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;
70 // Loop over the blocks, adding them to our set-vector, and aborting with an
71 // empty set if we encounter invalid blocks.
72 for (ArrayRef<BasicBlock *>::iterator I = BBs.begin(), E = BBs.end();
74 if (!Result.insert(*I))
77 if (!isBlockValidForExtraction(**I)) {
86 CodeExtractor::CodeExtractor(BasicBlock *BB, bool AggregateArgs)
87 : DT(0), AggregateArgs(AggregateArgs||AggregateArgsOpt),
88 Blocks(buildExtractionBlockSet(BB)), NumExitBlocks(~0U) {}
90 CodeExtractor::CodeExtractor(ArrayRef<BasicBlock *> BBs, DominatorTree *DT,
92 : DT(DT), AggregateArgs(AggregateArgs||AggregateArgsOpt),
93 Blocks(buildExtractionBlockSet(BBs)), NumExitBlocks(~0U) {}
95 CodeExtractor::CodeExtractor(DominatorTree &DT, Loop &L, bool AggregateArgs)
96 : DT(&DT), AggregateArgs(AggregateArgs||AggregateArgsOpt),
97 Blocks(buildExtractionBlockSet(L.getBlocks())), NumExitBlocks(~0U) {}
100 /// definedInRegion - Return true if the specified value is defined in the
101 /// extracted region.
102 static bool definedInRegion(const SetVector<BasicBlock *> &Blocks, Value *V) {
103 if (Instruction *I = dyn_cast<Instruction>(V))
104 if (Blocks.count(I->getParent()))
109 /// definedInCaller - Return true if the specified value is defined in the
110 /// function being code extracted, but not in the region being extracted.
111 /// These values must be passed in as live-ins to the function.
112 static bool definedInCaller(const SetVector<BasicBlock *> &Blocks, Value *V) {
113 if (isa<Argument>(V)) return true;
114 if (Instruction *I = dyn_cast<Instruction>(V))
115 if (!Blocks.count(I->getParent()))
120 /// severSplitPHINodes - If a PHI node has multiple inputs from outside of the
121 /// region, we need to split the entry block of the region so that the PHI node
122 /// is easier to deal with.
123 void CodeExtractor::severSplitPHINodes(BasicBlock *&Header) {
124 unsigned NumPredsFromRegion = 0;
125 unsigned NumPredsOutsideRegion = 0;
127 if (Header != &Header->getParent()->getEntryBlock()) {
128 PHINode *PN = dyn_cast<PHINode>(Header->begin());
129 if (!PN) return; // No PHI nodes.
131 // If the header node contains any PHI nodes, check to see if there is more
132 // than one entry from outside the region. If so, we need to sever the
133 // header block into two.
134 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
135 if (Blocks.count(PN->getIncomingBlock(i)))
136 ++NumPredsFromRegion;
138 ++NumPredsOutsideRegion;
140 // If there is one (or fewer) predecessor from outside the region, we don't
141 // need to do anything special.
142 if (NumPredsOutsideRegion <= 1) return;
145 // Otherwise, we need to split the header block into two pieces: one
146 // containing PHI nodes merging values from outside of the region, and a
147 // second that contains all of the code for the block and merges back any
148 // incoming values from inside of the region.
149 BasicBlock::iterator AfterPHIs = Header->getFirstNonPHI();
150 BasicBlock *NewBB = Header->splitBasicBlock(AfterPHIs,
151 Header->getName()+".ce");
153 // We only want to code extract the second block now, and it becomes the new
154 // header of the region.
155 BasicBlock *OldPred = Header;
156 Blocks.remove(OldPred);
157 Blocks.insert(NewBB);
160 // Okay, update dominator sets. The blocks that dominate the new one are the
161 // blocks that dominate TIBB plus the new block itself.
163 DT->splitBlock(NewBB);
165 // Okay, now we need to adjust the PHI nodes and any branches from within the
166 // region to go to the new header block instead of the old header block.
167 if (NumPredsFromRegion) {
168 PHINode *PN = cast<PHINode>(OldPred->begin());
169 // Loop over all of the predecessors of OldPred that are in the region,
170 // changing them to branch to NewBB instead.
171 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
172 if (Blocks.count(PN->getIncomingBlock(i))) {
173 TerminatorInst *TI = PN->getIncomingBlock(i)->getTerminator();
174 TI->replaceUsesOfWith(OldPred, NewBB);
177 // Okay, everything within the region is now branching to the right block, we
178 // just have to update the PHI nodes now, inserting PHI nodes into NewBB.
179 for (AfterPHIs = OldPred->begin(); isa<PHINode>(AfterPHIs); ++AfterPHIs) {
180 PHINode *PN = cast<PHINode>(AfterPHIs);
181 // Create a new PHI node in the new region, which has an incoming value
182 // from OldPred of PN.
183 PHINode *NewPN = PHINode::Create(PN->getType(), 1 + NumPredsFromRegion,
184 PN->getName()+".ce", NewBB->begin());
185 NewPN->addIncoming(PN, OldPred);
187 // Loop over all of the incoming value in PN, moving them to NewPN if they
188 // are from the extracted region.
189 for (unsigned i = 0; i != PN->getNumIncomingValues(); ++i) {
190 if (Blocks.count(PN->getIncomingBlock(i))) {
191 NewPN->addIncoming(PN->getIncomingValue(i), PN->getIncomingBlock(i));
192 PN->removeIncomingValue(i);
200 void CodeExtractor::splitReturnBlocks() {
201 for (SetVector<BasicBlock *>::iterator I = Blocks.begin(), E = Blocks.end();
203 if (ReturnInst *RI = dyn_cast<ReturnInst>((*I)->getTerminator())) {
204 BasicBlock *New = (*I)->splitBasicBlock(RI, (*I)->getName()+".ret");
206 // Old dominates New. New node dominates all other nodes dominated
208 DomTreeNode *OldNode = DT->getNode(*I);
209 SmallVector<DomTreeNode*, 8> Children;
210 for (DomTreeNode::iterator DI = OldNode->begin(), DE = OldNode->end();
212 Children.push_back(*DI);
214 DomTreeNode *NewNode = DT->addNewBlock(New, *I);
216 for (SmallVector<DomTreeNode*, 8>::iterator I = Children.begin(),
217 E = Children.end(); I != E; ++I)
218 DT->changeImmediateDominator(*I, NewNode);
223 // findInputsOutputs - Find inputs to, outputs from the code region.
225 void CodeExtractor::findInputsOutputs(ValueSet &inputs, ValueSet &outputs) {
226 std::set<BasicBlock*> ExitBlocks;
227 for (SetVector<BasicBlock*>::const_iterator ci = Blocks.begin(),
228 ce = Blocks.end(); ci != ce; ++ci) {
229 BasicBlock *BB = *ci;
231 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
232 // If a used value is defined outside the region, it's an input. If an
233 // instruction is used outside the region, it's an output.
234 for (User::op_iterator O = I->op_begin(), E = I->op_end(); O != E; ++O)
235 if (definedInCaller(Blocks, *O))
238 // Consider uses of this instruction (outputs).
239 for (Value::use_iterator UI = I->use_begin(), E = I->use_end();
241 if (!definedInRegion(Blocks, *UI)) {
247 // Keep track of the exit blocks from the region.
248 TerminatorInst *TI = BB->getTerminator();
249 for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
250 if (!Blocks.count(TI->getSuccessor(i)))
251 ExitBlocks.insert(TI->getSuccessor(i));
252 } // for: basic blocks
254 NumExitBlocks = ExitBlocks.size();
257 /// constructFunction - make a function based on inputs and outputs, as follows:
258 /// f(in0, ..., inN, out0, ..., outN)
260 Function *CodeExtractor::constructFunction(const ValueSet &inputs,
261 const ValueSet &outputs,
263 BasicBlock *newRootNode,
264 BasicBlock *newHeader,
265 Function *oldFunction,
267 DEBUG(dbgs() << "inputs: " << inputs.size() << "\n");
268 DEBUG(dbgs() << "outputs: " << outputs.size() << "\n");
270 // This function returns unsigned, outputs will go back by reference.
271 switch (NumExitBlocks) {
273 case 1: RetTy = Type::getVoidTy(header->getContext()); break;
274 case 2: RetTy = Type::getInt1Ty(header->getContext()); break;
275 default: RetTy = Type::getInt16Ty(header->getContext()); break;
278 std::vector<Type*> paramTy;
280 // Add the types of the input values to the function's argument list
281 for (ValueSet::const_iterator i = inputs.begin(), e = inputs.end();
283 const Value *value = *i;
284 DEBUG(dbgs() << "value used in func: " << *value << "\n");
285 paramTy.push_back(value->getType());
288 // Add the types of the output values to the function's argument list.
289 for (ValueSet::const_iterator I = outputs.begin(), E = outputs.end();
291 DEBUG(dbgs() << "instr used in func: " << **I << "\n");
293 paramTy.push_back((*I)->getType());
295 paramTy.push_back(PointerType::getUnqual((*I)->getType()));
298 DEBUG(dbgs() << "Function type: " << *RetTy << " f(");
299 for (std::vector<Type*>::iterator i = paramTy.begin(),
300 e = paramTy.end(); i != e; ++i)
301 DEBUG(dbgs() << **i << ", ");
302 DEBUG(dbgs() << ")\n");
304 if (AggregateArgs && (inputs.size() + outputs.size() > 0)) {
305 PointerType *StructPtr =
306 PointerType::getUnqual(StructType::get(M->getContext(), paramTy));
308 paramTy.push_back(StructPtr);
310 FunctionType *funcType =
311 FunctionType::get(RetTy, paramTy, false);
313 // Create the new function
314 Function *newFunction = Function::Create(funcType,
315 GlobalValue::InternalLinkage,
316 oldFunction->getName() + "_" +
317 header->getName(), M);
318 // If the old function is no-throw, so is the new one.
319 if (oldFunction->doesNotThrow())
320 newFunction->setDoesNotThrow(true);
322 newFunction->getBasicBlockList().push_back(newRootNode);
324 // Create an iterator to name all of the arguments we inserted.
325 Function::arg_iterator AI = newFunction->arg_begin();
327 // Rewrite all users of the inputs in the extracted region to use the
328 // arguments (or appropriate addressing into struct) instead.
329 for (unsigned i = 0, e = inputs.size(); i != e; ++i) {
333 Idx[0] = Constant::getNullValue(Type::getInt32Ty(header->getContext()));
334 Idx[1] = ConstantInt::get(Type::getInt32Ty(header->getContext()), i);
335 TerminatorInst *TI = newFunction->begin()->getTerminator();
336 GetElementPtrInst *GEP =
337 GetElementPtrInst::Create(AI, Idx, "gep_" + inputs[i]->getName(), TI);
338 RewriteVal = new LoadInst(GEP, "loadgep_" + inputs[i]->getName(), TI);
342 std::vector<User*> Users(inputs[i]->use_begin(), inputs[i]->use_end());
343 for (std::vector<User*>::iterator use = Users.begin(), useE = Users.end();
345 if (Instruction* inst = dyn_cast<Instruction>(*use))
346 if (Blocks.count(inst->getParent()))
347 inst->replaceUsesOfWith(inputs[i], RewriteVal);
350 // Set names for input and output arguments.
351 if (!AggregateArgs) {
352 AI = newFunction->arg_begin();
353 for (unsigned i = 0, e = inputs.size(); i != e; ++i, ++AI)
354 AI->setName(inputs[i]->getName());
355 for (unsigned i = 0, e = outputs.size(); i != e; ++i, ++AI)
356 AI->setName(outputs[i]->getName()+".out");
359 // Rewrite branches to basic blocks outside of the loop to new dummy blocks
360 // within the new function. This must be done before we lose track of which
361 // blocks were originally in the code region.
362 std::vector<User*> Users(header->use_begin(), header->use_end());
363 for (unsigned i = 0, e = Users.size(); i != e; ++i)
364 // The BasicBlock which contains the branch is not in the region
365 // modify the branch target to a new block
366 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(Users[i]))
367 if (!Blocks.count(TI->getParent()) &&
368 TI->getParent()->getParent() == oldFunction)
369 TI->replaceUsesOfWith(header, newHeader);
374 /// FindPhiPredForUseInBlock - Given a value and a basic block, find a PHI
375 /// that uses the value within the basic block, and return the predecessor
376 /// block associated with that use, or return 0 if none is found.
377 static BasicBlock* FindPhiPredForUseInBlock(Value* Used, BasicBlock* BB) {
378 for (Value::use_iterator UI = Used->use_begin(),
379 UE = Used->use_end(); UI != UE; ++UI) {
380 PHINode *P = dyn_cast<PHINode>(*UI);
381 if (P && P->getParent() == BB)
382 return P->getIncomingBlock(UI);
388 /// emitCallAndSwitchStatement - This method sets up the caller side by adding
389 /// the call instruction, splitting any PHI nodes in the header block as
392 emitCallAndSwitchStatement(Function *newFunction, BasicBlock *codeReplacer,
393 ValueSet &inputs, ValueSet &outputs) {
394 // Emit a call to the new function, passing in: *pointer to struct (if
395 // aggregating parameters), or plan inputs and allocated memory for outputs
396 std::vector<Value*> params, StructValues, ReloadOutputs, Reloads;
398 LLVMContext &Context = newFunction->getContext();
400 // Add inputs as params, or to be filled into the struct
401 for (ValueSet::iterator i = inputs.begin(), e = inputs.end(); i != e; ++i)
403 StructValues.push_back(*i);
405 params.push_back(*i);
407 // Create allocas for the outputs
408 for (ValueSet::iterator i = outputs.begin(), e = outputs.end(); i != e; ++i) {
410 StructValues.push_back(*i);
413 new AllocaInst((*i)->getType(), 0, (*i)->getName()+".loc",
414 codeReplacer->getParent()->begin()->begin());
415 ReloadOutputs.push_back(alloca);
416 params.push_back(alloca);
420 AllocaInst *Struct = 0;
421 if (AggregateArgs && (inputs.size() + outputs.size() > 0)) {
422 std::vector<Type*> ArgTypes;
423 for (ValueSet::iterator v = StructValues.begin(),
424 ve = StructValues.end(); v != ve; ++v)
425 ArgTypes.push_back((*v)->getType());
427 // Allocate a struct at the beginning of this function
428 Type *StructArgTy = StructType::get(newFunction->getContext(), ArgTypes);
430 new AllocaInst(StructArgTy, 0, "structArg",
431 codeReplacer->getParent()->begin()->begin());
432 params.push_back(Struct);
434 for (unsigned i = 0, e = inputs.size(); i != e; ++i) {
436 Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context));
437 Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), i);
438 GetElementPtrInst *GEP =
439 GetElementPtrInst::Create(Struct, Idx,
440 "gep_" + StructValues[i]->getName());
441 codeReplacer->getInstList().push_back(GEP);
442 StoreInst *SI = new StoreInst(StructValues[i], GEP);
443 codeReplacer->getInstList().push_back(SI);
447 // Emit the call to the function
448 CallInst *call = CallInst::Create(newFunction, params,
449 NumExitBlocks > 1 ? "targetBlock" : "");
450 codeReplacer->getInstList().push_back(call);
452 Function::arg_iterator OutputArgBegin = newFunction->arg_begin();
453 unsigned FirstOut = inputs.size();
455 std::advance(OutputArgBegin, inputs.size());
457 // Reload the outputs passed in by reference
458 for (unsigned i = 0, e = outputs.size(); i != e; ++i) {
462 Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context));
463 Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), FirstOut + i);
464 GetElementPtrInst *GEP
465 = GetElementPtrInst::Create(Struct, Idx,
466 "gep_reload_" + outputs[i]->getName());
467 codeReplacer->getInstList().push_back(GEP);
470 Output = ReloadOutputs[i];
472 LoadInst *load = new LoadInst(Output, outputs[i]->getName()+".reload");
473 Reloads.push_back(load);
474 codeReplacer->getInstList().push_back(load);
475 std::vector<User*> Users(outputs[i]->use_begin(), outputs[i]->use_end());
476 for (unsigned u = 0, e = Users.size(); u != e; ++u) {
477 Instruction *inst = cast<Instruction>(Users[u]);
478 if (!Blocks.count(inst->getParent()))
479 inst->replaceUsesOfWith(outputs[i], load);
483 // Now we can emit a switch statement using the call as a value.
484 SwitchInst *TheSwitch =
485 SwitchInst::Create(Constant::getNullValue(Type::getInt16Ty(Context)),
486 codeReplacer, 0, codeReplacer);
488 // Since there may be multiple exits from the original region, make the new
489 // function return an unsigned, switch on that number. This loop iterates
490 // over all of the blocks in the extracted region, updating any terminator
491 // instructions in the to-be-extracted region that branch to blocks that are
492 // not in the region to be extracted.
493 std::map<BasicBlock*, BasicBlock*> ExitBlockMap;
495 unsigned switchVal = 0;
496 for (SetVector<BasicBlock*>::const_iterator i = Blocks.begin(),
497 e = Blocks.end(); i != e; ++i) {
498 TerminatorInst *TI = (*i)->getTerminator();
499 for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
500 if (!Blocks.count(TI->getSuccessor(i))) {
501 BasicBlock *OldTarget = TI->getSuccessor(i);
502 // add a new basic block which returns the appropriate value
503 BasicBlock *&NewTarget = ExitBlockMap[OldTarget];
505 // If we don't already have an exit stub for this non-extracted
506 // destination, create one now!
507 NewTarget = BasicBlock::Create(Context,
508 OldTarget->getName() + ".exitStub",
510 unsigned SuccNum = switchVal++;
513 switch (NumExitBlocks) {
515 case 1: break; // No value needed.
516 case 2: // Conditional branch, return a bool
517 brVal = ConstantInt::get(Type::getInt1Ty(Context), !SuccNum);
520 brVal = ConstantInt::get(Type::getInt16Ty(Context), SuccNum);
524 ReturnInst *NTRet = ReturnInst::Create(Context, brVal, NewTarget);
526 // Update the switch instruction.
527 TheSwitch->addCase(ConstantInt::get(Type::getInt16Ty(Context),
531 // Restore values just before we exit
532 Function::arg_iterator OAI = OutputArgBegin;
533 for (unsigned out = 0, e = outputs.size(); out != e; ++out) {
534 // For an invoke, the normal destination is the only one that is
535 // dominated by the result of the invocation
536 BasicBlock *DefBlock = cast<Instruction>(outputs[out])->getParent();
538 bool DominatesDef = true;
540 if (InvokeInst *Invoke = dyn_cast<InvokeInst>(outputs[out])) {
541 DefBlock = Invoke->getNormalDest();
543 // Make sure we are looking at the original successor block, not
544 // at a newly inserted exit block, which won't be in the dominator
546 for (std::map<BasicBlock*, BasicBlock*>::iterator I =
547 ExitBlockMap.begin(), E = ExitBlockMap.end(); I != E; ++I)
548 if (DefBlock == I->second) {
553 // In the extract block case, if the block we are extracting ends
554 // with an invoke instruction, make sure that we don't emit a
555 // store of the invoke value for the unwind block.
556 if (!DT && DefBlock != OldTarget)
557 DominatesDef = false;
561 DominatesDef = DT->dominates(DefBlock, OldTarget);
563 // If the output value is used by a phi in the target block,
564 // then we need to test for dominance of the phi's predecessor
565 // instead. Unfortunately, this a little complicated since we
566 // have already rewritten uses of the value to uses of the reload.
567 BasicBlock* pred = FindPhiPredForUseInBlock(Reloads[out],
569 if (pred && DT && DT->dominates(DefBlock, pred))
576 Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context));
577 Idx[1] = ConstantInt::get(Type::getInt32Ty(Context),
579 GetElementPtrInst *GEP =
580 GetElementPtrInst::Create(OAI, Idx,
581 "gep_" + outputs[out]->getName(),
583 new StoreInst(outputs[out], GEP, NTRet);
585 new StoreInst(outputs[out], OAI, NTRet);
588 // Advance output iterator even if we don't emit a store
589 if (!AggregateArgs) ++OAI;
593 // rewrite the original branch instruction with this new target
594 TI->setSuccessor(i, NewTarget);
598 // Now that we've done the deed, simplify the switch instruction.
599 Type *OldFnRetTy = TheSwitch->getParent()->getParent()->getReturnType();
600 switch (NumExitBlocks) {
602 // There are no successors (the block containing the switch itself), which
603 // means that previously this was the last part of the function, and hence
604 // this should be rewritten as a `ret'
606 // Check if the function should return a value
607 if (OldFnRetTy->isVoidTy()) {
608 ReturnInst::Create(Context, 0, TheSwitch); // Return void
609 } else if (OldFnRetTy == TheSwitch->getCondition()->getType()) {
610 // return what we have
611 ReturnInst::Create(Context, TheSwitch->getCondition(), TheSwitch);
613 // Otherwise we must have code extracted an unwind or something, just
614 // return whatever we want.
615 ReturnInst::Create(Context,
616 Constant::getNullValue(OldFnRetTy), TheSwitch);
619 TheSwitch->eraseFromParent();
622 // Only a single destination, change the switch into an unconditional
624 BranchInst::Create(TheSwitch->getSuccessor(1), TheSwitch);
625 TheSwitch->eraseFromParent();
628 BranchInst::Create(TheSwitch->getSuccessor(1), TheSwitch->getSuccessor(2),
630 TheSwitch->eraseFromParent();
633 // Otherwise, make the default destination of the switch instruction be one
634 // of the other successors.
635 TheSwitch->setCondition(call);
636 TheSwitch->setDefaultDest(TheSwitch->getSuccessor(NumExitBlocks));
637 // Remove redundant case
638 TheSwitch->removeCase(SwitchInst::CaseIt(TheSwitch, NumExitBlocks-1));
643 void CodeExtractor::moveCodeToFunction(Function *newFunction) {
644 Function *oldFunc = (*Blocks.begin())->getParent();
645 Function::BasicBlockListType &oldBlocks = oldFunc->getBasicBlockList();
646 Function::BasicBlockListType &newBlocks = newFunction->getBasicBlockList();
648 for (SetVector<BasicBlock*>::const_iterator i = Blocks.begin(),
649 e = Blocks.end(); i != e; ++i) {
650 // Delete the basic block from the old function, and the list of blocks
651 oldBlocks.remove(*i);
653 // Insert this basic block into the new function
654 newBlocks.push_back(*i);
658 Function *CodeExtractor::extractCodeRegion() {
662 ValueSet inputs, outputs;
664 // Assumption: this is a single-entry code region, and the header is the first
665 // block in the region.
666 BasicBlock *header = *Blocks.begin();
668 for (SetVector<BasicBlock *>::iterator BI = llvm::next(Blocks.begin()),
671 for (pred_iterator PI = pred_begin(*BI), E = pred_end(*BI);
673 assert(Blocks.count(*PI) &&
674 "No blocks in this region may have entries from outside the region"
675 " except for the first block!");
677 // If we have to split PHI nodes or the entry block, do so now.
678 severSplitPHINodes(header);
680 // If we have any return instructions in the region, split those blocks so
681 // that the return is not in the region.
684 Function *oldFunction = header->getParent();
686 // This takes place of the original loop
687 BasicBlock *codeReplacer = BasicBlock::Create(header->getContext(),
688 "codeRepl", oldFunction,
691 // The new function needs a root node because other nodes can branch to the
692 // head of the region, but the entry node of a function cannot have preds.
693 BasicBlock *newFuncRoot = BasicBlock::Create(header->getContext(),
695 newFuncRoot->getInstList().push_back(BranchInst::Create(header));
697 // Find inputs to, outputs from the code region.
698 findInputsOutputs(inputs, outputs);
700 // Construct new function based on inputs/outputs & add allocas for all defs.
701 Function *newFunction = constructFunction(inputs, outputs, header,
703 codeReplacer, oldFunction,
704 oldFunction->getParent());
706 emitCallAndSwitchStatement(newFunction, codeReplacer, inputs, outputs);
708 moveCodeToFunction(newFunction);
710 // Loop over all of the PHI nodes in the header block, and change any
711 // references to the old incoming edge to be the new incoming edge.
712 for (BasicBlock::iterator I = header->begin(); isa<PHINode>(I); ++I) {
713 PHINode *PN = cast<PHINode>(I);
714 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
715 if (!Blocks.count(PN->getIncomingBlock(i)))
716 PN->setIncomingBlock(i, newFuncRoot);
719 // Look at all successors of the codeReplacer block. If any of these blocks
720 // had PHI nodes in them, we need to update the "from" block to be the code
721 // replacer, not the original block in the extracted region.
722 std::vector<BasicBlock*> Succs(succ_begin(codeReplacer),
723 succ_end(codeReplacer));
724 for (unsigned i = 0, e = Succs.size(); i != e; ++i)
725 for (BasicBlock::iterator I = Succs[i]->begin(); isa<PHINode>(I); ++I) {
726 PHINode *PN = cast<PHINode>(I);
727 std::set<BasicBlock*> ProcessedPreds;
728 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
729 if (Blocks.count(PN->getIncomingBlock(i))) {
730 if (ProcessedPreds.insert(PN->getIncomingBlock(i)).second)
731 PN->setIncomingBlock(i, codeReplacer);
733 // There were multiple entries in the PHI for this block, now there
734 // is only one, so remove the duplicated entries.
735 PN->removeIncomingValue(i, false);
741 //cerr << "NEW FUNCTION: " << *newFunction;
742 // verifyFunction(*newFunction);
744 // cerr << "OLD FUNCTION: " << *oldFunction;
745 // verifyFunction(*oldFunction);
747 DEBUG(if (verifyFunction(*newFunction))
748 report_fatal_error("verifyFunction failed!"));