summaryrefslogtreecommitdiff
path: root/lib/Transforms/Scalar/LoopPredication.cpp
blob: 52dea3254e79e5f98f36caa68630970553ad9a42 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
//===-- LoopPredication.cpp - Guard based loop predication pass -----------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// The LoopPredication pass tries to convert loop variant range checks to loop
// invariant by widening checks across loop iterations. For example, it will
// convert
//
//   for (i = 0; i < n; i++) {
//     guard(i < len);
//     ...
//   }
//
// to
//
//   for (i = 0; i < n; i++) {
//     guard(n - 1 < len);
//     ...
//   }
//
// After this transformation the condition of the guard is loop invariant, so
// loop-unswitch can later unswitch the loop by this condition which basically
// predicates the loop by the widened condition:
//
//   if (n - 1 < len)
//     for (i = 0; i < n; i++) {
//       ...
//     }
//   else
//     deoptimize
//
// It's tempting to rely on SCEV here, but it has proven to be problematic.
// Generally the facts SCEV provides about the increment step of add
// recurrences are true if the backedge of the loop is taken, which implicitly
// assumes that the guard doesn't fail. Using these facts to optimize the
// guard results in a circular logic where the guard is optimized under the
// assumption that it never fails.
//
// For example, in the loop below the induction variable will be marked as nuw
// basing on the guard. Basing on nuw the guard predicate will be considered
// monotonic. Given a monotonic condition it's tempting to replace the induction
// variable in the condition with its value on the last iteration. But this
// transformation is not correct, e.g. e = 4, b = 5 breaks the loop.
//
//   for (int i = b; i != e; i++)
//     guard(i u< len)
//
// One of the ways to reason about this problem is to use an inductive proof
// approach. Given the loop:
//
//   if (B(0)) {
//     do {
//       I = PHI(0, I.INC)
//       I.INC = I + Step
//       guard(G(I));
//     } while (B(I));
//   }
//
// where B(x) and G(x) are predicates that map integers to booleans, we want a
// loop invariant expression M such the following program has the same semantics
// as the above:
//
//   if (B(0)) {
//     do {
//       I = PHI(0, I.INC)
//       I.INC = I + Step
//       guard(G(0) && M);
//     } while (B(I));
//   }
//
// One solution for M is M = forall X . (G(X) && B(X)) => G(X + Step)
// 
// Informal proof that the transformation above is correct:
//
//   By the definition of guards we can rewrite the guard condition to:
//     G(I) && G(0) && M
//
//   Let's prove that for each iteration of the loop:
//     G(0) && M => G(I)
//   And the condition above can be simplified to G(Start) && M.
// 
//   Induction base.
//     G(0) && M => G(0)
//
//   Induction step. Assuming G(0) && M => G(I) on the subsequent
//   iteration:
//
//     B(I) is true because it's the backedge condition.
//     G(I) is true because the backedge is guarded by this condition.
//
//   So M = forall X . (G(X) && B(X)) => G(X + Step) implies G(I + Step).
//
// Note that we can use anything stronger than M, i.e. any condition which
// implies M.
//
// For now the transformation is limited to the following case:
//   * The loop has a single latch with the condition of the form:
//     B(X) = latchStart + X <pred> latchLimit,
//     where <pred> is u<, u<=, s<, or s<=.
//   * The step of the IV used in the latch condition is 1.
//   * The guard condition is of the form
//     G(X) = guardStart + X u< guardLimit
//
// For the ult latch comparison case M is:
//   forall X . guardStart + X u< guardLimit && latchStart + X <u latchLimit =>
//      guardStart + X + 1 u< guardLimit
//
// The only way the antecedent can be true and the consequent can be false is
// if
//   X == guardLimit - 1 - guardStart
// (and guardLimit is non-zero, but we won't use this latter fact).
// If X == guardLimit - 1 - guardStart then the second half of the antecedent is
//   latchStart + guardLimit - 1 - guardStart u< latchLimit
// and its negation is
//   latchStart + guardLimit - 1 - guardStart u>= latchLimit
//
// In other words, if
//   latchLimit u<= latchStart + guardLimit - 1 - guardStart
// then:
// (the ranges below are written in ConstantRange notation, where [A, B) is the
// set for (I = A; I != B; I++ /*maywrap*/) yield(I);)
//
//    forall X . guardStart + X u< guardLimit &&
//               latchStart + X u< latchLimit =>
//      guardStart + X + 1 u< guardLimit
// == forall X . guardStart + X u< guardLimit &&
//               latchStart + X u< latchStart + guardLimit - 1 - guardStart =>
//      guardStart + X + 1 u< guardLimit
// == forall X . (guardStart + X) in [0, guardLimit) &&
//               (latchStart + X) in [0, latchStart + guardLimit - 1 - guardStart) =>
//      (guardStart + X + 1) in [0, guardLimit)
// == forall X . X in [-guardStart, guardLimit - guardStart) &&
//               X in [-latchStart, guardLimit - 1 - guardStart) =>
//       X in [-guardStart - 1, guardLimit - guardStart - 1)
// == true
//
// So the widened condition is:
//   guardStart u< guardLimit &&
//   latchStart + guardLimit - 1 - guardStart u>= latchLimit
// Similarly for ule condition the widened condition is:
//   guardStart u< guardLimit &&
//   latchStart + guardLimit - 1 - guardStart u> latchLimit
// For slt condition the widened condition is:
//   guardStart u< guardLimit &&
//   latchStart + guardLimit - 1 - guardStart s>= latchLimit
// For sle condition the widened condition is:
//   guardStart u< guardLimit &&
//   latchStart + guardLimit - 1 - guardStart s> latchLimit
//
//===----------------------------------------------------------------------===//

#include "llvm/Transforms/Scalar/LoopPredication.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/LoopPass.h"
#include "llvm/Analysis/ScalarEvolution.h"
#include "llvm/Analysis/ScalarEvolutionExpander.h"
#include "llvm/Analysis/ScalarEvolutionExpressions.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/PatternMatch.h"
#include "llvm/Pass.h"
#include "llvm/Support/Debug.h"
#include "llvm/Transforms/Scalar.h"
#include "llvm/Transforms/Utils/LoopUtils.h"

#define DEBUG_TYPE "loop-predication"

using namespace llvm;

static cl::opt<bool> EnableIVTruncation("loop-predication-enable-iv-truncation",
                                        cl::Hidden, cl::init(true));

namespace {
class LoopPredication {
  /// Represents an induction variable check:
  ///   icmp Pred, <induction variable>, <loop invariant limit>
  struct LoopICmp {
    ICmpInst::Predicate Pred;
    const SCEVAddRecExpr *IV;
    const SCEV *Limit;
    LoopICmp(ICmpInst::Predicate Pred, const SCEVAddRecExpr *IV,
             const SCEV *Limit)
        : Pred(Pred), IV(IV), Limit(Limit) {}
    LoopICmp() {}
    void dump() {
      dbgs() << "LoopICmp Pred = " << Pred << ", IV = " << *IV
             << ", Limit = " << *Limit << "\n";
    }
  };

  ScalarEvolution *SE;

  Loop *L;
  const DataLayout *DL;
  BasicBlock *Preheader;
  LoopICmp LatchCheck;

  bool isSupportedStep(const SCEV* Step);
  Optional<LoopICmp> parseLoopICmp(ICmpInst *ICI) {
    return parseLoopICmp(ICI->getPredicate(), ICI->getOperand(0),
                         ICI->getOperand(1));
  }
  Optional<LoopICmp> parseLoopICmp(ICmpInst::Predicate Pred, Value *LHS,
                                   Value *RHS);

  Optional<LoopICmp> parseLoopLatchICmp();

  bool CanExpand(const SCEV* S);
  Value *expandCheck(SCEVExpander &Expander, IRBuilder<> &Builder,
                     ICmpInst::Predicate Pred, const SCEV *LHS, const SCEV *RHS,
                     Instruction *InsertAt);

  Optional<Value *> widenICmpRangeCheck(ICmpInst *ICI, SCEVExpander &Expander,
                                        IRBuilder<> &Builder);
  Optional<Value *> widenICmpRangeCheckIncrementingLoop(LoopICmp LatchCheck,
                                                        LoopICmp RangeCheck,
                                                        SCEVExpander &Expander,
                                                        IRBuilder<> &Builder);

  bool widenGuardConditions(IntrinsicInst *II, SCEVExpander &Expander);

  // When the IV type is wider than the range operand type, we can still do loop
  // predication, by generating SCEVs for the range and latch that are of the
  // same type. We achieve this by generating a SCEV truncate expression for the
  // latch IV. This is done iff truncation of the IV is a safe operation,
  // without loss of information.
  // Another way to achieve this is by generating a wider type SCEV for the
  // range check operand, however, this needs a more involved check that
  // operands do not overflow. This can lead to loss of information when the
  // range operand is of the form: add i32 %offset, %iv. We need to prove that
  // sext(x + y) is same as sext(x) + sext(y).
  // This function returns true if we can safely represent the IV type in
  // the RangeCheckType without loss of information.
  bool isSafeToTruncateWideIVType(Type *RangeCheckType);
  // Return the loopLatchCheck corresponding to the RangeCheckType if safe to do
  // so.
  Optional<LoopICmp> generateLoopLatchCheck(Type *RangeCheckType);
public:
  LoopPredication(ScalarEvolution *SE) : SE(SE){};
  bool runOnLoop(Loop *L);
};

class LoopPredicationLegacyPass : public LoopPass {
public:
  static char ID;
  LoopPredicationLegacyPass() : LoopPass(ID) {
    initializeLoopPredicationLegacyPassPass(*PassRegistry::getPassRegistry());
  }

  void getAnalysisUsage(AnalysisUsage &AU) const override {
    getLoopAnalysisUsage(AU);
  }

  bool runOnLoop(Loop *L, LPPassManager &LPM) override {
    if (skipLoop(L))
      return false;
    auto *SE = &getAnalysis<ScalarEvolutionWrapperPass>().getSE();
    LoopPredication LP(SE);
    return LP.runOnLoop(L);
  }
};

char LoopPredicationLegacyPass::ID = 0;
} // end namespace llvm

INITIALIZE_PASS_BEGIN(LoopPredicationLegacyPass, "loop-predication",
                      "Loop predication", false, false)
INITIALIZE_PASS_DEPENDENCY(LoopPass)
INITIALIZE_PASS_END(LoopPredicationLegacyPass, "loop-predication",
                    "Loop predication", false, false)

Pass *llvm::createLoopPredicationPass() {
  return new LoopPredicationLegacyPass();
}

PreservedAnalyses LoopPredicationPass::run(Loop &L, LoopAnalysisManager &AM,
                                           LoopStandardAnalysisResults &AR,
                                           LPMUpdater &U) {
  LoopPredication LP(&AR.SE);
  if (!LP.runOnLoop(&L))
    return PreservedAnalyses::all();

  return getLoopPassPreservedAnalyses();
}

Optional<LoopPredication::LoopICmp>
LoopPredication::parseLoopICmp(ICmpInst::Predicate Pred, Value *LHS,
                               Value *RHS) {
  const SCEV *LHSS = SE->getSCEV(LHS);
  if (isa<SCEVCouldNotCompute>(LHSS))
    return None;
  const SCEV *RHSS = SE->getSCEV(RHS);
  if (isa<SCEVCouldNotCompute>(RHSS))
    return None;

  // Canonicalize RHS to be loop invariant bound, LHS - a loop computable IV
  if (SE->isLoopInvariant(LHSS, L)) {
    std::swap(LHS, RHS);
    std::swap(LHSS, RHSS);
    Pred = ICmpInst::getSwappedPredicate(Pred);
  }

  const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(LHSS);
  if (!AR || AR->getLoop() != L)
    return None;

  return LoopICmp(Pred, AR, RHSS);
}

Value *LoopPredication::expandCheck(SCEVExpander &Expander,
                                    IRBuilder<> &Builder,
                                    ICmpInst::Predicate Pred, const SCEV *LHS,
                                    const SCEV *RHS, Instruction *InsertAt) {
  // TODO: we can check isLoopEntryGuardedByCond before emitting the check
 
  Type *Ty = LHS->getType();
  assert(Ty == RHS->getType() && "expandCheck operands have different types?");

  if (SE->isLoopEntryGuardedByCond(L, Pred, LHS, RHS))
    return Builder.getTrue();

  Value *LHSV = Expander.expandCodeFor(LHS, Ty, InsertAt);
  Value *RHSV = Expander.expandCodeFor(RHS, Ty, InsertAt);
  return Builder.CreateICmp(Pred, LHSV, RHSV);
}

Optional<LoopPredication::LoopICmp>
LoopPredication::generateLoopLatchCheck(Type *RangeCheckType) {

  auto *LatchType = LatchCheck.IV->getType();
  if (RangeCheckType == LatchType)
    return LatchCheck;
  // For now, bail out if latch type is narrower than range type.
  if (DL->getTypeSizeInBits(LatchType) < DL->getTypeSizeInBits(RangeCheckType))
    return None;
  if (!isSafeToTruncateWideIVType(RangeCheckType))
    return None;
  // We can now safely identify the truncated version of the IV and limit for
  // RangeCheckType.
  LoopICmp NewLatchCheck;
  NewLatchCheck.Pred = LatchCheck.Pred;
  NewLatchCheck.IV = dyn_cast<SCEVAddRecExpr>(
      SE->getTruncateExpr(LatchCheck.IV, RangeCheckType));
  if (!NewLatchCheck.IV)
    return None;
  NewLatchCheck.Limit = SE->getTruncateExpr(LatchCheck.Limit, RangeCheckType);
  DEBUG(dbgs() << "IV of type: " << *LatchType
               << "can be represented as range check type:" << *RangeCheckType
               << "\n");
  DEBUG(dbgs() << "LatchCheck.IV: " << *NewLatchCheck.IV << "\n");
  DEBUG(dbgs() << "LatchCheck.Limit: " << *NewLatchCheck.Limit << "\n");
  return NewLatchCheck;
}

bool LoopPredication::isSupportedStep(const SCEV* Step) {
  return Step->isOne();
}

bool LoopPredication::CanExpand(const SCEV* S) {
  return SE->isLoopInvariant(S, L) && isSafeToExpand(S, *SE);
}

Optional<Value *> LoopPredication::widenICmpRangeCheckIncrementingLoop(
    LoopPredication::LoopICmp LatchCheck, LoopPredication::LoopICmp RangeCheck,
    SCEVExpander &Expander, IRBuilder<> &Builder) {
  auto *Ty = RangeCheck.IV->getType();
  // Generate the widened condition for the forward loop:
  //   guardStart u< guardLimit &&
  //   latchLimit <pred> guardLimit - 1 - guardStart + latchStart
  // where <pred> depends on the latch condition predicate. See the file
  // header comment for the reasoning.
  // guardLimit - guardStart + latchStart - 1
  const SCEV *GuardStart = RangeCheck.IV->getStart();
  const SCEV *GuardLimit = RangeCheck.Limit;
  const SCEV *LatchStart = LatchCheck.IV->getStart();
  const SCEV *LatchLimit = LatchCheck.Limit;

  // guardLimit - guardStart + latchStart - 1
  const SCEV *RHS =
      SE->getAddExpr(SE->getMinusSCEV(GuardLimit, GuardStart),
                     SE->getMinusSCEV(LatchStart, SE->getOne(Ty)));
  if (!CanExpand(GuardStart) || !CanExpand(GuardLimit) ||
      !CanExpand(LatchLimit) || !CanExpand(RHS)) {
    DEBUG(dbgs() << "Can't expand limit check!\n");
    return None;
  }
  ICmpInst::Predicate LimitCheckPred;
  switch (LatchCheck.Pred) {
  case ICmpInst::ICMP_ULT:
    LimitCheckPred = ICmpInst::ICMP_ULE;
    break;
  case ICmpInst::ICMP_ULE:
    LimitCheckPred = ICmpInst::ICMP_ULT;
    break;
  case ICmpInst::ICMP_SLT:
    LimitCheckPred = ICmpInst::ICMP_SLE;
    break;
  case ICmpInst::ICMP_SLE:
    LimitCheckPred = ICmpInst::ICMP_SLT;
    break;
  default:
    llvm_unreachable("Unsupported loop latch!");
  }

  DEBUG(dbgs() << "LHS: " << *LatchLimit << "\n");
  DEBUG(dbgs() << "RHS: " << *RHS << "\n");
  DEBUG(dbgs() << "Pred: " << LimitCheckPred << "\n");

  Instruction *InsertAt = Preheader->getTerminator();
  auto *LimitCheck =
      expandCheck(Expander, Builder, LimitCheckPred, LatchLimit, RHS, InsertAt);
  auto *FirstIterationCheck = expandCheck(Expander, Builder, RangeCheck.Pred,
                                          GuardStart, GuardLimit, InsertAt);
  return Builder.CreateAnd(FirstIterationCheck, LimitCheck);
}
/// If ICI can be widened to a loop invariant condition emits the loop
/// invariant condition in the loop preheader and return it, otherwise
/// returns None.
Optional<Value *> LoopPredication::widenICmpRangeCheck(ICmpInst *ICI,
                                                       SCEVExpander &Expander,
                                                       IRBuilder<> &Builder) {
  DEBUG(dbgs() << "Analyzing ICmpInst condition:\n");
  DEBUG(ICI->dump());

  // parseLoopStructure guarantees that the latch condition is:
  //   ++i <pred> latchLimit, where <pred> is u<, u<=, s<, or s<=.
  // We are looking for the range checks of the form:
  //   i u< guardLimit
  auto RangeCheck = parseLoopICmp(ICI);
  if (!RangeCheck) {
    DEBUG(dbgs() << "Failed to parse the loop latch condition!\n");
    return None;
  }
  DEBUG(dbgs() << "Guard check:\n");
  DEBUG(RangeCheck->dump());
  if (RangeCheck->Pred != ICmpInst::ICMP_ULT) {
    DEBUG(dbgs() << "Unsupported range check predicate(" << RangeCheck->Pred
                 << ")!\n");
    return None;
  }
  auto *RangeCheckIV = RangeCheck->IV;
  if (!RangeCheckIV->isAffine()) {
    DEBUG(dbgs() << "Range check IV is not affine!\n");
    return None;
  }
  auto *Step = RangeCheckIV->getStepRecurrence(*SE);
  // We cannot just compare with latch IV step because the latch and range IVs
  // may have different types.
  if (!isSupportedStep(Step)) {
    DEBUG(dbgs() << "Range check and latch have IVs different steps!\n");
    return None;
  }
  auto *Ty = RangeCheckIV->getType();
  auto CurrLatchCheckOpt = generateLoopLatchCheck(Ty);
  if (!CurrLatchCheckOpt) {
    DEBUG(dbgs() << "Failed to generate a loop latch check "
                    "corresponding to range type: "
                 << *Ty << "\n");
    return None;
  }

  LoopICmp CurrLatchCheck = *CurrLatchCheckOpt;
  // At this point the range check step and latch step should have the same
  // value and type.
  assert(Step == CurrLatchCheck.IV->getStepRecurrence(*SE) &&
         "Range and latch should have same step recurrence!");

  return widenICmpRangeCheckIncrementingLoop(CurrLatchCheck, *RangeCheck,
                                             Expander, Builder);
}

bool LoopPredication::widenGuardConditions(IntrinsicInst *Guard,
                                           SCEVExpander &Expander) {
  DEBUG(dbgs() << "Processing guard:\n");
  DEBUG(Guard->dump());

  IRBuilder<> Builder(cast<Instruction>(Preheader->getTerminator()));

  // The guard condition is expected to be in form of:
  //   cond1 && cond2 && cond3 ...
  // Iterate over subconditions looking for for icmp conditions which can be
  // widened across loop iterations. Widening these conditions remember the
  // resulting list of subconditions in Checks vector.
  SmallVector<Value *, 4> Worklist(1, Guard->getOperand(0));
  SmallPtrSet<Value *, 4> Visited;

  SmallVector<Value *, 4> Checks;

  unsigned NumWidened = 0;
  do {
    Value *Condition = Worklist.pop_back_val();
    if (!Visited.insert(Condition).second)
      continue;

    Value *LHS, *RHS;
    using namespace llvm::PatternMatch;
    if (match(Condition, m_And(m_Value(LHS), m_Value(RHS)))) {
      Worklist.push_back(LHS);
      Worklist.push_back(RHS);
      continue;
    }

    if (ICmpInst *ICI = dyn_cast<ICmpInst>(Condition)) {
      if (auto NewRangeCheck = widenICmpRangeCheck(ICI, Expander, Builder)) {
        Checks.push_back(NewRangeCheck.getValue());
        NumWidened++;
        continue;
      }
    }

    // Save the condition as is if we can't widen it
    Checks.push_back(Condition);
  } while (Worklist.size() != 0);

  if (NumWidened == 0)
    return false;

  // Emit the new guard condition
  Builder.SetInsertPoint(Guard);
  Value *LastCheck = nullptr;
  for (auto *Check : Checks)
    if (!LastCheck)
      LastCheck = Check;
    else
      LastCheck = Builder.CreateAnd(LastCheck, Check);
  Guard->setOperand(0, LastCheck);

  DEBUG(dbgs() << "Widened checks = " << NumWidened << "\n");
  return true;
}

Optional<LoopPredication::LoopICmp> LoopPredication::parseLoopLatchICmp() {
  using namespace PatternMatch;

  BasicBlock *LoopLatch = L->getLoopLatch();
  if (!LoopLatch) {
    DEBUG(dbgs() << "The loop doesn't have a single latch!\n");
    return None;
  }

  ICmpInst::Predicate Pred;
  Value *LHS, *RHS;
  BasicBlock *TrueDest, *FalseDest;

  if (!match(LoopLatch->getTerminator(),
             m_Br(m_ICmp(Pred, m_Value(LHS), m_Value(RHS)), TrueDest,
                  FalseDest))) {
    DEBUG(dbgs() << "Failed to match the latch terminator!\n");
    return None;
  }
  assert((TrueDest == L->getHeader() || FalseDest == L->getHeader()) &&
         "One of the latch's destinations must be the header");
  if (TrueDest != L->getHeader())
    Pred = ICmpInst::getInversePredicate(Pred);

  auto Result = parseLoopICmp(Pred, LHS, RHS);
  if (!Result) {
    DEBUG(dbgs() << "Failed to parse the loop latch condition!\n");
    return None;
  }

  // Check affine first, so if it's not we don't try to compute the step
  // recurrence.
  if (!Result->IV->isAffine()) {
    DEBUG(dbgs() << "The induction variable is not affine!\n");
    return None;
  }

  auto *Step = Result->IV->getStepRecurrence(*SE);
  if (!isSupportedStep(Step)) {
    DEBUG(dbgs() << "Unsupported loop stride(" << *Step << ")!\n");
    return None;
  }

  auto IsUnsupportedPredicate = [](const SCEV *Step, ICmpInst::Predicate Pred) {
    assert(Step->isOne() && "expected Step to be one!");
    return Pred != ICmpInst::ICMP_ULT && Pred != ICmpInst::ICMP_SLT &&
           Pred != ICmpInst::ICMP_ULE && Pred != ICmpInst::ICMP_SLE;
  };

  if (IsUnsupportedPredicate(Step, Result->Pred)) {
    DEBUG(dbgs() << "Unsupported loop latch predicate(" << Result->Pred
                 << ")!\n");
    return None;
  }
  return Result;
}

// Returns true if its safe to truncate the IV to RangeCheckType.
bool LoopPredication::isSafeToTruncateWideIVType(Type *RangeCheckType) {
  if (!EnableIVTruncation)
    return false;
  assert(DL->getTypeSizeInBits(LatchCheck.IV->getType()) >
             DL->getTypeSizeInBits(RangeCheckType) &&
         "Expected latch check IV type to be larger than range check operand "
         "type!");
  // The start and end values of the IV should be known. This is to guarantee
  // that truncating the wide type will not lose information.
  auto *Limit = dyn_cast<SCEVConstant>(LatchCheck.Limit);
  auto *Start = dyn_cast<SCEVConstant>(LatchCheck.IV->getStart());
  if (!Limit || !Start)
    return false;
  // This check makes sure that the IV does not change sign during loop
  // iterations. Consider latchType = i64, LatchStart = 5, Pred = ICMP_SGE,
  // LatchEnd = 2, rangeCheckType = i32. If it's not a monotonic predicate, the
  // IV wraps around, and the truncation of the IV would lose the range of
  // iterations between 2^32 and 2^64.
  bool Increasing;
  if (!SE->isMonotonicPredicate(LatchCheck.IV, LatchCheck.Pred, Increasing))
    return false;
  // The active bits should be less than the bits in the RangeCheckType. This
  // guarantees that truncating the latch check to RangeCheckType is a safe
  // operation.
  auto RangeCheckTypeBitSize = DL->getTypeSizeInBits(RangeCheckType);
  return Start->getAPInt().getActiveBits() < RangeCheckTypeBitSize &&
         Limit->getAPInt().getActiveBits() < RangeCheckTypeBitSize;
}

bool LoopPredication::runOnLoop(Loop *Loop) {
  L = Loop;

  DEBUG(dbgs() << "Analyzing ");
  DEBUG(L->dump());

  Module *M = L->getHeader()->getModule();

  // There is nothing to do if the module doesn't use guards
  auto *GuardDecl =
      M->getFunction(Intrinsic::getName(Intrinsic::experimental_guard));
  if (!GuardDecl || GuardDecl->use_empty())
    return false;

  DL = &M->getDataLayout();

  Preheader = L->getLoopPreheader();
  if (!Preheader)
    return false;

  auto LatchCheckOpt = parseLoopLatchICmp();
  if (!LatchCheckOpt)
    return false;
  LatchCheck = *LatchCheckOpt;

  DEBUG(dbgs() << "Latch check:\n");
  DEBUG(LatchCheck.dump());

  // Collect all the guards into a vector and process later, so as not
  // to invalidate the instruction iterator.
  SmallVector<IntrinsicInst *, 4> Guards;
  for (const auto BB : L->blocks())
    for (auto &I : *BB)
      if (auto *II = dyn_cast<IntrinsicInst>(&I))
        if (II->getIntrinsicID() == Intrinsic::experimental_guard)
          Guards.push_back(II);

  if (Guards.empty())
    return false;

  SCEVExpander Expander(*SE, *DL, "loop-predication");

  bool Changed = false;
  for (auto *Guard : Guards)
    Changed |= widenGuardConditions(Guard, Expander);

  return Changed;
}