1 //===- TailRecursionElimination.cpp - Eliminate Tail Calls ----------------===//
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 transforms calls of the current function (self recursion) followed
11 // by a return instruction with a branch to the entry of the function, creating
12 // a loop. This pass also implements the following extensions to the basic
15 // 1. Trivial instructions between the call and return do not prevent the
16 // transformation from taking place, though currently the analysis cannot
17 // support moving any really useful instructions (only dead ones).
18 // 2. This pass transforms functions that are prevented from being tail
19 // recursive by an associative and commutative expression to use an
20 // accumulator variable, thus compiling the typical naive factorial or
21 // 'fib' implementation into efficient code.
22 // 3. TRE is performed if the function returns void, if the return
23 // returns the result returned by the call, or if the function returns a
24 // run-time constant on all exits from the function. It is possible, though
25 // unlikely, that the return returns something else (like constant 0), and
26 // can still be TRE'd. It can be TRE'd if ALL OTHER return instructions in
27 // the function return the exact same value.
28 // 4. If it can prove that callees do not access their caller stack frame,
29 // they are marked as eligible for tail call elimination (by the code
32 // There are several improvements that could be made:
34 // 1. If the function has any alloca instructions, these instructions will be
35 // moved out of the entry block of the function, causing them to be
36 // evaluated each time through the tail recursion. Safely keeping allocas
37 // in the entry block requires analysis to proves that the tail-called
38 // function does not read or write the stack object.
39 // 2. Tail recursion is only performed if the call immediately precedes the
40 // return instruction. It's possible that there could be a jump between
41 // the call and the return.
42 // 3. There can be intervening operations between the call and the return that
43 // prevent the TRE from occurring. For example, there could be GEP's and
44 // stores to memory that will not be read or written by the call. This
45 // requires some substantial analysis (such as with DSA) to prove safe to
46 // move ahead of the call, but doing so could allow many more TREs to be
47 // performed, for example in TreeAdd/TreeAlloc from the treeadd benchmark.
48 // 4. The algorithm we use to detect if callees access their caller stack
49 // frames is very primitive.
51 //===----------------------------------------------------------------------===//
53 #define DEBUG_TYPE "tailcallelim"
54 #include "llvm/Transforms/Scalar.h"
55 #include "llvm/ADT/STLExtras.h"
56 #include "llvm/ADT/Statistic.h"
57 #include "llvm/Analysis/CaptureTracking.h"
58 #include "llvm/Analysis/InlineCost.h"
59 #include "llvm/Analysis/InstructionSimplify.h"
60 #include "llvm/Analysis/Loads.h"
61 #include "llvm/Analysis/TargetTransformInfo.h"
62 #include "llvm/IR/Constants.h"
63 #include "llvm/IR/DerivedTypes.h"
64 #include "llvm/IR/Function.h"
65 #include "llvm/IR/Instructions.h"
66 #include "llvm/IR/IntrinsicInst.h"
67 #include "llvm/IR/Module.h"
68 #include "llvm/Pass.h"
69 #include "llvm/Support/CFG.h"
70 #include "llvm/Support/CallSite.h"
71 #include "llvm/Support/Debug.h"
72 #include "llvm/Support/raw_ostream.h"
73 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
74 #include "llvm/Transforms/Utils/Local.h"
77 STATISTIC(NumEliminated, "Number of tail calls removed");
78 STATISTIC(NumRetDuped, "Number of return duplicated");
79 STATISTIC(NumAccumAdded, "Number of accumulators introduced");
82 struct TailCallElim : public FunctionPass {
83 const TargetTransformInfo *TTI;
85 static char ID; // Pass identification, replacement for typeid
86 TailCallElim() : FunctionPass(ID) {
87 initializeTailCallElimPass(*PassRegistry::getPassRegistry());
90 virtual void getAnalysisUsage(AnalysisUsage &AU) const;
92 virtual bool runOnFunction(Function &F);
95 CallInst *FindTRECandidate(Instruction *I,
96 bool CannotTailCallElimCallsMarkedTail);
97 bool EliminateRecursiveTailCall(CallInst *CI, ReturnInst *Ret,
98 BasicBlock *&OldEntry,
99 bool &TailCallsAreMarkedTail,
100 SmallVector<PHINode*, 8> &ArgumentPHIs,
101 bool CannotTailCallElimCallsMarkedTail);
102 bool FoldReturnAndProcessPred(BasicBlock *BB,
103 ReturnInst *Ret, BasicBlock *&OldEntry,
104 bool &TailCallsAreMarkedTail,
105 SmallVector<PHINode*, 8> &ArgumentPHIs,
106 bool CannotTailCallElimCallsMarkedTail);
107 bool ProcessReturningBlock(ReturnInst *RI, BasicBlock *&OldEntry,
108 bool &TailCallsAreMarkedTail,
109 SmallVector<PHINode*, 8> &ArgumentPHIs,
110 bool CannotTailCallElimCallsMarkedTail);
111 bool CanMoveAboveCall(Instruction *I, CallInst *CI);
112 Value *CanTransformAccumulatorRecursion(Instruction *I, CallInst *CI);
116 char TailCallElim::ID = 0;
117 INITIALIZE_PASS_BEGIN(TailCallElim, "tailcallelim",
118 "Tail Call Elimination", false, false)
119 INITIALIZE_AG_DEPENDENCY(TargetTransformInfo)
120 INITIALIZE_PASS_END(TailCallElim, "tailcallelim",
121 "Tail Call Elimination", false, false)
123 // Public interface to the TailCallElimination pass
124 FunctionPass *llvm::createTailCallEliminationPass() {
125 return new TailCallElim();
128 void TailCallElim::getAnalysisUsage(AnalysisUsage &AU) const {
129 AU.addRequired<TargetTransformInfo>();
132 /// AllocaMightEscapeToCalls - Return true if this alloca may be accessed by
133 /// callees of this function. We only do very simple analysis right now, this
134 /// could be expanded in the future to use mod/ref information for particular
135 /// call sites if desired.
136 static bool AllocaMightEscapeToCalls(AllocaInst *AI) {
137 // FIXME: do simple 'address taken' analysis.
141 /// CheckForEscapingAllocas - Scan the specified basic block for alloca
142 /// instructions. If it contains any that might be accessed by calls, return
144 static bool CheckForEscapingAllocas(BasicBlock *BB,
145 bool &CannotTCETailMarkedCall) {
147 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
148 if (AllocaInst *AI = dyn_cast<AllocaInst>(I)) {
149 RetVal |= AllocaMightEscapeToCalls(AI);
151 // If this alloca is in the body of the function, or if it is a variable
152 // sized allocation, we cannot tail call eliminate calls marked 'tail'
153 // with this mechanism.
154 if (BB != &BB->getParent()->getEntryBlock() ||
155 !isa<ConstantInt>(AI->getArraySize()))
156 CannotTCETailMarkedCall = true;
161 bool TailCallElim::runOnFunction(Function &F) {
162 // If this function is a varargs function, we won't be able to PHI the args
163 // right, so don't even try to convert it...
164 if (F.getFunctionType()->isVarArg()) return false;
166 TTI = &getAnalysis<TargetTransformInfo>();
167 BasicBlock *OldEntry = 0;
168 bool TailCallsAreMarkedTail = false;
169 SmallVector<PHINode*, 8> ArgumentPHIs;
170 bool MadeChange = false;
171 bool FunctionContainsEscapingAllocas = false;
173 // CannotTCETailMarkedCall - If true, we cannot perform TCE on tail calls
174 // marked with the 'tail' attribute, because doing so would cause the stack
175 // size to increase (real TCE would deallocate variable sized allocas, TCE
177 bool CannotTCETailMarkedCall = false;
179 // Loop over the function, looking for any returning blocks, and keeping track
180 // of whether this function has any non-trivially used allocas.
181 for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
182 if (FunctionContainsEscapingAllocas && CannotTCETailMarkedCall)
185 FunctionContainsEscapingAllocas |=
186 CheckForEscapingAllocas(BB, CannotTCETailMarkedCall);
189 /// FIXME: The code generator produces really bad code when an 'escaping
190 /// alloca' is changed from being a static alloca to being a dynamic alloca.
191 /// Until this is resolved, disable this transformation if that would ever
192 /// happen. This bug is PR962.
193 if (FunctionContainsEscapingAllocas)
196 // Second pass, change any tail calls to loops.
197 for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
198 if (ReturnInst *Ret = dyn_cast<ReturnInst>(BB->getTerminator())) {
199 bool Change = ProcessReturningBlock(Ret, OldEntry, TailCallsAreMarkedTail,
200 ArgumentPHIs,CannotTCETailMarkedCall);
201 if (!Change && BB->getFirstNonPHIOrDbg() == Ret)
202 Change = FoldReturnAndProcessPred(BB, Ret, OldEntry,
203 TailCallsAreMarkedTail, ArgumentPHIs,
204 CannotTCETailMarkedCall);
205 MadeChange |= Change;
209 // If we eliminated any tail recursions, it's possible that we inserted some
210 // silly PHI nodes which just merge an initial value (the incoming operand)
211 // with themselves. Check to see if we did and clean up our mess if so. This
212 // occurs when a function passes an argument straight through to its tail
214 if (!ArgumentPHIs.empty()) {
215 for (unsigned i = 0, e = ArgumentPHIs.size(); i != e; ++i) {
216 PHINode *PN = ArgumentPHIs[i];
218 // If the PHI Node is a dynamic constant, replace it with the value it is.
219 if (Value *PNV = SimplifyInstruction(PN)) {
220 PN->replaceAllUsesWith(PNV);
221 PN->eraseFromParent();
226 // Finally, if this function contains no non-escaping allocas, or calls
227 // setjmp, mark all calls in the function as eligible for tail calls
228 //(there is no stack memory for them to access).
229 if (!FunctionContainsEscapingAllocas && !F.callsFunctionThatReturnsTwice())
230 for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
231 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
232 if (CallInst *CI = dyn_cast<CallInst>(I)) {
241 /// CanMoveAboveCall - Return true if it is safe to move the specified
242 /// instruction from after the call to before the call, assuming that all
243 /// instructions between the call and this instruction are movable.
245 bool TailCallElim::CanMoveAboveCall(Instruction *I, CallInst *CI) {
246 // FIXME: We can move load/store/call/free instructions above the call if the
247 // call does not mod/ref the memory location being processed.
248 if (I->mayHaveSideEffects()) // This also handles volatile loads.
251 if (LoadInst *L = dyn_cast<LoadInst>(I)) {
252 // Loads may always be moved above calls without side effects.
253 if (CI->mayHaveSideEffects()) {
254 // Non-volatile loads may be moved above a call with side effects if it
255 // does not write to memory and the load provably won't trap.
256 // FIXME: Writes to memory only matter if they may alias the pointer
257 // being loaded from.
258 if (CI->mayWriteToMemory() ||
259 !isSafeToLoadUnconditionally(L->getPointerOperand(), L,
265 // Otherwise, if this is a side-effect free instruction, check to make sure
266 // that it does not use the return value of the call. If it doesn't use the
267 // return value of the call, it must only use things that are defined before
268 // the call, or movable instructions between the call and the instruction
270 for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i)
271 if (I->getOperand(i) == CI)
276 // isDynamicConstant - Return true if the specified value is the same when the
277 // return would exit as it was when the initial iteration of the recursive
278 // function was executed.
280 // We currently handle static constants and arguments that are not modified as
281 // part of the recursion.
283 static bool isDynamicConstant(Value *V, CallInst *CI, ReturnInst *RI) {
284 if (isa<Constant>(V)) return true; // Static constants are always dyn consts
286 // Check to see if this is an immutable argument, if so, the value
287 // will be available to initialize the accumulator.
288 if (Argument *Arg = dyn_cast<Argument>(V)) {
289 // Figure out which argument number this is...
291 Function *F = CI->getParent()->getParent();
292 for (Function::arg_iterator AI = F->arg_begin(); &*AI != Arg; ++AI)
295 // If we are passing this argument into call as the corresponding
296 // argument operand, then the argument is dynamically constant.
297 // Otherwise, we cannot transform this function safely.
298 if (CI->getArgOperand(ArgNo) == Arg)
302 // Switch cases are always constant integers. If the value is being switched
303 // on and the return is only reachable from one of its cases, it's
304 // effectively constant.
305 if (BasicBlock *UniquePred = RI->getParent()->getUniquePredecessor())
306 if (SwitchInst *SI = dyn_cast<SwitchInst>(UniquePred->getTerminator()))
307 if (SI->getCondition() == V)
308 return SI->getDefaultDest() != RI->getParent();
310 // Not a constant or immutable argument, we can't safely transform.
314 // getCommonReturnValue - Check to see if the function containing the specified
315 // tail call consistently returns the same runtime-constant value at all exit
316 // points except for IgnoreRI. If so, return the returned value.
318 static Value *getCommonReturnValue(ReturnInst *IgnoreRI, CallInst *CI) {
319 Function *F = CI->getParent()->getParent();
320 Value *ReturnedValue = 0;
322 for (Function::iterator BBI = F->begin(), E = F->end(); BBI != E; ++BBI) {
323 ReturnInst *RI = dyn_cast<ReturnInst>(BBI->getTerminator());
324 if (RI == 0 || RI == IgnoreRI) continue;
326 // We can only perform this transformation if the value returned is
327 // evaluatable at the start of the initial invocation of the function,
328 // instead of at the end of the evaluation.
330 Value *RetOp = RI->getOperand(0);
331 if (!isDynamicConstant(RetOp, CI, RI))
334 if (ReturnedValue && RetOp != ReturnedValue)
335 return 0; // Cannot transform if differing values are returned.
336 ReturnedValue = RetOp;
338 return ReturnedValue;
341 /// CanTransformAccumulatorRecursion - If the specified instruction can be
342 /// transformed using accumulator recursion elimination, return the constant
343 /// which is the start of the accumulator value. Otherwise return null.
345 Value *TailCallElim::CanTransformAccumulatorRecursion(Instruction *I,
347 if (!I->isAssociative() || !I->isCommutative()) return 0;
348 assert(I->getNumOperands() == 2 &&
349 "Associative/commutative operations should have 2 args!");
351 // Exactly one operand should be the result of the call instruction.
352 if ((I->getOperand(0) == CI && I->getOperand(1) == CI) ||
353 (I->getOperand(0) != CI && I->getOperand(1) != CI))
356 // The only user of this instruction we allow is a single return instruction.
357 if (!I->hasOneUse() || !isa<ReturnInst>(I->use_back()))
360 // Ok, now we have to check all of the other return instructions in this
361 // function. If they return non-constants or differing values, then we cannot
362 // transform the function safely.
363 return getCommonReturnValue(cast<ReturnInst>(I->use_back()), CI);
366 static Instruction *FirstNonDbg(BasicBlock::iterator I) {
367 while (isa<DbgInfoIntrinsic>(I))
373 TailCallElim::FindTRECandidate(Instruction *TI,
374 bool CannotTailCallElimCallsMarkedTail) {
375 BasicBlock *BB = TI->getParent();
376 Function *F = BB->getParent();
378 if (&BB->front() == TI) // Make sure there is something before the terminator.
381 // Scan backwards from the return, checking to see if there is a tail call in
382 // this block. If so, set CI to it.
384 BasicBlock::iterator BBI = TI;
386 CI = dyn_cast<CallInst>(BBI);
387 if (CI && CI->getCalledFunction() == F)
390 if (BBI == BB->begin())
391 return 0; // Didn't find a potential tail call.
395 // If this call is marked as a tail call, and if there are dynamic allocas in
396 // the function, we cannot perform this optimization.
397 if (CI->isTailCall() && CannotTailCallElimCallsMarkedTail)
400 // As a special case, detect code like this:
401 // double fabs(double f) { return __builtin_fabs(f); } // a 'fabs' call
402 // and disable this xform in this case, because the code generator will
403 // lower the call to fabs into inline code.
404 if (BB == &F->getEntryBlock() &&
405 FirstNonDbg(BB->front()) == CI &&
406 FirstNonDbg(llvm::next(BB->begin())) == TI &&
407 CI->getCalledFunction() &&
408 !TTI->isLoweredToCall(CI->getCalledFunction())) {
409 // A single-block function with just a call and a return. Check that
410 // the arguments match.
411 CallSite::arg_iterator I = CallSite(CI).arg_begin(),
412 E = CallSite(CI).arg_end();
413 Function::arg_iterator FI = F->arg_begin(),
415 for (; I != E && FI != FE; ++I, ++FI)
416 if (*I != &*FI) break;
417 if (I == E && FI == FE)
424 bool TailCallElim::EliminateRecursiveTailCall(CallInst *CI, ReturnInst *Ret,
425 BasicBlock *&OldEntry,
426 bool &TailCallsAreMarkedTail,
427 SmallVector<PHINode*, 8> &ArgumentPHIs,
428 bool CannotTailCallElimCallsMarkedTail) {
429 // If we are introducing accumulator recursion to eliminate operations after
430 // the call instruction that are both associative and commutative, the initial
431 // value for the accumulator is placed in this variable. If this value is set
432 // then we actually perform accumulator recursion elimination instead of
433 // simple tail recursion elimination. If the operation is an LLVM instruction
434 // (eg: "add") then it is recorded in AccumulatorRecursionInstr. If not, then
435 // we are handling the case when the return instruction returns a constant C
436 // which is different to the constant returned by other return instructions
437 // (which is recorded in AccumulatorRecursionEliminationInitVal). This is a
438 // special case of accumulator recursion, the operation being "return C".
439 Value *AccumulatorRecursionEliminationInitVal = 0;
440 Instruction *AccumulatorRecursionInstr = 0;
442 // Ok, we found a potential tail call. We can currently only transform the
443 // tail call if all of the instructions between the call and the return are
444 // movable to above the call itself, leaving the call next to the return.
445 // Check that this is the case now.
446 BasicBlock::iterator BBI = CI;
447 for (++BBI; &*BBI != Ret; ++BBI) {
448 if (CanMoveAboveCall(BBI, CI)) continue;
450 // If we can't move the instruction above the call, it might be because it
451 // is an associative and commutative operation that could be transformed
452 // using accumulator recursion elimination. Check to see if this is the
453 // case, and if so, remember the initial accumulator value for later.
454 if ((AccumulatorRecursionEliminationInitVal =
455 CanTransformAccumulatorRecursion(BBI, CI))) {
456 // Yes, this is accumulator recursion. Remember which instruction
458 AccumulatorRecursionInstr = BBI;
460 return false; // Otherwise, we cannot eliminate the tail recursion!
464 // We can only transform call/return pairs that either ignore the return value
465 // of the call and return void, ignore the value of the call and return a
466 // constant, return the value returned by the tail call, or that are being
467 // accumulator recursion variable eliminated.
468 if (Ret->getNumOperands() == 1 && Ret->getReturnValue() != CI &&
469 !isa<UndefValue>(Ret->getReturnValue()) &&
470 AccumulatorRecursionEliminationInitVal == 0 &&
471 !getCommonReturnValue(0, CI)) {
472 // One case remains that we are able to handle: the current return
473 // instruction returns a constant, and all other return instructions
474 // return a different constant.
475 if (!isDynamicConstant(Ret->getReturnValue(), CI, Ret))
476 return false; // Current return instruction does not return a constant.
477 // Check that all other return instructions return a common constant. If
478 // so, record it in AccumulatorRecursionEliminationInitVal.
479 AccumulatorRecursionEliminationInitVal = getCommonReturnValue(Ret, CI);
480 if (!AccumulatorRecursionEliminationInitVal)
484 BasicBlock *BB = Ret->getParent();
485 Function *F = BB->getParent();
487 // OK! We can transform this tail call. If this is the first one found,
488 // create the new entry block, allowing us to branch back to the old entry.
490 OldEntry = &F->getEntryBlock();
491 BasicBlock *NewEntry = BasicBlock::Create(F->getContext(), "", F, OldEntry);
492 NewEntry->takeName(OldEntry);
493 OldEntry->setName("tailrecurse");
494 BranchInst::Create(OldEntry, NewEntry);
496 // If this tail call is marked 'tail' and if there are any allocas in the
497 // entry block, move them up to the new entry block.
498 TailCallsAreMarkedTail = CI->isTailCall();
499 if (TailCallsAreMarkedTail)
500 // Move all fixed sized allocas from OldEntry to NewEntry.
501 for (BasicBlock::iterator OEBI = OldEntry->begin(), E = OldEntry->end(),
502 NEBI = NewEntry->begin(); OEBI != E; )
503 if (AllocaInst *AI = dyn_cast<AllocaInst>(OEBI++))
504 if (isa<ConstantInt>(AI->getArraySize()))
505 AI->moveBefore(NEBI);
507 // Now that we have created a new block, which jumps to the entry
508 // block, insert a PHI node for each argument of the function.
509 // For now, we initialize each PHI to only have the real arguments
510 // which are passed in.
511 Instruction *InsertPos = OldEntry->begin();
512 for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end();
514 PHINode *PN = PHINode::Create(I->getType(), 2,
515 I->getName() + ".tr", InsertPos);
516 I->replaceAllUsesWith(PN); // Everyone use the PHI node now!
517 PN->addIncoming(I, NewEntry);
518 ArgumentPHIs.push_back(PN);
522 // If this function has self recursive calls in the tail position where some
523 // are marked tail and some are not, only transform one flavor or another. We
524 // have to choose whether we move allocas in the entry block to the new entry
525 // block or not, so we can't make a good choice for both. NOTE: We could do
526 // slightly better here in the case that the function has no entry block
528 if (TailCallsAreMarkedTail && !CI->isTailCall())
531 // Ok, now that we know we have a pseudo-entry block WITH all of the
532 // required PHI nodes, add entries into the PHI node for the actual
533 // parameters passed into the tail-recursive call.
534 for (unsigned i = 0, e = CI->getNumArgOperands(); i != e; ++i)
535 ArgumentPHIs[i]->addIncoming(CI->getArgOperand(i), BB);
537 // If we are introducing an accumulator variable to eliminate the recursion,
538 // do so now. Note that we _know_ that no subsequent tail recursion
539 // eliminations will happen on this function because of the way the
540 // accumulator recursion predicate is set up.
542 if (AccumulatorRecursionEliminationInitVal) {
543 Instruction *AccRecInstr = AccumulatorRecursionInstr;
544 // Start by inserting a new PHI node for the accumulator.
545 pred_iterator PB = pred_begin(OldEntry), PE = pred_end(OldEntry);
547 PHINode::Create(AccumulatorRecursionEliminationInitVal->getType(),
548 std::distance(PB, PE) + 1,
549 "accumulator.tr", OldEntry->begin());
551 // Loop over all of the predecessors of the tail recursion block. For the
552 // real entry into the function we seed the PHI with the initial value,
553 // computed earlier. For any other existing branches to this block (due to
554 // other tail recursions eliminated) the accumulator is not modified.
555 // Because we haven't added the branch in the current block to OldEntry yet,
556 // it will not show up as a predecessor.
557 for (pred_iterator PI = PB; PI != PE; ++PI) {
559 if (P == &F->getEntryBlock())
560 AccPN->addIncoming(AccumulatorRecursionEliminationInitVal, P);
562 AccPN->addIncoming(AccPN, P);
566 // Add an incoming argument for the current block, which is computed by
567 // our associative and commutative accumulator instruction.
568 AccPN->addIncoming(AccRecInstr, BB);
570 // Next, rewrite the accumulator recursion instruction so that it does not
571 // use the result of the call anymore, instead, use the PHI node we just
573 AccRecInstr->setOperand(AccRecInstr->getOperand(0) != CI, AccPN);
575 // Add an incoming argument for the current block, which is just the
576 // constant returned by the current return instruction.
577 AccPN->addIncoming(Ret->getReturnValue(), BB);
580 // Finally, rewrite any return instructions in the program to return the PHI
581 // node instead of the "initval" that they do currently. This loop will
582 // actually rewrite the return value we are destroying, but that's ok.
583 for (Function::iterator BBI = F->begin(), E = F->end(); BBI != E; ++BBI)
584 if (ReturnInst *RI = dyn_cast<ReturnInst>(BBI->getTerminator()))
585 RI->setOperand(0, AccPN);
589 // Now that all of the PHI nodes are in place, remove the call and
590 // ret instructions, replacing them with an unconditional branch.
591 BranchInst *NewBI = BranchInst::Create(OldEntry, Ret);
592 NewBI->setDebugLoc(CI->getDebugLoc());
594 BB->getInstList().erase(Ret); // Remove return.
595 BB->getInstList().erase(CI); // Remove call.
600 bool TailCallElim::FoldReturnAndProcessPred(BasicBlock *BB,
601 ReturnInst *Ret, BasicBlock *&OldEntry,
602 bool &TailCallsAreMarkedTail,
603 SmallVector<PHINode*, 8> &ArgumentPHIs,
604 bool CannotTailCallElimCallsMarkedTail) {
607 // If the return block contains nothing but the return and PHI's,
608 // there might be an opportunity to duplicate the return in its
609 // predecessors and perform TRC there. Look for predecessors that end
610 // in unconditional branch and recursive call(s).
611 SmallVector<BranchInst*, 8> UncondBranchPreds;
612 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
613 BasicBlock *Pred = *PI;
614 TerminatorInst *PTI = Pred->getTerminator();
615 if (BranchInst *BI = dyn_cast<BranchInst>(PTI))
616 if (BI->isUnconditional())
617 UncondBranchPreds.push_back(BI);
620 while (!UncondBranchPreds.empty()) {
621 BranchInst *BI = UncondBranchPreds.pop_back_val();
622 BasicBlock *Pred = BI->getParent();
623 if (CallInst *CI = FindTRECandidate(BI, CannotTailCallElimCallsMarkedTail)){
624 DEBUG(dbgs() << "FOLDING: " << *BB
625 << "INTO UNCOND BRANCH PRED: " << *Pred);
626 EliminateRecursiveTailCall(CI, FoldReturnIntoUncondBranch(Ret, BB, Pred),
627 OldEntry, TailCallsAreMarkedTail, ArgumentPHIs,
628 CannotTailCallElimCallsMarkedTail);
637 bool TailCallElim::ProcessReturningBlock(ReturnInst *Ret, BasicBlock *&OldEntry,
638 bool &TailCallsAreMarkedTail,
639 SmallVector<PHINode*, 8> &ArgumentPHIs,
640 bool CannotTailCallElimCallsMarkedTail) {
641 CallInst *CI = FindTRECandidate(Ret, CannotTailCallElimCallsMarkedTail);
645 return EliminateRecursiveTailCall(CI, Ret, OldEntry, TailCallsAreMarkedTail,
647 CannotTailCallElimCallsMarkedTail);