X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTransforms%2FUtils%2FCloneFunction.cpp;h=9876ee19a7b0bdbb3a2a31036eabfed3fb560183;hb=3885990176bdce382c45418a8e38ea6c089fa71c;hp=561b69dd1ee1cefa7ab49e60209bab6513d308bf;hpb=72f5f313d87558958696ce69593d82efcdfa9128;p=oota-llvm.git diff --git a/lib/Transforms/Utils/CloneFunction.cpp b/lib/Transforms/Utils/CloneFunction.cpp index 561b69dd1ee..9876ee19a7b 100644 --- a/lib/Transforms/Utils/CloneFunction.cpp +++ b/lib/Transforms/Utils/CloneFunction.cpp @@ -14,19 +14,23 @@ //===----------------------------------------------------------------------===// #include "llvm/Transforms/Utils/Cloning.h" -#include "llvm/Constants.h" -#include "llvm/DerivedTypes.h" -#include "llvm/Instructions.h" -#include "llvm/IntrinsicInst.h" -#include "llvm/GlobalVariable.h" -#include "llvm/Function.h" -#include "llvm/LLVMContext.h" -#include "llvm/Metadata.h" -#include "llvm/Support/CFG.h" -#include "llvm/Transforms/Utils/ValueMapper.h" -#include "llvm/Analysis/ConstantFolding.h" -#include "llvm/Analysis/DebugInfo.h" #include "llvm/ADT/SmallVector.h" +#include "llvm/Analysis/ConstantFolding.h" +#include "llvm/Analysis/InstructionSimplify.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; @@ -60,7 +64,6 @@ 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(); @@ -75,7 +78,9 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc, ValueToValueMapTy &VMap, bool ModuleLevelChanges, SmallVectorImpl &Returns, - const char *NameSuffix, ClonedCodeInfo *CodeInfo) { + const char *NameSuffix, ClonedCodeInfo *CodeInfo, + ValueMapTypeRemapper *TypeMapper, + ValueMaterializer *Materializer) { assert(NameSuffix && "NameSuffix cannot be null!"); #ifndef NDEBUG @@ -84,24 +89,28 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc, assert(VMap.count(I) && "No mapping from source argument specified!"); #endif - // Clone any attributes. - if (NewFunc->arg_size() == OldFunc->arg_size()) - NewFunc->copyAttributesFrom(OldFunc); - else { - //Some arguments were deleted with the VMap. Copy arguments one by one - for (Function::const_arg_iterator I = OldFunc->arg_begin(), - E = OldFunc->arg_end(); I != E; ++I) - if (Argument* Anew = dyn_cast(VMap[I])) - Anew->addAttr( OldFunc->getAttributes() - .getParamAttributes(I->getArgNo() + 1)); - NewFunc->setAttributes(NewFunc->getAttributes() - .addAttr(0, OldFunc->getAttributes() - .getRetAttributes())); - NewFunc->setAttributes(NewFunc->getAttributes() - .addAttr(~0, OldFunc->getAttributes() - .getFnAttributes())); + // 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); + + 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 @@ -113,8 +122,23 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc, // Create a new basic block and copy instructions into it! BasicBlock *CBB = CloneBasicBlock(&BB, VMap, NameSuffix, NewFunc, CodeInfo); - VMap[&BB] = CBB; // Add basic block mapping. + // 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); } @@ -126,7 +150,56 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc, // Loop over all instructions, fixing each one as we find it... for (BasicBlock::iterator II = BB->begin(); II != BB->end(); ++II) RemapInstruction(II, VMap, - ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges); + ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges, + TypeMapper, Materializer); +} + +// Find the MDNode which corresponds to the DISubprogram data that described F. +static MDNode* 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, DIArray SPs, Metadata *NewSP) { + SmallVector NewSPs; + NewSPs.reserve(SPs->getNumOperands() + 1); + for (unsigned I = 0, E = SPs->getNumOperands(); I != E; ++I) + NewSPs.push_back(SPs->getOperand(I)); + NewSPs.push_back(NewSP); + CU.replaceSubprograms(DIArray(MDNode::get(CU->getContext(), NewSPs))); +} + +// 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 MDNode *OldSubprogramMDNode = FindSubprogram(OldFunc, Finder); + if (!OldSubprogramMDNode) return; + + // Ensure that OldFunc appears in the map. + // (if it's already there it must point to NewFunc anyway) + VMap[OldFunc] = NewFunc; + DISubprogram NewSubprogram(MapMetadata(OldSubprogramMDNode, VMap)); + + for (DICompileUnit CU : Finder.compile_units()) { + DIArray Subprograms(CU.getSubprograms()); + + // If the compile unit's function list contains the old function, it should + // also contain the new one. + for (unsigned i = 0; i < Subprograms.getNumElements(); i++) { + if ((MDNode*)Subprograms.getElement(i) == OldSubprogramMDNode) { + AddOperand(CU, Subprograms, NewSubprogram); + break; + } + } + } } /// CloneFunction - Return a copy of the specified function, but without @@ -140,7 +213,7 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc, 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 VMap. If so, we need to not add the arguments to the arg ty vector @@ -166,6 +239,9 @@ Function *llvm::CloneFunction(const Function *F, ValueToValueMapTy &VMap, VMap[I] = DestI++; // Add mapping to VMap } + if (ModuleLevelChanges) + CloneDebugInfoMetadata(NewF, F, VMap); + SmallVector Returns; // Ignore returns cloned. CloneFunctionInto(NewF, F, VMap, ModuleLevelChanges, Returns, "", CodeInfo); return NewF; @@ -181,40 +257,44 @@ namespace { const Function *OldFunc; ValueToValueMapTy &VMap; bool ModuleLevelChanges; - SmallVectorImpl &Returns; const char *NameSuffix; ClonedCodeInfo *CodeInfo; - const TargetData *TD; + CloningDirector *Director; + ValueMapTypeRemapper *TypeMapper; + ValueMaterializer *Materializer; + public: PruningFunctionCloner(Function *newFunc, const Function *oldFunc, - ValueToValueMapTy &valueMap, - bool moduleLevelChanges, - SmallVectorImpl &returns, - const char *nameSuffix, - ClonedCodeInfo *codeInfo, - const TargetData *td) - : NewFunc(newFunc), OldFunc(oldFunc), - VMap(valueMap), ModuleLevelChanges(moduleLevelChanges), - Returns(returns), NameSuffix(nameSuffix), CodeInfo(codeInfo), TD(td) { + ValueToValueMapTy &valueMap, bool moduleLevelChanges, + const char *nameSuffix, ClonedCodeInfo *codeInfo, + CloningDirector *Director) + : NewFunc(newFunc), OldFunc(oldFunc), VMap(valueMap), + ModuleLevelChanges(moduleLevelChanges), NameSuffix(nameSuffix), + CodeInfo(codeInfo), Director(Director) { + // These are optional components. The Director may return null. + if (Director) { + TypeMapper = Director->getTypeRemapper(); + Materializer = Director->getValueMaterializer(); + } else { + TypeMapper = nullptr; + Materializer = nullptr; + } } /// CloneBlock - 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 VMap if they are available. - Constant *ConstantFoldMappedInstruction(const Instruction *I); }; } /// CloneBlock - 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){ - TrackingVH &BBEntry = VMap[BB]; + WeakVH &BBEntry = VMap[BB]; // Have we already cloned this block? if (BBEntry) return; @@ -224,25 +304,74 @@ void PruningFunctionCloner::CloneBlock(const BasicBlock *BB, 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)) { - VMap[II] = C; - continue; + // If the "Director" remaps the instruction, don't clone it. + if (Director) { + CloningDirector::CloningAction Action + = Director->handleInstruction(VMap, II, NewBB); + // If the cloning director says stop, we want to stop everything, not + // just break out of the loop (which would cause the terminator to be + // cloned). The cloning director is responsible for inserting a proper + // terminator into the new basic block in this case. + if (Action == CloningDirector::StopCloningBB) + return; + // If the cloning director says skip, continue to the next instruction. + // In this case, the cloning director is responsible for mapping the + // skipped instruction to some value that is defined in the new + // basic block. + if (Action == CloningDirector::SkipInstruction) + 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, + TypeMapper, Materializer); + + // 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); - NewBB->getInstList().push_back(NewInst); VMap[II] = NewInst; // Add instruction map to value. - + NewBB->getInstList().push_back(NewInst); hasCalls |= (isa(II) && !isa(II)); if (const AllocaInst *AI = dyn_cast(II)) { if (isa(AI->getArraySize())) @@ -255,12 +384,24 @@ void PruningFunctionCloner::CloneBlock(const BasicBlock *BB, // Finally, clone over the terminator. const TerminatorInst *OldTI = BB->getTerminator(); bool TerminatorDone = false; + if (Director) { + CloningDirector::CloningAction Action + = Director->handleInstruction(VMap, OldTI, NewBB); + // If the cloning director says stop, we want to stop everything, not + // just break out of the loop (which would cause the terminator to be + // cloned). The cloning director is responsible for inserting a proper + // terminator into the new basic block in this case. + if (Action == CloningDirector::StopCloningBB) + return; + assert(Action != CloningDirector::SkipInstruction && + "SkipInstruction is not valid for terminators."); + } if (const BranchInst *BI = dyn_cast(OldTI)) { if (BI->isConditional()) { // 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) { + if (!Cond) { Value *V = VMap[BI->getCondition()]; Cond = dyn_cast_or_null(V); } @@ -276,12 +417,13 @@ 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... + 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)); + SwitchInst::ConstCaseIt Case = SI->findCaseValue(Cond); + BasicBlock *Dest = const_cast(Case.getCaseSuccessor()); VMap[OldTI] = BranchInst::Create(Dest, NewBB); ToClone.push_back(Dest); TerminatorDone = true; @@ -303,90 +445,60 @@ void PruningFunctionCloner::CloneBlock(const BasicBlock *BB, 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 VMap 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), - VMap, - ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges))) - Ops.push_back(Op); - else - return 0; // All operands not constant! - - if (const CmpInst *CI = dyn_cast(I)) - return ConstantFoldCompareInstOperands(CI->getPredicate(), Ops[0], Ops[1], - TD); - - if (const LoadInst *LI = dyn_cast(I)) - if (ConstantExpr *CE = dyn_cast(Ops[0])) - if (!LI->isVolatile() && CE->getOpcode() == Instruction::GetElementPtr) - if (GlobalVariable *GV = dyn_cast(CE->getOperand(0))) - if (GV->isConstant() && GV->hasDefinitiveInitializer()) - return ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), - CE); - - return ConstantFoldInstOperands(I->getOpcode(), I->getType(), &Ops[0], - Ops.size(), TD); -} - -static DebugLoc -UpdateInlinedAtInfo(const DebugLoc &InsnDL, const DebugLoc &TheCallDL, - LLVMContext &Ctx) { - DebugLoc NewLoc = TheCallDL; - if (MDNode *IA = InsnDL.getInlinedAt(Ctx)) - NewLoc = UpdateInlinedAtInfo(DebugLoc::getFromDILocation(IA), TheCallDL, - Ctx); - - return DebugLoc::get(InsnDL.getLine(), InsnDL.getCol(), - InsnDL.getScope(Ctx), NewLoc.getAsMDNode(Ctx)); } -/// 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, +/// CloneAndPruneIntoFromInst - 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, + SmallVectorImpl &Returns, const char *NameSuffix, ClonedCodeInfo *CodeInfo, - const TargetData *TD, - Instruction *TheCall) { + CloningDirector *Director) { assert(NameSuffix && "NameSuffix cannot be null!"); - + + ValueMapTypeRemapper *TypeMapper = nullptr; + ValueMaterializer *Materializer = nullptr; + + if (Director) { + TypeMapper = Director->getTypeRemapper(); + Materializer = Director->getValueMaterializer(); + } + #ifndef NDEBUG - for (Function::const_arg_iterator II = OldFunc->arg_begin(), - E = OldFunc->arg_end(); II != E; ++II) - assert(VMap.count(II) && "No mapping from source argument specified!"); + // If the cloning starts at the begining of the function, verify that + // the function arguments are mapped. + if (!StartingInst) + for (Function::const_arg_iterator II = OldFunc->arg_begin(), + E = OldFunc->arg_end(); II != E; ++II) + assert(VMap.count(II) && "No mapping from source argument specified!"); #endif PruningFunctionCloner PFC(NewFunc, OldFunc, VMap, ModuleLevelChanges, - Returns, NameSuffix, CodeInfo, TD); + NameSuffix, CodeInfo, Director); + const BasicBlock *StartingBB; + if (StartingInst) + StartingBB = StartingInst->getParent(); + else { + StartingBB = &OldFunc->getEntryBlock(); + StartingInst = StartingBB->begin(); + } // Clone the entry block, and anything recursively reachable from it. std::vector CloneWorklist; - CloneWorklist.push_back(&OldFunc->getEntryBlock()); + PFC.CloneBlock(StartingBB, StartingInst, 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 @@ -399,69 +511,24 @@ void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, BI != BE; ++BI) { Value *V = VMap[BI]; BasicBlock *NewBB = cast_or_null(V); - if (NewBB == 0) continue; // Dead block. + 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 VMap to do all the hard work. - // - BasicBlock::iterator I = NewBB->begin(); - - DebugLoc TheCallDL; - if (TheCall) - TheCallDL = TheCall->getDebugLoc(); - + // 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) { - if (I->hasMetadata()) { - if (!TheCallDL.isUnknown()) { - DebugLoc IDL = I->getDebugLoc(); - if (!IDL.isUnknown()) { - DebugLoc NewDL = UpdateInlinedAtInfo(IDL, TheCallDL, - I->getContext()); - I->setDebugLoc(NewDL); - } - } else { - // The cloned instruction has dbg info but the call instruction - // does not have dbg info. Remove dbg info from cloned instruction. - I->setDebugLoc(DebugLoc()); - } - } - PHIToResolve.push_back(cast(OldI)); - } - } - - // FIXME: - // FIXME: - // FIXME: Unclone all this metadata stuff. - // FIXME: - // FIXME: - - // Otherwise, remap the rest of the instructions normally. - for (; I != NewBB->end(); ++I) { - if (I->hasMetadata()) { - if (!TheCallDL.isUnknown()) { - DebugLoc IDL = I->getDebugLoc(); - if (!IDL.isUnknown()) { - DebugLoc NewDL = UpdateInlinedAtInfo(IDL, TheCallDL, - I->getContext()); - I->setDebugLoc(NewDL); - } - } else { - // The cloned instruction has dbg info but the call instruction - // does not have dbg info. Remove dbg info from cloned instruction. - I->setDebugLoc(DebugLoc()); - } - } - RemapInstruction(I, VMap, - ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges); - } + for (BasicBlock::const_iterator I = BI->begin(), E = BI->end(); I != E; ++I) + if (const PHINode *PN = dyn_cast(I)) + PHIToResolve.push_back(PN); + else + break; + + // 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 @@ -543,31 +610,55 @@ void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, ++OldI; } } - // NOTE: We cannot eliminate single entry phi nodes here, because of - // VMap. Single entry phi nodes can have multiple VMap entries - // pointing at them. Thus, deleting one would require scanning the VMap - // 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(VMap[&OldFunc->getEntryBlock()]); + Function::iterator Begin = cast(VMap[StartingBB]); + 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 VMap 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(); @@ -583,4 +674,33 @@ void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, // 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]), + E = NewFunc->end(); + I != E; ++I) + if (ReturnInst *RI = dyn_cast(I->getTerminator())) + Returns.push_back(RI); +} + + +/// 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, + ValueToValueMapTy &VMap, + bool ModuleLevelChanges, + SmallVectorImpl &Returns, + const char *NameSuffix, + ClonedCodeInfo *CodeInfo, + Instruction *TheCall) { + CloneAndPruneIntoFromInst(NewFunc, OldFunc, OldFunc->front().begin(), VMap, + ModuleLevelChanges, Returns, NameSuffix, CodeInfo, + nullptr); }