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/Optional.h"
19 #include "llvm/ADT/STLExtras.h"
20 #include "llvm/ADT/SetVector.h"
21 #include "llvm/ADT/SmallPtrSet.h"
22 #include "llvm/ADT/SmallVector.h"
23 #include "llvm/Analysis/AssumptionCache.h"
24 #include "llvm/Analysis/BlockFrequencyInfo.h"
25 #include "llvm/Analysis/BlockFrequencyInfoImpl.h"
26 #include "llvm/Analysis/BranchProbabilityInfo.h"
27 #include "llvm/Analysis/LoopInfo.h"
28 #include "llvm/IR/Argument.h"
29 #include "llvm/IR/Attributes.h"
30 #include "llvm/IR/BasicBlock.h"
31 #include "llvm/IR/CFG.h"
32 #include "llvm/IR/Constant.h"
33 #include "llvm/IR/Constants.h"
34 #include "llvm/IR/DIBuilder.h"
35 #include "llvm/IR/DataLayout.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/Pass.h"
56 #include "llvm/Support/BlockFrequency.h"
57 #include "llvm/Support/BranchProbability.h"
58 #include "llvm/Support/Casting.h"
59 #include "llvm/Support/CommandLine.h"
60 #include "llvm/Support/Debug.h"
61 #include "llvm/Support/ErrorHandling.h"
62 #include "llvm/Support/raw_ostream.h"
63 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
64 #include "llvm/Transforms/Utils/Local.h"
65 #include <cassert>
66 #include <cstdint>
67 #include <iterator>
68 #include <map>
69 #include <set>
70 #include <utility>
71 #include <vector>
72 
73 using namespace llvm;
74 using namespace llvm::PatternMatch;
75 using ProfileCount = Function::ProfileCount;
76 
77 #define DEBUG_TYPE "code-extractor"
78 
79 // Provide a command-line option to aggregate function arguments into a struct
80 // for functions produced by the code extractor. This is useful when converting
81 // extracted functions to pthread-based code, as only one argument (void*) can
82 // be passed in to pthread_create().
83 static cl::opt<bool>
84 AggregateArgsOpt("aggregate-extracted-args", cl::Hidden,
85                  cl::desc("Aggregate arguments to code-extracted functions"));
86 
87 /// Test whether a block is valid for extraction.
isBlockValidForExtraction(const BasicBlock & BB,const SetVector<BasicBlock * > & Result,bool AllowVarArgs,bool AllowAlloca)88 static bool isBlockValidForExtraction(const BasicBlock &BB,
89                                       const SetVector<BasicBlock *> &Result,
90                                       bool AllowVarArgs, bool AllowAlloca) {
91   // taking the address of a basic block moved to another function is illegal
92   if (BB.hasAddressTaken())
93     return false;
94 
95   // don't hoist code that uses another basicblock address, as it's likely to
96   // lead to unexpected behavior, like cross-function jumps
97   SmallPtrSet<User const *, 16> Visited;
98   SmallVector<User const *, 16> ToVisit;
99 
100   for (Instruction const &Inst : BB)
101     ToVisit.push_back(&Inst);
102 
103   while (!ToVisit.empty()) {
104     User const *Curr = ToVisit.pop_back_val();
105     if (!Visited.insert(Curr).second)
106       continue;
107     if (isa<BlockAddress const>(Curr))
108       return false; // even a reference to self is likely to be not compatible
109 
110     if (isa<Instruction>(Curr) && cast<Instruction>(Curr)->getParent() != &BB)
111       continue;
112 
113     for (auto const &U : Curr->operands()) {
114       if (auto *UU = dyn_cast<User>(U))
115         ToVisit.push_back(UU);
116     }
117   }
118 
119   // If explicitly requested, allow vastart and alloca. For invoke instructions
120   // verify that extraction is valid.
121   for (BasicBlock::const_iterator I = BB.begin(), E = BB.end(); I != E; ++I) {
122     if (isa<AllocaInst>(I)) {
123        if (!AllowAlloca)
124          return false;
125        continue;
126     }
127 
128     if (const auto *II = dyn_cast<InvokeInst>(I)) {
129       // Unwind destination (either a landingpad, catchswitch, or cleanuppad)
130       // must be a part of the subgraph which is being extracted.
131       if (auto *UBB = II->getUnwindDest())
132         if (!Result.count(UBB))
133           return false;
134       continue;
135     }
136 
137     // All catch handlers of a catchswitch instruction as well as the unwind
138     // destination must be in the subgraph.
139     if (const auto *CSI = dyn_cast<CatchSwitchInst>(I)) {
140       if (auto *UBB = CSI->getUnwindDest())
141         if (!Result.count(UBB))
142           return false;
143       for (auto *HBB : CSI->handlers())
144         if (!Result.count(const_cast<BasicBlock*>(HBB)))
145           return false;
146       continue;
147     }
148 
149     // Make sure that entire catch handler is within subgraph. It is sufficient
150     // to check that catch return's block is in the list.
151     if (const auto *CPI = dyn_cast<CatchPadInst>(I)) {
152       for (const auto *U : CPI->users())
153         if (const auto *CRI = dyn_cast<CatchReturnInst>(U))
154           if (!Result.count(const_cast<BasicBlock*>(CRI->getParent())))
155             return false;
156       continue;
157     }
158 
159     // And do similar checks for cleanup handler - the entire handler must be
160     // in subgraph which is going to be extracted. For cleanup return should
161     // additionally check that the unwind destination is also in the subgraph.
162     if (const auto *CPI = dyn_cast<CleanupPadInst>(I)) {
163       for (const auto *U : CPI->users())
164         if (const auto *CRI = dyn_cast<CleanupReturnInst>(U))
165           if (!Result.count(const_cast<BasicBlock*>(CRI->getParent())))
166             return false;
167       continue;
168     }
169     if (const auto *CRI = dyn_cast<CleanupReturnInst>(I)) {
170       if (auto *UBB = CRI->getUnwindDest())
171         if (!Result.count(UBB))
172           return false;
173       continue;
174     }
175 
176     if (const CallInst *CI = dyn_cast<CallInst>(I)) {
177       if (const Function *F = CI->getCalledFunction()) {
178         auto IID = F->getIntrinsicID();
179         if (IID == Intrinsic::vastart) {
180           if (AllowVarArgs)
181             continue;
182           else
183             return false;
184         }
185 
186         // Currently, we miscompile outlined copies of eh_typid_for. There are
187         // proposals for fixing this in llvm.org/PR39545.
188         if (IID == Intrinsic::eh_typeid_for)
189           return false;
190       }
191     }
192   }
193 
194   return true;
195 }
196 
197 /// Build a set of blocks to extract if the input blocks are viable.
198 static SetVector<BasicBlock *>
buildExtractionBlockSet(ArrayRef<BasicBlock * > BBs,DominatorTree * DT,bool AllowVarArgs,bool AllowAlloca)199 buildExtractionBlockSet(ArrayRef<BasicBlock *> BBs, DominatorTree *DT,
200                         bool AllowVarArgs, bool AllowAlloca) {
201   assert(!BBs.empty() && "The set of blocks to extract must be non-empty");
202   SetVector<BasicBlock *> Result;
203 
204   // Loop over the blocks, adding them to our set-vector, and aborting with an
205   // empty set if we encounter invalid blocks.
206   for (BasicBlock *BB : BBs) {
207     // If this block is dead, don't process it.
208     if (DT && !DT->isReachableFromEntry(BB))
209       continue;
210 
211     if (!Result.insert(BB))
212       llvm_unreachable("Repeated basic blocks in extraction input");
213   }
214 
215   LLVM_DEBUG(dbgs() << "Region front block: " << Result.front()->getName()
216                     << '\n');
217 
218   for (auto *BB : Result) {
219     if (!isBlockValidForExtraction(*BB, Result, AllowVarArgs, AllowAlloca))
220       return {};
221 
222     // Make sure that the first block is not a landing pad.
223     if (BB == Result.front()) {
224       if (BB->isEHPad()) {
225         LLVM_DEBUG(dbgs() << "The first block cannot be an unwind block\n");
226         return {};
227       }
228       continue;
229     }
230 
231     // All blocks other than the first must not have predecessors outside of
232     // the subgraph which is being extracted.
233     for (auto *PBB : predecessors(BB))
234       if (!Result.count(PBB)) {
235         LLVM_DEBUG(dbgs() << "No blocks in this region may have entries from "
236                              "outside the region except for the first block!\n"
237                           << "Problematic source BB: " << BB->getName() << "\n"
238                           << "Problematic destination BB: " << PBB->getName()
239                           << "\n");
240         return {};
241       }
242   }
243 
244   return Result;
245 }
246 
CodeExtractor(ArrayRef<BasicBlock * > BBs,DominatorTree * DT,bool AggregateArgs,BlockFrequencyInfo * BFI,BranchProbabilityInfo * BPI,AssumptionCache * AC,bool AllowVarArgs,bool AllowAlloca,std::string Suffix)247 CodeExtractor::CodeExtractor(ArrayRef<BasicBlock *> BBs, DominatorTree *DT,
248                              bool AggregateArgs, BlockFrequencyInfo *BFI,
249                              BranchProbabilityInfo *BPI, AssumptionCache *AC,
250                              bool AllowVarArgs, bool AllowAlloca,
251                              std::string Suffix)
252     : DT(DT), AggregateArgs(AggregateArgs || AggregateArgsOpt), BFI(BFI),
253       BPI(BPI), AC(AC), AllowVarArgs(AllowVarArgs),
254       Blocks(buildExtractionBlockSet(BBs, DT, AllowVarArgs, AllowAlloca)),
255       Suffix(Suffix) {}
256 
CodeExtractor(DominatorTree & DT,Loop & L,bool AggregateArgs,BlockFrequencyInfo * BFI,BranchProbabilityInfo * BPI,AssumptionCache * AC,std::string Suffix)257 CodeExtractor::CodeExtractor(DominatorTree &DT, Loop &L, bool AggregateArgs,
258                              BlockFrequencyInfo *BFI,
259                              BranchProbabilityInfo *BPI, AssumptionCache *AC,
260                              std::string Suffix)
261     : DT(&DT), AggregateArgs(AggregateArgs || AggregateArgsOpt), BFI(BFI),
262       BPI(BPI), AC(AC), AllowVarArgs(false),
263       Blocks(buildExtractionBlockSet(L.getBlocks(), &DT,
264                                      /* AllowVarArgs */ false,
265                                      /* AllowAlloca */ false)),
266       Suffix(Suffix) {}
267 
268 /// definedInRegion - Return true if the specified value is defined in the
269 /// extracted region.
definedInRegion(const SetVector<BasicBlock * > & Blocks,Value * V)270 static bool definedInRegion(const SetVector<BasicBlock *> &Blocks, Value *V) {
271   if (Instruction *I = dyn_cast<Instruction>(V))
272     if (Blocks.count(I->getParent()))
273       return true;
274   return false;
275 }
276 
277 /// definedInCaller - Return true if the specified value is defined in the
278 /// function being code extracted, but not in the region being extracted.
279 /// These values must be passed in as live-ins to the function.
definedInCaller(const SetVector<BasicBlock * > & Blocks,Value * V)280 static bool definedInCaller(const SetVector<BasicBlock *> &Blocks, Value *V) {
281   if (isa<Argument>(V)) return true;
282   if (Instruction *I = dyn_cast<Instruction>(V))
283     if (!Blocks.count(I->getParent()))
284       return true;
285   return false;
286 }
287 
getCommonExitBlock(const SetVector<BasicBlock * > & Blocks)288 static BasicBlock *getCommonExitBlock(const SetVector<BasicBlock *> &Blocks) {
289   BasicBlock *CommonExitBlock = nullptr;
290   auto hasNonCommonExitSucc = [&](BasicBlock *Block) {
291     for (auto *Succ : successors(Block)) {
292       // Internal edges, ok.
293       if (Blocks.count(Succ))
294         continue;
295       if (!CommonExitBlock) {
296         CommonExitBlock = Succ;
297         continue;
298       }
299       if (CommonExitBlock != Succ)
300         return true;
301     }
302     return false;
303   };
304 
305   if (any_of(Blocks, hasNonCommonExitSucc))
306     return nullptr;
307 
308   return CommonExitBlock;
309 }
310 
CodeExtractorAnalysisCache(Function & F)311 CodeExtractorAnalysisCache::CodeExtractorAnalysisCache(Function &F) {
312   for (BasicBlock &BB : F) {
313     for (Instruction &II : BB.instructionsWithoutDebug())
314       if (auto *AI = dyn_cast<AllocaInst>(&II))
315         Allocas.push_back(AI);
316 
317     findSideEffectInfoForBlock(BB);
318   }
319 }
320 
findSideEffectInfoForBlock(BasicBlock & BB)321 void CodeExtractorAnalysisCache::findSideEffectInfoForBlock(BasicBlock &BB) {
322   for (Instruction &II : BB.instructionsWithoutDebug()) {
323     unsigned Opcode = II.getOpcode();
324     Value *MemAddr = nullptr;
325     switch (Opcode) {
326     case Instruction::Store:
327     case Instruction::Load: {
328       if (Opcode == Instruction::Store) {
329         StoreInst *SI = cast<StoreInst>(&II);
330         MemAddr = SI->getPointerOperand();
331       } else {
332         LoadInst *LI = cast<LoadInst>(&II);
333         MemAddr = LI->getPointerOperand();
334       }
335       // Global variable can not be aliased with locals.
336       if (isa<Constant>(MemAddr))
337         break;
338       Value *Base = MemAddr->stripInBoundsConstantOffsets();
339       if (!isa<AllocaInst>(Base)) {
340         SideEffectingBlocks.insert(&BB);
341         return;
342       }
343       BaseMemAddrs[&BB].insert(Base);
344       break;
345     }
346     default: {
347       IntrinsicInst *IntrInst = dyn_cast<IntrinsicInst>(&II);
348       if (IntrInst) {
349         if (IntrInst->isLifetimeStartOrEnd())
350           break;
351         SideEffectingBlocks.insert(&BB);
352         return;
353       }
354       // Treat all the other cases conservatively if it has side effects.
355       if (II.mayHaveSideEffects()) {
356         SideEffectingBlocks.insert(&BB);
357         return;
358       }
359     }
360     }
361   }
362 }
363 
doesBlockContainClobberOfAddr(BasicBlock & BB,AllocaInst * Addr) const364 bool CodeExtractorAnalysisCache::doesBlockContainClobberOfAddr(
365     BasicBlock &BB, AllocaInst *Addr) const {
366   if (SideEffectingBlocks.count(&BB))
367     return true;
368   auto It = BaseMemAddrs.find(&BB);
369   if (It != BaseMemAddrs.end())
370     return It->second.count(Addr);
371   return false;
372 }
373 
isLegalToShrinkwrapLifetimeMarkers(const CodeExtractorAnalysisCache & CEAC,Instruction * Addr) const374 bool CodeExtractor::isLegalToShrinkwrapLifetimeMarkers(
375     const CodeExtractorAnalysisCache &CEAC, Instruction *Addr) const {
376   AllocaInst *AI = cast<AllocaInst>(Addr->stripInBoundsConstantOffsets());
377   Function *Func = (*Blocks.begin())->getParent();
378   for (BasicBlock &BB : *Func) {
379     if (Blocks.count(&BB))
380       continue;
381     if (CEAC.doesBlockContainClobberOfAddr(BB, AI))
382       return false;
383   }
384   return true;
385 }
386 
387 BasicBlock *
findOrCreateBlockForHoisting(BasicBlock * CommonExitBlock)388 CodeExtractor::findOrCreateBlockForHoisting(BasicBlock *CommonExitBlock) {
389   BasicBlock *SinglePredFromOutlineRegion = nullptr;
390   assert(!Blocks.count(CommonExitBlock) &&
391          "Expect a block outside the region!");
392   for (auto *Pred : predecessors(CommonExitBlock)) {
393     if (!Blocks.count(Pred))
394       continue;
395     if (!SinglePredFromOutlineRegion) {
396       SinglePredFromOutlineRegion = Pred;
397     } else if (SinglePredFromOutlineRegion != Pred) {
398       SinglePredFromOutlineRegion = nullptr;
399       break;
400     }
401   }
402 
403   if (SinglePredFromOutlineRegion)
404     return SinglePredFromOutlineRegion;
405 
406 #ifndef NDEBUG
407   auto getFirstPHI = [](BasicBlock *BB) {
408     BasicBlock::iterator I = BB->begin();
409     PHINode *FirstPhi = nullptr;
410     while (I != BB->end()) {
411       PHINode *Phi = dyn_cast<PHINode>(I);
412       if (!Phi)
413         break;
414       if (!FirstPhi) {
415         FirstPhi = Phi;
416         break;
417       }
418     }
419     return FirstPhi;
420   };
421   // If there are any phi nodes, the single pred either exists or has already
422   // be created before code extraction.
423   assert(!getFirstPHI(CommonExitBlock) && "Phi not expected");
424 #endif
425 
426   BasicBlock *NewExitBlock = CommonExitBlock->splitBasicBlock(
427       CommonExitBlock->getFirstNonPHI()->getIterator());
428 
429   for (BasicBlock *Pred :
430        llvm::make_early_inc_range(predecessors(CommonExitBlock))) {
431     if (Blocks.count(Pred))
432       continue;
433     Pred->getTerminator()->replaceUsesOfWith(CommonExitBlock, NewExitBlock);
434   }
435   // Now add the old exit block to the outline region.
436   Blocks.insert(CommonExitBlock);
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
getLifetimeMarkers(const CodeExtractorAnalysisCache & CEAC,Instruction * Addr,BasicBlock * ExitBlock) const445 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 
findAllocas(const CodeExtractorAnalysisCache & CEAC,ValueSet & SinkCands,ValueSet & HoistCands,BasicBlock * & ExitBlock) const494 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 = Type::getInt8PtrTy(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 
isEligible() const619 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 
findInputsOutputs(ValueSet & Inputs,ValueSet & Outputs,const ValueSet & SinkCands) const646 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.
severSplitPHINodesOfEntry(BasicBlock * & Header)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", &NewBB->front());
726       PN->replaceAllUsesWith(NewPN);
727       NewPN->addIncoming(PN, OldPred);
728 
729       // Loop over all of the incoming value in PN, moving them to NewPN if they
730       // are from the extracted region.
731       for (unsigned i = 0; i != PN->getNumIncomingValues(); ++i) {
732         if (Blocks.count(PN->getIncomingBlock(i))) {
733           NewPN->addIncoming(PN->getIncomingValue(i), PN->getIncomingBlock(i));
734           PN->removeIncomingValue(i);
735           --i;
736         }
737       }
738     }
739   }
740 }
741 
742 /// severSplitPHINodesOfExits - if PHI nodes in exit blocks have inputs from
743 /// outlined region, we split these PHIs on two: one with inputs from region
744 /// and other with remaining incoming blocks; then first PHIs are placed in
745 /// outlined region.
severSplitPHINodesOfExits(const SmallPtrSetImpl<BasicBlock * > & Exits)746 void CodeExtractor::severSplitPHINodesOfExits(
747     const SmallPtrSetImpl<BasicBlock *> &Exits) {
748   for (BasicBlock *ExitBB : Exits) {
749     BasicBlock *NewBB = nullptr;
750 
751     for (PHINode &PN : ExitBB->phis()) {
752       // Find all incoming values from the outlining region.
753       SmallVector<unsigned, 2> IncomingVals;
754       for (unsigned i = 0; i < PN.getNumIncomingValues(); ++i)
755         if (Blocks.count(PN.getIncomingBlock(i)))
756           IncomingVals.push_back(i);
757 
758       // Do not process PHI if there is one (or fewer) predecessor from region.
759       // If PHI has exactly one predecessor from region, only this one incoming
760       // will be replaced on codeRepl block, so it should be safe to skip PHI.
761       if (IncomingVals.size() <= 1)
762         continue;
763 
764       // Create block for new PHIs and add it to the list of outlined if it
765       // wasn't done before.
766       if (!NewBB) {
767         NewBB = BasicBlock::Create(ExitBB->getContext(),
768                                    ExitBB->getName() + ".split",
769                                    ExitBB->getParent(), ExitBB);
770         SmallVector<BasicBlock *, 4> Preds(predecessors(ExitBB));
771         for (BasicBlock *PredBB : Preds)
772           if (Blocks.count(PredBB))
773             PredBB->getTerminator()->replaceUsesOfWith(ExitBB, NewBB);
774         BranchInst::Create(ExitBB, NewBB);
775         Blocks.insert(NewBB);
776       }
777 
778       // Split this PHI.
779       PHINode *NewPN =
780           PHINode::Create(PN.getType(), IncomingVals.size(),
781                           PN.getName() + ".ce", NewBB->getFirstNonPHI());
782       for (unsigned i : IncomingVals)
783         NewPN->addIncoming(PN.getIncomingValue(i), PN.getIncomingBlock(i));
784       for (unsigned i : reverse(IncomingVals))
785         PN.removeIncomingValue(i, false);
786       PN.addIncoming(NewPN, NewBB);
787     }
788   }
789 }
790 
splitReturnBlocks()791 void CodeExtractor::splitReturnBlocks() {
792   for (BasicBlock *Block : Blocks)
793     if (ReturnInst *RI = dyn_cast<ReturnInst>(Block->getTerminator())) {
794       BasicBlock *New =
795           Block->splitBasicBlock(RI->getIterator(), Block->getName() + ".ret");
796       if (DT) {
797         // Old dominates New. New node dominates all other nodes dominated
798         // by Old.
799         DomTreeNode *OldNode = DT->getNode(Block);
800         SmallVector<DomTreeNode *, 8> Children(OldNode->begin(),
801                                                OldNode->end());
802 
803         DomTreeNode *NewNode = DT->addNewBlock(New, Block);
804 
805         for (DomTreeNode *I : Children)
806           DT->changeImmediateDominator(I, NewNode);
807       }
808     }
809 }
810 
811 /// constructFunction - make a function based on inputs and outputs, as follows:
812 /// f(in0, ..., inN, out0, ..., outN)
constructFunction(const ValueSet & inputs,const ValueSet & outputs,BasicBlock * header,BasicBlock * newRootNode,BasicBlock * newHeader,Function * oldFunction,Module * M)813 Function *CodeExtractor::constructFunction(const ValueSet &inputs,
814                                            const ValueSet &outputs,
815                                            BasicBlock *header,
816                                            BasicBlock *newRootNode,
817                                            BasicBlock *newHeader,
818                                            Function *oldFunction,
819                                            Module *M) {
820   LLVM_DEBUG(dbgs() << "inputs: " << inputs.size() << "\n");
821   LLVM_DEBUG(dbgs() << "outputs: " << outputs.size() << "\n");
822 
823   // This function returns unsigned, outputs will go back by reference.
824   switch (NumExitBlocks) {
825   case 0:
826   case 1: RetTy = Type::getVoidTy(header->getContext()); break;
827   case 2: RetTy = Type::getInt1Ty(header->getContext()); break;
828   default: RetTy = Type::getInt16Ty(header->getContext()); break;
829   }
830 
831   std::vector<Type *> paramTy;
832 
833   // Add the types of the input values to the function's argument list
834   for (Value *value : inputs) {
835     LLVM_DEBUG(dbgs() << "value used in func: " << *value << "\n");
836     paramTy.push_back(value->getType());
837   }
838 
839   // Add the types of the output values to the function's argument list.
840   for (Value *output : outputs) {
841     LLVM_DEBUG(dbgs() << "instr used in func: " << *output << "\n");
842     if (AggregateArgs)
843       paramTy.push_back(output->getType());
844     else
845       paramTy.push_back(PointerType::getUnqual(output->getType()));
846   }
847 
848   LLVM_DEBUG({
849     dbgs() << "Function type: " << *RetTy << " f(";
850     for (Type *i : paramTy)
851       dbgs() << *i << ", ";
852     dbgs() << ")\n";
853   });
854 
855   StructType *StructTy = nullptr;
856   if (AggregateArgs && (inputs.size() + outputs.size() > 0)) {
857     StructTy = StructType::get(M->getContext(), paramTy);
858     paramTy.clear();
859     paramTy.push_back(PointerType::getUnqual(StructTy));
860   }
861   FunctionType *funcType =
862                   FunctionType::get(RetTy, paramTy,
863                                     AllowVarArgs && oldFunction->isVarArg());
864 
865   std::string SuffixToUse =
866       Suffix.empty()
867           ? (header->getName().empty() ? "extracted" : header->getName().str())
868           : Suffix;
869   // Create the new function
870   Function *newFunction = Function::Create(
871       funcType, GlobalValue::InternalLinkage, oldFunction->getAddressSpace(),
872       oldFunction->getName() + "." + SuffixToUse, M);
873   // If the old function is no-throw, so is the new one.
874   if (oldFunction->doesNotThrow())
875     newFunction->setDoesNotThrow();
876 
877   // Inherit the uwtable attribute if we need to.
878   if (oldFunction->hasUWTable())
879     newFunction->setHasUWTable();
880 
881   // Inherit all of the target dependent attributes and white-listed
882   // target independent attributes.
883   //  (e.g. If the extracted region contains a call to an x86.sse
884   //  instruction we need to make sure that the extracted region has the
885   //  "target-features" attribute allowing it to be lowered.
886   // FIXME: This should be changed to check to see if a specific
887   //           attribute can not be inherited.
888   for (const auto &Attr : oldFunction->getAttributes().getFnAttributes()) {
889     if (Attr.isStringAttribute()) {
890       if (Attr.getKindAsString() == "thunk")
891         continue;
892     } else
893       switch (Attr.getKindAsEnum()) {
894       // Those attributes cannot be propagated safely. Explicitly list them
895       // here so we get a warning if new attributes are added. This list also
896       // includes non-function attributes.
897       case Attribute::Alignment:
898       case Attribute::AllocSize:
899       case Attribute::ArgMemOnly:
900       case Attribute::Builtin:
901       case Attribute::ByVal:
902       case Attribute::Convergent:
903       case Attribute::Dereferenceable:
904       case Attribute::DereferenceableOrNull:
905       case Attribute::ElementType:
906       case Attribute::InAlloca:
907       case Attribute::InReg:
908       case Attribute::InaccessibleMemOnly:
909       case Attribute::InaccessibleMemOrArgMemOnly:
910       case Attribute::JumpTable:
911       case Attribute::Naked:
912       case Attribute::Nest:
913       case Attribute::NoAlias:
914       case Attribute::NoBuiltin:
915       case Attribute::NoCapture:
916       case Attribute::NoMerge:
917       case Attribute::NoReturn:
918       case Attribute::NoSync:
919       case Attribute::NoUndef:
920       case Attribute::None:
921       case Attribute::NonNull:
922       case Attribute::Preallocated:
923       case Attribute::ReadNone:
924       case Attribute::ReadOnly:
925       case Attribute::Returned:
926       case Attribute::ReturnsTwice:
927       case Attribute::SExt:
928       case Attribute::Speculatable:
929       case Attribute::StackAlignment:
930       case Attribute::StructRet:
931       case Attribute::SwiftError:
932       case Attribute::SwiftSelf:
933       case Attribute::SwiftAsync:
934       case Attribute::WillReturn:
935       case Attribute::WriteOnly:
936       case Attribute::ZExt:
937       case Attribute::ImmArg:
938       case Attribute::ByRef:
939       case Attribute::EndAttrKinds:
940       case Attribute::EmptyKey:
941       case Attribute::TombstoneKey:
942         continue;
943       // Those attributes should be safe to propagate to the extracted function.
944       case Attribute::AlwaysInline:
945       case Attribute::Cold:
946       case Attribute::Hot:
947       case Attribute::NoRecurse:
948       case Attribute::InlineHint:
949       case Attribute::MinSize:
950       case Attribute::NoCallback:
951       case Attribute::NoDuplicate:
952       case Attribute::NoFree:
953       case Attribute::NoImplicitFloat:
954       case Attribute::NoInline:
955       case Attribute::NonLazyBind:
956       case Attribute::NoRedZone:
957       case Attribute::NoUnwind:
958       case Attribute::NoSanitizeCoverage:
959       case Attribute::NullPointerIsValid:
960       case Attribute::OptForFuzzing:
961       case Attribute::OptimizeNone:
962       case Attribute::OptimizeForSize:
963       case Attribute::SafeStack:
964       case Attribute::ShadowCallStack:
965       case Attribute::SanitizeAddress:
966       case Attribute::SanitizeMemory:
967       case Attribute::SanitizeThread:
968       case Attribute::SanitizeHWAddress:
969       case Attribute::SanitizeMemTag:
970       case Attribute::SpeculativeLoadHardening:
971       case Attribute::StackProtect:
972       case Attribute::StackProtectReq:
973       case Attribute::StackProtectStrong:
974       case Attribute::StrictFP:
975       case Attribute::UWTable:
976       case Attribute::VScaleRange:
977       case Attribute::NoCfCheck:
978       case Attribute::MustProgress:
979       case Attribute::NoProfile:
980         break;
981       }
982 
983     newFunction->addFnAttr(Attr);
984   }
985   newFunction->getBasicBlockList().push_back(newRootNode);
986 
987   // Create an iterator to name all of the arguments we inserted.
988   Function::arg_iterator AI = newFunction->arg_begin();
989 
990   // Rewrite all users of the inputs in the extracted region to use the
991   // arguments (or appropriate addressing into struct) instead.
992   for (unsigned i = 0, e = inputs.size(); i != e; ++i) {
993     Value *RewriteVal;
994     if (AggregateArgs) {
995       Value *Idx[2];
996       Idx[0] = Constant::getNullValue(Type::getInt32Ty(header->getContext()));
997       Idx[1] = ConstantInt::get(Type::getInt32Ty(header->getContext()), i);
998       Instruction *TI = newFunction->begin()->getTerminator();
999       GetElementPtrInst *GEP = GetElementPtrInst::Create(
1000           StructTy, &*AI, Idx, "gep_" + inputs[i]->getName(), TI);
1001       RewriteVal = new LoadInst(StructTy->getElementType(i), GEP,
1002                                 "loadgep_" + inputs[i]->getName(), TI);
1003     } else
1004       RewriteVal = &*AI++;
1005 
1006     std::vector<User *> Users(inputs[i]->user_begin(), inputs[i]->user_end());
1007     for (User *use : Users)
1008       if (Instruction *inst = dyn_cast<Instruction>(use))
1009         if (Blocks.count(inst->getParent()))
1010           inst->replaceUsesOfWith(inputs[i], RewriteVal);
1011   }
1012 
1013   // Set names for input and output arguments.
1014   if (!AggregateArgs) {
1015     AI = newFunction->arg_begin();
1016     for (unsigned i = 0, e = inputs.size(); i != e; ++i, ++AI)
1017       AI->setName(inputs[i]->getName());
1018     for (unsigned i = 0, e = outputs.size(); i != e; ++i, ++AI)
1019       AI->setName(outputs[i]->getName()+".out");
1020   }
1021 
1022   // Rewrite branches to basic blocks outside of the loop to new dummy blocks
1023   // within the new function. This must be done before we lose track of which
1024   // blocks were originally in the code region.
1025   std::vector<User *> Users(header->user_begin(), header->user_end());
1026   for (auto &U : Users)
1027     // The BasicBlock which contains the branch is not in the region
1028     // modify the branch target to a new block
1029     if (Instruction *I = dyn_cast<Instruction>(U))
1030       if (I->isTerminator() && I->getFunction() == oldFunction &&
1031           !Blocks.count(I->getParent()))
1032         I->replaceUsesOfWith(header, newHeader);
1033 
1034   return newFunction;
1035 }
1036 
1037 /// Erase lifetime.start markers which reference inputs to the extraction
1038 /// region, and insert the referenced memory into \p LifetimesStart.
1039 ///
1040 /// The extraction region is defined by a set of blocks (\p Blocks), and a set
1041 /// of allocas which will be moved from the caller function into the extracted
1042 /// function (\p SunkAllocas).
eraseLifetimeMarkersOnInputs(const SetVector<BasicBlock * > & Blocks,const SetVector<Value * > & SunkAllocas,SetVector<Value * > & LifetimesStart)1043 static void eraseLifetimeMarkersOnInputs(const SetVector<BasicBlock *> &Blocks,
1044                                          const SetVector<Value *> &SunkAllocas,
1045                                          SetVector<Value *> &LifetimesStart) {
1046   for (BasicBlock *BB : Blocks) {
1047     for (auto It = BB->begin(), End = BB->end(); It != End;) {
1048       auto *II = dyn_cast<IntrinsicInst>(&*It);
1049       ++It;
1050       if (!II || !II->isLifetimeStartOrEnd())
1051         continue;
1052 
1053       // Get the memory operand of the lifetime marker. If the underlying
1054       // object is a sunk alloca, or is otherwise defined in the extraction
1055       // region, the lifetime marker must not be erased.
1056       Value *Mem = II->getOperand(1)->stripInBoundsOffsets();
1057       if (SunkAllocas.count(Mem) || definedInRegion(Blocks, Mem))
1058         continue;
1059 
1060       if (II->getIntrinsicID() == Intrinsic::lifetime_start)
1061         LifetimesStart.insert(Mem);
1062       II->eraseFromParent();
1063     }
1064   }
1065 }
1066 
1067 /// Insert lifetime start/end markers surrounding the call to the new function
1068 /// for objects defined in the caller.
insertLifetimeMarkersSurroundingCall(Module * M,ArrayRef<Value * > LifetimesStart,ArrayRef<Value * > LifetimesEnd,CallInst * TheCall)1069 static void insertLifetimeMarkersSurroundingCall(
1070     Module *M, ArrayRef<Value *> LifetimesStart, ArrayRef<Value *> LifetimesEnd,
1071     CallInst *TheCall) {
1072   LLVMContext &Ctx = M->getContext();
1073   auto Int8PtrTy = Type::getInt8PtrTy(Ctx);
1074   auto NegativeOne = ConstantInt::getSigned(Type::getInt64Ty(Ctx), -1);
1075   Instruction *Term = TheCall->getParent()->getTerminator();
1076 
1077   // The memory argument to a lifetime marker must be a i8*. Cache any bitcasts
1078   // needed to satisfy this requirement so they may be reused.
1079   DenseMap<Value *, Value *> Bitcasts;
1080 
1081   // Emit lifetime markers for the pointers given in \p Objects. Insert the
1082   // markers before the call if \p InsertBefore, and after the call otherwise.
1083   auto insertMarkers = [&](Function *MarkerFunc, ArrayRef<Value *> Objects,
1084                            bool InsertBefore) {
1085     for (Value *Mem : Objects) {
1086       assert((!isa<Instruction>(Mem) || cast<Instruction>(Mem)->getFunction() ==
1087                                             TheCall->getFunction()) &&
1088              "Input memory not defined in original function");
1089       Value *&MemAsI8Ptr = Bitcasts[Mem];
1090       if (!MemAsI8Ptr) {
1091         if (Mem->getType() == Int8PtrTy)
1092           MemAsI8Ptr = Mem;
1093         else
1094           MemAsI8Ptr =
1095               CastInst::CreatePointerCast(Mem, Int8PtrTy, "lt.cast", TheCall);
1096       }
1097 
1098       auto Marker = CallInst::Create(MarkerFunc, {NegativeOne, MemAsI8Ptr});
1099       if (InsertBefore)
1100         Marker->insertBefore(TheCall);
1101       else
1102         Marker->insertBefore(Term);
1103     }
1104   };
1105 
1106   if (!LifetimesStart.empty()) {
1107     auto StartFn = llvm::Intrinsic::getDeclaration(
1108         M, llvm::Intrinsic::lifetime_start, Int8PtrTy);
1109     insertMarkers(StartFn, LifetimesStart, /*InsertBefore=*/true);
1110   }
1111 
1112   if (!LifetimesEnd.empty()) {
1113     auto EndFn = llvm::Intrinsic::getDeclaration(
1114         M, llvm::Intrinsic::lifetime_end, Int8PtrTy);
1115     insertMarkers(EndFn, LifetimesEnd, /*InsertBefore=*/false);
1116   }
1117 }
1118 
1119 /// emitCallAndSwitchStatement - This method sets up the caller side by adding
1120 /// the call instruction, splitting any PHI nodes in the header block as
1121 /// necessary.
emitCallAndSwitchStatement(Function * newFunction,BasicBlock * codeReplacer,ValueSet & inputs,ValueSet & outputs)1122 CallInst *CodeExtractor::emitCallAndSwitchStatement(Function *newFunction,
1123                                                     BasicBlock *codeReplacer,
1124                                                     ValueSet &inputs,
1125                                                     ValueSet &outputs) {
1126   // Emit a call to the new function, passing in: *pointer to struct (if
1127   // aggregating parameters), or plan inputs and allocated memory for outputs
1128   std::vector<Value *> params, StructValues, ReloadOutputs, Reloads;
1129 
1130   Module *M = newFunction->getParent();
1131   LLVMContext &Context = M->getContext();
1132   const DataLayout &DL = M->getDataLayout();
1133   CallInst *call = nullptr;
1134 
1135   // Add inputs as params, or to be filled into the struct
1136   unsigned ArgNo = 0;
1137   SmallVector<unsigned, 1> SwiftErrorArgs;
1138   for (Value *input : inputs) {
1139     if (AggregateArgs)
1140       StructValues.push_back(input);
1141     else {
1142       params.push_back(input);
1143       if (input->isSwiftError())
1144         SwiftErrorArgs.push_back(ArgNo);
1145     }
1146     ++ArgNo;
1147   }
1148 
1149   // Create allocas for the outputs
1150   for (Value *output : outputs) {
1151     if (AggregateArgs) {
1152       StructValues.push_back(output);
1153     } else {
1154       AllocaInst *alloca =
1155         new AllocaInst(output->getType(), DL.getAllocaAddrSpace(),
1156                        nullptr, output->getName() + ".loc",
1157                        &codeReplacer->getParent()->front().front());
1158       ReloadOutputs.push_back(alloca);
1159       params.push_back(alloca);
1160     }
1161   }
1162 
1163   StructType *StructArgTy = nullptr;
1164   AllocaInst *Struct = nullptr;
1165   if (AggregateArgs && (inputs.size() + outputs.size() > 0)) {
1166     std::vector<Type *> ArgTypes;
1167     for (Value *V : StructValues)
1168       ArgTypes.push_back(V->getType());
1169 
1170     // Allocate a struct at the beginning of this function
1171     StructArgTy = StructType::get(newFunction->getContext(), ArgTypes);
1172     Struct = new AllocaInst(StructArgTy, DL.getAllocaAddrSpace(), nullptr,
1173                             "structArg",
1174                             &codeReplacer->getParent()->front().front());
1175     params.push_back(Struct);
1176 
1177     for (unsigned i = 0, e = inputs.size(); i != e; ++i) {
1178       Value *Idx[2];
1179       Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context));
1180       Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), i);
1181       GetElementPtrInst *GEP = GetElementPtrInst::Create(
1182           StructArgTy, Struct, Idx, "gep_" + StructValues[i]->getName());
1183       codeReplacer->getInstList().push_back(GEP);
1184       new StoreInst(StructValues[i], GEP, codeReplacer);
1185     }
1186   }
1187 
1188   // Emit the call to the function
1189   call = CallInst::Create(newFunction, params,
1190                           NumExitBlocks > 1 ? "targetBlock" : "");
1191   // Add debug location to the new call, if the original function has debug
1192   // info. In that case, the terminator of the entry block of the extracted
1193   // function contains the first debug location of the extracted function,
1194   // set in extractCodeRegion.
1195   if (codeReplacer->getParent()->getSubprogram()) {
1196     if (auto DL = newFunction->getEntryBlock().getTerminator()->getDebugLoc())
1197       call->setDebugLoc(DL);
1198   }
1199   codeReplacer->getInstList().push_back(call);
1200 
1201   // Set swifterror parameter attributes.
1202   for (unsigned SwiftErrArgNo : SwiftErrorArgs) {
1203     call->addParamAttr(SwiftErrArgNo, Attribute::SwiftError);
1204     newFunction->addParamAttr(SwiftErrArgNo, Attribute::SwiftError);
1205   }
1206 
1207   Function::arg_iterator OutputArgBegin = newFunction->arg_begin();
1208   unsigned FirstOut = inputs.size();
1209   if (!AggregateArgs)
1210     std::advance(OutputArgBegin, inputs.size());
1211 
1212   // Reload the outputs passed in by reference.
1213   for (unsigned i = 0, e = outputs.size(); i != e; ++i) {
1214     Value *Output = nullptr;
1215     if (AggregateArgs) {
1216       Value *Idx[2];
1217       Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context));
1218       Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), FirstOut + i);
1219       GetElementPtrInst *GEP = GetElementPtrInst::Create(
1220           StructArgTy, Struct, Idx, "gep_reload_" + outputs[i]->getName());
1221       codeReplacer->getInstList().push_back(GEP);
1222       Output = GEP;
1223     } else {
1224       Output = ReloadOutputs[i];
1225     }
1226     LoadInst *load = new LoadInst(outputs[i]->getType(), Output,
1227                                   outputs[i]->getName() + ".reload",
1228                                   codeReplacer);
1229     Reloads.push_back(load);
1230     std::vector<User *> Users(outputs[i]->user_begin(), outputs[i]->user_end());
1231     for (unsigned u = 0, e = Users.size(); u != e; ++u) {
1232       Instruction *inst = cast<Instruction>(Users[u]);
1233       if (!Blocks.count(inst->getParent()))
1234         inst->replaceUsesOfWith(outputs[i], load);
1235     }
1236   }
1237 
1238   // Now we can emit a switch statement using the call as a value.
1239   SwitchInst *TheSwitch =
1240       SwitchInst::Create(Constant::getNullValue(Type::getInt16Ty(Context)),
1241                          codeReplacer, 0, codeReplacer);
1242 
1243   // Since there may be multiple exits from the original region, make the new
1244   // function return an unsigned, switch on that number.  This loop iterates
1245   // over all of the blocks in the extracted region, updating any terminator
1246   // instructions in the to-be-extracted region that branch to blocks that are
1247   // not in the region to be extracted.
1248   std::map<BasicBlock *, BasicBlock *> ExitBlockMap;
1249 
1250   unsigned switchVal = 0;
1251   for (BasicBlock *Block : Blocks) {
1252     Instruction *TI = Block->getTerminator();
1253     for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
1254       if (!Blocks.count(TI->getSuccessor(i))) {
1255         BasicBlock *OldTarget = TI->getSuccessor(i);
1256         // add a new basic block which returns the appropriate value
1257         BasicBlock *&NewTarget = ExitBlockMap[OldTarget];
1258         if (!NewTarget) {
1259           // If we don't already have an exit stub for this non-extracted
1260           // destination, create one now!
1261           NewTarget = BasicBlock::Create(Context,
1262                                          OldTarget->getName() + ".exitStub",
1263                                          newFunction);
1264           unsigned SuccNum = switchVal++;
1265 
1266           Value *brVal = nullptr;
1267           switch (NumExitBlocks) {
1268           case 0:
1269           case 1: break;  // No value needed.
1270           case 2:         // Conditional branch, return a bool
1271             brVal = ConstantInt::get(Type::getInt1Ty(Context), !SuccNum);
1272             break;
1273           default:
1274             brVal = ConstantInt::get(Type::getInt16Ty(Context), SuccNum);
1275             break;
1276           }
1277 
1278           ReturnInst::Create(Context, brVal, NewTarget);
1279 
1280           // Update the switch instruction.
1281           TheSwitch->addCase(ConstantInt::get(Type::getInt16Ty(Context),
1282                                               SuccNum),
1283                              OldTarget);
1284         }
1285 
1286         // rewrite the original branch instruction with this new target
1287         TI->setSuccessor(i, NewTarget);
1288       }
1289   }
1290 
1291   // Store the arguments right after the definition of output value.
1292   // This should be proceeded after creating exit stubs to be ensure that invoke
1293   // result restore will be placed in the outlined function.
1294   Function::arg_iterator OAI = OutputArgBegin;
1295   for (unsigned i = 0, e = outputs.size(); i != e; ++i) {
1296     auto *OutI = dyn_cast<Instruction>(outputs[i]);
1297     if (!OutI)
1298       continue;
1299 
1300     // Find proper insertion point.
1301     BasicBlock::iterator InsertPt;
1302     // In case OutI is an invoke, we insert the store at the beginning in the
1303     // 'normal destination' BB. Otherwise we insert the store right after OutI.
1304     if (auto *InvokeI = dyn_cast<InvokeInst>(OutI))
1305       InsertPt = InvokeI->getNormalDest()->getFirstInsertionPt();
1306     else if (auto *Phi = dyn_cast<PHINode>(OutI))
1307       InsertPt = Phi->getParent()->getFirstInsertionPt();
1308     else
1309       InsertPt = std::next(OutI->getIterator());
1310 
1311     Instruction *InsertBefore = &*InsertPt;
1312     assert((InsertBefore->getFunction() == newFunction ||
1313             Blocks.count(InsertBefore->getParent())) &&
1314            "InsertPt should be in new function");
1315     assert(OAI != newFunction->arg_end() &&
1316            "Number of output arguments should match "
1317            "the amount of defined values");
1318     if (AggregateArgs) {
1319       Value *Idx[2];
1320       Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context));
1321       Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), FirstOut + i);
1322       GetElementPtrInst *GEP = GetElementPtrInst::Create(
1323           StructArgTy, &*OAI, Idx, "gep_" + outputs[i]->getName(),
1324           InsertBefore);
1325       new StoreInst(outputs[i], GEP, InsertBefore);
1326       // Since there should be only one struct argument aggregating
1327       // all the output values, we shouldn't increment OAI, which always
1328       // points to the struct argument, in this case.
1329     } else {
1330       new StoreInst(outputs[i], &*OAI, InsertBefore);
1331       ++OAI;
1332     }
1333   }
1334 
1335   // Now that we've done the deed, simplify the switch instruction.
1336   Type *OldFnRetTy = TheSwitch->getParent()->getParent()->getReturnType();
1337   switch (NumExitBlocks) {
1338   case 0:
1339     // There are no successors (the block containing the switch itself), which
1340     // means that previously this was the last part of the function, and hence
1341     // this should be rewritten as a `ret'
1342 
1343     // Check if the function should return a value
1344     if (OldFnRetTy->isVoidTy()) {
1345       ReturnInst::Create(Context, nullptr, TheSwitch);  // Return void
1346     } else if (OldFnRetTy == TheSwitch->getCondition()->getType()) {
1347       // return what we have
1348       ReturnInst::Create(Context, TheSwitch->getCondition(), TheSwitch);
1349     } else {
1350       // Otherwise we must have code extracted an unwind or something, just
1351       // return whatever we want.
1352       ReturnInst::Create(Context,
1353                          Constant::getNullValue(OldFnRetTy), TheSwitch);
1354     }
1355 
1356     TheSwitch->eraseFromParent();
1357     break;
1358   case 1:
1359     // Only a single destination, change the switch into an unconditional
1360     // branch.
1361     BranchInst::Create(TheSwitch->getSuccessor(1), TheSwitch);
1362     TheSwitch->eraseFromParent();
1363     break;
1364   case 2:
1365     BranchInst::Create(TheSwitch->getSuccessor(1), TheSwitch->getSuccessor(2),
1366                        call, TheSwitch);
1367     TheSwitch->eraseFromParent();
1368     break;
1369   default:
1370     // Otherwise, make the default destination of the switch instruction be one
1371     // of the other successors.
1372     TheSwitch->setCondition(call);
1373     TheSwitch->setDefaultDest(TheSwitch->getSuccessor(NumExitBlocks));
1374     // Remove redundant case
1375     TheSwitch->removeCase(SwitchInst::CaseIt(TheSwitch, NumExitBlocks-1));
1376     break;
1377   }
1378 
1379   // Insert lifetime markers around the reloads of any output values. The
1380   // allocas output values are stored in are only in-use in the codeRepl block.
1381   insertLifetimeMarkersSurroundingCall(M, ReloadOutputs, ReloadOutputs, call);
1382 
1383   return call;
1384 }
1385 
moveCodeToFunction(Function * newFunction)1386 void CodeExtractor::moveCodeToFunction(Function *newFunction) {
1387   Function *oldFunc = (*Blocks.begin())->getParent();
1388   Function::BasicBlockListType &oldBlocks = oldFunc->getBasicBlockList();
1389   Function::BasicBlockListType &newBlocks = newFunction->getBasicBlockList();
1390 
1391   for (BasicBlock *Block : Blocks) {
1392     // Delete the basic block from the old function, and the list of blocks
1393     oldBlocks.remove(Block);
1394 
1395     // Insert this basic block into the new function
1396     newBlocks.push_back(Block);
1397   }
1398 }
1399 
calculateNewCallTerminatorWeights(BasicBlock * CodeReplacer,DenseMap<BasicBlock *,BlockFrequency> & ExitWeights,BranchProbabilityInfo * BPI)1400 void CodeExtractor::calculateNewCallTerminatorWeights(
1401     BasicBlock *CodeReplacer,
1402     DenseMap<BasicBlock *, BlockFrequency> &ExitWeights,
1403     BranchProbabilityInfo *BPI) {
1404   using Distribution = BlockFrequencyInfoImplBase::Distribution;
1405   using BlockNode = BlockFrequencyInfoImplBase::BlockNode;
1406 
1407   // Update the branch weights for the exit block.
1408   Instruction *TI = CodeReplacer->getTerminator();
1409   SmallVector<unsigned, 8> BranchWeights(TI->getNumSuccessors(), 0);
1410 
1411   // Block Frequency distribution with dummy node.
1412   Distribution BranchDist;
1413 
1414   SmallVector<BranchProbability, 4> EdgeProbabilities(
1415       TI->getNumSuccessors(), BranchProbability::getUnknown());
1416 
1417   // Add each of the frequencies of the successors.
1418   for (unsigned i = 0, e = TI->getNumSuccessors(); i < e; ++i) {
1419     BlockNode ExitNode(i);
1420     uint64_t ExitFreq = ExitWeights[TI->getSuccessor(i)].getFrequency();
1421     if (ExitFreq != 0)
1422       BranchDist.addExit(ExitNode, ExitFreq);
1423     else
1424       EdgeProbabilities[i] = BranchProbability::getZero();
1425   }
1426 
1427   // Check for no total weight.
1428   if (BranchDist.Total == 0) {
1429     BPI->setEdgeProbability(CodeReplacer, EdgeProbabilities);
1430     return;
1431   }
1432 
1433   // Normalize the distribution so that they can fit in unsigned.
1434   BranchDist.normalize();
1435 
1436   // Create normalized branch weights and set the metadata.
1437   for (unsigned I = 0, E = BranchDist.Weights.size(); I < E; ++I) {
1438     const auto &Weight = BranchDist.Weights[I];
1439 
1440     // Get the weight and update the current BFI.
1441     BranchWeights[Weight.TargetNode.Index] = Weight.Amount;
1442     BranchProbability BP(Weight.Amount, BranchDist.Total);
1443     EdgeProbabilities[Weight.TargetNode.Index] = BP;
1444   }
1445   BPI->setEdgeProbability(CodeReplacer, EdgeProbabilities);
1446   TI->setMetadata(
1447       LLVMContext::MD_prof,
1448       MDBuilder(TI->getContext()).createBranchWeights(BranchWeights));
1449 }
1450 
1451 /// Erase debug info intrinsics which refer to values in \p F but aren't in
1452 /// \p F.
eraseDebugIntrinsicsWithNonLocalRefs(Function & F)1453 static void eraseDebugIntrinsicsWithNonLocalRefs(Function &F) {
1454   for (Instruction &I : instructions(F)) {
1455     SmallVector<DbgVariableIntrinsic *, 4> DbgUsers;
1456     findDbgUsers(DbgUsers, &I);
1457     for (DbgVariableIntrinsic *DVI : DbgUsers)
1458       if (DVI->getFunction() != &F)
1459         DVI->eraseFromParent();
1460   }
1461 }
1462 
1463 /// Fix up the debug info in the old and new functions by pointing line
1464 /// locations and debug intrinsics to the new subprogram scope, and by deleting
1465 /// intrinsics which point to values outside of the new function.
fixupDebugInfoPostExtraction(Function & OldFunc,Function & NewFunc,CallInst & TheCall)1466 static void fixupDebugInfoPostExtraction(Function &OldFunc, Function &NewFunc,
1467                                          CallInst &TheCall) {
1468   DISubprogram *OldSP = OldFunc.getSubprogram();
1469   LLVMContext &Ctx = OldFunc.getContext();
1470 
1471   if (!OldSP) {
1472     // Erase any debug info the new function contains.
1473     stripDebugInfo(NewFunc);
1474     // Make sure the old function doesn't contain any non-local metadata refs.
1475     eraseDebugIntrinsicsWithNonLocalRefs(NewFunc);
1476     return;
1477   }
1478 
1479   // Create a subprogram for the new function. Leave out a description of the
1480   // function arguments, as the parameters don't correspond to anything at the
1481   // source level.
1482   assert(OldSP->getUnit() && "Missing compile unit for subprogram");
1483   DIBuilder DIB(*OldFunc.getParent(), /*AllowUnresolved=*/false,
1484                 OldSP->getUnit());
1485   auto SPType = DIB.createSubroutineType(DIB.getOrCreateTypeArray(None));
1486   DISubprogram::DISPFlags SPFlags = DISubprogram::SPFlagDefinition |
1487                                     DISubprogram::SPFlagOptimized |
1488                                     DISubprogram::SPFlagLocalToUnit;
1489   auto NewSP = DIB.createFunction(
1490       OldSP->getUnit(), NewFunc.getName(), NewFunc.getName(), OldSP->getFile(),
1491       /*LineNo=*/0, SPType, /*ScopeLine=*/0, DINode::FlagZero, SPFlags);
1492   NewFunc.setSubprogram(NewSP);
1493 
1494   // Debug intrinsics in the new function need to be updated in one of two
1495   // ways:
1496   //  1) They need to be deleted, because they describe a value in the old
1497   //     function.
1498   //  2) They need to point to fresh metadata, e.g. because they currently
1499   //     point to a variable in the wrong scope.
1500   SmallDenseMap<DINode *, DINode *> RemappedMetadata;
1501   SmallVector<Instruction *, 4> DebugIntrinsicsToDelete;
1502   for (Instruction &I : instructions(NewFunc)) {
1503     auto *DII = dyn_cast<DbgInfoIntrinsic>(&I);
1504     if (!DII)
1505       continue;
1506 
1507     // Point the intrinsic to a fresh label within the new function.
1508     if (auto *DLI = dyn_cast<DbgLabelInst>(&I)) {
1509       DILabel *OldLabel = DLI->getLabel();
1510       DINode *&NewLabel = RemappedMetadata[OldLabel];
1511       if (!NewLabel)
1512         NewLabel = DILabel::get(Ctx, NewSP, OldLabel->getName(),
1513                                 OldLabel->getFile(), OldLabel->getLine());
1514       DLI->setArgOperand(0, MetadataAsValue::get(Ctx, NewLabel));
1515       continue;
1516     }
1517 
1518     auto IsInvalidLocation = [&NewFunc](Value *Location) {
1519       // Location is invalid if it isn't a constant or an instruction, or is an
1520       // instruction but isn't in the new function.
1521       if (!Location ||
1522           (!isa<Constant>(Location) && !isa<Instruction>(Location)))
1523         return true;
1524       Instruction *LocationInst = dyn_cast<Instruction>(Location);
1525       return LocationInst && LocationInst->getFunction() != &NewFunc;
1526     };
1527 
1528     auto *DVI = cast<DbgVariableIntrinsic>(DII);
1529     // If any of the used locations are invalid, delete the intrinsic.
1530     if (any_of(DVI->location_ops(), IsInvalidLocation)) {
1531       DebugIntrinsicsToDelete.push_back(DVI);
1532       continue;
1533     }
1534 
1535     // Point the intrinsic to a fresh variable within the new function.
1536     DILocalVariable *OldVar = DVI->getVariable();
1537     DINode *&NewVar = RemappedMetadata[OldVar];
1538     if (!NewVar)
1539       NewVar = DIB.createAutoVariable(
1540           NewSP, OldVar->getName(), OldVar->getFile(), OldVar->getLine(),
1541           OldVar->getType(), /*AlwaysPreserve=*/false, DINode::FlagZero,
1542           OldVar->getAlignInBits());
1543     DVI->setVariable(cast<DILocalVariable>(NewVar));
1544   }
1545   for (auto *DII : DebugIntrinsicsToDelete)
1546     DII->eraseFromParent();
1547   DIB.finalizeSubprogram(NewSP);
1548 
1549   // Fix up the scope information attached to the line locations in the new
1550   // function.
1551   for (Instruction &I : instructions(NewFunc)) {
1552     if (const DebugLoc &DL = I.getDebugLoc())
1553       I.setDebugLoc(DILocation::get(Ctx, DL.getLine(), DL.getCol(), NewSP));
1554 
1555     // Loop info metadata may contain line locations. Fix them up.
1556     auto updateLoopInfoLoc = [&Ctx, NewSP](Metadata *MD) -> Metadata * {
1557       if (auto *Loc = dyn_cast_or_null<DILocation>(MD))
1558         return DILocation::get(Ctx, Loc->getLine(), Loc->getColumn(), NewSP,
1559                                nullptr);
1560       return MD;
1561     };
1562     updateLoopMetadataDebugLocations(I, updateLoopInfoLoc);
1563   }
1564   if (!TheCall.getDebugLoc())
1565     TheCall.setDebugLoc(DILocation::get(Ctx, 0, 0, OldSP));
1566 
1567   eraseDebugIntrinsicsWithNonLocalRefs(NewFunc);
1568 }
1569 
1570 Function *
extractCodeRegion(const CodeExtractorAnalysisCache & CEAC)1571 CodeExtractor::extractCodeRegion(const CodeExtractorAnalysisCache &CEAC) {
1572   if (!isEligible())
1573     return nullptr;
1574 
1575   // Assumption: this is a single-entry code region, and the header is the first
1576   // block in the region.
1577   BasicBlock *header = *Blocks.begin();
1578   Function *oldFunction = header->getParent();
1579 
1580   // Calculate the entry frequency of the new function before we change the root
1581   //   block.
1582   BlockFrequency EntryFreq;
1583   if (BFI) {
1584     assert(BPI && "Both BPI and BFI are required to preserve profile info");
1585     for (BasicBlock *Pred : predecessors(header)) {
1586       if (Blocks.count(Pred))
1587         continue;
1588       EntryFreq +=
1589           BFI->getBlockFreq(Pred) * BPI->getEdgeProbability(Pred, header);
1590     }
1591   }
1592 
1593   // Remove @llvm.assume calls that will be moved to the new function from the
1594   // old function's assumption cache.
1595   for (BasicBlock *Block : Blocks) {
1596     for (auto It = Block->begin(), End = Block->end(); It != End;) {
1597       Instruction *I = &*It;
1598       ++It;
1599 
1600       if (auto *AI = dyn_cast<AssumeInst>(I)) {
1601         if (AC)
1602           AC->unregisterAssumption(AI);
1603         AI->eraseFromParent();
1604       }
1605     }
1606   }
1607 
1608   // If we have any return instructions in the region, split those blocks so
1609   // that the return is not in the region.
1610   splitReturnBlocks();
1611 
1612   // Calculate the exit blocks for the extracted region and the total exit
1613   // weights for each of those blocks.
1614   DenseMap<BasicBlock *, BlockFrequency> ExitWeights;
1615   SmallPtrSet<BasicBlock *, 1> ExitBlocks;
1616   for (BasicBlock *Block : Blocks) {
1617     for (BasicBlock *Succ : successors(Block)) {
1618       if (!Blocks.count(Succ)) {
1619         // Update the branch weight for this successor.
1620         if (BFI) {
1621           BlockFrequency &BF = ExitWeights[Succ];
1622           BF += BFI->getBlockFreq(Block) * BPI->getEdgeProbability(Block, Succ);
1623         }
1624         ExitBlocks.insert(Succ);
1625       }
1626     }
1627   }
1628   NumExitBlocks = ExitBlocks.size();
1629 
1630   // If we have to split PHI nodes of the entry or exit blocks, do so now.
1631   severSplitPHINodesOfEntry(header);
1632   severSplitPHINodesOfExits(ExitBlocks);
1633 
1634   // This takes place of the original loop
1635   BasicBlock *codeReplacer = BasicBlock::Create(header->getContext(),
1636                                                 "codeRepl", oldFunction,
1637                                                 header);
1638 
1639   // The new function needs a root node because other nodes can branch to the
1640   // head of the region, but the entry node of a function cannot have preds.
1641   BasicBlock *newFuncRoot = BasicBlock::Create(header->getContext(),
1642                                                "newFuncRoot");
1643   auto *BranchI = BranchInst::Create(header);
1644   // If the original function has debug info, we have to add a debug location
1645   // to the new branch instruction from the artificial entry block.
1646   // We use the debug location of the first instruction in the extracted
1647   // blocks, as there is no other equivalent line in the source code.
1648   if (oldFunction->getSubprogram()) {
1649     any_of(Blocks, [&BranchI](const BasicBlock *BB) {
1650       return any_of(*BB, [&BranchI](const Instruction &I) {
1651         if (!I.getDebugLoc())
1652           return false;
1653         BranchI->setDebugLoc(I.getDebugLoc());
1654         return true;
1655       });
1656     });
1657   }
1658   newFuncRoot->getInstList().push_back(BranchI);
1659 
1660   ValueSet inputs, outputs, SinkingCands, HoistingCands;
1661   BasicBlock *CommonExit = nullptr;
1662   findAllocas(CEAC, SinkingCands, HoistingCands, CommonExit);
1663   assert(HoistingCands.empty() || CommonExit);
1664 
1665   // Find inputs to, outputs from the code region.
1666   findInputsOutputs(inputs, outputs, SinkingCands);
1667 
1668   // Now sink all instructions which only have non-phi uses inside the region.
1669   // Group the allocas at the start of the block, so that any bitcast uses of
1670   // the allocas are well-defined.
1671   AllocaInst *FirstSunkAlloca = nullptr;
1672   for (auto *II : SinkingCands) {
1673     if (auto *AI = dyn_cast<AllocaInst>(II)) {
1674       AI->moveBefore(*newFuncRoot, newFuncRoot->getFirstInsertionPt());
1675       if (!FirstSunkAlloca)
1676         FirstSunkAlloca = AI;
1677     }
1678   }
1679   assert((SinkingCands.empty() || FirstSunkAlloca) &&
1680          "Did not expect a sink candidate without any allocas");
1681   for (auto *II : SinkingCands) {
1682     if (!isa<AllocaInst>(II)) {
1683       cast<Instruction>(II)->moveAfter(FirstSunkAlloca);
1684     }
1685   }
1686 
1687   if (!HoistingCands.empty()) {
1688     auto *HoistToBlock = findOrCreateBlockForHoisting(CommonExit);
1689     Instruction *TI = HoistToBlock->getTerminator();
1690     for (auto *II : HoistingCands)
1691       cast<Instruction>(II)->moveBefore(TI);
1692   }
1693 
1694   // Collect objects which are inputs to the extraction region and also
1695   // referenced by lifetime start markers within it. The effects of these
1696   // markers must be replicated in the calling function to prevent the stack
1697   // coloring pass from merging slots which store input objects.
1698   ValueSet LifetimesStart;
1699   eraseLifetimeMarkersOnInputs(Blocks, SinkingCands, LifetimesStart);
1700 
1701   // Construct new function based on inputs/outputs & add allocas for all defs.
1702   Function *newFunction =
1703       constructFunction(inputs, outputs, header, newFuncRoot, codeReplacer,
1704                         oldFunction, oldFunction->getParent());
1705 
1706   // Update the entry count of the function.
1707   if (BFI) {
1708     auto Count = BFI->getProfileCountFromFreq(EntryFreq.getFrequency());
1709     if (Count.hasValue())
1710       newFunction->setEntryCount(
1711           ProfileCount(Count.getValue(), Function::PCT_Real)); // FIXME
1712     BFI->setBlockFreq(codeReplacer, EntryFreq.getFrequency());
1713   }
1714 
1715   CallInst *TheCall =
1716       emitCallAndSwitchStatement(newFunction, codeReplacer, inputs, outputs);
1717 
1718   moveCodeToFunction(newFunction);
1719 
1720   // Replicate the effects of any lifetime start/end markers which referenced
1721   // input objects in the extraction region by placing markers around the call.
1722   insertLifetimeMarkersSurroundingCall(
1723       oldFunction->getParent(), LifetimesStart.getArrayRef(), {}, TheCall);
1724 
1725   // Propagate personality info to the new function if there is one.
1726   if (oldFunction->hasPersonalityFn())
1727     newFunction->setPersonalityFn(oldFunction->getPersonalityFn());
1728 
1729   // Update the branch weights for the exit block.
1730   if (BFI && NumExitBlocks > 1)
1731     calculateNewCallTerminatorWeights(codeReplacer, ExitWeights, BPI);
1732 
1733   // Loop over all of the PHI nodes in the header and exit blocks, and change
1734   // any references to the old incoming edge to be the new incoming edge.
1735   for (BasicBlock::iterator I = header->begin(); isa<PHINode>(I); ++I) {
1736     PHINode *PN = cast<PHINode>(I);
1737     for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
1738       if (!Blocks.count(PN->getIncomingBlock(i)))
1739         PN->setIncomingBlock(i, newFuncRoot);
1740   }
1741 
1742   for (BasicBlock *ExitBB : ExitBlocks)
1743     for (PHINode &PN : ExitBB->phis()) {
1744       Value *IncomingCodeReplacerVal = nullptr;
1745       for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) {
1746         // Ignore incoming values from outside of the extracted region.
1747         if (!Blocks.count(PN.getIncomingBlock(i)))
1748           continue;
1749 
1750         // Ensure that there is only one incoming value from codeReplacer.
1751         if (!IncomingCodeReplacerVal) {
1752           PN.setIncomingBlock(i, codeReplacer);
1753           IncomingCodeReplacerVal = PN.getIncomingValue(i);
1754         } else
1755           assert(IncomingCodeReplacerVal == PN.getIncomingValue(i) &&
1756                  "PHI has two incompatbile incoming values from codeRepl");
1757       }
1758     }
1759 
1760   fixupDebugInfoPostExtraction(*oldFunction, *newFunction, *TheCall);
1761 
1762   // Mark the new function `noreturn` if applicable. Terminators which resume
1763   // exception propagation are treated as returning instructions. This is to
1764   // avoid inserting traps after calls to outlined functions which unwind.
1765   bool doesNotReturn = none_of(*newFunction, [](const BasicBlock &BB) {
1766     const Instruction *Term = BB.getTerminator();
1767     return isa<ReturnInst>(Term) || isa<ResumeInst>(Term);
1768   });
1769   if (doesNotReturn)
1770     newFunction->setDoesNotReturn();
1771 
1772   LLVM_DEBUG(if (verifyFunction(*newFunction, &errs())) {
1773     newFunction->dump();
1774     report_fatal_error("verification of newFunction failed!");
1775   });
1776   LLVM_DEBUG(if (verifyFunction(*oldFunction))
1777              report_fatal_error("verification of oldFunction failed!"));
1778   LLVM_DEBUG(if (AC && verifyAssumptionCache(*oldFunction, *newFunction, AC))
1779                  report_fatal_error("Stale Asumption cache for old Function!"));
1780   return newFunction;
1781 }
1782 
verifyAssumptionCache(const Function & OldFunc,const Function & NewFunc,AssumptionCache * AC)1783 bool CodeExtractor::verifyAssumptionCache(const Function &OldFunc,
1784                                           const Function &NewFunc,
1785                                           AssumptionCache *AC) {
1786   for (auto AssumeVH : AC->assumptions()) {
1787     auto *I = dyn_cast_or_null<CallInst>(AssumeVH);
1788     if (!I)
1789       continue;
1790 
1791     // There shouldn't be any llvm.assume intrinsics in the new function.
1792     if (I->getFunction() != &OldFunc)
1793       return true;
1794 
1795     // There shouldn't be any stale affected values in the assumption cache
1796     // that were previously in the old function, but that have now been moved
1797     // to the new function.
1798     for (auto AffectedValVH : AC->assumptionsFor(I->getOperand(0))) {
1799       auto *AffectedCI = dyn_cast_or_null<CallInst>(AffectedValVH);
1800       if (!AffectedCI)
1801         continue;
1802       if (AffectedCI->getFunction() != &OldFunc)
1803         return true;
1804       auto *AssumedInst = cast<Instruction>(AffectedCI->getOperand(0));
1805       if (AssumedInst->getFunction() != &OldFunc)
1806         return true;
1807     }
1808   }
1809   return false;
1810 }
1811