1 //===- CodeExtractor.cpp - Pull code region into a new function -----------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the interface to tear out a code region, such as an
11 // individual loop or a parallel section, into a new function, replacing it with
12 // a call to the new function.
14 //===----------------------------------------------------------------------===//
16 #include "llvm/Transforms/Utils/CodeExtractor.h"
17 #include "llvm/Constants.h"
18 #include "llvm/DerivedTypes.h"
19 #include "llvm/Instructions.h"
20 #include "llvm/Intrinsics.h"
21 #include "llvm/LLVMContext.h"
22 #include "llvm/Module.h"
23 #include "llvm/Pass.h"
24 #include "llvm/Analysis/Dominators.h"
25 #include "llvm/Analysis/LoopInfo.h"
26 #include "llvm/Analysis/Verifier.h"
27 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
28 #include "llvm/Support/CommandLine.h"
29 #include "llvm/Support/Debug.h"
30 #include "llvm/Support/ErrorHandling.h"
31 #include "llvm/Support/raw_ostream.h"
32 #include "llvm/ADT/SetVector.h"
33 #include "llvm/ADT/StringExtras.h"
38 // Provide a command-line option to aggregate function arguments into a struct
39 // for functions produced by the code extractor. This is useful when converting
40 // extracted functions to pthread-based code, as only one argument (void*) can
41 // be passed in to pthread_create().
43 AggregateArgsOpt("aggregate-extracted-args", cl::Hidden,
44 cl::desc("Aggregate arguments to code-extracted functions"));
46 /// \brief Test whether a block is valid for extraction.
47 static bool isBlockValidForExtraction(const BasicBlock &BB) {
48 // Landing pads must be in the function where they were inserted for cleanup.
49 if (BB.isLandingPad())
52 // Don't hoist code containing allocas, invokes, or vastarts.
53 for (BasicBlock::const_iterator I = BB.begin(), E = BB.end(); I != E; ++I) {
54 if (isa<AllocaInst>(I) || isa<InvokeInst>(I))
56 if (const CallInst *CI = dyn_cast<CallInst>(I))
57 if (const Function *F = CI->getCalledFunction())
58 if (F->getIntrinsicID() == Intrinsic::vastart)
65 /// \brief Build a set of blocks to extract if the input blocks are viable.
66 static SetVector<BasicBlock *>
67 buildExtractionBlockSet(ArrayRef<BasicBlock *> BBs) {
68 SetVector<BasicBlock *> Result;
72 // Loop over the blocks, adding them to our set-vector, and aborting with an
73 // empty set if we encounter invalid blocks.
74 for (ArrayRef<BasicBlock *>::iterator I = BBs.begin(), E = BBs.end();
76 if (!Result.insert(*I))
77 llvm_unreachable("Repeated basic blocks in extraction input");
79 if (!isBlockValidForExtraction(**I)) {
86 for (ArrayRef<BasicBlock *>::iterator I = llvm::next(BBs.begin()),
89 for (pred_iterator PI = pred_begin(*I), PE = pred_end(*I);
91 assert(Result.count(*PI) &&
92 "No blocks in this region may have entries from outside the region"
93 " except for the first block!");
99 CodeExtractor::CodeExtractor(BasicBlock *BB, bool AggregateArgs)
100 : DT(0), AggregateArgs(AggregateArgs||AggregateArgsOpt),
101 Blocks(buildExtractionBlockSet(BB)), NumExitBlocks(~0U) {}
103 CodeExtractor::CodeExtractor(ArrayRef<BasicBlock *> BBs, DominatorTree *DT,
105 : DT(DT), AggregateArgs(AggregateArgs||AggregateArgsOpt),
106 Blocks(buildExtractionBlockSet(BBs)), NumExitBlocks(~0U) {}
108 CodeExtractor::CodeExtractor(DominatorTree &DT, Loop &L, bool AggregateArgs)
109 : DT(&DT), AggregateArgs(AggregateArgs||AggregateArgsOpt),
110 Blocks(buildExtractionBlockSet(L.getBlocks())), NumExitBlocks(~0U) {}
112 /// definedInRegion - Return true if the specified value is defined in the
113 /// extracted region.
114 static bool definedInRegion(const SetVector<BasicBlock *> &Blocks, Value *V) {
115 if (Instruction *I = dyn_cast<Instruction>(V))
116 if (Blocks.count(I->getParent()))
121 /// definedInCaller - Return true if the specified value is defined in the
122 /// function being code extracted, but not in the region being extracted.
123 /// These values must be passed in as live-ins to the function.
124 static bool definedInCaller(const SetVector<BasicBlock *> &Blocks, Value *V) {
125 if (isa<Argument>(V)) return true;
126 if (Instruction *I = dyn_cast<Instruction>(V))
127 if (!Blocks.count(I->getParent()))
132 void CodeExtractor::findInputsOutputs(ValueSet &Inputs,
133 ValueSet &Outputs) const {
134 for (SetVector<BasicBlock *>::const_iterator I = Blocks.begin(),
139 // If a used value is defined outside the region, it's an input. If an
140 // instruction is used outside the region, it's an output.
141 for (BasicBlock::iterator II = BB->begin(), IE = BB->end();
143 for (User::op_iterator OI = II->op_begin(), OE = II->op_end();
145 if (definedInCaller(Blocks, *OI))
148 for (Value::use_iterator UI = II->use_begin(), UE = II->use_end();
150 if (!definedInRegion(Blocks, *UI)) {
158 /// severSplitPHINodes - If a PHI node has multiple inputs from outside of the
159 /// region, we need to split the entry block of the region so that the PHI node
160 /// is easier to deal with.
161 void CodeExtractor::severSplitPHINodes(BasicBlock *&Header) {
162 unsigned NumPredsFromRegion = 0;
163 unsigned NumPredsOutsideRegion = 0;
165 if (Header != &Header->getParent()->getEntryBlock()) {
166 PHINode *PN = dyn_cast<PHINode>(Header->begin());
167 if (!PN) return; // No PHI nodes.
169 // If the header node contains any PHI nodes, check to see if there is more
170 // than one entry from outside the region. If so, we need to sever the
171 // header block into two.
172 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
173 if (Blocks.count(PN->getIncomingBlock(i)))
174 ++NumPredsFromRegion;
176 ++NumPredsOutsideRegion;
178 // If there is one (or fewer) predecessor from outside the region, we don't
179 // need to do anything special.
180 if (NumPredsOutsideRegion <= 1) return;
183 // Otherwise, we need to split the header block into two pieces: one
184 // containing PHI nodes merging values from outside of the region, and a
185 // second that contains all of the code for the block and merges back any
186 // incoming values from inside of the region.
187 BasicBlock::iterator AfterPHIs = Header->getFirstNonPHI();
188 BasicBlock *NewBB = Header->splitBasicBlock(AfterPHIs,
189 Header->getName()+".ce");
191 // We only want to code extract the second block now, and it becomes the new
192 // header of the region.
193 BasicBlock *OldPred = Header;
194 Blocks.remove(OldPred);
195 Blocks.insert(NewBB);
198 // Okay, update dominator sets. The blocks that dominate the new one are the
199 // blocks that dominate TIBB plus the new block itself.
201 DT->splitBlock(NewBB);
203 // Okay, now we need to adjust the PHI nodes and any branches from within the
204 // region to go to the new header block instead of the old header block.
205 if (NumPredsFromRegion) {
206 PHINode *PN = cast<PHINode>(OldPred->begin());
207 // Loop over all of the predecessors of OldPred that are in the region,
208 // changing them to branch to NewBB instead.
209 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
210 if (Blocks.count(PN->getIncomingBlock(i))) {
211 TerminatorInst *TI = PN->getIncomingBlock(i)->getTerminator();
212 TI->replaceUsesOfWith(OldPred, NewBB);
215 // Okay, everything within the region is now branching to the right block, we
216 // just have to update the PHI nodes now, inserting PHI nodes into NewBB.
217 for (AfterPHIs = OldPred->begin(); isa<PHINode>(AfterPHIs); ++AfterPHIs) {
218 PHINode *PN = cast<PHINode>(AfterPHIs);
219 // Create a new PHI node in the new region, which has an incoming value
220 // from OldPred of PN.
221 PHINode *NewPN = PHINode::Create(PN->getType(), 1 + NumPredsFromRegion,
222 PN->getName()+".ce", NewBB->begin());
223 NewPN->addIncoming(PN, OldPred);
225 // Loop over all of the incoming value in PN, moving them to NewPN if they
226 // are from the extracted region.
227 for (unsigned i = 0; i != PN->getNumIncomingValues(); ++i) {
228 if (Blocks.count(PN->getIncomingBlock(i))) {
229 NewPN->addIncoming(PN->getIncomingValue(i), PN->getIncomingBlock(i));
230 PN->removeIncomingValue(i);
238 void CodeExtractor::splitReturnBlocks() {
239 for (SetVector<BasicBlock *>::iterator I = Blocks.begin(), E = Blocks.end();
241 if (ReturnInst *RI = dyn_cast<ReturnInst>((*I)->getTerminator())) {
242 BasicBlock *New = (*I)->splitBasicBlock(RI, (*I)->getName()+".ret");
244 // Old dominates New. New node dominates all other nodes dominated
246 DomTreeNode *OldNode = DT->getNode(*I);
247 SmallVector<DomTreeNode*, 8> Children;
248 for (DomTreeNode::iterator DI = OldNode->begin(), DE = OldNode->end();
250 Children.push_back(*DI);
252 DomTreeNode *NewNode = DT->addNewBlock(New, *I);
254 for (SmallVector<DomTreeNode*, 8>::iterator I = Children.begin(),
255 E = Children.end(); I != E; ++I)
256 DT->changeImmediateDominator(*I, NewNode);
261 /// constructFunction - make a function based on inputs and outputs, as follows:
262 /// f(in0, ..., inN, out0, ..., outN)
264 Function *CodeExtractor::constructFunction(const ValueSet &inputs,
265 const ValueSet &outputs,
267 BasicBlock *newRootNode,
268 BasicBlock *newHeader,
269 Function *oldFunction,
271 DEBUG(dbgs() << "inputs: " << inputs.size() << "\n");
272 DEBUG(dbgs() << "outputs: " << outputs.size() << "\n");
274 // This function returns unsigned, outputs will go back by reference.
275 switch (NumExitBlocks) {
277 case 1: RetTy = Type::getVoidTy(header->getContext()); break;
278 case 2: RetTy = Type::getInt1Ty(header->getContext()); break;
279 default: RetTy = Type::getInt16Ty(header->getContext()); break;
282 std::vector<Type*> paramTy;
284 // Add the types of the input values to the function's argument list
285 for (ValueSet::const_iterator i = inputs.begin(), e = inputs.end();
287 const Value *value = *i;
288 DEBUG(dbgs() << "value used in func: " << *value << "\n");
289 paramTy.push_back(value->getType());
292 // Add the types of the output values to the function's argument list.
293 for (ValueSet::const_iterator I = outputs.begin(), E = outputs.end();
295 DEBUG(dbgs() << "instr used in func: " << **I << "\n");
297 paramTy.push_back((*I)->getType());
299 paramTy.push_back(PointerType::getUnqual((*I)->getType()));
302 DEBUG(dbgs() << "Function type: " << *RetTy << " f(");
303 for (std::vector<Type*>::iterator i = paramTy.begin(),
304 e = paramTy.end(); i != e; ++i)
305 DEBUG(dbgs() << **i << ", ");
306 DEBUG(dbgs() << ")\n");
308 if (AggregateArgs && (inputs.size() + outputs.size() > 0)) {
309 PointerType *StructPtr =
310 PointerType::getUnqual(StructType::get(M->getContext(), paramTy));
312 paramTy.push_back(StructPtr);
314 FunctionType *funcType =
315 FunctionType::get(RetTy, paramTy, false);
317 // Create the new function
318 Function *newFunction = Function::Create(funcType,
319 GlobalValue::InternalLinkage,
320 oldFunction->getName() + "_" +
321 header->getName(), M);
322 // If the old function is no-throw, so is the new one.
323 if (oldFunction->doesNotThrow())
324 newFunction->setDoesNotThrow(true);
326 newFunction->getBasicBlockList().push_back(newRootNode);
328 // Create an iterator to name all of the arguments we inserted.
329 Function::arg_iterator AI = newFunction->arg_begin();
331 // Rewrite all users of the inputs in the extracted region to use the
332 // arguments (or appropriate addressing into struct) instead.
333 for (unsigned i = 0, e = inputs.size(); i != e; ++i) {
337 Idx[0] = Constant::getNullValue(Type::getInt32Ty(header->getContext()));
338 Idx[1] = ConstantInt::get(Type::getInt32Ty(header->getContext()), i);
339 TerminatorInst *TI = newFunction->begin()->getTerminator();
340 GetElementPtrInst *GEP =
341 GetElementPtrInst::Create(AI, Idx, "gep_" + inputs[i]->getName(), TI);
342 RewriteVal = new LoadInst(GEP, "loadgep_" + inputs[i]->getName(), TI);
346 std::vector<User*> Users(inputs[i]->use_begin(), inputs[i]->use_end());
347 for (std::vector<User*>::iterator use = Users.begin(), useE = Users.end();
349 if (Instruction* inst = dyn_cast<Instruction>(*use))
350 if (Blocks.count(inst->getParent()))
351 inst->replaceUsesOfWith(inputs[i], RewriteVal);
354 // Set names for input and output arguments.
355 if (!AggregateArgs) {
356 AI = newFunction->arg_begin();
357 for (unsigned i = 0, e = inputs.size(); i != e; ++i, ++AI)
358 AI->setName(inputs[i]->getName());
359 for (unsigned i = 0, e = outputs.size(); i != e; ++i, ++AI)
360 AI->setName(outputs[i]->getName()+".out");
363 // Rewrite branches to basic blocks outside of the loop to new dummy blocks
364 // within the new function. This must be done before we lose track of which
365 // blocks were originally in the code region.
366 std::vector<User*> Users(header->use_begin(), header->use_end());
367 for (unsigned i = 0, e = Users.size(); i != e; ++i)
368 // The BasicBlock which contains the branch is not in the region
369 // modify the branch target to a new block
370 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(Users[i]))
371 if (!Blocks.count(TI->getParent()) &&
372 TI->getParent()->getParent() == oldFunction)
373 TI->replaceUsesOfWith(header, newHeader);
378 /// FindPhiPredForUseInBlock - Given a value and a basic block, find a PHI
379 /// that uses the value within the basic block, and return the predecessor
380 /// block associated with that use, or return 0 if none is found.
381 static BasicBlock* FindPhiPredForUseInBlock(Value* Used, BasicBlock* BB) {
382 for (Value::use_iterator UI = Used->use_begin(),
383 UE = Used->use_end(); UI != UE; ++UI) {
384 PHINode *P = dyn_cast<PHINode>(*UI);
385 if (P && P->getParent() == BB)
386 return P->getIncomingBlock(UI);
392 /// emitCallAndSwitchStatement - This method sets up the caller side by adding
393 /// the call instruction, splitting any PHI nodes in the header block as
396 emitCallAndSwitchStatement(Function *newFunction, BasicBlock *codeReplacer,
397 ValueSet &inputs, ValueSet &outputs) {
398 // Emit a call to the new function, passing in: *pointer to struct (if
399 // aggregating parameters), or plan inputs and allocated memory for outputs
400 std::vector<Value*> params, StructValues, ReloadOutputs, Reloads;
402 LLVMContext &Context = newFunction->getContext();
404 // Add inputs as params, or to be filled into the struct
405 for (ValueSet::iterator i = inputs.begin(), e = inputs.end(); i != e; ++i)
407 StructValues.push_back(*i);
409 params.push_back(*i);
411 // Create allocas for the outputs
412 for (ValueSet::iterator i = outputs.begin(), e = outputs.end(); i != e; ++i) {
414 StructValues.push_back(*i);
417 new AllocaInst((*i)->getType(), 0, (*i)->getName()+".loc",
418 codeReplacer->getParent()->begin()->begin());
419 ReloadOutputs.push_back(alloca);
420 params.push_back(alloca);
424 AllocaInst *Struct = 0;
425 if (AggregateArgs && (inputs.size() + outputs.size() > 0)) {
426 std::vector<Type*> ArgTypes;
427 for (ValueSet::iterator v = StructValues.begin(),
428 ve = StructValues.end(); v != ve; ++v)
429 ArgTypes.push_back((*v)->getType());
431 // Allocate a struct at the beginning of this function
432 Type *StructArgTy = StructType::get(newFunction->getContext(), ArgTypes);
434 new AllocaInst(StructArgTy, 0, "structArg",
435 codeReplacer->getParent()->begin()->begin());
436 params.push_back(Struct);
438 for (unsigned i = 0, e = inputs.size(); i != e; ++i) {
440 Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context));
441 Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), i);
442 GetElementPtrInst *GEP =
443 GetElementPtrInst::Create(Struct, Idx,
444 "gep_" + StructValues[i]->getName());
445 codeReplacer->getInstList().push_back(GEP);
446 StoreInst *SI = new StoreInst(StructValues[i], GEP);
447 codeReplacer->getInstList().push_back(SI);
451 // Emit the call to the function
452 CallInst *call = CallInst::Create(newFunction, params,
453 NumExitBlocks > 1 ? "targetBlock" : "");
454 codeReplacer->getInstList().push_back(call);
456 Function::arg_iterator OutputArgBegin = newFunction->arg_begin();
457 unsigned FirstOut = inputs.size();
459 std::advance(OutputArgBegin, inputs.size());
461 // Reload the outputs passed in by reference
462 for (unsigned i = 0, e = outputs.size(); i != e; ++i) {
466 Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context));
467 Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), FirstOut + i);
468 GetElementPtrInst *GEP
469 = GetElementPtrInst::Create(Struct, Idx,
470 "gep_reload_" + outputs[i]->getName());
471 codeReplacer->getInstList().push_back(GEP);
474 Output = ReloadOutputs[i];
476 LoadInst *load = new LoadInst(Output, outputs[i]->getName()+".reload");
477 Reloads.push_back(load);
478 codeReplacer->getInstList().push_back(load);
479 std::vector<User*> Users(outputs[i]->use_begin(), outputs[i]->use_end());
480 for (unsigned u = 0, e = Users.size(); u != e; ++u) {
481 Instruction *inst = cast<Instruction>(Users[u]);
482 if (!Blocks.count(inst->getParent()))
483 inst->replaceUsesOfWith(outputs[i], load);
487 // Now we can emit a switch statement using the call as a value.
488 SwitchInst *TheSwitch =
489 SwitchInst::Create(Constant::getNullValue(Type::getInt16Ty(Context)),
490 codeReplacer, 0, codeReplacer);
492 // Since there may be multiple exits from the original region, make the new
493 // function return an unsigned, switch on that number. This loop iterates
494 // over all of the blocks in the extracted region, updating any terminator
495 // instructions in the to-be-extracted region that branch to blocks that are
496 // not in the region to be extracted.
497 std::map<BasicBlock*, BasicBlock*> ExitBlockMap;
499 unsigned switchVal = 0;
500 for (SetVector<BasicBlock*>::const_iterator i = Blocks.begin(),
501 e = Blocks.end(); i != e; ++i) {
502 TerminatorInst *TI = (*i)->getTerminator();
503 for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
504 if (!Blocks.count(TI->getSuccessor(i))) {
505 BasicBlock *OldTarget = TI->getSuccessor(i);
506 // add a new basic block which returns the appropriate value
507 BasicBlock *&NewTarget = ExitBlockMap[OldTarget];
509 // If we don't already have an exit stub for this non-extracted
510 // destination, create one now!
511 NewTarget = BasicBlock::Create(Context,
512 OldTarget->getName() + ".exitStub",
514 unsigned SuccNum = switchVal++;
517 switch (NumExitBlocks) {
519 case 1: break; // No value needed.
520 case 2: // Conditional branch, return a bool
521 brVal = ConstantInt::get(Type::getInt1Ty(Context), !SuccNum);
524 brVal = ConstantInt::get(Type::getInt16Ty(Context), SuccNum);
528 ReturnInst *NTRet = ReturnInst::Create(Context, brVal, NewTarget);
530 // Update the switch instruction.
531 TheSwitch->addCase(ConstantInt::get(Type::getInt16Ty(Context),
535 // Restore values just before we exit
536 Function::arg_iterator OAI = OutputArgBegin;
537 for (unsigned out = 0, e = outputs.size(); out != e; ++out) {
538 // For an invoke, the normal destination is the only one that is
539 // dominated by the result of the invocation
540 BasicBlock *DefBlock = cast<Instruction>(outputs[out])->getParent();
542 bool DominatesDef = true;
544 if (InvokeInst *Invoke = dyn_cast<InvokeInst>(outputs[out])) {
545 DefBlock = Invoke->getNormalDest();
547 // Make sure we are looking at the original successor block, not
548 // at a newly inserted exit block, which won't be in the dominator
550 for (std::map<BasicBlock*, BasicBlock*>::iterator I =
551 ExitBlockMap.begin(), E = ExitBlockMap.end(); I != E; ++I)
552 if (DefBlock == I->second) {
557 // In the extract block case, if the block we are extracting ends
558 // with an invoke instruction, make sure that we don't emit a
559 // store of the invoke value for the unwind block.
560 if (!DT && DefBlock != OldTarget)
561 DominatesDef = false;
565 DominatesDef = DT->dominates(DefBlock, OldTarget);
567 // If the output value is used by a phi in the target block,
568 // then we need to test for dominance of the phi's predecessor
569 // instead. Unfortunately, this a little complicated since we
570 // have already rewritten uses of the value to uses of the reload.
571 BasicBlock* pred = FindPhiPredForUseInBlock(Reloads[out],
573 if (pred && DT && DT->dominates(DefBlock, pred))
580 Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context));
581 Idx[1] = ConstantInt::get(Type::getInt32Ty(Context),
583 GetElementPtrInst *GEP =
584 GetElementPtrInst::Create(OAI, Idx,
585 "gep_" + outputs[out]->getName(),
587 new StoreInst(outputs[out], GEP, NTRet);
589 new StoreInst(outputs[out], OAI, NTRet);
592 // Advance output iterator even if we don't emit a store
593 if (!AggregateArgs) ++OAI;
597 // rewrite the original branch instruction with this new target
598 TI->setSuccessor(i, NewTarget);
602 // Now that we've done the deed, simplify the switch instruction.
603 Type *OldFnRetTy = TheSwitch->getParent()->getParent()->getReturnType();
604 switch (NumExitBlocks) {
606 // There are no successors (the block containing the switch itself), which
607 // means that previously this was the last part of the function, and hence
608 // this should be rewritten as a `ret'
610 // Check if the function should return a value
611 if (OldFnRetTy->isVoidTy()) {
612 ReturnInst::Create(Context, 0, TheSwitch); // Return void
613 } else if (OldFnRetTy == TheSwitch->getCondition()->getType()) {
614 // return what we have
615 ReturnInst::Create(Context, TheSwitch->getCondition(), TheSwitch);
617 // Otherwise we must have code extracted an unwind or something, just
618 // return whatever we want.
619 ReturnInst::Create(Context,
620 Constant::getNullValue(OldFnRetTy), TheSwitch);
623 TheSwitch->eraseFromParent();
626 // Only a single destination, change the switch into an unconditional
628 BranchInst::Create(TheSwitch->getSuccessor(1), TheSwitch);
629 TheSwitch->eraseFromParent();
632 BranchInst::Create(TheSwitch->getSuccessor(1), TheSwitch->getSuccessor(2),
634 TheSwitch->eraseFromParent();
637 // Otherwise, make the default destination of the switch instruction be one
638 // of the other successors.
639 TheSwitch->setCondition(call);
640 TheSwitch->setDefaultDest(TheSwitch->getSuccessor(NumExitBlocks));
641 // Remove redundant case
642 TheSwitch->removeCase(SwitchInst::CaseIt(TheSwitch, NumExitBlocks-1));
647 void CodeExtractor::moveCodeToFunction(Function *newFunction) {
648 Function *oldFunc = (*Blocks.begin())->getParent();
649 Function::BasicBlockListType &oldBlocks = oldFunc->getBasicBlockList();
650 Function::BasicBlockListType &newBlocks = newFunction->getBasicBlockList();
652 for (SetVector<BasicBlock*>::const_iterator i = Blocks.begin(),
653 e = Blocks.end(); i != e; ++i) {
654 // Delete the basic block from the old function, and the list of blocks
655 oldBlocks.remove(*i);
657 // Insert this basic block into the new function
658 newBlocks.push_back(*i);
662 Function *CodeExtractor::extractCodeRegion() {
666 ValueSet inputs, outputs;
668 // Assumption: this is a single-entry code region, and the header is the first
669 // block in the region.
670 BasicBlock *header = *Blocks.begin();
672 // If we have to split PHI nodes or the entry block, do so now.
673 severSplitPHINodes(header);
675 // If we have any return instructions in the region, split those blocks so
676 // that the return is not in the region.
679 Function *oldFunction = header->getParent();
681 // This takes place of the original loop
682 BasicBlock *codeReplacer = BasicBlock::Create(header->getContext(),
683 "codeRepl", oldFunction,
686 // The new function needs a root node because other nodes can branch to the
687 // head of the region, but the entry node of a function cannot have preds.
688 BasicBlock *newFuncRoot = BasicBlock::Create(header->getContext(),
690 newFuncRoot->getInstList().push_back(BranchInst::Create(header));
692 // Find inputs to, outputs from the code region.
693 findInputsOutputs(inputs, outputs);
695 SmallPtrSet<BasicBlock *, 1> ExitBlocks;
696 for (SetVector<BasicBlock *>::iterator I = Blocks.begin(), E = Blocks.end();
698 for (succ_iterator SI = succ_begin(*I), SE = succ_end(*I); SI != SE; ++SI)
699 if (!Blocks.count(*SI))
700 ExitBlocks.insert(*SI);
701 NumExitBlocks = ExitBlocks.size();
703 // Construct new function based on inputs/outputs & add allocas for all defs.
704 Function *newFunction = constructFunction(inputs, outputs, header,
706 codeReplacer, oldFunction,
707 oldFunction->getParent());
709 emitCallAndSwitchStatement(newFunction, codeReplacer, inputs, outputs);
711 moveCodeToFunction(newFunction);
713 // Loop over all of the PHI nodes in the header block, and change any
714 // references to the old incoming edge to be the new incoming edge.
715 for (BasicBlock::iterator I = header->begin(); isa<PHINode>(I); ++I) {
716 PHINode *PN = cast<PHINode>(I);
717 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
718 if (!Blocks.count(PN->getIncomingBlock(i)))
719 PN->setIncomingBlock(i, newFuncRoot);
722 // Look at all successors of the codeReplacer block. If any of these blocks
723 // had PHI nodes in them, we need to update the "from" block to be the code
724 // replacer, not the original block in the extracted region.
725 std::vector<BasicBlock*> Succs(succ_begin(codeReplacer),
726 succ_end(codeReplacer));
727 for (unsigned i = 0, e = Succs.size(); i != e; ++i)
728 for (BasicBlock::iterator I = Succs[i]->begin(); isa<PHINode>(I); ++I) {
729 PHINode *PN = cast<PHINode>(I);
730 std::set<BasicBlock*> ProcessedPreds;
731 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
732 if (Blocks.count(PN->getIncomingBlock(i))) {
733 if (ProcessedPreds.insert(PN->getIncomingBlock(i)).second)
734 PN->setIncomingBlock(i, codeReplacer);
736 // There were multiple entries in the PHI for this block, now there
737 // is only one, so remove the duplicated entries.
738 PN->removeIncomingValue(i, false);
744 //cerr << "NEW FUNCTION: " << *newFunction;
745 // verifyFunction(*newFunction);
747 // cerr << "OLD FUNCTION: " << *oldFunction;
748 // verifyFunction(*oldFunction);
750 DEBUG(if (verifyFunction(*newFunction))
751 report_fatal_error("verifyFunction failed!"));