ExtInfo()); // Emit subprogram debug descriptor. if (CGDebugInfo *DI = getDebugInfo()) { DI->setLocation(StartLoc); DI->EmitFunctionStart(GD, FnType, CurFn, Builder); } EmitFunctionInstrumentation("__cyg_profile_func_enter"); // FIXME: Leaked. // CC info is ignored, hopefully? CurFnInfo = &CGM.getTypes().getFunctionInfo(FnRetTy, Args, FunctionType::ExtInfo()); if (RetTy->isVoidType()) { // Void type; nothing to return. ReturnValue = 0; } else if (CurFnInfo->getReturnInfo().getKind() == ABIArgInfo::Indirect && hasAggregateLLVMType(CurFnInfo->getReturnType())) { // Indirect aggregate return; emit returned value directly into sret slot. // This reduces code size, and affects correctness in C++. ReturnValue = CurFn->arg_begin(); } else { ReturnValue = CreateIRTemp(RetTy, "retval"); } EmitStartEHSpec(CurCodeDecl); EmitFunctionProlog(*CurFnInfo, CurFn, Args); if (CXXThisDecl) CXXThisValue = Builder.CreateLoad(LocalDeclMap[CXXThisDecl], "this"); if (CXXVTTDecl) CXXVTTValue = Builder.CreateLoad(LocalDeclMap[CXXVTTDecl], "vtt"); // If any of the arguments have a variably modified type, make sure to // emit the type size. for (FunctionArgList::const_iterator i = Args.begin(), e = Args.end(); i != e; ++i) { QualType Ty = i->second; if (Ty->isVariablyModifiedType()) EmitVLASize(Ty); } } void CodeGenFunction::EmitFunctionBody(FunctionArgList &Args) { const FunctionDecl *FD = cast(CurGD.getDecl()); assert(FD->getBody()); EmitStmt(FD->getBody()); } void CodeGenFunction::GenerateCode(GlobalDecl GD, llvm::Function *Fn) { const FunctionDecl *FD = cast(GD.getDecl()); // Check if we should generate debug info for this function. if (CGM.getDebugInfo() && !FD->hasAttr()) DebugInfo = CGM.getDebugInfo(); FunctionArgList Args; CurGD = GD; if (const CXXMethodDecl *MD = dyn_cast(FD)) { if (MD->isInstance()) { // Create the implicit 'this' decl. // FIXME: I'm not entirely sure I like using a fake decl just for code // generation. Maybe we can come up with a better way? CXXThisDecl = ImplicitParamDecl::Create(getContext(), 0, FD->getLocation(), &getContext().Idents.get("this"), MD->getThisType(getContext())); Args.push_back(std::make_pair(CXXThisDecl, CXXThisDecl->getType())); // Check if we need a VTT parameter as well. if (CodeGenVTables::needsVTTParameter(GD)) { // FIXME: The comment about using a fake decl above applies here too. QualType T = getContext().getPointerType(getContext().VoidPtrTy); CXXVTTDecl = ImplicitParamDecl::Create(getContext(), 0, FD->getLocation(), &getContext().Idents.get("vtt"), T); Args.push_back(std::make_pair(CXXVTTDecl, CXXVTTDecl->getType())); } } } if (FD->getNumParams()) { const FunctionProtoType* FProto = FD->getType()->getAs(); assert(FProto && "Function def must have prototype!"); for (unsigned i = 0, e = FD->getNumParams(); i != e; ++i) Args.push_back(std::make_pair(FD->getParamDecl(i), FProto->getArgType(i))); } SourceRange BodyRange; if (Stmt *Body = FD->getBody()) BodyRange = Body->getSourceRange(); // Emit the standard function prologue. StartFunction(GD, FD->getResultType(), Fn, Args, BodyRange.getBegin()); // Generate the body of the function. if (isa(FD)) EmitDestructorBody(Args); else if (isa(FD)) EmitConstructorBody(Args); else EmitFunctionBody(Args); // Emit the standard function epilogue. FinishFunction(BodyRange.getEnd()); // Destroy the 'this' declaration. if (CXXThisDecl) CXXThisDecl->Destroy(getContext()); // Destroy the VTT declaration. if (CXXVTTDecl) CXXVTTDecl->Destroy(getContext()); } /// ContainsLabel - Return true if the statement contains a label in it. If /// this statement is not executed normally, it not containing a label means /// that we can just remove the code. bool CodeGenFunction::ContainsLabel(const Stmt *S, bool IgnoreCaseStmts) { // Null statement, not a label! if (S == 0) return false; // If this is a label, we have to emit the code, consider something like: // if (0) { ... foo: bar(); } goto foo; if (isa(S)) return true; // If this is a case/default statement, and we haven't seen a switch, we have // to emit the code. if (isa(S) && !IgnoreCaseStmts) return true; // If this is a switch statement, we want to ignore cases below it. if (isa(S)) IgnoreCaseStmts = true; // Scan subexpressions for verboten labels. for (Stmt::const_child_iterator I = S->child_begin(), E = S->child_end(); I != E; ++I) if (ContainsLabel(*I, IgnoreCaseStmts)) return true; return false; } /// ConstantFoldsToSimpleInteger - If the sepcified expression does not fold to /// a constant, or if it does but contains a label, return 0. If it constant /// folds to 'true' and does not contain a label, return 1, if it constant folds /// to 'false' and does not contain a label, return -1. int CodeGenFunction::ConstantFoldsToSimpleInteger(const Expr *Cond) { // FIXME: Rename and handle conversion of other evaluatable things // to bool. Expr::EvalResult Result; if (!Cond->Evaluate(Result, getContext()) || !Result.Val.isInt() || Result.HasSideEffects) return 0; // Not foldable, not integer or not fully evaluatable. if (CodeGenFunction::ContainsLabel(Cond)) return 0; // Contains a label. return Result.Val.getInt().getBoolValue() ? 1 : -1; } /// EmitBranchOnBoolExpr - Emit a branch on a boolean condition (e.g. for an if /// statement) to the specified blocks. Based on the condition, this might try /// to simplify the codegen of the conditional based on the branch. /// void CodeGenFunction::EmitBranchOnBoolExpr(const Expr *Cond, llvm::BasicBlock *TrueBlock, llvm::BasicBlock *FalseBlock) { if (const ParenExpr *PE = dyn_cast(Cond)) return EmitBranchOnBoolExpr(PE->getSubExpr(), TrueBlock, FalseBlock); if (const BinaryOperator *CondBOp = dyn_cast(Cond)) { // Handle X && Y in a condition. if (CondBOp->getOpcode() == BinaryOperator::LAnd) { // If we have "1 && X", simplify the code. "0 && X" would have constant // folded if the case was simple enough. if (ConstantFoldsToSimpleInteger(CondBOp->getLHS()) == 1) { // br(1 && X) -> br(X). return EmitBranchOnBoolExpr(CondBOp->getRHS(), TrueBlock, FalseBlock); } // If we have "X && 1", simplify the code to use an uncond branch. // "X && 0" would have been constant folded to 0. if (ConstantFoldsToSimpleInteger(CondBOp->getRHS()) == 1) { // br(X && 1) -> br(X). return EmitBranchOnBoolExpr(CondBOp->getLHS(), TrueBlock, FalseBlock); } // Emit the LHS as a conditional. If the LHS conditional is false, we // want to jump to the FalseBlock. llvm::BasicBlock *LHSTrue = createBasicBlock("land.lhs.true"); EmitBranchOnBoolExpr(CondBOp->getLHS(), LHSTrue, FalseBlock); EmitBlock(LHSTrue); // Any temporaries created here are conditional. BeginConditionalBranch(); EmitBranchOnBoolExpr(CondBOp->getRHS(), TrueBlock, FalseBlock); EndConditionalBranch(); return; } else if (CondBOp->getOpcode() == BinaryOperator::LOr) { // If we have "0 || X", simplify the code. "1 || X" would have constant // folded if the case was simple enough. if (ConstantFoldsToSimpleInteger(CondBOp->getLHS()) == -1) { // br(0 || X) -> br(X). return EmitBranchOnBoolExpr(CondBOp->getRHS(), TrueBlock, FalseBlock); } // If we have "X || 0", simplify the code to use an uncond branch. // "X || 1" would have been constant folded to 1. if (ConstantFoldsToSimpleInteger(CondBOp->getRHS()) == -1) { // br(X || 0) -> br(X). return EmitBranchOnBoolExpr(CondBOp->getLHS(), TrueBlock, FalseBlock); } // Emit the LHS as a conditional. If the LHS conditional is true, we // want to jump to the TrueBlock. llvm::BasicBlock *LHSFalse = createBasicBlock("lor.lhs.false"); EmitBranchOnBoolExpr(CondBOp->getLHS(), TrueBlock, LHSFalse); EmitBlock(LHSFalse); // Any temporaries created here are conditional. BeginConditionalBranch(); EmitBranchOnBoolExpr(CondBOp->getRHS(), TrueBlock, FalseBlock); EndConditionalBranch(); return; } } if (const UnaryOperator *CondUOp = dyn_cast(Cond)) { // br(!x, t, f) -> br(x, f, t) if (CondUOp->getOpcode() == UnaryOperator::LNot) return EmitBranchOnBoolExpr(CondUOp->getSubExpr(), FalseBlock, TrueBlock); } if (const ConditionalOperator *CondOp = dyn_cast(Cond)) { // Handle ?: operator. // Just ignore GNU ?: extension. if (CondOp->getLHS()) { // br(c ? x : y, t, f) -> br(c, br(x, t, f), br(y, t, f)) llvm::BasicBlock *LHSBlock = createBasicBlock("cond.true"); llvm::BasicBlock *RHSBlock = createBasicBlock("cond.false"); EmitBranchOnBoolExpr(CondOp->getCond(), LHSBlock, RHSBlock); EmitBlock(LHSBlock); EmitBranchOnBoolExpr(CondOp->getLHS(), TrueBlock, FalseBlock); EmitBlock(RHSBlock); EmitBranchOnBoolExpr(CondOp->getRHS(), TrueBlock, FalseBlock); return; } } // Emit the code with the fully general case. llvm::Value *CondV = EvaluateExprAsBool(Cond); Builder.CreateCondBr(CondV, TrueBlock, FalseBlock); } /// ErrorUnsupported - Print out an error that codegen doesn't support the /// specified stmt yet. void CodeGenFunction::ErrorUnsupported(const Stmt *S, const char *Type, bool OmitOnError) { CGM.ErrorUnsupported(S, Type, OmitOnError); } void CodeGenFunction::EmitNullInitialization(llvm::Value *DestPtr, QualType Ty) { // If the type contains a pointer to data member we can't memset it to zero. // Instead, create a null constant and copy it to the destination. if (CGM.getTypes().ContainsPointerToDataMember(Ty)) { llvm::Constant *NullConstant = CGM.EmitNullConstant(Ty); llvm::GlobalVariable *NullVariable = new llvm::GlobalVariable(CGM.getModule(), NullConstant->getType(), /*isConstant=*/true, llvm::GlobalVariable::PrivateLinkage, NullConstant, llvm::Twine()); EmitAggregateCopy(DestPtr, NullVariable, Ty, /*isVolatile=*/false); return; } // Ignore empty classes in C++. if (getContext().getLangOptions().CPlusPlus) { if (const RecordType *RT = Ty->getAs()) { if (cast(RT->getDecl())->isEmpty()) return; } } // Otherwise, just memset the whole thing to zero. This is legal // because in LLVM, all default initializers (other than the ones we just // handled above) are guaranteed to have a bit pattern of all zeros. const llvm::Type *BP = llvm::Type::getInt8PtrTy(VMContext); if (DestPtr->getType() != BP) DestPtr = Builder.CreateBitCast(DestPtr, BP, "tmp"); // Get size and alignment info for this aggregate. std::pair TypeInfo = getContext().getTypeInfo(Ty); // Don't bother emitting a zero-byte memset. if (TypeInfo.first == 0) return; // FIXME: Handle variable sized types. Builder.CreateCall5(CGM.getMemSetFn(BP, IntPtrTy), DestPtr, llvm::Constant::getNullValue(llvm::Type::getInt8Ty(VMContext)), // TypeInfo.first describes size in bits. llvm::ConstantInt::get(IntPtrTy, TypeInfo.first/8), llvm::ConstantInt::get(Int32Ty, TypeInfo.second/8), llvm::ConstantInt::get(llvm::Type::getInt1Ty(VMContext), 0)); } llvm::BlockAddress *CodeGenFunction::GetAddrOfLabel(const LabelStmt *L) { // Make sure that there is a block for the indirect goto. if (IndirectBranch == 0) GetIndirectGotoBlock(); llvm::BasicBlock *BB = getJumpDestForLabel(L).Block; // Make sure the indirect branch includes all of the address-taken blocks. IndirectBranch->addDestination(BB); return llvm::BlockAddress::get(CurFn, BB); } llvm::BasicBlock *CodeGenFunction::GetIndirectGotoBlock() { // If we already made the indirect branch for indirect goto, return its block. if (IndirectBranch) return IndirectBranch->getParent(); CGBuilderTy TmpBuilder(createBasicBlock("indirectgoto")); const llvm::Type *Int8PtrTy = llvm::Type::getInt8PtrTy(VMContext); // Create the PHI node that indirect gotos will add entries to. llvm::Value *DestVal = TmpBuilder.CreatePHI(Int8PtrTy, "indirect.goto.dest"); // Create the indirect branch instruction. IndirectBranch = TmpBuilder.CreateIndirectBr(DestVal); return IndirectBranch->getParent(); } llvm::Value *CodeGenFunction::GetVLASize(const VariableArrayType *VAT) { llvm::Value *&SizeEntry = VLASizeMap[VAT->getSizeExpr()]; assert(SizeEntry && "Did not emit size for type"); return SizeEntry; } llvm::Value *CodeGenFunction::EmitVLASize(QualType Ty) { assert(Ty->isVariablyModifiedType() && "Must pass variably modified type to EmitVLASizes!"); EnsureInsertPoint(); if (const VariableArrayType *VAT = getContext().getAsVariableArrayType(Ty)) { llvm::Value *&SizeEntry = VLASizeMap[VAT->getSizeExpr()]; if (!SizeEntry) { const llvm::Type *SizeTy = ConvertType(getContext().getSizeType()); // Get the element size; QualType ElemTy = VAT->getElementType(); llvm::Value *ElemSize; if (ElemTy->isVariableArrayType()) ElemSize = EmitVLASize(ElemTy); else ElemSize = llvm::ConstantInt::get(SizeTy, getContext().getTypeSizeInChars(ElemTy).getQuantity()); llvm::Value *NumElements = EmitScalarExpr(VAT->getSizeExpr()); NumElements = Builder.CreateIntCast(NumElements, SizeTy, false, "tmp"); SizeEntry = Builder.CreateMul(ElemSize, NumElements); } return SizeEntry; } if (const ArrayType *AT = dyn_cast(Ty)) { EmitVLASize(AT->getElementType()); return 0; } const PointerType *PT = Ty->getAs(); assert(PT && "unknown VM type!"); EmitVLASize(PT->getPointeeType()); return 0; } llvm::Value* CodeGenFunction::EmitVAListRef(const Expr* E) { if (CGM.getContext().getBuiltinVaListType()->isArrayType()) return EmitScalarExpr(E); return EmitLValue(E).getAddress(); } /// Pops cleanup blocks until the given savepoint is reached. void CodeGenFunction::PopCleanupBlocks(EHScopeStack::stable_iterator Old) { assert(Old.isValid()); EHScopeStack::iterator E = EHStack.find(Old); while (EHStack.begin() != E) PopCleanupBlock(); } /// Destroys a cleanup if it was unused. static void DestroyCleanup(CodeGenFunction &CGF, llvm::BasicBlock *Entry, llvm::BasicBlock *Exit) { assert(Entry->use_empty() && "destroying cleanup with uses!"); assert(Exit->getTerminator() == 0 && "exit has terminator but entry has no predecessors!"); // This doesn't always remove the entire cleanup, but it's much // safer as long as we don't know what blocks belong to the cleanup. // A *much* better approach if we care about this inefficiency would // be to lazily emit the cleanup. // If the exit block is distinct from the entry, give it a branch to // an unreachable destination. This preserves the well-formedness // of the IR. if (Entry != Exit) llvm::BranchInst::Create(CGF.getUnreachableBlock(), Exit); assert(!Entry->getParent() && "cleanup entry already positioned?"); // We can't just delete the entry; we have to kill any references to // its instructions in other blocks. for (llvm::BasicBlock::iterator I = Entry->begin(), E = Entry->end(); I != E; ++I) if (!I->use_empty()) I->replaceAllUsesWith(llvm::UndefValue::get(I->getType())); delete Entry; } /// Creates a switch instruction to thread branches out of the given /// block (which is the exit block of a cleanup). static void CreateCleanupSwitch(CodeGenFunction &CGF, llvm::BasicBlock *Block) { if (Block->getTerminator()) { assert(isa(Block->getTerminator()) && "cleanup block already has a terminator, but it isn't a switch"); return; } llvm::Value *DestCodePtr = CGF.CreateTempAlloca(CGF.Builder.getInt32Ty(), "cleanup.dst"); CGBuilderTy Builder(Block); llvm::Value *DestCode = Builder.CreateLoad(DestCodePtr, "tmp"); // Create a switch instruction to determine where to jump next. Builder.CreateSwitch(DestCode, CGF.getUnreachableBlock()); } /// Attempts to reduce a cleanup's entry block to a fallthrough. This /// is basically llvm::MergeBlockIntoPredecessor, except /// simplified/optimized for the tighter constraints on cleanup /// blocks. static void SimplifyCleanupEntry(CodeGenFunction &CGF, llvm::BasicBlock *Entry) { llvm::BasicBlock *Pred = Entry->getSinglePredecessor(); if (!Pred) return; llvm::BranchInst *Br = dyn_cast(Pred->getTerminator()); if (!Br || Br->isConditional()) return; assert(Br->getSuccessor(0) == Entry); // If we were previously inserting at the end of the cleanup entry // block, we'll need to continue inserting at the end of the // predecessor. bool WasInsertBlock = CGF.Builder.GetInsertBlock() == Entry; assert(!WasInsertBlock || CGF.Builder.GetInsertPoint() == Entry->end()); // Kill the branch. Br->eraseFromParent(); // Merge the blocks. Pred->getInstList().splice(Pred->end(), Entry->getInstList()); // Kill the entry block. Entry->eraseFromParent(); if (WasInsertBlock) CGF.Builder.SetInsertPoint(Pred); } /// Attempts to reduce an cleanup's exit switch to an unconditional /// branch. static void SimplifyCleanupExit(llvm::BasicBlock *Exit) { llvm::TerminatorInst *Terminator = Exit->getTerminator(); assert(Terminator && "completed cleanup exit has no terminator"); llvm::SwitchInst *Switch = dyn_cast(Terminator); if (!Switch) return; if (Switch->getNumCases() != 2) return; // default + 1 llvm::LoadInst *Cond = cast(Switch->getCondition()); llvm::AllocaInst *CondVar = cast(Cond->getPointerOperand()); // Replace the switch instruction with an unconditional branch. llvm::BasicBlock *Dest = Switch->getSuccessor(1); // default is 0 Switch->eraseFromParent(); llvm::BranchInst::Create(Dest, Exit); // Delete all uses of the condition variable. Cond->eraseFromParent(); while (!CondVar->use_empty()) cast(*CondVar->use_begin())->eraseFromParent(); // Delete the condition variable itself. CondVar->eraseFromParent(); } /// Threads a branch fixup through a cleanup block. static void ThreadFixupThroughCleanup(CodeGenFunction &CGF, BranchFixup &Fixup, llvm::BasicBlock *Entry, llvm::BasicBlock *Exit) { if (!Exit->getTerminator()) CreateCleanupSwitch(CGF, Exit); // Find the switch and its destination index alloca. llvm::SwitchInst *Switch = cast(Exit->getTerminator()); llvm::Value *DestCodePtr = cast(Switch->getCondition())->getPointerOperand(); // Compute the index of the new case we're adding to the switch. unsigned Index = Switch->getNumCases(); const llvm::IntegerType *i32 = llvm::Type::getInt32Ty(CGF.getLLVMContext()); llvm::ConstantInt *IndexV = llvm::ConstantInt::get(i32, Index); // Set the index in the origin block. new llvm::StoreInst(IndexV, DestCodePtr, Fixup.Origin); // Add a case to the switch. Switch->addCase(IndexV, Fixup.Destination); // Change the last branch to point to the cleanup entry block. Fixup.LatestBranch->setSuccessor(Fixup.LatestBranchIndex, Entry); // And finally, update the fixup. Fixup.LatestBranch = Switch; Fixup.LatestBranchIndex = Index; } /// Try to simplify both the entry and exit edges of a cleanup. static void SimplifyCleanupEdges(CodeGenFunction &CGF, llvm::BasicBlock *Entry, llvm::BasicBlock *Exit) { // Given their current implementations, it's important to run these // in this order: SimplifyCleanupEntry will delete Entry if it can // be merged into its predecessor, which will then break // SimplifyCleanupExit if (as is common) Entry == Exit. SimplifyCleanupExit(Exit); SimplifyCleanupEntry(CGF, Entry); } static void EmitLazyCleanup(CodeGenFunction &CGF, EHScopeStack::LazyCleanup *Fn, bool ForEH) { if (ForEH) CGF.EHStack.pushTerminate(); Fn->Emit(CGF, ForEH); if (ForEH) CGF.EHStack.popTerminate(); assert(CGF.HaveInsertPoint() && "cleanup ended with no insertion point?"); } static void SplitAndEmitLazyCleanup(CodeGenFunction &CGF, EHScopeStack::LazyCleanup *Fn, bool ForEH, llvm::BasicBlock *Entry) { assert(Entry && "no entry block for cleanup"); // Remove the switch and load from the end of the entry block. llvm::Instruction *Switch = &Entry->getInstList().back(); Entry->getInstList().remove(Switch); assert(isa(Switch)); llvm::Instruction *Load = &Entry->getInstList().back(); Entry->getInstList().remove(Load); assert(isa(Load)); assert(Entry->getInstList().empty() && "lazy cleanup block not empty after removing load/switch pair?"); // Emit the actual cleanup at the end of the entry block. CGF.Builder.SetInsertPoint(Entry); EmitLazyCleanup(CGF, Fn, ForEH); // Put the load and switch at the end of the exit block. llvm::BasicBlock *Exit = CGF.Builder.GetInsertBlock(); Exit->getInstList().push_back(Load); Exit->getInstList().push_back(Switch); // Clean up the edges if possible. SimplifyCleanupEdges(CGF, Entry, Exit); CGF.Builder.ClearInsertionPoint(); } static void PopLazyCleanupBlock(CodeGenFunction &CGF) { assert(isa(*CGF.EHStack.begin()) && "top not a cleanup!"); EHLazyCleanupScope &Scope = cast(*CGF.EHStack.begin()); assert(Scope.getFixupDepth() <= CGF.EHStack.getNumBranchFixups()); // Check whether we need an EH cleanup. This is only true if we've // generated a lazy EH cleanup block. llvm::BasicBlock *EHEntry = Scope.getEHBlock(); bool RequiresEHCleanup = (EHEntry != 0); // Check the three conditions which might require a normal cleanup: // - whether there are branch fix-ups through this cleanup unsigned FixupDepth = Scope.getFixupDepth(); bool HasFixups = CGF.EHStack.getNumBranchFixups() != FixupDepth; // - whether control has already been threaded through this cleanup llvm::BasicBlock *NormalEntry = Scope.getNormalBlock(); bool HasExistingBranches = (NormalEntry != 0); // - whether there's a fallthrough llvm::BasicBlock *FallthroughSource = CGF.Builder.GetInsertBlock(); bool HasFallthrough = (FallthroughSource != 0); bool RequiresNormalCleanup = false; if (Scope.isNormalCleanup() && (HasFixups || HasExistingBranches || HasFallthrough)) { RequiresNormalCleanup = true; } // If we don't need the cleanup at all, we're done. if (!RequiresNormalCleanup && !RequiresEHCleanup) { CGF.EHStack.popCleanup(); assert(CGF.EHStack.getNumBranchFixups() == 0 || CGF.EHStack.hasNormalCleanups()); return; } // Copy the cleanup emission data out. Note that SmallVector // guarantees maximal alignment for its buffer regardless of its // type parameter. llvm::SmallVector CleanupBuffer; CleanupBuffer.reserve(Scope.getCleanupSize()); memcpy(CleanupBuffer.data(), Scope.getCleanupBuffer(), Scope.getCleanupSize()); CleanupBuffer.set_size(Scope.getCleanupSize()); EHScopeStack::LazyCleanup *Fn = reinterpret_cast(CleanupBuffer.data()); // We're done with the scope; pop it off so we can emit the cleanups. CGF.EHStack.popCleanup(); if (RequiresNormalCleanup) { // If we have a fallthrough and no other need for the cleanup, // emit it directly. if (HasFallthrough && !HasFixups && !HasExistingBranches) { EmitLazyCleanup(CGF, Fn, /*ForEH*/ false); // Otherwise, the best approach is to thread everything through // the cleanup block and then try to clean up after ourselves. } else { // Force the entry block to exist. if (!HasExistingBranches) { NormalEntry = CGF.createBasicBlock("cleanup"); CreateCleanupSwitch(CGF, NormalEntry); } CGF.EmitBlock(NormalEntry); // Thread the fallthrough edge through the (momentarily trivial) // cleanup. llvm::BasicBlock *FallthroughDestination = 0; if (HasFallthrough) { assert(isa(FallthroughSource->getTerminator())); FallthroughDestination = CGF.createBasicBlock("cleanup.cont"); BranchFixup Fix; Fix.Destination = FallthroughDestination; Fix.LatestBranch = FallthroughSource->getTerminator(); Fix.LatestBranchIndex = 0; Fix.Origin = Fix.LatestBranch; // Restore fixup invariant. EmitBlock added a branch to the // cleanup which we need to redirect to the destination. cast(Fix.LatestBranch) ->setSuccessor(0, Fix.Destination); ThreadFixupThroughCleanup(CGF, Fix, NormalEntry, NormalEntry); } // Thread any "real" fixups we need to thread. for (unsigned I = FixupDepth, E = CGF.EHStack.getNumBranchFixups(); I != E; ++I) if (CGF.EHStack.getBranchFixup(I).Destination) ThreadFixupThroughCleanup(CGF, CGF.EHStack.getBranchFixup(I), NormalEntry, NormalEntry); SplitAndEmitLazyCleanup(CGF, Fn, /*ForEH*/ false, NormalEntry); if (HasFallthrough) CGF.EmitBlock(FallthroughDestination); } } // Emit the EH cleanup if required. if (RequiresEHCleanup) { CGBuilderTy::InsertPoint SavedIP = CGF.Builder.saveAndClearIP(); CGF.EmitBlock(EHEntry); SplitAndEmitLazyCleanup(CGF, Fn, /*ForEH*/ true, EHEntry); CGF.Builder.restoreIP(SavedIP); } } /// Pops a cleanup block. If the block includes a normal cleanup, the /// current insertion point is threaded through the cleanup, as are /// any branch fixups on the cleanup. void CodeGenFunction::PopCleanupBlock() { assert(!EHStack.empty() && "cleanup stack is empty!"); if (isa(*EHStack.begin())) return PopLazyCleanupBlock(*this); assert(isa(*EHStack.begin()) && "top not a cleanup!"); EHCleanupScope &Scope = cast(*EHStack.begin()); assert(Scope.getFixupDepth() <= EHStack.getNumBranchFixups()); // Handle the EH cleanup if (1) there is one and (2) it's different // from the normal cleanup. if (Scope.isEHCleanup() && Scope.getEHEntry() != Scope.getNormalEntry()) { llvm::BasicBlock *EHEntry = Scope.getEHEntry(); llvm::BasicBlock *EHExit = Scope.getEHExit(); if (EHEntry->use_empty()) { DestroyCleanup(*this, EHEntry, EHExit); } else { // TODO: this isn't really the ideal location to put this EH // cleanup, but lazy emission is a better solution than trying // to pick a better spot. CGBuilderTy::InsertPoint SavedIP = Builder.saveAndClearIP(); EmitBlock(EHEntry); Builder.restoreIP(SavedIP); SimplifyCleanupEdges(*this, EHEntry, EHExit); } } // If we only have an EH cleanup, we don't really need to do much // here. Branch fixups just naturally drop down to the enclosing // cleanup scope. if (!Scope.isNormalCleanup()) { EHStack.popCleanup(); assert(EHStack.getNumBranchFixups() == 0 || EHStack.hasNormalCleanups()); return; } // Check whether the scope has any fixups that need to be threaded. unsigned FixupDepth = Scope.getFixupDepth(); bool HasFixups = EHStack.getNumBranchFixups() != FixupDepth; // Grab the entry and exit blocks. llvm::BasicBlock *Entry = Scope.getNormalEntry(); llvm::BasicBlock *Exit = Scope.getNormalExit(); // Check whether anything's been threaded through the cleanup already. assert((Exit->getTerminator() == 0) == Entry->use_empty() && "cleanup entry/exit mismatch"); bool HasExistingBranches = !Entry->use_empty(); // Check whether we need to emit a "fallthrough" branch through the // cleanup for the current insertion point. llvm::BasicBlock *FallThrough = Builder.GetInsertBlock(); if (FallThrough && FallThrough->getTerminator()) FallThrough = 0; // If *nothing* is using the cleanup, kill it. if (!FallThrough && !HasFixups && !HasExistingBranches) { EHStack.popCleanup(); DestroyCleanup(*this, Entry, Exit); return; } // Otherwise, add the block to the function. EmitBlock(Entry); if (FallThrough) Builder.SetInsertPoint(Exit); else Builder.ClearInsertionPoint(); // Fast case: if we don't have to add any fixups, and either // we don't have a fallthrough or the cleanup wasn't previously // used, then the setup above is sufficient. if (!HasFixups) { if (!FallThrough) { assert(HasExistingBranches && "no reason for cleanup but didn't kill before"); EHStack.popCleanup(); SimplifyCleanupEdges(*this, Entry, Exit); return; } else if (!HasExistingBranches) { assert(FallThrough && "no reason for cleanup but didn't kill before"); // We can't simplify the exit edge in this case because we're // already inserting at the end of the exit block. EHStack.popCleanup(); SimplifyCleanupEntry(*this, Entry); return; } } // Otherwise we're going to have to thread things through the cleanup. llvm::SmallVector Fixups; // Synthesize a fixup for the current insertion point. BranchFixup Cur; if (FallThrough) { Cur.Destination = createBasicBlock("cleanup.cont"); Cur.LatestBranch = FallThrough->getTerminator(); Cur.LatestBranchIndex = 0; Cur.Origin = Cur.LatestBranch; // Restore fixup invariant. EmitBlock added a branch to the cleanup // which we need to redirect to the destination. cast(Cur.LatestBranch)->setSuccessor(0, Cur.Destination); Fixups.push_back(&Cur); } else { Cur.Destination = 0; } // Collect any "real" fixups we need to thread. for (unsigned I = FixupDepth, E = EHStack.getNumBranchFixups(); I != E; ++I) if (EHStack.getBranchFixup(I).Destination) Fixups.push_back(&EHStack.getBranchFixup(I)); assert(!Fixups.empty() && "no fixups, invariants broken!"); // If there's only a single fixup to thread through, do so with // unconditional branches. This only happens if there's a single // branch and no fallthrough. if (Fixups.size() == 1 && !HasExistingBranches) { Fixups[0]->LatestBranch->setSuccessor(Fixups[0]->LatestBranchIndex, Entry); llvm::BranchInst *Br = llvm::BranchInst::Create(Fixups[0]->Destination, Exit); Fixups[0]->LatestBranch = Br; Fixups[0]->LatestBranchIndex = 0; // Otherwise, force a switch statement and thread everything through // the switch. } else { CreateCleanupSwitch(*this, Exit); for (unsigned I = 0, E = Fixups.size(); I != E; ++I) ThreadFixupThroughCleanup(*this, *Fixups[I], Entry, Exit); } // Emit the fallthrough destination block if necessary. if (Cur.Destination) EmitBlock(Cur.Destination); // We're finally done with the cleanup. EHStack.popCleanup(); } void CodeGenFunction::EmitBranchThroughCleanup(JumpDest Dest) { if (!HaveInsertPoint()) return; // Create the branch. llvm::BranchInst *BI = Builder.CreateBr(Dest.Block); // If we're not in a cleanup scope, we don't need to worry about // fixups. if (!EHStack.hasNormalCleanups()) { Builder.ClearInsertionPoint(); return; } // Initialize a fixup. BranchFixup Fixup; Fixup.Destination = Dest.Block; Fixup.Origin = BI; Fixup.LatestBranch = BI; Fixup.LatestBranchIndex = 0; // If we can't resolve the destination cleanup scope, just add this // to the current cleanup scope. if (!Dest.ScopeDepth.isValid()) { EHStack.addBranchFixup() = Fixup; Builder.ClearInsertionPoint(); return; } for (EHScopeStack::iterator I = EHStack.begin(), E = EHStack.find(Dest.ScopeDepth); I != E; ++I) { if (isa(*I)) { EHCleanupScope &Scope = cast(*I); if (Scope.isNormalCleanup()) ThreadFixupThroughCleanup(*this, Fixup, Scope.getNormalEntry(), Scope.getNormalExit()); } else if (isa(*I)) { EHLazyCleanupScope &Scope = cast(*I); if (Scope.isNormalCleanup()) { llvm::BasicBlock *Block = Scope.getNormalBlock(); if (!Block) { Block = createBasicBlock("cleanup"); Scope.setNormalBlock(Block); } ThreadFixupThroughCleanup(*this, Fixup, Block, Block); } } } Builder.ClearInsertionPoint(); } void CodeGenFunction::EmitBranchThroughEHCleanup(JumpDest Dest) { if (!HaveInsertPoint()) return; // Create the branch. llvm::BranchInst *BI = Builder.CreateBr(Dest.Block); // If we're not in a cleanup scope, we don't need to worry about // fixups. if (!EHStack.hasEHCleanups()) { Builder.ClearInsertionPoint(); return; } // Initialize a fixup. BranchFixup Fixup; Fixup.Destination = Dest.Block; Fixup.Origin = BI; Fixup.LatestBranch = BI; Fixup.LatestBranchIndex = 0; // We should never get invalid scope depths for these: invalid scope // depths only arise for as-yet-unemitted labels, and we can't do an // EH-unwind to one of those. assert(Dest.ScopeDepth.isValid() && "invalid scope depth on EH dest?"); for (EHScopeStack::iterator I = EHStack.begin(), E = EHStack.find(Dest.ScopeDepth); I != E; ++I) { if (isa(*I)) { EHCleanupScope &Scope = cast(*I); if (Scope.isEHCleanup()) ThreadFixupThroughCleanup(*this, Fixup, Scope.getEHEntry(), Scope.getEHExit()); } else if (isa(*I)) { EHLazyCleanupScope &Scope = cast(*I); if (Scope.isEHCleanup()) { llvm::BasicBlock *Block = Scope.getEHBlock(); if (!Block) { Block = createBasicBlock("eh.cleanup"); Scope.setEHBlock(Block); } ThreadFixupThroughCleanup(*this, Fixup, Block, Block); } } } Builder.ClearInsertionPoint(); }