xref: /freebsd/contrib/llvm-project/llvm/lib/Transforms/Utils/CodeExtractor.cpp (revision 9bc300465e48e19d794d88d0c158a2adb92c7197)
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