X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTransforms%2FUtils%2FCloneFunction.cpp;h=62fc2ec10b14ad8f85cbfd26a1f4d6f0c6b56dd6;hb=84bd6b0c31f41cdd1d859dab54b6bc1177c4c6bb;hp=dd1b31bde5ac3e077787a61c7d9b53d4b14043b5;hpb=8d8d6530b13ea092a69f7fdc89786e22c0f96827;p=oota-llvm.git diff --git a/lib/Transforms/Utils/CloneFunction.cpp b/lib/Transforms/Utils/CloneFunction.cpp index dd1b31bde5a..62fc2ec10b1 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. // //===----------------------------------------------------------------------===// // @@ -17,17 +17,25 @@ #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 "ValueMapper.h" -#include "llvm/Transforms/Utils/Local.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 using namespace llvm; // CloneBasicBlock - See comments in Cloning.h BasicBlock *llvm::CloneBasicBlock(const BasicBlock *BB, - std::map &ValueMap, - const char *NameSuffix, Function *F, + DenseMap &ValueMap, + 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; @@ -41,7 +49,7 @@ BasicBlock *llvm::CloneBasicBlock(const BasicBlock *BB, NewBB->getInstList().push_back(NewInst); ValueMap[II] = NewInst; // Add instruction map to value. - hasCalls |= isa(II); + hasCalls |= (isa(II) && !isa(II)); if (const AllocaInst *AI = dyn_cast(II)) { if (isa(AI->getArraySize())) hasStaticAllocas = true; @@ -55,7 +63,7 @@ BasicBlock *llvm::CloneBasicBlock(const BasicBlock *BB, CodeInfo->ContainsUnwinds |= isa(BB->getTerminator()); CodeInfo->ContainsDynamicAllocas |= hasDynamicAllocas; CodeInfo->ContainsDynamicAllocas |= hasStaticAllocas && - BB != &BB->getParent()->front(); + BB != &BB->getParent()->getEntryBlock(); } return NewBB; } @@ -64,8 +72,8 @@ BasicBlock *llvm::CloneBasicBlock(const BasicBlock *BB, // ArgMap values. // void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc, - std::map &ValueMap, - std::vector &Returns, + DenseMap &ValueMap, + SmallVectorImpl &Returns, const char *NameSuffix, ClonedCodeInfo *CodeInfo) { assert(NameSuffix && "NameSuffix cannot be null!"); @@ -75,6 +83,25 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc, assert(ValueMap.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())); + + } + // 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 // recursive functions into themselves. @@ -111,7 +138,7 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc, /// the function from their old to new values. /// Function *llvm::CloneFunction(const Function *F, - std::map &ValueMap, + DenseMap &ValueMap, ClonedCodeInfo *CodeInfo) { std::vector ArgTypes; @@ -128,7 +155,7 @@ Function *llvm::CloneFunction(const Function *F, 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(); @@ -139,7 +166,7 @@ Function *llvm::CloneFunction(const Function *F, ValueMap[I] = DestI++; // Add mapping to ValueMap } - std::vector Returns; // Ignore returns cloned... + SmallVector Returns; // Ignore returns cloned. CloneFunctionInto(NewF, F, ValueMap, Returns, "", CodeInfo); return NewF; } @@ -152,24 +179,26 @@ namespace { struct PruningFunctionCloner { Function *NewFunc; const Function *OldFunc; - std::map &ValueMap; - std::vector &Returns; + DenseMap &ValueMap; + SmallVectorImpl &Returns; const char *NameSuffix; ClonedCodeInfo *CodeInfo; - + const TargetData *TD; public: PruningFunctionCloner(Function *newFunc, const Function *oldFunc, - std::map &valueMap, - std::vector &returns, + DenseMap &valueMap, + SmallVectorImpl &returns, const char *nameSuffix, - ClonedCodeInfo *codeInfo) + ClonedCodeInfo *codeInfo, + const TargetData *td) : NewFunc(newFunc), OldFunc(oldFunc), ValueMap(valueMap), Returns(returns), - NameSuffix(nameSuffix), CodeInfo(codeInfo) { + 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); + void CloneBlock(const BasicBlock *BB, + std::vector &ToClone); public: /// ConstantFoldMappedInstruction - Constant fold the specified instruction, @@ -180,7 +209,8 @@ namespace { /// CloneBlock - The specified block is found to be reachable, clone it and /// anything that it can reach. -void PruningFunctionCloner::CloneBlock(const BasicBlock *BB) { +void PruningFunctionCloner::CloneBlock(const BasicBlock *BB, + std::vector &ToClone){ Value *&BBEntry = ValueMap[BB]; // Have we already cloned this block? @@ -188,14 +218,14 @@ void PruningFunctionCloner::CloneBlock(const BasicBlock *BB) { // Nope, clone it now. BasicBlock *NewBB; - BBEntry = NewBB = new BasicBlock(); + BBEntry = NewBB = BasicBlock::Create(BB->getContext()); if (BB->hasName()) NewBB->setName(BB->getName()+NameSuffix); 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 = 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. @@ -203,14 +233,14 @@ void PruningFunctionCloner::CloneBlock(const BasicBlock *BB) { ValueMap[II] = C; continue; } - + 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. - hasCalls |= isa(II); + hasCalls |= (isa(II) && !isa(II)); if (const AllocaInst *AI = dyn_cast(II)) { if (isa(AI->getArraySize())) hasStaticAllocas = true; @@ -219,9 +249,54 @@ void PruningFunctionCloner::CloneBlock(const BasicBlock *BB) { } } + // Finally, clone over the terminator. + const TerminatorInst *OldTI = BB->getTerminator(); + bool TerminatorDone = false; + 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) + Cond = dyn_cast_or_null(ValueMap[BI->getCondition()]); + + // Constant fold to uncond branch! + if (Cond) { + BasicBlock *Dest = BI->getSuccessor(!Cond->getZExtValue()); + ValueMap[OldTI] = BranchInst::Create(Dest, NewBB); + ToClone.push_back(Dest); + TerminatorDone = true; + } + } + } 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) { // Constant fold to uncond branch! + BasicBlock *Dest = SI->getSuccessor(SI->findCaseValue(Cond)); + ValueMap[OldTI] = BranchInst::Create(Dest, NewBB); + ToClone.push_back(Dest); + TerminatorDone = true; + } + } + + if (!TerminatorDone) { + 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. + + // 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)); + } + if (CodeInfo) { CodeInfo->ContainsCalls |= hasCalls; - CodeInfo->ContainsUnwinds |= isa(BB->getTerminator()); + CodeInfo->ContainsUnwinds |= isa(OldTI); CodeInfo->ContainsDynamicAllocas |= hasDynamicAllocas; CodeInfo->ContainsDynamicAllocas |= hasStaticAllocas && BB != &BB->getParent()->front(); @@ -229,27 +304,13 @@ void PruningFunctionCloner::CloneBlock(const BasicBlock *BB) { if (ReturnInst *RI = dyn_cast(NewBB->getTerminator())) Returns.push_back(RI); - - // Recursively clone any reachable successor blocks. - const TerminatorInst *TI = BB->getTerminator(); - for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i) - CloneBlock(TI->getSuccessor(i)); } /// ConstantFoldMappedInstruction - Constant fold the specified instruction, /// mapping its operands through ValueMap if they are available. Constant *PruningFunctionCloner:: ConstantFoldMappedInstruction(const Instruction *I) { - if (isa(I) || isa(I)) { - if (Constant *Op0 = dyn_cast_or_null(MapValue(I->getOperand(0), - ValueMap))) - if (Constant *Op1 = dyn_cast_or_null(MapValue(I->getOperand(1), - ValueMap))) - return ConstantExpr::get(I->getOpcode(), Op0, Op1); - return 0; - } - - std::vector Ops; + SmallVector Ops; for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) if (Constant *Op = dyn_cast_or_null(MapValue(I->getOperand(i), ValueMap))) @@ -257,45 +318,88 @@ ConstantFoldMappedInstruction(const Instruction *I) { else return 0; // All operands not constant! - return ConstantFoldInstOperands(I->getOpcode(), I->getType(), Ops); + 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 MDNode *UpdateInlinedAtInfo(MDNode *InsnMD, MDNode *TheCallMD) { + DILocation ILoc(InsnMD); + if (!ILoc.Verify()) return InsnMD; + + DILocation CallLoc(TheCallMD); + if (!CallLoc.Verify()) return InsnMD; + + DILocation OrigLocation = ILoc.getOrigLocation(); + MDNode *NewLoc = TheCallMD; + if (OrigLocation.Verify()) + NewLoc = UpdateInlinedAtInfo(OrigLocation.getNode(), TheCallMD); + + Value *MDVs[] = { + InsnMD->getOperand(0), // Line + InsnMD->getOperand(1), // Col + InsnMD->getOperand(2), // Scope + NewLoc + }; + return MDNode::get(InsnMD->getContext(), MDVs, 4); +} /// 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 exactly copy of the input, it can't 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, - std::map &ValueMap, - std::vector &Returns, + DenseMap &ValueMap, + SmallVectorImpl &Returns, const char *NameSuffix, - ClonedCodeInfo *CodeInfo) { + ClonedCodeInfo *CodeInfo, + const TargetData *TD, + Instruction *TheCall) { 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 (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!"); #endif - - PruningFunctionCloner PFC(NewFunc, OldFunc, ValueMap, Returns, - NameSuffix, CodeInfo); + + PruningFunctionCloner PFC(NewFunc, OldFunc, ValueMap, Returns, + NameSuffix, CodeInfo, TD); // Clone the entry block, and anything recursively reachable from it. - PFC.CloneBlock(&OldFunc->getEntryBlock()); + std::vector CloneWorklist; + CloneWorklist.push_back(&OldFunc->getEntryBlock()); + while (!CloneWorklist.empty()) { + const BasicBlock *BB = CloneWorklist.back(); + CloneWorklist.pop_back(); + PFC.CloneBlock(BB, CloneWorklist); + } // Loop over all of the basic blocks in the old function. If the block was // reachable, we have cloned it and the old block is now in the value map: // insert it into the new function in the right order. If not, ignore it. // + // Defer PHI resolution until rest of function is resolved. + SmallVector 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. - + // Add the new block to the new function. NewFunc->getBasicBlockList().push_back(NewBB); @@ -303,31 +407,174 @@ void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, // references as we go. This uses ValueMap to do all the hard work. // BasicBlock::iterator I = NewBB->begin(); + + unsigned DbgKind = OldFunc->getContext().getMDKindID("dbg"); + MDNode *TheCallMD = NULL; + if (TheCall && TheCall->hasMetadata()) + TheCallMD = TheCall->getMetadata(DbgKind); // Handle PHI nodes specially, as we have to remove references to dead // blocks. if (PHINode *PN = dyn_cast(I)) { - unsigned NumPreds = PN->getNumIncomingValues(); - for (; (PN = dyn_cast(I)); ++I) { - 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); - assert(InVal && "Unknown input value?"); - PN->setIncomingValue(pred, InVal); - PN->setIncomingBlock(pred, MappedBlock); + // 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 (TheCallMD) { + if (MDNode *IMD = I->getMetadata(DbgKind)) { + MDNode *NewMD = UpdateInlinedAtInfo(IMD, TheCallMD); + I->setMetadata(DbgKind, NewMD); + } } else { - PN->removeIncomingValue(pred, false); - --pred, --e; // Revisit the next entry. + // The cloned instruction has dbg info but the call instruction + // does not have dbg info. Remove dbg info from cloned instruction. + I->setMetadata(DbgKind, 0); } } + 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) + for (; I != NewBB->end(); ++I) { + if (I->hasMetadata()) { + if (TheCallMD) { + if (MDNode *IMD = I->getMetadata(DbgKind)) { + MDNode *NewMD = UpdateInlinedAtInfo(IMD, TheCallMD); + I->setMetadata(DbgKind, NewMD); + } + } else { + // The cloned instruction has dbg info but the call instruction + // does not have dbg info. Remove dbg info from cloned instruction. + I->setMetadata(DbgKind, 0); + } + } RemapInstruction(I, ValueMap); + } } -} + + // Defer PHI resolution until rest of function is resolved, PHI resolution + // requires the CFG to be up-to-date. + for (unsigned phino = 0, e = PHIToResolve.size(); phino != e; ) { + const PHINode *OPN = PHIToResolve[phino]; + unsigned NumPreds = OPN->getNumIncomingValues(); + const BasicBlock *OldBB = OPN->getParent(); + BasicBlock *NewBB = cast(ValueMap[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]); + 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); + assert(InVal && "Unknown input value?"); + PN->setIncomingValue(pred, InVal); + PN->setIncomingBlock(pred, MappedBlock); + } else { + PN->removeIncomingValue(pred, false); + --pred, --e; // Revisit the next entry. + } + } + } + + // The loop above has removed PHI entries for those blocks that are dead + // and has updated others. However, if a block is live (i.e. copied over) + // but its terminator has been changed to not go to this block, then our + // phi nodes will have invalid entries. Update the PHI nodes in this + // case. + PHINode *PN = cast(NewBB->begin()); + NumPreds = std::distance(pred_begin(NewBB), pred_end(NewBB)); + if (NumPreds != PN->getNumIncomingValues()) { + assert(NumPreds < PN->getNumIncomingValues()); + // Count how many times each predecessor comes to this block. + std::map PredCount; + for (pred_iterator PI = pred_begin(NewBB), E = pred_end(NewBB); + PI != E; ++PI) + --PredCount[*PI]; + + // Figure out how many entries to remove from each PHI. + for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) + ++PredCount[PN->getIncomingBlock(i)]; + + // At this point, the excess predecessor entries are positive in the + // map. Loop over all of the PHIs and remove excess predecessor + // entries. + BasicBlock::iterator I = NewBB->begin(); + for (; (PN = dyn_cast(I)); ++I) { + for (std::map::iterator PCI =PredCount.begin(), + E = PredCount.end(); PCI != E; ++PCI) { + BasicBlock *Pred = PCI->first; + for (unsigned NumToRemove = PCI->second; NumToRemove; --NumToRemove) + PN->removeIncomingValue(Pred, false); + } + } + } + + // If the loops above have made these phi nodes have 0 or 1 operand, + // replace them with undef or the input value. We must do this for + // correctness, because 0-operand phis are not valid. + PN = cast(NewBB->begin()); + if (PN->getNumIncomingValues() == 0) { + BasicBlock::iterator I = NewBB->begin(); + BasicBlock::const_iterator OldI = OldBB->begin(); + while ((PN = dyn_cast(I++))) { + Value *NV = UndefValue::get(PN->getType()); + PN->replaceAllUsesWith(NV); + assert(ValueMap[OldI] == PN && "ValueMap mismatch"); + ValueMap[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. + } + + // 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()]); + while (I != NewFunc->end()) { + 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())) { + ++I; continue; + } + + // 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); + // Remove the dest block. + Dest->eraseFromParent(); + + // Do not increment I, iteratively merge all things this block branches to. + } +}