Subversion Repositories QNX 8.QNX8 LLVM/Clang compiler suite

Rev

Details | Last modification | View Log | RSS feed

Rev Author Line No. Line
14 pmbaty 1
//===- llvm/Analysis/ValueTracking.h - Walk computations --------*- C++ -*-===//
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 contains routines that help analyze properties that chains of
10
// computations have.
11
//
12
//===----------------------------------------------------------------------===//
13
 
14
#ifndef LLVM_ANALYSIS_VALUETRACKING_H
15
#define LLVM_ANALYSIS_VALUETRACKING_H
16
 
17
#include "llvm/ADT/ArrayRef.h"
18
#include "llvm/ADT/SmallSet.h"
19
#include "llvm/IR/Constants.h"
20
#include "llvm/IR/DataLayout.h"
21
#include "llvm/IR/InstrTypes.h"
22
#include "llvm/IR/Intrinsics.h"
23
#include <cassert>
24
#include <cstdint>
25
 
26
namespace llvm {
27
 
28
class Operator;
29
class AddOperator;
30
class AllocaInst;
31
class APInt;
32
class AssumptionCache;
33
class DominatorTree;
34
class GEPOperator;
35
class LoadInst;
36
class WithOverflowInst;
37
struct KnownBits;
38
class Loop;
39
class LoopInfo;
40
class MDNode;
41
class OptimizationRemarkEmitter;
42
class StringRef;
43
class TargetLibraryInfo;
44
class Value;
45
 
46
constexpr unsigned MaxAnalysisRecursionDepth = 6;
47
 
48
/// Determine which bits of V are known to be either zero or one and return
49
/// them in the KnownZero/KnownOne bit sets.
50
///
51
/// This function is defined on values with integer type, values with pointer
52
/// type, and vectors of integers.  In the case
53
/// where V is a vector, the known zero and known one values are the
54
/// same width as the vector element, and the bit is set only if it is true
55
/// for all of the elements in the vector.
56
void computeKnownBits(const Value *V, KnownBits &Known, const DataLayout &DL,
57
                      unsigned Depth = 0, AssumptionCache *AC = nullptr,
58
                      const Instruction *CxtI = nullptr,
59
                      const DominatorTree *DT = nullptr,
60
                      OptimizationRemarkEmitter *ORE = nullptr,
61
                      bool UseInstrInfo = true);
62
 
63
/// Determine which bits of V are known to be either zero or one and return
64
/// them in the KnownZero/KnownOne bit sets.
65
///
66
/// This function is defined on values with integer type, values with pointer
67
/// type, and vectors of integers.  In the case
68
/// where V is a vector, the known zero and known one values are the
69
/// same width as the vector element, and the bit is set only if it is true
70
/// for all of the demanded elements in the vector.
71
void computeKnownBits(const Value *V, const APInt &DemandedElts,
72
                      KnownBits &Known, const DataLayout &DL,
73
                      unsigned Depth = 0, AssumptionCache *AC = nullptr,
74
                      const Instruction *CxtI = nullptr,
75
                      const DominatorTree *DT = nullptr,
76
                      OptimizationRemarkEmitter *ORE = nullptr,
77
                      bool UseInstrInfo = true);
78
 
79
/// Returns the known bits rather than passing by reference.
80
KnownBits computeKnownBits(const Value *V, const DataLayout &DL,
81
                           unsigned Depth = 0, AssumptionCache *AC = nullptr,
82
                           const Instruction *CxtI = nullptr,
83
                           const DominatorTree *DT = nullptr,
84
                           OptimizationRemarkEmitter *ORE = nullptr,
85
                           bool UseInstrInfo = true);
86
 
87
/// Returns the known bits rather than passing by reference.
88
KnownBits computeKnownBits(const Value *V, const APInt &DemandedElts,
89
                           const DataLayout &DL, unsigned Depth = 0,
90
                           AssumptionCache *AC = nullptr,
91
                           const Instruction *CxtI = nullptr,
92
                           const DominatorTree *DT = nullptr,
93
                           OptimizationRemarkEmitter *ORE = nullptr,
94
                           bool UseInstrInfo = true);
95
 
96
/// Compute known bits from the range metadata.
97
/// \p KnownZero the set of bits that are known to be zero
98
/// \p KnownOne the set of bits that are known to be one
99
void computeKnownBitsFromRangeMetadata(const MDNode &Ranges, KnownBits &Known);
100
 
101
/// Return true if LHS and RHS have no common bits set.
102
bool haveNoCommonBitsSet(const Value *LHS, const Value *RHS,
103
                         const DataLayout &DL, AssumptionCache *AC = nullptr,
104
                         const Instruction *CxtI = nullptr,
105
                         const DominatorTree *DT = nullptr,
106
                         bool UseInstrInfo = true);
107
 
108
/// Return true if the given value is known to have exactly one bit set when
109
/// defined. For vectors return true if every element is known to be a power
110
/// of two when defined. Supports values with integer or pointer type and
111
/// vectors of integers. If 'OrZero' is set, then return true if the given
112
/// value is either a power of two or zero.
113
bool isKnownToBeAPowerOfTwo(const Value *V, const DataLayout &DL,
114
                            bool OrZero = false, unsigned Depth = 0,
115
                            AssumptionCache *AC = nullptr,
116
                            const Instruction *CxtI = nullptr,
117
                            const DominatorTree *DT = nullptr,
118
                            bool UseInstrInfo = true);
119
 
120
bool isOnlyUsedInZeroEqualityComparison(const Instruction *CxtI);
121
 
122
/// Return true if the given value is known to be non-zero when defined. For
123
/// vectors, return true if every element is known to be non-zero when
124
/// defined. For pointers, if the context instruction and dominator tree are
125
/// specified, perform context-sensitive analysis and return true if the
126
/// pointer couldn't possibly be null at the specified instruction.
127
/// Supports values with integer or pointer type and vectors of integers.
128
bool isKnownNonZero(const Value *V, const DataLayout &DL, unsigned Depth = 0,
129
                    AssumptionCache *AC = nullptr,
130
                    const Instruction *CxtI = nullptr,
131
                    const DominatorTree *DT = nullptr,
132
                    bool UseInstrInfo = true);
133
 
134
/// Return true if the two given values are negation.
135
/// Currently can recoginze Value pair:
136
/// 1: <X, Y> if X = sub (0, Y) or Y = sub (0, X)
137
/// 2: <X, Y> if X = sub (A, B) and Y = sub (B, A)
138
bool isKnownNegation(const Value *X, const Value *Y, bool NeedNSW = false);
139
 
140
/// Returns true if the give value is known to be non-negative.
141
bool isKnownNonNegative(const Value *V, const DataLayout &DL,
142
                        unsigned Depth = 0, AssumptionCache *AC = nullptr,
143
                        const Instruction *CxtI = nullptr,
144
                        const DominatorTree *DT = nullptr,
145
                        bool UseInstrInfo = true);
146
 
147
/// Returns true if the given value is known be positive (i.e. non-negative
148
/// and non-zero).
149
bool isKnownPositive(const Value *V, const DataLayout &DL, unsigned Depth = 0,
150
                     AssumptionCache *AC = nullptr,
151
                     const Instruction *CxtI = nullptr,
152
                     const DominatorTree *DT = nullptr,
153
                     bool UseInstrInfo = true);
154
 
155
/// Returns true if the given value is known be negative (i.e. non-positive
156
/// and non-zero).
157
bool isKnownNegative(const Value *V, const DataLayout &DL, unsigned Depth = 0,
158
                     AssumptionCache *AC = nullptr,
159
                     const Instruction *CxtI = nullptr,
160
                     const DominatorTree *DT = nullptr,
161
                     bool UseInstrInfo = true);
162
 
163
/// Return true if the given values are known to be non-equal when defined.
164
/// Supports scalar integer types only.
165
bool isKnownNonEqual(const Value *V1, const Value *V2, const DataLayout &DL,
166
                     AssumptionCache *AC = nullptr,
167
                     const Instruction *CxtI = nullptr,
168
                     const DominatorTree *DT = nullptr,
169
                     bool UseInstrInfo = true);
170
 
171
/// Return true if 'V & Mask' is known to be zero. We use this predicate to
172
/// simplify operations downstream. Mask is known to be zero for bits that V
173
/// cannot have.
174
///
175
/// This function is defined on values with integer type, values with pointer
176
/// type, and vectors of integers.  In the case
177
/// where V is a vector, the mask, known zero, and known one values are the
178
/// same width as the vector element, and the bit is set only if it is true
179
/// for all of the elements in the vector.
180
bool MaskedValueIsZero(const Value *V, const APInt &Mask, const DataLayout &DL,
181
                       unsigned Depth = 0, AssumptionCache *AC = nullptr,
182
                       const Instruction *CxtI = nullptr,
183
                       const DominatorTree *DT = nullptr,
184
                       bool UseInstrInfo = true);
185
 
186
/// Return the number of times the sign bit of the register is replicated into
187
/// the other bits. We know that at least 1 bit is always equal to the sign
188
/// bit (itself), but other cases can give us information. For example,
189
/// immediately after an "ashr X, 2", we know that the top 3 bits are all
190
/// equal to each other, so we return 3. For vectors, return the number of
191
/// sign bits for the vector element with the mininum number of known sign
192
/// bits.
193
unsigned ComputeNumSignBits(const Value *Op, const DataLayout &DL,
194
                            unsigned Depth = 0, AssumptionCache *AC = nullptr,
195
                            const Instruction *CxtI = nullptr,
196
                            const DominatorTree *DT = nullptr,
197
                            bool UseInstrInfo = true);
198
 
199
/// Get the upper bound on bit size for this Value \p Op as a signed integer.
200
/// i.e.  x == sext(trunc(x to MaxSignificantBits) to bitwidth(x)).
201
/// Similar to the APInt::getSignificantBits function.
202
unsigned ComputeMaxSignificantBits(const Value *Op, const DataLayout &DL,
203
                                   unsigned Depth = 0,
204
                                   AssumptionCache *AC = nullptr,
205
                                   const Instruction *CxtI = nullptr,
206
                                   const DominatorTree *DT = nullptr);
207
 
208
/// Map a call instruction to an intrinsic ID.  Libcalls which have equivalent
209
/// intrinsics are treated as-if they were intrinsics.
210
Intrinsic::ID getIntrinsicForCallSite(const CallBase &CB,
211
                                      const TargetLibraryInfo *TLI);
212
 
213
/// Return true if we can prove that the specified FP value is never equal to
214
/// -0.0.
215
bool CannotBeNegativeZero(const Value *V, const TargetLibraryInfo *TLI,
216
                          unsigned Depth = 0);
217
 
218
/// Return true if we can prove that the specified FP value is either NaN or
219
/// never less than -0.0.
220
///
221
///      NaN --> true
222
///       +0 --> true
223
///       -0 --> true
224
///   x > +0 --> true
225
///   x < -0 --> false
226
bool CannotBeOrderedLessThanZero(const Value *V, const TargetLibraryInfo *TLI);
227
 
228
/// Return true if the floating-point scalar value is not an infinity or if
229
/// the floating-point vector value has no infinities. Return false if a value
230
/// could ever be infinity.
231
bool isKnownNeverInfinity(const Value *V, const TargetLibraryInfo *TLI,
232
                          unsigned Depth = 0);
233
 
234
/// Return true if the floating-point scalar value is not a NaN or if the
235
/// floating-point vector value has no NaN elements. Return false if a value
236
/// could ever be NaN.
237
bool isKnownNeverNaN(const Value *V, const TargetLibraryInfo *TLI,
238
                     unsigned Depth = 0);
239
 
240
/// Return true if we can prove that the specified FP value's sign bit is 0.
241
///
242
///      NaN --> true/false (depending on the NaN's sign bit)
243
///       +0 --> true
244
///       -0 --> false
245
///   x > +0 --> true
246
///   x < -0 --> false
247
bool SignBitMustBeZero(const Value *V, const TargetLibraryInfo *TLI);
248
 
249
/// If the specified value can be set by repeating the same byte in memory,
250
/// return the i8 value that it is represented with. This is true for all i8
251
/// values obviously, but is also true for i32 0, i32 -1, i16 0xF0F0, double
252
/// 0.0 etc. If the value can't be handled with a repeated byte store (e.g.
253
/// i16 0x1234), return null. If the value is entirely undef and padding,
254
/// return undef.
255
Value *isBytewiseValue(Value *V, const DataLayout &DL);
256
 
257
/// Given an aggregate and an sequence of indices, see if the scalar value
258
/// indexed is already around as a register, for example if it were inserted
259
/// directly into the aggregate.
260
///
261
/// If InsertBefore is not null, this function will duplicate (modified)
262
/// insertvalues when a part of a nested struct is extracted.
263
Value *FindInsertedValue(Value *V, ArrayRef<unsigned> idx_range,
264
                         Instruction *InsertBefore = nullptr);
265
 
266
/// Analyze the specified pointer to see if it can be expressed as a base
267
/// pointer plus a constant offset. Return the base and offset to the caller.
268
///
269
/// This is a wrapper around Value::stripAndAccumulateConstantOffsets that
270
/// creates and later unpacks the required APInt.
271
inline Value *GetPointerBaseWithConstantOffset(Value *Ptr, int64_t &Offset,
272
                                               const DataLayout &DL,
273
                                               bool AllowNonInbounds = true) {
274
  APInt OffsetAPInt(DL.getIndexTypeSizeInBits(Ptr->getType()), 0);
275
  Value *Base =
276
      Ptr->stripAndAccumulateConstantOffsets(DL, OffsetAPInt, AllowNonInbounds);
277
 
278
  Offset = OffsetAPInt.getSExtValue();
279
  return Base;
280
}
281
inline const Value *
282
GetPointerBaseWithConstantOffset(const Value *Ptr, int64_t &Offset,
283
                                 const DataLayout &DL,
284
                                 bool AllowNonInbounds = true) {
285
  return GetPointerBaseWithConstantOffset(const_cast<Value *>(Ptr), Offset, DL,
286
                                          AllowNonInbounds);
287
}
288
 
289
/// Returns true if the GEP is based on a pointer to a string (array of
290
// \p CharSize integers) and is indexing into this string.
291
bool isGEPBasedOnPointerToString(const GEPOperator *GEP, unsigned CharSize = 8);
292
 
293
/// Represents offset+length into a ConstantDataArray.
294
struct ConstantDataArraySlice {
295
  /// ConstantDataArray pointer. nullptr indicates a zeroinitializer (a valid
296
  /// initializer, it just doesn't fit the ConstantDataArray interface).
297
  const ConstantDataArray *Array;
298
 
299
  /// Slice starts at this Offset.
300
  uint64_t Offset;
301
 
302
  /// Length of the slice.
303
  uint64_t Length;
304
 
305
  /// Moves the Offset and adjusts Length accordingly.
306
  void move(uint64_t Delta) {
307
    assert(Delta < Length);
308
    Offset += Delta;
309
    Length -= Delta;
310
  }
311
 
312
  /// Convenience accessor for elements in the slice.
313
  uint64_t operator[](unsigned I) const {
314
    return Array == nullptr ? 0 : Array->getElementAsInteger(I + Offset);
315
  }
316
};
317
 
318
/// Returns true if the value \p V is a pointer into a ConstantDataArray.
319
/// If successful \p Slice will point to a ConstantDataArray info object
320
/// with an appropriate offset.
321
bool getConstantDataArrayInfo(const Value *V, ConstantDataArraySlice &Slice,
322
                              unsigned ElementSize, uint64_t Offset = 0);
323
 
324
/// This function computes the length of a null-terminated C string pointed to
325
/// by V. If successful, it returns true and returns the string in Str. If
326
/// unsuccessful, it returns false. This does not include the trailing null
327
/// character by default. If TrimAtNul is set to false, then this returns any
328
/// trailing null characters as well as any other characters that come after
329
/// it.
330
bool getConstantStringInfo(const Value *V, StringRef &Str,
331
                           bool TrimAtNul = true);
332
 
333
/// If we can compute the length of the string pointed to by the specified
334
/// pointer, return 'len+1'.  If we can't, return 0.
335
uint64_t GetStringLength(const Value *V, unsigned CharSize = 8);
336
 
337
/// This function returns call pointer argument that is considered the same by
338
/// aliasing rules. You CAN'T use it to replace one value with another. If
339
/// \p MustPreserveNullness is true, the call must preserve the nullness of
340
/// the pointer.
341
const Value *getArgumentAliasingToReturnedPointer(const CallBase *Call,
342
                                                  bool MustPreserveNullness);
343
inline Value *getArgumentAliasingToReturnedPointer(CallBase *Call,
344
                                                   bool MustPreserveNullness) {
345
  return const_cast<Value *>(getArgumentAliasingToReturnedPointer(
346
      const_cast<const CallBase *>(Call), MustPreserveNullness));
347
}
348
 
349
/// {launder,strip}.invariant.group returns pointer that aliases its argument,
350
/// and it only captures pointer by returning it.
351
/// These intrinsics are not marked as nocapture, because returning is
352
/// considered as capture. The arguments are not marked as returned neither,
353
/// because it would make it useless. If \p MustPreserveNullness is true,
354
/// the intrinsic must preserve the nullness of the pointer.
355
bool isIntrinsicReturningPointerAliasingArgumentWithoutCapturing(
356
    const CallBase *Call, bool MustPreserveNullness);
357
 
358
/// This method strips off any GEP address adjustments and pointer casts from
359
/// the specified value, returning the original object being addressed. Note
360
/// that the returned value has pointer type if the specified value does. If
361
/// the MaxLookup value is non-zero, it limits the number of instructions to
362
/// be stripped off.
363
const Value *getUnderlyingObject(const Value *V, unsigned MaxLookup = 6);
364
inline Value *getUnderlyingObject(Value *V, unsigned MaxLookup = 6) {
365
  // Force const to avoid infinite recursion.
366
  const Value *VConst = V;
367
  return const_cast<Value *>(getUnderlyingObject(VConst, MaxLookup));
368
}
369
 
370
/// This method is similar to getUnderlyingObject except that it can
371
/// look through phi and select instructions and return multiple objects.
372
///
373
/// If LoopInfo is passed, loop phis are further analyzed.  If a pointer
374
/// accesses different objects in each iteration, we don't look through the
375
/// phi node. E.g. consider this loop nest:
376
///
377
///   int **A;
378
///   for (i)
379
///     for (j) {
380
///        A[i][j] = A[i-1][j] * B[j]
381
///     }
382
///
383
/// This is transformed by Load-PRE to stash away A[i] for the next iteration
384
/// of the outer loop:
385
///
386
///   Curr = A[0];          // Prev_0
387
///   for (i: 1..N) {
388
///     Prev = Curr;        // Prev = PHI (Prev_0, Curr)
389
///     Curr = A[i];
390
///     for (j: 0..N) {
391
///        Curr[j] = Prev[j] * B[j]
392
///     }
393
///   }
394
///
395
/// Since A[i] and A[i-1] are independent pointers, getUnderlyingObjects
396
/// should not assume that Curr and Prev share the same underlying object thus
397
/// it shouldn't look through the phi above.
398
void getUnderlyingObjects(const Value *V,
399
                          SmallVectorImpl<const Value *> &Objects,
400
                          LoopInfo *LI = nullptr, unsigned MaxLookup = 6);
401
 
402
/// This is a wrapper around getUnderlyingObjects and adds support for basic
403
/// ptrtoint+arithmetic+inttoptr sequences.
404
bool getUnderlyingObjectsForCodeGen(const Value *V,
405
                                    SmallVectorImpl<Value *> &Objects);
406
 
407
/// Returns unique alloca where the value comes from, or nullptr.
408
/// If OffsetZero is true check that V points to the begining of the alloca.
409
AllocaInst *findAllocaForValue(Value *V, bool OffsetZero = false);
410
inline const AllocaInst *findAllocaForValue(const Value *V,
411
                                            bool OffsetZero = false) {
412
  return findAllocaForValue(const_cast<Value *>(V), OffsetZero);
413
}
414
 
415
/// Return true if the only users of this pointer are lifetime markers.
416
bool onlyUsedByLifetimeMarkers(const Value *V);
417
 
418
/// Return true if the only users of this pointer are lifetime markers or
419
/// droppable instructions.
420
bool onlyUsedByLifetimeMarkersOrDroppableInsts(const Value *V);
421
 
422
/// Return true if speculation of the given load must be suppressed to avoid
423
/// ordering or interfering with an active sanitizer.  If not suppressed,
424
/// dereferenceability and alignment must be proven separately.  Note: This
425
/// is only needed for raw reasoning; if you use the interface below
426
/// (isSafeToSpeculativelyExecute), this is handled internally.
427
bool mustSuppressSpeculation(const LoadInst &LI);
428
 
429
/// Return true if the instruction does not have any effects besides
430
/// calculating the result and does not have undefined behavior.
431
///
432
/// This method never returns true for an instruction that returns true for
433
/// mayHaveSideEffects; however, this method also does some other checks in
434
/// addition. It checks for undefined behavior, like dividing by zero or
435
/// loading from an invalid pointer (but not for undefined results, like a
436
/// shift with a shift amount larger than the width of the result). It checks
437
/// for malloc and alloca because speculatively executing them might cause a
438
/// memory leak. It also returns false for instructions related to control
439
/// flow, specifically terminators and PHI nodes.
440
///
441
/// If the CtxI is specified this method performs context-sensitive analysis
442
/// and returns true if it is safe to execute the instruction immediately
443
/// before the CtxI.
444
///
445
/// If the CtxI is NOT specified this method only looks at the instruction
446
/// itself and its operands, so if this method returns true, it is safe to
447
/// move the instruction as long as the correct dominance relationships for
448
/// the operands and users hold.
449
///
450
/// This method can return true for instructions that read memory;
451
/// for such instructions, moving them may change the resulting value.
452
bool isSafeToSpeculativelyExecute(const Instruction *I,
453
                                  const Instruction *CtxI = nullptr,
454
                                  AssumptionCache *AC = nullptr,
455
                                  const DominatorTree *DT = nullptr,
456
                                  const TargetLibraryInfo *TLI = nullptr);
457
 
458
/// This returns the same result as isSafeToSpeculativelyExecute if Opcode is
459
/// the actual opcode of Inst. If the provided and actual opcode differ, the
460
/// function (virtually) overrides the opcode of Inst with the provided
461
/// Opcode. There are come constraints in this case:
462
/// * If Opcode has a fixed number of operands (eg, as binary operators do),
463
///   then Inst has to have at least as many leading operands. The function
464
///   will ignore all trailing operands beyond that number.
465
/// * If Opcode allows for an arbitrary number of operands (eg, as CallInsts
466
///   do), then all operands are considered.
467
/// * The virtual instruction has to satisfy all typing rules of the provided
468
///   Opcode.
469
/// * This function is pessimistic in the following sense: If one actually
470
///   materialized the virtual instruction, then isSafeToSpeculativelyExecute
471
///   may say that the materialized instruction is speculatable whereas this
472
///   function may have said that the instruction wouldn't be speculatable.
473
///   This behavior is a shortcoming in the current implementation and not
474
///   intentional.
475
bool isSafeToSpeculativelyExecuteWithOpcode(
476
    unsigned Opcode, const Instruction *Inst, const Instruction *CtxI = nullptr,
477
    AssumptionCache *AC = nullptr, const DominatorTree *DT = nullptr,
478
    const TargetLibraryInfo *TLI = nullptr);
479
 
480
/// Returns true if the result or effects of the given instructions \p I
481
/// depend values not reachable through the def use graph.
482
/// * Memory dependence arises for example if the instruction reads from
483
///   memory or may produce effects or undefined behaviour. Memory dependent
484
///   instructions generally cannot be reorderd with respect to other memory
485
///   dependent instructions.
486
/// * Control dependence arises for example if the instruction may fault
487
///   if lifted above a throwing call or infinite loop.
488
bool mayHaveNonDefUseDependency(const Instruction &I);
489
 
490
/// Return true if it is an intrinsic that cannot be speculated but also
491
/// cannot trap.
492
bool isAssumeLikeIntrinsic(const Instruction *I);
493
 
494
/// Return true if it is valid to use the assumptions provided by an
495
/// assume intrinsic, I, at the point in the control-flow identified by the
496
/// context instruction, CxtI.
497
bool isValidAssumeForContext(const Instruction *I, const Instruction *CxtI,
498
                             const DominatorTree *DT = nullptr);
499
 
500
enum class OverflowResult {
501
  /// Always overflows in the direction of signed/unsigned min value.
502
  AlwaysOverflowsLow,
503
  /// Always overflows in the direction of signed/unsigned max value.
504
  AlwaysOverflowsHigh,
505
  /// May or may not overflow.
506
  MayOverflow,
507
  /// Never overflows.
508
  NeverOverflows,
509
};
510
 
511
OverflowResult computeOverflowForUnsignedMul(const Value *LHS, const Value *RHS,
512
                                             const DataLayout &DL,
513
                                             AssumptionCache *AC,
514
                                             const Instruction *CxtI,
515
                                             const DominatorTree *DT,
516
                                             bool UseInstrInfo = true);
517
OverflowResult computeOverflowForSignedMul(const Value *LHS, const Value *RHS,
518
                                           const DataLayout &DL,
519
                                           AssumptionCache *AC,
520
                                           const Instruction *CxtI,
521
                                           const DominatorTree *DT,
522
                                           bool UseInstrInfo = true);
523
OverflowResult computeOverflowForUnsignedAdd(const Value *LHS, const Value *RHS,
524
                                             const DataLayout &DL,
525
                                             AssumptionCache *AC,
526
                                             const Instruction *CxtI,
527
                                             const DominatorTree *DT,
528
                                             bool UseInstrInfo = true);
529
OverflowResult computeOverflowForSignedAdd(const Value *LHS, const Value *RHS,
530
                                           const DataLayout &DL,
531
                                           AssumptionCache *AC = nullptr,
532
                                           const Instruction *CxtI = nullptr,
533
                                           const DominatorTree *DT = nullptr);
534
/// This version also leverages the sign bit of Add if known.
535
OverflowResult computeOverflowForSignedAdd(const AddOperator *Add,
536
                                           const DataLayout &DL,
537
                                           AssumptionCache *AC = nullptr,
538
                                           const Instruction *CxtI = nullptr,
539
                                           const DominatorTree *DT = nullptr);
540
OverflowResult computeOverflowForUnsignedSub(const Value *LHS, const Value *RHS,
541
                                             const DataLayout &DL,
542
                                             AssumptionCache *AC,
543
                                             const Instruction *CxtI,
544
                                             const DominatorTree *DT);
545
OverflowResult computeOverflowForSignedSub(const Value *LHS, const Value *RHS,
546
                                           const DataLayout &DL,
547
                                           AssumptionCache *AC,
548
                                           const Instruction *CxtI,
549
                                           const DominatorTree *DT);
550
 
551
/// Returns true if the arithmetic part of the \p WO 's result is
552
/// used only along the paths control dependent on the computation
553
/// not overflowing, \p WO being an <op>.with.overflow intrinsic.
554
bool isOverflowIntrinsicNoWrap(const WithOverflowInst *WO,
555
                               const DominatorTree &DT);
556
 
557
/// Determine the possible constant range of an integer or vector of integer
558
/// value. This is intended as a cheap, non-recursive check.
559
ConstantRange computeConstantRange(const Value *V, bool ForSigned,
560
                                   bool UseInstrInfo = true,
561
                                   AssumptionCache *AC = nullptr,
562
                                   const Instruction *CtxI = nullptr,
563
                                   const DominatorTree *DT = nullptr,
564
                                   unsigned Depth = 0);
565
 
566
/// Return true if this function can prove that the instruction I will
567
/// always transfer execution to one of its successors (including the next
568
/// instruction that follows within a basic block). E.g. this is not
569
/// guaranteed for function calls that could loop infinitely.
570
///
571
/// In other words, this function returns false for instructions that may
572
/// transfer execution or fail to transfer execution in a way that is not
573
/// captured in the CFG nor in the sequence of instructions within a basic
574
/// block.
575
///
576
/// Undefined behavior is assumed not to happen, so e.g. division is
577
/// guaranteed to transfer execution to the following instruction even
578
/// though division by zero might cause undefined behavior.
579
bool isGuaranteedToTransferExecutionToSuccessor(const Instruction *I);
580
 
581
/// Returns true if this block does not contain a potential implicit exit.
582
/// This is equivelent to saying that all instructions within the basic block
583
/// are guaranteed to transfer execution to their successor within the basic
584
/// block. This has the same assumptions w.r.t. undefined behavior as the
585
/// instruction variant of this function.
586
bool isGuaranteedToTransferExecutionToSuccessor(const BasicBlock *BB);
587
 
588
/// Return true if every instruction in the range (Begin, End) is
589
/// guaranteed to transfer execution to its static successor. \p ScanLimit
590
/// bounds the search to avoid scanning huge blocks.
591
bool isGuaranteedToTransferExecutionToSuccessor(
592
    BasicBlock::const_iterator Begin, BasicBlock::const_iterator End,
593
    unsigned ScanLimit = 32);
594
 
595
/// Same as previous, but with range expressed via iterator_range.
596
bool isGuaranteedToTransferExecutionToSuccessor(
597
    iterator_range<BasicBlock::const_iterator> Range, unsigned ScanLimit = 32);
598
 
599
/// Return true if this function can prove that the instruction I
600
/// is executed for every iteration of the loop L.
601
///
602
/// Note that this currently only considers the loop header.
603
bool isGuaranteedToExecuteForEveryIteration(const Instruction *I,
604
                                            const Loop *L);
605
 
606
/// Return true if \p PoisonOp's user yields poison or raises UB if its
607
/// operand \p PoisonOp is poison.
608
///
609
/// If \p PoisonOp is a vector or an aggregate and the operation's result is a
610
/// single value, any poison element in /p PoisonOp should make the result
611
/// poison or raise UB.
612
///
613
/// To filter out operands that raise UB on poison, you can use
614
/// getGuaranteedNonPoisonOp.
615
bool propagatesPoison(const Use &PoisonOp);
616
 
617
/// Insert operands of I into Ops such that I will trigger undefined behavior
618
/// if I is executed and that operand has a poison value.
619
void getGuaranteedNonPoisonOps(const Instruction *I,
620
                               SmallVectorImpl<const Value *> &Ops);
621
 
622
/// Insert operands of I into Ops such that I will trigger undefined behavior
623
/// if I is executed and that operand is not a well-defined value
624
/// (i.e. has undef bits or poison).
625
void getGuaranteedWellDefinedOps(const Instruction *I,
626
                                 SmallVectorImpl<const Value *> &Ops);
627
 
628
/// Return true if the given instruction must trigger undefined behavior
629
/// when I is executed with any operands which appear in KnownPoison holding
630
/// a poison value at the point of execution.
631
bool mustTriggerUB(const Instruction *I,
632
                   const SmallSet<const Value *, 16> &KnownPoison);
633
 
634
/// Return true if this function can prove that if Inst is executed
635
/// and yields a poison value or undef bits, then that will trigger
636
/// undefined behavior.
637
///
638
/// Note that this currently only considers the basic block that is
639
/// the parent of Inst.
640
bool programUndefinedIfUndefOrPoison(const Instruction *Inst);
641
bool programUndefinedIfPoison(const Instruction *Inst);
642
 
643
/// canCreateUndefOrPoison returns true if Op can create undef or poison from
644
/// non-undef & non-poison operands.
645
/// For vectors, canCreateUndefOrPoison returns true if there is potential
646
/// poison or undef in any element of the result when vectors without
647
/// undef/poison poison are given as operands.
648
/// For example, given `Op = shl <2 x i32> %x, <0, 32>`, this function returns
649
/// true. If Op raises immediate UB but never creates poison or undef
650
/// (e.g. sdiv I, 0), canCreatePoison returns false.
651
///
652
/// \p ConsiderFlagsAndMetadata controls whether poison producing flags and
653
/// metadata on the instruction are considered.  This can be used to see if the
654
/// instruction could still introduce undef or poison even without poison
655
/// generating flags and metadata which might be on the instruction.
656
/// (i.e. could the result of Op->dropPoisonGeneratingFlags() still create
657
/// poison or undef)
658
///
659
/// canCreatePoison returns true if Op can create poison from non-poison
660
/// operands.
661
bool canCreateUndefOrPoison(const Operator *Op,
662
                            bool ConsiderFlagsAndMetadata = true);
663
bool canCreatePoison(const Operator *Op, bool ConsiderFlagsAndMetadata = true);
664
 
665
/// Return true if V is poison given that ValAssumedPoison is already poison.
666
/// For example, if ValAssumedPoison is `icmp X, 10` and V is `icmp X, 5`,
667
/// impliesPoison returns true.
668
bool impliesPoison(const Value *ValAssumedPoison, const Value *V);
669
 
670
/// Return true if this function can prove that V does not have undef bits
671
/// and is never poison. If V is an aggregate value or vector, check whether
672
/// all elements (except padding) are not undef or poison.
673
/// Note that this is different from canCreateUndefOrPoison because the
674
/// function assumes Op's operands are not poison/undef.
675
///
676
/// If CtxI and DT are specified this method performs flow-sensitive analysis
677
/// and returns true if it is guaranteed to be never undef or poison
678
/// immediately before the CtxI.
679
bool isGuaranteedNotToBeUndefOrPoison(const Value *V,
680
                                      AssumptionCache *AC = nullptr,
681
                                      const Instruction *CtxI = nullptr,
682
                                      const DominatorTree *DT = nullptr,
683
                                      unsigned Depth = 0);
684
bool isGuaranteedNotToBePoison(const Value *V, AssumptionCache *AC = nullptr,
685
                               const Instruction *CtxI = nullptr,
686
                               const DominatorTree *DT = nullptr,
687
                               unsigned Depth = 0);
688
 
689
/// Specific patterns of select instructions we can match.
690
enum SelectPatternFlavor {
691
  SPF_UNKNOWN = 0,
692
  SPF_SMIN,    /// Signed minimum
693
  SPF_UMIN,    /// Unsigned minimum
694
  SPF_SMAX,    /// Signed maximum
695
  SPF_UMAX,    /// Unsigned maximum
696
  SPF_FMINNUM, /// Floating point minnum
697
  SPF_FMAXNUM, /// Floating point maxnum
698
  SPF_ABS,     /// Absolute value
699
  SPF_NABS     /// Negated absolute value
700
};
701
 
702
/// Behavior when a floating point min/max is given one NaN and one
703
/// non-NaN as input.
704
enum SelectPatternNaNBehavior {
705
  SPNB_NA = 0,        /// NaN behavior not applicable.
706
  SPNB_RETURNS_NAN,   /// Given one NaN input, returns the NaN.
707
  SPNB_RETURNS_OTHER, /// Given one NaN input, returns the non-NaN.
708
  SPNB_RETURNS_ANY    /// Given one NaN input, can return either (or
709
                      /// it has been determined that no operands can
710
                      /// be NaN).
711
};
712
 
713
struct SelectPatternResult {
714
  SelectPatternFlavor Flavor;
715
  SelectPatternNaNBehavior NaNBehavior; /// Only applicable if Flavor is
716
                                        /// SPF_FMINNUM or SPF_FMAXNUM.
717
  bool Ordered; /// When implementing this min/max pattern as
718
                /// fcmp; select, does the fcmp have to be
719
                /// ordered?
720
 
721
  /// Return true if \p SPF is a min or a max pattern.
722
  static bool isMinOrMax(SelectPatternFlavor SPF) {
723
    return SPF != SPF_UNKNOWN && SPF != SPF_ABS && SPF != SPF_NABS;
724
  }
725
};
726
 
727
/// Pattern match integer [SU]MIN, [SU]MAX and ABS idioms, returning the kind
728
/// and providing the out parameter results if we successfully match.
729
///
730
/// For ABS/NABS, LHS will be set to the input to the abs idiom. RHS will be
731
/// the negation instruction from the idiom.
732
///
733
/// If CastOp is not nullptr, also match MIN/MAX idioms where the type does
734
/// not match that of the original select. If this is the case, the cast
735
/// operation (one of Trunc,SExt,Zext) that must be done to transform the
736
/// type of LHS and RHS into the type of V is returned in CastOp.
737
///
738
/// For example:
739
///   %1 = icmp slt i32 %a, i32 4
740
///   %2 = sext i32 %a to i64
741
///   %3 = select i1 %1, i64 %2, i64 4
742
///
743
/// -> LHS = %a, RHS = i32 4, *CastOp = Instruction::SExt
744
///
745
SelectPatternResult matchSelectPattern(Value *V, Value *&LHS, Value *&RHS,
746
                                       Instruction::CastOps *CastOp = nullptr,
747
                                       unsigned Depth = 0);
748
 
749
inline SelectPatternResult matchSelectPattern(const Value *V, const Value *&LHS,
750
                                              const Value *&RHS) {
751
  Value *L = const_cast<Value *>(LHS);
752
  Value *R = const_cast<Value *>(RHS);
753
  auto Result = matchSelectPattern(const_cast<Value *>(V), L, R);
754
  LHS = L;
755
  RHS = R;
756
  return Result;
757
}
758
 
759
/// Determine the pattern that a select with the given compare as its
760
/// predicate and given values as its true/false operands would match.
761
SelectPatternResult matchDecomposedSelectPattern(
762
    CmpInst *CmpI, Value *TrueVal, Value *FalseVal, Value *&LHS, Value *&RHS,
763
    Instruction::CastOps *CastOp = nullptr, unsigned Depth = 0);
764
 
765
/// Return the canonical comparison predicate for the specified
766
/// minimum/maximum flavor.
767
CmpInst::Predicate getMinMaxPred(SelectPatternFlavor SPF, bool Ordered = false);
768
 
769
/// Return the inverse minimum/maximum flavor of the specified flavor.
770
/// For example, signed minimum is the inverse of signed maximum.
771
SelectPatternFlavor getInverseMinMaxFlavor(SelectPatternFlavor SPF);
772
 
773
Intrinsic::ID getInverseMinMaxIntrinsic(Intrinsic::ID MinMaxID);
774
 
775
/// Return the minimum or maximum constant value for the specified integer
776
/// min/max flavor and type.
777
APInt getMinMaxLimit(SelectPatternFlavor SPF, unsigned BitWidth);
778
 
779
/// Check if the values in \p VL are select instructions that can be converted
780
/// to a min or max (vector) intrinsic. Returns the intrinsic ID, if such a
781
/// conversion is possible, together with a bool indicating whether all select
782
/// conditions are only used by the selects. Otherwise return
783
/// Intrinsic::not_intrinsic.
784
std::pair<Intrinsic::ID, bool>
785
canConvertToMinOrMaxIntrinsic(ArrayRef<Value *> VL);
786
 
787
/// Attempt to match a simple first order recurrence cycle of the form:
788
///   %iv = phi Ty [%Start, %Entry], [%Inc, %backedge]
789
///   %inc = binop %iv, %step
790
/// OR
791
///   %iv = phi Ty [%Start, %Entry], [%Inc, %backedge]
792
///   %inc = binop %step, %iv
793
///
794
/// A first order recurrence is a formula with the form: X_n = f(X_(n-1))
795
///
796
/// A couple of notes on subtleties in that definition:
797
/// * The Step does not have to be loop invariant.  In math terms, it can
798
///   be a free variable.  We allow recurrences with both constant and
799
///   variable coefficients. Callers may wish to filter cases where Step
800
///   does not dominate P.
801
/// * For non-commutative operators, we will match both forms.  This
802
///   results in some odd recurrence structures.  Callers may wish to filter
803
///   out recurrences where the phi is not the LHS of the returned operator.
804
/// * Because of the structure matched, the caller can assume as a post
805
///   condition of the match the presence of a Loop with P's parent as it's
806
///   header *except* in unreachable code.  (Dominance decays in unreachable
807
///   code.)
808
///
809
/// NOTE: This is intentional simple.  If you want the ability to analyze
810
/// non-trivial loop conditons, see ScalarEvolution instead.
811
bool matchSimpleRecurrence(const PHINode *P, BinaryOperator *&BO, Value *&Start,
812
                           Value *&Step);
813
 
814
/// Analogous to the above, but starting from the binary operator
815
bool matchSimpleRecurrence(const BinaryOperator *I, PHINode *&P, Value *&Start,
816
                           Value *&Step);
817
 
818
/// Return true if RHS is known to be implied true by LHS.  Return false if
819
/// RHS is known to be implied false by LHS.  Otherwise, return std::nullopt if
820
/// no implication can be made. A & B must be i1 (boolean) values or a vector of
821
/// such values. Note that the truth table for implication is the same as <=u on
822
/// i1 values (but not
823
/// <=s!).  The truth table for both is:
824
///    | T | F (B)
825
///  T | T | F
826
///  F | T | T
827
/// (A)
828
std::optional<bool> isImpliedCondition(const Value *LHS, const Value *RHS,
829
                                       const DataLayout &DL,
830
                                       bool LHSIsTrue = true,
831
                                       unsigned Depth = 0);
832
std::optional<bool> isImpliedCondition(const Value *LHS,
833
                                       CmpInst::Predicate RHSPred,
834
                                       const Value *RHSOp0, const Value *RHSOp1,
835
                                       const DataLayout &DL,
836
                                       bool LHSIsTrue = true,
837
                                       unsigned Depth = 0);
838
 
839
/// Return the boolean condition value in the context of the given instruction
840
/// if it is known based on dominating conditions.
841
std::optional<bool> isImpliedByDomCondition(const Value *Cond,
842
                                            const Instruction *ContextI,
843
                                            const DataLayout &DL);
844
std::optional<bool> isImpliedByDomCondition(CmpInst::Predicate Pred,
845
                                            const Value *LHS, const Value *RHS,
846
                                            const Instruction *ContextI,
847
                                            const DataLayout &DL);
848
 
849
/// If Ptr1 is provably equal to Ptr2 plus a constant offset, return that
850
/// offset. For example, Ptr1 might be &A[42], and Ptr2 might be &A[40]. In
851
/// this case offset would be -8.
852
std::optional<int64_t> isPointerOffset(const Value *Ptr1, const Value *Ptr2,
853
                                       const DataLayout &DL);
854
} // end namespace llvm
855
 
856
#endif // LLVM_ANALYSIS_VALUETRACKING_H