1 //===- InlineFunction.cpp - Code to perform function inlining -------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements inlining of a function into a call site, resolving
11 // parameters and the return value as appropriate.
13 //===----------------------------------------------------------------------===//
15 #include "llvm/ADT/DenseMap.h"
16 #include "llvm/ADT/None.h"
17 #include "llvm/ADT/Optional.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/ADT/StringExtras.h"
23 #include "llvm/ADT/iterator_range.h"
24 #include "llvm/Analysis/AliasAnalysis.h"
25 #include "llvm/Analysis/AssumptionCache.h"
26 #include "llvm/Analysis/BlockFrequencyInfo.h"
27 #include "llvm/Analysis/CallGraph.h"
28 #include "llvm/Analysis/CaptureTracking.h"
29 #include "llvm/Analysis/EHPersonalities.h"
30 #include "llvm/Analysis/InstructionSimplify.h"
31 #include "llvm/Analysis/ProfileSummaryInfo.h"
32 #include "llvm/Transforms/Utils/Local.h"
33 #include "llvm/Analysis/ValueTracking.h"
34 #include "llvm/Analysis/VectorUtils.h"
35 #include "llvm/IR/Argument.h"
36 #include "llvm/IR/BasicBlock.h"
37 #include "llvm/IR/CFG.h"
38 #include "llvm/IR/CallSite.h"
39 #include "llvm/IR/Constant.h"
40 #include "llvm/IR/Constants.h"
41 #include "llvm/IR/DIBuilder.h"
42 #include "llvm/IR/DataLayout.h"
43 #include "llvm/IR/DebugInfoMetadata.h"
44 #include "llvm/IR/DebugLoc.h"
45 #include "llvm/IR/DerivedTypes.h"
46 #include "llvm/IR/Dominators.h"
47 #include "llvm/IR/Function.h"
48 #include "llvm/IR/IRBuilder.h"
49 #include "llvm/IR/InstrTypes.h"
50 #include "llvm/IR/Instruction.h"
51 #include "llvm/IR/Instructions.h"
52 #include "llvm/IR/IntrinsicInst.h"
53 #include "llvm/IR/Intrinsics.h"
54 #include "llvm/IR/LLVMContext.h"
55 #include "llvm/IR/MDBuilder.h"
56 #include "llvm/IR/Metadata.h"
57 #include "llvm/IR/Module.h"
58 #include "llvm/IR/Type.h"
59 #include "llvm/IR/User.h"
60 #include "llvm/IR/Value.h"
61 #include "llvm/Support/Casting.h"
62 #include "llvm/Support/CommandLine.h"
63 #include "llvm/Support/ErrorHandling.h"
64 #include "llvm/Transforms/Utils/Cloning.h"
65 #include "llvm/Transforms/Utils/ValueMapper.h"
76 using ProfileCount = Function::ProfileCount;
79 EnableNoAliasConversion("enable-noalias-to-md-conversion", cl::init(true),
81 cl::desc("Convert noalias attributes to metadata during inlining."));
84 PreserveAlignmentAssumptions("preserve-alignment-assumptions-during-inlining",
85 cl::init(true), cl::Hidden,
86 cl::desc("Convert align attributes to assumptions during inlining."));
88 llvm::InlineResult llvm::InlineFunction(CallInst *CI, InlineFunctionInfo &IFI,
90 bool InsertLifetime) {
91 return InlineFunction(CallSite(CI), IFI, CalleeAAR, InsertLifetime);
94 llvm::InlineResult llvm::InlineFunction(InvokeInst *II, InlineFunctionInfo &IFI,
96 bool InsertLifetime) {
97 return InlineFunction(CallSite(II), IFI, CalleeAAR, InsertLifetime);
102 /// A class for recording information about inlining a landing pad.
103 class LandingPadInliningInfo {
104 /// Destination of the invoke's unwind.
105 BasicBlock *OuterResumeDest;
107 /// Destination for the callee's resume.
108 BasicBlock *InnerResumeDest = nullptr;
110 /// LandingPadInst associated with the invoke.
111 LandingPadInst *CallerLPad = nullptr;
113 /// PHI for EH values from landingpad insts.
114 PHINode *InnerEHValuesPHI = nullptr;
116 SmallVector<Value*, 8> UnwindDestPHIValues;
119 LandingPadInliningInfo(InvokeInst *II)
120 : OuterResumeDest(II->getUnwindDest()) {
121 // If there are PHI nodes in the unwind destination block, we need to keep
122 // track of which values came into them from the invoke before removing
123 // the edge from this block.
124 BasicBlock *InvokeBB = II->getParent();
125 BasicBlock::iterator I = OuterResumeDest->begin();
126 for (; isa<PHINode>(I); ++I) {
127 // Save the value to use for this edge.
128 PHINode *PHI = cast<PHINode>(I);
129 UnwindDestPHIValues.push_back(PHI->getIncomingValueForBlock(InvokeBB));
132 CallerLPad = cast<LandingPadInst>(I);
135 /// The outer unwind destination is the target of
136 /// unwind edges introduced for calls within the inlined function.
137 BasicBlock *getOuterResumeDest() const {
138 return OuterResumeDest;
141 BasicBlock *getInnerResumeDest();
143 LandingPadInst *getLandingPadInst() const { return CallerLPad; }
145 /// Forward the 'resume' instruction to the caller's landing pad block.
146 /// When the landing pad block has only one predecessor, this is
147 /// a simple branch. When there is more than one predecessor, we need to
148 /// split the landing pad block after the landingpad instruction and jump
150 void forwardResume(ResumeInst *RI,
151 SmallPtrSetImpl<LandingPadInst*> &InlinedLPads);
153 /// Add incoming-PHI values to the unwind destination block for the given
154 /// basic block, using the values for the original invoke's source block.
155 void addIncomingPHIValuesFor(BasicBlock *BB) const {
156 addIncomingPHIValuesForInto(BB, OuterResumeDest);
159 void addIncomingPHIValuesForInto(BasicBlock *src, BasicBlock *dest) const {
160 BasicBlock::iterator I = dest->begin();
161 for (unsigned i = 0, e = UnwindDestPHIValues.size(); i != e; ++i, ++I) {
162 PHINode *phi = cast<PHINode>(I);
163 phi->addIncoming(UnwindDestPHIValues[i], src);
168 } // end anonymous namespace
170 /// Get or create a target for the branch from ResumeInsts.
171 BasicBlock *LandingPadInliningInfo::getInnerResumeDest() {
172 if (InnerResumeDest) return InnerResumeDest;
174 // Split the landing pad.
175 BasicBlock::iterator SplitPoint = ++CallerLPad->getIterator();
177 OuterResumeDest->splitBasicBlock(SplitPoint,
178 OuterResumeDest->getName() + ".body");
180 // The number of incoming edges we expect to the inner landing pad.
181 const unsigned PHICapacity = 2;
183 // Create corresponding new PHIs for all the PHIs in the outer landing pad.
184 Instruction *InsertPoint = &InnerResumeDest->front();
185 BasicBlock::iterator I = OuterResumeDest->begin();
186 for (unsigned i = 0, e = UnwindDestPHIValues.size(); i != e; ++i, ++I) {
187 PHINode *OuterPHI = cast<PHINode>(I);
188 PHINode *InnerPHI = PHINode::Create(OuterPHI->getType(), PHICapacity,
189 OuterPHI->getName() + ".lpad-body",
191 OuterPHI->replaceAllUsesWith(InnerPHI);
192 InnerPHI->addIncoming(OuterPHI, OuterResumeDest);
195 // Create a PHI for the exception values.
196 InnerEHValuesPHI = PHINode::Create(CallerLPad->getType(), PHICapacity,
197 "eh.lpad-body", InsertPoint);
198 CallerLPad->replaceAllUsesWith(InnerEHValuesPHI);
199 InnerEHValuesPHI->addIncoming(CallerLPad, OuterResumeDest);
202 return InnerResumeDest;
205 /// Forward the 'resume' instruction to the caller's landing pad block.
206 /// When the landing pad block has only one predecessor, this is a simple
207 /// branch. When there is more than one predecessor, we need to split the
208 /// landing pad block after the landingpad instruction and jump to there.
209 void LandingPadInliningInfo::forwardResume(
210 ResumeInst *RI, SmallPtrSetImpl<LandingPadInst *> &InlinedLPads) {
211 BasicBlock *Dest = getInnerResumeDest();
212 BasicBlock *Src = RI->getParent();
214 BranchInst::Create(Dest, Src);
216 // Update the PHIs in the destination. They were inserted in an order which
218 addIncomingPHIValuesForInto(Src, Dest);
220 InnerEHValuesPHI->addIncoming(RI->getOperand(0), Src);
221 RI->eraseFromParent();
224 /// Helper for getUnwindDestToken/getUnwindDestTokenHelper.
225 static Value *getParentPad(Value *EHPad) {
226 if (auto *FPI = dyn_cast<FuncletPadInst>(EHPad))
227 return FPI->getParentPad();
228 return cast<CatchSwitchInst>(EHPad)->getParentPad();
231 using UnwindDestMemoTy = DenseMap<Instruction *, Value *>;
233 /// Helper for getUnwindDestToken that does the descendant-ward part of
235 static Value *getUnwindDestTokenHelper(Instruction *EHPad,
236 UnwindDestMemoTy &MemoMap) {
237 SmallVector<Instruction *, 8> Worklist(1, EHPad);
239 while (!Worklist.empty()) {
240 Instruction *CurrentPad = Worklist.pop_back_val();
241 // We only put pads on the worklist that aren't in the MemoMap. When
242 // we find an unwind dest for a pad we may update its ancestors, but
243 // the queue only ever contains uncles/great-uncles/etc. of CurrentPad,
244 // so they should never get updated while queued on the worklist.
245 assert(!MemoMap.count(CurrentPad));
246 Value *UnwindDestToken = nullptr;
247 if (auto *CatchSwitch = dyn_cast<CatchSwitchInst>(CurrentPad)) {
248 if (CatchSwitch->hasUnwindDest()) {
249 UnwindDestToken = CatchSwitch->getUnwindDest()->getFirstNonPHI();
251 // Catchswitch doesn't have a 'nounwind' variant, and one might be
252 // annotated as "unwinds to caller" when really it's nounwind (see
253 // e.g. SimplifyCFGOpt::SimplifyUnreachable), so we can't infer the
254 // parent's unwind dest from this. We can check its catchpads'
255 // descendants, since they might include a cleanuppad with an
256 // "unwinds to caller" cleanupret, which can be trusted.
257 for (auto HI = CatchSwitch->handler_begin(),
258 HE = CatchSwitch->handler_end();
259 HI != HE && !UnwindDestToken; ++HI) {
260 BasicBlock *HandlerBlock = *HI;
261 auto *CatchPad = cast<CatchPadInst>(HandlerBlock->getFirstNonPHI());
262 for (User *Child : CatchPad->users()) {
263 // Intentionally ignore invokes here -- since the catchswitch is
264 // marked "unwind to caller", it would be a verifier error if it
265 // contained an invoke which unwinds out of it, so any invoke we'd
266 // encounter must unwind to some child of the catch.
267 if (!isa<CleanupPadInst>(Child) && !isa<CatchSwitchInst>(Child))
270 Instruction *ChildPad = cast<Instruction>(Child);
271 auto Memo = MemoMap.find(ChildPad);
272 if (Memo == MemoMap.end()) {
273 // Haven't figured out this child pad yet; queue it.
274 Worklist.push_back(ChildPad);
277 // We've already checked this child, but might have found that
278 // it offers no proof either way.
279 Value *ChildUnwindDestToken = Memo->second;
280 if (!ChildUnwindDestToken)
282 // We already know the child's unwind dest, which can either
283 // be ConstantTokenNone to indicate unwind to caller, or can
284 // be another child of the catchpad. Only the former indicates
285 // the unwind dest of the catchswitch.
286 if (isa<ConstantTokenNone>(ChildUnwindDestToken)) {
287 UnwindDestToken = ChildUnwindDestToken;
290 assert(getParentPad(ChildUnwindDestToken) == CatchPad);
295 auto *CleanupPad = cast<CleanupPadInst>(CurrentPad);
296 for (User *U : CleanupPad->users()) {
297 if (auto *CleanupRet = dyn_cast<CleanupReturnInst>(U)) {
298 if (BasicBlock *RetUnwindDest = CleanupRet->getUnwindDest())
299 UnwindDestToken = RetUnwindDest->getFirstNonPHI();
301 UnwindDestToken = ConstantTokenNone::get(CleanupPad->getContext());
304 Value *ChildUnwindDestToken;
305 if (auto *Invoke = dyn_cast<InvokeInst>(U)) {
306 ChildUnwindDestToken = Invoke->getUnwindDest()->getFirstNonPHI();
307 } else if (isa<CleanupPadInst>(U) || isa<CatchSwitchInst>(U)) {
308 Instruction *ChildPad = cast<Instruction>(U);
309 auto Memo = MemoMap.find(ChildPad);
310 if (Memo == MemoMap.end()) {
311 // Haven't resolved this child yet; queue it and keep searching.
312 Worklist.push_back(ChildPad);
315 // We've checked this child, but still need to ignore it if it
316 // had no proof either way.
317 ChildUnwindDestToken = Memo->second;
318 if (!ChildUnwindDestToken)
321 // Not a relevant user of the cleanuppad
324 // In a well-formed program, the child/invoke must either unwind to
325 // an(other) child of the cleanup, or exit the cleanup. In the
326 // first case, continue searching.
327 if (isa<Instruction>(ChildUnwindDestToken) &&
328 getParentPad(ChildUnwindDestToken) == CleanupPad)
330 UnwindDestToken = ChildUnwindDestToken;
334 // If we haven't found an unwind dest for CurrentPad, we may have queued its
335 // children, so move on to the next in the worklist.
336 if (!UnwindDestToken)
339 // Now we know that CurrentPad unwinds to UnwindDestToken. It also exits
340 // any ancestors of CurrentPad up to but not including UnwindDestToken's
341 // parent pad. Record this in the memo map, and check to see if the
342 // original EHPad being queried is one of the ones exited.
344 if (auto *UnwindPad = dyn_cast<Instruction>(UnwindDestToken))
345 UnwindParent = getParentPad(UnwindPad);
347 UnwindParent = nullptr;
348 bool ExitedOriginalPad = false;
349 for (Instruction *ExitedPad = CurrentPad;
350 ExitedPad && ExitedPad != UnwindParent;
351 ExitedPad = dyn_cast<Instruction>(getParentPad(ExitedPad))) {
352 // Skip over catchpads since they just follow their catchswitches.
353 if (isa<CatchPadInst>(ExitedPad))
355 MemoMap[ExitedPad] = UnwindDestToken;
356 ExitedOriginalPad |= (ExitedPad == EHPad);
359 if (ExitedOriginalPad)
360 return UnwindDestToken;
362 // Continue the search.
365 // No definitive information is contained within this funclet.
369 /// Given an EH pad, find where it unwinds. If it unwinds to an EH pad,
370 /// return that pad instruction. If it unwinds to caller, return
371 /// ConstantTokenNone. If it does not have a definitive unwind destination,
374 /// This routine gets invoked for calls in funclets in inlinees when inlining
375 /// an invoke. Since many funclets don't have calls inside them, it's queried
376 /// on-demand rather than building a map of pads to unwind dests up front.
377 /// Determining a funclet's unwind dest may require recursively searching its
378 /// descendants, and also ancestors and cousins if the descendants don't provide
379 /// an answer. Since most funclets will have their unwind dest immediately
380 /// available as the unwind dest of a catchswitch or cleanupret, this routine
381 /// searches top-down from the given pad and then up. To avoid worst-case
382 /// quadratic run-time given that approach, it uses a memo map to avoid
383 /// re-processing funclet trees. The callers that rewrite the IR as they go
384 /// take advantage of this, for correctness, by checking/forcing rewritten
385 /// pads' entries to match the original callee view.
386 static Value *getUnwindDestToken(Instruction *EHPad,
387 UnwindDestMemoTy &MemoMap) {
388 // Catchpads unwind to the same place as their catchswitch;
389 // redirct any queries on catchpads so the code below can
390 // deal with just catchswitches and cleanuppads.
391 if (auto *CPI = dyn_cast<CatchPadInst>(EHPad))
392 EHPad = CPI->getCatchSwitch();
394 // Check if we've already determined the unwind dest for this pad.
395 auto Memo = MemoMap.find(EHPad);
396 if (Memo != MemoMap.end())
399 // Search EHPad and, if necessary, its descendants.
400 Value *UnwindDestToken = getUnwindDestTokenHelper(EHPad, MemoMap);
401 assert((UnwindDestToken == nullptr) != (MemoMap.count(EHPad) != 0));
403 return UnwindDestToken;
405 // No information is available for this EHPad from itself or any of its
406 // descendants. An unwind all the way out to a pad in the caller would
407 // need also to agree with the unwind dest of the parent funclet, so
408 // search up the chain to try to find a funclet with information. Put
409 // null entries in the memo map to avoid re-processing as we go up.
410 MemoMap[EHPad] = nullptr;
412 SmallPtrSet<Instruction *, 4> TempMemos;
413 TempMemos.insert(EHPad);
415 Instruction *LastUselessPad = EHPad;
416 Value *AncestorToken;
417 for (AncestorToken = getParentPad(EHPad);
418 auto *AncestorPad = dyn_cast<Instruction>(AncestorToken);
419 AncestorToken = getParentPad(AncestorToken)) {
420 // Skip over catchpads since they just follow their catchswitches.
421 if (isa<CatchPadInst>(AncestorPad))
423 // If the MemoMap had an entry mapping AncestorPad to nullptr, since we
424 // haven't yet called getUnwindDestTokenHelper for AncestorPad in this
425 // call to getUnwindDestToken, that would mean that AncestorPad had no
426 // information in itself, its descendants, or its ancestors. If that
427 // were the case, then we should also have recorded the lack of information
428 // for the descendant that we're coming from. So assert that we don't
429 // find a null entry in the MemoMap for AncestorPad.
430 assert(!MemoMap.count(AncestorPad) || MemoMap[AncestorPad]);
431 auto AncestorMemo = MemoMap.find(AncestorPad);
432 if (AncestorMemo == MemoMap.end()) {
433 UnwindDestToken = getUnwindDestTokenHelper(AncestorPad, MemoMap);
435 UnwindDestToken = AncestorMemo->second;
439 LastUselessPad = AncestorPad;
440 MemoMap[LastUselessPad] = nullptr;
442 TempMemos.insert(LastUselessPad);
446 // We know that getUnwindDestTokenHelper was called on LastUselessPad and
447 // returned nullptr (and likewise for EHPad and any of its ancestors up to
448 // LastUselessPad), so LastUselessPad has no information from below. Since
449 // getUnwindDestTokenHelper must investigate all downward paths through
450 // no-information nodes to prove that a node has no information like this,
451 // and since any time it finds information it records it in the MemoMap for
452 // not just the immediately-containing funclet but also any ancestors also
453 // exited, it must be the case that, walking downward from LastUselessPad,
454 // visiting just those nodes which have not been mapped to an unwind dest
455 // by getUnwindDestTokenHelper (the nullptr TempMemos notwithstanding, since
456 // they are just used to keep getUnwindDestTokenHelper from repeating work),
457 // any node visited must have been exhaustively searched with no information
459 SmallVector<Instruction *, 8> Worklist(1, LastUselessPad);
460 while (!Worklist.empty()) {
461 Instruction *UselessPad = Worklist.pop_back_val();
462 auto Memo = MemoMap.find(UselessPad);
463 if (Memo != MemoMap.end() && Memo->second) {
464 // Here the name 'UselessPad' is a bit of a misnomer, because we've found
465 // that it is a funclet that does have information about unwinding to
466 // a particular destination; its parent was a useless pad.
467 // Since its parent has no information, the unwind edge must not escape
468 // the parent, and must target a sibling of this pad. This local unwind
469 // gives us no information about EHPad. Leave it and the subtree rooted
471 assert(getParentPad(Memo->second) == getParentPad(UselessPad));
474 // We know we don't have information for UselesPad. If it has an entry in
475 // the MemoMap (mapping it to nullptr), it must be one of the TempMemos
476 // added on this invocation of getUnwindDestToken; if a previous invocation
477 // recorded nullptr, it would have had to prove that the ancestors of
478 // UselessPad, which include LastUselessPad, had no information, and that
479 // in turn would have required proving that the descendants of
480 // LastUselesPad, which include EHPad, have no information about
481 // LastUselessPad, which would imply that EHPad was mapped to nullptr in
482 // the MemoMap on that invocation, which isn't the case if we got here.
483 assert(!MemoMap.count(UselessPad) || TempMemos.count(UselessPad));
484 // Assert as we enumerate users that 'UselessPad' doesn't have any unwind
485 // information that we'd be contradicting by making a map entry for it
486 // (which is something that getUnwindDestTokenHelper must have proved for
487 // us to get here). Just assert on is direct users here; the checks in
488 // this downward walk at its descendants will verify that they don't have
489 // any unwind edges that exit 'UselessPad' either (i.e. they either have no
490 // unwind edges or unwind to a sibling).
491 MemoMap[UselessPad] = UnwindDestToken;
492 if (auto *CatchSwitch = dyn_cast<CatchSwitchInst>(UselessPad)) {
493 assert(CatchSwitch->getUnwindDest() == nullptr && "Expected useless pad");
494 for (BasicBlock *HandlerBlock : CatchSwitch->handlers()) {
495 auto *CatchPad = HandlerBlock->getFirstNonPHI();
496 for (User *U : CatchPad->users()) {
498 (!isa<InvokeInst>(U) ||
500 cast<InvokeInst>(U)->getUnwindDest()->getFirstNonPHI()) ==
502 "Expected useless pad");
503 if (isa<CatchSwitchInst>(U) || isa<CleanupPadInst>(U))
504 Worklist.push_back(cast<Instruction>(U));
508 assert(isa<CleanupPadInst>(UselessPad));
509 for (User *U : UselessPad->users()) {
510 assert(!isa<CleanupReturnInst>(U) && "Expected useless pad");
511 assert((!isa<InvokeInst>(U) ||
513 cast<InvokeInst>(U)->getUnwindDest()->getFirstNonPHI()) ==
515 "Expected useless pad");
516 if (isa<CatchSwitchInst>(U) || isa<CleanupPadInst>(U))
517 Worklist.push_back(cast<Instruction>(U));
522 return UnwindDestToken;
525 /// When we inline a basic block into an invoke,
526 /// we have to turn all of the calls that can throw into invokes.
527 /// This function analyze BB to see if there are any calls, and if so,
528 /// it rewrites them to be invokes that jump to InvokeDest and fills in the PHI
529 /// nodes in that block with the values specified in InvokeDestPHIValues.
530 static BasicBlock *HandleCallsInBlockInlinedThroughInvoke(
531 BasicBlock *BB, BasicBlock *UnwindEdge,
532 UnwindDestMemoTy *FuncletUnwindMap = nullptr) {
533 for (BasicBlock::iterator BBI = BB->begin(), E = BB->end(); BBI != E; ) {
534 Instruction *I = &*BBI++;
536 // We only need to check for function calls: inlined invoke
537 // instructions require no special handling.
538 CallInst *CI = dyn_cast<CallInst>(I);
540 if (!CI || CI->doesNotThrow() || isa<InlineAsm>(CI->getCalledValue()))
543 // We do not need to (and in fact, cannot) convert possibly throwing calls
544 // to @llvm.experimental_deoptimize (resp. @llvm.experimental.guard) into
545 // invokes. The caller's "segment" of the deoptimization continuation
546 // attached to the newly inlined @llvm.experimental_deoptimize
547 // (resp. @llvm.experimental.guard) call should contain the exception
548 // handling logic, if any.
549 if (auto *F = CI->getCalledFunction())
550 if (F->getIntrinsicID() == Intrinsic::experimental_deoptimize ||
551 F->getIntrinsicID() == Intrinsic::experimental_guard)
554 if (auto FuncletBundle = CI->getOperandBundle(LLVMContext::OB_funclet)) {
555 // This call is nested inside a funclet. If that funclet has an unwind
556 // destination within the inlinee, then unwinding out of this call would
557 // be UB. Rewriting this call to an invoke which targets the inlined
558 // invoke's unwind dest would give the call's parent funclet multiple
559 // unwind destinations, which is something that subsequent EH table
560 // generation can't handle and that the veirifer rejects. So when we
561 // see such a call, leave it as a call.
562 auto *FuncletPad = cast<Instruction>(FuncletBundle->Inputs[0]);
563 Value *UnwindDestToken =
564 getUnwindDestToken(FuncletPad, *FuncletUnwindMap);
565 if (UnwindDestToken && !isa<ConstantTokenNone>(UnwindDestToken))
568 Instruction *MemoKey;
569 if (auto *CatchPad = dyn_cast<CatchPadInst>(FuncletPad))
570 MemoKey = CatchPad->getCatchSwitch();
572 MemoKey = FuncletPad;
573 assert(FuncletUnwindMap->count(MemoKey) &&
574 (*FuncletUnwindMap)[MemoKey] == UnwindDestToken &&
575 "must get memoized to avoid confusing later searches");
579 changeToInvokeAndSplitBasicBlock(CI, UnwindEdge);
585 /// If we inlined an invoke site, we need to convert calls
586 /// in the body of the inlined function into invokes.
588 /// II is the invoke instruction being inlined. FirstNewBlock is the first
589 /// block of the inlined code (the last block is the end of the function),
590 /// and InlineCodeInfo is information about the code that got inlined.
591 static void HandleInlinedLandingPad(InvokeInst *II, BasicBlock *FirstNewBlock,
592 ClonedCodeInfo &InlinedCodeInfo) {
593 BasicBlock *InvokeDest = II->getUnwindDest();
595 Function *Caller = FirstNewBlock->getParent();
597 // The inlined code is currently at the end of the function, scan from the
598 // start of the inlined code to its end, checking for stuff we need to
600 LandingPadInliningInfo Invoke(II);
602 // Get all of the inlined landing pad instructions.
603 SmallPtrSet<LandingPadInst*, 16> InlinedLPads;
604 for (Function::iterator I = FirstNewBlock->getIterator(), E = Caller->end();
606 if (InvokeInst *II = dyn_cast<InvokeInst>(I->getTerminator()))
607 InlinedLPads.insert(II->getLandingPadInst());
609 // Append the clauses from the outer landing pad instruction into the inlined
610 // landing pad instructions.
611 LandingPadInst *OuterLPad = Invoke.getLandingPadInst();
612 for (LandingPadInst *InlinedLPad : InlinedLPads) {
613 unsigned OuterNum = OuterLPad->getNumClauses();
614 InlinedLPad->reserveClauses(OuterNum);
615 for (unsigned OuterIdx = 0; OuterIdx != OuterNum; ++OuterIdx)
616 InlinedLPad->addClause(OuterLPad->getClause(OuterIdx));
617 if (OuterLPad->isCleanup())
618 InlinedLPad->setCleanup(true);
621 for (Function::iterator BB = FirstNewBlock->getIterator(), E = Caller->end();
623 if (InlinedCodeInfo.ContainsCalls)
624 if (BasicBlock *NewBB = HandleCallsInBlockInlinedThroughInvoke(
625 &*BB, Invoke.getOuterResumeDest()))
626 // Update any PHI nodes in the exceptional block to indicate that there
627 // is now a new entry in them.
628 Invoke.addIncomingPHIValuesFor(NewBB);
630 // Forward any resumes that are remaining here.
631 if (ResumeInst *RI = dyn_cast<ResumeInst>(BB->getTerminator()))
632 Invoke.forwardResume(RI, InlinedLPads);
635 // Now that everything is happy, we have one final detail. The PHI nodes in
636 // the exception destination block still have entries due to the original
637 // invoke instruction. Eliminate these entries (which might even delete the
639 InvokeDest->removePredecessor(II->getParent());
642 /// If we inlined an invoke site, we need to convert calls
643 /// in the body of the inlined function into invokes.
645 /// II is the invoke instruction being inlined. FirstNewBlock is the first
646 /// block of the inlined code (the last block is the end of the function),
647 /// and InlineCodeInfo is information about the code that got inlined.
648 static void HandleInlinedEHPad(InvokeInst *II, BasicBlock *FirstNewBlock,
649 ClonedCodeInfo &InlinedCodeInfo) {
650 BasicBlock *UnwindDest = II->getUnwindDest();
651 Function *Caller = FirstNewBlock->getParent();
653 assert(UnwindDest->getFirstNonPHI()->isEHPad() && "unexpected BasicBlock!");
655 // If there are PHI nodes in the unwind destination block, we need to keep
656 // track of which values came into them from the invoke before removing the
657 // edge from this block.
658 SmallVector<Value *, 8> UnwindDestPHIValues;
659 BasicBlock *InvokeBB = II->getParent();
660 for (Instruction &I : *UnwindDest) {
661 // Save the value to use for this edge.
662 PHINode *PHI = dyn_cast<PHINode>(&I);
665 UnwindDestPHIValues.push_back(PHI->getIncomingValueForBlock(InvokeBB));
668 // Add incoming-PHI values to the unwind destination block for the given basic
669 // block, using the values for the original invoke's source block.
670 auto UpdatePHINodes = [&](BasicBlock *Src) {
671 BasicBlock::iterator I = UnwindDest->begin();
672 for (Value *V : UnwindDestPHIValues) {
673 PHINode *PHI = cast<PHINode>(I);
674 PHI->addIncoming(V, Src);
679 // This connects all the instructions which 'unwind to caller' to the invoke
681 UnwindDestMemoTy FuncletUnwindMap;
682 for (Function::iterator BB = FirstNewBlock->getIterator(), E = Caller->end();
684 if (auto *CRI = dyn_cast<CleanupReturnInst>(BB->getTerminator())) {
685 if (CRI->unwindsToCaller()) {
686 auto *CleanupPad = CRI->getCleanupPad();
687 CleanupReturnInst::Create(CleanupPad, UnwindDest, CRI);
688 CRI->eraseFromParent();
689 UpdatePHINodes(&*BB);
690 // Finding a cleanupret with an unwind destination would confuse
691 // subsequent calls to getUnwindDestToken, so map the cleanuppad
692 // to short-circuit any such calls and recognize this as an "unwind
693 // to caller" cleanup.
694 assert(!FuncletUnwindMap.count(CleanupPad) ||
695 isa<ConstantTokenNone>(FuncletUnwindMap[CleanupPad]));
696 FuncletUnwindMap[CleanupPad] =
697 ConstantTokenNone::get(Caller->getContext());
701 Instruction *I = BB->getFirstNonPHI();
705 Instruction *Replacement = nullptr;
706 if (auto *CatchSwitch = dyn_cast<CatchSwitchInst>(I)) {
707 if (CatchSwitch->unwindsToCaller()) {
708 Value *UnwindDestToken;
709 if (auto *ParentPad =
710 dyn_cast<Instruction>(CatchSwitch->getParentPad())) {
711 // This catchswitch is nested inside another funclet. If that
712 // funclet has an unwind destination within the inlinee, then
713 // unwinding out of this catchswitch would be UB. Rewriting this
714 // catchswitch to unwind to the inlined invoke's unwind dest would
715 // give the parent funclet multiple unwind destinations, which is
716 // something that subsequent EH table generation can't handle and
717 // that the veirifer rejects. So when we see such a call, leave it
718 // as "unwind to caller".
719 UnwindDestToken = getUnwindDestToken(ParentPad, FuncletUnwindMap);
720 if (UnwindDestToken && !isa<ConstantTokenNone>(UnwindDestToken))
723 // This catchswitch has no parent to inherit constraints from, and
724 // none of its descendants can have an unwind edge that exits it and
725 // targets another funclet in the inlinee. It may or may not have a
726 // descendant that definitively has an unwind to caller. In either
727 // case, we'll have to assume that any unwinds out of it may need to
728 // be routed to the caller, so treat it as though it has a definitive
730 UnwindDestToken = ConstantTokenNone::get(Caller->getContext());
732 auto *NewCatchSwitch = CatchSwitchInst::Create(
733 CatchSwitch->getParentPad(), UnwindDest,
734 CatchSwitch->getNumHandlers(), CatchSwitch->getName(),
736 for (BasicBlock *PadBB : CatchSwitch->handlers())
737 NewCatchSwitch->addHandler(PadBB);
738 // Propagate info for the old catchswitch over to the new one in
739 // the unwind map. This also serves to short-circuit any subsequent
740 // checks for the unwind dest of this catchswitch, which would get
741 // confused if they found the outer handler in the callee.
742 FuncletUnwindMap[NewCatchSwitch] = UnwindDestToken;
743 Replacement = NewCatchSwitch;
745 } else if (!isa<FuncletPadInst>(I)) {
746 llvm_unreachable("unexpected EHPad!");
750 Replacement->takeName(I);
751 I->replaceAllUsesWith(Replacement);
752 I->eraseFromParent();
753 UpdatePHINodes(&*BB);
757 if (InlinedCodeInfo.ContainsCalls)
758 for (Function::iterator BB = FirstNewBlock->getIterator(),
761 if (BasicBlock *NewBB = HandleCallsInBlockInlinedThroughInvoke(
762 &*BB, UnwindDest, &FuncletUnwindMap))
763 // Update any PHI nodes in the exceptional block to indicate that there
764 // is now a new entry in them.
765 UpdatePHINodes(NewBB);
767 // Now that everything is happy, we have one final detail. The PHI nodes in
768 // the exception destination block still have entries due to the original
769 // invoke instruction. Eliminate these entries (which might even delete the
771 UnwindDest->removePredecessor(InvokeBB);
774 /// When inlining a call site that has !llvm.mem.parallel_loop_access or
775 /// llvm.access.group metadata, that metadata should be propagated to all
776 /// memory-accessing cloned instructions.
777 static void PropagateParallelLoopAccessMetadata(CallSite CS,
778 ValueToValueMapTy &VMap) {
780 CS.getInstruction()->getMetadata(LLVMContext::MD_mem_parallel_loop_access);
781 MDNode *CallAccessGroup =
782 CS.getInstruction()->getMetadata(LLVMContext::MD_access_group);
783 if (!M && !CallAccessGroup)
786 for (ValueToValueMapTy::iterator VMI = VMap.begin(), VMIE = VMap.end();
787 VMI != VMIE; ++VMI) {
791 Instruction *NI = dyn_cast<Instruction>(VMI->second);
797 NI->getMetadata(LLVMContext::MD_mem_parallel_loop_access)) {
798 M = MDNode::concatenate(PM, M);
799 NI->setMetadata(LLVMContext::MD_mem_parallel_loop_access, M);
800 } else if (NI->mayReadOrWriteMemory()) {
801 NI->setMetadata(LLVMContext::MD_mem_parallel_loop_access, M);
805 if (NI->mayReadOrWriteMemory()) {
806 MDNode *UnitedAccGroups = uniteAccessGroups(
807 NI->getMetadata(LLVMContext::MD_access_group), CallAccessGroup);
808 NI->setMetadata(LLVMContext::MD_access_group, UnitedAccGroups);
813 /// When inlining a function that contains noalias scope metadata,
814 /// this metadata needs to be cloned so that the inlined blocks
815 /// have different "unique scopes" at every call site. Were this not done, then
816 /// aliasing scopes from a function inlined into a caller multiple times could
817 /// not be differentiated (and this would lead to miscompiles because the
818 /// non-aliasing property communicated by the metadata could have
819 /// call-site-specific control dependencies).
820 static void CloneAliasScopeMetadata(CallSite CS, ValueToValueMapTy &VMap) {
821 const Function *CalledFunc = CS.getCalledFunction();
822 SetVector<const MDNode *> MD;
824 // Note: We could only clone the metadata if it is already used in the
825 // caller. I'm omitting that check here because it might confuse
826 // inter-procedural alias analysis passes. We can revisit this if it becomes
827 // an efficiency or overhead problem.
829 for (const BasicBlock &I : *CalledFunc)
830 for (const Instruction &J : I) {
831 if (const MDNode *M = J.getMetadata(LLVMContext::MD_alias_scope))
833 if (const MDNode *M = J.getMetadata(LLVMContext::MD_noalias))
840 // Walk the existing metadata, adding the complete (perhaps cyclic) chain to
842 SmallVector<const Metadata *, 16> Queue(MD.begin(), MD.end());
843 while (!Queue.empty()) {
844 const MDNode *M = cast<MDNode>(Queue.pop_back_val());
845 for (unsigned i = 0, ie = M->getNumOperands(); i != ie; ++i)
846 if (const MDNode *M1 = dyn_cast<MDNode>(M->getOperand(i)))
851 // Now we have a complete set of all metadata in the chains used to specify
852 // the noalias scopes and the lists of those scopes.
853 SmallVector<TempMDTuple, 16> DummyNodes;
854 DenseMap<const MDNode *, TrackingMDNodeRef> MDMap;
855 for (const MDNode *I : MD) {
856 DummyNodes.push_back(MDTuple::getTemporary(CalledFunc->getContext(), None));
857 MDMap[I].reset(DummyNodes.back().get());
860 // Create new metadata nodes to replace the dummy nodes, replacing old
861 // metadata references with either a dummy node or an already-created new
863 for (const MDNode *I : MD) {
864 SmallVector<Metadata *, 4> NewOps;
865 for (unsigned i = 0, ie = I->getNumOperands(); i != ie; ++i) {
866 const Metadata *V = I->getOperand(i);
867 if (const MDNode *M = dyn_cast<MDNode>(V))
868 NewOps.push_back(MDMap[M]);
870 NewOps.push_back(const_cast<Metadata *>(V));
873 MDNode *NewM = MDNode::get(CalledFunc->getContext(), NewOps);
874 MDTuple *TempM = cast<MDTuple>(MDMap[I]);
875 assert(TempM->isTemporary() && "Expected temporary node");
877 TempM->replaceAllUsesWith(NewM);
880 // Now replace the metadata in the new inlined instructions with the
881 // repacements from the map.
882 for (ValueToValueMapTy::iterator VMI = VMap.begin(), VMIE = VMap.end();
883 VMI != VMIE; ++VMI) {
887 Instruction *NI = dyn_cast<Instruction>(VMI->second);
891 if (MDNode *M = NI->getMetadata(LLVMContext::MD_alias_scope)) {
892 MDNode *NewMD = MDMap[M];
893 // If the call site also had alias scope metadata (a list of scopes to
894 // which instructions inside it might belong), propagate those scopes to
895 // the inlined instructions.
897 CS.getInstruction()->getMetadata(LLVMContext::MD_alias_scope))
898 NewMD = MDNode::concatenate(NewMD, CSM);
899 NI->setMetadata(LLVMContext::MD_alias_scope, NewMD);
900 } else if (NI->mayReadOrWriteMemory()) {
902 CS.getInstruction()->getMetadata(LLVMContext::MD_alias_scope))
903 NI->setMetadata(LLVMContext::MD_alias_scope, M);
906 if (MDNode *M = NI->getMetadata(LLVMContext::MD_noalias)) {
907 MDNode *NewMD = MDMap[M];
908 // If the call site also had noalias metadata (a list of scopes with
909 // which instructions inside it don't alias), propagate those scopes to
910 // the inlined instructions.
912 CS.getInstruction()->getMetadata(LLVMContext::MD_noalias))
913 NewMD = MDNode::concatenate(NewMD, CSM);
914 NI->setMetadata(LLVMContext::MD_noalias, NewMD);
915 } else if (NI->mayReadOrWriteMemory()) {
916 if (MDNode *M = CS.getInstruction()->getMetadata(LLVMContext::MD_noalias))
917 NI->setMetadata(LLVMContext::MD_noalias, M);
922 /// If the inlined function has noalias arguments,
923 /// then add new alias scopes for each noalias argument, tag the mapped noalias
924 /// parameters with noalias metadata specifying the new scope, and tag all
925 /// non-derived loads, stores and memory intrinsics with the new alias scopes.
926 static void AddAliasScopeMetadata(CallSite CS, ValueToValueMapTy &VMap,
927 const DataLayout &DL, AAResults *CalleeAAR) {
928 if (!EnableNoAliasConversion)
931 const Function *CalledFunc = CS.getCalledFunction();
932 SmallVector<const Argument *, 4> NoAliasArgs;
934 for (const Argument &Arg : CalledFunc->args())
935 if (Arg.hasNoAliasAttr() && !Arg.use_empty())
936 NoAliasArgs.push_back(&Arg);
938 if (NoAliasArgs.empty())
941 // To do a good job, if a noalias variable is captured, we need to know if
942 // the capture point dominates the particular use we're considering.
944 DT.recalculate(const_cast<Function&>(*CalledFunc));
946 // noalias indicates that pointer values based on the argument do not alias
947 // pointer values which are not based on it. So we add a new "scope" for each
948 // noalias function argument. Accesses using pointers based on that argument
949 // become part of that alias scope, accesses using pointers not based on that
950 // argument are tagged as noalias with that scope.
952 DenseMap<const Argument *, MDNode *> NewScopes;
953 MDBuilder MDB(CalledFunc->getContext());
955 // Create a new scope domain for this function.
957 MDB.createAnonymousAliasScopeDomain(CalledFunc->getName());
958 for (unsigned i = 0, e = NoAliasArgs.size(); i != e; ++i) {
959 const Argument *A = NoAliasArgs[i];
961 std::string Name = CalledFunc->getName();
964 Name += A->getName();
966 Name += ": argument ";
970 // Note: We always create a new anonymous root here. This is true regardless
971 // of the linkage of the callee because the aliasing "scope" is not just a
972 // property of the callee, but also all control dependencies in the caller.
973 MDNode *NewScope = MDB.createAnonymousAliasScope(NewDomain, Name);
974 NewScopes.insert(std::make_pair(A, NewScope));
977 // Iterate over all new instructions in the map; for all memory-access
978 // instructions, add the alias scope metadata.
979 for (ValueToValueMapTy::iterator VMI = VMap.begin(), VMIE = VMap.end();
980 VMI != VMIE; ++VMI) {
981 if (const Instruction *I = dyn_cast<Instruction>(VMI->first)) {
985 Instruction *NI = dyn_cast<Instruction>(VMI->second);
989 bool IsArgMemOnlyCall = false, IsFuncCall = false;
990 SmallVector<const Value *, 2> PtrArgs;
992 if (const LoadInst *LI = dyn_cast<LoadInst>(I))
993 PtrArgs.push_back(LI->getPointerOperand());
994 else if (const StoreInst *SI = dyn_cast<StoreInst>(I))
995 PtrArgs.push_back(SI->getPointerOperand());
996 else if (const VAArgInst *VAAI = dyn_cast<VAArgInst>(I))
997 PtrArgs.push_back(VAAI->getPointerOperand());
998 else if (const AtomicCmpXchgInst *CXI = dyn_cast<AtomicCmpXchgInst>(I))
999 PtrArgs.push_back(CXI->getPointerOperand());
1000 else if (const AtomicRMWInst *RMWI = dyn_cast<AtomicRMWInst>(I))
1001 PtrArgs.push_back(RMWI->getPointerOperand());
1002 else if (const auto *Call = dyn_cast<CallBase>(I)) {
1003 // If we know that the call does not access memory, then we'll still
1004 // know that about the inlined clone of this call site, and we don't
1005 // need to add metadata.
1006 if (Call->doesNotAccessMemory())
1011 FunctionModRefBehavior MRB = CalleeAAR->getModRefBehavior(Call);
1012 if (MRB == FMRB_OnlyAccessesArgumentPointees ||
1013 MRB == FMRB_OnlyReadsArgumentPointees)
1014 IsArgMemOnlyCall = true;
1017 for (Value *Arg : Call->args()) {
1018 // We need to check the underlying objects of all arguments, not just
1019 // the pointer arguments, because we might be passing pointers as
1021 // However, if we know that the call only accesses pointer arguments,
1022 // then we only need to check the pointer arguments.
1023 if (IsArgMemOnlyCall && !Arg->getType()->isPointerTy())
1026 PtrArgs.push_back(Arg);
1030 // If we found no pointers, then this instruction is not suitable for
1031 // pairing with an instruction to receive aliasing metadata.
1032 // However, if this is a call, this we might just alias with none of the
1033 // noalias arguments.
1034 if (PtrArgs.empty() && !IsFuncCall)
1037 // It is possible that there is only one underlying object, but you
1038 // need to go through several PHIs to see it, and thus could be
1039 // repeated in the Objects list.
1040 SmallPtrSet<const Value *, 4> ObjSet;
1041 SmallVector<Metadata *, 4> Scopes, NoAliases;
1043 SmallSetVector<const Argument *, 4> NAPtrArgs;
1044 for (const Value *V : PtrArgs) {
1045 SmallVector<Value *, 4> Objects;
1046 GetUnderlyingObjects(const_cast<Value*>(V),
1047 Objects, DL, /* LI = */ nullptr);
1049 for (Value *O : Objects)
1053 // Figure out if we're derived from anything that is not a noalias
1055 bool CanDeriveViaCapture = false, UsesAliasingPtr = false;
1056 for (const Value *V : ObjSet) {
1057 // Is this value a constant that cannot be derived from any pointer
1058 // value (we need to exclude constant expressions, for example, that
1059 // are formed from arithmetic on global symbols).
1060 bool IsNonPtrConst = isa<ConstantInt>(V) || isa<ConstantFP>(V) ||
1061 isa<ConstantPointerNull>(V) ||
1062 isa<ConstantDataVector>(V) || isa<UndefValue>(V);
1066 // If this is anything other than a noalias argument, then we cannot
1067 // completely describe the aliasing properties using alias.scope
1068 // metadata (and, thus, won't add any).
1069 if (const Argument *A = dyn_cast<Argument>(V)) {
1070 if (!A->hasNoAliasAttr())
1071 UsesAliasingPtr = true;
1073 UsesAliasingPtr = true;
1076 // If this is not some identified function-local object (which cannot
1077 // directly alias a noalias argument), or some other argument (which,
1078 // by definition, also cannot alias a noalias argument), then we could
1079 // alias a noalias argument that has been captured).
1080 if (!isa<Argument>(V) &&
1081 !isIdentifiedFunctionLocal(const_cast<Value*>(V)))
1082 CanDeriveViaCapture = true;
1085 // A function call can always get captured noalias pointers (via other
1086 // parameters, globals, etc.).
1087 if (IsFuncCall && !IsArgMemOnlyCall)
1088 CanDeriveViaCapture = true;
1090 // First, we want to figure out all of the sets with which we definitely
1091 // don't alias. Iterate over all noalias set, and add those for which:
1092 // 1. The noalias argument is not in the set of objects from which we
1093 // definitely derive.
1094 // 2. The noalias argument has not yet been captured.
1095 // An arbitrary function that might load pointers could see captured
1096 // noalias arguments via other noalias arguments or globals, and so we
1097 // must always check for prior capture.
1098 for (const Argument *A : NoAliasArgs) {
1099 if (!ObjSet.count(A) && (!CanDeriveViaCapture ||
1100 // It might be tempting to skip the
1101 // PointerMayBeCapturedBefore check if
1102 // A->hasNoCaptureAttr() is true, but this is
1103 // incorrect because nocapture only guarantees
1104 // that no copies outlive the function, not
1105 // that the value cannot be locally captured.
1106 !PointerMayBeCapturedBefore(A,
1107 /* ReturnCaptures */ false,
1108 /* StoreCaptures */ false, I, &DT)))
1109 NoAliases.push_back(NewScopes[A]);
1112 if (!NoAliases.empty())
1113 NI->setMetadata(LLVMContext::MD_noalias,
1114 MDNode::concatenate(
1115 NI->getMetadata(LLVMContext::MD_noalias),
1116 MDNode::get(CalledFunc->getContext(), NoAliases)));
1118 // Next, we want to figure out all of the sets to which we might belong.
1119 // We might belong to a set if the noalias argument is in the set of
1120 // underlying objects. If there is some non-noalias argument in our list
1121 // of underlying objects, then we cannot add a scope because the fact
1122 // that some access does not alias with any set of our noalias arguments
1123 // cannot itself guarantee that it does not alias with this access
1124 // (because there is some pointer of unknown origin involved and the
1125 // other access might also depend on this pointer). We also cannot add
1126 // scopes to arbitrary functions unless we know they don't access any
1127 // non-parameter pointer-values.
1128 bool CanAddScopes = !UsesAliasingPtr;
1129 if (CanAddScopes && IsFuncCall)
1130 CanAddScopes = IsArgMemOnlyCall;
1133 for (const Argument *A : NoAliasArgs) {
1134 if (ObjSet.count(A))
1135 Scopes.push_back(NewScopes[A]);
1138 if (!Scopes.empty())
1140 LLVMContext::MD_alias_scope,
1141 MDNode::concatenate(NI->getMetadata(LLVMContext::MD_alias_scope),
1142 MDNode::get(CalledFunc->getContext(), Scopes)));
1147 /// If the inlined function has non-byval align arguments, then
1148 /// add @llvm.assume-based alignment assumptions to preserve this information.
1149 static void AddAlignmentAssumptions(CallSite CS, InlineFunctionInfo &IFI) {
1150 if (!PreserveAlignmentAssumptions || !IFI.GetAssumptionCache)
1153 AssumptionCache *AC = &(*IFI.GetAssumptionCache)(*CS.getCaller());
1154 auto &DL = CS.getCaller()->getParent()->getDataLayout();
1156 // To avoid inserting redundant assumptions, we should check for assumptions
1157 // already in the caller. To do this, we might need a DT of the caller.
1159 bool DTCalculated = false;
1161 Function *CalledFunc = CS.getCalledFunction();
1162 for (Argument &Arg : CalledFunc->args()) {
1163 unsigned Align = Arg.getType()->isPointerTy() ? Arg.getParamAlignment() : 0;
1164 if (Align && !Arg.hasByValOrInAllocaAttr() && !Arg.hasNUses(0)) {
1165 if (!DTCalculated) {
1166 DT.recalculate(*CS.getCaller());
1167 DTCalculated = true;
1170 // If we can already prove the asserted alignment in the context of the
1171 // caller, then don't bother inserting the assumption.
1172 Value *ArgVal = CS.getArgument(Arg.getArgNo());
1173 if (getKnownAlignment(ArgVal, DL, CS.getInstruction(), AC, &DT) >= Align)
1176 CallInst *NewAsmp = IRBuilder<>(CS.getInstruction())
1177 .CreateAlignmentAssumption(DL, ArgVal, Align);
1178 AC->registerAssumption(NewAsmp);
1183 /// Once we have cloned code over from a callee into the caller,
1184 /// update the specified callgraph to reflect the changes we made.
1185 /// Note that it's possible that not all code was copied over, so only
1186 /// some edges of the callgraph may remain.
1187 static void UpdateCallGraphAfterInlining(CallSite CS,
1188 Function::iterator FirstNewBlock,
1189 ValueToValueMapTy &VMap,
1190 InlineFunctionInfo &IFI) {
1191 CallGraph &CG = *IFI.CG;
1192 const Function *Caller = CS.getCaller();
1193 const Function *Callee = CS.getCalledFunction();
1194 CallGraphNode *CalleeNode = CG[Callee];
1195 CallGraphNode *CallerNode = CG[Caller];
1197 // Since we inlined some uninlined call sites in the callee into the caller,
1198 // add edges from the caller to all of the callees of the callee.
1199 CallGraphNode::iterator I = CalleeNode->begin(), E = CalleeNode->end();
1201 // Consider the case where CalleeNode == CallerNode.
1202 CallGraphNode::CalledFunctionsVector CallCache;
1203 if (CalleeNode == CallerNode) {
1204 CallCache.assign(I, E);
1205 I = CallCache.begin();
1206 E = CallCache.end();
1209 for (; I != E; ++I) {
1210 const Value *OrigCall = I->first;
1212 ValueToValueMapTy::iterator VMI = VMap.find(OrigCall);
1213 // Only copy the edge if the call was inlined!
1214 if (VMI == VMap.end() || VMI->second == nullptr)
1217 // If the call was inlined, but then constant folded, there is no edge to
1218 // add. Check for this case.
1219 Instruction *NewCall = dyn_cast<Instruction>(VMI->second);
1223 // We do not treat intrinsic calls like real function calls because we
1224 // expect them to become inline code; do not add an edge for an intrinsic.
1225 CallSite CS = CallSite(NewCall);
1226 if (CS && CS.getCalledFunction() && CS.getCalledFunction()->isIntrinsic())
1229 // Remember that this call site got inlined for the client of
1231 IFI.InlinedCalls.push_back(NewCall);
1233 // It's possible that inlining the callsite will cause it to go from an
1234 // indirect to a direct call by resolving a function pointer. If this
1235 // happens, set the callee of the new call site to a more precise
1236 // destination. This can also happen if the call graph node of the caller
1237 // was just unnecessarily imprecise.
1238 if (!I->second->getFunction())
1239 if (Function *F = CallSite(NewCall).getCalledFunction()) {
1240 // Indirect call site resolved to direct call.
1241 CallerNode->addCalledFunction(CallSite(NewCall), CG[F]);
1246 CallerNode->addCalledFunction(CallSite(NewCall), I->second);
1249 // Update the call graph by deleting the edge from Callee to Caller. We must
1250 // do this after the loop above in case Caller and Callee are the same.
1251 CallerNode->removeCallEdgeFor(CS);
1254 static void HandleByValArgumentInit(Value *Dst, Value *Src, Module *M,
1255 BasicBlock *InsertBlock,
1256 InlineFunctionInfo &IFI) {
1257 Type *AggTy = cast<PointerType>(Src->getType())->getElementType();
1258 IRBuilder<> Builder(InsertBlock, InsertBlock->begin());
1260 Value *Size = Builder.getInt64(M->getDataLayout().getTypeStoreSize(AggTy));
1262 // Always generate a memcpy of alignment 1 here because we don't know
1263 // the alignment of the src pointer. Other optimizations can infer
1264 // better alignment.
1265 Builder.CreateMemCpy(Dst, /*DstAlign*/1, Src, /*SrcAlign*/1, Size);
1268 /// When inlining a call site that has a byval argument,
1269 /// we have to make the implicit memcpy explicit by adding it.
1270 static Value *HandleByValArgument(Value *Arg, Instruction *TheCall,
1271 const Function *CalledFunc,
1272 InlineFunctionInfo &IFI,
1273 unsigned ByValAlignment) {
1274 PointerType *ArgTy = cast<PointerType>(Arg->getType());
1275 Type *AggTy = ArgTy->getElementType();
1277 Function *Caller = TheCall->getFunction();
1278 const DataLayout &DL = Caller->getParent()->getDataLayout();
1280 // If the called function is readonly, then it could not mutate the caller's
1281 // copy of the byval'd memory. In this case, it is safe to elide the copy and
1283 if (CalledFunc->onlyReadsMemory()) {
1284 // If the byval argument has a specified alignment that is greater than the
1285 // passed in pointer, then we either have to round up the input pointer or
1286 // give up on this transformation.
1287 if (ByValAlignment <= 1) // 0 = unspecified, 1 = no particular alignment.
1290 AssumptionCache *AC =
1291 IFI.GetAssumptionCache ? &(*IFI.GetAssumptionCache)(*Caller) : nullptr;
1293 // If the pointer is already known to be sufficiently aligned, or if we can
1294 // round it up to a larger alignment, then we don't need a temporary.
1295 if (getOrEnforceKnownAlignment(Arg, ByValAlignment, DL, TheCall, AC) >=
1299 // Otherwise, we have to make a memcpy to get a safe alignment. This is bad
1300 // for code quality, but rarely happens and is required for correctness.
1303 // Create the alloca. If we have DataLayout, use nice alignment.
1304 unsigned Align = DL.getPrefTypeAlignment(AggTy);
1306 // If the byval had an alignment specified, we *must* use at least that
1307 // alignment, as it is required by the byval argument (and uses of the
1308 // pointer inside the callee).
1309 Align = std::max(Align, ByValAlignment);
1311 Value *NewAlloca = new AllocaInst(AggTy, DL.getAllocaAddrSpace(),
1312 nullptr, Align, Arg->getName(),
1313 &*Caller->begin()->begin());
1314 IFI.StaticAllocas.push_back(cast<AllocaInst>(NewAlloca));
1316 // Uses of the argument in the function should use our new alloca
1321 // Check whether this Value is used by a lifetime intrinsic.
1322 static bool isUsedByLifetimeMarker(Value *V) {
1323 for (User *U : V->users())
1324 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(U))
1325 if (II->isLifetimeStartOrEnd())
1330 // Check whether the given alloca already has
1331 // lifetime.start or lifetime.end intrinsics.
1332 static bool hasLifetimeMarkers(AllocaInst *AI) {
1333 Type *Ty = AI->getType();
1334 Type *Int8PtrTy = Type::getInt8PtrTy(Ty->getContext(),
1335 Ty->getPointerAddressSpace());
1336 if (Ty == Int8PtrTy)
1337 return isUsedByLifetimeMarker(AI);
1339 // Do a scan to find all the casts to i8*.
1340 for (User *U : AI->users()) {
1341 if (U->getType() != Int8PtrTy) continue;
1342 if (U->stripPointerCasts() != AI) continue;
1343 if (isUsedByLifetimeMarker(U))
1349 /// Return the result of AI->isStaticAlloca() if AI were moved to the entry
1350 /// block. Allocas used in inalloca calls and allocas of dynamic array size
1351 /// cannot be static.
1352 static bool allocaWouldBeStaticInEntry(const AllocaInst *AI ) {
1353 return isa<Constant>(AI->getArraySize()) && !AI->isUsedWithInAlloca();
1356 /// Update inlined instructions' line numbers to
1357 /// to encode location where these instructions are inlined.
1358 static void fixupLineNumbers(Function *Fn, Function::iterator FI,
1359 Instruction *TheCall, bool CalleeHasDebugInfo) {
1360 const DebugLoc &TheCallDL = TheCall->getDebugLoc();
1364 auto &Ctx = Fn->getContext();
1365 DILocation *InlinedAtNode = TheCallDL;
1367 // Create a unique call site, not to be confused with any other call from the
1369 InlinedAtNode = DILocation::getDistinct(
1370 Ctx, InlinedAtNode->getLine(), InlinedAtNode->getColumn(),
1371 InlinedAtNode->getScope(), InlinedAtNode->getInlinedAt());
1373 // Cache the inlined-at nodes as they're built so they are reused, without
1374 // this every instruction's inlined-at chain would become distinct from each
1376 DenseMap<const MDNode *, MDNode *> IANodes;
1378 for (; FI != Fn->end(); ++FI) {
1379 for (BasicBlock::iterator BI = FI->begin(), BE = FI->end();
1381 if (DebugLoc DL = BI->getDebugLoc()) {
1382 auto IA = DebugLoc::appendInlinedAt(DL, InlinedAtNode, BI->getContext(),
1384 auto IDL = DebugLoc::get(DL.getLine(), DL.getCol(), DL.getScope(), IA);
1385 BI->setDebugLoc(IDL);
1389 if (CalleeHasDebugInfo)
1392 // If the inlined instruction has no line number, make it look as if it
1393 // originates from the call location. This is important for
1394 // ((__always_inline__, __nodebug__)) functions which must use caller
1395 // location for all instructions in their function body.
1397 // Don't update static allocas, as they may get moved later.
1398 if (auto *AI = dyn_cast<AllocaInst>(BI))
1399 if (allocaWouldBeStaticInEntry(AI))
1402 BI->setDebugLoc(TheCallDL);
1407 /// Update the block frequencies of the caller after a callee has been inlined.
1409 /// Each block cloned into the caller has its block frequency scaled by the
1410 /// ratio of CallSiteFreq/CalleeEntryFreq. This ensures that the cloned copy of
1411 /// callee's entry block gets the same frequency as the callsite block and the
1412 /// relative frequencies of all cloned blocks remain the same after cloning.
1413 static void updateCallerBFI(BasicBlock *CallSiteBlock,
1414 const ValueToValueMapTy &VMap,
1415 BlockFrequencyInfo *CallerBFI,
1416 BlockFrequencyInfo *CalleeBFI,
1417 const BasicBlock &CalleeEntryBlock) {
1418 SmallPtrSet<BasicBlock *, 16> ClonedBBs;
1419 for (auto const &Entry : VMap) {
1420 if (!isa<BasicBlock>(Entry.first) || !Entry.second)
1422 auto *OrigBB = cast<BasicBlock>(Entry.first);
1423 auto *ClonedBB = cast<BasicBlock>(Entry.second);
1424 uint64_t Freq = CalleeBFI->getBlockFreq(OrigBB).getFrequency();
1425 if (!ClonedBBs.insert(ClonedBB).second) {
1426 // Multiple blocks in the callee might get mapped to one cloned block in
1427 // the caller since we prune the callee as we clone it. When that happens,
1428 // we want to use the maximum among the original blocks' frequencies.
1429 uint64_t NewFreq = CallerBFI->getBlockFreq(ClonedBB).getFrequency();
1433 CallerBFI->setBlockFreq(ClonedBB, Freq);
1435 BasicBlock *EntryClone = cast<BasicBlock>(VMap.lookup(&CalleeEntryBlock));
1436 CallerBFI->setBlockFreqAndScale(
1437 EntryClone, CallerBFI->getBlockFreq(CallSiteBlock).getFrequency(),
1441 /// Update the branch metadata for cloned call instructions.
1442 static void updateCallProfile(Function *Callee, const ValueToValueMapTy &VMap,
1443 const ProfileCount &CalleeEntryCount,
1444 const Instruction *TheCall,
1445 ProfileSummaryInfo *PSI,
1446 BlockFrequencyInfo *CallerBFI) {
1447 if (!CalleeEntryCount.hasValue() || CalleeEntryCount.isSynthetic() ||
1448 CalleeEntryCount.getCount() < 1)
1450 auto CallSiteCount = PSI ? PSI->getProfileCount(TheCall, CallerBFI) : None;
1451 uint64_t CallCount =
1452 std::min(CallSiteCount.hasValue() ? CallSiteCount.getValue() : 0,
1453 CalleeEntryCount.getCount());
1455 for (auto const &Entry : VMap)
1456 if (isa<CallInst>(Entry.first))
1457 if (auto *CI = dyn_cast_or_null<CallInst>(Entry.second))
1458 CI->updateProfWeight(CallCount, CalleeEntryCount.getCount());
1459 for (BasicBlock &BB : *Callee)
1460 // No need to update the callsite if it is pruned during inlining.
1461 if (VMap.count(&BB))
1462 for (Instruction &I : BB)
1463 if (CallInst *CI = dyn_cast<CallInst>(&I))
1464 CI->updateProfWeight(CalleeEntryCount.getCount() - CallCount,
1465 CalleeEntryCount.getCount());
1468 /// Update the entry count of callee after inlining.
1470 /// The callsite's block count is subtracted from the callee's function entry
1472 static void updateCalleeCount(BlockFrequencyInfo *CallerBFI, BasicBlock *CallBB,
1473 Instruction *CallInst, Function *Callee,
1474 ProfileSummaryInfo *PSI) {
1475 // If the callee has a original count of N, and the estimated count of
1476 // callsite is M, the new callee count is set to N - M. M is estimated from
1477 // the caller's entry count, its entry block frequency and the block frequency
1479 auto CalleeCount = Callee->getEntryCount();
1480 if (!CalleeCount.hasValue() || !PSI)
1482 auto CallCount = PSI->getProfileCount(CallInst, CallerBFI);
1483 if (!CallCount.hasValue())
1485 // Since CallSiteCount is an estimate, it could exceed the original callee
1486 // count and has to be set to 0.
1487 if (CallCount.getValue() > CalleeCount.getCount())
1488 CalleeCount.setCount(0);
1490 CalleeCount.setCount(CalleeCount.getCount() - CallCount.getValue());
1491 Callee->setEntryCount(CalleeCount);
1494 /// This function inlines the called function into the basic block of the
1495 /// caller. This returns false if it is not possible to inline this call.
1496 /// The program is still in a well defined state if this occurs though.
1498 /// Note that this only does one level of inlining. For example, if the
1499 /// instruction 'call B' is inlined, and 'B' calls 'C', then the call to 'C' now
1500 /// exists in the instruction stream. Similarly this will inline a recursive
1501 /// function by one level.
1502 llvm::InlineResult llvm::InlineFunction(CallSite CS, InlineFunctionInfo &IFI,
1503 AAResults *CalleeAAR,
1504 bool InsertLifetime,
1505 Function *ForwardVarArgsTo) {
1506 Instruction *TheCall = CS.getInstruction();
1507 assert(TheCall->getParent() && TheCall->getFunction()
1508 && "Instruction not in function!");
1510 // If IFI has any state in it, zap it before we fill it in.
1513 Function *CalledFunc = CS.getCalledFunction();
1514 if (!CalledFunc || // Can't inline external function or indirect
1515 CalledFunc->isDeclaration()) // call!
1516 return "external or indirect";
1518 // The inliner does not know how to inline through calls with operand bundles
1520 if (CS.hasOperandBundles()) {
1521 for (int i = 0, e = CS.getNumOperandBundles(); i != e; ++i) {
1522 uint32_t Tag = CS.getOperandBundleAt(i).getTagID();
1523 // ... but it knows how to inline through "deopt" operand bundles ...
1524 if (Tag == LLVMContext::OB_deopt)
1526 // ... and "funclet" operand bundles.
1527 if (Tag == LLVMContext::OB_funclet)
1530 return "unsupported operand bundle";
1534 // If the call to the callee cannot throw, set the 'nounwind' flag on any
1535 // calls that we inline.
1536 bool MarkNoUnwind = CS.doesNotThrow();
1538 BasicBlock *OrigBB = TheCall->getParent();
1539 Function *Caller = OrigBB->getParent();
1541 // GC poses two hazards to inlining, which only occur when the callee has GC:
1542 // 1. If the caller has no GC, then the callee's GC must be propagated to the
1544 // 2. If the caller has a differing GC, it is invalid to inline.
1545 if (CalledFunc->hasGC()) {
1546 if (!Caller->hasGC())
1547 Caller->setGC(CalledFunc->getGC());
1548 else if (CalledFunc->getGC() != Caller->getGC())
1549 return "incompatible GC";
1552 // Get the personality function from the callee if it contains a landing pad.
1553 Constant *CalledPersonality =
1554 CalledFunc->hasPersonalityFn()
1555 ? CalledFunc->getPersonalityFn()->stripPointerCasts()
1558 // Find the personality function used by the landing pads of the caller. If it
1559 // exists, then check to see that it matches the personality function used in
1561 Constant *CallerPersonality =
1562 Caller->hasPersonalityFn()
1563 ? Caller->getPersonalityFn()->stripPointerCasts()
1565 if (CalledPersonality) {
1566 if (!CallerPersonality)
1567 Caller->setPersonalityFn(CalledPersonality);
1568 // If the personality functions match, then we can perform the
1569 // inlining. Otherwise, we can't inline.
1570 // TODO: This isn't 100% true. Some personality functions are proper
1571 // supersets of others and can be used in place of the other.
1572 else if (CalledPersonality != CallerPersonality)
1573 return "incompatible personality";
1576 // We need to figure out which funclet the callsite was in so that we may
1577 // properly nest the callee.
1578 Instruction *CallSiteEHPad = nullptr;
1579 if (CallerPersonality) {
1580 EHPersonality Personality = classifyEHPersonality(CallerPersonality);
1581 if (isScopedEHPersonality(Personality)) {
1582 Optional<OperandBundleUse> ParentFunclet =
1583 CS.getOperandBundle(LLVMContext::OB_funclet);
1585 CallSiteEHPad = cast<FuncletPadInst>(ParentFunclet->Inputs.front());
1587 // OK, the inlining site is legal. What about the target function?
1589 if (CallSiteEHPad) {
1590 if (Personality == EHPersonality::MSVC_CXX) {
1591 // The MSVC personality cannot tolerate catches getting inlined into
1592 // cleanup funclets.
1593 if (isa<CleanupPadInst>(CallSiteEHPad)) {
1594 // Ok, the call site is within a cleanuppad. Let's check the callee
1596 for (const BasicBlock &CalledBB : *CalledFunc) {
1597 if (isa<CatchSwitchInst>(CalledBB.getFirstNonPHI()))
1598 return "catch in cleanup funclet";
1601 } else if (isAsynchronousEHPersonality(Personality)) {
1602 // SEH is even less tolerant, there may not be any sort of exceptional
1603 // funclet in the callee.
1604 for (const BasicBlock &CalledBB : *CalledFunc) {
1605 if (CalledBB.isEHPad())
1606 return "SEH in cleanup funclet";
1613 // Determine if we are dealing with a call in an EHPad which does not unwind
1615 bool EHPadForCallUnwindsLocally = false;
1616 if (CallSiteEHPad && CS.isCall()) {
1617 UnwindDestMemoTy FuncletUnwindMap;
1618 Value *CallSiteUnwindDestToken =
1619 getUnwindDestToken(CallSiteEHPad, FuncletUnwindMap);
1621 EHPadForCallUnwindsLocally =
1622 CallSiteUnwindDestToken &&
1623 !isa<ConstantTokenNone>(CallSiteUnwindDestToken);
1626 // Get an iterator to the last basic block in the function, which will have
1627 // the new function inlined after it.
1628 Function::iterator LastBlock = --Caller->end();
1630 // Make sure to capture all of the return instructions from the cloned
1632 SmallVector<ReturnInst*, 8> Returns;
1633 ClonedCodeInfo InlinedFunctionInfo;
1634 Function::iterator FirstNewBlock;
1636 { // Scope to destroy VMap after cloning.
1637 ValueToValueMapTy VMap;
1638 // Keep a list of pair (dst, src) to emit byval initializations.
1639 SmallVector<std::pair<Value*, Value*>, 4> ByValInit;
1641 auto &DL = Caller->getParent()->getDataLayout();
1643 // Calculate the vector of arguments to pass into the function cloner, which
1644 // matches up the formal to the actual argument values.
1645 CallSite::arg_iterator AI = CS.arg_begin();
1647 for (Function::arg_iterator I = CalledFunc->arg_begin(),
1648 E = CalledFunc->arg_end(); I != E; ++I, ++AI, ++ArgNo) {
1649 Value *ActualArg = *AI;
1651 // When byval arguments actually inlined, we need to make the copy implied
1652 // by them explicit. However, we don't do this if the callee is readonly
1653 // or readnone, because the copy would be unneeded: the callee doesn't
1654 // modify the struct.
1655 if (CS.isByValArgument(ArgNo)) {
1656 ActualArg = HandleByValArgument(ActualArg, TheCall, CalledFunc, IFI,
1657 CalledFunc->getParamAlignment(ArgNo));
1658 if (ActualArg != *AI)
1659 ByValInit.push_back(std::make_pair(ActualArg, (Value*) *AI));
1662 VMap[&*I] = ActualArg;
1665 // Add alignment assumptions if necessary. We do this before the inlined
1666 // instructions are actually cloned into the caller so that we can easily
1667 // check what will be known at the start of the inlined code.
1668 AddAlignmentAssumptions(CS, IFI);
1670 // We want the inliner to prune the code as it copies. We would LOVE to
1671 // have no dead or constant instructions leftover after inlining occurs
1672 // (which can happen, e.g., because an argument was constant), but we'll be
1673 // happy with whatever the cloner can do.
1674 CloneAndPruneFunctionInto(Caller, CalledFunc, VMap,
1675 /*ModuleLevelChanges=*/false, Returns, ".i",
1676 &InlinedFunctionInfo, TheCall);
1677 // Remember the first block that is newly cloned over.
1678 FirstNewBlock = LastBlock; ++FirstNewBlock;
1680 if (IFI.CallerBFI != nullptr && IFI.CalleeBFI != nullptr)
1681 // Update the BFI of blocks cloned into the caller.
1682 updateCallerBFI(OrigBB, VMap, IFI.CallerBFI, IFI.CalleeBFI,
1683 CalledFunc->front());
1685 updateCallProfile(CalledFunc, VMap, CalledFunc->getEntryCount(), TheCall,
1686 IFI.PSI, IFI.CallerBFI);
1687 // Update the profile count of callee.
1688 updateCalleeCount(IFI.CallerBFI, OrigBB, TheCall, CalledFunc, IFI.PSI);
1690 // Inject byval arguments initialization.
1691 for (std::pair<Value*, Value*> &Init : ByValInit)
1692 HandleByValArgumentInit(Init.first, Init.second, Caller->getParent(),
1693 &*FirstNewBlock, IFI);
1695 Optional<OperandBundleUse> ParentDeopt =
1696 CS.getOperandBundle(LLVMContext::OB_deopt);
1698 SmallVector<OperandBundleDef, 2> OpDefs;
1700 for (auto &VH : InlinedFunctionInfo.OperandBundleCallSites) {
1701 Instruction *I = dyn_cast_or_null<Instruction>(VH);
1702 if (!I) continue; // instruction was DCE'd or RAUW'ed to undef
1707 OpDefs.reserve(ICS.getNumOperandBundles());
1709 for (unsigned i = 0, e = ICS.getNumOperandBundles(); i < e; ++i) {
1710 auto ChildOB = ICS.getOperandBundleAt(i);
1711 if (ChildOB.getTagID() != LLVMContext::OB_deopt) {
1712 // If the inlined call has other operand bundles, let them be
1713 OpDefs.emplace_back(ChildOB);
1717 // It may be useful to separate this logic (of handling operand
1718 // bundles) out to a separate "policy" component if this gets crowded.
1719 // Prepend the parent's deoptimization continuation to the newly
1720 // inlined call's deoptimization continuation.
1721 std::vector<Value *> MergedDeoptArgs;
1722 MergedDeoptArgs.reserve(ParentDeopt->Inputs.size() +
1723 ChildOB.Inputs.size());
1725 MergedDeoptArgs.insert(MergedDeoptArgs.end(),
1726 ParentDeopt->Inputs.begin(),
1727 ParentDeopt->Inputs.end());
1728 MergedDeoptArgs.insert(MergedDeoptArgs.end(), ChildOB.Inputs.begin(),
1729 ChildOB.Inputs.end());
1731 OpDefs.emplace_back("deopt", std::move(MergedDeoptArgs));
1734 Instruction *NewI = nullptr;
1735 if (isa<CallInst>(I))
1736 NewI = CallInst::Create(cast<CallInst>(I), OpDefs, I);
1738 NewI = InvokeInst::Create(cast<InvokeInst>(I), OpDefs, I);
1740 // Note: the RAUW does the appropriate fixup in VMap, so we need to do
1741 // this even if the call returns void.
1742 I->replaceAllUsesWith(NewI);
1745 I->eraseFromParent();
1749 // Update the callgraph if requested.
1751 UpdateCallGraphAfterInlining(CS, FirstNewBlock, VMap, IFI);
1753 // For 'nodebug' functions, the associated DISubprogram is always null.
1754 // Conservatively avoid propagating the callsite debug location to
1755 // instructions inlined from a function whose DISubprogram is not null.
1756 fixupLineNumbers(Caller, FirstNewBlock, TheCall,
1757 CalledFunc->getSubprogram() != nullptr);
1759 // Clone existing noalias metadata if necessary.
1760 CloneAliasScopeMetadata(CS, VMap);
1762 // Add noalias metadata if necessary.
1763 AddAliasScopeMetadata(CS, VMap, DL, CalleeAAR);
1765 // Propagate llvm.mem.parallel_loop_access if necessary.
1766 PropagateParallelLoopAccessMetadata(CS, VMap);
1768 // Register any cloned assumptions.
1769 if (IFI.GetAssumptionCache)
1770 for (BasicBlock &NewBlock :
1771 make_range(FirstNewBlock->getIterator(), Caller->end()))
1772 for (Instruction &I : NewBlock) {
1773 if (auto *II = dyn_cast<IntrinsicInst>(&I))
1774 if (II->getIntrinsicID() == Intrinsic::assume)
1775 (*IFI.GetAssumptionCache)(*Caller).registerAssumption(II);
1779 // If there are any alloca instructions in the block that used to be the entry
1780 // block for the callee, move them to the entry block of the caller. First
1781 // calculate which instruction they should be inserted before. We insert the
1782 // instructions at the end of the current alloca list.
1784 BasicBlock::iterator InsertPoint = Caller->begin()->begin();
1785 for (BasicBlock::iterator I = FirstNewBlock->begin(),
1786 E = FirstNewBlock->end(); I != E; ) {
1787 AllocaInst *AI = dyn_cast<AllocaInst>(I++);
1790 // If the alloca is now dead, remove it. This often occurs due to code
1792 if (AI->use_empty()) {
1793 AI->eraseFromParent();
1797 if (!allocaWouldBeStaticInEntry(AI))
1800 // Keep track of the static allocas that we inline into the caller.
1801 IFI.StaticAllocas.push_back(AI);
1803 // Scan for the block of allocas that we can move over, and move them
1805 while (isa<AllocaInst>(I) &&
1806 allocaWouldBeStaticInEntry(cast<AllocaInst>(I))) {
1807 IFI.StaticAllocas.push_back(cast<AllocaInst>(I));
1811 // Transfer all of the allocas over in a block. Using splice means
1812 // that the instructions aren't removed from the symbol table, then
1814 Caller->getEntryBlock().getInstList().splice(
1815 InsertPoint, FirstNewBlock->getInstList(), AI->getIterator(), I);
1817 // Move any dbg.declares describing the allocas into the entry basic block.
1818 DIBuilder DIB(*Caller->getParent());
1819 for (auto &AI : IFI.StaticAllocas)
1820 replaceDbgDeclareForAlloca(AI, AI, DIB, DIExpression::NoDeref, 0,
1821 DIExpression::NoDeref);
1824 SmallVector<Value*,4> VarArgsToForward;
1825 SmallVector<AttributeSet, 4> VarArgsAttrs;
1826 for (unsigned i = CalledFunc->getFunctionType()->getNumParams();
1827 i < CS.getNumArgOperands(); i++) {
1828 VarArgsToForward.push_back(CS.getArgOperand(i));
1829 VarArgsAttrs.push_back(CS.getAttributes().getParamAttributes(i));
1832 bool InlinedMustTailCalls = false, InlinedDeoptimizeCalls = false;
1833 if (InlinedFunctionInfo.ContainsCalls) {
1834 CallInst::TailCallKind CallSiteTailKind = CallInst::TCK_None;
1835 if (CallInst *CI = dyn_cast<CallInst>(TheCall))
1836 CallSiteTailKind = CI->getTailCallKind();
1838 // For inlining purposes, the "notail" marker is the same as no marker.
1839 if (CallSiteTailKind == CallInst::TCK_NoTail)
1840 CallSiteTailKind = CallInst::TCK_None;
1842 for (Function::iterator BB = FirstNewBlock, E = Caller->end(); BB != E;
1844 for (auto II = BB->begin(); II != BB->end();) {
1845 Instruction &I = *II++;
1846 CallInst *CI = dyn_cast<CallInst>(&I);
1850 // Forward varargs from inlined call site to calls to the
1851 // ForwardVarArgsTo function, if requested, and to musttail calls.
1852 if (!VarArgsToForward.empty() &&
1853 ((ForwardVarArgsTo &&
1854 CI->getCalledFunction() == ForwardVarArgsTo) ||
1855 CI->isMustTailCall())) {
1856 // Collect attributes for non-vararg parameters.
1857 AttributeList Attrs = CI->getAttributes();
1858 SmallVector<AttributeSet, 8> ArgAttrs;
1859 if (!Attrs.isEmpty() || !VarArgsAttrs.empty()) {
1860 for (unsigned ArgNo = 0;
1861 ArgNo < CI->getFunctionType()->getNumParams(); ++ArgNo)
1862 ArgAttrs.push_back(Attrs.getParamAttributes(ArgNo));
1865 // Add VarArg attributes.
1866 ArgAttrs.append(VarArgsAttrs.begin(), VarArgsAttrs.end());
1867 Attrs = AttributeList::get(CI->getContext(), Attrs.getFnAttributes(),
1868 Attrs.getRetAttributes(), ArgAttrs);
1869 // Add VarArgs to existing parameters.
1870 SmallVector<Value *, 6> Params(CI->arg_operands());
1871 Params.append(VarArgsToForward.begin(), VarArgsToForward.end());
1873 CallInst::Create(CI->getCalledFunction() ? CI->getCalledFunction()
1874 : CI->getCalledValue(),
1876 NewCI->setDebugLoc(CI->getDebugLoc());
1877 NewCI->setAttributes(Attrs);
1878 NewCI->setCallingConv(CI->getCallingConv());
1879 CI->replaceAllUsesWith(NewCI);
1880 CI->eraseFromParent();
1884 if (Function *F = CI->getCalledFunction())
1885 InlinedDeoptimizeCalls |=
1886 F->getIntrinsicID() == Intrinsic::experimental_deoptimize;
1888 // We need to reduce the strength of any inlined tail calls. For
1889 // musttail, we have to avoid introducing potential unbounded stack
1890 // growth. For example, if functions 'f' and 'g' are mutually recursive
1891 // with musttail, we can inline 'g' into 'f' so long as we preserve
1892 // musttail on the cloned call to 'f'. If either the inlined call site
1893 // or the cloned call site is *not* musttail, the program already has
1894 // one frame of stack growth, so it's safe to remove musttail. Here is
1895 // a table of example transformations:
1897 // f -> musttail g -> musttail f ==> f -> musttail f
1898 // f -> musttail g -> tail f ==> f -> tail f
1899 // f -> g -> musttail f ==> f -> f
1900 // f -> g -> tail f ==> f -> f
1902 // Inlined notail calls should remain notail calls.
1903 CallInst::TailCallKind ChildTCK = CI->getTailCallKind();
1904 if (ChildTCK != CallInst::TCK_NoTail)
1905 ChildTCK = std::min(CallSiteTailKind, ChildTCK);
1906 CI->setTailCallKind(ChildTCK);
1907 InlinedMustTailCalls |= CI->isMustTailCall();
1909 // Calls inlined through a 'nounwind' call site should be marked
1912 CI->setDoesNotThrow();
1917 // Leave lifetime markers for the static alloca's, scoping them to the
1918 // function we just inlined.
1919 if (InsertLifetime && !IFI.StaticAllocas.empty()) {
1920 IRBuilder<> builder(&FirstNewBlock->front());
1921 for (unsigned ai = 0, ae = IFI.StaticAllocas.size(); ai != ae; ++ai) {
1922 AllocaInst *AI = IFI.StaticAllocas[ai];
1923 // Don't mark swifterror allocas. They can't have bitcast uses.
1924 if (AI->isSwiftError())
1927 // If the alloca is already scoped to something smaller than the whole
1928 // function then there's no need to add redundant, less accurate markers.
1929 if (hasLifetimeMarkers(AI))
1932 // Try to determine the size of the allocation.
1933 ConstantInt *AllocaSize = nullptr;
1934 if (ConstantInt *AIArraySize =
1935 dyn_cast<ConstantInt>(AI->getArraySize())) {
1936 auto &DL = Caller->getParent()->getDataLayout();
1937 Type *AllocaType = AI->getAllocatedType();
1938 uint64_t AllocaTypeSize = DL.getTypeAllocSize(AllocaType);
1939 uint64_t AllocaArraySize = AIArraySize->getLimitedValue();
1941 // Don't add markers for zero-sized allocas.
1942 if (AllocaArraySize == 0)
1945 // Check that array size doesn't saturate uint64_t and doesn't
1946 // overflow when it's multiplied by type size.
1947 if (AllocaArraySize != std::numeric_limits<uint64_t>::max() &&
1948 std::numeric_limits<uint64_t>::max() / AllocaArraySize >=
1950 AllocaSize = ConstantInt::get(Type::getInt64Ty(AI->getContext()),
1951 AllocaArraySize * AllocaTypeSize);
1955 builder.CreateLifetimeStart(AI, AllocaSize);
1956 for (ReturnInst *RI : Returns) {
1957 // Don't insert llvm.lifetime.end calls between a musttail or deoptimize
1958 // call and a return. The return kills all local allocas.
1959 if (InlinedMustTailCalls &&
1960 RI->getParent()->getTerminatingMustTailCall())
1962 if (InlinedDeoptimizeCalls &&
1963 RI->getParent()->getTerminatingDeoptimizeCall())
1965 IRBuilder<>(RI).CreateLifetimeEnd(AI, AllocaSize);
1970 // If the inlined code contained dynamic alloca instructions, wrap the inlined
1971 // code with llvm.stacksave/llvm.stackrestore intrinsics.
1972 if (InlinedFunctionInfo.ContainsDynamicAllocas) {
1973 Module *M = Caller->getParent();
1974 // Get the two intrinsics we care about.
1975 Function *StackSave = Intrinsic::getDeclaration(M, Intrinsic::stacksave);
1976 Function *StackRestore=Intrinsic::getDeclaration(M,Intrinsic::stackrestore);
1978 // Insert the llvm.stacksave.
1979 CallInst *SavedPtr = IRBuilder<>(&*FirstNewBlock, FirstNewBlock->begin())
1980 .CreateCall(StackSave, {}, "savedstack");
1982 // Insert a call to llvm.stackrestore before any return instructions in the
1983 // inlined function.
1984 for (ReturnInst *RI : Returns) {
1985 // Don't insert llvm.stackrestore calls between a musttail or deoptimize
1986 // call and a return. The return will restore the stack pointer.
1987 if (InlinedMustTailCalls && RI->getParent()->getTerminatingMustTailCall())
1989 if (InlinedDeoptimizeCalls && RI->getParent()->getTerminatingDeoptimizeCall())
1991 IRBuilder<>(RI).CreateCall(StackRestore, SavedPtr);
1995 // If we are inlining for an invoke instruction, we must make sure to rewrite
1996 // any call instructions into invoke instructions. This is sensitive to which
1997 // funclet pads were top-level in the inlinee, so must be done before
1998 // rewriting the "parent pad" links.
1999 if (auto *II = dyn_cast<InvokeInst>(TheCall)) {
2000 BasicBlock *UnwindDest = II->getUnwindDest();
2001 Instruction *FirstNonPHI = UnwindDest->getFirstNonPHI();
2002 if (isa<LandingPadInst>(FirstNonPHI)) {
2003 HandleInlinedLandingPad(II, &*FirstNewBlock, InlinedFunctionInfo);
2005 HandleInlinedEHPad(II, &*FirstNewBlock, InlinedFunctionInfo);
2009 // Update the lexical scopes of the new funclets and callsites.
2010 // Anything that had 'none' as its parent is now nested inside the callsite's
2013 if (CallSiteEHPad) {
2014 for (Function::iterator BB = FirstNewBlock->getIterator(),
2017 // Add bundle operands to any top-level call sites.
2018 SmallVector<OperandBundleDef, 1> OpBundles;
2019 for (BasicBlock::iterator BBI = BB->begin(), E = BB->end(); BBI != E;) {
2020 Instruction *I = &*BBI++;
2025 // Skip call sites which are nounwind intrinsics.
2027 dyn_cast<Function>(CS.getCalledValue()->stripPointerCasts());
2028 if (CalledFn && CalledFn->isIntrinsic() && CS.doesNotThrow())
2031 // Skip call sites which already have a "funclet" bundle.
2032 if (CS.getOperandBundle(LLVMContext::OB_funclet))
2035 CS.getOperandBundlesAsDefs(OpBundles);
2036 OpBundles.emplace_back("funclet", CallSiteEHPad);
2038 Instruction *NewInst;
2040 NewInst = CallInst::Create(cast<CallInst>(I), OpBundles, I);
2042 NewInst = InvokeInst::Create(cast<InvokeInst>(I), OpBundles, I);
2043 NewInst->takeName(I);
2044 I->replaceAllUsesWith(NewInst);
2045 I->eraseFromParent();
2050 // It is problematic if the inlinee has a cleanupret which unwinds to
2051 // caller and we inline it into a call site which doesn't unwind but into
2052 // an EH pad that does. Such an edge must be dynamically unreachable.
2053 // As such, we replace the cleanupret with unreachable.
2054 if (auto *CleanupRet = dyn_cast<CleanupReturnInst>(BB->getTerminator()))
2055 if (CleanupRet->unwindsToCaller() && EHPadForCallUnwindsLocally)
2056 changeToUnreachable(CleanupRet, /*UseLLVMTrap=*/false);
2058 Instruction *I = BB->getFirstNonPHI();
2062 if (auto *CatchSwitch = dyn_cast<CatchSwitchInst>(I)) {
2063 if (isa<ConstantTokenNone>(CatchSwitch->getParentPad()))
2064 CatchSwitch->setParentPad(CallSiteEHPad);
2066 auto *FPI = cast<FuncletPadInst>(I);
2067 if (isa<ConstantTokenNone>(FPI->getParentPad()))
2068 FPI->setParentPad(CallSiteEHPad);
2073 if (InlinedDeoptimizeCalls) {
2074 // We need to at least remove the deoptimizing returns from the Return set,
2075 // so that the control flow from those returns does not get merged into the
2076 // caller (but terminate it instead). If the caller's return type does not
2077 // match the callee's return type, we also need to change the return type of
2079 if (Caller->getReturnType() == TheCall->getType()) {
2080 auto NewEnd = llvm::remove_if(Returns, [](ReturnInst *RI) {
2081 return RI->getParent()->getTerminatingDeoptimizeCall() != nullptr;
2083 Returns.erase(NewEnd, Returns.end());
2085 SmallVector<ReturnInst *, 8> NormalReturns;
2086 Function *NewDeoptIntrinsic = Intrinsic::getDeclaration(
2087 Caller->getParent(), Intrinsic::experimental_deoptimize,
2088 {Caller->getReturnType()});
2090 for (ReturnInst *RI : Returns) {
2091 CallInst *DeoptCall = RI->getParent()->getTerminatingDeoptimizeCall();
2093 NormalReturns.push_back(RI);
2097 // The calling convention on the deoptimize call itself may be bogus,
2098 // since the code we're inlining may have undefined behavior (and may
2099 // never actually execute at runtime); but all
2100 // @llvm.experimental.deoptimize declarations have to have the same
2101 // calling convention in a well-formed module.
2102 auto CallingConv = DeoptCall->getCalledFunction()->getCallingConv();
2103 NewDeoptIntrinsic->setCallingConv(CallingConv);
2104 auto *CurBB = RI->getParent();
2105 RI->eraseFromParent();
2107 SmallVector<Value *, 4> CallArgs(DeoptCall->arg_begin(),
2108 DeoptCall->arg_end());
2110 SmallVector<OperandBundleDef, 1> OpBundles;
2111 DeoptCall->getOperandBundlesAsDefs(OpBundles);
2112 DeoptCall->eraseFromParent();
2113 assert(!OpBundles.empty() &&
2114 "Expected at least the deopt operand bundle");
2116 IRBuilder<> Builder(CurBB);
2117 CallInst *NewDeoptCall =
2118 Builder.CreateCall(NewDeoptIntrinsic, CallArgs, OpBundles);
2119 NewDeoptCall->setCallingConv(CallingConv);
2120 if (NewDeoptCall->getType()->isVoidTy())
2121 Builder.CreateRetVoid();
2123 Builder.CreateRet(NewDeoptCall);
2126 // Leave behind the normal returns so we can merge control flow.
2127 std::swap(Returns, NormalReturns);
2131 // Handle any inlined musttail call sites. In order for a new call site to be
2132 // musttail, the source of the clone and the inlined call site must have been
2133 // musttail. Therefore it's safe to return without merging control into the
2135 if (InlinedMustTailCalls) {
2136 // Check if we need to bitcast the result of any musttail calls.
2137 Type *NewRetTy = Caller->getReturnType();
2138 bool NeedBitCast = !TheCall->use_empty() && TheCall->getType() != NewRetTy;
2140 // Handle the returns preceded by musttail calls separately.
2141 SmallVector<ReturnInst *, 8> NormalReturns;
2142 for (ReturnInst *RI : Returns) {
2143 CallInst *ReturnedMustTail =
2144 RI->getParent()->getTerminatingMustTailCall();
2145 if (!ReturnedMustTail) {
2146 NormalReturns.push_back(RI);
2152 // Delete the old return and any preceding bitcast.
2153 BasicBlock *CurBB = RI->getParent();
2154 auto *OldCast = dyn_cast_or_null<BitCastInst>(RI->getReturnValue());
2155 RI->eraseFromParent();
2157 OldCast->eraseFromParent();
2159 // Insert a new bitcast and return with the right type.
2160 IRBuilder<> Builder(CurBB);
2161 Builder.CreateRet(Builder.CreateBitCast(ReturnedMustTail, NewRetTy));
2164 // Leave behind the normal returns so we can merge control flow.
2165 std::swap(Returns, NormalReturns);
2168 // Now that all of the transforms on the inlined code have taken place but
2169 // before we splice the inlined code into the CFG and lose track of which
2170 // blocks were actually inlined, collect the call sites. We only do this if
2171 // call graph updates weren't requested, as those provide value handle based
2172 // tracking of inlined call sites instead.
2173 if (InlinedFunctionInfo.ContainsCalls && !IFI.CG) {
2174 // Otherwise just collect the raw call sites that were inlined.
2175 for (BasicBlock &NewBB :
2176 make_range(FirstNewBlock->getIterator(), Caller->end()))
2177 for (Instruction &I : NewBB)
2178 if (auto CS = CallSite(&I))
2179 IFI.InlinedCallSites.push_back(CS);
2182 // If we cloned in _exactly one_ basic block, and if that block ends in a
2183 // return instruction, we splice the body of the inlined callee directly into
2184 // the calling basic block.
2185 if (Returns.size() == 1 && std::distance(FirstNewBlock, Caller->end()) == 1) {
2186 // Move all of the instructions right before the call.
2187 OrigBB->getInstList().splice(TheCall->getIterator(),
2188 FirstNewBlock->getInstList(),
2189 FirstNewBlock->begin(), FirstNewBlock->end());
2190 // Remove the cloned basic block.
2191 Caller->getBasicBlockList().pop_back();
2193 // If the call site was an invoke instruction, add a branch to the normal
2195 if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
2196 BranchInst *NewBr = BranchInst::Create(II->getNormalDest(), TheCall);
2197 NewBr->setDebugLoc(Returns[0]->getDebugLoc());
2200 // If the return instruction returned a value, replace uses of the call with
2201 // uses of the returned value.
2202 if (!TheCall->use_empty()) {
2203 ReturnInst *R = Returns[0];
2204 if (TheCall == R->getReturnValue())
2205 TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
2207 TheCall->replaceAllUsesWith(R->getReturnValue());
2209 // Since we are now done with the Call/Invoke, we can delete it.
2210 TheCall->eraseFromParent();
2212 // Since we are now done with the return instruction, delete it also.
2213 Returns[0]->eraseFromParent();
2215 // We are now done with the inlining.
2219 // Otherwise, we have the normal case, of more than one block to inline or
2220 // multiple return sites.
2222 // We want to clone the entire callee function into the hole between the
2223 // "starter" and "ender" blocks. How we accomplish this depends on whether
2224 // this is an invoke instruction or a call instruction.
2225 BasicBlock *AfterCallBB;
2226 BranchInst *CreatedBranchToNormalDest = nullptr;
2227 if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
2229 // Add an unconditional branch to make this look like the CallInst case...
2230 CreatedBranchToNormalDest = BranchInst::Create(II->getNormalDest(), TheCall);
2232 // Split the basic block. This guarantees that no PHI nodes will have to be
2233 // updated due to new incoming edges, and make the invoke case more
2234 // symmetric to the call case.
2236 OrigBB->splitBasicBlock(CreatedBranchToNormalDest->getIterator(),
2237 CalledFunc->getName() + ".exit");
2239 } else { // It's a call
2240 // If this is a call instruction, we need to split the basic block that
2241 // the call lives in.
2243 AfterCallBB = OrigBB->splitBasicBlock(TheCall->getIterator(),
2244 CalledFunc->getName() + ".exit");
2247 if (IFI.CallerBFI) {
2248 // Copy original BB's block frequency to AfterCallBB
2249 IFI.CallerBFI->setBlockFreq(
2250 AfterCallBB, IFI.CallerBFI->getBlockFreq(OrigBB).getFrequency());
2253 // Change the branch that used to go to AfterCallBB to branch to the first
2254 // basic block of the inlined function.
2256 Instruction *Br = OrigBB->getTerminator();
2257 assert(Br && Br->getOpcode() == Instruction::Br &&
2258 "splitBasicBlock broken!");
2259 Br->setOperand(0, &*FirstNewBlock);
2261 // Now that the function is correct, make it a little bit nicer. In
2262 // particular, move the basic blocks inserted from the end of the function
2263 // into the space made by splitting the source basic block.
2264 Caller->getBasicBlockList().splice(AfterCallBB->getIterator(),
2265 Caller->getBasicBlockList(), FirstNewBlock,
2268 // Handle all of the return instructions that we just cloned in, and eliminate
2269 // any users of the original call/invoke instruction.
2270 Type *RTy = CalledFunc->getReturnType();
2272 PHINode *PHI = nullptr;
2273 if (Returns.size() > 1) {
2274 // The PHI node should go at the front of the new basic block to merge all
2275 // possible incoming values.
2276 if (!TheCall->use_empty()) {
2277 PHI = PHINode::Create(RTy, Returns.size(), TheCall->getName(),
2278 &AfterCallBB->front());
2279 // Anything that used the result of the function call should now use the
2280 // PHI node as their operand.
2281 TheCall->replaceAllUsesWith(PHI);
2284 // Loop over all of the return instructions adding entries to the PHI node
2287 for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
2288 ReturnInst *RI = Returns[i];
2289 assert(RI->getReturnValue()->getType() == PHI->getType() &&
2290 "Ret value not consistent in function!");
2291 PHI->addIncoming(RI->getReturnValue(), RI->getParent());
2295 // Add a branch to the merge points and remove return instructions.
2297 for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
2298 ReturnInst *RI = Returns[i];
2299 BranchInst* BI = BranchInst::Create(AfterCallBB, RI);
2300 Loc = RI->getDebugLoc();
2301 BI->setDebugLoc(Loc);
2302 RI->eraseFromParent();
2304 // We need to set the debug location to *somewhere* inside the
2305 // inlined function. The line number may be nonsensical, but the
2306 // instruction will at least be associated with the right
2308 if (CreatedBranchToNormalDest)
2309 CreatedBranchToNormalDest->setDebugLoc(Loc);
2310 } else if (!Returns.empty()) {
2311 // Otherwise, if there is exactly one return value, just replace anything
2312 // using the return value of the call with the computed value.
2313 if (!TheCall->use_empty()) {
2314 if (TheCall == Returns[0]->getReturnValue())
2315 TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
2317 TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
2320 // Update PHI nodes that use the ReturnBB to use the AfterCallBB.
2321 BasicBlock *ReturnBB = Returns[0]->getParent();
2322 ReturnBB->replaceAllUsesWith(AfterCallBB);
2324 // Splice the code from the return block into the block that it will return
2325 // to, which contains the code that was after the call.
2326 AfterCallBB->getInstList().splice(AfterCallBB->begin(),
2327 ReturnBB->getInstList());
2329 if (CreatedBranchToNormalDest)
2330 CreatedBranchToNormalDest->setDebugLoc(Returns[0]->getDebugLoc());
2332 // Delete the return instruction now and empty ReturnBB now.
2333 Returns[0]->eraseFromParent();
2334 ReturnBB->eraseFromParent();
2335 } else if (!TheCall->use_empty()) {
2336 // No returns, but something is using the return value of the call. Just
2338 TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
2341 // Since we are now done with the Call/Invoke, we can delete it.
2342 TheCall->eraseFromParent();
2344 // If we inlined any musttail calls and the original return is now
2345 // unreachable, delete it. It can only contain a bitcast and ret.
2346 if (InlinedMustTailCalls && pred_begin(AfterCallBB) == pred_end(AfterCallBB))
2347 AfterCallBB->eraseFromParent();
2349 // We should always be able to fold the entry block of the function into the
2350 // single predecessor of the block...
2351 assert(cast<BranchInst>(Br)->isUnconditional() && "splitBasicBlock broken!");
2352 BasicBlock *CalleeEntry = cast<BranchInst>(Br)->getSuccessor(0);
2354 // Splice the code entry block into calling block, right before the
2355 // unconditional branch.
2356 CalleeEntry->replaceAllUsesWith(OrigBB); // Update PHI nodes
2357 OrigBB->getInstList().splice(Br->getIterator(), CalleeEntry->getInstList());
2359 // Remove the unconditional branch.
2360 OrigBB->getInstList().erase(Br);
2362 // Now we can remove the CalleeEntry block, which is now empty.
2363 Caller->getBasicBlockList().erase(CalleeEntry);
2365 // If we inserted a phi node, check to see if it has a single value (e.g. all
2366 // the entries are the same or undef). If so, remove the PHI so it doesn't
2367 // block other optimizations.
2369 AssumptionCache *AC =
2370 IFI.GetAssumptionCache ? &(*IFI.GetAssumptionCache)(*Caller) : nullptr;
2371 auto &DL = Caller->getParent()->getDataLayout();
2372 if (Value *V = SimplifyInstruction(PHI, {DL, nullptr, nullptr, AC})) {
2373 PHI->replaceAllUsesWith(V);
2374 PHI->eraseFromParent();