Bug Summary

File:lib/Transforms/Scalar/TailRecursionElimination.cpp
Location:line 597, column 14
Description:Called C++ object pointer is null

Annotated Source Code

1//===- TailRecursionElimination.cpp - Eliminate Tail Calls ----------------===//
2//
3// The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
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
13// algorithm:
14//
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
30// generator).
31//
32// There are several improvements that could be made:
33//
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.
50//
51//===----------------------------------------------------------------------===//
52
53#include "llvm/Transforms/Scalar.h"
54#include "llvm/ADT/STLExtras.h"
55#include "llvm/ADT/SmallPtrSet.h"
56#include "llvm/ADT/Statistic.h"
57#include "llvm/Analysis/GlobalsModRef.h"
58#include "llvm/Analysis/CFG.h"
59#include "llvm/Analysis/CaptureTracking.h"
60#include "llvm/Analysis/InlineCost.h"
61#include "llvm/Analysis/InstructionSimplify.h"
62#include "llvm/Analysis/Loads.h"
63#include "llvm/Analysis/TargetTransformInfo.h"
64#include "llvm/IR/CFG.h"
65#include "llvm/IR/CallSite.h"
66#include "llvm/IR/Constants.h"
67#include "llvm/IR/DataLayout.h"
68#include "llvm/IR/DerivedTypes.h"
69#include "llvm/IR/DiagnosticInfo.h"
70#include "llvm/IR/Function.h"
71#include "llvm/IR/Instructions.h"
72#include "llvm/IR/IntrinsicInst.h"
73#include "llvm/IR/Module.h"
74#include "llvm/IR/ValueHandle.h"
75#include "llvm/Pass.h"
76#include "llvm/Support/Debug.h"
77#include "llvm/Support/raw_ostream.h"
78#include "llvm/Transforms/Utils/BasicBlockUtils.h"
79#include "llvm/Transforms/Utils/Local.h"
80using namespace llvm;
81
82#define DEBUG_TYPE"tailcallelim" "tailcallelim"
83
84STATISTIC(NumEliminated, "Number of tail calls removed")static llvm::Statistic NumEliminated = { "tailcallelim", "Number of tail calls removed"
, 0, 0 };
85STATISTIC(NumRetDuped, "Number of return duplicated")static llvm::Statistic NumRetDuped = { "tailcallelim", "Number of return duplicated"
, 0, 0 };
86STATISTIC(NumAccumAdded, "Number of accumulators introduced")static llvm::Statistic NumAccumAdded = { "tailcallelim", "Number of accumulators introduced"
, 0, 0 };
87
88namespace {
89 struct TailCallElim : public FunctionPass {
90 const TargetTransformInfo *TTI;
91
92 static char ID; // Pass identification, replacement for typeid
93 TailCallElim() : FunctionPass(ID) {
94 initializeTailCallElimPass(*PassRegistry::getPassRegistry());
95 }
96
97 void getAnalysisUsage(AnalysisUsage &AU) const override;
98
99 bool runOnFunction(Function &F) override;
100
101 private:
102 bool runTRE(Function &F);
103 bool markTails(Function &F, bool &AllCallsAreTailCalls);
104
105 CallInst *FindTRECandidate(Instruction *I,
106 bool CannotTailCallElimCallsMarkedTail);
107 bool EliminateRecursiveTailCall(CallInst *CI, ReturnInst *Ret,
108 BasicBlock *&OldEntry,
109 bool &TailCallsAreMarkedTail,
110 SmallVectorImpl<PHINode *> &ArgumentPHIs,
111 bool CannotTailCallElimCallsMarkedTail);
112 bool FoldReturnAndProcessPred(BasicBlock *BB,
113 ReturnInst *Ret, BasicBlock *&OldEntry,
114 bool &TailCallsAreMarkedTail,
115 SmallVectorImpl<PHINode *> &ArgumentPHIs,
116 bool CannotTailCallElimCallsMarkedTail);
117 bool ProcessReturningBlock(ReturnInst *RI, BasicBlock *&OldEntry,
118 bool &TailCallsAreMarkedTail,
119 SmallVectorImpl<PHINode *> &ArgumentPHIs,
120 bool CannotTailCallElimCallsMarkedTail);
121 bool CanMoveAboveCall(Instruction *I, CallInst *CI);
122 Value *CanTransformAccumulatorRecursion(Instruction *I, CallInst *CI);
123 };
124}
125
126char TailCallElim::ID = 0;
127INITIALIZE_PASS_BEGIN(TailCallElim, "tailcallelim",static void* initializeTailCallElimPassOnce(PassRegistry &
Registry) {
128 "Tail Call Elimination", false, false)static void* initializeTailCallElimPassOnce(PassRegistry &
Registry) {
129INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)initializeTargetTransformInfoWrapperPassPass(Registry);
130INITIALIZE_PASS_END(TailCallElim, "tailcallelim",PassInfo *PI = new PassInfo("Tail Call Elimination", "tailcallelim"
, & TailCallElim ::ID, PassInfo::NormalCtor_t(callDefaultCtor
< TailCallElim >), false, false); Registry.registerPass
(*PI, true); return PI; } void llvm::initializeTailCallElimPass
(PassRegistry &Registry) { static volatile sys::cas_flag initialized
= 0; sys::cas_flag old_val = sys::CompareAndSwap(&initialized
, 1, 0); if (old_val == 0) { initializeTailCallElimPassOnce(Registry
); sys::MemoryFence(); ; ; initialized = 2; ; } else { sys::cas_flag
tmp = initialized; sys::MemoryFence(); while (tmp != 2) { tmp
= initialized; sys::MemoryFence(); } } ; }
131 "Tail Call Elimination", false, false)PassInfo *PI = new PassInfo("Tail Call Elimination", "tailcallelim"
, & TailCallElim ::ID, PassInfo::NormalCtor_t(callDefaultCtor
< TailCallElim >), false, false); Registry.registerPass
(*PI, true); return PI; } void llvm::initializeTailCallElimPass
(PassRegistry &Registry) { static volatile sys::cas_flag initialized
= 0; sys::cas_flag old_val = sys::CompareAndSwap(&initialized
, 1, 0); if (old_val == 0) { initializeTailCallElimPassOnce(Registry
); sys::MemoryFence(); ; ; initialized = 2; ; } else { sys::cas_flag
tmp = initialized; sys::MemoryFence(); while (tmp != 2) { tmp
= initialized; sys::MemoryFence(); } } ; }
132
133// Public interface to the TailCallElimination pass
134FunctionPass *llvm::createTailCallEliminationPass() {
135 return new TailCallElim();
136}
137
138void TailCallElim::getAnalysisUsage(AnalysisUsage &AU) const {
139 AU.addRequired<TargetTransformInfoWrapperPass>();
140 AU.addPreserved<GlobalsAAWrapperPass>();
141}
142
143/// \brief Scan the specified function for alloca instructions.
144/// If it contains any dynamic allocas, returns false.
145static bool CanTRE(Function &F) {
146 // Because of PR962, we don't TRE dynamic allocas.
147 for (auto &BB : F) {
148 for (auto &I : BB) {
149 if (AllocaInst *AI = dyn_cast<AllocaInst>(&I)) {
150 if (!AI->isStaticAlloca())
151 return false;
152 }
153 }
154 }
155
156 return true;
157}
158
159bool TailCallElim::runOnFunction(Function &F) {
160 if (skipFunction(F))
1
Taking false branch
→
161 return false;
162
163 if (F.getFnAttribute("disable-tail-calls").getValueAsString() == "true")
2
←
Taking false branch
→
164 return false;
165
166 bool AllCallsAreTailCalls = false;
167 bool Modified = markTails(F, AllCallsAreTailCalls);
168 if (AllCallsAreTailCalls)
3
←
Assuming 'AllCallsAreTailCalls' is not equal to 0
→
4
←
Taking true branch
→
169 Modified |= runTRE(F);
5
←
Calling 'TailCallElim::runTRE'
→
170 return Modified;
171}
172
173namespace {
174struct AllocaDerivedValueTracker {
175 // Start at a root value and walk its use-def chain to mark calls that use the
176 // value or a derived value in AllocaUsers, and places where it may escape in
177 // EscapePoints.
178 void walk(Value *Root) {
179 SmallVector<Use *, 32> Worklist;
180 SmallPtrSet<Use *, 32> Visited;
181
182 auto AddUsesToWorklist = [&](Value *V) {
183 for (auto &U : V->uses()) {
184 if (!Visited.insert(&U).second)
185 continue;
186 Worklist.push_back(&U);
187 }
188 };
189
190 AddUsesToWorklist(Root);
191
192 while (!Worklist.empty()) {
193 Use *U = Worklist.pop_back_val();
194 Instruction *I = cast<Instruction>(U->getUser());
195
196 switch (I->getOpcode()) {
197 case Instruction::Call:
198 case Instruction::Invoke: {
199 CallSite CS(I);
200 bool IsNocapture =
201 CS.isDataOperand(U) && CS.doesNotCapture(CS.getDataOperandNo(U));
202 callUsesLocalStack(CS, IsNocapture);
203 if (IsNocapture) {
204 // If the alloca-derived argument is passed in as nocapture, then it
205 // can't propagate to the call's return. That would be capturing.
206 continue;
207 }
208 break;
209 }
210 case Instruction::Load: {
211 // The result of a load is not alloca-derived (unless an alloca has
212 // otherwise escaped, but this is a local analysis).
213 continue;
214 }
215 case Instruction::Store: {
216 if (U->getOperandNo() == 0)
217 EscapePoints.insert(I);
218 continue; // Stores have no users to analyze.
219 }
220 case Instruction::BitCast:
221 case Instruction::GetElementPtr:
222 case Instruction::PHI:
223 case Instruction::Select:
224 case Instruction::AddrSpaceCast:
225 break;
226 default:
227 EscapePoints.insert(I);
228 break;
229 }
230
231 AddUsesToWorklist(I);
232 }
233 }
234
235 void callUsesLocalStack(CallSite CS, bool IsNocapture) {
236 // Add it to the list of alloca users.
237 AllocaUsers.insert(CS.getInstruction());
238
239 // If it's nocapture then it can't capture this alloca.
240 if (IsNocapture)
241 return;
242
243 // If it can write to memory, it can leak the alloca value.
244 if (!CS.onlyReadsMemory())
245 EscapePoints.insert(CS.getInstruction());
246 }
247
248 SmallPtrSet<Instruction *, 32> AllocaUsers;
249 SmallPtrSet<Instruction *, 32> EscapePoints;
250};
251}
252
253bool TailCallElim::markTails(Function &F, bool &AllCallsAreTailCalls) {
254 if (F.callsFunctionThatReturnsTwice())
255 return false;
256 AllCallsAreTailCalls = true;
257
258 // The local stack holds all alloca instructions and all byval arguments.
259 AllocaDerivedValueTracker Tracker;
260 for (Argument &Arg : F.args()) {
261 if (Arg.hasByValAttr())
262 Tracker.walk(&Arg);
263 }
264 for (auto &BB : F) {
265 for (auto &I : BB)
266 if (AllocaInst *AI = dyn_cast<AllocaInst>(&I))
267 Tracker.walk(AI);
268 }
269
270 bool Modified = false;
271
272 // Track whether a block is reachable after an alloca has escaped. Blocks that
273 // contain the escaping instruction will be marked as being visited without an
274 // escaped alloca, since that is how the block began.
275 enum VisitType {
276 UNVISITED,
277 UNESCAPED,
278 ESCAPED
279 };
280 DenseMap<BasicBlock *, VisitType> Visited;
281
282 // We propagate the fact that an alloca has escaped from block to successor.
283 // Visit the blocks that are propagating the escapedness first. To do this, we
284 // maintain two worklists.
285 SmallVector<BasicBlock *, 32> WorklistUnescaped, WorklistEscaped;
286
287 // We may enter a block and visit it thinking that no alloca has escaped yet,
288 // then see an escape point and go back around a loop edge and come back to
289 // the same block twice. Because of this, we defer setting tail on calls when
290 // we first encounter them in a block. Every entry in this list does not
291 // statically use an alloca via use-def chain analysis, but may find an alloca
292 // through other means if the block turns out to be reachable after an escape
293 // point.
294 SmallVector<CallInst *, 32> DeferredTails;
295
296 BasicBlock *BB = &F.getEntryBlock();
297 VisitType Escaped = UNESCAPED;
298 do {
299 for (auto &I : *BB) {
300 if (Tracker.EscapePoints.count(&I))
301 Escaped = ESCAPED;
302
303 CallInst *CI = dyn_cast<CallInst>(&I);
304 if (!CI || CI->isTailCall())
305 continue;
306
307 bool IsNoTail = CI->isNoTailCall();
308
309 if (!IsNoTail && CI->doesNotAccessMemory()) {
310 // A call to a readnone function whose arguments are all things computed
311 // outside this function can be marked tail. Even if you stored the
312 // alloca address into a global, a readnone function can't load the
313 // global anyhow.
314 //
315 // Note that this runs whether we know an alloca has escaped or not. If
316 // it has, then we can't trust Tracker.AllocaUsers to be accurate.
317 bool SafeToTail = true;
318 for (auto &Arg : CI->arg_operands()) {
319 if (isa<Constant>(Arg.getUser()))
320 continue;
321 if (Argument *A = dyn_cast<Argument>(Arg.getUser()))
322 if (!A->hasByValAttr())
323 continue;
324 SafeToTail = false;
325 break;
326 }
327 if (SafeToTail) {
328 emitOptimizationRemark(
329 F.getContext(), "tailcallelim", F, CI->getDebugLoc(),
330 "marked this readnone call a tail call candidate");
331 CI->setTailCall();
332 Modified = true;
333 continue;
334 }
335 }
336
337 if (!IsNoTail && Escaped == UNESCAPED && !Tracker.AllocaUsers.count(CI)) {
338 DeferredTails.push_back(CI);
339 } else {
340 AllCallsAreTailCalls = false;
341 }
342 }
343
344 for (auto *SuccBB : make_range(succ_begin(BB), succ_end(BB))) {
345 auto &State = Visited[SuccBB];
346 if (State < Escaped) {
347 State = Escaped;
348 if (State == ESCAPED)
349 WorklistEscaped.push_back(SuccBB);
350 else
351 WorklistUnescaped.push_back(SuccBB);
352 }
353 }
354
355 if (!WorklistEscaped.empty()) {
356 BB = WorklistEscaped.pop_back_val();
357 Escaped = ESCAPED;
358 } else {
359 BB = nullptr;
360 while (!WorklistUnescaped.empty()) {
361 auto *NextBB = WorklistUnescaped.pop_back_val();
362 if (Visited[NextBB] == UNESCAPED) {
363 BB = NextBB;
364 Escaped = UNESCAPED;
365 break;
366 }
367 }
368 }
369 } while (BB);
370
371 for (CallInst *CI : DeferredTails) {
372 if (Visited[CI->getParent()] != ESCAPED) {
373 // If the escape point was part way through the block, calls after the
374 // escape point wouldn't have been put into DeferredTails.
375 emitOptimizationRemark(F.getContext(), "tailcallelim", F,
376 CI->getDebugLoc(),
377 "marked this call a tail call candidate");
378 CI->setTailCall();
379 Modified = true;
380 } else {
381 AllCallsAreTailCalls = false;
382 }
383 }
384
385 return Modified;
386}
387
388bool TailCallElim::runTRE(Function &F) {
389 // If this function is a varargs function, we won't be able to PHI the args
390 // right, so don't even try to convert it...
391 if (F.getFunctionType()->isVarArg()) return false;
6
←
Taking false branch
→
392
393 TTI = &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
394 BasicBlock *OldEntry = nullptr;
395 bool TailCallsAreMarkedTail = false;
396 SmallVector<PHINode*, 8> ArgumentPHIs;
397 bool MadeChange = false;
398
399 // If false, we cannot perform TRE on tail calls marked with the 'tail'
400 // attribute, because doing so would cause the stack size to increase (real
401 // TRE would deallocate variable sized allocas, TRE doesn't).
402 bool CanTRETailMarkedCall = CanTRE(F);
403
404 // Change any tail recursive calls to loops.
405 //
406 // FIXME: The code generator produces really bad code when an 'escaping
407 // alloca' is changed from being a static alloca to being a dynamic alloca.
408 // Until this is resolved, disable this transformation if that would ever
409 // happen. This bug is PR962.
410 for (Function::iterator BBI = F.begin(), E = F.end(); BBI != E; /*in loop*/) {
7
←
Loop condition is true. Entering loop body
→
9
←
Loop condition is true. Entering loop body
→
11
←
Loop condition is true. Entering loop body
→
13
←
Loop condition is true. Entering loop body
→
411 BasicBlock *BB = &*BBI++; // FoldReturnAndProcessPred may delete BB.
412 if (ReturnInst *Ret = dyn_cast<ReturnInst>(BB->getTerminator())) {
8
←
Taking false branch
→
10
←
Taking false branch
→
12
←
Taking false branch
→
14
←
Assuming 'Ret' is non-null
→
15
←
Taking true branch
→
413 bool Change = ProcessReturningBlock(Ret, OldEntry, TailCallsAreMarkedTail,
16
←
Calling 'TailCallElim::ProcessReturningBlock'
→
414 ArgumentPHIs, !CanTRETailMarkedCall);
415 if (!Change && BB->getFirstNonPHIOrDbg() == Ret)
416 Change = FoldReturnAndProcessPred(BB, Ret, OldEntry,
417 TailCallsAreMarkedTail, ArgumentPHIs,
418 !CanTRETailMarkedCall);
419 MadeChange |= Change;
420 }
421 }
422
423 // If we eliminated any tail recursions, it's possible that we inserted some
424 // silly PHI nodes which just merge an initial value (the incoming operand)
425 // with themselves. Check to see if we did and clean up our mess if so. This
426 // occurs when a function passes an argument straight through to its tail
427 // call.
428 for (PHINode *PN : ArgumentPHIs) {
429 // If the PHI Node is a dynamic constant, replace it with the value it is.
430 if (Value *PNV = SimplifyInstruction(PN, F.getParent()->getDataLayout())) {
431 PN->replaceAllUsesWith(PNV);
432 PN->eraseFromParent();
433 }
434 }
435
436 return MadeChange;
437}
438
439
440/// Return true if it is safe to move the specified
441/// instruction from after the call to before the call, assuming that all
442/// instructions between the call and this instruction are movable.
443///
444bool TailCallElim::CanMoveAboveCall(Instruction *I, CallInst *CI) {
445 // FIXME: We can move load/store/call/free instructions above the call if the
446 // call does not mod/ref the memory location being processed.
447 if (I->mayHaveSideEffects()) // This also handles volatile loads.
448 return false;
449
450 if (LoadInst *L = dyn_cast<LoadInst>(I)) {
451 // Loads may always be moved above calls without side effects.
452 if (CI->mayHaveSideEffects()) {
453 // Non-volatile loads may be moved above a call with side effects if it
454 // does not write to memory and the load provably won't trap.
455 // FIXME: Writes to memory only matter if they may alias the pointer
456 // being loaded from.
457 const DataLayout &DL = L->getModule()->getDataLayout();
458 if (CI->mayWriteToMemory() ||
459 !isSafeToLoadUnconditionally(L->getPointerOperand(),
460 L->getAlignment(), DL, L))
461 return false;
462 }
463 }
464
465 // Otherwise, if this is a side-effect free instruction, check to make sure
466 // that it does not use the return value of the call. If it doesn't use the
467 // return value of the call, it must only use things that are defined before
468 // the call, or movable instructions between the call and the instruction
469 // itself.
470 return std::find(I->op_begin(), I->op_end(), CI) == I->op_end();
471}
472
473/// Return true if the specified value is the same when the return would exit
474/// as it was when the initial iteration of the recursive function was executed.
475///
476/// We currently handle static constants and arguments that are not modified as
477/// part of the recursion.
478static bool isDynamicConstant(Value *V, CallInst *CI, ReturnInst *RI) {
479 if (isa<Constant>(V)) return true; // Static constants are always dyn consts
480
481 // Check to see if this is an immutable argument, if so, the value
482 // will be available to initialize the accumulator.
483 if (Argument *Arg = dyn_cast<Argument>(V)) {
484 // Figure out which argument number this is...
485 unsigned ArgNo = 0;
486 Function *F = CI->getParent()->getParent();
487 for (Function::arg_iterator AI = F->arg_begin(); &*AI != Arg; ++AI)
488 ++ArgNo;
489
490 // If we are passing this argument into call as the corresponding
491 // argument operand, then the argument is dynamically constant.
492 // Otherwise, we cannot transform this function safely.
493 if (CI->getArgOperand(ArgNo) == Arg)
494 return true;
495 }
496
497 // Switch cases are always constant integers. If the value is being switched
498 // on and the return is only reachable from one of its cases, it's
499 // effectively constant.
500 if (BasicBlock *UniquePred = RI->getParent()->getUniquePredecessor())
501 if (SwitchInst *SI = dyn_cast<SwitchInst>(UniquePred->getTerminator()))
502 if (SI->getCondition() == V)
503 return SI->getDefaultDest() != RI->getParent();
504
505 // Not a constant or immutable argument, we can't safely transform.
506 return false;
507}
508
509/// Check to see if the function containing the specified tail call consistently
510/// returns the same runtime-constant value at all exit points except for
511/// IgnoreRI. If so, return the returned value.
512static Value *getCommonReturnValue(ReturnInst *IgnoreRI, CallInst *CI) {
513 Function *F = CI->getParent()->getParent();
514 Value *ReturnedValue = nullptr;
515
516 for (Function::iterator BBI = F->begin(), E = F->end(); BBI != E; ++BBI) {
517 ReturnInst *RI = dyn_cast<ReturnInst>(BBI->getTerminator());
518 if (RI == nullptr || RI == IgnoreRI) continue;
519
520 // We can only perform this transformation if the value returned is
521 // evaluatable at the start of the initial invocation of the function,
522 // instead of at the end of the evaluation.
523 //
524 Value *RetOp = RI->getOperand(0);
525 if (!isDynamicConstant(RetOp, CI, RI))
526 return nullptr;
527
528 if (ReturnedValue && RetOp != ReturnedValue)
529 return nullptr; // Cannot transform if differing values are returned.
530 ReturnedValue = RetOp;
531 }
532 return ReturnedValue;
533}
534
535/// If the specified instruction can be transformed using accumulator recursion
536/// elimination, return the constant which is the start of the accumulator
537/// value. Otherwise return null.
538Value *TailCallElim::CanTransformAccumulatorRecursion(Instruction *I,
539 CallInst *CI) {
540 if (!I->isAssociative() || !I->isCommutative()) return nullptr;
541 assert(I->getNumOperands() == 2 &&((I->getNumOperands() == 2 && "Associative/commutative operations should have 2 args!"
) ? static_cast<void> (0) : __assert_fail ("I->getNumOperands() == 2 && \"Associative/commutative operations should have 2 args!\""
, "/tmp/buildd/llvm-toolchain-snapshot-3.9~svn271203/lib/Transforms/Scalar/TailRecursionElimination.cpp"
, 542, __PRETTY_FUNCTION__))
542 "Associative/commutative operations should have 2 args!")((I->getNumOperands() == 2 && "Associative/commutative operations should have 2 args!"
) ? static_cast<void> (0) : __assert_fail ("I->getNumOperands() == 2 && \"Associative/commutative operations should have 2 args!\""
, "/tmp/buildd/llvm-toolchain-snapshot-3.9~svn271203/lib/Transforms/Scalar/TailRecursionElimination.cpp"
, 542, __PRETTY_FUNCTION__));
543
544 // Exactly one operand should be the result of the call instruction.
545 if ((I->getOperand(0) == CI && I->getOperand(1) == CI) ||
546 (I->getOperand(0) != CI && I->getOperand(1) != CI))
547 return nullptr;
548
549 // The only user of this instruction we allow is a single return instruction.
550 if (!I->hasOneUse() || !isa<ReturnInst>(I->user_back()))
551 return nullptr;
552
553 // Ok, now we have to check all of the other return instructions in this
554 // function. If they return non-constants or differing values, then we cannot
555 // transform the function safely.
556 return getCommonReturnValue(cast<ReturnInst>(I->user_back()), CI);
557}
558
559static Instruction *FirstNonDbg(BasicBlock::iterator I) {
560 while (isa<DbgInfoIntrinsic>(I))
561 ++I;
562 return &*I;
563}
564
565CallInst*
566TailCallElim::FindTRECandidate(Instruction *TI,
567 bool CannotTailCallElimCallsMarkedTail) {
568 BasicBlock *BB = TI->getParent();
569 Function *F = BB->getParent();
18
←
'F' initialized here
→
570
571 if (&BB->front() == TI) // Make sure there is something before the terminator.
19
←
Taking false branch
→
572 return nullptr;
573
574 // Scan backwards from the return, checking to see if there is a tail call in
575 // this block. If so, set CI to it.
576 CallInst *CI = nullptr;
577 BasicBlock::iterator BBI(TI);
578 while (true) {
20
←
Loop condition is true. Entering loop body
→
579 CI = dyn_cast<CallInst>(BBI);
580 if (CI && CI->getCalledFunction() == F)
21
←
Assuming pointer value is null
→
22
←
Taking true branch
→
581 break;
23
←
 Execution continues on line 590
→
582
583 if (BBI == BB->begin())
584 return nullptr; // Didn't find a potential tail call.
585 --BBI;
586 }
587
588 // If this call is marked as a tail call, and if there are dynamic allocas in
589 // the function, we cannot perform this optimization.
590 if (CI->isTailCall() && CannotTailCallElimCallsMarkedTail)
24
←
Taking false branch
→
591 return nullptr;
592
593 // As a special case, detect code like this:
594 // double fabs(double f) { return __builtin_fabs(f); } // a 'fabs' call
595 // and disable this xform in this case, because the code generator will
596 // lower the call to fabs into inline code.
597 if (BB == &F->getEntryBlock() &&
25
←
Called C++ object pointer is null
598 FirstNonDbg(BB->front().getIterator()) == CI &&
599 FirstNonDbg(std::next(BB->begin())) == TI && CI->getCalledFunction() &&
600 !TTI->isLoweredToCall(CI->getCalledFunction())) {
601 // A single-block function with just a call and a return. Check that
602 // the arguments match.
603 CallSite::arg_iterator I = CallSite(CI).arg_begin(),
604 E = CallSite(CI).arg_end();
605 Function::arg_iterator FI = F->arg_begin(),
606 FE = F->arg_end();
607 for (; I != E && FI != FE; ++I, ++FI)
608 if (*I != &*FI) break;
609 if (I == E && FI == FE)
610 return nullptr;
611 }
612
613 return CI;
614}
615
616bool TailCallElim::EliminateRecursiveTailCall(CallInst *CI, ReturnInst *Ret,
617 BasicBlock *&OldEntry,
618 bool &TailCallsAreMarkedTail,
619 SmallVectorImpl<PHINode *> &ArgumentPHIs,
620 bool CannotTailCallElimCallsMarkedTail) {
621 // If we are introducing accumulator recursion to eliminate operations after
622 // the call instruction that are both associative and commutative, the initial
623 // value for the accumulator is placed in this variable. If this value is set
624 // then we actually perform accumulator recursion elimination instead of
625 // simple tail recursion elimination. If the operation is an LLVM instruction
626 // (eg: "add") then it is recorded in AccumulatorRecursionInstr. If not, then
627 // we are handling the case when the return instruction returns a constant C
628 // which is different to the constant returned by other return instructions
629 // (which is recorded in AccumulatorRecursionEliminationInitVal). This is a
630 // special case of accumulator recursion, the operation being "return C".
631 Value *AccumulatorRecursionEliminationInitVal = nullptr;
632 Instruction *AccumulatorRecursionInstr = nullptr;
633
634 // Ok, we found a potential tail call. We can currently only transform the
635 // tail call if all of the instructions between the call and the return are
636 // movable to above the call itself, leaving the call next to the return.
637 // Check that this is the case now.
638 BasicBlock::iterator BBI(CI);
639 for (++BBI; &*BBI != Ret; ++BBI) {
640 if (CanMoveAboveCall(&*BBI, CI)) continue;
641
642 // If we can't move the instruction above the call, it might be because it
643 // is an associative and commutative operation that could be transformed
644 // using accumulator recursion elimination. Check to see if this is the
645 // case, and if so, remember the initial accumulator value for later.
646 if ((AccumulatorRecursionEliminationInitVal =
647 CanTransformAccumulatorRecursion(&*BBI, CI))) {
648 // Yes, this is accumulator recursion. Remember which instruction
649 // accumulates.
650 AccumulatorRecursionInstr = &*BBI;
651 } else {
652 return false; // Otherwise, we cannot eliminate the tail recursion!
653 }
654 }
655
656 // We can only transform call/return pairs that either ignore the return value
657 // of the call and return void, ignore the value of the call and return a
658 // constant, return the value returned by the tail call, or that are being
659 // accumulator recursion variable eliminated.
660 if (Ret->getNumOperands() == 1 && Ret->getReturnValue() != CI &&
661 !isa<UndefValue>(Ret->getReturnValue()) &&
662 AccumulatorRecursionEliminationInitVal == nullptr &&
663 !getCommonReturnValue(nullptr, CI)) {
664 // One case remains that we are able to handle: the current return
665 // instruction returns a constant, and all other return instructions
666 // return a different constant.
667 if (!isDynamicConstant(Ret->getReturnValue(), CI, Ret))
668 return false; // Current return instruction does not return a constant.
669 // Check that all other return instructions return a common constant. If
670 // so, record it in AccumulatorRecursionEliminationInitVal.
671 AccumulatorRecursionEliminationInitVal = getCommonReturnValue(Ret, CI);
672 if (!AccumulatorRecursionEliminationInitVal)
673 return false;
674 }
675
676 BasicBlock *BB = Ret->getParent();
677 Function *F = BB->getParent();
678
679 emitOptimizationRemark(F->getContext(), "tailcallelim", *F, CI->getDebugLoc(),
680 "transforming tail recursion to loop");
681
682 // OK! We can transform this tail call. If this is the first one found,
683 // create the new entry block, allowing us to branch back to the old entry.
684 if (!OldEntry) {
685 OldEntry = &F->getEntryBlock();
686 BasicBlock *NewEntry = BasicBlock::Create(F->getContext(), "", F, OldEntry);
687 NewEntry->takeName(OldEntry);
688 OldEntry->setName("tailrecurse");
689 BranchInst::Create(OldEntry, NewEntry);
690
691 // If this tail call is marked 'tail' and if there are any allocas in the
692 // entry block, move them up to the new entry block.
693 TailCallsAreMarkedTail = CI->isTailCall();
694 if (TailCallsAreMarkedTail)
695 // Move all fixed sized allocas from OldEntry to NewEntry.
696 for (BasicBlock::iterator OEBI = OldEntry->begin(), E = OldEntry->end(),
697 NEBI = NewEntry->begin(); OEBI != E; )
698 if (AllocaInst *AI = dyn_cast<AllocaInst>(OEBI++))
699 if (isa<ConstantInt>(AI->getArraySize()))
700 AI->moveBefore(&*NEBI);
701
702 // Now that we have created a new block, which jumps to the entry
703 // block, insert a PHI node for each argument of the function.
704 // For now, we initialize each PHI to only have the real arguments
705 // which are passed in.
706 Instruction *InsertPos = &OldEntry->front();
707 for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end();
708 I != E; ++I) {
709 PHINode *PN = PHINode::Create(I->getType(), 2,
710 I->getName() + ".tr", InsertPos);
711 I->replaceAllUsesWith(PN); // Everyone use the PHI node now!
712 PN->addIncoming(&*I, NewEntry);
713 ArgumentPHIs.push_back(PN);
714 }
715 }
716
717 // If this function has self recursive calls in the tail position where some
718 // are marked tail and some are not, only transform one flavor or another. We
719 // have to choose whether we move allocas in the entry block to the new entry
720 // block or not, so we can't make a good choice for both. NOTE: We could do
721 // slightly better here in the case that the function has no entry block
722 // allocas.
723 if (TailCallsAreMarkedTail && !CI->isTailCall())
724 return false;
725
726 // Ok, now that we know we have a pseudo-entry block WITH all of the
727 // required PHI nodes, add entries into the PHI node for the actual
728 // parameters passed into the tail-recursive call.
729 for (unsigned i = 0, e = CI->getNumArgOperands(); i != e; ++i)
730 ArgumentPHIs[i]->addIncoming(CI->getArgOperand(i), BB);
731
732 // If we are introducing an accumulator variable to eliminate the recursion,
733 // do so now. Note that we _know_ that no subsequent tail recursion
734 // eliminations will happen on this function because of the way the
735 // accumulator recursion predicate is set up.
736 //
737 if (AccumulatorRecursionEliminationInitVal) {
738 Instruction *AccRecInstr = AccumulatorRecursionInstr;
739 // Start by inserting a new PHI node for the accumulator.
740 pred_iterator PB = pred_begin(OldEntry), PE = pred_end(OldEntry);
741 PHINode *AccPN = PHINode::Create(
742 AccumulatorRecursionEliminationInitVal->getType(),
743 std::distance(PB, PE) + 1, "accumulator.tr", &OldEntry->front());
744
745 // Loop over all of the predecessors of the tail recursion block. For the
746 // real entry into the function we seed the PHI with the initial value,
747 // computed earlier. For any other existing branches to this block (due to
748 // other tail recursions eliminated) the accumulator is not modified.
749 // Because we haven't added the branch in the current block to OldEntry yet,
750 // it will not show up as a predecessor.
751 for (pred_iterator PI = PB; PI != PE; ++PI) {
752 BasicBlock *P = *PI;
753 if (P == &F->getEntryBlock())
754 AccPN->addIncoming(AccumulatorRecursionEliminationInitVal, P);
755 else
756 AccPN->addIncoming(AccPN, P);
757 }
758
759 if (AccRecInstr) {
760 // Add an incoming argument for the current block, which is computed by
761 // our associative and commutative accumulator instruction.
762 AccPN->addIncoming(AccRecInstr, BB);
763
764 // Next, rewrite the accumulator recursion instruction so that it does not
765 // use the result of the call anymore, instead, use the PHI node we just
766 // inserted.
767 AccRecInstr->setOperand(AccRecInstr->getOperand(0) != CI, AccPN);
768 } else {
769 // Add an incoming argument for the current block, which is just the
770 // constant returned by the current return instruction.
771 AccPN->addIncoming(Ret->getReturnValue(), BB);
772 }
773
774 // Finally, rewrite any return instructions in the program to return the PHI
775 // node instead of the "initval" that they do currently. This loop will
776 // actually rewrite the return value we are destroying, but that's ok.
777 for (Function::iterator BBI = F->begin(), E = F->end(); BBI != E; ++BBI)
778 if (ReturnInst *RI = dyn_cast<ReturnInst>(BBI->getTerminator()))
779 RI->setOperand(0, AccPN);
780 ++NumAccumAdded;
781 }
782
783 // Now that all of the PHI nodes are in place, remove the call and
784 // ret instructions, replacing them with an unconditional branch.
785 BranchInst *NewBI = BranchInst::Create(OldEntry, Ret);
786 NewBI->setDebugLoc(CI->getDebugLoc());
787
788 BB->getInstList().erase(Ret); // Remove return.
789 BB->getInstList().erase(CI); // Remove call.
790 ++NumEliminated;
791 return true;
792}
793
794bool TailCallElim::FoldReturnAndProcessPred(BasicBlock *BB,
795 ReturnInst *Ret, BasicBlock *&OldEntry,
796 bool &TailCallsAreMarkedTail,
797 SmallVectorImpl<PHINode *> &ArgumentPHIs,
798 bool CannotTailCallElimCallsMarkedTail) {
799 bool Change = false;
800
801 // If the return block contains nothing but the return and PHI's,
802 // there might be an opportunity to duplicate the return in its
803 // predecessors and perform TRC there. Look for predecessors that end
804 // in unconditional branch and recursive call(s).
805 SmallVector<BranchInst*, 8> UncondBranchPreds;
806 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
807 BasicBlock *Pred = *PI;
808 TerminatorInst *PTI = Pred->getTerminator();
809 if (BranchInst *BI = dyn_cast<BranchInst>(PTI))
810 if (BI->isUnconditional())
811 UncondBranchPreds.push_back(BI);
812 }
813
814 while (!UncondBranchPreds.empty()) {
815 BranchInst *BI = UncondBranchPreds.pop_back_val();
816 BasicBlock *Pred = BI->getParent();
817 if (CallInst *CI = FindTRECandidate(BI, CannotTailCallElimCallsMarkedTail)){
818 DEBUG(dbgs() << "FOLDING: " << *BBdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("tailcallelim")) { dbgs() << "FOLDING: " << *BB <<
"INTO UNCOND BRANCH PRED: " << *Pred; } } while (0)
819 << "INTO UNCOND BRANCH PRED: " << *Pred)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("tailcallelim")) { dbgs() << "FOLDING: " << *BB <<
"INTO UNCOND BRANCH PRED: " << *Pred; } } while (0);
820 ReturnInst *RI = FoldReturnIntoUncondBranch(Ret, BB, Pred);
821
822 // Cleanup: if all predecessors of BB have been eliminated by
823 // FoldReturnIntoUncondBranch, delete it. It is important to empty it,
824 // because the ret instruction in there is still using a value which
825 // EliminateRecursiveTailCall will attempt to remove.
826 if (!BB->hasAddressTaken() && pred_begin(BB) == pred_end(BB))
827 BB->eraseFromParent();
828
829 EliminateRecursiveTailCall(CI, RI, OldEntry, TailCallsAreMarkedTail,
830 ArgumentPHIs,
831 CannotTailCallElimCallsMarkedTail);
832 ++NumRetDuped;
833 Change = true;
834 }
835 }
836
837 return Change;
838}
839
840bool
841TailCallElim::ProcessReturningBlock(ReturnInst *Ret, BasicBlock *&OldEntry,
842 bool &TailCallsAreMarkedTail,
843 SmallVectorImpl<PHINode *> &ArgumentPHIs,
844 bool CannotTailCallElimCallsMarkedTail) {
845 CallInst *CI = FindTRECandidate(Ret, CannotTailCallElimCallsMarkedTail);
17
←
Calling 'TailCallElim::FindTRECandidate'
→
846 if (!CI)
847 return false;
848
849 return EliminateRecursiveTailCall(CI, Ret, OldEntry, TailCallsAreMarkedTail,
850 ArgumentPHIs,
851 CannotTailCallElimCallsMarkedTail);
852}