xref: /freebsd/contrib/llvm-project/llvm/lib/Transforms/Utils/CodeExtractor.cpp (revision eaff4c4f92dc447cf5f0bbf80449cad7738fdf39)
1  //===- CodeExtractor.cpp - Pull code region into a new function -----------===//
2  //
3  // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4  // See https://llvm.org/LICENSE.txt for license information.
5  // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6  //
7  //===----------------------------------------------------------------------===//
8  //
9  // This file implements the interface to tear out a code region, such as an
10  // individual loop or a parallel section, into a new function, replacing it with
11  // a call to the new function.
12  //
13  //===----------------------------------------------------------------------===//
14  
15  #include "llvm/Transforms/Utils/CodeExtractor.h"
16  #include "llvm/ADT/ArrayRef.h"
17  #include "llvm/ADT/DenseMap.h"
18  #include "llvm/ADT/STLExtras.h"
19  #include "llvm/ADT/SetVector.h"
20  #include "llvm/ADT/SmallPtrSet.h"
21  #include "llvm/ADT/SmallVector.h"
22  #include "llvm/Analysis/AssumptionCache.h"
23  #include "llvm/Analysis/BlockFrequencyInfo.h"
24  #include "llvm/Analysis/BlockFrequencyInfoImpl.h"
25  #include "llvm/Analysis/BranchProbabilityInfo.h"
26  #include "llvm/Analysis/LoopInfo.h"
27  #include "llvm/IR/Argument.h"
28  #include "llvm/IR/Attributes.h"
29  #include "llvm/IR/BasicBlock.h"
30  #include "llvm/IR/CFG.h"
31  #include "llvm/IR/Constant.h"
32  #include "llvm/IR/Constants.h"
33  #include "llvm/IR/DIBuilder.h"
34  #include "llvm/IR/DataLayout.h"
35  #include "llvm/IR/DebugInfo.h"
36  #include "llvm/IR/DebugInfoMetadata.h"
37  #include "llvm/IR/DerivedTypes.h"
38  #include "llvm/IR/Dominators.h"
39  #include "llvm/IR/Function.h"
40  #include "llvm/IR/GlobalValue.h"
41  #include "llvm/IR/InstIterator.h"
42  #include "llvm/IR/InstrTypes.h"
43  #include "llvm/IR/Instruction.h"
44  #include "llvm/IR/Instructions.h"
45  #include "llvm/IR/IntrinsicInst.h"
46  #include "llvm/IR/Intrinsics.h"
47  #include "llvm/IR/LLVMContext.h"
48  #include "llvm/IR/MDBuilder.h"
49  #include "llvm/IR/Module.h"
50  #include "llvm/IR/PatternMatch.h"
51  #include "llvm/IR/Type.h"
52  #include "llvm/IR/User.h"
53  #include "llvm/IR/Value.h"
54  #include "llvm/IR/Verifier.h"
55  #include "llvm/Support/BlockFrequency.h"
56  #include "llvm/Support/BranchProbability.h"
57  #include "llvm/Support/Casting.h"
58  #include "llvm/Support/CommandLine.h"
59  #include "llvm/Support/Debug.h"
60  #include "llvm/Support/ErrorHandling.h"
61  #include "llvm/Support/raw_ostream.h"
62  #include "llvm/Transforms/Utils/BasicBlockUtils.h"
63  #include <cassert>
64  #include <cstdint>
65  #include <iterator>
66  #include <map>
67  #include <utility>
68  #include <vector>
69  
70  using namespace llvm;
71  using namespace llvm::PatternMatch;
72  using ProfileCount = Function::ProfileCount;
73  
74  #define DEBUG_TYPE "code-extractor"
75  
76  // Provide a command-line option to aggregate function arguments into a struct
77  // for functions produced by the code extractor. This is useful when converting
78  // extracted functions to pthread-based code, as only one argument (void*) can
79  // be passed in to pthread_create().
80  static cl::opt<bool>
81  AggregateArgsOpt("aggregate-extracted-args", cl::Hidden,
82                   cl::desc("Aggregate arguments to code-extracted functions"));
83  
84  /// Test whether a block is valid for extraction.
85  static bool isBlockValidForExtraction(const BasicBlock &BB,
86                                        const SetVector<BasicBlock *> &Result,
87                                        bool AllowVarArgs, bool AllowAlloca) {
88    // taking the address of a basic block moved to another function is illegal
89    if (BB.hasAddressTaken())
90      return false;
91  
92    // don't hoist code that uses another basicblock address, as it's likely to
93    // lead to unexpected behavior, like cross-function jumps
94    SmallPtrSet<User const *, 16> Visited;
95    SmallVector<User const *, 16> ToVisit;
96  
97    for (Instruction const &Inst : BB)
98      ToVisit.push_back(&Inst);
99  
100    while (!ToVisit.empty()) {
101      User const *Curr = ToVisit.pop_back_val();
102      if (!Visited.insert(Curr).second)
103        continue;
104      if (isa<BlockAddress const>(Curr))
105        return false; // even a reference to self is likely to be not compatible
106  
107      if (isa<Instruction>(Curr) && cast<Instruction>(Curr)->getParent() != &BB)
108        continue;
109  
110      for (auto const &U : Curr->operands()) {
111        if (auto *UU = dyn_cast<User>(U))
112          ToVisit.push_back(UU);
113      }
114    }
115  
116    // If explicitly requested, allow vastart and alloca. For invoke instructions
117    // verify that extraction is valid.
118    for (BasicBlock::const_iterator I = BB.begin(), E = BB.end(); I != E; ++I) {
119      if (isa<AllocaInst>(I)) {
120         if (!AllowAlloca)
121           return false;
122         continue;
123      }
124  
125      if (const auto *II = dyn_cast<InvokeInst>(I)) {
126        // Unwind destination (either a landingpad, catchswitch, or cleanuppad)
127        // must be a part of the subgraph which is being extracted.
128        if (auto *UBB = II->getUnwindDest())
129          if (!Result.count(UBB))
130            return false;
131        continue;
132      }
133  
134      // All catch handlers of a catchswitch instruction as well as the unwind
135      // destination must be in the subgraph.
136      if (const auto *CSI = dyn_cast<CatchSwitchInst>(I)) {
137        if (auto *UBB = CSI->getUnwindDest())
138          if (!Result.count(UBB))
139            return false;
140        for (const auto *HBB : CSI->handlers())
141          if (!Result.count(const_cast<BasicBlock*>(HBB)))
142            return false;
143        continue;
144      }
145  
146      // Make sure that entire catch handler is within subgraph. It is sufficient
147      // to check that catch return's block is in the list.
148      if (const auto *CPI = dyn_cast<CatchPadInst>(I)) {
149        for (const auto *U : CPI->users())
150          if (const auto *CRI = dyn_cast<CatchReturnInst>(U))
151            if (!Result.count(const_cast<BasicBlock*>(CRI->getParent())))
152              return false;
153        continue;
154      }
155  
156      // And do similar checks for cleanup handler - the entire handler must be
157      // in subgraph which is going to be extracted. For cleanup return should
158      // additionally check that the unwind destination is also in the subgraph.
159      if (const auto *CPI = dyn_cast<CleanupPadInst>(I)) {
160        for (const auto *U : CPI->users())
161          if (const auto *CRI = dyn_cast<CleanupReturnInst>(U))
162            if (!Result.count(const_cast<BasicBlock*>(CRI->getParent())))
163              return false;
164        continue;
165      }
166      if (const auto *CRI = dyn_cast<CleanupReturnInst>(I)) {
167        if (auto *UBB = CRI->getUnwindDest())
168          if (!Result.count(UBB))
169            return false;
170        continue;
171      }
172  
173      if (const CallInst *CI = dyn_cast<CallInst>(I)) {
174        if (const Function *F = CI->getCalledFunction()) {
175          auto IID = F->getIntrinsicID();
176          if (IID == Intrinsic::vastart) {
177            if (AllowVarArgs)
178              continue;
179            else
180              return false;
181          }
182  
183          // Currently, we miscompile outlined copies of eh_typid_for. There are
184          // proposals for fixing this in llvm.org/PR39545.
185          if (IID == Intrinsic::eh_typeid_for)
186            return false;
187        }
188      }
189    }
190  
191    return true;
192  }
193  
194  /// Build a set of blocks to extract if the input blocks are viable.
195  static SetVector<BasicBlock *>
196  buildExtractionBlockSet(ArrayRef<BasicBlock *> BBs, DominatorTree *DT,
197                          bool AllowVarArgs, bool AllowAlloca) {
198    assert(!BBs.empty() && "The set of blocks to extract must be non-empty");
199    SetVector<BasicBlock *> Result;
200  
201    // Loop over the blocks, adding them to our set-vector, and aborting with an
202    // empty set if we encounter invalid blocks.
203    for (BasicBlock *BB : BBs) {
204      // If this block is dead, don't process it.
205      if (DT && !DT->isReachableFromEntry(BB))
206        continue;
207  
208      if (!Result.insert(BB))
209        llvm_unreachable("Repeated basic blocks in extraction input");
210    }
211  
212    LLVM_DEBUG(dbgs() << "Region front block: " << Result.front()->getName()
213                      << '\n');
214  
215    for (auto *BB : Result) {
216      if (!isBlockValidForExtraction(*BB, Result, AllowVarArgs, AllowAlloca))
217        return {};
218  
219      // Make sure that the first block is not a landing pad.
220      if (BB == Result.front()) {
221        if (BB->isEHPad()) {
222          LLVM_DEBUG(dbgs() << "The first block cannot be an unwind block\n");
223          return {};
224        }
225        continue;
226      }
227  
228      // All blocks other than the first must not have predecessors outside of
229      // the subgraph which is being extracted.
230      for (auto *PBB : predecessors(BB))
231        if (!Result.count(PBB)) {
232          LLVM_DEBUG(dbgs() << "No blocks in this region may have entries from "
233                               "outside the region except for the first block!\n"
234                            << "Problematic source BB: " << BB->getName() << "\n"
235                            << "Problematic destination BB: " << PBB->getName()
236                            << "\n");
237          return {};
238        }
239    }
240  
241    return Result;
242  }
243  
244  CodeExtractor::CodeExtractor(ArrayRef<BasicBlock *> BBs, DominatorTree *DT,
245                               bool AggregateArgs, BlockFrequencyInfo *BFI,
246                               BranchProbabilityInfo *BPI, AssumptionCache *AC,
247                               bool AllowVarArgs, bool AllowAlloca,
248                               BasicBlock *AllocationBlock, std::string Suffix,
249                               bool ArgsInZeroAddressSpace)
250      : DT(DT), AggregateArgs(AggregateArgs || AggregateArgsOpt), BFI(BFI),
251        BPI(BPI), AC(AC), AllocationBlock(AllocationBlock),
252        AllowVarArgs(AllowVarArgs),
253        Blocks(buildExtractionBlockSet(BBs, DT, AllowVarArgs, AllowAlloca)),
254        Suffix(Suffix), ArgsInZeroAddressSpace(ArgsInZeroAddressSpace) {}
255  
256  CodeExtractor::CodeExtractor(DominatorTree &DT, Loop &L, bool AggregateArgs,
257                               BlockFrequencyInfo *BFI,
258                               BranchProbabilityInfo *BPI, AssumptionCache *AC,
259                               std::string Suffix)
260      : DT(&DT), AggregateArgs(AggregateArgs || AggregateArgsOpt), BFI(BFI),
261        BPI(BPI), AC(AC), AllocationBlock(nullptr), AllowVarArgs(false),
262        Blocks(buildExtractionBlockSet(L.getBlocks(), &DT,
263                                       /* AllowVarArgs */ false,
264                                       /* AllowAlloca */ false)),
265        Suffix(Suffix) {}
266  
267  /// definedInRegion - Return true if the specified value is defined in the
268  /// extracted region.
269  static bool definedInRegion(const SetVector<BasicBlock *> &Blocks, Value *V) {
270    if (Instruction *I = dyn_cast<Instruction>(V))
271      if (Blocks.count(I->getParent()))
272        return true;
273    return false;
274  }
275  
276  /// definedInCaller - Return true if the specified value is defined in the
277  /// function being code extracted, but not in the region being extracted.
278  /// These values must be passed in as live-ins to the function.
279  static bool definedInCaller(const SetVector<BasicBlock *> &Blocks, Value *V) {
280    if (isa<Argument>(V)) return true;
281    if (Instruction *I = dyn_cast<Instruction>(V))
282      if (!Blocks.count(I->getParent()))
283        return true;
284    return false;
285  }
286  
287  static BasicBlock *getCommonExitBlock(const SetVector<BasicBlock *> &Blocks) {
288    BasicBlock *CommonExitBlock = nullptr;
289    auto hasNonCommonExitSucc = [&](BasicBlock *Block) {
290      for (auto *Succ : successors(Block)) {
291        // Internal edges, ok.
292        if (Blocks.count(Succ))
293          continue;
294        if (!CommonExitBlock) {
295          CommonExitBlock = Succ;
296          continue;
297        }
298        if (CommonExitBlock != Succ)
299          return true;
300      }
301      return false;
302    };
303  
304    if (any_of(Blocks, hasNonCommonExitSucc))
305      return nullptr;
306  
307    return CommonExitBlock;
308  }
309  
310  CodeExtractorAnalysisCache::CodeExtractorAnalysisCache(Function &F) {
311    for (BasicBlock &BB : F) {
312      for (Instruction &II : BB.instructionsWithoutDebug())
313        if (auto *AI = dyn_cast<AllocaInst>(&II))
314          Allocas.push_back(AI);
315  
316      findSideEffectInfoForBlock(BB);
317    }
318  }
319  
320  void CodeExtractorAnalysisCache::findSideEffectInfoForBlock(BasicBlock &BB) {
321    for (Instruction &II : BB.instructionsWithoutDebug()) {
322      unsigned Opcode = II.getOpcode();
323      Value *MemAddr = nullptr;
324      switch (Opcode) {
325      case Instruction::Store:
326      case Instruction::Load: {
327        if (Opcode == Instruction::Store) {
328          StoreInst *SI = cast<StoreInst>(&II);
329          MemAddr = SI->getPointerOperand();
330        } else {
331          LoadInst *LI = cast<LoadInst>(&II);
332          MemAddr = LI->getPointerOperand();
333        }
334        // Global variable can not be aliased with locals.
335        if (isa<Constant>(MemAddr))
336          break;
337        Value *Base = MemAddr->stripInBoundsConstantOffsets();
338        if (!isa<AllocaInst>(Base)) {
339          SideEffectingBlocks.insert(&BB);
340          return;
341        }
342        BaseMemAddrs[&BB].insert(Base);
343        break;
344      }
345      default: {
346        IntrinsicInst *IntrInst = dyn_cast<IntrinsicInst>(&II);
347        if (IntrInst) {
348          if (IntrInst->isLifetimeStartOrEnd())
349            break;
350          SideEffectingBlocks.insert(&BB);
351          return;
352        }
353        // Treat all the other cases conservatively if it has side effects.
354        if (II.mayHaveSideEffects()) {
355          SideEffectingBlocks.insert(&BB);
356          return;
357        }
358      }
359      }
360    }
361  }
362  
363  bool CodeExtractorAnalysisCache::doesBlockContainClobberOfAddr(
364      BasicBlock &BB, AllocaInst *Addr) const {
365    if (SideEffectingBlocks.count(&BB))
366      return true;
367    auto It = BaseMemAddrs.find(&BB);
368    if (It != BaseMemAddrs.end())
369      return It->second.count(Addr);
370    return false;
371  }
372  
373  bool CodeExtractor::isLegalToShrinkwrapLifetimeMarkers(
374      const CodeExtractorAnalysisCache &CEAC, Instruction *Addr) const {
375    AllocaInst *AI = cast<AllocaInst>(Addr->stripInBoundsConstantOffsets());
376    Function *Func = (*Blocks.begin())->getParent();
377    for (BasicBlock &BB : *Func) {
378      if (Blocks.count(&BB))
379        continue;
380      if (CEAC.doesBlockContainClobberOfAddr(BB, AI))
381        return false;
382    }
383    return true;
384  }
385  
386  BasicBlock *
387  CodeExtractor::findOrCreateBlockForHoisting(BasicBlock *CommonExitBlock) {
388    BasicBlock *SinglePredFromOutlineRegion = nullptr;
389    assert(!Blocks.count(CommonExitBlock) &&
390           "Expect a block outside the region!");
391    for (auto *Pred : predecessors(CommonExitBlock)) {
392      if (!Blocks.count(Pred))
393        continue;
394      if (!SinglePredFromOutlineRegion) {
395        SinglePredFromOutlineRegion = Pred;
396      } else if (SinglePredFromOutlineRegion != Pred) {
397        SinglePredFromOutlineRegion = nullptr;
398        break;
399      }
400    }
401  
402    if (SinglePredFromOutlineRegion)
403      return SinglePredFromOutlineRegion;
404  
405  #ifndef NDEBUG
406    auto getFirstPHI = [](BasicBlock *BB) {
407      BasicBlock::iterator I = BB->begin();
408      PHINode *FirstPhi = nullptr;
409      while (I != BB->end()) {
410        PHINode *Phi = dyn_cast<PHINode>(I);
411        if (!Phi)
412          break;
413        if (!FirstPhi) {
414          FirstPhi = Phi;
415          break;
416        }
417      }
418      return FirstPhi;
419    };
420    // If there are any phi nodes, the single pred either exists or has already
421    // be created before code extraction.
422    assert(!getFirstPHI(CommonExitBlock) && "Phi not expected");
423  #endif
424  
425    BasicBlock *NewExitBlock = CommonExitBlock->splitBasicBlock(
426        CommonExitBlock->getFirstNonPHI()->getIterator());
427  
428    for (BasicBlock *Pred :
429         llvm::make_early_inc_range(predecessors(CommonExitBlock))) {
430      if (Blocks.count(Pred))
431        continue;
432      Pred->getTerminator()->replaceUsesOfWith(CommonExitBlock, NewExitBlock);
433    }
434    // Now add the old exit block to the outline region.
435    Blocks.insert(CommonExitBlock);
436    OldTargets.push_back(NewExitBlock);
437    return CommonExitBlock;
438  }
439  
440  // Find the pair of life time markers for address 'Addr' that are either
441  // defined inside the outline region or can legally be shrinkwrapped into the
442  // outline region. If there are not other untracked uses of the address, return
443  // the pair of markers if found; otherwise return a pair of nullptr.
444  CodeExtractor::LifetimeMarkerInfo
445  CodeExtractor::getLifetimeMarkers(const CodeExtractorAnalysisCache &CEAC,
446                                    Instruction *Addr,
447                                    BasicBlock *ExitBlock) const {
448    LifetimeMarkerInfo Info;
449  
450    for (User *U : Addr->users()) {
451      IntrinsicInst *IntrInst = dyn_cast<IntrinsicInst>(U);
452      if (IntrInst) {
453        // We don't model addresses with multiple start/end markers, but the
454        // markers do not need to be in the region.
455        if (IntrInst->getIntrinsicID() == Intrinsic::lifetime_start) {
456          if (Info.LifeStart)
457            return {};
458          Info.LifeStart = IntrInst;
459          continue;
460        }
461        if (IntrInst->getIntrinsicID() == Intrinsic::lifetime_end) {
462          if (Info.LifeEnd)
463            return {};
464          Info.LifeEnd = IntrInst;
465          continue;
466        }
467        // At this point, permit debug uses outside of the region.
468        // This is fixed in a later call to fixupDebugInfoPostExtraction().
469        if (isa<DbgInfoIntrinsic>(IntrInst))
470          continue;
471      }
472      // Find untracked uses of the address, bail.
473      if (!definedInRegion(Blocks, U))
474        return {};
475    }
476  
477    if (!Info.LifeStart || !Info.LifeEnd)
478      return {};
479  
480    Info.SinkLifeStart = !definedInRegion(Blocks, Info.LifeStart);
481    Info.HoistLifeEnd = !definedInRegion(Blocks, Info.LifeEnd);
482    // Do legality check.
483    if ((Info.SinkLifeStart || Info.HoistLifeEnd) &&
484        !isLegalToShrinkwrapLifetimeMarkers(CEAC, Addr))
485      return {};
486  
487    // Check to see if we have a place to do hoisting, if not, bail.
488    if (Info.HoistLifeEnd && !ExitBlock)
489      return {};
490  
491    return Info;
492  }
493  
494  void CodeExtractor::findAllocas(const CodeExtractorAnalysisCache &CEAC,
495                                  ValueSet &SinkCands, ValueSet &HoistCands,
496                                  BasicBlock *&ExitBlock) const {
497    Function *Func = (*Blocks.begin())->getParent();
498    ExitBlock = getCommonExitBlock(Blocks);
499  
500    auto moveOrIgnoreLifetimeMarkers =
501        [&](const LifetimeMarkerInfo &LMI) -> bool {
502      if (!LMI.LifeStart)
503        return false;
504      if (LMI.SinkLifeStart) {
505        LLVM_DEBUG(dbgs() << "Sinking lifetime.start: " << *LMI.LifeStart
506                          << "\n");
507        SinkCands.insert(LMI.LifeStart);
508      }
509      if (LMI.HoistLifeEnd) {
510        LLVM_DEBUG(dbgs() << "Hoisting lifetime.end: " << *LMI.LifeEnd << "\n");
511        HoistCands.insert(LMI.LifeEnd);
512      }
513      return true;
514    };
515  
516    // Look up allocas in the original function in CodeExtractorAnalysisCache, as
517    // this is much faster than walking all the instructions.
518    for (AllocaInst *AI : CEAC.getAllocas()) {
519      BasicBlock *BB = AI->getParent();
520      if (Blocks.count(BB))
521        continue;
522  
523      // As a prior call to extractCodeRegion() may have shrinkwrapped the alloca,
524      // check whether it is actually still in the original function.
525      Function *AIFunc = BB->getParent();
526      if (AIFunc != Func)
527        continue;
528  
529      LifetimeMarkerInfo MarkerInfo = getLifetimeMarkers(CEAC, AI, ExitBlock);
530      bool Moved = moveOrIgnoreLifetimeMarkers(MarkerInfo);
531      if (Moved) {
532        LLVM_DEBUG(dbgs() << "Sinking alloca: " << *AI << "\n");
533        SinkCands.insert(AI);
534        continue;
535      }
536  
537      // Find bitcasts in the outlined region that have lifetime marker users
538      // outside that region. Replace the lifetime marker use with an
539      // outside region bitcast to avoid unnecessary alloca/reload instructions
540      // and extra lifetime markers.
541      SmallVector<Instruction *, 2> LifetimeBitcastUsers;
542      for (User *U : AI->users()) {
543        if (!definedInRegion(Blocks, U))
544          continue;
545  
546        if (U->stripInBoundsConstantOffsets() != AI)
547          continue;
548  
549        Instruction *Bitcast = cast<Instruction>(U);
550        for (User *BU : Bitcast->users()) {
551          IntrinsicInst *IntrInst = dyn_cast<IntrinsicInst>(BU);
552          if (!IntrInst)
553            continue;
554  
555          if (!IntrInst->isLifetimeStartOrEnd())
556            continue;
557  
558          if (definedInRegion(Blocks, IntrInst))
559            continue;
560  
561          LLVM_DEBUG(dbgs() << "Replace use of extracted region bitcast"
562                            << *Bitcast << " in out-of-region lifetime marker "
563                            << *IntrInst << "\n");
564          LifetimeBitcastUsers.push_back(IntrInst);
565        }
566      }
567  
568      for (Instruction *I : LifetimeBitcastUsers) {
569        Module *M = AIFunc->getParent();
570        LLVMContext &Ctx = M->getContext();
571        auto *Int8PtrTy = PointerType::getUnqual(Ctx);
572        CastInst *CastI =
573            CastInst::CreatePointerCast(AI, Int8PtrTy, "lt.cast", I);
574        I->replaceUsesOfWith(I->getOperand(1), CastI);
575      }
576  
577      // Follow any bitcasts.
578      SmallVector<Instruction *, 2> Bitcasts;
579      SmallVector<LifetimeMarkerInfo, 2> BitcastLifetimeInfo;
580      for (User *U : AI->users()) {
581        if (U->stripInBoundsConstantOffsets() == AI) {
582          Instruction *Bitcast = cast<Instruction>(U);
583          LifetimeMarkerInfo LMI = getLifetimeMarkers(CEAC, Bitcast, ExitBlock);
584          if (LMI.LifeStart) {
585            Bitcasts.push_back(Bitcast);
586            BitcastLifetimeInfo.push_back(LMI);
587            continue;
588          }
589        }
590  
591        // Found unknown use of AI.
592        if (!definedInRegion(Blocks, U)) {
593          Bitcasts.clear();
594          break;
595        }
596      }
597  
598      // Either no bitcasts reference the alloca or there are unknown uses.
599      if (Bitcasts.empty())
600        continue;
601  
602      LLVM_DEBUG(dbgs() << "Sinking alloca (via bitcast): " << *AI << "\n");
603      SinkCands.insert(AI);
604      for (unsigned I = 0, E = Bitcasts.size(); I != E; ++I) {
605        Instruction *BitcastAddr = Bitcasts[I];
606        const LifetimeMarkerInfo &LMI = BitcastLifetimeInfo[I];
607        assert(LMI.LifeStart &&
608               "Unsafe to sink bitcast without lifetime markers");
609        moveOrIgnoreLifetimeMarkers(LMI);
610        if (!definedInRegion(Blocks, BitcastAddr)) {
611          LLVM_DEBUG(dbgs() << "Sinking bitcast-of-alloca: " << *BitcastAddr
612                            << "\n");
613          SinkCands.insert(BitcastAddr);
614        }
615      }
616    }
617  }
618  
619  bool CodeExtractor::isEligible() const {
620    if (Blocks.empty())
621      return false;
622    BasicBlock *Header = *Blocks.begin();
623    Function *F = Header->getParent();
624  
625    // For functions with varargs, check that varargs handling is only done in the
626    // outlined function, i.e vastart and vaend are only used in outlined blocks.
627    if (AllowVarArgs && F->getFunctionType()->isVarArg()) {
628      auto containsVarArgIntrinsic = [](const Instruction &I) {
629        if (const CallInst *CI = dyn_cast<CallInst>(&I))
630          if (const Function *Callee = CI->getCalledFunction())
631            return Callee->getIntrinsicID() == Intrinsic::vastart ||
632                   Callee->getIntrinsicID() == Intrinsic::vaend;
633        return false;
634      };
635  
636      for (auto &BB : *F) {
637        if (Blocks.count(&BB))
638          continue;
639        if (llvm::any_of(BB, containsVarArgIntrinsic))
640          return false;
641      }
642    }
643    return true;
644  }
645  
646  void CodeExtractor::findInputsOutputs(ValueSet &Inputs, ValueSet &Outputs,
647                                        const ValueSet &SinkCands) const {
648    for (BasicBlock *BB : Blocks) {
649      // If a used value is defined outside the region, it's an input.  If an
650      // instruction is used outside the region, it's an output.
651      for (Instruction &II : *BB) {
652        for (auto &OI : II.operands()) {
653          Value *V = OI;
654          if (!SinkCands.count(V) && definedInCaller(Blocks, V))
655            Inputs.insert(V);
656        }
657  
658        for (User *U : II.users())
659          if (!definedInRegion(Blocks, U)) {
660            Outputs.insert(&II);
661            break;
662          }
663      }
664    }
665  }
666  
667  /// severSplitPHINodesOfEntry - If a PHI node has multiple inputs from outside
668  /// of the region, we need to split the entry block of the region so that the
669  /// PHI node is easier to deal with.
670  void CodeExtractor::severSplitPHINodesOfEntry(BasicBlock *&Header) {
671    unsigned NumPredsFromRegion = 0;
672    unsigned NumPredsOutsideRegion = 0;
673  
674    if (Header != &Header->getParent()->getEntryBlock()) {
675      PHINode *PN = dyn_cast<PHINode>(Header->begin());
676      if (!PN) return;  // No PHI nodes.
677  
678      // If the header node contains any PHI nodes, check to see if there is more
679      // than one entry from outside the region.  If so, we need to sever the
680      // header block into two.
681      for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
682        if (Blocks.count(PN->getIncomingBlock(i)))
683          ++NumPredsFromRegion;
684        else
685          ++NumPredsOutsideRegion;
686  
687      // If there is one (or fewer) predecessor from outside the region, we don't
688      // need to do anything special.
689      if (NumPredsOutsideRegion <= 1) return;
690    }
691  
692    // Otherwise, we need to split the header block into two pieces: one
693    // containing PHI nodes merging values from outside of the region, and a
694    // second that contains all of the code for the block and merges back any
695    // incoming values from inside of the region.
696    BasicBlock *NewBB = SplitBlock(Header, Header->getFirstNonPHI(), DT);
697  
698    // We only want to code extract the second block now, and it becomes the new
699    // header of the region.
700    BasicBlock *OldPred = Header;
701    Blocks.remove(OldPred);
702    Blocks.insert(NewBB);
703    Header = NewBB;
704  
705    // Okay, now we need to adjust the PHI nodes and any branches from within the
706    // region to go to the new header block instead of the old header block.
707    if (NumPredsFromRegion) {
708      PHINode *PN = cast<PHINode>(OldPred->begin());
709      // Loop over all of the predecessors of OldPred that are in the region,
710      // changing them to branch to NewBB instead.
711      for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
712        if (Blocks.count(PN->getIncomingBlock(i))) {
713          Instruction *TI = PN->getIncomingBlock(i)->getTerminator();
714          TI->replaceUsesOfWith(OldPred, NewBB);
715        }
716  
717      // Okay, everything within the region is now branching to the right block, we
718      // just have to update the PHI nodes now, inserting PHI nodes into NewBB.
719      BasicBlock::iterator AfterPHIs;
720      for (AfterPHIs = OldPred->begin(); isa<PHINode>(AfterPHIs); ++AfterPHIs) {
721        PHINode *PN = cast<PHINode>(AfterPHIs);
722        // Create a new PHI node in the new region, which has an incoming value
723        // from OldPred of PN.
724        PHINode *NewPN = PHINode::Create(PN->getType(), 1 + NumPredsFromRegion,
725                                         PN->getName() + ".ce");
726        NewPN->insertBefore(NewBB->begin());
727        PN->replaceAllUsesWith(NewPN);
728        NewPN->addIncoming(PN, OldPred);
729  
730        // Loop over all of the incoming value in PN, moving them to NewPN if they
731        // are from the extracted region.
732        for (unsigned i = 0; i != PN->getNumIncomingValues(); ++i) {
733          if (Blocks.count(PN->getIncomingBlock(i))) {
734            NewPN->addIncoming(PN->getIncomingValue(i), PN->getIncomingBlock(i));
735            PN->removeIncomingValue(i);
736            --i;
737          }
738        }
739      }
740    }
741  }
742  
743  /// severSplitPHINodesOfExits - if PHI nodes in exit blocks have inputs from
744  /// outlined region, we split these PHIs on two: one with inputs from region
745  /// and other with remaining incoming blocks; then first PHIs are placed in
746  /// outlined region.
747  void CodeExtractor::severSplitPHINodesOfExits(
748      const SmallPtrSetImpl<BasicBlock *> &Exits) {
749    for (BasicBlock *ExitBB : Exits) {
750      BasicBlock *NewBB = nullptr;
751  
752      for (PHINode &PN : ExitBB->phis()) {
753        // Find all incoming values from the outlining region.
754        SmallVector<unsigned, 2> IncomingVals;
755        for (unsigned i = 0; i < PN.getNumIncomingValues(); ++i)
756          if (Blocks.count(PN.getIncomingBlock(i)))
757            IncomingVals.push_back(i);
758  
759        // Do not process PHI if there is one (or fewer) predecessor from region.
760        // If PHI has exactly one predecessor from region, only this one incoming
761        // will be replaced on codeRepl block, so it should be safe to skip PHI.
762        if (IncomingVals.size() <= 1)
763          continue;
764  
765        // Create block for new PHIs and add it to the list of outlined if it
766        // wasn't done before.
767        if (!NewBB) {
768          NewBB = BasicBlock::Create(ExitBB->getContext(),
769                                     ExitBB->getName() + ".split",
770                                     ExitBB->getParent(), ExitBB);
771          NewBB->IsNewDbgInfoFormat = ExitBB->IsNewDbgInfoFormat;
772          SmallVector<BasicBlock *, 4> Preds(predecessors(ExitBB));
773          for (BasicBlock *PredBB : Preds)
774            if (Blocks.count(PredBB))
775              PredBB->getTerminator()->replaceUsesOfWith(ExitBB, NewBB);
776          BranchInst::Create(ExitBB, NewBB);
777          Blocks.insert(NewBB);
778        }
779  
780        // Split this PHI.
781        PHINode *NewPN = PHINode::Create(PN.getType(), IncomingVals.size(),
782                                         PN.getName() + ".ce");
783        NewPN->insertBefore(NewBB->getFirstNonPHIIt());
784        for (unsigned i : IncomingVals)
785          NewPN->addIncoming(PN.getIncomingValue(i), PN.getIncomingBlock(i));
786        for (unsigned i : reverse(IncomingVals))
787          PN.removeIncomingValue(i, false);
788        PN.addIncoming(NewPN, NewBB);
789      }
790    }
791  }
792  
793  void CodeExtractor::splitReturnBlocks() {
794    for (BasicBlock *Block : Blocks)
795      if (ReturnInst *RI = dyn_cast<ReturnInst>(Block->getTerminator())) {
796        BasicBlock *New =
797            Block->splitBasicBlock(RI->getIterator(), Block->getName() + ".ret");
798        if (DT) {
799          // Old dominates New. New node dominates all other nodes dominated
800          // by Old.
801          DomTreeNode *OldNode = DT->getNode(Block);
802          SmallVector<DomTreeNode *, 8> Children(OldNode->begin(),
803                                                 OldNode->end());
804  
805          DomTreeNode *NewNode = DT->addNewBlock(New, Block);
806  
807          for (DomTreeNode *I : Children)
808            DT->changeImmediateDominator(I, NewNode);
809        }
810      }
811  }
812  
813  /// constructFunction - make a function based on inputs and outputs, as follows:
814  /// f(in0, ..., inN, out0, ..., outN)
815  Function *CodeExtractor::constructFunction(const ValueSet &inputs,
816                                             const ValueSet &outputs,
817                                             BasicBlock *header,
818                                             BasicBlock *newRootNode,
819                                             BasicBlock *newHeader,
820                                             Function *oldFunction,
821                                             Module *M) {
822    LLVM_DEBUG(dbgs() << "inputs: " << inputs.size() << "\n");
823    LLVM_DEBUG(dbgs() << "outputs: " << outputs.size() << "\n");
824  
825    // This function returns unsigned, outputs will go back by reference.
826    switch (NumExitBlocks) {
827    case 0:
828    case 1: RetTy = Type::getVoidTy(header->getContext()); break;
829    case 2: RetTy = Type::getInt1Ty(header->getContext()); break;
830    default: RetTy = Type::getInt16Ty(header->getContext()); break;
831    }
832  
833    std::vector<Type *> ParamTy;
834    std::vector<Type *> AggParamTy;
835    ValueSet StructValues;
836    const DataLayout &DL = M->getDataLayout();
837  
838    // Add the types of the input values to the function's argument list
839    for (Value *value : inputs) {
840      LLVM_DEBUG(dbgs() << "value used in func: " << *value << "\n");
841      if (AggregateArgs && !ExcludeArgsFromAggregate.contains(value)) {
842        AggParamTy.push_back(value->getType());
843        StructValues.insert(value);
844      } else
845        ParamTy.push_back(value->getType());
846    }
847  
848    // Add the types of the output values to the function's argument list.
849    for (Value *output : outputs) {
850      LLVM_DEBUG(dbgs() << "instr used in func: " << *output << "\n");
851      if (AggregateArgs && !ExcludeArgsFromAggregate.contains(output)) {
852        AggParamTy.push_back(output->getType());
853        StructValues.insert(output);
854      } else
855        ParamTy.push_back(
856            PointerType::get(output->getType(), DL.getAllocaAddrSpace()));
857    }
858  
859    assert(
860        (ParamTy.size() + AggParamTy.size()) ==
861            (inputs.size() + outputs.size()) &&
862        "Number of scalar and aggregate params does not match inputs, outputs");
863    assert((StructValues.empty() || AggregateArgs) &&
864           "Expeced StructValues only with AggregateArgs set");
865  
866    // Concatenate scalar and aggregate params in ParamTy.
867    size_t NumScalarParams = ParamTy.size();
868    StructType *StructTy = nullptr;
869    if (AggregateArgs && !AggParamTy.empty()) {
870      StructTy = StructType::get(M->getContext(), AggParamTy);
871      ParamTy.push_back(PointerType::get(
872          StructTy, ArgsInZeroAddressSpace ? 0 : DL.getAllocaAddrSpace()));
873    }
874  
875    LLVM_DEBUG({
876      dbgs() << "Function type: " << *RetTy << " f(";
877      for (Type *i : ParamTy)
878        dbgs() << *i << ", ";
879      dbgs() << ")\n";
880    });
881  
882    FunctionType *funcType = FunctionType::get(
883        RetTy, ParamTy, AllowVarArgs && oldFunction->isVarArg());
884  
885    std::string SuffixToUse =
886        Suffix.empty()
887            ? (header->getName().empty() ? "extracted" : header->getName().str())
888            : Suffix;
889    // Create the new function
890    Function *newFunction = Function::Create(
891        funcType, GlobalValue::InternalLinkage, oldFunction->getAddressSpace(),
892        oldFunction->getName() + "." + SuffixToUse, M);
893    newFunction->IsNewDbgInfoFormat = oldFunction->IsNewDbgInfoFormat;
894  
895    // Inherit all of the target dependent attributes and white-listed
896    // target independent attributes.
897    //  (e.g. If the extracted region contains a call to an x86.sse
898    //  instruction we need to make sure that the extracted region has the
899    //  "target-features" attribute allowing it to be lowered.
900    // FIXME: This should be changed to check to see if a specific
901    //           attribute can not be inherited.
902    for (const auto &Attr : oldFunction->getAttributes().getFnAttrs()) {
903      if (Attr.isStringAttribute()) {
904        if (Attr.getKindAsString() == "thunk")
905          continue;
906      } else
907        switch (Attr.getKindAsEnum()) {
908        // Those attributes cannot be propagated safely. Explicitly list them
909        // here so we get a warning if new attributes are added.
910        case Attribute::AllocSize:
911        case Attribute::Builtin:
912        case Attribute::Convergent:
913        case Attribute::JumpTable:
914        case Attribute::Naked:
915        case Attribute::NoBuiltin:
916        case Attribute::NoMerge:
917        case Attribute::NoReturn:
918        case Attribute::NoSync:
919        case Attribute::ReturnsTwice:
920        case Attribute::Speculatable:
921        case Attribute::StackAlignment:
922        case Attribute::WillReturn:
923        case Attribute::AllocKind:
924        case Attribute::PresplitCoroutine:
925        case Attribute::Memory:
926        case Attribute::NoFPClass:
927        case Attribute::CoroDestroyOnlyWhenComplete:
928          continue;
929        // Those attributes should be safe to propagate to the extracted function.
930        case Attribute::AlwaysInline:
931        case Attribute::Cold:
932        case Attribute::DisableSanitizerInstrumentation:
933        case Attribute::FnRetThunkExtern:
934        case Attribute::Hot:
935        case Attribute::NoRecurse:
936        case Attribute::InlineHint:
937        case Attribute::MinSize:
938        case Attribute::NoCallback:
939        case Attribute::NoDuplicate:
940        case Attribute::NoFree:
941        case Attribute::NoImplicitFloat:
942        case Attribute::NoInline:
943        case Attribute::NonLazyBind:
944        case Attribute::NoRedZone:
945        case Attribute::NoUnwind:
946        case Attribute::NoSanitizeBounds:
947        case Attribute::NoSanitizeCoverage:
948        case Attribute::NullPointerIsValid:
949        case Attribute::OptimizeForDebugging:
950        case Attribute::OptForFuzzing:
951        case Attribute::OptimizeNone:
952        case Attribute::OptimizeForSize:
953        case Attribute::SafeStack:
954        case Attribute::ShadowCallStack:
955        case Attribute::SanitizeAddress:
956        case Attribute::SanitizeMemory:
957        case Attribute::SanitizeThread:
958        case Attribute::SanitizeHWAddress:
959        case Attribute::SanitizeMemTag:
960        case Attribute::SpeculativeLoadHardening:
961        case Attribute::StackProtect:
962        case Attribute::StackProtectReq:
963        case Attribute::StackProtectStrong:
964        case Attribute::StrictFP:
965        case Attribute::UWTable:
966        case Attribute::VScaleRange:
967        case Attribute::NoCfCheck:
968        case Attribute::MustProgress:
969        case Attribute::NoProfile:
970        case Attribute::SkipProfile:
971          break;
972        // These attributes cannot be applied to functions.
973        case Attribute::Alignment:
974        case Attribute::AllocatedPointer:
975        case Attribute::AllocAlign:
976        case Attribute::ByVal:
977        case Attribute::Dereferenceable:
978        case Attribute::DereferenceableOrNull:
979        case Attribute::ElementType:
980        case Attribute::InAlloca:
981        case Attribute::InReg:
982        case Attribute::Nest:
983        case Attribute::NoAlias:
984        case Attribute::NoCapture:
985        case Attribute::NoUndef:
986        case Attribute::NonNull:
987        case Attribute::Preallocated:
988        case Attribute::ReadNone:
989        case Attribute::ReadOnly:
990        case Attribute::Returned:
991        case Attribute::SExt:
992        case Attribute::StructRet:
993        case Attribute::SwiftError:
994        case Attribute::SwiftSelf:
995        case Attribute::SwiftAsync:
996        case Attribute::ZExt:
997        case Attribute::ImmArg:
998        case Attribute::ByRef:
999        case Attribute::WriteOnly:
1000        case Attribute::Writable:
1001        case Attribute::DeadOnUnwind:
1002        //  These are not really attributes.
1003        case Attribute::None:
1004        case Attribute::EndAttrKinds:
1005        case Attribute::EmptyKey:
1006        case Attribute::TombstoneKey:
1007          llvm_unreachable("Not a function attribute");
1008        }
1009  
1010      newFunction->addFnAttr(Attr);
1011    }
1012    newFunction->insert(newFunction->end(), newRootNode);
1013  
1014    // Create scalar and aggregate iterators to name all of the arguments we
1015    // inserted.
1016    Function::arg_iterator ScalarAI = newFunction->arg_begin();
1017    Function::arg_iterator AggAI = std::next(ScalarAI, NumScalarParams);
1018  
1019    // Rewrite all users of the inputs in the extracted region to use the
1020    // arguments (or appropriate addressing into struct) instead.
1021    for (unsigned i = 0, e = inputs.size(), aggIdx = 0; i != e; ++i) {
1022      Value *RewriteVal;
1023      if (AggregateArgs && StructValues.contains(inputs[i])) {
1024        Value *Idx[2];
1025        Idx[0] = Constant::getNullValue(Type::getInt32Ty(header->getContext()));
1026        Idx[1] = ConstantInt::get(Type::getInt32Ty(header->getContext()), aggIdx);
1027        Instruction *TI = newFunction->begin()->getTerminator();
1028        GetElementPtrInst *GEP = GetElementPtrInst::Create(
1029            StructTy, &*AggAI, Idx, "gep_" + inputs[i]->getName(), TI);
1030        RewriteVal = new LoadInst(StructTy->getElementType(aggIdx), GEP,
1031                                  "loadgep_" + inputs[i]->getName(), TI);
1032        ++aggIdx;
1033      } else
1034        RewriteVal = &*ScalarAI++;
1035  
1036      std::vector<User *> Users(inputs[i]->user_begin(), inputs[i]->user_end());
1037      for (User *use : Users)
1038        if (Instruction *inst = dyn_cast<Instruction>(use))
1039          if (Blocks.count(inst->getParent()))
1040            inst->replaceUsesOfWith(inputs[i], RewriteVal);
1041    }
1042  
1043    // Set names for input and output arguments.
1044    if (NumScalarParams) {
1045      ScalarAI = newFunction->arg_begin();
1046      for (unsigned i = 0, e = inputs.size(); i != e; ++i, ++ScalarAI)
1047        if (!StructValues.contains(inputs[i]))
1048          ScalarAI->setName(inputs[i]->getName());
1049      for (unsigned i = 0, e = outputs.size(); i != e; ++i, ++ScalarAI)
1050        if (!StructValues.contains(outputs[i]))
1051          ScalarAI->setName(outputs[i]->getName() + ".out");
1052    }
1053  
1054    // Rewrite branches to basic blocks outside of the loop to new dummy blocks
1055    // within the new function. This must be done before we lose track of which
1056    // blocks were originally in the code region.
1057    std::vector<User *> Users(header->user_begin(), header->user_end());
1058    for (auto &U : Users)
1059      // The BasicBlock which contains the branch is not in the region
1060      // modify the branch target to a new block
1061      if (Instruction *I = dyn_cast<Instruction>(U))
1062        if (I->isTerminator() && I->getFunction() == oldFunction &&
1063            !Blocks.count(I->getParent()))
1064          I->replaceUsesOfWith(header, newHeader);
1065  
1066    return newFunction;
1067  }
1068  
1069  /// Erase lifetime.start markers which reference inputs to the extraction
1070  /// region, and insert the referenced memory into \p LifetimesStart.
1071  ///
1072  /// The extraction region is defined by a set of blocks (\p Blocks), and a set
1073  /// of allocas which will be moved from the caller function into the extracted
1074  /// function (\p SunkAllocas).
1075  static void eraseLifetimeMarkersOnInputs(const SetVector<BasicBlock *> &Blocks,
1076                                           const SetVector<Value *> &SunkAllocas,
1077                                           SetVector<Value *> &LifetimesStart) {
1078    for (BasicBlock *BB : Blocks) {
1079      for (Instruction &I : llvm::make_early_inc_range(*BB)) {
1080        auto *II = dyn_cast<IntrinsicInst>(&I);
1081        if (!II || !II->isLifetimeStartOrEnd())
1082          continue;
1083  
1084        // Get the memory operand of the lifetime marker. If the underlying
1085        // object is a sunk alloca, or is otherwise defined in the extraction
1086        // region, the lifetime marker must not be erased.
1087        Value *Mem = II->getOperand(1)->stripInBoundsOffsets();
1088        if (SunkAllocas.count(Mem) || definedInRegion(Blocks, Mem))
1089          continue;
1090  
1091        if (II->getIntrinsicID() == Intrinsic::lifetime_start)
1092          LifetimesStart.insert(Mem);
1093        II->eraseFromParent();
1094      }
1095    }
1096  }
1097  
1098  /// Insert lifetime start/end markers surrounding the call to the new function
1099  /// for objects defined in the caller.
1100  static void insertLifetimeMarkersSurroundingCall(
1101      Module *M, ArrayRef<Value *> LifetimesStart, ArrayRef<Value *> LifetimesEnd,
1102      CallInst *TheCall) {
1103    LLVMContext &Ctx = M->getContext();
1104    auto NegativeOne = ConstantInt::getSigned(Type::getInt64Ty(Ctx), -1);
1105    Instruction *Term = TheCall->getParent()->getTerminator();
1106  
1107    // Emit lifetime markers for the pointers given in \p Objects. Insert the
1108    // markers before the call if \p InsertBefore, and after the call otherwise.
1109    auto insertMarkers = [&](Intrinsic::ID MarkerFunc, ArrayRef<Value *> Objects,
1110                             bool InsertBefore) {
1111      for (Value *Mem : Objects) {
1112        assert((!isa<Instruction>(Mem) || cast<Instruction>(Mem)->getFunction() ==
1113                                              TheCall->getFunction()) &&
1114               "Input memory not defined in original function");
1115  
1116        Function *Func = Intrinsic::getDeclaration(M, MarkerFunc, Mem->getType());
1117        auto Marker = CallInst::Create(Func, {NegativeOne, Mem});
1118        if (InsertBefore)
1119          Marker->insertBefore(TheCall);
1120        else
1121          Marker->insertBefore(Term);
1122      }
1123    };
1124  
1125    if (!LifetimesStart.empty()) {
1126      insertMarkers(Intrinsic::lifetime_start, LifetimesStart,
1127                    /*InsertBefore=*/true);
1128    }
1129  
1130    if (!LifetimesEnd.empty()) {
1131      insertMarkers(Intrinsic::lifetime_end, LifetimesEnd,
1132                    /*InsertBefore=*/false);
1133    }
1134  }
1135  
1136  /// emitCallAndSwitchStatement - This method sets up the caller side by adding
1137  /// the call instruction, splitting any PHI nodes in the header block as
1138  /// necessary.
1139  CallInst *CodeExtractor::emitCallAndSwitchStatement(Function *newFunction,
1140                                                      BasicBlock *codeReplacer,
1141                                                      ValueSet &inputs,
1142                                                      ValueSet &outputs) {
1143    // Emit a call to the new function, passing in: *pointer to struct (if
1144    // aggregating parameters), or plan inputs and allocated memory for outputs
1145    std::vector<Value *> params, ReloadOutputs, Reloads;
1146    ValueSet StructValues;
1147  
1148    Module *M = newFunction->getParent();
1149    LLVMContext &Context = M->getContext();
1150    const DataLayout &DL = M->getDataLayout();
1151    CallInst *call = nullptr;
1152  
1153    // Add inputs as params, or to be filled into the struct
1154    unsigned ScalarInputArgNo = 0;
1155    SmallVector<unsigned, 1> SwiftErrorArgs;
1156    for (Value *input : inputs) {
1157      if (AggregateArgs && !ExcludeArgsFromAggregate.contains(input))
1158        StructValues.insert(input);
1159      else {
1160        params.push_back(input);
1161        if (input->isSwiftError())
1162          SwiftErrorArgs.push_back(ScalarInputArgNo);
1163      }
1164      ++ScalarInputArgNo;
1165    }
1166  
1167    // Create allocas for the outputs
1168    unsigned ScalarOutputArgNo = 0;
1169    for (Value *output : outputs) {
1170      if (AggregateArgs && !ExcludeArgsFromAggregate.contains(output)) {
1171        StructValues.insert(output);
1172      } else {
1173        AllocaInst *alloca =
1174          new AllocaInst(output->getType(), DL.getAllocaAddrSpace(),
1175                         nullptr, output->getName() + ".loc",
1176                         &codeReplacer->getParent()->front().front());
1177        ReloadOutputs.push_back(alloca);
1178        params.push_back(alloca);
1179        ++ScalarOutputArgNo;
1180      }
1181    }
1182  
1183    StructType *StructArgTy = nullptr;
1184    AllocaInst *Struct = nullptr;
1185    unsigned NumAggregatedInputs = 0;
1186    if (AggregateArgs && !StructValues.empty()) {
1187      std::vector<Type *> ArgTypes;
1188      for (Value *V : StructValues)
1189        ArgTypes.push_back(V->getType());
1190  
1191      // Allocate a struct at the beginning of this function
1192      StructArgTy = StructType::get(newFunction->getContext(), ArgTypes);
1193      Struct = new AllocaInst(
1194          StructArgTy, DL.getAllocaAddrSpace(), nullptr, "structArg",
1195          AllocationBlock ? &*AllocationBlock->getFirstInsertionPt()
1196                          : &codeReplacer->getParent()->front().front());
1197  
1198      if (ArgsInZeroAddressSpace && DL.getAllocaAddrSpace() != 0) {
1199        auto *StructSpaceCast = new AddrSpaceCastInst(
1200            Struct, PointerType ::get(Context, 0), "structArg.ascast");
1201        StructSpaceCast->insertAfter(Struct);
1202        params.push_back(StructSpaceCast);
1203      } else {
1204        params.push_back(Struct);
1205      }
1206      // Store aggregated inputs in the struct.
1207      for (unsigned i = 0, e = StructValues.size(); i != e; ++i) {
1208        if (inputs.contains(StructValues[i])) {
1209          Value *Idx[2];
1210          Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context));
1211          Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), i);
1212          GetElementPtrInst *GEP = GetElementPtrInst::Create(
1213              StructArgTy, Struct, Idx, "gep_" + StructValues[i]->getName());
1214          GEP->insertInto(codeReplacer, codeReplacer->end());
1215          new StoreInst(StructValues[i], GEP, codeReplacer);
1216          NumAggregatedInputs++;
1217        }
1218      }
1219    }
1220  
1221    // Emit the call to the function
1222    call = CallInst::Create(newFunction, params,
1223                            NumExitBlocks > 1 ? "targetBlock" : "");
1224    // Add debug location to the new call, if the original function has debug
1225    // info. In that case, the terminator of the entry block of the extracted
1226    // function contains the first debug location of the extracted function,
1227    // set in extractCodeRegion.
1228    if (codeReplacer->getParent()->getSubprogram()) {
1229      if (auto DL = newFunction->getEntryBlock().getTerminator()->getDebugLoc())
1230        call->setDebugLoc(DL);
1231    }
1232    call->insertInto(codeReplacer, codeReplacer->end());
1233  
1234    // Set swifterror parameter attributes.
1235    for (unsigned SwiftErrArgNo : SwiftErrorArgs) {
1236      call->addParamAttr(SwiftErrArgNo, Attribute::SwiftError);
1237      newFunction->addParamAttr(SwiftErrArgNo, Attribute::SwiftError);
1238    }
1239  
1240    // Reload the outputs passed in by reference, use the struct if output is in
1241    // the aggregate or reload from the scalar argument.
1242    for (unsigned i = 0, e = outputs.size(), scalarIdx = 0,
1243                  aggIdx = NumAggregatedInputs;
1244         i != e; ++i) {
1245      Value *Output = nullptr;
1246      if (AggregateArgs && StructValues.contains(outputs[i])) {
1247        Value *Idx[2];
1248        Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context));
1249        Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), aggIdx);
1250        GetElementPtrInst *GEP = GetElementPtrInst::Create(
1251            StructArgTy, Struct, Idx, "gep_reload_" + outputs[i]->getName());
1252        GEP->insertInto(codeReplacer, codeReplacer->end());
1253        Output = GEP;
1254        ++aggIdx;
1255      } else {
1256        Output = ReloadOutputs[scalarIdx];
1257        ++scalarIdx;
1258      }
1259      LoadInst *load = new LoadInst(outputs[i]->getType(), Output,
1260                                    outputs[i]->getName() + ".reload",
1261                                    codeReplacer);
1262      Reloads.push_back(load);
1263      std::vector<User *> Users(outputs[i]->user_begin(), outputs[i]->user_end());
1264      for (User *U : Users) {
1265        Instruction *inst = cast<Instruction>(U);
1266        if (!Blocks.count(inst->getParent()))
1267          inst->replaceUsesOfWith(outputs[i], load);
1268      }
1269    }
1270  
1271    // Now we can emit a switch statement using the call as a value.
1272    SwitchInst *TheSwitch =
1273        SwitchInst::Create(Constant::getNullValue(Type::getInt16Ty(Context)),
1274                           codeReplacer, 0, codeReplacer);
1275  
1276    // Since there may be multiple exits from the original region, make the new
1277    // function return an unsigned, switch on that number.  This loop iterates
1278    // over all of the blocks in the extracted region, updating any terminator
1279    // instructions in the to-be-extracted region that branch to blocks that are
1280    // not in the region to be extracted.
1281    std::map<BasicBlock *, BasicBlock *> ExitBlockMap;
1282  
1283    // Iterate over the previously collected targets, and create new blocks inside
1284    // the function to branch to.
1285    unsigned switchVal = 0;
1286    for (BasicBlock *OldTarget : OldTargets) {
1287      if (Blocks.count(OldTarget))
1288        continue;
1289      BasicBlock *&NewTarget = ExitBlockMap[OldTarget];
1290      if (NewTarget)
1291        continue;
1292  
1293      // If we don't already have an exit stub for this non-extracted
1294      // destination, create one now!
1295      NewTarget = BasicBlock::Create(Context,
1296                                      OldTarget->getName() + ".exitStub",
1297                                      newFunction);
1298      unsigned SuccNum = switchVal++;
1299  
1300      Value *brVal = nullptr;
1301      assert(NumExitBlocks < 0xffff && "too many exit blocks for switch");
1302      switch (NumExitBlocks) {
1303      case 0:
1304      case 1: break;  // No value needed.
1305      case 2:         // Conditional branch, return a bool
1306        brVal = ConstantInt::get(Type::getInt1Ty(Context), !SuccNum);
1307        break;
1308      default:
1309        brVal = ConstantInt::get(Type::getInt16Ty(Context), SuccNum);
1310        break;
1311      }
1312  
1313      ReturnInst::Create(Context, brVal, NewTarget);
1314  
1315      // Update the switch instruction.
1316      TheSwitch->addCase(ConstantInt::get(Type::getInt16Ty(Context),
1317                                          SuccNum),
1318                          OldTarget);
1319    }
1320  
1321    for (BasicBlock *Block : Blocks) {
1322      Instruction *TI = Block->getTerminator();
1323      for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i) {
1324        if (Blocks.count(TI->getSuccessor(i)))
1325          continue;
1326        BasicBlock *OldTarget = TI->getSuccessor(i);
1327        // add a new basic block which returns the appropriate value
1328        BasicBlock *NewTarget = ExitBlockMap[OldTarget];
1329        assert(NewTarget && "Unknown target block!");
1330  
1331        // rewrite the original branch instruction with this new target
1332        TI->setSuccessor(i, NewTarget);
1333     }
1334    }
1335  
1336    // Store the arguments right after the definition of output value.
1337    // This should be proceeded after creating exit stubs to be ensure that invoke
1338    // result restore will be placed in the outlined function.
1339    Function::arg_iterator ScalarOutputArgBegin = newFunction->arg_begin();
1340    std::advance(ScalarOutputArgBegin, ScalarInputArgNo);
1341    Function::arg_iterator AggOutputArgBegin = newFunction->arg_begin();
1342    std::advance(AggOutputArgBegin, ScalarInputArgNo + ScalarOutputArgNo);
1343  
1344    for (unsigned i = 0, e = outputs.size(), aggIdx = NumAggregatedInputs; i != e;
1345         ++i) {
1346      auto *OutI = dyn_cast<Instruction>(outputs[i]);
1347      if (!OutI)
1348        continue;
1349  
1350      // Find proper insertion point.
1351      BasicBlock::iterator InsertPt;
1352      // In case OutI is an invoke, we insert the store at the beginning in the
1353      // 'normal destination' BB. Otherwise we insert the store right after OutI.
1354      if (auto *InvokeI = dyn_cast<InvokeInst>(OutI))
1355        InsertPt = InvokeI->getNormalDest()->getFirstInsertionPt();
1356      else if (auto *Phi = dyn_cast<PHINode>(OutI))
1357        InsertPt = Phi->getParent()->getFirstInsertionPt();
1358      else
1359        InsertPt = std::next(OutI->getIterator());
1360  
1361      Instruction *InsertBefore = &*InsertPt;
1362      assert((InsertBefore->getFunction() == newFunction ||
1363              Blocks.count(InsertBefore->getParent())) &&
1364             "InsertPt should be in new function");
1365      if (AggregateArgs && StructValues.contains(outputs[i])) {
1366        assert(AggOutputArgBegin != newFunction->arg_end() &&
1367               "Number of aggregate output arguments should match "
1368               "the number of defined values");
1369        Value *Idx[2];
1370        Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context));
1371        Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), aggIdx);
1372        GetElementPtrInst *GEP = GetElementPtrInst::Create(
1373            StructArgTy, &*AggOutputArgBegin, Idx, "gep_" + outputs[i]->getName(),
1374            InsertBefore);
1375        new StoreInst(outputs[i], GEP, InsertBefore);
1376        ++aggIdx;
1377        // Since there should be only one struct argument aggregating
1378        // all the output values, we shouldn't increment AggOutputArgBegin, which
1379        // always points to the struct argument, in this case.
1380      } else {
1381        assert(ScalarOutputArgBegin != newFunction->arg_end() &&
1382               "Number of scalar output arguments should match "
1383               "the number of defined values");
1384        new StoreInst(outputs[i], &*ScalarOutputArgBegin, InsertBefore);
1385        ++ScalarOutputArgBegin;
1386      }
1387    }
1388  
1389    // Now that we've done the deed, simplify the switch instruction.
1390    Type *OldFnRetTy = TheSwitch->getParent()->getParent()->getReturnType();
1391    switch (NumExitBlocks) {
1392    case 0:
1393      // There are no successors (the block containing the switch itself), which
1394      // means that previously this was the last part of the function, and hence
1395      // this should be rewritten as a `ret'
1396  
1397      // Check if the function should return a value
1398      if (OldFnRetTy->isVoidTy()) {
1399        ReturnInst::Create(Context, nullptr, TheSwitch);  // Return void
1400      } else if (OldFnRetTy == TheSwitch->getCondition()->getType()) {
1401        // return what we have
1402        ReturnInst::Create(Context, TheSwitch->getCondition(), TheSwitch);
1403      } else {
1404        // Otherwise we must have code extracted an unwind or something, just
1405        // return whatever we want.
1406        ReturnInst::Create(Context,
1407                           Constant::getNullValue(OldFnRetTy), TheSwitch);
1408      }
1409  
1410      TheSwitch->eraseFromParent();
1411      break;
1412    case 1:
1413      // Only a single destination, change the switch into an unconditional
1414      // branch.
1415      BranchInst::Create(TheSwitch->getSuccessor(1), TheSwitch);
1416      TheSwitch->eraseFromParent();
1417      break;
1418    case 2:
1419      BranchInst::Create(TheSwitch->getSuccessor(1), TheSwitch->getSuccessor(2),
1420                         call, TheSwitch);
1421      TheSwitch->eraseFromParent();
1422      break;
1423    default:
1424      // Otherwise, make the default destination of the switch instruction be one
1425      // of the other successors.
1426      TheSwitch->setCondition(call);
1427      TheSwitch->setDefaultDest(TheSwitch->getSuccessor(NumExitBlocks));
1428      // Remove redundant case
1429      TheSwitch->removeCase(SwitchInst::CaseIt(TheSwitch, NumExitBlocks-1));
1430      break;
1431    }
1432  
1433    // Insert lifetime markers around the reloads of any output values. The
1434    // allocas output values are stored in are only in-use in the codeRepl block.
1435    insertLifetimeMarkersSurroundingCall(M, ReloadOutputs, ReloadOutputs, call);
1436  
1437    return call;
1438  }
1439  
1440  void CodeExtractor::moveCodeToFunction(Function *newFunction) {
1441    auto newFuncIt = newFunction->front().getIterator();
1442    for (BasicBlock *Block : Blocks) {
1443      // Delete the basic block from the old function, and the list of blocks
1444      Block->removeFromParent();
1445  
1446      // Insert this basic block into the new function
1447      // Insert the original blocks after the entry block created
1448      // for the new function. The entry block may be followed
1449      // by a set of exit blocks at this point, but these exit
1450      // blocks better be placed at the end of the new function.
1451      newFuncIt = newFunction->insert(std::next(newFuncIt), Block);
1452    }
1453  }
1454  
1455  void CodeExtractor::calculateNewCallTerminatorWeights(
1456      BasicBlock *CodeReplacer,
1457      DenseMap<BasicBlock *, BlockFrequency> &ExitWeights,
1458      BranchProbabilityInfo *BPI) {
1459    using Distribution = BlockFrequencyInfoImplBase::Distribution;
1460    using BlockNode = BlockFrequencyInfoImplBase::BlockNode;
1461  
1462    // Update the branch weights for the exit block.
1463    Instruction *TI = CodeReplacer->getTerminator();
1464    SmallVector<unsigned, 8> BranchWeights(TI->getNumSuccessors(), 0);
1465  
1466    // Block Frequency distribution with dummy node.
1467    Distribution BranchDist;
1468  
1469    SmallVector<BranchProbability, 4> EdgeProbabilities(
1470        TI->getNumSuccessors(), BranchProbability::getUnknown());
1471  
1472    // Add each of the frequencies of the successors.
1473    for (unsigned i = 0, e = TI->getNumSuccessors(); i < e; ++i) {
1474      BlockNode ExitNode(i);
1475      uint64_t ExitFreq = ExitWeights[TI->getSuccessor(i)].getFrequency();
1476      if (ExitFreq != 0)
1477        BranchDist.addExit(ExitNode, ExitFreq);
1478      else
1479        EdgeProbabilities[i] = BranchProbability::getZero();
1480    }
1481  
1482    // Check for no total weight.
1483    if (BranchDist.Total == 0) {
1484      BPI->setEdgeProbability(CodeReplacer, EdgeProbabilities);
1485      return;
1486    }
1487  
1488    // Normalize the distribution so that they can fit in unsigned.
1489    BranchDist.normalize();
1490  
1491    // Create normalized branch weights and set the metadata.
1492    for (unsigned I = 0, E = BranchDist.Weights.size(); I < E; ++I) {
1493      const auto &Weight = BranchDist.Weights[I];
1494  
1495      // Get the weight and update the current BFI.
1496      BranchWeights[Weight.TargetNode.Index] = Weight.Amount;
1497      BranchProbability BP(Weight.Amount, BranchDist.Total);
1498      EdgeProbabilities[Weight.TargetNode.Index] = BP;
1499    }
1500    BPI->setEdgeProbability(CodeReplacer, EdgeProbabilities);
1501    TI->setMetadata(
1502        LLVMContext::MD_prof,
1503        MDBuilder(TI->getContext()).createBranchWeights(BranchWeights));
1504  }
1505  
1506  /// Erase debug info intrinsics which refer to values in \p F but aren't in
1507  /// \p F.
1508  static void eraseDebugIntrinsicsWithNonLocalRefs(Function &F) {
1509    for (Instruction &I : instructions(F)) {
1510      SmallVector<DbgVariableIntrinsic *, 4> DbgUsers;
1511      SmallVector<DPValue *, 4> DPValues;
1512      findDbgUsers(DbgUsers, &I, &DPValues);
1513      for (DbgVariableIntrinsic *DVI : DbgUsers)
1514        if (DVI->getFunction() != &F)
1515          DVI->eraseFromParent();
1516      for (DPValue *DPV : DPValues)
1517        if (DPV->getFunction() != &F)
1518          DPV->eraseFromParent();
1519    }
1520  }
1521  
1522  /// Fix up the debug info in the old and new functions by pointing line
1523  /// locations and debug intrinsics to the new subprogram scope, and by deleting
1524  /// intrinsics which point to values outside of the new function.
1525  static void fixupDebugInfoPostExtraction(Function &OldFunc, Function &NewFunc,
1526                                           CallInst &TheCall) {
1527    DISubprogram *OldSP = OldFunc.getSubprogram();
1528    LLVMContext &Ctx = OldFunc.getContext();
1529  
1530    if (!OldSP) {
1531      // Erase any debug info the new function contains.
1532      stripDebugInfo(NewFunc);
1533      // Make sure the old function doesn't contain any non-local metadata refs.
1534      eraseDebugIntrinsicsWithNonLocalRefs(NewFunc);
1535      return;
1536    }
1537  
1538    // Create a subprogram for the new function. Leave out a description of the
1539    // function arguments, as the parameters don't correspond to anything at the
1540    // source level.
1541    assert(OldSP->getUnit() && "Missing compile unit for subprogram");
1542    DIBuilder DIB(*OldFunc.getParent(), /*AllowUnresolved=*/false,
1543                  OldSP->getUnit());
1544    auto SPType =
1545        DIB.createSubroutineType(DIB.getOrCreateTypeArray(std::nullopt));
1546    DISubprogram::DISPFlags SPFlags = DISubprogram::SPFlagDefinition |
1547                                      DISubprogram::SPFlagOptimized |
1548                                      DISubprogram::SPFlagLocalToUnit;
1549    auto NewSP = DIB.createFunction(
1550        OldSP->getUnit(), NewFunc.getName(), NewFunc.getName(), OldSP->getFile(),
1551        /*LineNo=*/0, SPType, /*ScopeLine=*/0, DINode::FlagZero, SPFlags);
1552    NewFunc.setSubprogram(NewSP);
1553  
1554    auto IsInvalidLocation = [&NewFunc](Value *Location) {
1555      // Location is invalid if it isn't a constant or an instruction, or is an
1556      // instruction but isn't in the new function.
1557      if (!Location ||
1558          (!isa<Constant>(Location) && !isa<Instruction>(Location)))
1559        return true;
1560      Instruction *LocationInst = dyn_cast<Instruction>(Location);
1561      return LocationInst && LocationInst->getFunction() != &NewFunc;
1562    };
1563  
1564    // Debug intrinsics in the new function need to be updated in one of two
1565    // ways:
1566    //  1) They need to be deleted, because they describe a value in the old
1567    //     function.
1568    //  2) They need to point to fresh metadata, e.g. because they currently
1569    //     point to a variable in the wrong scope.
1570    SmallDenseMap<DINode *, DINode *> RemappedMetadata;
1571    SmallVector<Instruction *, 4> DebugIntrinsicsToDelete;
1572    SmallVector<DPValue *, 4> DPVsToDelete;
1573    DenseMap<const MDNode *, MDNode *> Cache;
1574  
1575    auto GetUpdatedDIVariable = [&](DILocalVariable *OldVar) {
1576      DINode *&NewVar = RemappedMetadata[OldVar];
1577      if (!NewVar) {
1578        DILocalScope *NewScope = DILocalScope::cloneScopeForSubprogram(
1579            *OldVar->getScope(), *NewSP, Ctx, Cache);
1580        NewVar = DIB.createAutoVariable(
1581            NewScope, OldVar->getName(), OldVar->getFile(), OldVar->getLine(),
1582            OldVar->getType(), /*AlwaysPreserve=*/false, DINode::FlagZero,
1583            OldVar->getAlignInBits());
1584      }
1585      return cast<DILocalVariable>(NewVar);
1586    };
1587  
1588    auto UpdateDPValuesOnInst = [&](Instruction &I) -> void {
1589      for (auto &DPV : I.getDbgValueRange()) {
1590        // Apply the two updates that dbg.values get: invalid operands, and
1591        // variable metadata fixup.
1592        if (any_of(DPV.location_ops(), IsInvalidLocation)) {
1593          DPVsToDelete.push_back(&DPV);
1594          continue;
1595        }
1596        if (DPV.isDbgAssign() && IsInvalidLocation(DPV.getAddress())) {
1597          DPVsToDelete.push_back(&DPV);
1598          continue;
1599        }
1600        if (!DPV.getDebugLoc().getInlinedAt())
1601          DPV.setVariable(GetUpdatedDIVariable(DPV.getVariable()));
1602        DPV.setDebugLoc(DebugLoc::replaceInlinedAtSubprogram(DPV.getDebugLoc(),
1603                                                             *NewSP, Ctx, Cache));
1604      }
1605    };
1606  
1607    for (Instruction &I : instructions(NewFunc)) {
1608      UpdateDPValuesOnInst(I);
1609  
1610      auto *DII = dyn_cast<DbgInfoIntrinsic>(&I);
1611      if (!DII)
1612        continue;
1613  
1614      // Point the intrinsic to a fresh label within the new function if the
1615      // intrinsic was not inlined from some other function.
1616      if (auto *DLI = dyn_cast<DbgLabelInst>(&I)) {
1617        if (DLI->getDebugLoc().getInlinedAt())
1618          continue;
1619        DILabel *OldLabel = DLI->getLabel();
1620        DINode *&NewLabel = RemappedMetadata[OldLabel];
1621        if (!NewLabel) {
1622          DILocalScope *NewScope = DILocalScope::cloneScopeForSubprogram(
1623              *OldLabel->getScope(), *NewSP, Ctx, Cache);
1624          NewLabel = DILabel::get(Ctx, NewScope, OldLabel->getName(),
1625                                  OldLabel->getFile(), OldLabel->getLine());
1626        }
1627        DLI->setArgOperand(0, MetadataAsValue::get(Ctx, NewLabel));
1628        continue;
1629      }
1630  
1631      auto *DVI = cast<DbgVariableIntrinsic>(DII);
1632      // If any of the used locations are invalid, delete the intrinsic.
1633      if (any_of(DVI->location_ops(), IsInvalidLocation)) {
1634        DebugIntrinsicsToDelete.push_back(DVI);
1635        continue;
1636      }
1637      // DbgAssign intrinsics have an extra Value argument:
1638      if (auto *DAI = dyn_cast<DbgAssignIntrinsic>(DVI);
1639          DAI && IsInvalidLocation(DAI->getAddress())) {
1640        DebugIntrinsicsToDelete.push_back(DVI);
1641        continue;
1642      }
1643      // If the variable was in the scope of the old function, i.e. it was not
1644      // inlined, point the intrinsic to a fresh variable within the new function.
1645      if (!DVI->getDebugLoc().getInlinedAt())
1646        DVI->setVariable(GetUpdatedDIVariable(DVI->getVariable()));
1647    }
1648  
1649    for (auto *DII : DebugIntrinsicsToDelete)
1650      DII->eraseFromParent();
1651    for (auto *DPV : DPVsToDelete)
1652      DPV->getMarker()->MarkedInstr->dropOneDbgValue(DPV);
1653    DIB.finalizeSubprogram(NewSP);
1654  
1655    // Fix up the scope information attached to the line locations in the new
1656    // function.
1657    for (Instruction &I : instructions(NewFunc)) {
1658      if (const DebugLoc &DL = I.getDebugLoc())
1659        I.setDebugLoc(
1660            DebugLoc::replaceInlinedAtSubprogram(DL, *NewSP, Ctx, Cache));
1661  
1662      // Loop info metadata may contain line locations. Fix them up.
1663      auto updateLoopInfoLoc = [&Ctx, &Cache, NewSP](Metadata *MD) -> Metadata * {
1664        if (auto *Loc = dyn_cast_or_null<DILocation>(MD))
1665          return DebugLoc::replaceInlinedAtSubprogram(Loc, *NewSP, Ctx, Cache);
1666        return MD;
1667      };
1668      updateLoopMetadataDebugLocations(I, updateLoopInfoLoc);
1669    }
1670    if (!TheCall.getDebugLoc())
1671      TheCall.setDebugLoc(DILocation::get(Ctx, 0, 0, OldSP));
1672  
1673    eraseDebugIntrinsicsWithNonLocalRefs(NewFunc);
1674  }
1675  
1676  Function *
1677  CodeExtractor::extractCodeRegion(const CodeExtractorAnalysisCache &CEAC) {
1678    ValueSet Inputs, Outputs;
1679    return extractCodeRegion(CEAC, Inputs, Outputs);
1680  }
1681  
1682  Function *
1683  CodeExtractor::extractCodeRegion(const CodeExtractorAnalysisCache &CEAC,
1684                                   ValueSet &inputs, ValueSet &outputs) {
1685    if (!isEligible())
1686      return nullptr;
1687  
1688    // Assumption: this is a single-entry code region, and the header is the first
1689    // block in the region.
1690    BasicBlock *header = *Blocks.begin();
1691    Function *oldFunction = header->getParent();
1692  
1693    // Calculate the entry frequency of the new function before we change the root
1694    //   block.
1695    BlockFrequency EntryFreq;
1696    if (BFI) {
1697      assert(BPI && "Both BPI and BFI are required to preserve profile info");
1698      for (BasicBlock *Pred : predecessors(header)) {
1699        if (Blocks.count(Pred))
1700          continue;
1701        EntryFreq +=
1702            BFI->getBlockFreq(Pred) * BPI->getEdgeProbability(Pred, header);
1703      }
1704    }
1705  
1706    // Remove @llvm.assume calls that will be moved to the new function from the
1707    // old function's assumption cache.
1708    for (BasicBlock *Block : Blocks) {
1709      for (Instruction &I : llvm::make_early_inc_range(*Block)) {
1710        if (auto *AI = dyn_cast<AssumeInst>(&I)) {
1711          if (AC)
1712            AC->unregisterAssumption(AI);
1713          AI->eraseFromParent();
1714        }
1715      }
1716    }
1717  
1718    // If we have any return instructions in the region, split those blocks so
1719    // that the return is not in the region.
1720    splitReturnBlocks();
1721  
1722    // Calculate the exit blocks for the extracted region and the total exit
1723    // weights for each of those blocks.
1724    DenseMap<BasicBlock *, BlockFrequency> ExitWeights;
1725    SmallPtrSet<BasicBlock *, 1> ExitBlocks;
1726    for (BasicBlock *Block : Blocks) {
1727      for (BasicBlock *Succ : successors(Block)) {
1728        if (!Blocks.count(Succ)) {
1729          // Update the branch weight for this successor.
1730          if (BFI) {
1731            BlockFrequency &BF = ExitWeights[Succ];
1732            BF += BFI->getBlockFreq(Block) * BPI->getEdgeProbability(Block, Succ);
1733          }
1734          ExitBlocks.insert(Succ);
1735        }
1736      }
1737    }
1738    NumExitBlocks = ExitBlocks.size();
1739  
1740    for (BasicBlock *Block : Blocks) {
1741      for (BasicBlock *OldTarget : successors(Block))
1742        if (!Blocks.contains(OldTarget))
1743          OldTargets.push_back(OldTarget);
1744    }
1745  
1746    // If we have to split PHI nodes of the entry or exit blocks, do so now.
1747    severSplitPHINodesOfEntry(header);
1748    severSplitPHINodesOfExits(ExitBlocks);
1749  
1750    // This takes place of the original loop
1751    BasicBlock *codeReplacer = BasicBlock::Create(header->getContext(),
1752                                                  "codeRepl", oldFunction,
1753                                                  header);
1754    codeReplacer->IsNewDbgInfoFormat = oldFunction->IsNewDbgInfoFormat;
1755  
1756    // The new function needs a root node because other nodes can branch to the
1757    // head of the region, but the entry node of a function cannot have preds.
1758    BasicBlock *newFuncRoot = BasicBlock::Create(header->getContext(),
1759                                                 "newFuncRoot");
1760    newFuncRoot->IsNewDbgInfoFormat = oldFunction->IsNewDbgInfoFormat;
1761  
1762    auto *BranchI = BranchInst::Create(header);
1763    // If the original function has debug info, we have to add a debug location
1764    // to the new branch instruction from the artificial entry block.
1765    // We use the debug location of the first instruction in the extracted
1766    // blocks, as there is no other equivalent line in the source code.
1767    if (oldFunction->getSubprogram()) {
1768      any_of(Blocks, [&BranchI](const BasicBlock *BB) {
1769        return any_of(*BB, [&BranchI](const Instruction &I) {
1770          if (!I.getDebugLoc())
1771            return false;
1772          BranchI->setDebugLoc(I.getDebugLoc());
1773          return true;
1774        });
1775      });
1776    }
1777    BranchI->insertInto(newFuncRoot, newFuncRoot->end());
1778  
1779    ValueSet SinkingCands, HoistingCands;
1780    BasicBlock *CommonExit = nullptr;
1781    findAllocas(CEAC, SinkingCands, HoistingCands, CommonExit);
1782    assert(HoistingCands.empty() || CommonExit);
1783  
1784    // Find inputs to, outputs from the code region.
1785    findInputsOutputs(inputs, outputs, SinkingCands);
1786  
1787    // Now sink all instructions which only have non-phi uses inside the region.
1788    // Group the allocas at the start of the block, so that any bitcast uses of
1789    // the allocas are well-defined.
1790    AllocaInst *FirstSunkAlloca = nullptr;
1791    for (auto *II : SinkingCands) {
1792      if (auto *AI = dyn_cast<AllocaInst>(II)) {
1793        AI->moveBefore(*newFuncRoot, newFuncRoot->getFirstInsertionPt());
1794        if (!FirstSunkAlloca)
1795          FirstSunkAlloca = AI;
1796      }
1797    }
1798    assert((SinkingCands.empty() || FirstSunkAlloca) &&
1799           "Did not expect a sink candidate without any allocas");
1800    for (auto *II : SinkingCands) {
1801      if (!isa<AllocaInst>(II)) {
1802        cast<Instruction>(II)->moveAfter(FirstSunkAlloca);
1803      }
1804    }
1805  
1806    if (!HoistingCands.empty()) {
1807      auto *HoistToBlock = findOrCreateBlockForHoisting(CommonExit);
1808      Instruction *TI = HoistToBlock->getTerminator();
1809      for (auto *II : HoistingCands)
1810        cast<Instruction>(II)->moveBefore(TI);
1811    }
1812  
1813    // Collect objects which are inputs to the extraction region and also
1814    // referenced by lifetime start markers within it. The effects of these
1815    // markers must be replicated in the calling function to prevent the stack
1816    // coloring pass from merging slots which store input objects.
1817    ValueSet LifetimesStart;
1818    eraseLifetimeMarkersOnInputs(Blocks, SinkingCands, LifetimesStart);
1819  
1820    // Construct new function based on inputs/outputs & add allocas for all defs.
1821    Function *newFunction =
1822        constructFunction(inputs, outputs, header, newFuncRoot, codeReplacer,
1823                          oldFunction, oldFunction->getParent());
1824  
1825    // Update the entry count of the function.
1826    if (BFI) {
1827      auto Count = BFI->getProfileCountFromFreq(EntryFreq);
1828      if (Count)
1829        newFunction->setEntryCount(
1830            ProfileCount(*Count, Function::PCT_Real)); // FIXME
1831      BFI->setBlockFreq(codeReplacer, EntryFreq);
1832    }
1833  
1834    CallInst *TheCall =
1835        emitCallAndSwitchStatement(newFunction, codeReplacer, inputs, outputs);
1836  
1837    moveCodeToFunction(newFunction);
1838  
1839    // Replicate the effects of any lifetime start/end markers which referenced
1840    // input objects in the extraction region by placing markers around the call.
1841    insertLifetimeMarkersSurroundingCall(
1842        oldFunction->getParent(), LifetimesStart.getArrayRef(), {}, TheCall);
1843  
1844    // Propagate personality info to the new function if there is one.
1845    if (oldFunction->hasPersonalityFn())
1846      newFunction->setPersonalityFn(oldFunction->getPersonalityFn());
1847  
1848    // Update the branch weights for the exit block.
1849    if (BFI && NumExitBlocks > 1)
1850      calculateNewCallTerminatorWeights(codeReplacer, ExitWeights, BPI);
1851  
1852    // Loop over all of the PHI nodes in the header and exit blocks, and change
1853    // any references to the old incoming edge to be the new incoming edge.
1854    for (BasicBlock::iterator I = header->begin(); isa<PHINode>(I); ++I) {
1855      PHINode *PN = cast<PHINode>(I);
1856      for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
1857        if (!Blocks.count(PN->getIncomingBlock(i)))
1858          PN->setIncomingBlock(i, newFuncRoot);
1859    }
1860  
1861    for (BasicBlock *ExitBB : ExitBlocks)
1862      for (PHINode &PN : ExitBB->phis()) {
1863        Value *IncomingCodeReplacerVal = nullptr;
1864        for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) {
1865          // Ignore incoming values from outside of the extracted region.
1866          if (!Blocks.count(PN.getIncomingBlock(i)))
1867            continue;
1868  
1869          // Ensure that there is only one incoming value from codeReplacer.
1870          if (!IncomingCodeReplacerVal) {
1871            PN.setIncomingBlock(i, codeReplacer);
1872            IncomingCodeReplacerVal = PN.getIncomingValue(i);
1873          } else
1874            assert(IncomingCodeReplacerVal == PN.getIncomingValue(i) &&
1875                   "PHI has two incompatbile incoming values from codeRepl");
1876        }
1877      }
1878  
1879    fixupDebugInfoPostExtraction(*oldFunction, *newFunction, *TheCall);
1880  
1881    // Mark the new function `noreturn` if applicable. Terminators which resume
1882    // exception propagation are treated as returning instructions. This is to
1883    // avoid inserting traps after calls to outlined functions which unwind.
1884    bool doesNotReturn = none_of(*newFunction, [](const BasicBlock &BB) {
1885      const Instruction *Term = BB.getTerminator();
1886      return isa<ReturnInst>(Term) || isa<ResumeInst>(Term);
1887    });
1888    if (doesNotReturn)
1889      newFunction->setDoesNotReturn();
1890  
1891    LLVM_DEBUG(if (verifyFunction(*newFunction, &errs())) {
1892      newFunction->dump();
1893      report_fatal_error("verification of newFunction failed!");
1894    });
1895    LLVM_DEBUG(if (verifyFunction(*oldFunction))
1896               report_fatal_error("verification of oldFunction failed!"));
1897    LLVM_DEBUG(if (AC && verifyAssumptionCache(*oldFunction, *newFunction, AC))
1898                   report_fatal_error("Stale Asumption cache for old Function!"));
1899    return newFunction;
1900  }
1901  
1902  bool CodeExtractor::verifyAssumptionCache(const Function &OldFunc,
1903                                            const Function &NewFunc,
1904                                            AssumptionCache *AC) {
1905    for (auto AssumeVH : AC->assumptions()) {
1906      auto *I = dyn_cast_or_null<CallInst>(AssumeVH);
1907      if (!I)
1908        continue;
1909  
1910      // There shouldn't be any llvm.assume intrinsics in the new function.
1911      if (I->getFunction() != &OldFunc)
1912        return true;
1913  
1914      // There shouldn't be any stale affected values in the assumption cache
1915      // that were previously in the old function, but that have now been moved
1916      // to the new function.
1917      for (auto AffectedValVH : AC->assumptionsFor(I->getOperand(0))) {
1918        auto *AffectedCI = dyn_cast_or_null<CallInst>(AffectedValVH);
1919        if (!AffectedCI)
1920          continue;
1921        if (AffectedCI->getFunction() != &OldFunc)
1922          return true;
1923        auto *AssumedInst = cast<Instruction>(AffectedCI->getOperand(0));
1924        if (AssumedInst->getFunction() != &OldFunc)
1925          return true;
1926      }
1927    }
1928    return false;
1929  }
1930  
1931  void CodeExtractor::excludeArgFromAggregate(Value *Arg) {
1932    ExcludeArgsFromAggregate.insert(Arg);
1933  }
1934