//===- 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/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Instructions.h"
+#include "llvm/Intrinsics.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 "Support/Debug.h"
-#include "Support/StringExtras.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Compiler.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/ADT/StringExtras.h"
#include <algorithm>
#include <set>
using namespace llvm;
+// Provide a command-line option to aggregate function arguments into a struct
+// for functions produced by the code extrator. 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"));
+
namespace {
- class CodeExtractor {
+ class VISIBILITY_HIDDEN CodeExtractor {
typedef std::vector<Value*> Values;
std::set<BasicBlock*> BlocksToExtract;
- DominatorSet *DS;
+ DominatorTree* DT;
+ bool AggregateArgs;
+ unsigned NumExitBlocks;
+ const Type *RetTy;
public:
- CodeExtractor(DominatorSet *ds = 0) : DS(ds) {}
+ 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(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;
+ }
+
+ /// 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 severSplitPHINodes(BasicBlock *&Header);
+ void splitReturnBlocks();
+ void findInputsOutputs(Values &inputs, Values &outputs);
Function *constructFunction(const Values &inputs,
const Values &outputs,
};
}
-void CodeExtractor::findInputsOutputs(Values &inputs, Values &outputs,
- BasicBlock *newHeader,
- BasicBlock *newRootNode) {
- for (std::set<BasicBlock*>::const_iterator ci = BlocksToExtract.begin(),
+/// 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
+ ++NumPredsOutsideRegion;
+
+ // If there is one (or fewer) predecessor from outside the region, we don't
+ // need to do anything special.
+ if (NumPredsOutsideRegion <= 1) return;
+ }
+
+ // 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->begin();
+ while (isa<PHINode>(AfterPHIs)) ++AfterPHIs;
+ 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);
+ }
+
+ // 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 = new PHINode(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;
+ }
+ }
+ }
+ }
+}
+
+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");
+}
+
+// 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 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.
- if (PHINode *PN = dyn_cast<PHINode>(I)) {
- for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
- Value *V = PN->getIncomingValue(i);
- if (!BlocksToExtract.count(PN->getIncomingBlock(i)) &&
- (isa<Instruction>(V) || isa<Argument>(V)))
- inputs.push_back(V);
- }
- } 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)) {
- // Check if definition of this operand is within the loop
- if (!BlocksToExtract.count(opI->getParent()))
- inputs.push_back(opI);
- } else if (isa<Argument>(*op)) {
- inputs.push_back(*op);
- }
- }
-
- // Consider uses of this instruction (outputs)
+ 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 (!BlocksToExtract.count(cast<Instruction>(*UI)->getParent())) {
+ if (!definedInRegion(*UI)) {
outputs.push_back(I);
break;
}
} // 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:
BasicBlock *newHeader,
Function *oldFunction,
Module *M) {
- DEBUG(std::cerr << "inputs: " << inputs.size() << "\n");
- DEBUG(std::cerr << "outputs: " << outputs.size() << "\n");
+ DOUT << "inputs: " << inputs.size() << "\n";
+ DOUT << "outputs: " << outputs.size() << "\n";
// 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.
for (Values::const_iterator I = outputs.begin(), E = outputs.end();
I != E; ++I) {
- DEBUG(std::cerr << "instr used in func: " << *I << "\n");
- paramTy.push_back(PointerType::get((*I)->getType()));
+ DOUT << "instr used in func: " << **I << "\n";
+ if (AggregateArgs)
+ paramTy.push_back((*I)->getType());
+ else
+ paramTy.push_back(PointerType::getUnqual((*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");
+ DOUT << **i << ", ";
+ DOUT << ")\n";
- const FunctionType *funcType = FunctionType::get(retTy, paramTy, false);
+ if (AggregateArgs && (inputs.size() + outputs.size() > 0)) {
+ PointerType *StructPtr = PointerType::getUnqual(StructType::get(paramTy));
+ paramTy.clear();
+ paramTy.push_back(StructPtr);
+ }
+ const FunctionType *funcType = FunctionType::get(RetTy, paramTy, false);
// Create the new function
Function *newFunction = new Function(funcType,
GlobalValue::InternalLinkage,
- oldFunction->getName() + "_code", M);
+ oldFunction->getName() + "_" +
+ header->getName(), M);
newFunction->getBasicBlockList().push_back(newRootNode);
// Create an iterator to name all of the arguments we inserted.
- Function::aiterator AI = newFunction->abegin();
+ Function::arg_iterator AI = newFunction->arg_begin();
// Rewrite all users of the inputs in the extracted region to use the
- // arguments instead.
- for (unsigned i = 0, e = inputs.size(); i != e; ++i, ++AI) {
- AI->setName(inputs[i]->getName());
+ // 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] = Constant::getNullValue(Type::Int32Ty);
+ Idx[1] = ConstantInt::get(Type::Int32Ty, i);
+ std::string GEPname = "gep_" + inputs[i]->getName();
+ TerminatorInst *TI = newFunction->begin()->getTerminator();
+ GetElementPtrInst *GEP = new GetElementPtrInst(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 (BlocksToExtract.count(inst->getParent()))
- inst->replaceUsesOfWith(inputs[i], AI);
+ inst->replaceUsesOfWith(inputs[i], RewriteVal);
}
- // Set names for all of the output arguments.
- for (unsigned i = 0, e = outputs.size(); i != e; ++i, ++AI)
- AI->setName(outputs[i]->getName()+".out");
-
+ // 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
return newFunction;
}
-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);
+/// 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) {
+ // 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);
+
+ // 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((*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
- newBlocks.push_back(*i);
+ 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 = StructType::get(ArgTypes);
+ Struct =
+ new AllocaInst(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] = Constant::getNullValue(Type::Int32Ty);
+ Idx[1] = ConstantInt::get(Type::Int32Ty, i);
+ GetElementPtrInst *GEP =
+ new GetElementPtrInst(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,
- 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(inputs);
+ // Emit the call to the function
+ CallInst *call = new CallInst(newFunction, params.begin(), params.end(),
+ NumExitBlocks > 1 ? "targetBlock" : "");
+ codeReplacer->getInstList().push_back(call);
- // Get an iterator to the first output argument.
- Function::aiterator OutputArgBegin = newFunction->abegin();
- std::advance(OutputArgBegin, inputs.size());
+ 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 = outputs[i];
- // Create allocas for scalar outputs
- AllocaInst *alloca =
- new AllocaInst(outputs[i]->getType(), 0, Output->getName()+".loc",
- codeReplacer->getParent()->begin()->begin());
- params.push_back(alloca);
-
- LoadInst *load = new LoadInst(alloca, Output->getName()+".reload");
+ Value *Output = 0;
+ if (AggregateArgs) {
+ Value *Idx[2];
+ Idx[0] = Constant::getNullValue(Type::Int32Ty);
+ Idx[1] = ConstantInt::get(Type::Int32Ty, FirstOut + i);
+ GetElementPtrInst *GEP
+ = new GetElementPtrInst(Struct, Idx, Idx + 2,
+ "gep_reload_" + outputs[i]->getName());
+ codeReplacer->getInstList().push_back(GEP);
+ Output = GEP;
+ } else {
+ 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) {
}
}
- CallInst *call = new CallInst(newFunction, params, "targetBlock");
- codeReplacer->getInstList().push_front(call);
-
// Now we can emit a switch statement using the call as a value.
- SwitchInst *TheSwitch = new SwitchInst(call, codeReplacer, codeReplacer);
+ SwitchInst *TheSwitch =
+ new SwitchInst(ConstantInt::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. This loop iterates
// destination, create one now!
NewTarget = new BasicBlock(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 = ConstantInt::get(Type::Int1Ty, !SuccNum);
+ break;
+ default:
+ brVal = ConstantInt::get(Type::Int16Ty, SuccNum);
+ break;
+ }
- ConstantUInt *brVal = ConstantUInt::get(Type::UShortTy, switchVal++);
ReturnInst *NTRet = new ReturnInst(brVal, NewTarget);
// Update the switch instruction.
- TheSwitch->addCase(brVal, OldTarget);
+ TheSwitch->addCase(ConstantInt::get(Type::Int16Ty, SuccNum),
+ OldTarget);
// Restore values just before we exit
- // FIXME: Use a GetElementPtr to bunch the outputs in a struct
- Function::aiterator OAI = OutputArgBegin;
- for (unsigned out = 0, e = outputs.size(); out != e; ++out, ++OAI)
- if (!DS ||
- DS->dominates(cast<Instruction>(outputs[out])->getParent(),
- TI->getParent()))
- new StoreInst(outputs[out], OAI, NTRet);
+ 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] = Constant::getNullValue(Type::Int32Ty);
+ Idx[1] = ConstantInt::get(Type::Int32Ty,FirstOut+out);
+ GetElementPtrInst *GEP =
+ new GetElementPtrInst(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
}
}
- // Now that we've done the deed, make the default destination of the switch
- // instruction be one of the exit blocks of the region.
- if (TheSwitch->getNumSuccessors() > 1) {
- // FIXME: this is broken w.r.t. PHI nodes, but the old code was more broken.
- // This edge is not traversable.
- TheSwitch->setSuccessor(0, TheSwitch->getSuccessor(1));
+ // 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) {
+ new ReturnInst(0, TheSwitch); // Return void
+ } else if (OldFnRetTy == TheSwitch->getCondition()->getType()) {
+ // return what we have
+ new ReturnInst(TheSwitch->getCondition(), TheSwitch);
+ } else {
+ // Otherwise we must have code extracted an unwind or something, just
+ // return whatever we want.
+ new ReturnInst(Constant::getNullValue(OldFnRetTy), TheSwitch);
+ }
+
+ TheSwitch->getParent()->getInstList().erase(TheSwitch);
+ break;
+ case 1:
+ // Only a single destination, change the switch into an unconditional
+ // branch.
+ new BranchInst(TheSwitch->getSuccessor(1), TheSwitch);
+ TheSwitch->getParent()->getInstList().erase(TheSwitch);
+ break;
+ case 2:
+ new BranchInst(TheSwitch->getSuccessor(1), TheSwitch->getSuccessor(2),
+ call, TheSwitch);
+ TheSwitch->getParent()->getInstList().erase(TheSwitch);
+ 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
// Assumption: this is a single-entry code region, and the header is the first
// 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();
// This takes place of the original loop
- BasicBlock *codeReplacer = new BasicBlock("codeRepl", oldFunction);
+ BasicBlock *codeReplacer = new BasicBlock("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
+ // head of the region, but the entry node of a function cannot have preds.
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(inputs, outputs, codeReplacer, newFuncRoot);
-
- // Step 2: Construct new function based on inputs/outputs,
- // Add allocas for all defs
- Function *newFunction = constructFunction(inputs, outputs, code[0],
- newFuncRoot,
+ // 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());
moveCodeToFunction(newFunction);
- // Loop over all of the PHI nodes in the entry block (code[0]), and change any
+ // 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 = code[0]->begin();
- PHINode *PN = dyn_cast<PHINode>(I); ++I)
+ 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);
+ }
+
+ // 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;
+ }
+ }
+ }
+
+ //cerr << "NEW FUNCTION: " << *newFunction;
+ // verifyFunction(*newFunction);
+
+ // cerr << "OLD FUNCTION: " << *oldFunction;
+ // verifyFunction(*oldFunction);
DEBUG(if (verifyFunction(*newFunction)) abort());
return 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;
+}
+
+
/// ExtractCodeRegion - slurp a sequence of basic blocks into a brand new
/// function
///
-Function* llvm::ExtractCodeRegion(DominatorSet &DS,
- const std::vector<BasicBlock*> &code) {
- return CodeExtractor(&DS).ExtractCodeRegion(code);
+Function* llvm::ExtractCodeRegion(DominatorTree &DT,
+ const std::vector<BasicBlock*> &code,
+ bool AggregateArgs) {
+ return CodeExtractor(&DT, AggregateArgs).ExtractCodeRegion(code);
}
/// ExtractBasicBlock - slurp a natural loop into a brand new function
///
-Function* llvm::ExtractLoop(DominatorSet &DS, Loop *L) {
- return CodeExtractor(&DS).ExtractCodeRegion(L->getBlocks());
+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) {
+Function* llvm::ExtractBasicBlock(BasicBlock *BB, bool AggregateArgs) {
std::vector<BasicBlock*> Blocks;
Blocks.push_back(BB);
- return CodeExtractor().ExtractCodeRegion(Blocks);
+ return CodeExtractor(0, AggregateArgs).ExtractCodeRegion(Blocks);
}