X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FTransforms%2FUtils%2FCloneFunction.cpp;h=995d6e4baa754a38a3d07a16b7eaee5d82c4b986;hp=f05085fca1639da1c9803fe2a66fefa7c9a5d7a5;hb=da92e119096b6e2e93efbf44ed07ea0a715cef9c;hpb=dc024674ff96820d6020757b48d47f46d4c07db2 diff --git a/lib/Transforms/Utils/CloneFunction.cpp b/lib/Transforms/Utils/CloneFunction.cpp index f05085fca16..6454afb8bc4 100644 --- a/lib/Transforms/Utils/CloneFunction.cpp +++ b/lib/Transforms/Utils/CloneFunction.cpp @@ -2,8 +2,8 @@ // // The LLVM Compiler Infrastructure // -// This file was developed by the LLVM research group and is distributed under -// the University of Illinois Open Source License. See LICENSE.TXT for details. +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // @@ -14,24 +14,33 @@ //===----------------------------------------------------------------------===// #include "llvm/Transforms/Utils/Cloning.h" -#include "llvm/Constants.h" -#include "llvm/DerivedTypes.h" -#include "llvm/Instructions.h" -#include "llvm/Function.h" -#include "llvm/Support/CFG.h" -#include "llvm/Support/Compiler.h" -#include "llvm/Transforms/Utils/ValueMapper.h" -#include "llvm/Analysis/ConstantFolding.h" #include "llvm/ADT/SmallVector.h" +#include "llvm/Analysis/ConstantFolding.h" +#include "llvm/Analysis/InstructionSimplify.h" +#include "llvm/Analysis/LoopInfo.h" +#include "llvm/IR/CFG.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DebugInfo.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/Function.h" +#include "llvm/IR/GlobalVariable.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/LLVMContext.h" +#include "llvm/IR/Metadata.h" +#include "llvm/IR/Module.h" +#include "llvm/Transforms/Utils/BasicBlockUtils.h" +#include "llvm/Transforms/Utils/Local.h" +#include "llvm/Transforms/Utils/ValueMapper.h" #include using namespace llvm; -// CloneBasicBlock - See comments in Cloning.h +/// See comments in Cloning.h. BasicBlock *llvm::CloneBasicBlock(const BasicBlock *BB, - DenseMap &ValueMap, - const char *NameSuffix, Function *F, + ValueToValueMapTy &VMap, + const Twine &NameSuffix, Function *F, ClonedCodeInfo *CodeInfo) { - BasicBlock *NewBB = new BasicBlock("", F); + BasicBlock *NewBB = BasicBlock::Create(BB->getContext(), "", F); if (BB->hasName()) NewBB->setName(BB->getName()+NameSuffix); bool hasCalls = false, hasDynamicAllocas = false, hasStaticAllocas = false; @@ -43,9 +52,9 @@ BasicBlock *llvm::CloneBasicBlock(const BasicBlock *BB, if (II->hasName()) NewInst->setName(II->getName()+NameSuffix); NewBB->getInstList().push_back(NewInst); - ValueMap[II] = NewInst; // Add instruction map to value. - - hasCalls |= isa(II); + VMap[&*II] = NewInst; // Add instruction map to value. + + hasCalls |= (isa(II) && !isa(II)); if (const AllocaInst *AI = dyn_cast(II)) { if (isa(AI->getArraySize())) hasStaticAllocas = true; @@ -56,7 +65,6 @@ BasicBlock *llvm::CloneBasicBlock(const BasicBlock *BB, if (CodeInfo) { CodeInfo->ContainsCalls |= hasCalls; - CodeInfo->ContainsUnwinds |= isa(BB->getTerminator()); CodeInfo->ContainsDynamicAllocas |= hasDynamicAllocas; CodeInfo->ContainsDynamicAllocas |= hasStaticAllocas && BB != &BB->getParent()->getEntryBlock(); @@ -65,22 +73,51 @@ BasicBlock *llvm::CloneBasicBlock(const BasicBlock *BB, } // Clone OldFunc into NewFunc, transforming the old arguments into references to -// ArgMap values. +// VMap values. // void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc, - DenseMap &ValueMap, - std::vector &Returns, - const char *NameSuffix, ClonedCodeInfo *CodeInfo) { + ValueToValueMapTy &VMap, + bool ModuleLevelChanges, + SmallVectorImpl &Returns, + const char *NameSuffix, ClonedCodeInfo *CodeInfo, + ValueMapTypeRemapper *TypeMapper, + ValueMaterializer *Materializer) { assert(NameSuffix && "NameSuffix cannot be null!"); #ifndef NDEBUG - for (Function::const_arg_iterator I = OldFunc->arg_begin(), - E = OldFunc->arg_end(); I != E; ++I) - assert(ValueMap.count(I) && "No mapping from source argument specified!"); + for (const Argument &I : OldFunc->args()) + assert(VMap.count(&I) && "No mapping from source argument specified!"); #endif - // Clone the parameter attributes - NewFunc->setParamAttrs(OldFunc->getParamAttrs()); + // Copy all attributes other than those stored in the AttributeSet. We need + // to remap the parameter indices of the AttributeSet. + AttributeSet NewAttrs = NewFunc->getAttributes(); + NewFunc->copyAttributesFrom(OldFunc); + NewFunc->setAttributes(NewAttrs); + + // Fix up the personality function that got copied over. + if (OldFunc->hasPersonalityFn()) + NewFunc->setPersonalityFn( + MapValue(OldFunc->getPersonalityFn(), VMap, + ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges, + TypeMapper, Materializer)); + + AttributeSet OldAttrs = OldFunc->getAttributes(); + // Clone any argument attributes that are present in the VMap. + for (const Argument &OldArg : OldFunc->args()) + if (Argument *NewArg = dyn_cast(VMap[&OldArg])) { + AttributeSet attrs = + OldAttrs.getParamAttributes(OldArg.getArgNo() + 1); + if (attrs.getNumSlots() > 0) + NewArg->addAttr(attrs); + } + + NewFunc->setAttributes( + NewFunc->getAttributes() + .addAttributes(NewFunc->getContext(), AttributeSet::ReturnIndex, + OldAttrs.getRetAttributes()) + .addAttributes(NewFunc->getContext(), AttributeSet::FunctionIndex, + OldAttrs.getFnAttributes())); // Loop over all of the basic blocks in the function, cloning them as // appropriate. Note that we save BE this way in order to handle cloning of @@ -91,137 +128,233 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc, const BasicBlock &BB = *BI; // Create a new basic block and copy instructions into it! - BasicBlock *CBB = CloneBasicBlock(&BB, ValueMap, NameSuffix, NewFunc, - CodeInfo); - ValueMap[&BB] = CBB; // Add basic block mapping. + BasicBlock *CBB = CloneBasicBlock(&BB, VMap, NameSuffix, NewFunc, CodeInfo); + + // Add basic block mapping. + VMap[&BB] = CBB; + + // It is only legal to clone a function if a block address within that + // function is never referenced outside of the function. Given that, we + // want to map block addresses from the old function to block addresses in + // the clone. (This is different from the generic ValueMapper + // implementation, which generates an invalid blockaddress when + // cloning a function.) + if (BB.hasAddressTaken()) { + Constant *OldBBAddr = BlockAddress::get(const_cast(OldFunc), + const_cast(&BB)); + VMap[OldBBAddr] = BlockAddress::get(NewFunc, CBB); + } + // Note return instructions for the caller. if (ReturnInst *RI = dyn_cast(CBB->getTerminator())) Returns.push_back(RI); } // Loop over all of the instructions in the function, fixing up operand - // references as we go. This uses ValueMap to do all the hard work. - // - for (Function::iterator BB = cast(ValueMap[OldFunc->begin()]), - BE = NewFunc->end(); BB != BE; ++BB) + // references as we go. This uses VMap to do all the hard work. + for (Function::iterator BB = + cast(VMap[&OldFunc->front()])->getIterator(), + BE = NewFunc->end(); + BB != BE; ++BB) // Loop over all instructions, fixing each one as we find it... - for (BasicBlock::iterator II = BB->begin(); II != BB->end(); ++II) - RemapInstruction(II, ValueMap); + for (Instruction &II : *BB) + RemapInstruction(&II, VMap, + ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges, + TypeMapper, Materializer); +} + +// Find the MDNode which corresponds to the subprogram data that described F. +static DISubprogram *FindSubprogram(const Function *F, + DebugInfoFinder &Finder) { + for (DISubprogram *Subprogram : Finder.subprograms()) { + if (Subprogram->describes(F)) + return Subprogram; + } + return nullptr; +} + +// Add an operand to an existing MDNode. The new operand will be added at the +// back of the operand list. +static void AddOperand(DICompileUnit *CU, DISubprogramArray SPs, + Metadata *NewSP) { + SmallVector NewSPs; + NewSPs.reserve(SPs.size() + 1); + for (auto *SP : SPs) + NewSPs.push_back(SP); + NewSPs.push_back(NewSP); + CU->replaceSubprograms(MDTuple::get(CU->getContext(), NewSPs)); } -/// CloneFunction - Return a copy of the specified function, but without +// Clone the module-level debug info associated with OldFunc. The cloned data +// will point to NewFunc instead. +static void CloneDebugInfoMetadata(Function *NewFunc, const Function *OldFunc, + ValueToValueMapTy &VMap) { + DebugInfoFinder Finder; + Finder.processModule(*OldFunc->getParent()); + + const DISubprogram *OldSubprogramMDNode = FindSubprogram(OldFunc, Finder); + if (!OldSubprogramMDNode) return; + + auto *NewSubprogram = + cast(MapMetadata(OldSubprogramMDNode, VMap)); + NewFunc->setSubprogram(NewSubprogram); + + for (auto *CU : Finder.compile_units()) { + auto Subprograms = CU->getSubprograms(); + // If the compile unit's function list contains the old function, it should + // also contain the new one. + for (auto *SP : Subprograms) { + if (SP == OldSubprogramMDNode) { + AddOperand(CU, Subprograms, NewSubprogram); + break; + } + } + } +} + +/// Return a copy of the specified function, but without /// embedding the function into another module. Also, any references specified -/// in the ValueMap are changed to refer to their mapped value instead of the -/// original one. If any of the arguments to the function are in the ValueMap, -/// the arguments are deleted from the resultant function. The ValueMap is +/// in the VMap are changed to refer to their mapped value instead of the +/// original one. If any of the arguments to the function are in the VMap, +/// the arguments are deleted from the resultant function. The VMap is /// updated to include mappings from all of the instructions and basicblocks in /// the function from their old to new values. /// -Function *llvm::CloneFunction(const Function *F, - DenseMap &ValueMap, +Function *llvm::CloneFunction(const Function *F, ValueToValueMapTy &VMap, + bool ModuleLevelChanges, ClonedCodeInfo *CodeInfo) { - std::vector ArgTypes; + std::vector ArgTypes; // The user might be deleting arguments to the function by specifying them in - // the ValueMap. If so, we need to not add the arguments to the arg ty vector + // the VMap. If so, we need to not add the arguments to the arg ty vector // - for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end(); - I != E; ++I) - if (ValueMap.count(I) == 0) // Haven't mapped the argument to anything yet? - ArgTypes.push_back(I->getType()); + for (const Argument &I : F->args()) + if (VMap.count(&I) == 0) // Haven't mapped the argument to anything yet? + ArgTypes.push_back(I.getType()); // Create a new function type... FunctionType *FTy = FunctionType::get(F->getFunctionType()->getReturnType(), ArgTypes, F->getFunctionType()->isVarArg()); // Create the new function... - Function *NewF = new Function(FTy, F->getLinkage(), F->getName()); + Function *NewF = Function::Create(FTy, F->getLinkage(), F->getName()); // Loop over the arguments, copying the names of the mapped arguments over... Function::arg_iterator DestI = NewF->arg_begin(); - for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end(); - I != E; ++I) - if (ValueMap.count(I) == 0) { // Is this argument preserved? - DestI->setName(I->getName()); // Copy the name over... - ValueMap[I] = DestI++; // Add mapping to ValueMap + for (const Argument & I : F->args()) + if (VMap.count(&I) == 0) { // Is this argument preserved? + DestI->setName(I.getName()); // Copy the name over... + VMap[&I] = &*DestI++; // Add mapping to VMap } - std::vector Returns; // Ignore returns cloned... - CloneFunctionInto(NewF, F, ValueMap, Returns, "", CodeInfo); + if (ModuleLevelChanges) + CloneDebugInfoMetadata(NewF, F, VMap); + + SmallVector Returns; // Ignore returns cloned. + CloneFunctionInto(NewF, F, VMap, ModuleLevelChanges, Returns, "", CodeInfo); return NewF; } namespace { - /// PruningFunctionCloner - This class is a private class used to implement - /// the CloneAndPruneFunctionInto method. - struct VISIBILITY_HIDDEN PruningFunctionCloner { + /// This is a private class used to implement CloneAndPruneFunctionInto. + struct PruningFunctionCloner { Function *NewFunc; const Function *OldFunc; - DenseMap &ValueMap; - std::vector &Returns; + ValueToValueMapTy &VMap; + bool ModuleLevelChanges; const char *NameSuffix; ClonedCodeInfo *CodeInfo; - const TargetData *TD; public: PruningFunctionCloner(Function *newFunc, const Function *oldFunc, - DenseMap &valueMap, - std::vector &returns, - const char *nameSuffix, - ClonedCodeInfo *codeInfo, - const TargetData *td) - : NewFunc(newFunc), OldFunc(oldFunc), ValueMap(valueMap), Returns(returns), - NameSuffix(nameSuffix), CodeInfo(codeInfo), TD(td) { - } + ValueToValueMapTy &valueMap, bool moduleLevelChanges, + const char *nameSuffix, ClonedCodeInfo *codeInfo) + : NewFunc(newFunc), OldFunc(oldFunc), VMap(valueMap), + ModuleLevelChanges(moduleLevelChanges), NameSuffix(nameSuffix), + CodeInfo(codeInfo) {} - /// CloneBlock - The specified block is found to be reachable, clone it and + /// The specified block is found to be reachable, clone it and /// anything that it can reach. - void CloneBlock(const BasicBlock *BB, + void CloneBlock(const BasicBlock *BB, + BasicBlock::const_iterator StartingInst, std::vector &ToClone); - - public: - /// ConstantFoldMappedInstruction - Constant fold the specified instruction, - /// mapping its operands through ValueMap if they are available. - Constant *ConstantFoldMappedInstruction(const Instruction *I); }; } -/// CloneBlock - The specified block is found to be reachable, clone it and +/// The specified block is found to be reachable, clone it and /// anything that it can reach. void PruningFunctionCloner::CloneBlock(const BasicBlock *BB, + BasicBlock::const_iterator StartingInst, std::vector &ToClone){ - Value *&BBEntry = ValueMap[BB]; + WeakVH &BBEntry = VMap[BB]; // Have we already cloned this block? if (BBEntry) return; // Nope, clone it now. BasicBlock *NewBB; - BBEntry = NewBB = new BasicBlock(); + BBEntry = NewBB = BasicBlock::Create(BB->getContext()); if (BB->hasName()) NewBB->setName(BB->getName()+NameSuffix); + // It is only legal to clone a function if a block address within that + // function is never referenced outside of the function. Given that, we + // want to map block addresses from the old function to block addresses in + // the clone. (This is different from the generic ValueMapper + // implementation, which generates an invalid blockaddress when + // cloning a function.) + // + // Note that we don't need to fix the mapping for unreachable blocks; + // the default mapping there is safe. + if (BB->hasAddressTaken()) { + Constant *OldBBAddr = BlockAddress::get(const_cast(OldFunc), + const_cast(BB)); + VMap[OldBBAddr] = BlockAddress::get(NewFunc, NewBB); + } + bool hasCalls = false, hasDynamicAllocas = false, hasStaticAllocas = false; - + // Loop over all instructions, and copy them over, DCE'ing as we go. This // loop doesn't include the terminator. - for (BasicBlock::const_iterator II = BB->begin(), IE = --BB->end(); + for (BasicBlock::const_iterator II = StartingInst, IE = --BB->end(); II != IE; ++II) { - // If this instruction constant folds, don't bother cloning the instruction, - // instead, just add the constant to the value map. - if (Constant *C = ConstantFoldMappedInstruction(II)) { - ValueMap[II] = C; - continue; - } - + Instruction *NewInst = II->clone(); + + // Eagerly remap operands to the newly cloned instruction, except for PHI + // nodes for which we defer processing until we update the CFG. + if (!isa(NewInst)) { + RemapInstruction(NewInst, VMap, + ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges); + + // If we can simplify this instruction to some other value, simply add + // a mapping to that value rather than inserting a new instruction into + // the basic block. + if (Value *V = + SimplifyInstruction(NewInst, BB->getModule()->getDataLayout())) { + // On the off-chance that this simplifies to an instruction in the old + // function, map it back into the new function. + if (Value *MappedV = VMap.lookup(V)) + V = MappedV; + + VMap[&*II] = V; + delete NewInst; + continue; + } + } + if (II->hasName()) NewInst->setName(II->getName()+NameSuffix); + VMap[&*II] = NewInst; // Add instruction map to value. NewBB->getInstList().push_back(NewInst); - ValueMap[II] = NewInst; // Add instruction map to value. - - hasCalls |= isa(II); + hasCalls |= (isa(II) && !isa(II)); + + if (CodeInfo) + if (auto CS = ImmutableCallSite(&*II)) + if (CS.hasOperandBundles()) + CodeInfo->OperandBundleCallSites.push_back(NewInst); + if (const AllocaInst *AI = dyn_cast(II)) { if (isa(AI->getArraySize())) hasStaticAllocas = true; @@ -238,13 +371,15 @@ void PruningFunctionCloner::CloneBlock(const BasicBlock *BB, // If the condition was a known constant in the callee... ConstantInt *Cond = dyn_cast(BI->getCondition()); // Or is a known constant in the caller... - if (Cond == 0) - Cond = dyn_cast_or_null(ValueMap[BI->getCondition()]); + if (!Cond) { + Value *V = VMap[BI->getCondition()]; + Cond = dyn_cast_or_null(V); + } // Constant fold to uncond branch! if (Cond) { BasicBlock *Dest = BI->getSuccessor(!Cond->getZExtValue()); - ValueMap[OldTI] = new BranchInst(Dest, NewBB); + VMap[OldTI] = BranchInst::Create(Dest, NewBB); ToClone.push_back(Dest); TerminatorDone = true; } @@ -252,11 +387,14 @@ void PruningFunctionCloner::CloneBlock(const BasicBlock *BB, } else if (const SwitchInst *SI = dyn_cast(OldTI)) { // If switching on a value known constant in the caller. ConstantInt *Cond = dyn_cast(SI->getCondition()); - if (Cond == 0) // Or known constant after constant prop in the callee... - Cond = dyn_cast_or_null(ValueMap[SI->getCondition()]); + if (!Cond) { // Or known constant after constant prop in the callee... + Value *V = VMap[SI->getCondition()]; + Cond = dyn_cast_or_null(V); + } if (Cond) { // Constant fold to uncond branch! - BasicBlock *Dest = SI->getSuccessor(SI->findCaseValue(Cond)); - ValueMap[OldTI] = new BranchInst(Dest, NewBB); + SwitchInst::ConstCaseIt Case = SI->findCaseValue(Cond); + BasicBlock *Dest = const_cast(Case.getCaseSuccessor()); + VMap[OldTI] = BranchInst::Create(Dest, NewBB); ToClone.push_back(Dest); TerminatorDone = true; } @@ -267,72 +405,67 @@ void PruningFunctionCloner::CloneBlock(const BasicBlock *BB, if (OldTI->hasName()) NewInst->setName(OldTI->getName()+NameSuffix); NewBB->getInstList().push_back(NewInst); - ValueMap[OldTI] = NewInst; // Add instruction map to value. - + VMap[OldTI] = NewInst; // Add instruction map to value. + + if (CodeInfo) + if (auto CS = ImmutableCallSite(OldTI)) + if (CS.hasOperandBundles()) + CodeInfo->OperandBundleCallSites.push_back(NewInst); + // Recursively clone any reachable successor blocks. const TerminatorInst *TI = BB->getTerminator(); - for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i) - ToClone.push_back(TI->getSuccessor(i)); + for (const BasicBlock *Succ : TI->successors()) + ToClone.push_back(Succ); } if (CodeInfo) { CodeInfo->ContainsCalls |= hasCalls; - CodeInfo->ContainsUnwinds |= isa(OldTI); CodeInfo->ContainsDynamicAllocas |= hasDynamicAllocas; CodeInfo->ContainsDynamicAllocas |= hasStaticAllocas && BB != &BB->getParent()->front(); } - - if (ReturnInst *RI = dyn_cast(NewBB->getTerminator())) - Returns.push_back(RI); } -/// ConstantFoldMappedInstruction - Constant fold the specified instruction, -/// mapping its operands through ValueMap if they are available. -Constant *PruningFunctionCloner:: -ConstantFoldMappedInstruction(const Instruction *I) { - SmallVector Ops; - for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) - if (Constant *Op = dyn_cast_or_null(MapValue(I->getOperand(i), - ValueMap))) - Ops.push_back(Op); - else - return 0; // All operands not constant! - - return ConstantFoldInstOperands(I, &Ops[0], Ops.size(), TD); -} - -/// CloneAndPruneFunctionInto - This works exactly like CloneFunctionInto, -/// except that it does some simple constant prop and DCE on the fly. The -/// effect of this is to copy significantly less code in cases where (for -/// example) a function call with constant arguments is inlined, and those -/// constant arguments cause a significant amount of code in the callee to be -/// dead. Since this doesn't produce an exact copy of the input, it can't be -/// used for things like CloneFunction or CloneModule. -void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, - DenseMap &ValueMap, - std::vector &Returns, - const char *NameSuffix, - ClonedCodeInfo *CodeInfo, - const TargetData *TD) { +/// This works like CloneAndPruneFunctionInto, except that it does not clone the +/// entire function. Instead it starts at an instruction provided by the caller +/// and copies (and prunes) only the code reachable from that instruction. +void llvm::CloneAndPruneIntoFromInst(Function *NewFunc, const Function *OldFunc, + const Instruction *StartingInst, + ValueToValueMapTy &VMap, + bool ModuleLevelChanges, + SmallVectorImpl &Returns, + const char *NameSuffix, + ClonedCodeInfo *CodeInfo) { assert(NameSuffix && "NameSuffix cannot be null!"); - + + ValueMapTypeRemapper *TypeMapper = nullptr; + ValueMaterializer *Materializer = nullptr; + #ifndef NDEBUG - for (Function::const_arg_iterator II = OldFunc->arg_begin(), - E = OldFunc->arg_end(); II != E; ++II) - assert(ValueMap.count(II) && "No mapping from source argument specified!"); + // If the cloning starts at the beginning of the function, verify that + // the function arguments are mapped. + if (!StartingInst) + for (const Argument &II : OldFunc->args()) + assert(VMap.count(&II) && "No mapping from source argument specified!"); #endif - - PruningFunctionCloner PFC(NewFunc, OldFunc, ValueMap, Returns, - NameSuffix, CodeInfo, TD); + + PruningFunctionCloner PFC(NewFunc, OldFunc, VMap, ModuleLevelChanges, + NameSuffix, CodeInfo); + const BasicBlock *StartingBB; + if (StartingInst) + StartingBB = StartingInst->getParent(); + else { + StartingBB = &OldFunc->getEntryBlock(); + StartingInst = &StartingBB->front(); + } // Clone the entry block, and anything recursively reachable from it. std::vector CloneWorklist; - CloneWorklist.push_back(&OldFunc->getEntryBlock()); + PFC.CloneBlock(StartingBB, StartingInst->getIterator(), CloneWorklist); while (!CloneWorklist.empty()) { const BasicBlock *BB = CloneWorklist.back(); CloneWorklist.pop_back(); - PFC.CloneBlock(BB, CloneWorklist); + PFC.CloneBlock(BB, BB->begin(), CloneWorklist); } // Loop over all of the basic blocks in the old function. If the block was @@ -340,32 +473,35 @@ void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, // insert it into the new function in the right order. If not, ignore it. // // Defer PHI resolution until rest of function is resolved. - std::vector PHIToResolve; - for (Function::const_iterator BI = OldFunc->begin(), BE = OldFunc->end(); - BI != BE; ++BI) { - BasicBlock *NewBB = cast_or_null(ValueMap[BI]); - if (NewBB == 0) continue; // Dead block. + SmallVector PHIToResolve; + for (const BasicBlock &BI : *OldFunc) { + Value *V = VMap[&BI]; + BasicBlock *NewBB = cast_or_null(V); + if (!NewBB) continue; // Dead block. // Add the new block to the new function. NewFunc->getBasicBlockList().push_back(NewBB); - - // Loop over all of the instructions in the block, fixing up operand - // references as we go. This uses ValueMap to do all the hard work. - // - BasicBlock::iterator I = NewBB->begin(); - + // Handle PHI nodes specially, as we have to remove references to dead // blocks. - if (PHINode *PN = dyn_cast(I)) { - // Skip over all PHI nodes, remembering them for later. - BasicBlock::const_iterator OldI = BI->begin(); - for (; (PN = dyn_cast(I)); ++I, ++OldI) - PHIToResolve.push_back(cast(OldI)); + for (BasicBlock::const_iterator I = BI.begin(), E = BI.end(); I != E; ++I) { + // PHI nodes may have been remapped to non-PHI nodes by the caller or + // during the cloning process. + if (const PHINode *PN = dyn_cast(I)) { + if (isa(VMap[PN])) + PHIToResolve.push_back(PN); + else + break; + } else { + break; + } } - - // Otherwise, remap the rest of the instructions normally. - for (; I != NewBB->end(); ++I) - RemapInstruction(I, ValueMap); + + // Finally, remap the terminator instructions, as those can't be remapped + // until all BBs are mapped. + RemapInstruction(NewBB->getTerminator(), VMap, + ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges, + TypeMapper, Materializer); } // Defer PHI resolution until rest of function is resolved, PHI resolution @@ -374,18 +510,20 @@ void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, const PHINode *OPN = PHIToResolve[phino]; unsigned NumPreds = OPN->getNumIncomingValues(); const BasicBlock *OldBB = OPN->getParent(); - BasicBlock *NewBB = cast(ValueMap[OldBB]); + BasicBlock *NewBB = cast(VMap[OldBB]); // Map operands for blocks that are live and remove operands for blocks // that are dead. for (; phino != PHIToResolve.size() && PHIToResolve[phino]->getParent() == OldBB; ++phino) { OPN = PHIToResolve[phino]; - PHINode *PN = cast(ValueMap[OPN]); + PHINode *PN = cast(VMap[OPN]); for (unsigned pred = 0, e = NumPreds; pred != e; ++pred) { - if (BasicBlock *MappedBlock = - cast_or_null(ValueMap[PN->getIncomingBlock(pred)])) { - Value *InVal = MapValue(PN->getIncomingValue(pred), ValueMap); + Value *V = VMap[PN->getIncomingBlock(pred)]; + if (BasicBlock *MappedBlock = cast_or_null(V)) { + Value *InVal = MapValue(PN->getIncomingValue(pred), + VMap, + ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges); assert(InVal && "Unknown input value?"); PN->setIncomingValue(pred, InVal); PN->setIncomingBlock(pred, MappedBlock); @@ -439,50 +577,168 @@ void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, while ((PN = dyn_cast(I++))) { Value *NV = UndefValue::get(PN->getType()); PN->replaceAllUsesWith(NV); - assert(ValueMap[OldI] == PN && "ValueMap mismatch"); - ValueMap[OldI] = NV; + assert(VMap[&*OldI] == PN && "VMap mismatch"); + VMap[&*OldI] = NV; PN->eraseFromParent(); ++OldI; } } - // NOTE: We cannot eliminate single entry phi nodes here, because of - // ValueMap. Single entry phi nodes can have multiple ValueMap entries - // pointing at them. Thus, deleting one would require scanning the ValueMap - // to update any entries in it that would require that. This would be - // really slow. } - + + // Make a second pass over the PHINodes now that all of them have been + // remapped into the new function, simplifying the PHINode and performing any + // recursive simplifications exposed. This will transparently update the + // WeakVH in the VMap. Notably, we rely on that so that if we coalesce + // two PHINodes, the iteration over the old PHIs remains valid, and the + // mapping will just map us to the new node (which may not even be a PHI + // node). + for (unsigned Idx = 0, Size = PHIToResolve.size(); Idx != Size; ++Idx) + if (PHINode *PN = dyn_cast(VMap[PHIToResolve[Idx]])) + recursivelySimplifyInstruction(PN); + // Now that the inlined function body has been fully constructed, go through - // and zap unconditional fall-through branches. This happen all the time when + // and zap unconditional fall-through branches. This happens all the time when // specializing code: code specialization turns conditional branches into // uncond branches, and this code folds them. - Function::iterator I = cast(ValueMap[&OldFunc->getEntryBlock()]); + Function::iterator Begin = cast(VMap[StartingBB])->getIterator(); + Function::iterator I = Begin; while (I != NewFunc->end()) { + // Check if this block has become dead during inlining or other + // simplifications. Note that the first block will appear dead, as it has + // not yet been wired up properly. + if (I != Begin && (pred_begin(&*I) == pred_end(&*I) || + I->getSinglePredecessor() == &*I)) { + BasicBlock *DeadBB = &*I++; + DeleteDeadBlock(DeadBB); + continue; + } + + // We need to simplify conditional branches and switches with a constant + // operand. We try to prune these out when cloning, but if the + // simplification required looking through PHI nodes, those are only + // available after forming the full basic block. That may leave some here, + // and we still want to prune the dead code as early as possible. + ConstantFoldTerminator(&*I); + BranchInst *BI = dyn_cast(I->getTerminator()); if (!BI || BI->isConditional()) { ++I; continue; } - // Note that we can't eliminate uncond branches if the destination has - // single-entry PHI nodes. Eliminating the single-entry phi nodes would - // require scanning the ValueMap to update any entries that point to the phi - // node. BasicBlock *Dest = BI->getSuccessor(0); - if (!Dest->getSinglePredecessor() || isa(Dest->begin())) { + if (!Dest->getSinglePredecessor()) { ++I; continue; } - + + // We shouldn't be able to get single-entry PHI nodes here, as instsimplify + // above should have zapped all of them.. + assert(!isa(Dest->begin())); + // We know all single-entry PHI nodes in the inlined function have been // removed, so we just need to splice the blocks. BI->eraseFromParent(); + // Make all PHI nodes that referred to Dest now refer to I as their source. + Dest->replaceAllUsesWith(&*I); + // Move all the instructions in the succ to the pred. I->getInstList().splice(I->end(), Dest->getInstList()); - // Make all PHI nodes that referred to Dest now refer to I as their source. - Dest->replaceAllUsesWith(I); - // Remove the dest block. Dest->eraseFromParent(); // Do not increment I, iteratively merge all things this block branches to. } + + // Make a final pass over the basic blocks from the old function to gather + // any return instructions which survived folding. We have to do this here + // because we can iteratively remove and merge returns above. + for (Function::iterator I = cast(VMap[StartingBB])->getIterator(), + E = NewFunc->end(); + I != E; ++I) + if (ReturnInst *RI = dyn_cast(I->getTerminator())) + Returns.push_back(RI); +} + + +/// This works exactly like CloneFunctionInto, +/// except that it does some simple constant prop and DCE on the fly. The +/// effect of this is to copy significantly less code in cases where (for +/// example) a function call with constant arguments is inlined, and those +/// constant arguments cause a significant amount of code in the callee to be +/// dead. Since this doesn't produce an exact copy of the input, it can't be +/// used for things like CloneFunction or CloneModule. +void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, + ValueToValueMapTy &VMap, + bool ModuleLevelChanges, + SmallVectorImpl &Returns, + const char *NameSuffix, + ClonedCodeInfo *CodeInfo, + Instruction *TheCall) { + CloneAndPruneIntoFromInst(NewFunc, OldFunc, &OldFunc->front().front(), VMap, + ModuleLevelChanges, Returns, NameSuffix, CodeInfo); +} + +/// \brief Remaps instructions in \p Blocks using the mapping in \p VMap. +void llvm::remapInstructionsInBlocks( + const SmallVectorImpl &Blocks, ValueToValueMapTy &VMap) { + // Rewrite the code to refer to itself. + for (auto *BB : Blocks) + for (auto &Inst : *BB) + RemapInstruction(&Inst, VMap, + RF_NoModuleLevelChanges | RF_IgnoreMissingEntries); +} + +/// \brief Clones a loop \p OrigLoop. Returns the loop and the blocks in \p +/// Blocks. +/// +/// Updates LoopInfo and DominatorTree assuming the loop is dominated by block +/// \p LoopDomBB. Insert the new blocks before block specified in \p Before. +Loop *llvm::cloneLoopWithPreheader(BasicBlock *Before, BasicBlock *LoopDomBB, + Loop *OrigLoop, ValueToValueMapTy &VMap, + const Twine &NameSuffix, LoopInfo *LI, + DominatorTree *DT, + SmallVectorImpl &Blocks) { + Function *F = OrigLoop->getHeader()->getParent(); + Loop *ParentLoop = OrigLoop->getParentLoop(); + + Loop *NewLoop = new Loop(); + if (ParentLoop) + ParentLoop->addChildLoop(NewLoop); + else + LI->addTopLevelLoop(NewLoop); + + BasicBlock *OrigPH = OrigLoop->getLoopPreheader(); + assert(OrigPH && "No preheader"); + BasicBlock *NewPH = CloneBasicBlock(OrigPH, VMap, NameSuffix, F); + // To rename the loop PHIs. + VMap[OrigPH] = NewPH; + Blocks.push_back(NewPH); + + // Update LoopInfo. + if (ParentLoop) + ParentLoop->addBasicBlockToLoop(NewPH, *LI); + + // Update DominatorTree. + DT->addNewBlock(NewPH, LoopDomBB); + + for (BasicBlock *BB : OrigLoop->getBlocks()) { + BasicBlock *NewBB = CloneBasicBlock(BB, VMap, NameSuffix, F); + VMap[BB] = NewBB; + + // Update LoopInfo. + NewLoop->addBasicBlockToLoop(NewBB, *LI); + + // Update DominatorTree. + BasicBlock *IDomBB = DT->getNode(BB)->getIDom()->getBlock(); + DT->addNewBlock(NewBB, cast(VMap[IDomBB])); + + Blocks.push_back(NewBB); + } + + // Move them physically from the end of the block list. + F->getBasicBlockList().splice(Before->getIterator(), F->getBasicBlockList(), + NewPH); + F->getBasicBlockList().splice(Before->getIterator(), F->getBasicBlockList(), + NewLoop->getHeader()->getIterator(), F->end()); + + return NewLoop; }