LLVM 19.0.0git
TailRecursionElimination.cpp
Go to the documentation of this file.
1//===- TailRecursionElimination.cpp - Eliminate Tail Calls ----------------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file transforms calls of the current function (self recursion) followed
10// by a return instruction with a branch to the entry of the function, creating
11// a loop. This pass also implements the following extensions to the basic
12// algorithm:
13//
14// 1. Trivial instructions between the call and return do not prevent the
15// transformation from taking place, though currently the analysis cannot
16// support moving any really useful instructions (only dead ones).
17// 2. This pass transforms functions that are prevented from being tail
18// recursive by an associative and commutative expression to use an
19// accumulator variable, thus compiling the typical naive factorial or
20// 'fib' implementation into efficient code.
21// 3. TRE is performed if the function returns void, if the return
22// returns the result returned by the call, or if the function returns a
23// run-time constant on all exits from the function. It is possible, though
24// unlikely, that the return returns something else (like constant 0), and
25// can still be TRE'd. It can be TRE'd if ALL OTHER return instructions in
26// the function return the exact same value.
27// 4. If it can prove that callees do not access their caller stack frame,
28// they are marked as eligible for tail call elimination (by the code
29// generator).
30//
31// There are several improvements that could be made:
32//
33// 1. If the function has any alloca instructions, these instructions will be
34// moved out of the entry block of the function, causing them to be
35// evaluated each time through the tail recursion. Safely keeping allocas
36// in the entry block requires analysis to proves that the tail-called
37// function does not read or write the stack object.
38// 2. Tail recursion is only performed if the call immediately precedes the
39// return instruction. It's possible that there could be a jump between
40// the call and the return.
41// 3. There can be intervening operations between the call and the return that
42// prevent the TRE from occurring. For example, there could be GEP's and
43// stores to memory that will not be read or written by the call. This
44// requires some substantial analysis (such as with DSA) to prove safe to
45// move ahead of the call, but doing so could allow many more TREs to be
46// performed, for example in TreeAdd/TreeAlloc from the treeadd benchmark.
47// 4. The algorithm we use to detect if callees access their caller stack
48// frames is very primitive.
49//
50//===----------------------------------------------------------------------===//
51
53#include "llvm/ADT/STLExtras.h"
55#include "llvm/ADT/Statistic.h"
59#include "llvm/Analysis/Loads.h"
64#include "llvm/IR/CFG.h"
65#include "llvm/IR/Constants.h"
66#include "llvm/IR/DataLayout.h"
69#include "llvm/IR/Dominators.h"
70#include "llvm/IR/Function.h"
71#include "llvm/IR/IRBuilder.h"
75#include "llvm/IR/Module.h"
77#include "llvm/Pass.h"
78#include "llvm/Support/Debug.h"
82using namespace llvm;
83
84#define DEBUG_TYPE "tailcallelim"
85
86STATISTIC(NumEliminated, "Number of tail calls removed");
87STATISTIC(NumRetDuped, "Number of return duplicated");
88STATISTIC(NumAccumAdded, "Number of accumulators introduced");
89
90/// Scan the specified function for alloca instructions.
91/// If it contains any dynamic allocas, returns false.
92static bool canTRE(Function &F) {
93 // TODO: We don't do TRE if dynamic allocas are used.
94 // Dynamic allocas allocate stack space which should be
95 // deallocated before new iteration started. That is
96 // currently not implemented.
98 auto *AI = dyn_cast<AllocaInst>(&I);
99 return !AI || AI->isStaticAlloca();
100 });
101}
102
103namespace {
104struct AllocaDerivedValueTracker {
105 // Start at a root value and walk its use-def chain to mark calls that use the
106 // value or a derived value in AllocaUsers, and places where it may escape in
107 // EscapePoints.
108 void walk(Value *Root) {
109 SmallVector<Use *, 32> Worklist;
111
112 auto AddUsesToWorklist = [&](Value *V) {
113 for (auto &U : V->uses()) {
114 if (!Visited.insert(&U).second)
115 continue;
116 Worklist.push_back(&U);
117 }
118 };
119
120 AddUsesToWorklist(Root);
121
122 while (!Worklist.empty()) {
123 Use *U = Worklist.pop_back_val();
124 Instruction *I = cast<Instruction>(U->getUser());
125
126 switch (I->getOpcode()) {
127 case Instruction::Call:
128 case Instruction::Invoke: {
129 auto &CB = cast<CallBase>(*I);
130 // If the alloca-derived argument is passed byval it is not an escape
131 // point, or a use of an alloca. Calling with byval copies the contents
132 // of the alloca into argument registers or stack slots, which exist
133 // beyond the lifetime of the current frame.
134 if (CB.isArgOperand(U) && CB.isByValArgument(CB.getArgOperandNo(U)))
135 continue;
136 bool IsNocapture =
137 CB.isDataOperand(U) && CB.doesNotCapture(CB.getDataOperandNo(U));
138 callUsesLocalStack(CB, IsNocapture);
139 if (IsNocapture) {
140 // If the alloca-derived argument is passed in as nocapture, then it
141 // can't propagate to the call's return. That would be capturing.
142 continue;
143 }
144 break;
145 }
146 case Instruction::Load: {
147 // The result of a load is not alloca-derived (unless an alloca has
148 // otherwise escaped, but this is a local analysis).
149 continue;
150 }
151 case Instruction::Store: {
152 if (U->getOperandNo() == 0)
153 EscapePoints.insert(I);
154 continue; // Stores have no users to analyze.
155 }
156 case Instruction::BitCast:
157 case Instruction::GetElementPtr:
158 case Instruction::PHI:
159 case Instruction::Select:
160 case Instruction::AddrSpaceCast:
161 break;
162 default:
163 EscapePoints.insert(I);
164 break;
165 }
166
167 AddUsesToWorklist(I);
168 }
169 }
170
171 void callUsesLocalStack(CallBase &CB, bool IsNocapture) {
172 // Add it to the list of alloca users.
173 AllocaUsers.insert(&CB);
174
175 // If it's nocapture then it can't capture this alloca.
176 if (IsNocapture)
177 return;
178
179 // If it can write to memory, it can leak the alloca value.
180 if (!CB.onlyReadsMemory())
181 EscapePoints.insert(&CB);
182 }
183
186};
187}
188
190 if (F.callsFunctionThatReturnsTwice())
191 return false;
192
193 // The local stack holds all alloca instructions and all byval arguments.
194 AllocaDerivedValueTracker Tracker;
195 for (Argument &Arg : F.args()) {
196 if (Arg.hasByValAttr())
197 Tracker.walk(&Arg);
198 }
199 for (auto &BB : F) {
200 for (auto &I : BB)
201 if (AllocaInst *AI = dyn_cast<AllocaInst>(&I))
202 Tracker.walk(AI);
203 }
204
205 bool Modified = false;
206
207 // Track whether a block is reachable after an alloca has escaped. Blocks that
208 // contain the escaping instruction will be marked as being visited without an
209 // escaped alloca, since that is how the block began.
210 enum VisitType {
211 UNVISITED,
212 UNESCAPED,
213 ESCAPED
214 };
216
217 // We propagate the fact that an alloca has escaped from block to successor.
218 // Visit the blocks that are propagating the escapedness first. To do this, we
219 // maintain two worklists.
220 SmallVector<BasicBlock *, 32> WorklistUnescaped, WorklistEscaped;
221
222 // We may enter a block and visit it thinking that no alloca has escaped yet,
223 // then see an escape point and go back around a loop edge and come back to
224 // the same block twice. Because of this, we defer setting tail on calls when
225 // we first encounter them in a block. Every entry in this list does not
226 // statically use an alloca via use-def chain analysis, but may find an alloca
227 // through other means if the block turns out to be reachable after an escape
228 // point.
229 SmallVector<CallInst *, 32> DeferredTails;
230
231 BasicBlock *BB = &F.getEntryBlock();
232 VisitType Escaped = UNESCAPED;
233 do {
234 for (auto &I : *BB) {
235 if (Tracker.EscapePoints.count(&I))
236 Escaped = ESCAPED;
237
238 CallInst *CI = dyn_cast<CallInst>(&I);
239 // A PseudoProbeInst has the IntrInaccessibleMemOnly tag hence it is
240 // considered accessing memory and will be marked as a tail call if we
241 // don't bail out here.
242 if (!CI || CI->isTailCall() || isa<DbgInfoIntrinsic>(&I) ||
243 isa<PseudoProbeInst>(&I))
244 continue;
245
246 // Special-case operand bundles "clang.arc.attachedcall", "ptrauth", and
247 // "kcfi".
248 bool IsNoTail = CI->isNoTailCall() ||
252
253 if (!IsNoTail && CI->doesNotAccessMemory()) {
254 // A call to a readnone function whose arguments are all things computed
255 // outside this function can be marked tail. Even if you stored the
256 // alloca address into a global, a readnone function can't load the
257 // global anyhow.
258 //
259 // Note that this runs whether we know an alloca has escaped or not. If
260 // it has, then we can't trust Tracker.AllocaUsers to be accurate.
261 bool SafeToTail = true;
262 for (auto &Arg : CI->args()) {
263 if (isa<Constant>(Arg.getUser()))
264 continue;
265 if (Argument *A = dyn_cast<Argument>(Arg.getUser()))
266 if (!A->hasByValAttr())
267 continue;
268 SafeToTail = false;
269 break;
270 }
271 if (SafeToTail) {
272 using namespace ore;
273 ORE->emit([&]() {
274 return OptimizationRemark(DEBUG_TYPE, "tailcall-readnone", CI)
275 << "marked as tail call candidate (readnone)";
276 });
277 CI->setTailCall();
278 Modified = true;
279 continue;
280 }
281 }
282
283 if (!IsNoTail && Escaped == UNESCAPED && !Tracker.AllocaUsers.count(CI))
284 DeferredTails.push_back(CI);
285 }
286
287 for (auto *SuccBB : successors(BB)) {
288 auto &State = Visited[SuccBB];
289 if (State < Escaped) {
290 State = Escaped;
291 if (State == ESCAPED)
292 WorklistEscaped.push_back(SuccBB);
293 else
294 WorklistUnescaped.push_back(SuccBB);
295 }
296 }
297
298 if (!WorklistEscaped.empty()) {
299 BB = WorklistEscaped.pop_back_val();
300 Escaped = ESCAPED;
301 } else {
302 BB = nullptr;
303 while (!WorklistUnescaped.empty()) {
304 auto *NextBB = WorklistUnescaped.pop_back_val();
305 if (Visited[NextBB] == UNESCAPED) {
306 BB = NextBB;
307 Escaped = UNESCAPED;
308 break;
309 }
310 }
311 }
312 } while (BB);
313
314 for (CallInst *CI : DeferredTails) {
315 if (Visited[CI->getParent()] != ESCAPED) {
316 // If the escape point was part way through the block, calls after the
317 // escape point wouldn't have been put into DeferredTails.
318 LLVM_DEBUG(dbgs() << "Marked as tail call candidate: " << *CI << "\n");
319 CI->setTailCall();
320 Modified = true;
321 }
322 }
323
324 return Modified;
325}
326
327/// Return true if it is safe to move the specified
328/// instruction from after the call to before the call, assuming that all
329/// instructions between the call and this instruction are movable.
330///
332 if (isa<DbgInfoIntrinsic>(I))
333 return true;
334
335 if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(I))
336 if (II->getIntrinsicID() == Intrinsic::lifetime_end &&
337 llvm::findAllocaForValue(II->getArgOperand(1)))
338 return true;
339
340 // FIXME: We can move load/store/call/free instructions above the call if the
341 // call does not mod/ref the memory location being processed.
342 if (I->mayHaveSideEffects()) // This also handles volatile loads.
343 return false;
344
345 if (LoadInst *L = dyn_cast<LoadInst>(I)) {
346 // Loads may always be moved above calls without side effects.
347 if (CI->mayHaveSideEffects()) {
348 // Non-volatile loads may be moved above a call with side effects if it
349 // does not write to memory and the load provably won't trap.
350 // Writes to memory only matter if they may alias the pointer
351 // being loaded from.
352 const DataLayout &DL = L->getModule()->getDataLayout();
353 if (isModSet(AA->getModRefInfo(CI, MemoryLocation::get(L))) ||
354 !isSafeToLoadUnconditionally(L->getPointerOperand(), L->getType(),
355 L->getAlign(), DL, L))
356 return false;
357 }
358 }
359
360 // Otherwise, if this is a side-effect free instruction, check to make sure
361 // that it does not use the return value of the call. If it doesn't use the
362 // return value of the call, it must only use things that are defined before
363 // the call, or movable instructions between the call and the instruction
364 // itself.
365 return !is_contained(I->operands(), CI);
366}
367
369 if (!I->isAssociative() || !I->isCommutative())
370 return false;
371
372 assert(I->getNumOperands() >= 2 &&
373 "Associative/commutative operations should have at least 2 args!");
374
375 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
376 // Accumulators must have an identity.
377 if (!ConstantExpr::getIntrinsicIdentity(II->getIntrinsicID(), I->getType()))
378 return false;
379 }
380
381 // Exactly one operand should be the result of the call instruction.
382 if ((I->getOperand(0) == CI && I->getOperand(1) == CI) ||
383 (I->getOperand(0) != CI && I->getOperand(1) != CI))
384 return false;
385
386 // The only user of this instruction we allow is a single return instruction.
387 if (!I->hasOneUse() || !isa<ReturnInst>(I->user_back()))
388 return false;
389
390 return true;
391}
392
394 while (isa<DbgInfoIntrinsic>(I))
395 ++I;
396 return &*I;
397}
398
399namespace {
400class TailRecursionEliminator {
401 Function &F;
403 AliasAnalysis *AA;
405 DomTreeUpdater &DTU;
406
407 // The below are shared state we want to have available when eliminating any
408 // calls in the function. There values should be populated by
409 // createTailRecurseLoopHeader the first time we find a call we can eliminate.
410 BasicBlock *HeaderBB = nullptr;
411 SmallVector<PHINode *, 8> ArgumentPHIs;
412
413 // PHI node to store our return value.
414 PHINode *RetPN = nullptr;
415
416 // i1 PHI node to track if we have a valid return value stored in RetPN.
417 PHINode *RetKnownPN = nullptr;
418
419 // Vector of select instructions we insereted. These selects use RetKnownPN
420 // to either propagate RetPN or select a new return value.
422
423 // The below are shared state needed when performing accumulator recursion.
424 // There values should be populated by insertAccumulator the first time we
425 // find an elimination that requires an accumulator.
426
427 // PHI node to store our current accumulated value.
428 PHINode *AccPN = nullptr;
429
430 // The instruction doing the accumulating.
431 Instruction *AccumulatorRecursionInstr = nullptr;
432
433 TailRecursionEliminator(Function &F, const TargetTransformInfo *TTI,
435 DomTreeUpdater &DTU)
436 : F(F), TTI(TTI), AA(AA), ORE(ORE), DTU(DTU) {}
437
438 CallInst *findTRECandidate(BasicBlock *BB);
439
440 void createTailRecurseLoopHeader(CallInst *CI);
441
442 void insertAccumulator(Instruction *AccRecInstr);
443
444 bool eliminateCall(CallInst *CI);
445
446 void cleanupAndFinalize();
447
448 bool processBlock(BasicBlock &BB);
449
450 void copyByValueOperandIntoLocalTemp(CallInst *CI, int OpndIdx);
451
452 void copyLocalTempOfByValueOperandIntoArguments(CallInst *CI, int OpndIdx);
453
454public:
455 static bool eliminate(Function &F, const TargetTransformInfo *TTI,
457 DomTreeUpdater &DTU);
458};
459} // namespace
460
461CallInst *TailRecursionEliminator::findTRECandidate(BasicBlock *BB) {
462 Instruction *TI = BB->getTerminator();
463
464 if (&BB->front() == TI) // Make sure there is something before the terminator.
465 return nullptr;
466
467 // Scan backwards from the return, checking to see if there is a tail call in
468 // this block. If so, set CI to it.
469 CallInst *CI = nullptr;
470 BasicBlock::iterator BBI(TI);
471 while (true) {
472 CI = dyn_cast<CallInst>(BBI);
473 if (CI && CI->getCalledFunction() == &F)
474 break;
475
476 if (BBI == BB->begin())
477 return nullptr; // Didn't find a potential tail call.
478 --BBI;
479 }
480
481 assert((!CI->isTailCall() || !CI->isNoTailCall()) &&
482 "Incompatible call site attributes(Tail,NoTail)");
483 if (!CI->isTailCall())
484 return nullptr;
485
486 // As a special case, detect code like this:
487 // double fabs(double f) { return __builtin_fabs(f); } // a 'fabs' call
488 // and disable this xform in this case, because the code generator will
489 // lower the call to fabs into inline code.
490 if (BB == &F.getEntryBlock() &&
491 firstNonDbg(BB->front().getIterator()) == CI &&
492 firstNonDbg(std::next(BB->begin())) == TI && CI->getCalledFunction() &&
494 // A single-block function with just a call and a return. Check that
495 // the arguments match.
496 auto I = CI->arg_begin(), E = CI->arg_end();
497 Function::arg_iterator FI = F.arg_begin(), FE = F.arg_end();
498 for (; I != E && FI != FE; ++I, ++FI)
499 if (*I != &*FI) break;
500 if (I == E && FI == FE)
501 return nullptr;
502 }
503
504 return CI;
505}
506
507void TailRecursionEliminator::createTailRecurseLoopHeader(CallInst *CI) {
508 HeaderBB = &F.getEntryBlock();
509 BasicBlock *NewEntry = BasicBlock::Create(F.getContext(), "", &F, HeaderBB);
510 NewEntry->takeName(HeaderBB);
511 HeaderBB->setName("tailrecurse");
512 BranchInst *BI = BranchInst::Create(HeaderBB, NewEntry);
513 BI->setDebugLoc(CI->getDebugLoc());
514
515 // Move all fixed sized allocas from HeaderBB to NewEntry.
516 for (BasicBlock::iterator OEBI = HeaderBB->begin(), E = HeaderBB->end(),
517 NEBI = NewEntry->begin();
518 OEBI != E;)
519 if (AllocaInst *AI = dyn_cast<AllocaInst>(OEBI++))
520 if (isa<ConstantInt>(AI->getArraySize()))
521 AI->moveBefore(&*NEBI);
522
523 // Now that we have created a new block, which jumps to the entry
524 // block, insert a PHI node for each argument of the function.
525 // For now, we initialize each PHI to only have the real arguments
526 // which are passed in.
527 BasicBlock::iterator InsertPos = HeaderBB->begin();
528 for (Function::arg_iterator I = F.arg_begin(), E = F.arg_end(); I != E; ++I) {
529 PHINode *PN = PHINode::Create(I->getType(), 2, I->getName() + ".tr");
530 PN->insertBefore(InsertPos);
531 I->replaceAllUsesWith(PN); // Everyone use the PHI node now!
532 PN->addIncoming(&*I, NewEntry);
533 ArgumentPHIs.push_back(PN);
534 }
535
536 // If the function doen't return void, create the RetPN and RetKnownPN PHI
537 // nodes to track our return value. We initialize RetPN with poison and
538 // RetKnownPN with false since we can't know our return value at function
539 // entry.
540 Type *RetType = F.getReturnType();
541 if (!RetType->isVoidTy()) {
542 Type *BoolType = Type::getInt1Ty(F.getContext());
543 RetPN = PHINode::Create(RetType, 2, "ret.tr");
544 RetPN->insertBefore(InsertPos);
545 RetKnownPN = PHINode::Create(BoolType, 2, "ret.known.tr");
546 RetKnownPN->insertBefore(InsertPos);
547
548 RetPN->addIncoming(PoisonValue::get(RetType), NewEntry);
549 RetKnownPN->addIncoming(ConstantInt::getFalse(BoolType), NewEntry);
550 }
551
552 // The entry block was changed from HeaderBB to NewEntry.
553 // The forward DominatorTree needs to be recalculated when the EntryBB is
554 // changed. In this corner-case we recalculate the entire tree.
555 DTU.recalculate(*NewEntry->getParent());
556}
557
558void TailRecursionEliminator::insertAccumulator(Instruction *AccRecInstr) {
559 assert(!AccPN && "Trying to insert multiple accumulators");
560
561 AccumulatorRecursionInstr = AccRecInstr;
562
563 // Start by inserting a new PHI node for the accumulator.
564 pred_iterator PB = pred_begin(HeaderBB), PE = pred_end(HeaderBB);
565 AccPN = PHINode::Create(F.getReturnType(), std::distance(PB, PE) + 1,
566 "accumulator.tr");
567 AccPN->insertBefore(HeaderBB->begin());
568
569 // Loop over all of the predecessors of the tail recursion block. For the
570 // real entry into the function we seed the PHI with the identity constant for
571 // the accumulation operation. For any other existing branches to this block
572 // (due to other tail recursions eliminated) the accumulator is not modified.
573 // Because we haven't added the branch in the current block to HeaderBB yet,
574 // it will not show up as a predecessor.
575 for (pred_iterator PI = PB; PI != PE; ++PI) {
576 BasicBlock *P = *PI;
577 if (P == &F.getEntryBlock()) {
578 Constant *Identity =
579 ConstantExpr::getIdentity(AccRecInstr, AccRecInstr->getType());
580 AccPN->addIncoming(Identity, P);
581 } else {
582 AccPN->addIncoming(AccPN, P);
583 }
584 }
585
586 ++NumAccumAdded;
587}
588
589// Creates a copy of contents of ByValue operand of the specified
590// call instruction into the newly created temporarily variable.
591void TailRecursionEliminator::copyByValueOperandIntoLocalTemp(CallInst *CI,
592 int OpndIdx) {
593 Type *AggTy = CI->getParamByValType(OpndIdx);
594 assert(AggTy);
595 const DataLayout &DL = F.getParent()->getDataLayout();
596
597 // Get alignment of byVal operand.
598 Align Alignment(CI->getParamAlign(OpndIdx).valueOrOne());
599
600 // Create alloca for temporarily byval operands.
601 // Put alloca into the entry block.
602 Value *NewAlloca = new AllocaInst(
603 AggTy, DL.getAllocaAddrSpace(), nullptr, Alignment,
604 CI->getArgOperand(OpndIdx)->getName(), F.getEntryBlock().begin());
605
606 IRBuilder<> Builder(CI);
607 Value *Size = Builder.getInt64(DL.getTypeAllocSize(AggTy));
608
609 // Copy data from byvalue operand into the temporarily variable.
610 Builder.CreateMemCpy(NewAlloca, /*DstAlign*/ Alignment,
611 CI->getArgOperand(OpndIdx),
612 /*SrcAlign*/ Alignment, Size);
613 CI->setArgOperand(OpndIdx, NewAlloca);
614}
615
616// Creates a copy from temporarily variable(keeping value of ByVal argument)
617// into the corresponding function argument location.
618void TailRecursionEliminator::copyLocalTempOfByValueOperandIntoArguments(
619 CallInst *CI, int OpndIdx) {
620 Type *AggTy = CI->getParamByValType(OpndIdx);
621 assert(AggTy);
622 const DataLayout &DL = F.getParent()->getDataLayout();
623
624 // Get alignment of byVal operand.
625 Align Alignment(CI->getParamAlign(OpndIdx).valueOrOne());
626
627 IRBuilder<> Builder(CI);
628 Value *Size = Builder.getInt64(DL.getTypeAllocSize(AggTy));
629
630 // Copy data from the temporarily variable into corresponding
631 // function argument location.
632 Builder.CreateMemCpy(F.getArg(OpndIdx), /*DstAlign*/ Alignment,
633 CI->getArgOperand(OpndIdx),
634 /*SrcAlign*/ Alignment, Size);
635}
636
637bool TailRecursionEliminator::eliminateCall(CallInst *CI) {
638 ReturnInst *Ret = cast<ReturnInst>(CI->getParent()->getTerminator());
639
640 // Ok, we found a potential tail call. We can currently only transform the
641 // tail call if all of the instructions between the call and the return are
642 // movable to above the call itself, leaving the call next to the return.
643 // Check that this is the case now.
644 Instruction *AccRecInstr = nullptr;
645 BasicBlock::iterator BBI(CI);
646 for (++BBI; &*BBI != Ret; ++BBI) {
647 if (canMoveAboveCall(&*BBI, CI, AA))
648 continue;
649
650 // If we can't move the instruction above the call, it might be because it
651 // is an associative and commutative operation that could be transformed
652 // using accumulator recursion elimination. Check to see if this is the
653 // case, and if so, remember which instruction accumulates for later.
654 if (AccPN || !canTransformAccumulatorRecursion(&*BBI, CI))
655 return false; // We cannot eliminate the tail recursion!
656
657 // Yes, this is accumulator recursion. Remember which instruction
658 // accumulates.
659 AccRecInstr = &*BBI;
660 }
661
662 BasicBlock *BB = Ret->getParent();
663
664 using namespace ore;
665 ORE->emit([&]() {
666 return OptimizationRemark(DEBUG_TYPE, "tailcall-recursion", CI)
667 << "transforming tail recursion into loop";
668 });
669
670 // OK! We can transform this tail call. If this is the first one found,
671 // create the new entry block, allowing us to branch back to the old entry.
672 if (!HeaderBB)
673 createTailRecurseLoopHeader(CI);
674
675 // Copy values of ByVal operands into local temporarily variables.
676 for (unsigned I = 0, E = CI->arg_size(); I != E; ++I) {
677 if (CI->isByValArgument(I))
678 copyByValueOperandIntoLocalTemp(CI, I);
679 }
680
681 // Ok, now that we know we have a pseudo-entry block WITH all of the
682 // required PHI nodes, add entries into the PHI node for the actual
683 // parameters passed into the tail-recursive call.
684 for (unsigned I = 0, E = CI->arg_size(); I != E; ++I) {
685 if (CI->isByValArgument(I)) {
686 copyLocalTempOfByValueOperandIntoArguments(CI, I);
687 // When eliminating a tail call, we modify the values of the arguments.
688 // Therefore, if the byval parameter has a readonly attribute, we have to
689 // remove it. It is safe because, from the perspective of a caller, the
690 // byval parameter is always treated as "readonly," even if the readonly
691 // attribute is removed.
692 F.removeParamAttr(I, Attribute::ReadOnly);
693 ArgumentPHIs[I]->addIncoming(F.getArg(I), BB);
694 } else
695 ArgumentPHIs[I]->addIncoming(CI->getArgOperand(I), BB);
696 }
697
698 if (AccRecInstr) {
699 insertAccumulator(AccRecInstr);
700
701 // Rewrite the accumulator recursion instruction so that it does not use
702 // the result of the call anymore, instead, use the PHI node we just
703 // inserted.
704 AccRecInstr->setOperand(AccRecInstr->getOperand(0) != CI, AccPN);
705 }
706
707 // Update our return value tracking
708 if (RetPN) {
709 if (Ret->getReturnValue() == CI || AccRecInstr) {
710 // Defer selecting a return value
711 RetPN->addIncoming(RetPN, BB);
712 RetKnownPN->addIncoming(RetKnownPN, BB);
713 } else {
714 // We found a return value we want to use, insert a select instruction to
715 // select it if we don't already know what our return value will be and
716 // store the result in our return value PHI node.
717 SelectInst *SI =
718 SelectInst::Create(RetKnownPN, RetPN, Ret->getReturnValue(),
719 "current.ret.tr", Ret->getIterator());
720 RetSelects.push_back(SI);
721
722 RetPN->addIncoming(SI, BB);
723 RetKnownPN->addIncoming(ConstantInt::getTrue(RetKnownPN->getType()), BB);
724 }
725
726 if (AccPN)
727 AccPN->addIncoming(AccRecInstr ? AccRecInstr : AccPN, BB);
728 }
729
730 // Now that all of the PHI nodes are in place, remove the call and
731 // ret instructions, replacing them with an unconditional branch.
732 BranchInst *NewBI = BranchInst::Create(HeaderBB, Ret->getIterator());
733 NewBI->setDebugLoc(CI->getDebugLoc());
734
735 Ret->eraseFromParent(); // Remove return.
736 CI->eraseFromParent(); // Remove call.
737 DTU.applyUpdates({{DominatorTree::Insert, BB, HeaderBB}});
738 ++NumEliminated;
739 return true;
740}
741
742void TailRecursionEliminator::cleanupAndFinalize() {
743 // If we eliminated any tail recursions, it's possible that we inserted some
744 // silly PHI nodes which just merge an initial value (the incoming operand)
745 // with themselves. Check to see if we did and clean up our mess if so. This
746 // occurs when a function passes an argument straight through to its tail
747 // call.
748 for (PHINode *PN : ArgumentPHIs) {
749 // If the PHI Node is a dynamic constant, replace it with the value it is.
750 if (Value *PNV = simplifyInstruction(PN, F.getParent()->getDataLayout())) {
751 PN->replaceAllUsesWith(PNV);
752 PN->eraseFromParent();
753 }
754 }
755
756 if (RetPN) {
757 if (RetSelects.empty()) {
758 // If we didn't insert any select instructions, then we know we didn't
759 // store a return value and we can remove the PHI nodes we inserted.
760 RetPN->dropAllReferences();
761 RetPN->eraseFromParent();
762
763 RetKnownPN->dropAllReferences();
764 RetKnownPN->eraseFromParent();
765
766 if (AccPN) {
767 // We need to insert a copy of our accumulator instruction before any
768 // return in the function, and return its result instead.
769 Instruction *AccRecInstr = AccumulatorRecursionInstr;
770 for (BasicBlock &BB : F) {
771 ReturnInst *RI = dyn_cast<ReturnInst>(BB.getTerminator());
772 if (!RI)
773 continue;
774
775 Instruction *AccRecInstrNew = AccRecInstr->clone();
776 AccRecInstrNew->setName("accumulator.ret.tr");
777 AccRecInstrNew->setOperand(AccRecInstr->getOperand(0) == AccPN,
778 RI->getOperand(0));
779 AccRecInstrNew->insertBefore(RI);
780 RI->setOperand(0, AccRecInstrNew);
781 }
782 }
783 } else {
784 // We need to insert a select instruction before any return left in the
785 // function to select our stored return value if we have one.
786 for (BasicBlock &BB : F) {
787 ReturnInst *RI = dyn_cast<ReturnInst>(BB.getTerminator());
788 if (!RI)
789 continue;
790
791 SelectInst *SI =
792 SelectInst::Create(RetKnownPN, RetPN, RI->getOperand(0),
793 "current.ret.tr", RI->getIterator());
794 RetSelects.push_back(SI);
795 RI->setOperand(0, SI);
796 }
797
798 if (AccPN) {
799 // We need to insert a copy of our accumulator instruction before any
800 // of the selects we inserted, and select its result instead.
801 Instruction *AccRecInstr = AccumulatorRecursionInstr;
802 for (SelectInst *SI : RetSelects) {
803 Instruction *AccRecInstrNew = AccRecInstr->clone();
804 AccRecInstrNew->setName("accumulator.ret.tr");
805 AccRecInstrNew->setOperand(AccRecInstr->getOperand(0) == AccPN,
806 SI->getFalseValue());
807 AccRecInstrNew->insertBefore(SI);
808 SI->setFalseValue(AccRecInstrNew);
809 }
810 }
811 }
812 }
813}
814
815bool TailRecursionEliminator::processBlock(BasicBlock &BB) {
816 Instruction *TI = BB.getTerminator();
817
818 if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
819 if (BI->isConditional())
820 return false;
821
822 BasicBlock *Succ = BI->getSuccessor(0);
823 ReturnInst *Ret = dyn_cast<ReturnInst>(Succ->getFirstNonPHIOrDbg(true));
824
825 if (!Ret)
826 return false;
827
828 CallInst *CI = findTRECandidate(&BB);
829
830 if (!CI)
831 return false;
832
833 LLVM_DEBUG(dbgs() << "FOLDING: " << *Succ
834 << "INTO UNCOND BRANCH PRED: " << BB);
835 FoldReturnIntoUncondBranch(Ret, Succ, &BB, &DTU);
836 ++NumRetDuped;
837
838 // If all predecessors of Succ have been eliminated by
839 // FoldReturnIntoUncondBranch, delete it. It is important to empty it,
840 // because the ret instruction in there is still using a value which
841 // eliminateCall will attempt to remove. This block can only contain
842 // instructions that can't have uses, therefore it is safe to remove.
843 if (pred_empty(Succ))
844 DTU.deleteBB(Succ);
845
846 eliminateCall(CI);
847 return true;
848 } else if (isa<ReturnInst>(TI)) {
849 CallInst *CI = findTRECandidate(&BB);
850
851 if (CI)
852 return eliminateCall(CI);
853 }
854
855 return false;
856}
857
858bool TailRecursionEliminator::eliminate(Function &F,
860 AliasAnalysis *AA,
862 DomTreeUpdater &DTU) {
863 if (F.getFnAttribute("disable-tail-calls").getValueAsBool())
864 return false;
865
866 bool MadeChange = false;
867 MadeChange |= markTails(F, ORE);
868
869 // If this function is a varargs function, we won't be able to PHI the args
870 // right, so don't even try to convert it...
871 if (F.getFunctionType()->isVarArg())
872 return MadeChange;
873
874 if (!canTRE(F))
875 return MadeChange;
876
877 // Change any tail recursive calls to loops.
878 TailRecursionEliminator TRE(F, TTI, AA, ORE, DTU);
879
880 for (BasicBlock &BB : F)
881 MadeChange |= TRE.processBlock(BB);
882
883 TRE.cleanupAndFinalize();
884
885 return MadeChange;
886}
887
888namespace {
889struct TailCallElim : public FunctionPass {
890 static char ID; // Pass identification, replacement for typeid
891 TailCallElim() : FunctionPass(ID) {
893 }
894
895 void getAnalysisUsage(AnalysisUsage &AU) const override {
902 }
903
904 bool runOnFunction(Function &F) override {
905 if (skipFunction(F))
906 return false;
907
908 auto *DTWP = getAnalysisIfAvailable<DominatorTreeWrapperPass>();
909 auto *DT = DTWP ? &DTWP->getDomTree() : nullptr;
910 auto *PDTWP = getAnalysisIfAvailable<PostDominatorTreeWrapperPass>();
911 auto *PDT = PDTWP ? &PDTWP->getPostDomTree() : nullptr;
912 // There is no noticable performance difference here between Lazy and Eager
913 // UpdateStrategy based on some test results. It is feasible to switch the
914 // UpdateStrategy to Lazy if we find it profitable later.
915 DomTreeUpdater DTU(DT, PDT, DomTreeUpdater::UpdateStrategy::Eager);
916
917 return TailRecursionEliminator::eliminate(
918 F, &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F),
919 &getAnalysis<AAResultsWrapperPass>().getAAResults(),
920 &getAnalysis<OptimizationRemarkEmitterWrapperPass>().getORE(), DTU);
921 }
922};
923}
924
925char TailCallElim::ID = 0;
926INITIALIZE_PASS_BEGIN(TailCallElim, "tailcallelim", "Tail Call Elimination",
927 false, false)
932
933// Public interface to the TailCallElimination pass
935 return new TailCallElim();
936}
937
940
946 // There is no noticable performance difference here between Lazy and Eager
947 // UpdateStrategy based on some test results. It is feasible to switch the
948 // UpdateStrategy to Lazy if we find it profitable later.
950 bool Changed = TailRecursionEliminator::eliminate(F, &TTI, &AA, &ORE, DTU);
951
952 if (!Changed)
953 return PreservedAnalyses::all();
957 return PA;
958}
MachineBasicBlock MachineBasicBlock::iterator DebugLoc DL
Expand Atomic instructions
static GCRegistry::Add< ErlangGC > A("erlang", "erlang-compatible garbage collector")
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
This file contains the declarations for the subclasses of Constant, which represent the different fla...
#define LLVM_DEBUG(X)
Definition: Debug.h:101
uint64_t Size
This is the interface for a simple mod/ref and alias analysis over globals.
This file provides various utilities for inspecting and working with the control flow graph in LLVM I...
#define F(x, y, z)
Definition: MD5.cpp:55
#define I(x, y, z)
Definition: MD5.cpp:58
Module.h This file contains the declarations for the Module class.
#define P(N)
PassBuilder PB(Machine, PassOpts->PTO, std::nullopt, &PIC)
#define INITIALIZE_PASS_DEPENDENCY(depName)
Definition: PassSupport.h:55
#define INITIALIZE_PASS_END(passName, arg, name, cfg, analysis)
Definition: PassSupport.h:59
#define INITIALIZE_PASS_BEGIN(passName, arg, name, cfg, analysis)
Definition: PassSupport.h:52
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
This file contains some templates that are useful if you are working with the STL at all.
This file defines the SmallPtrSet class.
This file defines the 'Statistic' class, which is designed to be an easy way to expose various metric...
#define STATISTIC(VARNAME, DESC)
Definition: Statistic.h:167
static bool canTRE(Function &F)
Scan the specified function for alloca instructions.
static bool canMoveAboveCall(Instruction *I, CallInst *CI, AliasAnalysis *AA)
Return true if it is safe to move the specified instruction from after the call to before the call,...
Tail Call Elimination
static Instruction * firstNonDbg(BasicBlock::iterator I)
#define DEBUG_TYPE
static bool canTransformAccumulatorRecursion(Instruction *I, CallInst *CI)
static bool markTails(Function &F, OptimizationRemarkEmitter *ORE)
This pass exposes codegen information to IR-level passes.
A manager for alias analyses.
A wrapper pass to provide the legacy pass manager access to a suitably prepared AAResults object.
ModRefInfo getModRefInfo(const Instruction *I, const std::optional< MemoryLocation > &OptLoc)
Check whether or not an instruction may read or write the optionally specified memory location.
an instruction to allocate memory on the stack
Definition: Instructions.h:59
A container for analyses that lazily runs them and caches their results.
Definition: PassManager.h:348
PassT::Result * getCachedResult(IRUnitT &IR) const
Get the cached result of an analysis pass for a given IR unit.
Definition: PassManager.h:519
PassT::Result & getResult(IRUnitT &IR, ExtraArgTs... ExtraArgs)
Get the result of an analysis pass for a given IR unit.
Definition: PassManager.h:500
Represent the analysis usage information of a pass.
AnalysisUsage & addRequired()
AnalysisUsage & addPreserved()
Add the specified Pass class to the set of analyses preserved by this pass.
This class represents an incoming formal argument to a Function.
Definition: Argument.h:28
LLVM Basic Block Representation.
Definition: BasicBlock.h:60
iterator begin()
Instruction iterator methods.
Definition: BasicBlock.h:429
const Instruction & front() const
Definition: BasicBlock.h:452
static BasicBlock * Create(LLVMContext &Context, const Twine &Name="", Function *Parent=nullptr, BasicBlock *InsertBefore=nullptr)
Creates a new BasicBlock.
Definition: BasicBlock.h:198
const Function * getParent() const
Return the enclosing method, or null if none.
Definition: BasicBlock.h:205
const Instruction * getFirstNonPHIOrDbg(bool SkipPseudoOp=true) const
Returns a pointer to the first instruction in this block that is not a PHINode or a debug intrinsic,...
Definition: BasicBlock.cpp:366
InstListType::iterator iterator
Instruction iterators...
Definition: BasicBlock.h:164
const Instruction * getTerminator() const LLVM_READONLY
Returns the terminator instruction if the block is well formed or null if the block is not well forme...
Definition: BasicBlock.h:220
Conditional or Unconditional Branch instruction.
static BranchInst * Create(BasicBlock *IfTrue, BasicBlock::iterator InsertBefore)
bool isConditional() const
BasicBlock * getSuccessor(unsigned i) const
Base class for all callable instructions (InvokeInst and CallInst) Holds everything related to callin...
Definition: InstrTypes.h:1455
Function * getCalledFunction() const
Returns the function called, or null if this is an indirect function invocation or the function signa...
Definition: InstrTypes.h:1703
bool doesNotAccessMemory(unsigned OpNo) const
Definition: InstrTypes.h:2037
User::op_iterator arg_begin()
Return the iterator pointing to the beginning of the argument list.
Definition: InstrTypes.h:1623
bool isByValArgument(unsigned ArgNo) const
Determine whether this argument is passed by value.
Definition: InstrTypes.h:2001
MaybeAlign getParamAlign(unsigned ArgNo) const
Extract the alignment for a call or parameter (0=unknown).
Definition: InstrTypes.h:2065
bool onlyReadsMemory(unsigned OpNo) const
Definition: InstrTypes.h:2043
Type * getParamByValType(unsigned ArgNo) const
Extract the byval type for a call or parameter.
Definition: InstrTypes.h:2074
bool hasOperandBundlesOtherThan(ArrayRef< uint32_t > IDs) const
Return true if this operand bundle user contains operand bundles with tags other than those specified...
Definition: InstrTypes.h:2431
Value * getArgOperand(unsigned i) const
Definition: InstrTypes.h:1648
void setArgOperand(unsigned i, Value *v)
Definition: InstrTypes.h:1653
User::op_iterator arg_end()
Return the iterator pointing to the end of the argument list.
Definition: InstrTypes.h:1629
iterator_range< User::op_iterator > args()
Iteration adapter for range-for loops.
Definition: InstrTypes.h:1639
unsigned arg_size() const
Definition: InstrTypes.h:1646
This class represents a function call, abstracting a target machine's calling convention.
bool isNoTailCall() const
bool isTailCall() const
void setTailCall(bool IsTc=true)
static Constant * getIdentity(Instruction *I, Type *Ty, bool AllowRHSConstant=false, bool NSZ=false)
Return the identity constant for a binary or intrinsic Instruction.
Definition: Constants.cpp:2660
static Constant * getIntrinsicIdentity(Intrinsic::ID, Type *Ty)
Definition: Constants.cpp:2643
static ConstantInt * getTrue(LLVMContext &Context)
Definition: Constants.cpp:849
static ConstantInt * getFalse(LLVMContext &Context)
Definition: Constants.cpp:856
This is an important base class in LLVM.
Definition: Constant.h:41
A parsed version of the target data layout string in and methods for querying it.
Definition: DataLayout.h:110
Analysis pass which computes a DominatorTree.
Definition: Dominators.h:279
Legacy analysis pass which computes a DominatorTree.
Definition: Dominators.h:317
FunctionPass class - This class is used to implement most global optimizations.
Definition: Pass.h:311
virtual bool runOnFunction(Function &F)=0
runOnFunction - Virtual method overriden by subclasses to do the per-function processing of the pass.
bool skipFunction(const Function &F) const
Optional passes call this function to check whether the pass should be skipped.
Definition: Pass.cpp:178
Legacy wrapper pass to provide the GlobalsAAResult object.
This provides a uniform API for creating instructions and inserting them into a basic block: either a...
Definition: IRBuilder.h:2649
Instruction * clone() const
Create a copy of 'this' instruction that is identical in all ways except the following:
void insertBefore(Instruction *InsertPos)
Insert an unlinked instruction into a basic block immediately before the specified instruction.
const DebugLoc & getDebugLoc() const
Return the debug location for this node as a DebugLoc.
Definition: Instruction.h:453
const BasicBlock * getParent() const
Definition: Instruction.h:151
InstListType::iterator eraseFromParent()
This method unlinks 'this' from the containing basic block and deletes it.
bool mayHaveSideEffects() const LLVM_READONLY
Return true if the instruction may have side effects.
void setDebugLoc(DebugLoc Loc)
Set the debug location information for this instruction.
Definition: Instruction.h:450
A wrapper class for inspecting calls to intrinsic functions.
Definition: IntrinsicInst.h:47
An instruction for reading from memory.
Definition: Instructions.h:184
static MemoryLocation get(const LoadInst *LI)
Return a location with information about the memory reference by the given instruction.
OptimizationRemarkEmitter legacy analysis pass.
The optimization diagnostic interface.
void emit(DiagnosticInfoOptimizationBase &OptDiag)
Output the remark via the diagnostic handler and to the optimization record file.
Diagnostic information for applied optimization remarks.
void addIncoming(Value *V, BasicBlock *BB)
Add an incoming value to the end of the PHI list.
static PHINode * Create(Type *Ty, unsigned NumReservedValues, const Twine &NameStr, BasicBlock::iterator InsertBefore)
Constructors - NumReservedValues is a hint for the number of incoming edges that this phi node will h...
static PassRegistry * getPassRegistry()
getPassRegistry - Access the global registry object, which is automatically initialized at applicatio...
virtual void getAnalysisUsage(AnalysisUsage &) const
getAnalysisUsage - This function should be overriden by passes that need analysis information to do t...
Definition: Pass.cpp:98
static PoisonValue * get(Type *T)
Static factory methods - Return an 'poison' object of the specified type.
Definition: Constants.cpp:1827
Analysis pass which computes a PostDominatorTree.
A set of analyses that are preserved following a run of a transformation pass.
Definition: Analysis.h:109
static PreservedAnalyses all()
Construct a special preserved set that preserves all passes.
Definition: Analysis.h:115
void preserve()
Mark an analysis as preserved.
Definition: Analysis.h:129
Return a value (possibly void), from a function.
This class represents the LLVM 'select' instruction.
static SelectInst * Create(Value *C, Value *S1, Value *S2, const Twine &NameStr, BasicBlock::iterator InsertBefore, Instruction *MDFrom=nullptr)
std::pair< iterator, bool > insert(PtrType Ptr)
Inserts Ptr if and only if there is no element in the container equal to Ptr.
Definition: SmallPtrSet.h:342
SmallPtrSet - This class implements a set which is optimized for holding SmallSize or less elements.
Definition: SmallPtrSet.h:427
bool empty() const
Definition: SmallVector.h:94
void push_back(const T &Elt)
Definition: SmallVector.h:426
This is a 'vector' (really, a variable-sized array), optimized for the case when the array is small.
Definition: SmallVector.h:1209
Analysis pass providing the TargetTransformInfo.
Wrapper pass for TargetTransformInfo.
This pass provides access to the codegen interfaces that are needed for IR-level transformations.
bool isLoweredToCall(const Function *F) const
Test whether calls to a function lower to actual program function calls.
The instances of the Type class are immutable: once they are created, they are never changed.
Definition: Type.h:45
static IntegerType * getInt1Ty(LLVMContext &C)
bool isVoidTy() const
Return true if this is 'void'.
Definition: Type.h:140
A Use represents the edge between a Value definition and its users.
Definition: Use.h:43
void setOperand(unsigned i, Value *Val)
Definition: User.h:174
Value * getOperand(unsigned i) const
Definition: User.h:169
LLVM Value Representation.
Definition: Value.h:74
Type * getType() const
All values are typed, get the type of this value.
Definition: Value.h:255
void setName(const Twine &Name)
Change the name of the value.
Definition: Value.cpp:377
void replaceAllUsesWith(Value *V)
Change all uses of this to point to a new Value.
Definition: Value.cpp:534
StringRef getName() const
Return a constant reference to the value's name.
Definition: Value.cpp:309
void takeName(Value *V)
Transfer the name from V to this value.
Definition: Value.cpp:383
self_iterator getIterator()
Definition: ilist_node.h:109
unsigned ID
LLVM IR allows to use arbitrary numbers as calling convention identifiers.
Definition: CallingConv.h:24
@ Tail
Attemps to make calls as fast as possible while guaranteeing that tail call optimization can always b...
Definition: CallingConv.h:76
This is an optimization pass for GlobalISel generic memory operations.
Definition: AddressRanges.h:18
bool all_of(R &&range, UnaryPredicate P)
Provide wrappers to std::all_of which take ranges instead of having to pass begin/end explicitly.
Definition: STLExtras.h:1731
FunctionPass * createTailCallEliminationPass()
AllocaInst * findAllocaForValue(Value *V, bool OffsetZero=false)
Returns unique alloca where the value comes from, or nullptr.
auto successors(const MachineBasicBlock *BB)
void initializeTailCallElimPass(PassRegistry &)
ReturnInst * FoldReturnIntoUncondBranch(ReturnInst *RI, BasicBlock *BB, BasicBlock *Pred, DomTreeUpdater *DTU=nullptr)
This method duplicates the specified return instruction into a predecessor which ends in an unconditi...
Value * simplifyInstruction(Instruction *I, const SimplifyQuery &Q)
See if we can compute a simplified version of this instruction.
Interval::pred_iterator pred_end(Interval *I)
Definition: Interval.h:112
bool isModSet(const ModRefInfo MRI)
Definition: ModRef.h:48
raw_ostream & dbgs()
dbgs() - This returns a reference to a raw_ostream for debugging messages.
Definition: Debug.cpp:163
Interval::pred_iterator pred_begin(Interval *I)
pred_begin/pred_end - define methods so that Intervals may be used just like BasicBlocks can with the...
Definition: Interval.h:109
bool is_contained(R &&Range, const E &Element)
Returns true if Element is found in Range.
Definition: STLExtras.h:1888
bool isSafeToLoadUnconditionally(Value *V, Align Alignment, APInt &Size, const DataLayout &DL, Instruction *ScanFrom=nullptr, AssumptionCache *AC=nullptr, const DominatorTree *DT=nullptr, const TargetLibraryInfo *TLI=nullptr)
Return true if we know that executing a load from this value cannot trap.
Definition: Loads.cpp:350
bool pred_empty(const BasicBlock *BB)
Definition: CFG.h:118
This struct is a compact representation of a valid (non-zero power of two) alignment.
Definition: Alignment.h:39
Align valueOrOne() const
For convenience, returns a valid alignment or 1 if undefined.
Definition: Alignment.h:141
PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM)