X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTransforms%2FUtils%2FCloneFunction.cpp;h=be8d39e128a53e7d3f118f2124fd43417c028490;hb=defce4cfd607e013169033164def7a3e558bfd54;hp=2b2fcb1052c2ea117ed8c87848d62c9fa465f86a;hpb=9e9a0d5fc26878e51a58a8b57900fcbf952c2691;p=oota-llvm.git diff --git a/lib/Transforms/Utils/CloneFunction.cpp b/lib/Transforms/Utils/CloneFunction.cpp index 2b2fcb1052c..be8d39e128a 100644 --- a/lib/Transforms/Utils/CloneFunction.cpp +++ b/lib/Transforms/Utils/CloneFunction.cpp @@ -14,28 +14,31 @@ //===----------------------------------------------------------------------===// #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/ADT/SmallVector.h" +#include "llvm/Analysis/ConstantFolding.h" +#include "llvm/Analysis/InstructionSimplify.h" +#include "llvm/DebugInfo.h" +#include "llvm/IR/Constants.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/Support/CFG.h" -#include "llvm/Support/Compiler.h" +#include "llvm/Transforms/Utils/BasicBlockUtils.h" +#include "llvm/Transforms/Utils/Local.h" #include "llvm/Transforms/Utils/ValueMapper.h" -#include "llvm/Analysis/ConstantFolding.h" -#include "llvm/Analysis/DebugInfo.h" -#include "llvm/ADT/SmallVector.h" #include using namespace llvm; // CloneBasicBlock - 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 = BasicBlock::Create("", F); + BasicBlock *NewBB = BasicBlock::Create(BB->getContext(), "", F); if (BB->hasName()) NewBB->setName(BB->getName()+NameSuffix); bool hasCalls = false, hasDynamicAllocas = false, hasStaticAllocas = false; @@ -43,11 +46,11 @@ BasicBlock *llvm::CloneBasicBlock(const BasicBlock *BB, // Loop over all instructions, and copy them over. for (BasicBlock::const_iterator II = BB->begin(), IE = BB->end(); II != IE; ++II) { - Instruction *NewInst = II->clone(BB->getContext()); + Instruction *NewInst = II->clone(); if (II->hasName()) NewInst->setName(II->getName()+NameSuffix); NewBB->getInstList().push_back(NewInst); - ValueMap[II] = NewInst; // Add instruction map to value. + VMap[II] = NewInst; // Add instruction map to value. hasCalls |= (isa(II) && !isa(II)); if (const AllocaInst *AI = dyn_cast(II)) { @@ -60,7 +63,6 @@ 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(); @@ -69,38 +71,42 @@ 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) { 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!"); + 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 ValueMap. 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(ValueMap[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())); + AttributeSet OldAttrs = OldFunc->getAttributes(); + // Clone any argument attributes that are present in the VMap. + for (Function::const_arg_iterator I = OldFunc->arg_begin(), + E = OldFunc->arg_end(); + I != E; ++I) + if (Argument *Anew = dyn_cast(VMap[I])) { + AttributeSet attrs = + OldAttrs.getParamAttributes(I->getArgNo() + 1); + if (attrs.getNumSlots() > 0) + Anew->addAttr(attrs); + } - } + NewFunc->setAttributes(NewFunc->getAttributes() + .addAttributes(NewFunc->getContext(), + AttributeSet::ReturnIndex, + OldAttrs.getRetAttributes())); + NewFunc->setAttributes(NewFunc->getAttributes() + .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 @@ -111,43 +117,58 @@ 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()]), + // references as we go. This uses VMap to do all the hard work. + for (Function::iterator BB = cast(VMap[OldFunc->begin()]), 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); + RemapInstruction(II, VMap, + ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges, + TypeMapper); } /// CloneFunction - 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? + if (VMap.count(I) == 0) // Haven't mapped the argument to anything yet? ArgTypes.push_back(I->getType()); // Create a new function type... @@ -161,13 +182,13 @@ Function *llvm::CloneFunction(const Function *F, 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? + if (VMap.count(I) == 0) { // Is this argument preserved? DestI->setName(I->getName()); // Copy the name over... - ValueMap[I] = DestI++; // Add mapping to ValueMap + VMap[I] = DestI++; // Add mapping to VMap } - std::vector Returns; // Ignore returns cloned... - CloneFunctionInto(NewF, F, ValueMap, Returns, "", CodeInfo); + SmallVector Returns; // Ignore returns cloned. + CloneFunctionInto(NewF, F, VMap, ModuleLevelChanges, Returns, "", CodeInfo); return NewF; } @@ -176,35 +197,30 @@ Function *llvm::CloneFunction(const Function *F, namespace { /// PruningFunctionCloner - This class is a private class used to implement /// the CloneAndPruneFunctionInto method. - struct VISIBILITY_HIDDEN PruningFunctionCloner { + struct PruningFunctionCloner { Function *NewFunc; const Function *OldFunc; - DenseMap &ValueMap; - std::vector &Returns; + ValueToValueMapTy &VMap; + bool ModuleLevelChanges; const char *NameSuffix; ClonedCodeInfo *CodeInfo; - const TargetData *TD; - Value *DbgFnStart; + const DataLayout *TD; public: PruningFunctionCloner(Function *newFunc, const Function *oldFunc, - DenseMap &valueMap, - std::vector &returns, + ValueToValueMapTy &valueMap, + bool moduleLevelChanges, const char *nameSuffix, ClonedCodeInfo *codeInfo, - const TargetData *td) - : NewFunc(newFunc), OldFunc(oldFunc), ValueMap(valueMap), Returns(returns), - NameSuffix(nameSuffix), CodeInfo(codeInfo), TD(td), DbgFnStart(NULL) { + const DataLayout *td) + : NewFunc(newFunc), OldFunc(oldFunc), + VMap(valueMap), ModuleLevelChanges(moduleLevelChanges), + NameSuffix(nameSuffix), CodeInfo(codeInfo), TD(td) { } /// CloneBlock - The specified block is found to be reachable, clone it and /// anything that it can reach. void CloneBlock(const BasicBlock *BB, std::vector &ToClone); - - public: - /// ConstantFoldMappedInstruction - Constant fold the specified instruction, - /// mapping its operands through ValueMap if they are available. - Constant *ConstantFoldMappedInstruction(const Instruction *I); }; } @@ -212,48 +228,65 @@ namespace { /// anything that it can reach. void PruningFunctionCloner::CloneBlock(const BasicBlock *BB, 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 = BasicBlock::Create(); + 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(); 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; - } - - // Do not clone llvm.dbg.region.end. It will be adjusted by the inliner. - if (const DbgFuncStartInst *DFSI = dyn_cast(II)) { - if (DbgFnStart == NULL) { - DISubprogram SP(cast(DFSI->getSubprogram())); - if (SP.describes(BB->getParent())) - DbgFnStart = DFSI->getSubprogram(); - } - } - if (const DbgRegionEndInst *DREIS = dyn_cast(II)) { - if (DREIS->getContext() == DbgFnStart) + 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, TD)) { + // 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; + } } - - Instruction *NewInst = II->clone(BB->getContext()); + 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) && !isa(II)); if (const AllocaInst *AI = dyn_cast(II)) { if (isa(AI->getArraySize())) @@ -271,13 +304,15 @@ 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 == 0) { + 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] = BranchInst::Create(Dest, NewBB); + VMap[OldTI] = BranchInst::Create(Dest, NewBB); ToClone.push_back(Dest); TerminatorDone = true; } @@ -285,22 +320,25 @@ 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 == 0) { // 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] = BranchInst::Create(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; } } if (!TerminatorDone) { - Instruction *NewInst = OldTI->clone(BB->getContext()); + Instruction *NewInst = OldTI->clone(); 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. // Recursively clone any reachable successor blocks. const TerminatorInst *TI = BB->getTerminator(); @@ -310,46 +348,10 @@ 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 ValueMap if they are available. -Constant *PruningFunctionCloner:: -ConstantFoldMappedInstruction(const Instruction *I) { - LLVMContext &Context = I->getContext(); - - SmallVector Ops; - for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) - if (Constant *Op = dyn_cast_or_null(MapValue(I->getOperand(i), - ValueMap, - Context))) - 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.size(), - Context, 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, Context); - - return ConstantFoldInstOperands(I->getOpcode(), I->getType(), &Ops[0], - Ops.size(), Context, TD); } /// CloneAndPruneFunctionInto - This works exactly like CloneFunctionInto, @@ -360,21 +362,22 @@ ConstantFoldMappedInstruction(const Instruction *I) { /// 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, + ValueToValueMapTy &VMap, + bool ModuleLevelChanges, + SmallVectorImpl &Returns, const char *NameSuffix, ClonedCodeInfo *CodeInfo, - const TargetData *TD) { + const DataLayout *TD, + Instruction *TheCall) { assert(NameSuffix && "NameSuffix cannot be null!"); - LLVMContext &Context = OldFunc->getContext(); #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!"); + assert(VMap.count(II) && "No mapping from source argument specified!"); #endif - PruningFunctionCloner PFC(NewFunc, OldFunc, ValueMap, Returns, + PruningFunctionCloner PFC(NewFunc, OldFunc, VMap, ModuleLevelChanges, NameSuffix, CodeInfo, TD); // Clone the entry block, and anything recursively reachable from it. @@ -391,32 +394,28 @@ 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; + SmallVector PHIToResolve; for (Function::const_iterator BI = OldFunc->begin(), BE = OldFunc->end(); BI != BE; ++BI) { - BasicBlock *NewBB = cast_or_null(ValueMap[BI]); + Value *V = VMap[BI]; + BasicBlock *NewBB = cast_or_null(V); if (NewBB == 0) 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)); - } - - // Otherwise, remap the rest of the instructions normally. - for (; I != NewBB->end(); ++I) - RemapInstruction(I, ValueMap); + 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); } // Defer PHI resolution until rest of function is resolved, PHI resolution @@ -425,19 +424,20 @@ 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 *V = VMap[PN->getIncomingBlock(pred)]; + if (BasicBlock *MappedBlock = cast_or_null(V)) { Value *InVal = MapValue(PN->getIncomingValue(pred), - ValueMap, Context); + VMap, + ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges); assert(InVal && "Unknown input value?"); PN->setIncomingValue(pred, InVal); PN->setIncomingBlock(pred, MappedBlock); @@ -491,50 +491,83 @@ 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, TD); + // Now that the inlined function body has been fully constructed, go through // and zap unconditional fall-through branches. This happen 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[&OldFunc->getEntryBlock()]); + 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(); - // 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); + // Move all the instructions in the succ to the pred. + I->getInstList().splice(I->end(), Dest->getInstList()); + // 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 theh 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[&OldFunc->getEntryBlock()]), + E = NewFunc->end(); + I != E; ++I) + if (ReturnInst *RI = dyn_cast(I->getTerminator())) + Returns.push_back(RI); }