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/Analysis/SimplifyQuery.h" 19#include "llvm/Analysis/WithCache.h" 20#include "llvm/IR/Constants.h" 21#include "llvm/IR/DataLayout.h" 22#include "llvm/IR/FMF.h" 23#include "llvm/IR/InstrTypes.h" 24#include "llvm/IR/Intrinsics.h" 25#include <cassert> 26#include <cstdint> 27 28namespace llvm { 29 30class Operator; 31class AddOperator; 32class AllocaInst; 33class APInt; 34class AssumptionCache; 35class DominatorTree; 36class GEPOperator; 37class LoadInst; 38class WithOverflowInst; 39struct KnownBits; 40class Loop; 41class LoopInfo; 42class MDNode; 43class StringRef; 44class TargetLibraryInfo; 45class Value; 46 47constexpr unsigned MaxAnalysisRecursionDepth = 6; 48 49/// Determine which bits of V are known to be either zero or one and return 50/// them in the KnownZero/KnownOne bit sets. 51/// 52/// This function is defined on values with integer type, values with pointer 53/// type, and vectors of integers. In the case 54/// where V is a vector, the known zero and known one values are the 55/// same width as the vector element, and the bit is set only if it is true 56/// for all of the elements in the vector. 57void computeKnownBits(const Value *V, KnownBits &Known, const DataLayout &DL, 58 unsigned Depth = 0, AssumptionCache *AC = nullptr, 59 const Instruction *CxtI = nullptr, 60 const DominatorTree *DT = nullptr, 61 bool UseInstrInfo = true); 62 63/// Returns the known bits rather than passing by reference. 64KnownBits computeKnownBits(const Value *V, const DataLayout &DL, 65 unsigned Depth = 0, AssumptionCache *AC = nullptr, 66 const Instruction *CxtI = nullptr, 67 const DominatorTree *DT = nullptr, 68 bool UseInstrInfo = true); 69 70/// Returns the known bits rather than passing by reference. 71KnownBits computeKnownBits(const Value *V, const APInt &DemandedElts, 72 const DataLayout &DL, unsigned Depth = 0, 73 AssumptionCache *AC = nullptr, 74 const Instruction *CxtI = nullptr, 75 const DominatorTree *DT = nullptr, 76 bool UseInstrInfo = true); 77 78KnownBits computeKnownBits(const Value *V, const APInt &DemandedElts, 79 unsigned Depth, const SimplifyQuery &Q); 80 81KnownBits computeKnownBits(const Value *V, unsigned Depth, 82 const SimplifyQuery &Q); 83 84void computeKnownBits(const Value *V, KnownBits &Known, unsigned Depth, 85 const SimplifyQuery &Q); 86 87/// Compute known bits from the range metadata. 88/// \p KnownZero the set of bits that are known to be zero 89/// \p KnownOne the set of bits that are known to be one 90void computeKnownBitsFromRangeMetadata(const MDNode &Ranges, KnownBits &Known); 91 92/// Merge bits known from context-dependent facts into Known. 93void computeKnownBitsFromContext(const Value *V, KnownBits &Known, 94 unsigned Depth, const SimplifyQuery &Q); 95 96/// Using KnownBits LHS/RHS produce the known bits for logic op (and/xor/or). 97KnownBits analyzeKnownBitsFromAndXorOr(const Operator *I, 98 const KnownBits &KnownLHS, 99 const KnownBits &KnownRHS, 100 unsigned Depth, const SimplifyQuery &SQ); 101 102/// Return true if LHS and RHS have no common bits set. 103bool haveNoCommonBitsSet(const WithCache<const Value *> &LHSCache, 104 const WithCache<const Value *> &RHSCache, 105 const SimplifyQuery &SQ); 106 107/// Return true if the given value is known to have exactly one bit set when 108/// defined. For vectors return true if every element is known to be a power 109/// of two when defined. Supports values with integer or pointer type and 110/// vectors of integers. If 'OrZero' is set, then return true if the given 111/// value is either a power of two or zero. 112bool isKnownToBeAPowerOfTwo(const Value *V, const DataLayout &DL, 113 bool OrZero = false, unsigned Depth = 0, 114 AssumptionCache *AC = nullptr, 115 const Instruction *CxtI = nullptr, 116 const DominatorTree *DT = nullptr, 117 bool UseInstrInfo = true); 118 119bool isOnlyUsedInZeroEqualityComparison(const Instruction *CxtI); 120 121/// Return true if the given value is known to be non-zero when defined. For 122/// vectors, return true if every element is known to be non-zero when 123/// defined. For pointers, if the context instruction and dominator tree are 124/// specified, perform context-sensitive analysis and return true if the 125/// pointer couldn't possibly be null at the specified instruction. 126/// Supports values with integer or pointer type and vectors of integers. 127bool isKnownNonZero(const Value *V, const DataLayout &DL, unsigned Depth = 0, 128 AssumptionCache *AC = nullptr, 129 const Instruction *CxtI = nullptr, 130 const DominatorTree *DT = nullptr, 131 bool UseInstrInfo = true); 132 133/// Return true if the two given values are negation. 134/// Currently can recoginze Value pair: 135/// 1: <X, Y> if X = sub (0, Y) or Y = sub (0, X) 136/// 2: <X, Y> if X = sub (A, B) and Y = sub (B, A) 137bool isKnownNegation(const Value *X, const Value *Y, bool NeedNSW = false); 138 139/// Returns true if the give value is known to be non-negative. 140bool isKnownNonNegative(const Value *V, const SimplifyQuery &SQ, 141 unsigned Depth = 0); 142 143/// Returns true if the given value is known be positive (i.e. non-negative 144/// and non-zero). 145bool isKnownPositive(const Value *V, const SimplifyQuery &SQ, 146 unsigned Depth = 0); 147 148/// Returns true if the given value is known be negative (i.e. non-positive 149/// and non-zero). 150bool isKnownNegative(const Value *V, const SimplifyQuery &DL, 151 unsigned Depth = 0); 152 153/// Return true if the given values are known to be non-equal when defined. 154/// Supports scalar integer types only. 155bool isKnownNonEqual(const Value *V1, const Value *V2, const DataLayout &DL, 156 AssumptionCache *AC = nullptr, 157 const Instruction *CxtI = nullptr, 158 const DominatorTree *DT = nullptr, 159 bool UseInstrInfo = true); 160 161/// Return true if 'V & Mask' is known to be zero. We use this predicate to 162/// simplify operations downstream. Mask is known to be zero for bits that V 163/// cannot have. 164/// 165/// This function is defined on values with integer type, values with pointer 166/// type, and vectors of integers. In the case 167/// where V is a vector, the mask, known zero, and known one values are the 168/// same width as the vector element, and the bit is set only if it is true 169/// for all of the elements in the vector. 170bool MaskedValueIsZero(const Value *V, const APInt &Mask, 171 const SimplifyQuery &DL, unsigned Depth = 0); 172 173/// Return the number of times the sign bit of the register is replicated into 174/// the other bits. We know that at least 1 bit is always equal to the sign 175/// bit (itself), but other cases can give us information. For example, 176/// immediately after an "ashr X, 2", we know that the top 3 bits are all 177/// equal to each other, so we return 3. For vectors, return the number of 178/// sign bits for the vector element with the mininum number of known sign 179/// bits. 180unsigned ComputeNumSignBits(const Value *Op, 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/// Get the upper bound on bit size for this Value \p Op as a signed integer. 187/// i.e. x == sext(trunc(x to MaxSignificantBits) to bitwidth(x)). 188/// Similar to the APInt::getSignificantBits function. 189unsigned ComputeMaxSignificantBits(const Value *Op, const DataLayout &DL, 190 unsigned Depth = 0, 191 AssumptionCache *AC = nullptr, 192 const Instruction *CxtI = nullptr, 193 const DominatorTree *DT = nullptr); 194 195/// Map a call instruction to an intrinsic ID. Libcalls which have equivalent 196/// intrinsics are treated as-if they were intrinsics. 197Intrinsic::ID getIntrinsicForCallSite(const CallBase &CB, 198 const TargetLibraryInfo *TLI); 199 200/// Returns a pair of values, which if passed to llvm.is.fpclass, returns the 201/// same result as an fcmp with the given operands. 202/// 203/// If \p LookThroughSrc is true, consider the input value when computing the 204/// mask. 205/// 206/// If \p LookThroughSrc is false, ignore the source value (i.e. the first pair 207/// element will always be LHS. 208std::pair<Value *, FPClassTest> fcmpToClassTest(CmpInst::Predicate Pred, 209 const Function &F, Value *LHS, 210 Value *RHS, 211 bool LookThroughSrc = true); 212std::pair<Value *, FPClassTest> fcmpToClassTest(CmpInst::Predicate Pred, 213 const Function &F, Value *LHS, 214 const APFloat *ConstRHS, 215 bool LookThroughSrc = true); 216 217struct KnownFPClass { 218 /// Floating-point classes the value could be one of. 219 FPClassTest KnownFPClasses = fcAllFlags; 220 221 /// std::nullopt if the sign bit is unknown, true if the sign bit is 222 /// definitely set or false if the sign bit is definitely unset. 223 std::optional<bool> SignBit; 224 225 /// Return true if it's known this can never be one of the mask entries. 226 bool isKnownNever(FPClassTest Mask) const { 227 return (KnownFPClasses & Mask) == fcNone; 228 } 229 230 bool isUnknown() const { 231 return KnownFPClasses == fcAllFlags && !SignBit; 232 } 233 234 /// Return true if it's known this can never be a nan. 235 bool isKnownNeverNaN() const { 236 return isKnownNever(fcNan); 237 } 238 239 /// Return true if it's known this can never be an infinity. 240 bool isKnownNeverInfinity() const { 241 return isKnownNever(fcInf); 242 } 243 244 /// Return true if it's known this can never be +infinity. 245 bool isKnownNeverPosInfinity() const { 246 return isKnownNever(fcPosInf); 247 } 248 249 /// Return true if it's known this can never be -infinity. 250 bool isKnownNeverNegInfinity() const { 251 return isKnownNever(fcNegInf); 252 } 253 254 /// Return true if it's known this can never be a subnormal 255 bool isKnownNeverSubnormal() const { 256 return isKnownNever(fcSubnormal); 257 } 258 259 /// Return true if it's known this can never be a positive subnormal 260 bool isKnownNeverPosSubnormal() const { 261 return isKnownNever(fcPosSubnormal); 262 } 263 264 /// Return true if it's known this can never be a negative subnormal 265 bool isKnownNeverNegSubnormal() const { 266 return isKnownNever(fcNegSubnormal); 267 } 268 269 /// Return true if it's known this can never be a zero. This means a literal 270 /// [+-]0, and does not include denormal inputs implicitly treated as [+-]0. 271 bool isKnownNeverZero() const { 272 return isKnownNever(fcZero); 273 } 274 275 /// Return true if it's known this can never be a literal positive zero. 276 bool isKnownNeverPosZero() const { 277 return isKnownNever(fcPosZero); 278 } 279 280 /// Return true if it's known this can never be a negative zero. This means a 281 /// literal -0 and does not include denormal inputs implicitly treated as -0. 282 bool isKnownNeverNegZero() const { 283 return isKnownNever(fcNegZero); 284 } 285 286 /// Return true if it's know this can never be interpreted as a zero. This 287 /// extends isKnownNeverZero to cover the case where the assumed 288 /// floating-point mode for the function interprets denormals as zero. 289 bool isKnownNeverLogicalZero(const Function &F, Type *Ty) const; 290 291 /// Return true if it's know this can never be interpreted as a negative zero. 292 bool isKnownNeverLogicalNegZero(const Function &F, Type *Ty) const; 293 294 /// Return true if it's know this can never be interpreted as a positive zero. 295 bool isKnownNeverLogicalPosZero(const Function &F, Type *Ty) const; 296 297 static constexpr FPClassTest OrderedLessThanZeroMask = 298 fcNegSubnormal | fcNegNormal | fcNegInf; 299 static constexpr FPClassTest OrderedGreaterThanZeroMask = 300 fcPosSubnormal | fcPosNormal | fcPosInf; 301 302 /// Return true if we can prove that the analyzed floating-point value is 303 /// either NaN or never less than -0.0. 304 /// 305 /// NaN --> true 306 /// +0 --> true 307 /// -0 --> true 308 /// x > +0 --> true 309 /// x < -0 --> false 310 bool cannotBeOrderedLessThanZero() const { 311 return isKnownNever(OrderedLessThanZeroMask); 312 } 313 314 /// Return true if we can prove that the analyzed floating-point value is 315 /// either NaN or never greater than -0.0. 316 /// NaN --> true 317 /// +0 --> true 318 /// -0 --> true 319 /// x > +0 --> false 320 /// x < -0 --> true 321 bool cannotBeOrderedGreaterThanZero() const { 322 return isKnownNever(OrderedGreaterThanZeroMask); 323 } 324 325 KnownFPClass &operator|=(const KnownFPClass &RHS) { 326 KnownFPClasses = KnownFPClasses | RHS.KnownFPClasses; 327 328 if (SignBit != RHS.SignBit) 329 SignBit = std::nullopt; 330 return *this; 331 } 332 333 void knownNot(FPClassTest RuleOut) { 334 KnownFPClasses = KnownFPClasses & ~RuleOut; 335 } 336 337 void fneg() { 338 KnownFPClasses = llvm::fneg(KnownFPClasses); 339 if (SignBit) 340 SignBit = !*SignBit; 341 } 342 343 void fabs() { 344 if (KnownFPClasses & fcNegZero) 345 KnownFPClasses |= fcPosZero; 346 347 if (KnownFPClasses & fcNegInf) 348 KnownFPClasses |= fcPosInf; 349 350 if (KnownFPClasses & fcNegSubnormal) 351 KnownFPClasses |= fcPosSubnormal; 352 353 if (KnownFPClasses & fcNegNormal) 354 KnownFPClasses |= fcPosNormal; 355 356 signBitMustBeZero(); 357 } 358 359 /// Return true if the sign bit must be 0, ignoring the sign of nans. 360 bool signBitIsZeroOrNaN() const { 361 return isKnownNever(fcNegative); 362 } 363 364 /// Assume the sign bit is zero. 365 void signBitMustBeZero() { 366 KnownFPClasses &= (fcPositive | fcNan); 367 SignBit = false; 368 } 369 370 void copysign(const KnownFPClass &Sign) { 371 // Don't know anything about the sign of the source. Expand the possible set 372 // to its opposite sign pair. 373 if (KnownFPClasses & fcZero) 374 KnownFPClasses |= fcZero; 375 if (KnownFPClasses & fcSubnormal) 376 KnownFPClasses |= fcSubnormal; 377 if (KnownFPClasses & fcNormal) 378 KnownFPClasses |= fcNormal; 379 if (KnownFPClasses & fcInf) 380 KnownFPClasses |= fcInf; 381 382 // Sign bit is exactly preserved even for nans. 383 SignBit = Sign.SignBit; 384 385 // Clear sign bits based on the input sign mask. 386 if (Sign.isKnownNever(fcPositive | fcNan) || (SignBit && *SignBit)) 387 KnownFPClasses &= (fcNegative | fcNan); 388 if (Sign.isKnownNever(fcNegative | fcNan) || (SignBit && !*SignBit)) 389 KnownFPClasses &= (fcPositive | fcNan); 390 } 391 392 // Propagate knowledge that a non-NaN source implies the result can also not 393 // be a NaN. For unconstrained operations, signaling nans are not guaranteed 394 // to be quieted but cannot be introduced. 395 void propagateNaN(const KnownFPClass &Src, bool PreserveSign = false) { 396 if (Src.isKnownNever(fcNan)) { 397 knownNot(fcNan); 398 if (PreserveSign) 399 SignBit = Src.SignBit; 400 } else if (Src.isKnownNever(fcSNan)) 401 knownNot(fcSNan); 402 } 403 404 /// Propagate knowledge from a source value that could be a denormal or 405 /// zero. We have to be conservative since output flushing is not guaranteed, 406 /// so known-never-zero may not hold. 407 /// 408 /// This assumes a copy-like operation and will replace any currently known 409 /// information. 410 void propagateDenormal(const KnownFPClass &Src, const Function &F, Type *Ty); 411 412 /// Report known classes if \p Src is evaluated through a potentially 413 /// canonicalizing operation. We can assume signaling nans will not be 414 /// introduced, but cannot assume a denormal will be flushed under FTZ/DAZ. 415 /// 416 /// This assumes a copy-like operation and will replace any currently known 417 /// information. 418 void propagateCanonicalizingSrc(const KnownFPClass &Src, const Function &F, 419 Type *Ty); 420 421 void resetAll() { *this = KnownFPClass(); } 422}; 423 424inline KnownFPClass operator|(KnownFPClass LHS, const KnownFPClass &RHS) { 425 LHS |= RHS; 426 return LHS; 427} 428 429inline KnownFPClass operator|(const KnownFPClass &LHS, KnownFPClass &&RHS) { 430 RHS |= LHS; 431 return std::move(RHS); 432} 433 434/// Determine which floating-point classes are valid for \p V, and return them 435/// in KnownFPClass bit sets. 436/// 437/// This function is defined on values with floating-point type, values vectors 438/// of floating-point type, and arrays of floating-point type. 439 440/// \p InterestedClasses is a compile time optimization hint for which floating 441/// point classes should be queried. Queries not specified in \p 442/// InterestedClasses should be reliable if they are determined during the 443/// query. 444KnownFPClass computeKnownFPClass( 445 const Value *V, const APInt &DemandedElts, const DataLayout &DL, 446 FPClassTest InterestedClasses = fcAllFlags, unsigned Depth = 0, 447 const TargetLibraryInfo *TLI = nullptr, AssumptionCache *AC = nullptr, 448 const Instruction *CxtI = nullptr, const DominatorTree *DT = nullptr, 449 bool UseInstrInfo = true); 450 451KnownFPClass computeKnownFPClass( 452 const Value *V, const DataLayout &DL, 453 FPClassTest InterestedClasses = fcAllFlags, unsigned Depth = 0, 454 const TargetLibraryInfo *TLI = nullptr, AssumptionCache *AC = nullptr, 455 const Instruction *CxtI = nullptr, const DominatorTree *DT = nullptr, 456 bool UseInstrInfo = true); 457 458/// Wrapper to account for known fast math flags at the use instruction. 459inline KnownFPClass computeKnownFPClass( 460 const Value *V, FastMathFlags FMF, const DataLayout &DL, 461 FPClassTest InterestedClasses = fcAllFlags, unsigned Depth = 0, 462 const TargetLibraryInfo *TLI = nullptr, AssumptionCache *AC = nullptr, 463 const Instruction *CxtI = nullptr, const DominatorTree *DT = nullptr, 464 bool UseInstrInfo = true) { 465 if (FMF.noNaNs()) 466 InterestedClasses &= ~fcNan; 467 if (FMF.noInfs()) 468 InterestedClasses &= ~fcInf; 469 470 KnownFPClass Result = computeKnownFPClass(V, DL, InterestedClasses, Depth, 471 TLI, AC, CxtI, DT, UseInstrInfo); 472 473 if (FMF.noNaNs()) 474 Result.KnownFPClasses &= ~fcNan; 475 if (FMF.noInfs()) 476 Result.KnownFPClasses &= ~fcInf; 477 return Result; 478} 479 480/// Return true if we can prove that the specified FP value is never equal to 481/// -0.0. Users should use caution when considering PreserveSign 482/// denormal-fp-math. 483inline bool cannotBeNegativeZero(const Value *V, const DataLayout &DL, 484 const TargetLibraryInfo *TLI = nullptr, 485 unsigned Depth = 0, 486 AssumptionCache *AC = nullptr, 487 const Instruction *CtxI = nullptr, 488 const DominatorTree *DT = nullptr, 489 bool UseInstrInfo = true) { 490 KnownFPClass Known = computeKnownFPClass(V, DL, fcNegZero, Depth, TLI, AC, 491 CtxI, DT, UseInstrInfo); 492 return Known.isKnownNeverNegZero(); 493} 494 495/// Return true if we can prove that the specified FP value is either NaN or 496/// never less than -0.0. 497/// 498/// NaN --> true 499/// +0 --> true 500/// -0 --> true 501/// x > +0 --> true 502/// x < -0 --> false 503inline bool cannotBeOrderedLessThanZero(const Value *V, const DataLayout &DL, 504 const TargetLibraryInfo *TLI = nullptr, 505 unsigned Depth = 0, 506 AssumptionCache *AC = nullptr, 507 const Instruction *CtxI = nullptr, 508 const DominatorTree *DT = nullptr, 509 bool UseInstrInfo = true) { 510 KnownFPClass Known = 511 computeKnownFPClass(V, DL, KnownFPClass::OrderedLessThanZeroMask, Depth, 512 TLI, AC, CtxI, DT, UseInstrInfo); 513 return Known.cannotBeOrderedLessThanZero(); 514} 515 516/// Return true if the floating-point scalar value is not an infinity or if 517/// the floating-point vector value has no infinities. Return false if a value 518/// could ever be infinity. 519inline bool isKnownNeverInfinity(const Value *V, const DataLayout &DL, 520 const TargetLibraryInfo *TLI = nullptr, 521 unsigned Depth = 0, 522 AssumptionCache *AC = nullptr, 523 const Instruction *CtxI = nullptr, 524 const DominatorTree *DT = nullptr, 525 bool UseInstrInfo = true) { 526 KnownFPClass Known = computeKnownFPClass(V, DL, fcInf, Depth, TLI, AC, CtxI, 527 DT, UseInstrInfo); 528 return Known.isKnownNeverInfinity(); 529} 530 531/// Return true if the floating-point value can never contain a NaN or infinity. 532inline bool isKnownNeverInfOrNaN( 533 const Value *V, const DataLayout &DL, const TargetLibraryInfo *TLI, 534 unsigned Depth = 0, AssumptionCache *AC = nullptr, 535 const Instruction *CtxI = nullptr, const DominatorTree *DT = nullptr, 536 bool UseInstrInfo = true) { 537 KnownFPClass Known = computeKnownFPClass(V, DL, fcInf | fcNan, Depth, TLI, AC, 538 CtxI, DT, UseInstrInfo); 539 return Known.isKnownNeverNaN() && Known.isKnownNeverInfinity(); 540} 541 542/// Return true if the floating-point scalar value is not a NaN or if the 543/// floating-point vector value has no NaN elements. Return false if a value 544/// could ever be NaN. 545inline bool isKnownNeverNaN(const Value *V, const DataLayout &DL, 546 const TargetLibraryInfo *TLI, unsigned Depth = 0, 547 AssumptionCache *AC = nullptr, 548 const Instruction *CtxI = nullptr, 549 const DominatorTree *DT = nullptr, 550 bool UseInstrInfo = true) { 551 KnownFPClass Known = computeKnownFPClass(V, DL, fcNan, Depth, TLI, AC, CtxI, 552 DT, UseInstrInfo); 553 return Known.isKnownNeverNaN(); 554} 555 556/// Return true if we can prove that the specified FP value's sign bit is 0. 557/// 558/// NaN --> true/false (depending on the NaN's sign bit) 559/// +0 --> true 560/// -0 --> false 561/// x > +0 --> true 562/// x < -0 --> false 563bool SignBitMustBeZero(const Value *V, const DataLayout &DL, 564 const TargetLibraryInfo *TLI); 565 566/// If the specified value can be set by repeating the same byte in memory, 567/// return the i8 value that it is represented with. This is true for all i8 568/// values obviously, but is also true for i32 0, i32 -1, i16 0xF0F0, double 569/// 0.0 etc. If the value can't be handled with a repeated byte store (e.g. 570/// i16 0x1234), return null. If the value is entirely undef and padding, 571/// return undef. 572Value *isBytewiseValue(Value *V, const DataLayout &DL); 573 574/// Given an aggregate and an sequence of indices, see if the scalar value 575/// indexed is already around as a register, for example if it were inserted 576/// directly into the aggregate. 577/// 578/// If InsertBefore is not null, this function will duplicate (modified) 579/// insertvalues when a part of a nested struct is extracted. 580Value *FindInsertedValue(Value *V, ArrayRef<unsigned> idx_range, 581 Instruction *InsertBefore = nullptr); 582 583/// Analyze the specified pointer to see if it can be expressed as a base 584/// pointer plus a constant offset. Return the base and offset to the caller. 585/// 586/// This is a wrapper around Value::stripAndAccumulateConstantOffsets that 587/// creates and later unpacks the required APInt. 588inline Value *GetPointerBaseWithConstantOffset(Value *Ptr, int64_t &Offset, 589 const DataLayout &DL, 590 bool AllowNonInbounds = true) { 591 APInt OffsetAPInt(DL.getIndexTypeSizeInBits(Ptr->getType()), 0); 592 Value *Base = 593 Ptr->stripAndAccumulateConstantOffsets(DL, OffsetAPInt, AllowNonInbounds); 594 595 Offset = OffsetAPInt.getSExtValue(); 596 return Base; 597} 598inline const Value * 599GetPointerBaseWithConstantOffset(const Value *Ptr, int64_t &Offset, 600 const DataLayout &DL, 601 bool AllowNonInbounds = true) { 602 return GetPointerBaseWithConstantOffset(const_cast<Value *>(Ptr), Offset, DL, 603 AllowNonInbounds); 604} 605 606/// Returns true if the GEP is based on a pointer to a string (array of 607// \p CharSize integers) and is indexing into this string. 608bool isGEPBasedOnPointerToString(const GEPOperator *GEP, unsigned CharSize = 8); 609 610/// Represents offset+length into a ConstantDataArray. 611struct ConstantDataArraySlice { 612 /// ConstantDataArray pointer. nullptr indicates a zeroinitializer (a valid 613 /// initializer, it just doesn't fit the ConstantDataArray interface). 614 const ConstantDataArray *Array; 615 616 /// Slice starts at this Offset. 617 uint64_t Offset; 618 619 /// Length of the slice. 620 uint64_t Length; 621 622 /// Moves the Offset and adjusts Length accordingly. 623 void move(uint64_t Delta) { 624 assert(Delta < Length); 625 Offset += Delta; 626 Length -= Delta; 627 } 628 629 /// Convenience accessor for elements in the slice. 630 uint64_t operator[](unsigned I) const { 631 return Array == nullptr ? 0 : Array->getElementAsInteger(I + Offset); 632 } 633}; 634 635/// Returns true if the value \p V is a pointer into a ConstantDataArray. 636/// If successful \p Slice will point to a ConstantDataArray info object 637/// with an appropriate offset. 638bool getConstantDataArrayInfo(const Value *V, ConstantDataArraySlice &Slice, 639 unsigned ElementSize, uint64_t Offset = 0); 640 641/// This function computes the length of a null-terminated C string pointed to 642/// by V. If successful, it returns true and returns the string in Str. If 643/// unsuccessful, it returns false. This does not include the trailing null 644/// character by default. If TrimAtNul is set to false, then this returns any 645/// trailing null characters as well as any other characters that come after 646/// it. 647bool getConstantStringInfo(const Value *V, StringRef &Str, 648 bool TrimAtNul = true); 649 650/// If we can compute the length of the string pointed to by the specified 651/// pointer, return 'len+1'. If we can't, return 0. 652uint64_t GetStringLength(const Value *V, unsigned CharSize = 8); 653 654/// This function returns call pointer argument that is considered the same by 655/// aliasing rules. You CAN'T use it to replace one value with another. If 656/// \p MustPreserveNullness is true, the call must preserve the nullness of 657/// the pointer. 658const Value *getArgumentAliasingToReturnedPointer(const CallBase *Call, 659 bool MustPreserveNullness); 660inline Value *getArgumentAliasingToReturnedPointer(CallBase *Call, 661 bool MustPreserveNullness) { 662 return const_cast<Value *>(getArgumentAliasingToReturnedPointer( 663 const_cast<const CallBase *>(Call), MustPreserveNullness)); 664} 665 666/// {launder,strip}.invariant.group returns pointer that aliases its argument, 667/// and it only captures pointer by returning it. 668/// These intrinsics are not marked as nocapture, because returning is 669/// considered as capture. The arguments are not marked as returned neither, 670/// because it would make it useless. If \p MustPreserveNullness is true, 671/// the intrinsic must preserve the nullness of the pointer. 672bool isIntrinsicReturningPointerAliasingArgumentWithoutCapturing( 673 const CallBase *Call, bool MustPreserveNullness); 674 675/// This method strips off any GEP address adjustments and pointer casts from 676/// the specified value, returning the original object being addressed. Note 677/// that the returned value has pointer type if the specified value does. If 678/// the MaxLookup value is non-zero, it limits the number of instructions to 679/// be stripped off. 680const Value *getUnderlyingObject(const Value *V, unsigned MaxLookup = 6); 681inline Value *getUnderlyingObject(Value *V, unsigned MaxLookup = 6) { 682 // Force const to avoid infinite recursion. 683 const Value *VConst = V; 684 return const_cast<Value *>(getUnderlyingObject(VConst, MaxLookup)); 685} 686 687/// This method is similar to getUnderlyingObject except that it can 688/// look through phi and select instructions and return multiple objects. 689/// 690/// If LoopInfo is passed, loop phis are further analyzed. If a pointer 691/// accesses different objects in each iteration, we don't look through the 692/// phi node. E.g. consider this loop nest: 693/// 694/// int **A; 695/// for (i) 696/// for (j) { 697/// A[i][j] = A[i-1][j] * B[j] 698/// } 699/// 700/// This is transformed by Load-PRE to stash away A[i] for the next iteration 701/// of the outer loop: 702/// 703/// Curr = A[0]; // Prev_0 704/// for (i: 1..N) { 705/// Prev = Curr; // Prev = PHI (Prev_0, Curr) 706/// Curr = A[i]; 707/// for (j: 0..N) { 708/// Curr[j] = Prev[j] * B[j] 709/// } 710/// } 711/// 712/// Since A[i] and A[i-1] are independent pointers, getUnderlyingObjects 713/// should not assume that Curr and Prev share the same underlying object thus 714/// it shouldn't look through the phi above. 715void getUnderlyingObjects(const Value *V, 716 SmallVectorImpl<const Value *> &Objects, 717 LoopInfo *LI = nullptr, unsigned MaxLookup = 6); 718 719/// This is a wrapper around getUnderlyingObjects and adds support for basic 720/// ptrtoint+arithmetic+inttoptr sequences. 721bool getUnderlyingObjectsForCodeGen(const Value *V, 722 SmallVectorImpl<Value *> &Objects); 723 724/// Returns unique alloca where the value comes from, or nullptr. 725/// If OffsetZero is true check that V points to the begining of the alloca. 726AllocaInst *findAllocaForValue(Value *V, bool OffsetZero = false); 727inline const AllocaInst *findAllocaForValue(const Value *V, 728 bool OffsetZero = false) { 729 return findAllocaForValue(const_cast<Value *>(V), OffsetZero); 730} 731 732/// Return true if the only users of this pointer are lifetime markers. 733bool onlyUsedByLifetimeMarkers(const Value *V); 734 735/// Return true if the only users of this pointer are lifetime markers or 736/// droppable instructions. 737bool onlyUsedByLifetimeMarkersOrDroppableInsts(const Value *V); 738 739/// Return true if speculation of the given load must be suppressed to avoid 740/// ordering or interfering with an active sanitizer. If not suppressed, 741/// dereferenceability and alignment must be proven separately. Note: This 742/// is only needed for raw reasoning; if you use the interface below 743/// (isSafeToSpeculativelyExecute), this is handled internally. 744bool mustSuppressSpeculation(const LoadInst &LI); 745 746/// Return true if the instruction does not have any effects besides 747/// calculating the result and does not have undefined behavior. 748/// 749/// This method never returns true for an instruction that returns true for 750/// mayHaveSideEffects; however, this method also does some other checks in 751/// addition. It checks for undefined behavior, like dividing by zero or 752/// loading from an invalid pointer (but not for undefined results, like a 753/// shift with a shift amount larger than the width of the result). It checks 754/// for malloc and alloca because speculatively executing them might cause a 755/// memory leak. It also returns false for instructions related to control 756/// flow, specifically terminators and PHI nodes. 757/// 758/// If the CtxI is specified this method performs context-sensitive analysis 759/// and returns true if it is safe to execute the instruction immediately 760/// before the CtxI. 761/// 762/// If the CtxI is NOT specified this method only looks at the instruction 763/// itself and its operands, so if this method returns true, it is safe to 764/// move the instruction as long as the correct dominance relationships for 765/// the operands and users hold. 766/// 767/// This method can return true for instructions that read memory; 768/// for such instructions, moving them may change the resulting value. 769bool isSafeToSpeculativelyExecute(const Instruction *I, 770 const Instruction *CtxI = nullptr, 771 AssumptionCache *AC = nullptr, 772 const DominatorTree *DT = nullptr, 773 const TargetLibraryInfo *TLI = nullptr); 774 775/// This returns the same result as isSafeToSpeculativelyExecute if Opcode is 776/// the actual opcode of Inst. If the provided and actual opcode differ, the 777/// function (virtually) overrides the opcode of Inst with the provided 778/// Opcode. There are come constraints in this case: 779/// * If Opcode has a fixed number of operands (eg, as binary operators do), 780/// then Inst has to have at least as many leading operands. The function 781/// will ignore all trailing operands beyond that number. 782/// * If Opcode allows for an arbitrary number of operands (eg, as CallInsts 783/// do), then all operands are considered. 784/// * The virtual instruction has to satisfy all typing rules of the provided 785/// Opcode. 786/// * This function is pessimistic in the following sense: If one actually 787/// materialized the virtual instruction, then isSafeToSpeculativelyExecute 788/// may say that the materialized instruction is speculatable whereas this 789/// function may have said that the instruction wouldn't be speculatable. 790/// This behavior is a shortcoming in the current implementation and not 791/// intentional. 792bool isSafeToSpeculativelyExecuteWithOpcode( 793 unsigned Opcode, const Instruction *Inst, const Instruction *CtxI = nullptr, 794 AssumptionCache *AC = nullptr, const DominatorTree *DT = nullptr, 795 const TargetLibraryInfo *TLI = nullptr); 796 797/// Returns true if the result or effects of the given instructions \p I 798/// depend values not reachable through the def use graph. 799/// * Memory dependence arises for example if the instruction reads from 800/// memory or may produce effects or undefined behaviour. Memory dependent 801/// instructions generally cannot be reorderd with respect to other memory 802/// dependent instructions. 803/// * Control dependence arises for example if the instruction may fault 804/// if lifted above a throwing call or infinite loop. 805bool mayHaveNonDefUseDependency(const Instruction &I); 806 807/// Return true if it is an intrinsic that cannot be speculated but also 808/// cannot trap. 809bool isAssumeLikeIntrinsic(const Instruction *I); 810 811/// Return true if it is valid to use the assumptions provided by an 812/// assume intrinsic, I, at the point in the control-flow identified by the 813/// context instruction, CxtI. By default, ephemeral values of the assumption 814/// are treated as an invalid context, to prevent the assumption from being used 815/// to optimize away its argument. If the caller can ensure that this won't 816/// happen, it can call with AllowEphemerals set to true to get more valid 817/// assumptions. 818bool isValidAssumeForContext(const Instruction *I, const Instruction *CxtI, 819 const DominatorTree *DT = nullptr, 820 bool AllowEphemerals = false); 821 822enum class OverflowResult { 823 /// Always overflows in the direction of signed/unsigned min value. 824 AlwaysOverflowsLow, 825 /// Always overflows in the direction of signed/unsigned max value. 826 AlwaysOverflowsHigh, 827 /// May or may not overflow. 828 MayOverflow, 829 /// Never overflows. 830 NeverOverflows, 831}; 832 833OverflowResult computeOverflowForUnsignedMul(const Value *LHS, const Value *RHS, 834 const SimplifyQuery &SQ); 835OverflowResult computeOverflowForSignedMul(const Value *LHS, const Value *RHS, 836 const SimplifyQuery &SQ); 837OverflowResult 838computeOverflowForUnsignedAdd(const WithCache<const Value *> &LHS, 839 const WithCache<const Value *> &RHS, 840 const SimplifyQuery &SQ); 841OverflowResult computeOverflowForSignedAdd(const WithCache<const Value *> &LHS, 842 const WithCache<const Value *> &RHS, 843 const SimplifyQuery &SQ); 844/// This version also leverages the sign bit of Add if known. 845OverflowResult computeOverflowForSignedAdd(const AddOperator *Add, 846 const SimplifyQuery &SQ); 847OverflowResult computeOverflowForUnsignedSub(const Value *LHS, const Value *RHS, 848 const SimplifyQuery &SQ); 849OverflowResult computeOverflowForSignedSub(const Value *LHS, const Value *RHS, 850 const SimplifyQuery &SQ); 851 852/// Returns true if the arithmetic part of the \p WO 's result is 853/// used only along the paths control dependent on the computation 854/// not overflowing, \p WO being an <op>.with.overflow intrinsic. 855bool isOverflowIntrinsicNoWrap(const WithOverflowInst *WO, 856 const DominatorTree &DT); 857 858/// Determine the possible constant range of vscale with the given bit width, 859/// based on the vscale_range function attribute. 860ConstantRange getVScaleRange(const Function *F, unsigned BitWidth); 861 862/// Determine the possible constant range of an integer or vector of integer 863/// value. This is intended as a cheap, non-recursive check. 864ConstantRange computeConstantRange(const Value *V, bool ForSigned, 865 bool UseInstrInfo = true, 866 AssumptionCache *AC = nullptr, 867 const Instruction *CtxI = nullptr, 868 const DominatorTree *DT = nullptr, 869 unsigned Depth = 0); 870 871/// Combine constant ranges from computeConstantRange() and computeKnownBits(). 872ConstantRange 873computeConstantRangeIncludingKnownBits(const WithCache<const Value *> &V, 874 bool ForSigned, const SimplifyQuery &SQ); 875 876/// Return true if this function can prove that the instruction I will 877/// always transfer execution to one of its successors (including the next 878/// instruction that follows within a basic block). E.g. this is not 879/// guaranteed for function calls that could loop infinitely. 880/// 881/// In other words, this function returns false for instructions that may 882/// transfer execution or fail to transfer execution in a way that is not 883/// captured in the CFG nor in the sequence of instructions within a basic 884/// block. 885/// 886/// Undefined behavior is assumed not to happen, so e.g. division is 887/// guaranteed to transfer execution to the following instruction even 888/// though division by zero might cause undefined behavior. 889bool isGuaranteedToTransferExecutionToSuccessor(const Instruction *I); 890 891/// Returns true if this block does not contain a potential implicit exit. 892/// This is equivelent to saying that all instructions within the basic block 893/// are guaranteed to transfer execution to their successor within the basic 894/// block. This has the same assumptions w.r.t. undefined behavior as the 895/// instruction variant of this function. 896bool isGuaranteedToTransferExecutionToSuccessor(const BasicBlock *BB); 897 898/// Return true if every instruction in the range (Begin, End) is 899/// guaranteed to transfer execution to its static successor. \p ScanLimit 900/// bounds the search to avoid scanning huge blocks. 901bool isGuaranteedToTransferExecutionToSuccessor( 902 BasicBlock::const_iterator Begin, BasicBlock::const_iterator End, 903 unsigned ScanLimit = 32); 904 905/// Same as previous, but with range expressed via iterator_range. 906bool isGuaranteedToTransferExecutionToSuccessor( 907 iterator_range<BasicBlock::const_iterator> Range, unsigned ScanLimit = 32); 908 909/// Return true if this function can prove that the instruction I 910/// is executed for every iteration of the loop L. 911/// 912/// Note that this currently only considers the loop header. 913bool isGuaranteedToExecuteForEveryIteration(const Instruction *I, 914 const Loop *L); 915 916/// Return true if \p PoisonOp's user yields poison or raises UB if its 917/// operand \p PoisonOp is poison. 918/// 919/// If \p PoisonOp is a vector or an aggregate and the operation's result is a 920/// single value, any poison element in /p PoisonOp should make the result 921/// poison or raise UB. 922/// 923/// To filter out operands that raise UB on poison, you can use 924/// getGuaranteedNonPoisonOp. 925bool propagatesPoison(const Use &PoisonOp); 926 927/// Insert operands of I into Ops such that I will trigger undefined behavior 928/// if I is executed and that operand has a poison value. 929void getGuaranteedNonPoisonOps(const Instruction *I, 930 SmallVectorImpl<const Value *> &Ops); 931 932/// Insert operands of I into Ops such that I will trigger undefined behavior 933/// if I is executed and that operand is not a well-defined value 934/// (i.e. has undef bits or poison). 935void getGuaranteedWellDefinedOps(const Instruction *I, 936 SmallVectorImpl<const Value *> &Ops); 937 938/// Return true if the given instruction must trigger undefined behavior 939/// when I is executed with any operands which appear in KnownPoison holding 940/// a poison value at the point of execution. 941bool mustTriggerUB(const Instruction *I, 942 const SmallPtrSetImpl<const Value *> &KnownPoison); 943 944/// Return true if this function can prove that if Inst is executed 945/// and yields a poison value or undef bits, then that will trigger 946/// undefined behavior. 947/// 948/// Note that this currently only considers the basic block that is 949/// the parent of Inst. 950bool programUndefinedIfUndefOrPoison(const Instruction *Inst); 951bool programUndefinedIfPoison(const Instruction *Inst); 952 953/// canCreateUndefOrPoison returns true if Op can create undef or poison from 954/// non-undef & non-poison operands. 955/// For vectors, canCreateUndefOrPoison returns true if there is potential 956/// poison or undef in any element of the result when vectors without 957/// undef/poison poison are given as operands. 958/// For example, given `Op = shl <2 x i32> %x, <0, 32>`, this function returns 959/// true. If Op raises immediate UB but never creates poison or undef 960/// (e.g. sdiv I, 0), canCreatePoison returns false. 961/// 962/// \p ConsiderFlagsAndMetadata controls whether poison producing flags and 963/// metadata on the instruction are considered. This can be used to see if the 964/// instruction could still introduce undef or poison even without poison 965/// generating flags and metadata which might be on the instruction. 966/// (i.e. could the result of Op->dropPoisonGeneratingFlags() still create 967/// poison or undef) 968/// 969/// canCreatePoison returns true if Op can create poison from non-poison 970/// operands. 971bool canCreateUndefOrPoison(const Operator *Op, 972 bool ConsiderFlagsAndMetadata = true); 973bool canCreatePoison(const Operator *Op, bool ConsiderFlagsAndMetadata = true); 974 975/// Return true if V is poison given that ValAssumedPoison is already poison. 976/// For example, if ValAssumedPoison is `icmp X, 10` and V is `icmp X, 5`, 977/// impliesPoison returns true. 978bool impliesPoison(const Value *ValAssumedPoison, const Value *V); 979 980/// Return true if this function can prove that V does not have undef bits 981/// and is never poison. If V is an aggregate value or vector, check whether 982/// all elements (except padding) are not undef or poison. 983/// Note that this is different from canCreateUndefOrPoison because the 984/// function assumes Op's operands are not poison/undef. 985/// 986/// If CtxI and DT are specified this method performs flow-sensitive analysis 987/// and returns true if it is guaranteed to be never undef or poison 988/// immediately before the CtxI. 989bool isGuaranteedNotToBeUndefOrPoison(const Value *V, 990 AssumptionCache *AC = nullptr, 991 const Instruction *CtxI = nullptr, 992 const DominatorTree *DT = nullptr, 993 unsigned Depth = 0); 994 995/// Returns true if V cannot be poison, but may be undef. 996bool isGuaranteedNotToBePoison(const Value *V, AssumptionCache *AC = nullptr, 997 const Instruction *CtxI = nullptr, 998 const DominatorTree *DT = nullptr, 999 unsigned Depth = 0); 1000 1001/// Returns true if V cannot be undef, but may be poison. 1002bool isGuaranteedNotToBeUndef(const Value *V, AssumptionCache *AC = nullptr, 1003 const Instruction *CtxI = nullptr, 1004 const DominatorTree *DT = nullptr, 1005 unsigned Depth = 0); 1006 1007/// Return true if undefined behavior would provable be executed on the path to 1008/// OnPathTo if Root produced a posion result. Note that this doesn't say 1009/// anything about whether OnPathTo is actually executed or whether Root is 1010/// actually poison. This can be used to assess whether a new use of Root can 1011/// be added at a location which is control equivalent with OnPathTo (such as 1012/// immediately before it) without introducing UB which didn't previously 1013/// exist. Note that a false result conveys no information. 1014bool mustExecuteUBIfPoisonOnPathTo(Instruction *Root, 1015 Instruction *OnPathTo, 1016 DominatorTree *DT); 1017 1018/// Specific patterns of select instructions we can match. 1019enum SelectPatternFlavor { 1020 SPF_UNKNOWN = 0, 1021 SPF_SMIN, /// Signed minimum 1022 SPF_UMIN, /// Unsigned minimum 1023 SPF_SMAX, /// Signed maximum 1024 SPF_UMAX, /// Unsigned maximum 1025 SPF_FMINNUM, /// Floating point minnum 1026 SPF_FMAXNUM, /// Floating point maxnum 1027 SPF_ABS, /// Absolute value 1028 SPF_NABS /// Negated absolute value 1029}; 1030 1031/// Behavior when a floating point min/max is given one NaN and one 1032/// non-NaN as input. 1033enum SelectPatternNaNBehavior { 1034 SPNB_NA = 0, /// NaN behavior not applicable. 1035 SPNB_RETURNS_NAN, /// Given one NaN input, returns the NaN. 1036 SPNB_RETURNS_OTHER, /// Given one NaN input, returns the non-NaN. 1037 SPNB_RETURNS_ANY /// Given one NaN input, can return either (or 1038 /// it has been determined that no operands can 1039 /// be NaN). 1040}; 1041 1042struct SelectPatternResult { 1043 SelectPatternFlavor Flavor; 1044 SelectPatternNaNBehavior NaNBehavior; /// Only applicable if Flavor is 1045 /// SPF_FMINNUM or SPF_FMAXNUM. 1046 bool Ordered; /// When implementing this min/max pattern as 1047 /// fcmp; select, does the fcmp have to be 1048 /// ordered? 1049 1050 /// Return true if \p SPF is a min or a max pattern. 1051 static bool isMinOrMax(SelectPatternFlavor SPF) { 1052 return SPF != SPF_UNKNOWN && SPF != SPF_ABS && SPF != SPF_NABS; 1053 } 1054}; 1055 1056/// Pattern match integer [SU]MIN, [SU]MAX and ABS idioms, returning the kind 1057/// and providing the out parameter results if we successfully match. 1058/// 1059/// For ABS/NABS, LHS will be set to the input to the abs idiom. RHS will be 1060/// the negation instruction from the idiom. 1061/// 1062/// If CastOp is not nullptr, also match MIN/MAX idioms where the type does 1063/// not match that of the original select. If this is the case, the cast 1064/// operation (one of Trunc,SExt,Zext) that must be done to transform the 1065/// type of LHS and RHS into the type of V is returned in CastOp. 1066/// 1067/// For example: 1068/// %1 = icmp slt i32 %a, i32 4 1069/// %2 = sext i32 %a to i64 1070/// %3 = select i1 %1, i64 %2, i64 4 1071/// 1072/// -> LHS = %a, RHS = i32 4, *CastOp = Instruction::SExt 1073/// 1074SelectPatternResult matchSelectPattern(Value *V, Value *&LHS, Value *&RHS, 1075 Instruction::CastOps *CastOp = nullptr, 1076 unsigned Depth = 0); 1077 1078inline SelectPatternResult matchSelectPattern(const Value *V, const Value *&LHS, 1079 const Value *&RHS) { 1080 Value *L = const_cast<Value *>(LHS); 1081 Value *R = const_cast<Value *>(RHS); 1082 auto Result = matchSelectPattern(const_cast<Value *>(V), L, R); 1083 LHS = L; 1084 RHS = R; 1085 return Result; 1086} 1087 1088/// Determine the pattern that a select with the given compare as its 1089/// predicate and given values as its true/false operands would match. 1090SelectPatternResult matchDecomposedSelectPattern( 1091 CmpInst *CmpI, Value *TrueVal, Value *FalseVal, Value *&LHS, Value *&RHS, 1092 Instruction::CastOps *CastOp = nullptr, unsigned Depth = 0); 1093 1094/// Return the canonical comparison predicate for the specified 1095/// minimum/maximum flavor. 1096CmpInst::Predicate getMinMaxPred(SelectPatternFlavor SPF, bool Ordered = false); 1097 1098/// Return the inverse minimum/maximum flavor of the specified flavor. 1099/// For example, signed minimum is the inverse of signed maximum. 1100SelectPatternFlavor getInverseMinMaxFlavor(SelectPatternFlavor SPF); 1101 1102Intrinsic::ID getInverseMinMaxIntrinsic(Intrinsic::ID MinMaxID); 1103 1104/// Return the minimum or maximum constant value for the specified integer 1105/// min/max flavor and type. 1106APInt getMinMaxLimit(SelectPatternFlavor SPF, unsigned BitWidth); 1107 1108/// Check if the values in \p VL are select instructions that can be converted 1109/// to a min or max (vector) intrinsic. Returns the intrinsic ID, if such a 1110/// conversion is possible, together with a bool indicating whether all select 1111/// conditions are only used by the selects. Otherwise return 1112/// Intrinsic::not_intrinsic. 1113std::pair<Intrinsic::ID, bool> 1114canConvertToMinOrMaxIntrinsic(ArrayRef<Value *> VL); 1115 1116/// Attempt to match a simple first order recurrence cycle of the form: 1117/// %iv = phi Ty [%Start, %Entry], [%Inc, %backedge] 1118/// %inc = binop %iv, %step 1119/// OR 1120/// %iv = phi Ty [%Start, %Entry], [%Inc, %backedge] 1121/// %inc = binop %step, %iv 1122/// 1123/// A first order recurrence is a formula with the form: X_n = f(X_(n-1)) 1124/// 1125/// A couple of notes on subtleties in that definition: 1126/// * The Step does not have to be loop invariant. In math terms, it can 1127/// be a free variable. We allow recurrences with both constant and 1128/// variable coefficients. Callers may wish to filter cases where Step 1129/// does not dominate P. 1130/// * For non-commutative operators, we will match both forms. This 1131/// results in some odd recurrence structures. Callers may wish to filter 1132/// out recurrences where the phi is not the LHS of the returned operator. 1133/// * Because of the structure matched, the caller can assume as a post 1134/// condition of the match the presence of a Loop with P's parent as it's 1135/// header *except* in unreachable code. (Dominance decays in unreachable 1136/// code.) 1137/// 1138/// NOTE: This is intentional simple. If you want the ability to analyze 1139/// non-trivial loop conditons, see ScalarEvolution instead. 1140bool matchSimpleRecurrence(const PHINode *P, BinaryOperator *&BO, Value *&Start, 1141 Value *&Step); 1142 1143/// Analogous to the above, but starting from the binary operator 1144bool matchSimpleRecurrence(const BinaryOperator *I, PHINode *&P, Value *&Start, 1145 Value *&Step); 1146 1147/// Return true if RHS is known to be implied true by LHS. Return false if 1148/// RHS is known to be implied false by LHS. Otherwise, return std::nullopt if 1149/// no implication can be made. A & B must be i1 (boolean) values or a vector of 1150/// such values. Note that the truth table for implication is the same as <=u on 1151/// i1 values (but not 1152/// <=s!). The truth table for both is: 1153/// | T | F (B) 1154/// T | T | F 1155/// F | T | T 1156/// (A) 1157std::optional<bool> isImpliedCondition(const Value *LHS, const Value *RHS, 1158 const DataLayout &DL, 1159 bool LHSIsTrue = true, 1160 unsigned Depth = 0); 1161std::optional<bool> isImpliedCondition(const Value *LHS, 1162 CmpInst::Predicate RHSPred, 1163 const Value *RHSOp0, const Value *RHSOp1, 1164 const DataLayout &DL, 1165 bool LHSIsTrue = true, 1166 unsigned Depth = 0); 1167 1168/// Return the boolean condition value in the context of the given instruction 1169/// if it is known based on dominating conditions. 1170std::optional<bool> isImpliedByDomCondition(const Value *Cond, 1171 const Instruction *ContextI, 1172 const DataLayout &DL); 1173std::optional<bool> isImpliedByDomCondition(CmpInst::Predicate Pred, 1174 const Value *LHS, const Value *RHS, 1175 const Instruction *ContextI, 1176 const DataLayout &DL); 1177} // end namespace llvm 1178 1179#endif // LLVM_ANALYSIS_VALUETRACKING_H 1180