//===- CodeExtractor.cpp - Pull code region into a new function -----------===// // // 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 implements the interface to tear out a code region, such as an // individual loop or a parallel section, into a new function, replacing it with // a call to the new function. // //===----------------------------------------------------------------------===// #include "llvm/BasicBlock.h" #include "llvm/Constants.h" #include "llvm/DerivedTypes.h" #include "llvm/Instructions.h" #include "llvm/Module.h" #include "llvm/Pass.h" #include "llvm/Analysis/LoopInfo.h" #include "llvm/Transforms/Utils/BasicBlockUtils.h" #include "llvm/Transforms/Utils/FunctionUtils.h" #include "Support/Debug.h" #include "Support/StringExtras.h" #include #include #include using namespace llvm; namespace { inline bool contains(const std::vector &V, const BasicBlock *BB){ return std::find(V.begin(), V.end(), BB) != V.end(); } /// getFunctionArg - Return a pointer to F's ARGNOth argument. /// Argument *getFunctionArg(Function *F, unsigned argno) { Function::aiterator ai = F->abegin(); while (argno) { ++ai; --argno; } return &*ai; } struct CodeExtractor { typedef std::vector Values; typedef std::vector > PhiValChangesTy; typedef std::map PhiVal2ArgTy; PhiVal2ArgTy PhiVal2Arg; public: Function *ExtractCodeRegion(const std::vector &code); private: void findInputsOutputs(const std::vector &code, Values &inputs, Values &outputs, BasicBlock *newHeader, BasicBlock *newRootNode); void processPhiNodeInputs(PHINode *Phi, const std::vector &code, Values &inputs, BasicBlock *newHeader, BasicBlock *newRootNode); void rewritePhiNodes(Function *F, BasicBlock *newFuncRoot); Function *constructFunction(const Values &inputs, const Values &outputs, BasicBlock *newRootNode, BasicBlock *newHeader, const std::vector &code, Function *oldFunction, Module *M); void moveCodeToFunction(const std::vector &code, Function *newFunction); void emitCallAndSwitchStatement(Function *newFunction, BasicBlock *newHeader, const std::vector &code, Values &inputs, Values &outputs); }; } void CodeExtractor::processPhiNodeInputs(PHINode *Phi, const std::vector &code, Values &inputs, BasicBlock *codeReplacer, BasicBlock *newFuncRoot) { // Separate incoming values and BasicBlocks as internal/external. We ignore // the case where both the value and BasicBlock are internal, because we don't // need to do a thing. std::vector EValEBB; std::vector EValIBB; std::vector IValEBB; for (unsigned i = 0, e = Phi->getNumIncomingValues(); i != e; ++i) { Value *phiVal = Phi->getIncomingValue(i); if (Instruction *Inst = dyn_cast(phiVal)) { if (contains(code, Inst->getParent())) { if (!contains(code, Phi->getIncomingBlock(i))) IValEBB.push_back(i); } else { if (contains(code, Phi->getIncomingBlock(i))) EValIBB.push_back(i); else EValEBB.push_back(i); } } else if (Constant *Const = dyn_cast(phiVal)) { // Constants are internal, but considered `external' if they are coming // from an external block. if (!contains(code, Phi->getIncomingBlock(i))) EValEBB.push_back(i); } else if (Argument *Arg = dyn_cast(phiVal)) { // arguments are external if (contains(code, Phi->getIncomingBlock(i))) EValIBB.push_back(i); else EValEBB.push_back(i); } else { phiVal->dump(); assert(0 && "Unhandled input in a Phi node"); } } // Both value and block are external. Need to group all of // these, have an external phi, pass the result as an // argument, and have THIS phi use that result. if (EValEBB.size() > 0) { if (EValEBB.size() == 1) { // Now if it's coming from the newFuncRoot, it's that funky input unsigned phiIdx = EValEBB[0]; if (!dyn_cast(Phi->getIncomingValue(phiIdx))) { PhiVal2Arg[Phi].push_back(std::make_pair(phiIdx, inputs.size())); // We can just pass this value in as argument inputs.push_back(Phi->getIncomingValue(phiIdx)); } Phi->setIncomingBlock(phiIdx, newFuncRoot); } else { PHINode *externalPhi = new PHINode(Phi->getType(), "extPhi"); codeReplacer->getInstList().insert(codeReplacer->begin(), externalPhi); for (std::vector::iterator i = EValEBB.begin(), e = EValEBB.end(); i != e; ++i) { externalPhi->addIncoming(Phi->getIncomingValue(*i), Phi->getIncomingBlock(*i)); // We make these values invalid instead of deleting them because that // would shift the indices of other values... The fixPhiNodes should // clean these phi nodes up later. Phi->setIncomingValue(*i, 0); Phi->setIncomingBlock(*i, 0); } PhiVal2Arg[Phi].push_back(std::make_pair(Phi->getNumIncomingValues(), inputs.size())); // We can just pass this value in as argument inputs.push_back(externalPhi); } } // When the value is external, but block internal... // just pass it in as argument, no change to phi node for (std::vector::iterator i = EValIBB.begin(), e = EValIBB.end(); i != e; ++i) { // rewrite the phi input node to be an argument PhiVal2Arg[Phi].push_back(std::make_pair(*i, inputs.size())); inputs.push_back(Phi->getIncomingValue(*i)); } // Value internal, block external // this can happen if we are extracting a part of a loop for (std::vector::iterator i = IValEBB.begin(), e = IValEBB.end(); i != e; ++i) { assert(0 && "Cannot (YET) handle internal values via external blocks"); } } void CodeExtractor::findInputsOutputs(const std::vector &code, Values &inputs, Values &outputs, BasicBlock *newHeader, BasicBlock *newRootNode) { for (std::vector::const_iterator ci = code.begin(), ce = code.end(); ci != ce; ++ci) { BasicBlock *BB = *ci; for (BasicBlock::iterator BBi = BB->begin(), BBe = BB->end(); BBi != BBe; ++BBi) { // If a use is defined outside the region, it's an input. // If a def is used outside the region, it's an output. if (Instruction *I = dyn_cast(&*BBi)) { // If it's a phi node if (PHINode *Phi = dyn_cast(I)) { processPhiNodeInputs(Phi, code, inputs, newHeader, newRootNode); } else { // All other instructions go through the generic input finder // Loop over the operands of each instruction (inputs) for (User::op_iterator op = I->op_begin(), opE = I->op_end(); op != opE; ++op) { if (Instruction *opI = dyn_cast(op->get())) { // Check if definition of this operand is within the loop if (!contains(code, opI->getParent())) { // add this operand to the inputs inputs.push_back(opI); } } } } // Consider uses of this instruction (outputs) for (Value::use_iterator use = I->use_begin(), useE = I->use_end(); use != useE; ++use) { if (Instruction* inst = dyn_cast(*use)) { if (!contains(code, inst->getParent())) { // add this op to the outputs outputs.push_back(I); } } } } /* if */ } /* for: insts */ } /* for: basic blocks */ } void CodeExtractor::rewritePhiNodes(Function *F, BasicBlock *newFuncRoot) { // Write any changes that were saved before: use function arguments as inputs for (PhiVal2ArgTy::iterator i = PhiVal2Arg.begin(), e = PhiVal2Arg.end(); i != e; ++i) { PHINode *phi = (*i).first; PhiValChangesTy &values = (*i).second; for (unsigned cIdx = 0, ce = values.size(); cIdx != ce; ++cIdx) { unsigned phiValueIdx = values[cIdx].first, argNum = values[cIdx].second; if (phiValueIdx < phi->getNumIncomingValues()) phi->setIncomingValue(phiValueIdx, getFunctionArg(F, argNum)); else phi->addIncoming(getFunctionArg(F, argNum), newFuncRoot); } } // Delete any invalid Phi node inputs that were marked as NULL previously for (PhiVal2ArgTy::iterator i = PhiVal2Arg.begin(), e = PhiVal2Arg.end(); i != e; ++i) { PHINode *phi = (*i).first; for (unsigned idx = 0, end = phi->getNumIncomingValues(); idx != end; ++idx) { if (phi->getIncomingValue(idx) == 0 && phi->getIncomingBlock(idx) == 0) { phi->removeIncomingValue(idx); --idx; --end; } } } // We are done with the saved values PhiVal2Arg.clear(); } /// constructFunction - make a function based on inputs and outputs, as follows: /// f(in0, ..., inN, out0, ..., outN) /// Function *CodeExtractor::constructFunction(const Values &inputs, const Values &outputs, BasicBlock *newRootNode, BasicBlock *newHeader, const std::vector &code, Function *oldFunction, Module *M) { DEBUG(std::cerr << "inputs: " << inputs.size() << "\n"); DEBUG(std::cerr << "outputs: " << outputs.size() << "\n"); BasicBlock *header = code[0]; // This function returns unsigned, outputs will go back by reference. Type *retTy = Type::UShortTy; std::vector paramTy; // Add the types of the input values to the function's argument list for (Values::const_iterator i = inputs.begin(), e = inputs.end(); i != e; ++i) { const Value *value = *i; DEBUG(std::cerr << "value used in func: " << value << "\n"); paramTy.push_back(value->getType()); } // Add the types of the output values to the function's argument list, but // make them pointer types for scalars for (Values::const_iterator i = outputs.begin(), e = outputs.end(); i != e; ++i) { const Value *value = *i; DEBUG(std::cerr << "instr used in func: " << value << "\n"); const Type *valueType = value->getType(); // Convert scalar types into a pointer of that type if (valueType->isPrimitiveType()) { valueType = PointerType::get(valueType); } paramTy.push_back(valueType); } DEBUG(std::cerr << "Function type: " << retTy << " f("); for (std::vector::iterator i = paramTy.begin(), e = paramTy.end(); i != e; ++i) DEBUG(std::cerr << (*i) << ", "); DEBUG(std::cerr << ")\n"); const FunctionType *funcType = FunctionType::get(retTy, paramTy, false); // Create the new function Function *newFunction = new Function(funcType, GlobalValue::InternalLinkage, oldFunction->getName() + "_code", M); newFunction->getBasicBlockList().push_back(newRootNode); for (unsigned i = 0, e = inputs.size(); i != e; ++i) { std::vector Users(inputs[i]->use_begin(), inputs[i]->use_end()); for (std::vector::iterator use = Users.begin(), useE = Users.end(); use != useE; ++use) { if (Instruction* inst = dyn_cast(*use)) { if (contains(code, inst->getParent())) { inst->replaceUsesOfWith(inputs[i], getFunctionArg(newFunction, i)); } } } } // Rewrite branches to basic blocks outside of the loop to new dummy blocks // within the new function. This must be done before we lose track of which // blocks were originally in the code region. std::vector Users(header->use_begin(), header->use_end()); for (std::vector::iterator i = Users.begin(), e = Users.end(); i != e; ++i) { if (BranchInst *inst = dyn_cast(*i)) { BasicBlock *BB = inst->getParent(); if (!contains(code, BB) && BB->getParent() == oldFunction) { // The BasicBlock which contains the branch is not in the region // modify the branch target to a new block inst->replaceUsesOfWith(header, newHeader); } } } return newFunction; } void CodeExtractor::moveCodeToFunction(const std::vector &code, Function *newFunction) { for (std::vector::const_iterator i = code.begin(), e =code.end(); i != e; ++i) { BasicBlock *BB = *i; Function *oldFunc = BB->getParent(); Function::BasicBlockListType &oldBlocks = oldFunc->getBasicBlockList(); // Delete the basic block from the old function, and the list of blocks oldBlocks.remove(BB); // Insert this basic block into the new function Function::BasicBlockListType &newBlocks = newFunction->getBasicBlockList(); newBlocks.push_back(BB); } } void CodeExtractor::emitCallAndSwitchStatement(Function *newFunction, BasicBlock *codeReplacer, const std::vector &code, Values &inputs, Values &outputs) { // Emit a call to the new function, passing allocated memory for outputs and // just plain inputs for non-scalars std::vector params; BasicBlock *codeReplacerTail = new BasicBlock("codeReplTail", codeReplacer->getParent()); for (Values::const_iterator i = inputs.begin(), e = inputs.end(); i != e; ++i) params.push_back(*i); for (Values::const_iterator i = outputs.begin(), e = outputs.end(); i != e; ++i) { // Create allocas for scalar outputs if ((*i)->getType()->isPrimitiveType()) { Constant *one = ConstantUInt::get(Type::UIntTy, 1); AllocaInst *alloca = new AllocaInst((*i)->getType(), one); codeReplacer->getInstList().push_back(alloca); params.push_back(alloca); LoadInst *load = new LoadInst(alloca, "alloca"); codeReplacerTail->getInstList().push_back(load); std::vector Users((*i)->use_begin(), (*i)->use_end()); for (std::vector::iterator use = Users.begin(), useE =Users.end(); use != useE; ++use) { if (Instruction* inst = dyn_cast(*use)) { if (!contains(code, inst->getParent())) { inst->replaceUsesOfWith(*i, load); } } } } else { params.push_back(*i); } } CallInst *call = new CallInst(newFunction, params, "targetBlock"); codeReplacer->getInstList().push_back(call); codeReplacer->getInstList().push_back(new BranchInst(codeReplacerTail)); // Now we can emit a switch statement using the call as a value. // FIXME: perhaps instead of default being self BB, it should be a second // dummy block which asserts that the value is not within the range...? //BasicBlock *defaultBlock = new BasicBlock("defaultBlock", oldF); //insert abort() ? //defaultBlock->getInstList().push_back(new BranchInst(codeReplacer)); SwitchInst *switchInst = new SwitchInst(call, codeReplacerTail, codeReplacerTail); // Since there may be multiple exits from the original region, make the new // function return an unsigned, switch on that number unsigned switchVal = 0; for (std::vector::const_iterator i =code.begin(), e = code.end(); i != e; ++i) { BasicBlock *BB = *i; // rewrite the terminator of the original BasicBlock Instruction *term = BB->getTerminator(); if (BranchInst *brInst = dyn_cast(term)) { // Restore values just before we exit // FIXME: Use a GetElementPtr to bunch the outputs in a struct for (unsigned outIdx = 0, outE = outputs.size(); outIdx != outE; ++outIdx) { new StoreInst(outputs[outIdx], getFunctionArg(newFunction, outIdx), brInst); } // Rewrite branches into exists which return a value based on which // exit we take from this function if (brInst->isUnconditional()) { if (!contains(code, brInst->getSuccessor(0))) { ConstantUInt *brVal = ConstantUInt::get(Type::UShortTy, switchVal); ReturnInst *newRet = new ReturnInst(brVal); // add a new target to the switch switchInst->addCase(brVal, brInst->getSuccessor(0)); ++switchVal; // rewrite the branch with a return BasicBlock::iterator ii(brInst); ReplaceInstWithInst(BB->getInstList(), ii, newRet); delete brInst; } } else { // Replace the conditional branch to branch // to two new blocks, each of which returns a different code. for (unsigned idx = 0; idx < 2; ++idx) { BasicBlock *oldTarget = brInst->getSuccessor(idx); if (!contains(code, oldTarget)) { // add a new basic block which returns the appropriate value BasicBlock *newTarget = new BasicBlock("newTarget", newFunction); ConstantUInt *brVal = ConstantUInt::get(Type::UShortTy, switchVal); ReturnInst *newRet = new ReturnInst(brVal); newTarget->getInstList().push_back(newRet); // rewrite the original branch instruction with this new target brInst->setSuccessor(idx, newTarget); // the switch statement knows what to do with this value switchInst->addCase(brVal, oldTarget); ++switchVal; } } } } else if (ReturnInst *retTerm = dyn_cast(term)) { assert(0 && "Cannot handle return instructions just yet."); // FIXME: what if the terminator is a return!??! // Need to rewrite: add new basic block, move the return there // treat the original as an unconditional branch to that basicblock } else if (SwitchInst *swTerm = dyn_cast(term)) { assert(0 && "Cannot handle switch instructions just yet."); } else if (InvokeInst *invInst = dyn_cast(term)) { assert(0 && "Cannot handle invoke instructions just yet."); } else { assert(0 && "Unrecognized terminator, or badly-formed BasicBlock."); } } } /// ExtractRegion - Removes a loop from a function, replaces it with a call to /// new function. Returns pointer to the new function. /// /// algorithm: /// /// find inputs and outputs for the region /// /// for inputs: add to function as args, map input instr* to arg# /// for outputs: add allocas for scalars, /// add to func as args, map output instr* to arg# /// /// rewrite func to use argument #s instead of instr* /// /// for each scalar output in the function: at every exit, store intermediate /// computed result back into memory. /// Function *CodeExtractor::ExtractCodeRegion(const std::vector &code) { // 1) Find inputs, outputs // 2) Construct new function // * Add allocas for defs, pass as args by reference // * Pass in uses as args // 3) Move code region, add call instr to func // Values inputs, outputs; // Assumption: this is a single-entry code region, and the header is the first // block in the region. FIXME: is this true for a list of blocks from a // natural function? BasicBlock *header = code[0]; Function *oldFunction = header->getParent(); Module *module = oldFunction->getParent(); // This takes place of the original loop BasicBlock *codeReplacer = new BasicBlock("codeRepl", oldFunction); // The new function needs a root node because other nodes can branch to the // head of the loop, and the root cannot have predecessors BasicBlock *newFuncRoot = new BasicBlock("newFuncRoot"); newFuncRoot->getInstList().push_back(new BranchInst(header)); // Find inputs to, outputs from the code region // // If one of the inputs is coming from a different basic block and it's in a // phi node, we need to rewrite the phi node: // // * All the inputs which involve basic blocks OUTSIDE of this region go into // a NEW phi node that takes care of finding which value really came in. // The result of this phi is passed to the function as an argument. // // * All the other phi values stay. // // FIXME: PHI nodes' incoming blocks aren't being rewritten to accomodate for // blocks moving to a new function. // SOLUTION: move Phi nodes out of the loop header into the codeReplacer, pass // the values as parameters to the function findInputsOutputs(code, inputs, outputs, codeReplacer, newFuncRoot); // Step 2: Construct new function based on inputs/outputs, // Add allocas for all defs Function *newFunction = constructFunction(inputs, outputs, newFuncRoot, codeReplacer, code, oldFunction, module); rewritePhiNodes(newFunction, newFuncRoot); emitCallAndSwitchStatement(newFunction, codeReplacer, code, inputs, outputs); moveCodeToFunction(code, newFunction); return newFunction; } Function* llvm::ExtractLoop(Loop *L) { CodeExtractor CE; return CE.ExtractCodeRegion(L->getBlocks()); }