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/Transforms/Utils/Cloning.h"
16 #include "llvm/ADT/SmallVector.h"
17 #include "llvm/ADT/StringExtras.h"
18 #include "llvm/Analysis/CallGraph.h"
19 #include "llvm/Analysis/InstructionSimplify.h"
20 #include "llvm/DebugInfo.h"
21 #include "llvm/IR/Attributes.h"
22 #include "llvm/IR/Constants.h"
23 #include "llvm/IR/DataLayout.h"
24 #include "llvm/IR/DerivedTypes.h"
25 #include "llvm/IR/IRBuilder.h"
26 #include "llvm/IR/Instructions.h"
27 #include "llvm/IR/IntrinsicInst.h"
28 #include "llvm/IR/Intrinsics.h"
29 #include "llvm/IR/Module.h"
30 #include "llvm/Support/CallSite.h"
31 #include "llvm/Transforms/Utils/Local.h"
34 bool llvm::InlineFunction(CallInst *CI, InlineFunctionInfo &IFI,
35 bool InsertLifetime) {
36 return InlineFunction(CallSite(CI), IFI, InsertLifetime);
38 bool llvm::InlineFunction(InvokeInst *II, InlineFunctionInfo &IFI,
39 bool InsertLifetime) {
40 return InlineFunction(CallSite(II), IFI, InsertLifetime);
44 /// A class for recording information about inlining through an invoke.
45 class InvokeInliningInfo {
46 BasicBlock *OuterResumeDest; ///< Destination of the invoke's unwind.
47 BasicBlock *InnerResumeDest; ///< Destination for the callee's resume.
48 LandingPadInst *CallerLPad; ///< LandingPadInst associated with the invoke.
49 PHINode *InnerEHValuesPHI; ///< PHI for EH values from landingpad insts.
50 SmallVector<Value*, 8> UnwindDestPHIValues;
53 InvokeInliningInfo(InvokeInst *II)
54 : OuterResumeDest(II->getUnwindDest()), InnerResumeDest(0),
55 CallerLPad(0), InnerEHValuesPHI(0) {
56 // If there are PHI nodes in the unwind destination block, we need to keep
57 // track of which values came into them from the invoke before removing
58 // the edge from this block.
59 llvm::BasicBlock *InvokeBB = II->getParent();
60 BasicBlock::iterator I = OuterResumeDest->begin();
61 for (; isa<PHINode>(I); ++I) {
62 // Save the value to use for this edge.
63 PHINode *PHI = cast<PHINode>(I);
64 UnwindDestPHIValues.push_back(PHI->getIncomingValueForBlock(InvokeBB));
67 CallerLPad = cast<LandingPadInst>(I);
70 /// getOuterResumeDest - The outer unwind destination is the target of
71 /// unwind edges introduced for calls within the inlined function.
72 BasicBlock *getOuterResumeDest() const {
73 return OuterResumeDest;
76 BasicBlock *getInnerResumeDest();
78 LandingPadInst *getLandingPadInst() const { return CallerLPad; }
80 /// forwardResume - Forward the 'resume' instruction to the caller's landing
81 /// pad block. When the landing pad block has only one predecessor, this is
82 /// a simple branch. When there is more than one predecessor, we need to
83 /// split the landing pad block after the landingpad instruction and jump
85 void forwardResume(ResumeInst *RI,
86 SmallPtrSet<LandingPadInst*, 16> &InlinedLPads);
88 /// addIncomingPHIValuesFor - Add incoming-PHI values to the unwind
89 /// destination block for the given basic block, using the values for the
90 /// original invoke's source block.
91 void addIncomingPHIValuesFor(BasicBlock *BB) const {
92 addIncomingPHIValuesForInto(BB, OuterResumeDest);
95 void addIncomingPHIValuesForInto(BasicBlock *src, BasicBlock *dest) const {
96 BasicBlock::iterator I = dest->begin();
97 for (unsigned i = 0, e = UnwindDestPHIValues.size(); i != e; ++i, ++I) {
98 PHINode *phi = cast<PHINode>(I);
99 phi->addIncoming(UnwindDestPHIValues[i], src);
105 /// getInnerResumeDest - Get or create a target for the branch from ResumeInsts.
106 BasicBlock *InvokeInliningInfo::getInnerResumeDest() {
107 if (InnerResumeDest) return InnerResumeDest;
109 // Split the landing pad.
110 BasicBlock::iterator SplitPoint = CallerLPad; ++SplitPoint;
112 OuterResumeDest->splitBasicBlock(SplitPoint,
113 OuterResumeDest->getName() + ".body");
115 // The number of incoming edges we expect to the inner landing pad.
116 const unsigned PHICapacity = 2;
118 // Create corresponding new PHIs for all the PHIs in the outer landing pad.
119 BasicBlock::iterator InsertPoint = InnerResumeDest->begin();
120 BasicBlock::iterator I = OuterResumeDest->begin();
121 for (unsigned i = 0, e = UnwindDestPHIValues.size(); i != e; ++i, ++I) {
122 PHINode *OuterPHI = cast<PHINode>(I);
123 PHINode *InnerPHI = PHINode::Create(OuterPHI->getType(), PHICapacity,
124 OuterPHI->getName() + ".lpad-body",
126 OuterPHI->replaceAllUsesWith(InnerPHI);
127 InnerPHI->addIncoming(OuterPHI, OuterResumeDest);
130 // Create a PHI for the exception values.
131 InnerEHValuesPHI = PHINode::Create(CallerLPad->getType(), PHICapacity,
132 "eh.lpad-body", InsertPoint);
133 CallerLPad->replaceAllUsesWith(InnerEHValuesPHI);
134 InnerEHValuesPHI->addIncoming(CallerLPad, OuterResumeDest);
137 return InnerResumeDest;
140 /// forwardResume - Forward the 'resume' instruction to the caller's landing pad
141 /// block. When the landing pad block has only one predecessor, this is a simple
142 /// branch. When there is more than one predecessor, we need to split the
143 /// landing pad block after the landingpad instruction and jump to there.
144 void InvokeInliningInfo::forwardResume(ResumeInst *RI,
145 SmallPtrSet<LandingPadInst*, 16> &InlinedLPads) {
146 BasicBlock *Dest = getInnerResumeDest();
147 LandingPadInst *OuterLPad = getLandingPadInst();
148 BasicBlock *Src = RI->getParent();
150 BranchInst::Create(Dest, Src);
152 // Update the PHIs in the destination. They were inserted in an order which
154 addIncomingPHIValuesForInto(Src, Dest);
156 InnerEHValuesPHI->addIncoming(RI->getOperand(0), Src);
157 RI->eraseFromParent();
159 // Append the clauses from the outer landing pad instruction into the inlined
160 // landing pad instructions.
161 for (SmallPtrSet<LandingPadInst*, 16>::iterator I = InlinedLPads.begin(),
162 E = InlinedLPads.end(); I != E; ++I) {
163 LandingPadInst *InlinedLPad = *I;
164 for (unsigned OuterIdx = 0, OuterNum = OuterLPad->getNumClauses();
165 OuterIdx != OuterNum; ++OuterIdx)
166 InlinedLPad->addClause(OuterLPad->getClause(OuterIdx));
170 /// HandleCallsInBlockInlinedThroughInvoke - When we inline a basic block into
171 /// an invoke, we have to turn all of the calls that can throw into
172 /// invokes. This function analyze BB to see if there are any calls, and if so,
173 /// it rewrites them to be invokes that jump to InvokeDest and fills in the PHI
174 /// nodes in that block with the values specified in InvokeDestPHIValues.
176 /// Returns true to indicate that the next block should be skipped.
177 static bool HandleCallsInBlockInlinedThroughInvoke(BasicBlock *BB,
178 InvokeInliningInfo &Invoke) {
179 LandingPadInst *LPI = Invoke.getLandingPadInst();
181 for (BasicBlock::iterator BBI = BB->begin(), E = BB->end(); BBI != E; ) {
182 Instruction *I = BBI++;
184 if (LandingPadInst *L = dyn_cast<LandingPadInst>(I)) {
185 unsigned NumClauses = LPI->getNumClauses();
186 L->reserveClauses(NumClauses);
187 for (unsigned i = 0; i != NumClauses; ++i)
188 L->addClause(LPI->getClause(i));
191 // We only need to check for function calls: inlined invoke
192 // instructions require no special handling.
193 CallInst *CI = dyn_cast<CallInst>(I);
195 // If this call cannot unwind, don't convert it to an invoke.
196 if (!CI || CI->doesNotThrow())
199 // Convert this function call into an invoke instruction. First, split the
201 BasicBlock *Split = BB->splitBasicBlock(CI, CI->getName()+".noexc");
203 // Delete the unconditional branch inserted by splitBasicBlock
204 BB->getInstList().pop_back();
206 // Create the new invoke instruction.
207 ImmutableCallSite CS(CI);
208 SmallVector<Value*, 8> InvokeArgs(CS.arg_begin(), CS.arg_end());
209 InvokeInst *II = InvokeInst::Create(CI->getCalledValue(), Split,
210 Invoke.getOuterResumeDest(),
211 InvokeArgs, CI->getName(), BB);
212 II->setCallingConv(CI->getCallingConv());
213 II->setAttributes(CI->getAttributes());
215 // Make sure that anything using the call now uses the invoke! This also
216 // updates the CallGraph if present, because it uses a WeakVH.
217 CI->replaceAllUsesWith(II);
219 // Delete the original call
220 Split->getInstList().pop_front();
222 // Update any PHI nodes in the exceptional block to indicate that there is
223 // now a new entry in them.
224 Invoke.addIncomingPHIValuesFor(BB);
231 /// HandleInlinedInvoke - If we inlined an invoke site, we need to convert calls
232 /// in the body of the inlined function into invokes.
234 /// II is the invoke instruction being inlined. FirstNewBlock is the first
235 /// block of the inlined code (the last block is the end of the function),
236 /// and InlineCodeInfo is information about the code that got inlined.
237 static void HandleInlinedInvoke(InvokeInst *II, BasicBlock *FirstNewBlock,
238 ClonedCodeInfo &InlinedCodeInfo) {
239 BasicBlock *InvokeDest = II->getUnwindDest();
241 Function *Caller = FirstNewBlock->getParent();
243 // The inlined code is currently at the end of the function, scan from the
244 // start of the inlined code to its end, checking for stuff we need to
246 InvokeInliningInfo Invoke(II);
248 // Get all of the inlined landing pad instructions.
249 SmallPtrSet<LandingPadInst*, 16> InlinedLPads;
250 for (Function::iterator I = FirstNewBlock, E = Caller->end(); I != E; ++I)
251 if (InvokeInst *II = dyn_cast<InvokeInst>(I->getTerminator()))
252 InlinedLPads.insert(II->getLandingPadInst());
254 for (Function::iterator BB = FirstNewBlock, E = Caller->end(); BB != E; ++BB){
255 if (InlinedCodeInfo.ContainsCalls)
256 if (HandleCallsInBlockInlinedThroughInvoke(BB, Invoke)) {
257 // Honor a request to skip the next block.
262 // Forward any resumes that are remaining here.
263 if (ResumeInst *RI = dyn_cast<ResumeInst>(BB->getTerminator()))
264 Invoke.forwardResume(RI, InlinedLPads);
267 // Now that everything is happy, we have one final detail. The PHI nodes in
268 // the exception destination block still have entries due to the original
269 // invoke instruction. Eliminate these entries (which might even delete the
271 InvokeDest->removePredecessor(II->getParent());
274 /// UpdateCallGraphAfterInlining - Once we have cloned code over from a callee
275 /// into the caller, update the specified callgraph to reflect the changes we
276 /// made. Note that it's possible that not all code was copied over, so only
277 /// some edges of the callgraph may remain.
278 static void UpdateCallGraphAfterInlining(CallSite CS,
279 Function::iterator FirstNewBlock,
280 ValueToValueMapTy &VMap,
281 InlineFunctionInfo &IFI) {
282 CallGraph &CG = *IFI.CG;
283 const Function *Caller = CS.getInstruction()->getParent()->getParent();
284 const Function *Callee = CS.getCalledFunction();
285 CallGraphNode *CalleeNode = CG[Callee];
286 CallGraphNode *CallerNode = CG[Caller];
288 // Since we inlined some uninlined call sites in the callee into the caller,
289 // add edges from the caller to all of the callees of the callee.
290 CallGraphNode::iterator I = CalleeNode->begin(), E = CalleeNode->end();
292 // Consider the case where CalleeNode == CallerNode.
293 CallGraphNode::CalledFunctionsVector CallCache;
294 if (CalleeNode == CallerNode) {
295 CallCache.assign(I, E);
296 I = CallCache.begin();
300 for (; I != E; ++I) {
301 const Value *OrigCall = I->first;
303 ValueToValueMapTy::iterator VMI = VMap.find(OrigCall);
304 // Only copy the edge if the call was inlined!
305 if (VMI == VMap.end() || VMI->second == 0)
308 // If the call was inlined, but then constant folded, there is no edge to
309 // add. Check for this case.
310 Instruction *NewCall = dyn_cast<Instruction>(VMI->second);
311 if (NewCall == 0) continue;
313 // Remember that this call site got inlined for the client of
315 IFI.InlinedCalls.push_back(NewCall);
317 // It's possible that inlining the callsite will cause it to go from an
318 // indirect to a direct call by resolving a function pointer. If this
319 // happens, set the callee of the new call site to a more precise
320 // destination. This can also happen if the call graph node of the caller
321 // was just unnecessarily imprecise.
322 if (I->second->getFunction() == 0)
323 if (Function *F = CallSite(NewCall).getCalledFunction()) {
324 // Indirect call site resolved to direct call.
325 CallerNode->addCalledFunction(CallSite(NewCall), CG[F]);
330 CallerNode->addCalledFunction(CallSite(NewCall), I->second);
333 // Update the call graph by deleting the edge from Callee to Caller. We must
334 // do this after the loop above in case Caller and Callee are the same.
335 CallerNode->removeCallEdgeFor(CS);
338 /// HandleByValArgument - When inlining a call site that has a byval argument,
339 /// we have to make the implicit memcpy explicit by adding it.
340 static Value *HandleByValArgument(Value *Arg, Instruction *TheCall,
341 const Function *CalledFunc,
342 InlineFunctionInfo &IFI,
343 unsigned ByValAlignment) {
344 Type *AggTy = cast<PointerType>(Arg->getType())->getElementType();
346 // If the called function is readonly, then it could not mutate the caller's
347 // copy of the byval'd memory. In this case, it is safe to elide the copy and
349 if (CalledFunc->onlyReadsMemory()) {
350 // If the byval argument has a specified alignment that is greater than the
351 // passed in pointer, then we either have to round up the input pointer or
352 // give up on this transformation.
353 if (ByValAlignment <= 1) // 0 = unspecified, 1 = no particular alignment.
356 // If the pointer is already known to be sufficiently aligned, or if we can
357 // round it up to a larger alignment, then we don't need a temporary.
358 if (getOrEnforceKnownAlignment(Arg, ByValAlignment,
359 IFI.TD) >= ByValAlignment)
362 // Otherwise, we have to make a memcpy to get a safe alignment. This is bad
363 // for code quality, but rarely happens and is required for correctness.
366 LLVMContext &Context = Arg->getContext();
368 Type *VoidPtrTy = Type::getInt8PtrTy(Context);
370 // Create the alloca. If we have DataLayout, use nice alignment.
373 Align = IFI.TD->getPrefTypeAlignment(AggTy);
375 // If the byval had an alignment specified, we *must* use at least that
376 // alignment, as it is required by the byval argument (and uses of the
377 // pointer inside the callee).
378 Align = std::max(Align, ByValAlignment);
380 Function *Caller = TheCall->getParent()->getParent();
382 Value *NewAlloca = new AllocaInst(AggTy, 0, Align, Arg->getName(),
383 &*Caller->begin()->begin());
385 Type *Tys[3] = {VoidPtrTy, VoidPtrTy, Type::getInt64Ty(Context)};
386 Function *MemCpyFn = Intrinsic::getDeclaration(Caller->getParent(),
389 Value *DestCast = new BitCastInst(NewAlloca, VoidPtrTy, "tmp", TheCall);
390 Value *SrcCast = new BitCastInst(Arg, VoidPtrTy, "tmp", TheCall);
394 Size = ConstantExpr::getSizeOf(AggTy);
396 Size = ConstantInt::get(Type::getInt64Ty(Context),
397 IFI.TD->getTypeStoreSize(AggTy));
399 // Always generate a memcpy of alignment 1 here because we don't know
400 // the alignment of the src pointer. Other optimizations can infer
402 Value *CallArgs[] = {
403 DestCast, SrcCast, Size,
404 ConstantInt::get(Type::getInt32Ty(Context), 1),
405 ConstantInt::getFalse(Context) // isVolatile
407 IRBuilder<>(TheCall).CreateCall(MemCpyFn, CallArgs);
409 // Uses of the argument in the function should use our new alloca
414 // isUsedByLifetimeMarker - Check whether this Value is used by a lifetime
416 static bool isUsedByLifetimeMarker(Value *V) {
417 for (Value::use_iterator UI = V->use_begin(), UE = V->use_end(); UI != UE;
419 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(*UI)) {
420 switch (II->getIntrinsicID()) {
422 case Intrinsic::lifetime_start:
423 case Intrinsic::lifetime_end:
431 // hasLifetimeMarkers - Check whether the given alloca already has
432 // lifetime.start or lifetime.end intrinsics.
433 static bool hasLifetimeMarkers(AllocaInst *AI) {
434 Type *Int8PtrTy = Type::getInt8PtrTy(AI->getType()->getContext());
435 if (AI->getType() == Int8PtrTy)
436 return isUsedByLifetimeMarker(AI);
438 // Do a scan to find all the casts to i8*.
439 for (Value::use_iterator I = AI->use_begin(), E = AI->use_end(); I != E;
441 if (I->getType() != Int8PtrTy) continue;
442 if (I->stripPointerCasts() != AI) continue;
443 if (isUsedByLifetimeMarker(*I))
449 /// updateInlinedAtInfo - Helper function used by fixupLineNumbers to
450 /// recursively update InlinedAtEntry of a DebugLoc.
451 static DebugLoc updateInlinedAtInfo(const DebugLoc &DL,
452 const DebugLoc &InlinedAtDL,
454 if (MDNode *IA = DL.getInlinedAt(Ctx)) {
455 DebugLoc NewInlinedAtDL
456 = updateInlinedAtInfo(DebugLoc::getFromDILocation(IA), InlinedAtDL, Ctx);
457 return DebugLoc::get(DL.getLine(), DL.getCol(), DL.getScope(Ctx),
458 NewInlinedAtDL.getAsMDNode(Ctx));
461 return DebugLoc::get(DL.getLine(), DL.getCol(), DL.getScope(Ctx),
462 InlinedAtDL.getAsMDNode(Ctx));
465 /// fixupLineNumbers - Update inlined instructions' line numbers to
466 /// to encode location where these instructions are inlined.
467 static void fixupLineNumbers(Function *Fn, Function::iterator FI,
468 Instruction *TheCall) {
469 DebugLoc TheCallDL = TheCall->getDebugLoc();
470 if (TheCallDL.isUnknown())
473 for (; FI != Fn->end(); ++FI) {
474 for (BasicBlock::iterator BI = FI->begin(), BE = FI->end();
476 DebugLoc DL = BI->getDebugLoc();
477 if (!DL.isUnknown()) {
478 BI->setDebugLoc(updateInlinedAtInfo(DL, TheCallDL, BI->getContext()));
479 if (DbgValueInst *DVI = dyn_cast<DbgValueInst>(BI)) {
480 LLVMContext &Ctx = BI->getContext();
481 MDNode *InlinedAt = BI->getDebugLoc().getInlinedAt(Ctx);
482 DVI->setOperand(2, createInlinedVariable(DVI->getVariable(),
490 /// InlineFunction - This function inlines the called function into the basic
491 /// block of the caller. This returns false if it is not possible to inline
492 /// this call. The program is still in a well defined state if this occurs
495 /// Note that this only does one level of inlining. For example, if the
496 /// instruction 'call B' is inlined, and 'B' calls 'C', then the call to 'C' now
497 /// exists in the instruction stream. Similarly this will inline a recursive
498 /// function by one level.
499 bool llvm::InlineFunction(CallSite CS, InlineFunctionInfo &IFI,
500 bool InsertLifetime) {
501 Instruction *TheCall = CS.getInstruction();
502 assert(TheCall->getParent() && TheCall->getParent()->getParent() &&
503 "Instruction not in function!");
505 // If IFI has any state in it, zap it before we fill it in.
508 const Function *CalledFunc = CS.getCalledFunction();
509 if (CalledFunc == 0 || // Can't inline external function or indirect
510 CalledFunc->isDeclaration() || // call, or call to a vararg function!
511 CalledFunc->getFunctionType()->isVarArg()) return false;
513 // If the call to the callee is not a tail call, we must clear the 'tail'
514 // flags on any calls that we inline.
515 bool MustClearTailCallFlags =
516 !(isa<CallInst>(TheCall) && cast<CallInst>(TheCall)->isTailCall());
518 // If the call to the callee cannot throw, set the 'nounwind' flag on any
519 // calls that we inline.
520 bool MarkNoUnwind = CS.doesNotThrow();
522 BasicBlock *OrigBB = TheCall->getParent();
523 Function *Caller = OrigBB->getParent();
525 // GC poses two hazards to inlining, which only occur when the callee has GC:
526 // 1. If the caller has no GC, then the callee's GC must be propagated to the
528 // 2. If the caller has a differing GC, it is invalid to inline.
529 if (CalledFunc->hasGC()) {
530 if (!Caller->hasGC())
531 Caller->setGC(CalledFunc->getGC());
532 else if (CalledFunc->getGC() != Caller->getGC())
536 // Get the personality function from the callee if it contains a landing pad.
537 Value *CalleePersonality = 0;
538 for (Function::const_iterator I = CalledFunc->begin(), E = CalledFunc->end();
540 if (const InvokeInst *II = dyn_cast<InvokeInst>(I->getTerminator())) {
541 const BasicBlock *BB = II->getUnwindDest();
542 const LandingPadInst *LP = BB->getLandingPadInst();
543 CalleePersonality = LP->getPersonalityFn();
547 // Find the personality function used by the landing pads of the caller. If it
548 // exists, then check to see that it matches the personality function used in
550 if (CalleePersonality) {
551 for (Function::const_iterator I = Caller->begin(), E = Caller->end();
553 if (const InvokeInst *II = dyn_cast<InvokeInst>(I->getTerminator())) {
554 const BasicBlock *BB = II->getUnwindDest();
555 const LandingPadInst *LP = BB->getLandingPadInst();
557 // If the personality functions match, then we can perform the
558 // inlining. Otherwise, we can't inline.
559 // TODO: This isn't 100% true. Some personality functions are proper
560 // supersets of others and can be used in place of the other.
561 if (LP->getPersonalityFn() != CalleePersonality)
568 // Get an iterator to the last basic block in the function, which will have
569 // the new function inlined after it.
570 Function::iterator LastBlock = &Caller->back();
572 // Make sure to capture all of the return instructions from the cloned
574 SmallVector<ReturnInst*, 8> Returns;
575 ClonedCodeInfo InlinedFunctionInfo;
576 Function::iterator FirstNewBlock;
578 { // Scope to destroy VMap after cloning.
579 ValueToValueMapTy VMap;
581 assert(CalledFunc->arg_size() == CS.arg_size() &&
582 "No varargs calls can be inlined!");
584 // Calculate the vector of arguments to pass into the function cloner, which
585 // matches up the formal to the actual argument values.
586 CallSite::arg_iterator AI = CS.arg_begin();
588 for (Function::const_arg_iterator I = CalledFunc->arg_begin(),
589 E = CalledFunc->arg_end(); I != E; ++I, ++AI, ++ArgNo) {
590 Value *ActualArg = *AI;
592 // When byval arguments actually inlined, we need to make the copy implied
593 // by them explicit. However, we don't do this if the callee is readonly
594 // or readnone, because the copy would be unneeded: the callee doesn't
595 // modify the struct.
596 if (CS.isByValArgument(ArgNo)) {
597 ActualArg = HandleByValArgument(ActualArg, TheCall, CalledFunc, IFI,
598 CalledFunc->getParamAlignment(ArgNo+1));
600 // Calls that we inline may use the new alloca, so we need to clear
601 // their 'tail' flags if HandleByValArgument introduced a new alloca and
602 // the callee has calls.
603 MustClearTailCallFlags |= ActualArg != *AI;
609 // We want the inliner to prune the code as it copies. We would LOVE to
610 // have no dead or constant instructions leftover after inlining occurs
611 // (which can happen, e.g., because an argument was constant), but we'll be
612 // happy with whatever the cloner can do.
613 CloneAndPruneFunctionInto(Caller, CalledFunc, VMap,
614 /*ModuleLevelChanges=*/false, Returns, ".i",
615 &InlinedFunctionInfo, IFI.TD, TheCall);
617 // Remember the first block that is newly cloned over.
618 FirstNewBlock = LastBlock; ++FirstNewBlock;
620 // Update the callgraph if requested.
622 UpdateCallGraphAfterInlining(CS, FirstNewBlock, VMap, IFI);
624 // Update inlined instructions' line number information.
625 fixupLineNumbers(Caller, FirstNewBlock, TheCall);
628 // If there are any alloca instructions in the block that used to be the entry
629 // block for the callee, move them to the entry block of the caller. First
630 // calculate which instruction they should be inserted before. We insert the
631 // instructions at the end of the current alloca list.
633 BasicBlock::iterator InsertPoint = Caller->begin()->begin();
634 for (BasicBlock::iterator I = FirstNewBlock->begin(),
635 E = FirstNewBlock->end(); I != E; ) {
636 AllocaInst *AI = dyn_cast<AllocaInst>(I++);
637 if (AI == 0) continue;
639 // If the alloca is now dead, remove it. This often occurs due to code
641 if (AI->use_empty()) {
642 AI->eraseFromParent();
646 if (!isa<Constant>(AI->getArraySize()))
649 // Keep track of the static allocas that we inline into the caller.
650 IFI.StaticAllocas.push_back(AI);
652 // Scan for the block of allocas that we can move over, and move them
654 while (isa<AllocaInst>(I) &&
655 isa<Constant>(cast<AllocaInst>(I)->getArraySize())) {
656 IFI.StaticAllocas.push_back(cast<AllocaInst>(I));
660 // Transfer all of the allocas over in a block. Using splice means
661 // that the instructions aren't removed from the symbol table, then
663 Caller->getEntryBlock().getInstList().splice(InsertPoint,
664 FirstNewBlock->getInstList(),
669 // Leave lifetime markers for the static alloca's, scoping them to the
670 // function we just inlined.
671 if (InsertLifetime && !IFI.StaticAllocas.empty()) {
672 IRBuilder<> builder(FirstNewBlock->begin());
673 for (unsigned ai = 0, ae = IFI.StaticAllocas.size(); ai != ae; ++ai) {
674 AllocaInst *AI = IFI.StaticAllocas[ai];
676 // If the alloca is already scoped to something smaller than the whole
677 // function then there's no need to add redundant, less accurate markers.
678 if (hasLifetimeMarkers(AI))
681 // Try to determine the size of the allocation.
682 ConstantInt *AllocaSize = 0;
683 if (ConstantInt *AIArraySize =
684 dyn_cast<ConstantInt>(AI->getArraySize())) {
686 Type *AllocaType = AI->getAllocatedType();
687 uint64_t AllocaTypeSize = IFI.TD->getTypeAllocSize(AllocaType);
688 uint64_t AllocaArraySize = AIArraySize->getLimitedValue();
689 assert(AllocaArraySize > 0 && "array size of AllocaInst is zero");
690 // Check that array size doesn't saturate uint64_t and doesn't
691 // overflow when it's multiplied by type size.
692 if (AllocaArraySize != ~0ULL &&
693 UINT64_MAX / AllocaArraySize >= AllocaTypeSize) {
694 AllocaSize = ConstantInt::get(Type::getInt64Ty(AI->getContext()),
695 AllocaArraySize * AllocaTypeSize);
700 builder.CreateLifetimeStart(AI, AllocaSize);
701 for (unsigned ri = 0, re = Returns.size(); ri != re; ++ri) {
702 IRBuilder<> builder(Returns[ri]);
703 builder.CreateLifetimeEnd(AI, AllocaSize);
708 // If the inlined code contained dynamic alloca instructions, wrap the inlined
709 // code with llvm.stacksave/llvm.stackrestore intrinsics.
710 if (InlinedFunctionInfo.ContainsDynamicAllocas) {
711 Module *M = Caller->getParent();
712 // Get the two intrinsics we care about.
713 Function *StackSave = Intrinsic::getDeclaration(M, Intrinsic::stacksave);
714 Function *StackRestore=Intrinsic::getDeclaration(M,Intrinsic::stackrestore);
716 // Insert the llvm.stacksave.
717 CallInst *SavedPtr = IRBuilder<>(FirstNewBlock, FirstNewBlock->begin())
718 .CreateCall(StackSave, "savedstack");
720 // Insert a call to llvm.stackrestore before any return instructions in the
722 for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
723 IRBuilder<>(Returns[i]).CreateCall(StackRestore, SavedPtr);
727 // If we are inlining tail call instruction through a call site that isn't
728 // marked 'tail', we must remove the tail marker for any calls in the inlined
729 // code. Also, calls inlined through a 'nounwind' call site should be marked
731 if (InlinedFunctionInfo.ContainsCalls &&
732 (MustClearTailCallFlags || MarkNoUnwind)) {
733 for (Function::iterator BB = FirstNewBlock, E = Caller->end();
735 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
736 if (CallInst *CI = dyn_cast<CallInst>(I)) {
737 if (MustClearTailCallFlags)
738 CI->setTailCall(false);
740 CI->setDoesNotThrow();
744 // If we are inlining for an invoke instruction, we must make sure to rewrite
745 // any call instructions into invoke instructions.
746 if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall))
747 HandleInlinedInvoke(II, FirstNewBlock, InlinedFunctionInfo);
749 // If we cloned in _exactly one_ basic block, and if that block ends in a
750 // return instruction, we splice the body of the inlined callee directly into
751 // the calling basic block.
752 if (Returns.size() == 1 && std::distance(FirstNewBlock, Caller->end()) == 1) {
753 // Move all of the instructions right before the call.
754 OrigBB->getInstList().splice(TheCall, FirstNewBlock->getInstList(),
755 FirstNewBlock->begin(), FirstNewBlock->end());
756 // Remove the cloned basic block.
757 Caller->getBasicBlockList().pop_back();
759 // If the call site was an invoke instruction, add a branch to the normal
761 if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
762 BranchInst *NewBr = BranchInst::Create(II->getNormalDest(), TheCall);
763 NewBr->setDebugLoc(Returns[0]->getDebugLoc());
766 // If the return instruction returned a value, replace uses of the call with
767 // uses of the returned value.
768 if (!TheCall->use_empty()) {
769 ReturnInst *R = Returns[0];
770 if (TheCall == R->getReturnValue())
771 TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
773 TheCall->replaceAllUsesWith(R->getReturnValue());
775 // Since we are now done with the Call/Invoke, we can delete it.
776 TheCall->eraseFromParent();
778 // Since we are now done with the return instruction, delete it also.
779 Returns[0]->eraseFromParent();
781 // We are now done with the inlining.
785 // Otherwise, we have the normal case, of more than one block to inline or
786 // multiple return sites.
788 // We want to clone the entire callee function into the hole between the
789 // "starter" and "ender" blocks. How we accomplish this depends on whether
790 // this is an invoke instruction or a call instruction.
791 BasicBlock *AfterCallBB;
792 BranchInst *CreatedBranchToNormalDest = NULL;
793 if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
795 // Add an unconditional branch to make this look like the CallInst case...
796 CreatedBranchToNormalDest = BranchInst::Create(II->getNormalDest(), TheCall);
798 // Split the basic block. This guarantees that no PHI nodes will have to be
799 // updated due to new incoming edges, and make the invoke case more
800 // symmetric to the call case.
801 AfterCallBB = OrigBB->splitBasicBlock(CreatedBranchToNormalDest,
802 CalledFunc->getName()+".exit");
804 } else { // It's a call
805 // If this is a call instruction, we need to split the basic block that
806 // the call lives in.
808 AfterCallBB = OrigBB->splitBasicBlock(TheCall,
809 CalledFunc->getName()+".exit");
812 // Change the branch that used to go to AfterCallBB to branch to the first
813 // basic block of the inlined function.
815 TerminatorInst *Br = OrigBB->getTerminator();
816 assert(Br && Br->getOpcode() == Instruction::Br &&
817 "splitBasicBlock broken!");
818 Br->setOperand(0, FirstNewBlock);
821 // Now that the function is correct, make it a little bit nicer. In
822 // particular, move the basic blocks inserted from the end of the function
823 // into the space made by splitting the source basic block.
824 Caller->getBasicBlockList().splice(AfterCallBB, Caller->getBasicBlockList(),
825 FirstNewBlock, Caller->end());
827 // Handle all of the return instructions that we just cloned in, and eliminate
828 // any users of the original call/invoke instruction.
829 Type *RTy = CalledFunc->getReturnType();
832 if (Returns.size() > 1) {
833 // The PHI node should go at the front of the new basic block to merge all
834 // possible incoming values.
835 if (!TheCall->use_empty()) {
836 PHI = PHINode::Create(RTy, Returns.size(), TheCall->getName(),
837 AfterCallBB->begin());
838 // Anything that used the result of the function call should now use the
839 // PHI node as their operand.
840 TheCall->replaceAllUsesWith(PHI);
843 // Loop over all of the return instructions adding entries to the PHI node
846 for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
847 ReturnInst *RI = Returns[i];
848 assert(RI->getReturnValue()->getType() == PHI->getType() &&
849 "Ret value not consistent in function!");
850 PHI->addIncoming(RI->getReturnValue(), RI->getParent());
855 // Add a branch to the merge points and remove return instructions.
857 for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
858 ReturnInst *RI = Returns[i];
859 BranchInst* BI = BranchInst::Create(AfterCallBB, RI);
860 Loc = RI->getDebugLoc();
861 BI->setDebugLoc(Loc);
862 RI->eraseFromParent();
864 // We need to set the debug location to *somewhere* inside the
865 // inlined function. The line number may be nonsensical, but the
866 // instruction will at least be associated with the right
868 if (CreatedBranchToNormalDest)
869 CreatedBranchToNormalDest->setDebugLoc(Loc);
870 } else if (!Returns.empty()) {
871 // Otherwise, if there is exactly one return value, just replace anything
872 // using the return value of the call with the computed value.
873 if (!TheCall->use_empty()) {
874 if (TheCall == Returns[0]->getReturnValue())
875 TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
877 TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
880 // Update PHI nodes that use the ReturnBB to use the AfterCallBB.
881 BasicBlock *ReturnBB = Returns[0]->getParent();
882 ReturnBB->replaceAllUsesWith(AfterCallBB);
884 // Splice the code from the return block into the block that it will return
885 // to, which contains the code that was after the call.
886 AfterCallBB->getInstList().splice(AfterCallBB->begin(),
887 ReturnBB->getInstList());
889 if (CreatedBranchToNormalDest)
890 CreatedBranchToNormalDest->setDebugLoc(Returns[0]->getDebugLoc());
892 // Delete the return instruction now and empty ReturnBB now.
893 Returns[0]->eraseFromParent();
894 ReturnBB->eraseFromParent();
895 } else if (!TheCall->use_empty()) {
896 // No returns, but something is using the return value of the call. Just
898 TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
901 // Since we are now done with the Call/Invoke, we can delete it.
902 TheCall->eraseFromParent();
904 // We should always be able to fold the entry block of the function into the
905 // single predecessor of the block...
906 assert(cast<BranchInst>(Br)->isUnconditional() && "splitBasicBlock broken!");
907 BasicBlock *CalleeEntry = cast<BranchInst>(Br)->getSuccessor(0);
909 // Splice the code entry block into calling block, right before the
910 // unconditional branch.
911 CalleeEntry->replaceAllUsesWith(OrigBB); // Update PHI nodes
912 OrigBB->getInstList().splice(Br, CalleeEntry->getInstList());
914 // Remove the unconditional branch.
915 OrigBB->getInstList().erase(Br);
917 // Now we can remove the CalleeEntry block, which is now empty.
918 Caller->getBasicBlockList().erase(CalleeEntry);
920 // If we inserted a phi node, check to see if it has a single value (e.g. all
921 // the entries are the same or undef). If so, remove the PHI so it doesn't
922 // block other optimizations.
924 if (Value *V = SimplifyInstruction(PHI, IFI.TD)) {
925 PHI->replaceAllUsesWith(V);
926 PHI->eraseFromParent();