//===- 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 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
//
//===----------------------------------------------------------------------===//
-#include "llvm/BasicBlock.h"
+#include "llvm/Transforms/Utils/FunctionUtils.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Instructions.h"
+#include "llvm/Intrinsics.h"
+#include "llvm/LLVMContext.h"
#include "llvm/Module.h"
#include "llvm/Pass.h"
+#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/Analysis/Verifier.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
-#include "llvm/Transforms/Utils/FunctionUtils.h"
-#include "Support/Debug.h"
-#include "Support/StringExtras.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Compiler.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/ADT/StringExtras.h"
#include <algorithm>
-#include <map>
-#include <vector>
+#include <set>
using namespace llvm;
-namespace {
-
- inline bool contains(const std::vector<BasicBlock*> &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;
- }
+// Provide a command-line option to aggregate function arguments into a struct
+// for functions produced by the code extractor. This is useful when converting
+// extracted functions to pthread-based code, as only one argument (void*) can
+// be passed in to pthread_create().
+static cl::opt<bool>
+AggregateArgsOpt("aggregate-extracted-args", cl::Hidden,
+ cl::desc("Aggregate arguments to code-extracted functions"));
- struct CodeExtractor {
+namespace {
+ class VISIBILITY_HIDDEN CodeExtractor {
typedef std::vector<Value*> Values;
- typedef std::vector<std::pair<unsigned, unsigned> > PhiValChangesTy;
- typedef std::map<PHINode*, PhiValChangesTy> PhiVal2ArgTy;
- PhiVal2ArgTy PhiVal2Arg;
-
+ std::set<BasicBlock*> BlocksToExtract;
+ DominatorTree* DT;
+ bool AggregateArgs;
+ unsigned NumExitBlocks;
+ const Type *RetTy;
public:
+ CodeExtractor(DominatorTree* dt = 0, bool AggArgs = false)
+ : DT(dt), AggregateArgs(AggArgs||AggregateArgsOpt), NumExitBlocks(~0U) {}
+
Function *ExtractCodeRegion(const std::vector<BasicBlock*> &code);
+ bool isEligible(const std::vector<BasicBlock*> &code);
+
private:
- void findInputsOutputs(const std::vector<BasicBlock*> &code,
- Values &inputs,
- Values &outputs,
- BasicBlock *newHeader,
- BasicBlock *newRootNode);
+ /// definedInRegion - Return true if the specified value is defined in the
+ /// extracted region.
+ bool definedInRegion(Value *V) const {
+ if (Instruction *I = dyn_cast<Instruction>(V))
+ if (BlocksToExtract.count(I->getParent()))
+ return true;
+ return false;
+ }
- void processPhiNodeInputs(PHINode *Phi,
- const std::vector<BasicBlock*> &code,
- Values &inputs,
- BasicBlock *newHeader,
- BasicBlock *newRootNode);
+ /// definedInCaller - Return true if the specified value is defined in the
+ /// function being code extracted, but not in the region being extracted.
+ /// These values must be passed in as live-ins to the function.
+ bool definedInCaller(Value *V) const {
+ if (isa<Argument>(V)) return true;
+ if (Instruction *I = dyn_cast<Instruction>(V))
+ if (!BlocksToExtract.count(I->getParent()))
+ return true;
+ return false;
+ }
- void rewritePhiNodes(Function *F, BasicBlock *newFuncRoot);
+ void severSplitPHINodes(BasicBlock *&Header);
+ void splitReturnBlocks();
+ void findInputsOutputs(Values &inputs, Values &outputs);
Function *constructFunction(const Values &inputs,
const Values &outputs,
+ BasicBlock *header,
BasicBlock *newRootNode, BasicBlock *newHeader,
- const std::vector<BasicBlock*> &code,
Function *oldFunction, Module *M);
- void moveCodeToFunction(const std::vector<BasicBlock*> &code,
- Function *newFunction);
+ void moveCodeToFunction(Function *newFunction);
void emitCallAndSwitchStatement(Function *newFunction,
BasicBlock *newHeader,
- const std::vector<BasicBlock*> &code,
Values &inputs,
Values &outputs);
};
}
-void CodeExtractor::processPhiNodeInputs(PHINode *Phi,
- const std::vector<BasicBlock*> &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<unsigned> EValEBB;
- std::vector<unsigned> EValIBB;
- std::vector<unsigned> IValEBB;
-
- for (unsigned i = 0, e = Phi->getNumIncomingValues(); i != e; ++i) {
- Value *phiVal = Phi->getIncomingValue(i);
- if (Instruction *Inst = dyn_cast<Instruction>(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<Constant>(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<Argument>(phiVal)) {
- // arguments are external
- if (contains(code, Phi->getIncomingBlock(i)))
- EValIBB.push_back(i);
+/// severSplitPHINodes - If a PHI node has multiple inputs from outside of the
+/// region, we need to split the entry block of the region so that the PHI node
+/// is easier to deal with.
+void CodeExtractor::severSplitPHINodes(BasicBlock *&Header) {
+ bool HasPredsFromRegion = false;
+ unsigned NumPredsOutsideRegion = 0;
+
+ if (Header != &Header->getParent()->getEntryBlock()) {
+ PHINode *PN = dyn_cast<PHINode>(Header->begin());
+ if (!PN) return; // No PHI nodes.
+
+ // If the header node contains any PHI nodes, check to see if there is more
+ // than one entry from outside the region. If so, we need to sever the
+ // header block into two.
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
+ if (BlocksToExtract.count(PN->getIncomingBlock(i)))
+ HasPredsFromRegion = true;
else
- EValEBB.push_back(i);
- } else {
- phiVal->dump();
- assert(0 && "Unhandled input in a Phi node");
- }
+ ++NumPredsOutsideRegion;
+
+ // If there is one (or fewer) predecessor from outside the region, we don't
+ // need to do anything special.
+ if (NumPredsOutsideRegion <= 1) return;
}
- // 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<Constant>(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));
+ // Otherwise, we need to split the header block into two pieces: one
+ // containing PHI nodes merging values from outside of the region, and a
+ // second that contains all of the code for the block and merges back any
+ // incoming values from inside of the region.
+ BasicBlock::iterator AfterPHIs = Header->getFirstNonPHI();
+ BasicBlock *NewBB = Header->splitBasicBlock(AfterPHIs,
+ Header->getName()+".ce");
+
+ // We only want to code extract the second block now, and it becomes the new
+ // header of the region.
+ BasicBlock *OldPred = Header;
+ BlocksToExtract.erase(OldPred);
+ BlocksToExtract.insert(NewBB);
+ Header = NewBB;
+
+ // Okay, update dominator sets. The blocks that dominate the new one are the
+ // blocks that dominate TIBB plus the new block itself.
+ if (DT)
+ DT->splitBlock(NewBB);
+
+ // Okay, now we need to adjust the PHI nodes and any branches from within the
+ // region to go to the new header block instead of the old header block.
+ if (HasPredsFromRegion) {
+ PHINode *PN = cast<PHINode>(OldPred->begin());
+ // Loop over all of the predecessors of OldPred that are in the region,
+ // changing them to branch to NewBB instead.
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
+ if (BlocksToExtract.count(PN->getIncomingBlock(i))) {
+ TerminatorInst *TI = PN->getIncomingBlock(i)->getTerminator();
+ TI->replaceUsesOfWith(OldPred, NewBB);
}
- Phi->setIncomingBlock(phiIdx, newFuncRoot);
- } else {
- PHINode *externalPhi = new PHINode(Phi->getType(), "extPhi");
- codeReplacer->getInstList().insert(codeReplacer->begin(), externalPhi);
- for (std::vector<unsigned>::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);
+
+ // Okay, everthing within the region is now branching to the right block, we
+ // just have to update the PHI nodes now, inserting PHI nodes into NewBB.
+ for (AfterPHIs = OldPred->begin(); isa<PHINode>(AfterPHIs); ++AfterPHIs) {
+ PHINode *PN = cast<PHINode>(AfterPHIs);
+ // Create a new PHI node in the new region, which has an incoming value
+ // from OldPred of PN.
+ PHINode *NewPN = PHINode::Create(PN->getType(), PN->getName()+".ce",
+ NewBB->begin());
+ NewPN->addIncoming(PN, OldPred);
+
+ // Loop over all of the incoming value in PN, moving them to NewPN if they
+ // are from the extracted region.
+ for (unsigned i = 0; i != PN->getNumIncomingValues(); ++i) {
+ if (BlocksToExtract.count(PN->getIncomingBlock(i))) {
+ NewPN->addIncoming(PN->getIncomingValue(i), PN->getIncomingBlock(i));
+ PN->removeIncomingValue(i);
+ --i;
+ }
}
- 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<unsigned>::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<unsigned>::iterator i = IValEBB.begin(),
- e = IValEBB.end(); i != e; ++i)
- {
- assert(0 && "Cannot (YET) handle internal values via external blocks");
- }
}
+void CodeExtractor::splitReturnBlocks() {
+ for (std::set<BasicBlock*>::iterator I = BlocksToExtract.begin(),
+ E = BlocksToExtract.end(); I != E; ++I)
+ if (ReturnInst *RI = dyn_cast<ReturnInst>((*I)->getTerminator()))
+ (*I)->splitBasicBlock(RI, (*I)->getName()+".ret");
+}
-void CodeExtractor::findInputsOutputs(const std::vector<BasicBlock*> &code,
- Values &inputs,
- Values &outputs,
- BasicBlock *newHeader,
- BasicBlock *newRootNode)
-{
- for (std::vector<BasicBlock*>::const_iterator ci = code.begin(),
- ce = code.end(); ci != ce; ++ci) {
+// findInputsOutputs - Find inputs to, outputs from the code region.
+//
+void CodeExtractor::findInputsOutputs(Values &inputs, Values &outputs) {
+ std::set<BasicBlock*> ExitBlocks;
+ for (std::set<BasicBlock*>::const_iterator ci = BlocksToExtract.begin(),
+ ce = BlocksToExtract.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<Instruction>(&*BBi)) {
- // If it's a phi node
- if (PHINode *Phi = dyn_cast<PHINode>(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<Instruction>(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<Instruction>(*use)) {
- if (!contains(code, inst->getParent())) {
- // add this op to the outputs
- outputs.push_back(I);
- }
- }
+ for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
+ // If a used value is defined outside the region, it's an input. If an
+ // instruction is used outside the region, it's an output.
+ for (User::op_iterator O = I->op_begin(), E = I->op_end(); O != E; ++O)
+ if (definedInCaller(*O))
+ inputs.push_back(*O);
+
+ // Consider uses of this instruction (outputs).
+ for (Value::use_iterator UI = I->use_begin(), E = I->use_end();
+ UI != E; ++UI)
+ if (!definedInRegion(*UI)) {
+ outputs.push_back(I);
+ break;
}
- } /* 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();
+ } // for: insts
+
+ // Keep track of the exit blocks from the region.
+ TerminatorInst *TI = BB->getTerminator();
+ for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
+ if (!BlocksToExtract.count(TI->getSuccessor(i)))
+ ExitBlocks.insert(TI->getSuccessor(i));
+ } // for: basic blocks
+
+ NumExitBlocks = ExitBlocks.size();
+
+ // Eliminate duplicates.
+ std::sort(inputs.begin(), inputs.end());
+ inputs.erase(std::unique(inputs.begin(), inputs.end()), inputs.end());
+ std::sort(outputs.begin(), outputs.end());
+ outputs.erase(std::unique(outputs.begin(), outputs.end()), outputs.end());
}
-
/// 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 *header,
BasicBlock *newRootNode,
BasicBlock *newHeader,
- const std::vector<BasicBlock*> &code,
- Function *oldFunction, Module *M) {
- DEBUG(std::cerr << "inputs: " << inputs.size() << "\n");
- DEBUG(std::cerr << "outputs: " << outputs.size() << "\n");
- BasicBlock *header = code[0];
+ Function *oldFunction,
+ Module *M) {
+ DOUT << "inputs: " << inputs.size() << "\n";
+ DOUT << "outputs: " << outputs.size() << "\n";
+
+ LLVMContext *Context = header->getContext();
// This function returns unsigned, outputs will go back by reference.
- Type *retTy = Type::UShortTy;
+ switch (NumExitBlocks) {
+ case 0:
+ case 1: RetTy = Type::VoidTy; break;
+ case 2: RetTy = Type::Int1Ty; break;
+ default: RetTy = Type::Int16Ty; break;
+ }
+
std::vector<const Type*> 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");
+ DOUT << "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);
+ // Add the types of the output values to the function's argument list.
+ for (Values::const_iterator I = outputs.begin(), E = outputs.end();
+ I != E; ++I) {
+ DOUT << "instr used in func: " << **I << "\n";
+ if (AggregateArgs)
+ paramTy.push_back((*I)->getType());
+ else
+ paramTy.push_back(
+ header->getContext()->getPointerTypeUnqual((*I)->getType()));
}
- DEBUG(std::cerr << "Function type: " << retTy << " f(");
+ DOUT << "Function type: " << *RetTy << " f(";
for (std::vector<const Type*>::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);
+ DOUT << **i << ", ";
+ DOUT << ")\n";
+
+ if (AggregateArgs && (inputs.size() + outputs.size() > 0)) {
+ PointerType *StructPtr =
+ Context->getPointerTypeUnqual(Context->getStructType(paramTy));
+ paramTy.clear();
+ paramTy.push_back(StructPtr);
+ }
+ const FunctionType *funcType =
+ Context->getFunctionType(RetTy, paramTy, false);
// Create the new function
- Function *newFunction = new Function(funcType,
- GlobalValue::InternalLinkage,
- oldFunction->getName() + "_code", M);
+ Function *newFunction = Function::Create(funcType,
+ GlobalValue::InternalLinkage,
+ oldFunction->getName() + "_" +
+ header->getName(), M);
+ // If the old function is no-throw, so is the new one.
+ if (oldFunction->doesNotThrow())
+ newFunction->setDoesNotThrow(true);
+
newFunction->getBasicBlockList().push_back(newRootNode);
+ // Create an iterator to name all of the arguments we inserted.
+ Function::arg_iterator AI = newFunction->arg_begin();
+
+ // Rewrite all users of the inputs in the extracted region to use the
+ // arguments (or appropriate addressing into struct) instead.
for (unsigned i = 0, e = inputs.size(); i != e; ++i) {
+ Value *RewriteVal;
+ if (AggregateArgs) {
+ Value *Idx[2];
+ Idx[0] = Context->getNullValue(Type::Int32Ty);
+ Idx[1] = Context->getConstantInt(Type::Int32Ty, i);
+ std::string GEPname = "gep_" + inputs[i]->getName();
+ TerminatorInst *TI = newFunction->begin()->getTerminator();
+ GetElementPtrInst *GEP = GetElementPtrInst::Create(AI, Idx, Idx+2,
+ GEPname, TI);
+ RewriteVal = new LoadInst(GEP, "load" + GEPname, TI);
+ } else
+ RewriteVal = AI++;
+
std::vector<User*> Users(inputs[i]->use_begin(), inputs[i]->use_end());
for (std::vector<User*>::iterator use = Users.begin(), useE = Users.end();
- use != useE; ++use) {
- if (Instruction* inst = dyn_cast<Instruction>(*use)) {
- if (contains(code, inst->getParent())) {
- inst->replaceUsesOfWith(inputs[i], getFunctionArg(newFunction, i));
- }
- }
- }
+ use != useE; ++use)
+ if (Instruction* inst = dyn_cast<Instruction>(*use))
+ if (BlocksToExtract.count(inst->getParent()))
+ inst->replaceUsesOfWith(inputs[i], RewriteVal);
+ }
+
+ // Set names for input and output arguments.
+ if (!AggregateArgs) {
+ AI = newFunction->arg_begin();
+ for (unsigned i = 0, e = inputs.size(); i != e; ++i, ++AI)
+ AI->setName(inputs[i]->getName());
+ for (unsigned i = 0, e = outputs.size(); i != e; ++i, ++AI)
+ AI->setName(outputs[i]->getName()+".out");
}
// 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<User*> Users(header->use_begin(), header->use_end());
- for (std::vector<User*>::iterator i = Users.begin(), e = Users.end();
- i != e; ++i) {
- if (BranchInst *inst = dyn_cast<BranchInst>(*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);
- }
- }
- }
+ for (unsigned i = 0, e = Users.size(); i != e; ++i)
+ // The BasicBlock which contains the branch is not in the region
+ // modify the branch target to a new block
+ if (TerminatorInst *TI = dyn_cast<TerminatorInst>(Users[i]))
+ if (!BlocksToExtract.count(TI->getParent()) &&
+ TI->getParent()->getParent() == oldFunction)
+ TI->replaceUsesOfWith(header, newHeader);
return newFunction;
}
-void CodeExtractor::moveCodeToFunction(const std::vector<BasicBlock*> &code,
- Function *newFunction)
-{
- for (std::vector<BasicBlock*>::const_iterator i = code.begin(), e =code.end();
- i != e; ++i) {
- BasicBlock *BB = *i;
- Function *oldFunc = BB->getParent();
- Function::BasicBlockListType &oldBlocks = oldFunc->getBasicBlockList();
+/// emitCallAndSwitchStatement - This method sets up the caller side by adding
+/// the call instruction, splitting any PHI nodes in the header block as
+/// necessary.
+void CodeExtractor::
+emitCallAndSwitchStatement(Function *newFunction, BasicBlock *codeReplacer,
+ Values &inputs, Values &outputs) {
+ LLVMContext *Context = codeReplacer->getContext();
+
+ // Emit a call to the new function, passing in: *pointer to struct (if
+ // aggregating parameters), or plan inputs and allocated memory for outputs
+ std::vector<Value*> params, StructValues, ReloadOutputs;
+
+ // Add inputs as params, or to be filled into the struct
+ for (Values::iterator i = inputs.begin(), e = inputs.end(); i != e; ++i)
+ if (AggregateArgs)
+ StructValues.push_back(*i);
+ else
+ params.push_back(*i);
- // Delete the basic block from the old function, and the list of blocks
- oldBlocks.remove(BB);
+ // Create allocas for the outputs
+ for (Values::iterator i = outputs.begin(), e = outputs.end(); i != e; ++i) {
+ if (AggregateArgs) {
+ StructValues.push_back(*i);
+ } else {
+ AllocaInst *alloca =
+ new AllocaInst(*codeReplacer->getContext(),
+ (*i)->getType(), 0, (*i)->getName()+".loc",
+ codeReplacer->getParent()->begin()->begin());
+ ReloadOutputs.push_back(alloca);
+ params.push_back(alloca);
+ }
+ }
- // Insert this basic block into the new function
- Function::BasicBlockListType &newBlocks = newFunction->getBasicBlockList();
- newBlocks.push_back(BB);
+ AllocaInst *Struct = 0;
+ if (AggregateArgs && (inputs.size() + outputs.size() > 0)) {
+ std::vector<const Type*> ArgTypes;
+ for (Values::iterator v = StructValues.begin(),
+ ve = StructValues.end(); v != ve; ++v)
+ ArgTypes.push_back((*v)->getType());
+
+ // Allocate a struct at the beginning of this function
+ Type *StructArgTy = Context->getStructType(ArgTypes);
+ Struct =
+ new AllocaInst(*codeReplacer->getContext(), StructArgTy, 0, "structArg",
+ codeReplacer->getParent()->begin()->begin());
+ params.push_back(Struct);
+
+ for (unsigned i = 0, e = inputs.size(); i != e; ++i) {
+ Value *Idx[2];
+ Idx[0] = Context->getNullValue(Type::Int32Ty);
+ Idx[1] = Context->getConstantInt(Type::Int32Ty, i);
+ GetElementPtrInst *GEP =
+ GetElementPtrInst::Create(Struct, Idx, Idx + 2,
+ "gep_" + StructValues[i]->getName());
+ codeReplacer->getInstList().push_back(GEP);
+ StoreInst *SI = new StoreInst(StructValues[i], GEP);
+ codeReplacer->getInstList().push_back(SI);
+ }
}
-}
-void
-CodeExtractor::emitCallAndSwitchStatement(Function *newFunction,
- BasicBlock *codeReplacer,
- const std::vector<BasicBlock*> &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<Value*> 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);
+ // Emit the call to the function
+ CallInst *call = CallInst::Create(newFunction, params.begin(), params.end(),
+ NumExitBlocks > 1 ? "targetBlock" : "");
+ codeReplacer->getInstList().push_back(call);
- LoadInst *load = new LoadInst(alloca, "alloca");
- codeReplacerTail->getInstList().push_back(load);
- std::vector<User*> Users((*i)->use_begin(), (*i)->use_end());
- for (std::vector<User*>::iterator use = Users.begin(), useE =Users.end();
- use != useE; ++use) {
- if (Instruction* inst = dyn_cast<Instruction>(*use)) {
- if (!contains(code, inst->getParent())) {
- inst->replaceUsesOfWith(*i, load);
- }
- }
- }
+ Function::arg_iterator OutputArgBegin = newFunction->arg_begin();
+ unsigned FirstOut = inputs.size();
+ if (!AggregateArgs)
+ std::advance(OutputArgBegin, inputs.size());
+
+ // Reload the outputs passed in by reference
+ for (unsigned i = 0, e = outputs.size(); i != e; ++i) {
+ Value *Output = 0;
+ if (AggregateArgs) {
+ Value *Idx[2];
+ Idx[0] = Context->getNullValue(Type::Int32Ty);
+ Idx[1] = Context->getConstantInt(Type::Int32Ty, FirstOut + i);
+ GetElementPtrInst *GEP
+ = GetElementPtrInst::Create(Struct, Idx, Idx + 2,
+ "gep_reload_" + outputs[i]->getName());
+ codeReplacer->getInstList().push_back(GEP);
+ Output = GEP;
} else {
- params.push_back(*i);
+ Output = ReloadOutputs[i];
+ }
+ LoadInst *load = new LoadInst(Output, outputs[i]->getName()+".reload");
+ codeReplacer->getInstList().push_back(load);
+ std::vector<User*> Users(outputs[i]->use_begin(), outputs[i]->use_end());
+ for (unsigned u = 0, e = Users.size(); u != e; ++u) {
+ Instruction *inst = cast<Instruction>(Users[u]);
+ if (!BlocksToExtract.count(inst->getParent()))
+ inst->replaceUsesOfWith(outputs[i], load);
}
}
- 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);
+ SwitchInst *TheSwitch =
+ SwitchInst::Create(Context->getNullValue(Type::Int16Ty),
+ codeReplacer, 0, codeReplacer);
// Since there may be multiple exits from the original region, make the new
- // function return an unsigned, switch on that number
+ // function return an unsigned, switch on that number. This loop iterates
+ // over all of the blocks in the extracted region, updating any terminator
+ // instructions in the to-be-extracted region that branch to blocks that are
+ // not in the region to be extracted.
+ std::map<BasicBlock*, BasicBlock*> ExitBlockMap;
+
unsigned switchVal = 0;
- for (std::vector<BasicBlock*>::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<BranchInst>(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);
- }
+ for (std::set<BasicBlock*>::const_iterator i = BlocksToExtract.begin(),
+ e = BlocksToExtract.end(); i != e; ++i) {
+ TerminatorInst *TI = (*i)->getTerminator();
+ for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
+ if (!BlocksToExtract.count(TI->getSuccessor(i))) {
+ BasicBlock *OldTarget = TI->getSuccessor(i);
+ // add a new basic block which returns the appropriate value
+ BasicBlock *&NewTarget = ExitBlockMap[OldTarget];
+ if (!NewTarget) {
+ // If we don't already have an exit stub for this non-extracted
+ // destination, create one now!
+ NewTarget = BasicBlock::Create(OldTarget->getName() + ".exitStub",
+ newFunction);
+ unsigned SuccNum = switchVal++;
+
+ Value *brVal = 0;
+ switch (NumExitBlocks) {
+ case 0:
+ case 1: break; // No value needed.
+ case 2: // Conditional branch, return a bool
+ brVal = Context->getConstantInt(Type::Int1Ty, !SuccNum);
+ break;
+ default:
+ brVal = Context->getConstantInt(Type::Int16Ty, SuccNum);
+ break;
+ }
- // 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;
+ ReturnInst *NTRet = ReturnInst::Create(brVal, NewTarget);
+
+ // Update the switch instruction.
+ TheSwitch->addCase(Context->getConstantInt(Type::Int16Ty, SuccNum),
+ OldTarget);
+
+ // Restore values just before we exit
+ Function::arg_iterator OAI = OutputArgBegin;
+ for (unsigned out = 0, e = outputs.size(); out != e; ++out) {
+ // For an invoke, the normal destination is the only one that is
+ // dominated by the result of the invocation
+ BasicBlock *DefBlock = cast<Instruction>(outputs[out])->getParent();
+
+ bool DominatesDef = true;
+
+ if (InvokeInst *Invoke = dyn_cast<InvokeInst>(outputs[out])) {
+ DefBlock = Invoke->getNormalDest();
+
+ // Make sure we are looking at the original successor block, not
+ // at a newly inserted exit block, which won't be in the dominator
+ // info.
+ for (std::map<BasicBlock*, BasicBlock*>::iterator I =
+ ExitBlockMap.begin(), E = ExitBlockMap.end(); I != E; ++I)
+ if (DefBlock == I->second) {
+ DefBlock = I->first;
+ break;
+ }
+
+ // In the extract block case, if the block we are extracting ends
+ // with an invoke instruction, make sure that we don't emit a
+ // store of the invoke value for the unwind block.
+ if (!DT && DefBlock != OldTarget)
+ DominatesDef = false;
+ }
+
+ if (DT)
+ DominatesDef = DT->dominates(DefBlock, OldTarget);
+
+ if (DominatesDef) {
+ if (AggregateArgs) {
+ Value *Idx[2];
+ Idx[0] = Context->getNullValue(Type::Int32Ty);
+ Idx[1] = Context->getConstantInt(Type::Int32Ty,FirstOut+out);
+ GetElementPtrInst *GEP =
+ GetElementPtrInst::Create(OAI, Idx, Idx + 2,
+ "gep_" + outputs[out]->getName(),
+ NTRet);
+ new StoreInst(outputs[out], GEP, NTRet);
+ } else {
+ new StoreInst(outputs[out], OAI, NTRet);
+ }
+ }
+ // Advance output iterator even if we don't emit a store
+ if (!AggregateArgs) ++OAI;
}
}
+
+ // rewrite the original branch instruction with this new target
+ TI->setSuccessor(i, NewTarget);
}
- } else if (ReturnInst *retTerm = dyn_cast<ReturnInst>(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<SwitchInst>(term)) {
- assert(0 && "Cannot handle switch instructions just yet.");
- } else if (InvokeInst *invInst = dyn_cast<InvokeInst>(term)) {
- assert(0 && "Cannot handle invoke instructions just yet.");
+ }
+
+ // Now that we've done the deed, simplify the switch instruction.
+ const Type *OldFnRetTy = TheSwitch->getParent()->getParent()->getReturnType();
+ switch (NumExitBlocks) {
+ case 0:
+ // There are no successors (the block containing the switch itself), which
+ // means that previously this was the last part of the function, and hence
+ // this should be rewritten as a `ret'
+
+ // Check if the function should return a value
+ if (OldFnRetTy == Type::VoidTy) {
+ ReturnInst::Create(0, TheSwitch); // Return void
+ } else if (OldFnRetTy == TheSwitch->getCondition()->getType()) {
+ // return what we have
+ ReturnInst::Create(TheSwitch->getCondition(), TheSwitch);
} else {
- assert(0 && "Unrecognized terminator, or badly-formed BasicBlock.");
+ // Otherwise we must have code extracted an unwind or something, just
+ // return whatever we want.
+ ReturnInst::Create(Context->getNullValue(OldFnRetTy), TheSwitch);
}
+
+ TheSwitch->eraseFromParent();
+ break;
+ case 1:
+ // Only a single destination, change the switch into an unconditional
+ // branch.
+ BranchInst::Create(TheSwitch->getSuccessor(1), TheSwitch);
+ TheSwitch->eraseFromParent();
+ break;
+ case 2:
+ BranchInst::Create(TheSwitch->getSuccessor(1), TheSwitch->getSuccessor(2),
+ call, TheSwitch);
+ TheSwitch->eraseFromParent();
+ break;
+ default:
+ // Otherwise, make the default destination of the switch instruction be one
+ // of the other successors.
+ TheSwitch->setOperand(0, call);
+ TheSwitch->setSuccessor(0, TheSwitch->getSuccessor(NumExitBlocks));
+ TheSwitch->removeCase(NumExitBlocks); // Remove redundant case
+ break;
}
}
+void CodeExtractor::moveCodeToFunction(Function *newFunction) {
+ Function *oldFunc = (*BlocksToExtract.begin())->getParent();
+ Function::BasicBlockListType &oldBlocks = oldFunc->getBasicBlockList();
+ Function::BasicBlockListType &newBlocks = newFunction->getBasicBlockList();
+
+ for (std::set<BasicBlock*>::const_iterator i = BlocksToExtract.begin(),
+ e = BlocksToExtract.end(); i != e; ++i) {
+ // Delete the basic block from the old function, and the list of blocks
+ oldBlocks.remove(*i);
+
+ // Insert this basic block into the new function
+ newBlocks.push_back(*i);
+ }
+}
/// ExtractRegion - Removes a loop from a function, replaces it with a call to
/// new function. Returns pointer to the new function.
///
/// 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,
+/// 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
+/// for each scalar output in the function: at every exit, store intermediate
/// computed result back into memory.
///
-Function *CodeExtractor::ExtractCodeRegion(const std::vector<BasicBlock*> &code)
-{
+Function *CodeExtractor::
+ExtractCodeRegion(const std::vector<BasicBlock*> &code) {
+ if (!isEligible(code))
+ return 0;
+
// 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
- //
+ //
+ BlocksToExtract.insert(code.begin(), code.end());
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?
+ // block in the region.
BasicBlock *header = code[0];
+
+ for (unsigned i = 1, e = code.size(); i != e; ++i)
+ for (pred_iterator PI = pred_begin(code[i]), E = pred_end(code[i]);
+ PI != E; ++PI)
+ assert(BlocksToExtract.count(*PI) &&
+ "No blocks in this region may have entries from outside the region"
+ " except for the first block!");
+
+ // If we have to split PHI nodes or the entry block, do so now.
+ severSplitPHINodes(header);
+
+ // If we have any return instructions in the region, split those blocks so
+ // that the return is not in the region.
+ splitReturnBlocks();
+
Function *oldFunction = header->getParent();
- Module *module = oldFunction->getParent();
// This takes place of the original loop
- BasicBlock *codeReplacer = new BasicBlock("codeRepl", oldFunction);
+ BasicBlock *codeReplacer = BasicBlock::Create("codeRepl", oldFunction,
+ header);
// 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));
+ // head of the region, but the entry node of a function cannot have preds.
+ BasicBlock *newFuncRoot = BasicBlock::Create("newFuncRoot");
+ newFuncRoot->getInstList().push_back(BranchInst::Create(header));
+
+ // Find inputs to, outputs from the code region.
+ findInputsOutputs(inputs, outputs);
+
+ // Construct new function based on inputs/outputs & add allocas for all defs.
+ Function *newFunction = constructFunction(inputs, outputs, header,
+ newFuncRoot,
+ codeReplacer, oldFunction,
+ oldFunction->getParent());
+
+ emitCallAndSwitchStatement(newFunction, codeReplacer, inputs, outputs);
+
+ moveCodeToFunction(newFunction);
+
+ // Loop over all of the PHI nodes in the header block, and change any
+ // references to the old incoming edge to be the new incoming edge.
+ for (BasicBlock::iterator I = header->begin(); isa<PHINode>(I); ++I) {
+ PHINode *PN = cast<PHINode>(I);
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
+ if (!BlocksToExtract.count(PN->getIncomingBlock(i)))
+ PN->setIncomingBlock(i, newFuncRoot);
+ }
- // 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);
+ // Look at all successors of the codeReplacer block. If any of these blocks
+ // had PHI nodes in them, we need to update the "from" block to be the code
+ // replacer, not the original block in the extracted region.
+ std::vector<BasicBlock*> Succs(succ_begin(codeReplacer),
+ succ_end(codeReplacer));
+ for (unsigned i = 0, e = Succs.size(); i != e; ++i)
+ for (BasicBlock::iterator I = Succs[i]->begin(); isa<PHINode>(I); ++I) {
+ PHINode *PN = cast<PHINode>(I);
+ std::set<BasicBlock*> ProcessedPreds;
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
+ if (BlocksToExtract.count(PN->getIncomingBlock(i))) {
+ if (ProcessedPreds.insert(PN->getIncomingBlock(i)).second)
+ PN->setIncomingBlock(i, codeReplacer);
+ else {
+ // There were multiple entries in the PHI for this block, now there
+ // is only one, so remove the duplicated entries.
+ PN->removeIncomingValue(i, false);
+ --i; --e;
+ }
+ }
+ }
- // Step 2: Construct new function based on inputs/outputs,
- // Add allocas for all defs
- Function *newFunction = constructFunction(inputs, outputs, newFuncRoot,
- codeReplacer, code,
- oldFunction, module);
+ //cerr << "NEW FUNCTION: " << *newFunction;
+ // verifyFunction(*newFunction);
- rewritePhiNodes(newFunction, newFuncRoot);
+ // cerr << "OLD FUNCTION: " << *oldFunction;
+ // verifyFunction(*oldFunction);
- emitCallAndSwitchStatement(newFunction, codeReplacer, code, inputs, outputs);
+ DEBUG(if (verifyFunction(*newFunction))
+ llvm_report_error("verifyFunction failed!"));
+ return newFunction;
+}
- moveCodeToFunction(code, newFunction);
+bool CodeExtractor::isEligible(const std::vector<BasicBlock*> &code) {
+ // Deny code region if it contains allocas or vastarts.
+ for (std::vector<BasicBlock*>::const_iterator BB = code.begin(), e=code.end();
+ BB != e; ++BB)
+ for (BasicBlock::const_iterator I = (*BB)->begin(), Ie = (*BB)->end();
+ I != Ie; ++I)
+ if (isa<AllocaInst>(*I))
+ return false;
+ else if (const CallInst *CI = dyn_cast<CallInst>(I))
+ if (const Function *F = CI->getCalledFunction())
+ if (F->getIntrinsicID() == Intrinsic::vastart)
+ return false;
+ return true;
+}
- return newFunction;
+
+/// ExtractCodeRegion - slurp a sequence of basic blocks into a brand new
+/// function
+///
+Function* llvm::ExtractCodeRegion(DominatorTree &DT,
+ const std::vector<BasicBlock*> &code,
+ bool AggregateArgs) {
+ return CodeExtractor(&DT, AggregateArgs).ExtractCodeRegion(code);
}
-Function* llvm::ExtractLoop(Loop *L) {
- CodeExtractor CE;
- return CE.ExtractCodeRegion(L->getBlocks());
+/// ExtractBasicBlock - slurp a natural loop into a brand new function
+///
+Function* llvm::ExtractLoop(DominatorTree &DT, Loop *L, bool AggregateArgs) {
+ return CodeExtractor(&DT, AggregateArgs).ExtractCodeRegion(L->getBlocks());
}
+/// ExtractBasicBlock - slurp a basic block into a brand new function
+///
+Function* llvm::ExtractBasicBlock(BasicBlock *BB, bool AggregateArgs) {
+ std::vector<BasicBlock*> Blocks;
+ Blocks.push_back(BB);
+ return CodeExtractor(0, AggregateArgs).ExtractCodeRegion(Blocks);
+}