//===- 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
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/Verifier.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
-#include "Support/CommandLine.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;
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, bool AggArgs = false)
- : DS(ds), AggregateArgs(AggregateArgsOpt), NumExitBlocks(~0U) {}
+ CodeExtractor(DominatorTree* dt = 0, bool AggArgs = false)
+ : DT(dt), AggregateArgs(AggArgs||AggregateArgsOpt), NumExitBlocks(~0U) {}
Function *ExtractCodeRegion(const std::vector<BasicBlock*> &code);
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 HasPredsFromRegion = false;
unsigned NumPredsOutsideRegion = 0;
- if (Header != &Header->getParent()->front()) {
+ if (Header != &Header->getParent()->getEntryBlock()) {
PHINode *PN = dyn_cast<PHINode>(Header->begin());
if (!PN) return; // No PHI nodes.
// Okay, update dominator sets. The blocks that dominate the new one are the
// blocks that dominate TIBB plus the new block itself.
- if (DS) {
- DominatorSet::DomSetType DomSet = DS->getDominators(OldPred);
- DomSet.insert(NewBB); // A block always dominates itself.
- DS->addBasicBlock(NewBB, DomSet);
-
- // Additionally, NewBB dominates all blocks in the function that are
- // dominated by OldPred.
- Function *F = Header->getParent();
- for (Function::iterator I = F->begin(), E = F->end(); I != E; ++I)
- if (DS->properlyDominates(OldPred, I))
- DS->addDominator(I, NewBB);
- }
+ 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.
// 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();
- PHINode *PN = dyn_cast<PHINode>(AfterPHIs); ++AfterPHIs) {
+ 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",
//
void CodeExtractor::findInputsOutputs(Values &inputs, Values &outputs) {
std::set<BasicBlock*> ExitBlocks;
- for (std::set<BasicBlock*>::const_iterator ci = BlocksToExtract.begin(),
+ for (std::set<BasicBlock*>::const_iterator ci = BlocksToExtract.begin(),
ce = BlocksToExtract.end(); ci != ce; ++ci) {
BasicBlock *BB = *ci;
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)
} // 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.
switch (NumExitBlocks) {
case 0:
case 1: RetTy = Type::VoidTy; break;
- case 2: RetTy = Type::BoolTy; break;
- default: RetTy = Type::UShortTy; break;
+ case 2: RetTy = Type::Int1Ty; break;
+ default: RetTy = Type::Int16Ty; break;
}
std::vector<const Type*> paramTy;
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");
+ DOUT << "instr used in func: " << **I << "\n";
if (AggregateArgs)
paramTy.push_back((*I)->getType());
else
- paramTy.push_back(PointerType::get((*I)->getType()));
+ paramTy.push_back(PointerType::getUnqual((*I)->getType()));
}
- DEBUG(std::cerr << "Function type: " << RetTy << " f(");
- DEBUG(for (std::vector<const Type*>::iterator i = paramTy.begin(),
- e = paramTy.end(); i != e; ++i) std::cerr << *i << ", ");
- DEBUG(std::cerr << ")\n");
+ DOUT << "Function type: " << *RetTy << " f(";
+ for (std::vector<const Type*>::iterator i = paramTy.begin(),
+ e = paramTy.end(); i != e; ++i)
+ DOUT << **i << ", ";
+ DOUT << ")\n";
if (AggregateArgs && (inputs.size() + outputs.size() > 0)) {
- PointerType *StructPtr = PointerType::get(StructType::get(paramTy));
+ PointerType *StructPtr = PointerType::getUnqual(StructType::get(paramTy));
paramTy.clear();
paramTy.push_back(StructPtr);
}
// 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 (or appropriate addressing into struct) instead.
for (unsigned i = 0, e = inputs.size(); i != e; ++i) {
Value *RewriteVal;
if (AggregateArgs) {
- std::vector<Value*> Indices;
- Indices.push_back(Constant::getNullValue(Type::UIntTy));
- Indices.push_back(ConstantUInt::get(Type::UIntTy, i));
+ 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, Indices, GEPname, TI);
+ GetElementPtrInst *GEP = new GetElementPtrInst(AI, Idx, Idx+2,
+ GEPname, TI);
RewriteVal = new LoadInst(GEP, "load" + GEPname, TI);
} else
RewriteVal = AI++;
// Set names for input and output arguments.
if (!AggregateArgs) {
- AI = newFunction->abegin();
+ 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");
+ AI->setName(outputs[i]->getName()+".out");
}
// Rewrite branches to basic blocks outside of the loop to new dummy blocks
// Allocate a struct at the beginning of this function
Type *StructArgTy = StructType::get(ArgTypes);
- Struct =
- new AllocaInst(StructArgTy, 0, "structArg",
+ Struct =
+ new AllocaInst(StructArgTy, 0, "structArg",
codeReplacer->getParent()->begin()->begin());
params.push_back(Struct);
for (unsigned i = 0, e = inputs.size(); i != e; ++i) {
- std::vector<Value*> Indices;
- Indices.push_back(Constant::getNullValue(Type::UIntTy));
- Indices.push_back(ConstantUInt::get(Type::UIntTy, i));
+ Value *Idx[2];
+ Idx[0] = Constant::getNullValue(Type::Int32Ty);
+ Idx[1] = ConstantInt::get(Type::Int32Ty, i);
GetElementPtrInst *GEP =
- new GetElementPtrInst(Struct, Indices,
- "gep_" + StructValues[i]->getName(), 0);
+ 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);
}
- }
+ }
// Emit the call to the function
- CallInst *call = new CallInst(newFunction, params,
- NumExitBlocks > 1 ? "targetBlock": "");
+ CallInst *call = new CallInst(newFunction, params.begin(), params.end(),
+ NumExitBlocks > 1 ? "targetBlock" : "");
codeReplacer->getInstList().push_back(call);
- Function::aiterator OutputArgBegin = newFunction->abegin();
+ Function::arg_iterator OutputArgBegin = newFunction->arg_begin();
unsigned FirstOut = inputs.size();
if (!AggregateArgs)
std::advance(OutputArgBegin, inputs.size());
for (unsigned i = 0, e = outputs.size(); i != e; ++i) {
Value *Output = 0;
if (AggregateArgs) {
- std::vector<Value*> Indices;
- Indices.push_back(Constant::getNullValue(Type::UIntTy));
- Indices.push_back(ConstantUInt::get(Type::UIntTy, FirstOut + i));
- GetElementPtrInst *GEP
- = new GetElementPtrInst(Struct, Indices,
- "gep_reload_" + outputs[i]->getName(), 0);
+ 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 {
// Now we can emit a switch statement using the call as a value.
SwitchInst *TheSwitch =
- new SwitchInst(ConstantUInt::getNullValue(Type::UShortTy),
- codeReplacer, codeReplacer);
+ 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
case 0:
case 1: break; // No value needed.
case 2: // Conditional branch, return a bool
- brVal = SuccNum ? ConstantBool::False : ConstantBool::True;
+ brVal = ConstantInt::get(Type::Int1Ty, !SuccNum);
break;
default:
- brVal = ConstantUInt::get(Type::UShortTy, SuccNum);
+ brVal = ConstantInt::get(Type::Int16Ty, SuccNum);
break;
}
ReturnInst *NTRet = new ReturnInst(brVal, NewTarget);
// Update the switch instruction.
- TheSwitch->addCase(ConstantUInt::get(Type::UShortTy, SuccNum),
+ TheSwitch->addCase(ConstantInt::get(Type::Int16Ty, SuccNum),
OldTarget);
// Restore values just before we exit
- Function::aiterator OAI = OutputArgBegin;
+ 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();
- if (InvokeInst *Invoke = dyn_cast<InvokeInst>(outputs[out]))
+
+ bool DominatesDef = true;
+
+ if (InvokeInst *Invoke = dyn_cast<InvokeInst>(outputs[out])) {
DefBlock = Invoke->getNormalDest();
- if (!DS || DS->dominates(DefBlock, TI->getParent()))
+
+ // 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) {
- std::vector<Value*> Indices;
- Indices.push_back(Constant::getNullValue(Type::UIntTy));
- Indices.push_back(ConstantUInt::get(Type::UIntTy,FirstOut+out));
+ Value *Idx[2];
+ Idx[0] = Constant::getNullValue(Type::Int32Ty);
+ Idx[1] = ConstantInt::get(Type::Int32Ty,FirstOut+out);
GetElementPtrInst *GEP =
- new GetElementPtrInst(OAI, Indices,
- "gep_" + outputs[out]->getName(),
+ new GetElementPtrInst(OAI, Idx, Idx + 2,
+ "gep_" + outputs[out]->getName(),
NTRet);
new StoreInst(outputs[out], GEP, NTRet);
- } else
+ } else {
new StoreInst(outputs[out], OAI, NTRet);
+ }
+ }
// Advance output iterator even if we don't emit a store
if (!AggregateArgs) ++OAI;
}
}
// Now that we've done the deed, simplify the switch instruction.
+ const Type *OldFnRetTy = TheSwitch->getParent()->getParent()->getReturnType();
switch (NumExitBlocks) {
case 0:
- // There is only 1 successor (the block containing the switch itself), which
+ // 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 (TheSwitch->getParent()->getParent()->getReturnType() != Type::VoidTy &&
- TheSwitch->getParent()->getParent()->getReturnType() ==
- TheSwitch->getCondition()->getType())
+ 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
- // just return
- new ReturnInst(0, 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;
///
/// 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;
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);
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 region, but the entry node of a function cannot have preds.
// Construct new function based on inputs/outputs & add allocas for all defs.
Function *newFunction = constructFunction(inputs, outputs, header,
- newFuncRoot,
+ newFuncRoot,
codeReplacer, oldFunction,
oldFunction->getParent());
// 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();
- 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
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();
- PHINode *PN = dyn_cast<PHINode>(I); ++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)))
- PN->setIncomingBlock(i, codeReplacer);
-
- //std::cerr << "NEW FUNCTION: " << *newFunction;
+ 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);
- // std::cerr << "OLD FUNCTION: " << *oldFunction;
+ // cerr << "OLD FUNCTION: " << *oldFunction;
// verifyFunction(*oldFunction);
DEBUG(if (verifyFunction(*newFunction)) abort());
/// ExtractCodeRegion - slurp a sequence of basic blocks into a brand new
/// function
///
-Function* llvm::ExtractCodeRegion(DominatorSet &DS,
+Function* llvm::ExtractCodeRegion(DominatorTree &DT,
const std::vector<BasicBlock*> &code,
bool AggregateArgs) {
- return CodeExtractor(&DS, AggregateArgs).ExtractCodeRegion(code);
+ return CodeExtractor(&DT, AggregateArgs).ExtractCodeRegion(code);
}
/// ExtractBasicBlock - slurp a natural loop into a brand new function
///
-Function* llvm::ExtractLoop(DominatorSet &DS, Loop *L, bool AggregateArgs) {
- return CodeExtractor(&DS, AggregateArgs).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, bool AggregateArgs) {
std::vector<BasicBlock*> Blocks;
Blocks.push_back(BB);
- return CodeExtractor(0, AggregateArgs).ExtractCodeRegion(Blocks);
+ return CodeExtractor(0, AggregateArgs).ExtractCodeRegion(Blocks);
}