1//===- llvm/Analysis/ScalarEvolution.h - Scalar Evolution -------*- C++ -*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// The ScalarEvolution class is an LLVM pass which can be used to analyze and 11// categorize scalar expressions in loops. It specializes in recognizing 12// general induction variables, representing them with the abstract and opaque 13// SCEV class. Given this analysis, trip counts of loops and other important 14// properties can be obtained. 15// 16// This analysis is primarily useful for induction variable substitution and 17// strength reduction. 18// 19//===----------------------------------------------------------------------===// 20 21#ifndef LLVM_ANALYSIS_SCALAREVOLUTION_H 22#define LLVM_ANALYSIS_SCALAREVOLUTION_H 23 24#include "llvm/ADT/DenseSet.h" 25#include "llvm/ADT/FoldingSet.h" 26#include "llvm/ADT/SetVector.h" 27#include "llvm/Analysis/LoopInfo.h" 28#include "llvm/IR/ConstantRange.h" 29#include "llvm/IR/Instructions.h" 30#include "llvm/IR/Operator.h" 31#include "llvm/IR/PassManager.h" 32#include "llvm/IR/ValueHandle.h" 33#include "llvm/IR/ValueMap.h" 34#include "llvm/Pass.h" 35#include "llvm/Support/Allocator.h" 36#include "llvm/Support/DataTypes.h" 37 38namespace llvm { 39 class APInt; 40 class AssumptionCache; 41 class Constant; 42 class ConstantInt; 43 class DominatorTree; 44 class Type; 45 class ScalarEvolution; 46 class DataLayout; 47 class TargetLibraryInfo; 48 class LLVMContext; 49 class Operator; 50 class SCEV; 51 class SCEVAddRecExpr; 52 class SCEVConstant; 53 class SCEVExpander; 54 class SCEVPredicate; 55 class SCEVUnknown; 56 class Function; 57 58 template <> struct FoldingSetTrait<SCEV>; 59 template <> struct FoldingSetTrait<SCEVPredicate>; 60 61 /// This class represents an analyzed expression in the program. These are 62 /// opaque objects that the client is not allowed to do much with directly. 63 /// 64 class SCEV : public FoldingSetNode { 65 friend struct FoldingSetTrait<SCEV>; 66 67 /// A reference to an Interned FoldingSetNodeID for this node. The 68 /// ScalarEvolution's BumpPtrAllocator holds the data. 69 FoldingSetNodeIDRef FastID; 70 71 // The SCEV baseclass this node corresponds to 72 const unsigned short SCEVType; 73 74 protected: 75 /// This field is initialized to zero and may be used in subclasses to store 76 /// miscellaneous information. 77 unsigned short SubclassData; 78 79 private: 80 SCEV(const SCEV &) = delete; 81 void operator=(const SCEV &) = delete; 82 83 public: 84 /// NoWrapFlags are bitfield indices into SubclassData. 85 /// 86 /// Add and Mul expressions may have no-unsigned-wrap <NUW> or 87 /// no-signed-wrap <NSW> properties, which are derived from the IR 88 /// operator. NSW is a misnomer that we use to mean no signed overflow or 89 /// underflow. 90 /// 91 /// AddRec expressions may have a no-self-wraparound <NW> property if, in 92 /// the integer domain, abs(step) * max-iteration(loop) <= 93 /// unsigned-max(bitwidth). This means that the recurrence will never reach 94 /// its start value if the step is non-zero. Computing the same value on 95 /// each iteration is not considered wrapping, and recurrences with step = 0 96 /// are trivially <NW>. <NW> is independent of the sign of step and the 97 /// value the add recurrence starts with. 98 /// 99 /// Note that NUW and NSW are also valid properties of a recurrence, and 100 /// either implies NW. For convenience, NW will be set for a recurrence 101 /// whenever either NUW or NSW are set. 102 enum NoWrapFlags { FlagAnyWrap = 0, // No guarantee. 103 FlagNW = (1 << 0), // No self-wrap. 104 FlagNUW = (1 << 1), // No unsigned wrap. 105 FlagNSW = (1 << 2), // No signed wrap. 106 NoWrapMask = (1 << 3) -1 }; 107 108 explicit SCEV(const FoldingSetNodeIDRef ID, unsigned SCEVTy) : 109 FastID(ID), SCEVType(SCEVTy), SubclassData(0) {} 110 111 unsigned getSCEVType() const { return SCEVType; } 112 113 /// Return the LLVM type of this SCEV expression. 114 /// 115 Type *getType() const; 116 117 /// Return true if the expression is a constant zero. 118 /// 119 bool isZero() const; 120 121 /// Return true if the expression is a constant one. 122 /// 123 bool isOne() const; 124 125 /// Return true if the expression is a constant all-ones value. 126 /// 127 bool isAllOnesValue() const; 128 129 /// Return true if the specified scev is negated, but not a constant. 130 bool isNonConstantNegative() const; 131 132 /// Print out the internal representation of this scalar to the specified 133 /// stream. This should really only be used for debugging purposes. 134 void print(raw_ostream &OS) const; 135 136 /// This method is used for debugging. 137 /// 138 void dump() const; 139 }; 140 141 // Specialize FoldingSetTrait for SCEV to avoid needing to compute 142 // temporary FoldingSetNodeID values. 143 template<> struct FoldingSetTrait<SCEV> : DefaultFoldingSetTrait<SCEV> { 144 static void Profile(const SCEV &X, FoldingSetNodeID& ID) { 145 ID = X.FastID; 146 } 147 static bool Equals(const SCEV &X, const FoldingSetNodeID &ID, 148 unsigned IDHash, FoldingSetNodeID &TempID) { 149 return ID == X.FastID; 150 } 151 static unsigned ComputeHash(const SCEV &X, FoldingSetNodeID &TempID) { 152 return X.FastID.ComputeHash(); 153 } 154 }; 155 156 inline raw_ostream &operator<<(raw_ostream &OS, const SCEV &S) { 157 S.print(OS); 158 return OS; 159 } 160 161 /// An object of this class is returned by queries that could not be answered. 162 /// For example, if you ask for the number of iterations of a linked-list 163 /// traversal loop, you will get one of these. None of the standard SCEV 164 /// operations are valid on this class, it is just a marker. 165 struct SCEVCouldNotCompute : public SCEV { 166 SCEVCouldNotCompute(); 167 168 /// Methods for support type inquiry through isa, cast, and dyn_cast: 169 static bool classof(const SCEV *S); 170 }; 171 172 /// This class represents an assumption made using SCEV expressions which can 173 /// be checked at run-time. 174 class SCEVPredicate : public FoldingSetNode { 175 friend struct FoldingSetTrait<SCEVPredicate>; 176 177 /// A reference to an Interned FoldingSetNodeID for this node. The 178 /// ScalarEvolution's BumpPtrAllocator holds the data. 179 FoldingSetNodeIDRef FastID; 180 181 public: 182 enum SCEVPredicateKind { P_Union, P_Equal, P_Wrap }; 183 184 protected: 185 SCEVPredicateKind Kind; 186 ~SCEVPredicate() = default; 187 SCEVPredicate(const SCEVPredicate&) = default; 188 SCEVPredicate &operator=(const SCEVPredicate&) = default; 189 190 public: 191 SCEVPredicate(const FoldingSetNodeIDRef ID, SCEVPredicateKind Kind); 192 193 SCEVPredicateKind getKind() const { return Kind; } 194 195 /// Returns the estimated complexity of this predicate. This is roughly 196 /// measured in the number of run-time checks required. 197 virtual unsigned getComplexity() const { return 1; } 198 199 /// Returns true if the predicate is always true. This means that no 200 /// assumptions were made and nothing needs to be checked at run-time. 201 virtual bool isAlwaysTrue() const = 0; 202 203 /// Returns true if this predicate implies \p N. 204 virtual bool implies(const SCEVPredicate *N) const = 0; 205 206 /// Prints a textual representation of this predicate with an indentation of 207 /// \p Depth. 208 virtual void print(raw_ostream &OS, unsigned Depth = 0) const = 0; 209 210 /// Returns the SCEV to which this predicate applies, or nullptr if this is 211 /// a SCEVUnionPredicate. 212 virtual const SCEV *getExpr() const = 0; 213 }; 214 215 inline raw_ostream &operator<<(raw_ostream &OS, const SCEVPredicate &P) { 216 P.print(OS); 217 return OS; 218 } 219 220 // Specialize FoldingSetTrait for SCEVPredicate to avoid needing to compute 221 // temporary FoldingSetNodeID values. 222 template <> 223 struct FoldingSetTrait<SCEVPredicate> 224 : DefaultFoldingSetTrait<SCEVPredicate> { 225 226 static void Profile(const SCEVPredicate &X, FoldingSetNodeID &ID) { 227 ID = X.FastID; 228 } 229 230 static bool Equals(const SCEVPredicate &X, const FoldingSetNodeID &ID, 231 unsigned IDHash, FoldingSetNodeID &TempID) { 232 return ID == X.FastID; 233 } 234 static unsigned ComputeHash(const SCEVPredicate &X, 235 FoldingSetNodeID &TempID) { 236 return X.FastID.ComputeHash(); 237 } 238 }; 239 240 /// This class represents an assumption that two SCEV expressions are equal, 241 /// and this can be checked at run-time. We assume that the left hand side is 242 /// a SCEVUnknown and the right hand side a constant. 243 class SCEVEqualPredicate final : public SCEVPredicate { 244 /// We assume that LHS == RHS, where LHS is a SCEVUnknown and RHS a 245 /// constant. 246 const SCEVUnknown *LHS; 247 const SCEVConstant *RHS; 248 249 public: 250 SCEVEqualPredicate(const FoldingSetNodeIDRef ID, const SCEVUnknown *LHS, 251 const SCEVConstant *RHS); 252 253 /// Implementation of the SCEVPredicate interface 254 bool implies(const SCEVPredicate *N) const override; 255 void print(raw_ostream &OS, unsigned Depth = 0) const override; 256 bool isAlwaysTrue() const override; 257 const SCEV *getExpr() const override; 258 259 /// Returns the left hand side of the equality. 260 const SCEVUnknown *getLHS() const { return LHS; } 261 262 /// Returns the right hand side of the equality. 263 const SCEVConstant *getRHS() const { return RHS; } 264 265 /// Methods for support type inquiry through isa, cast, and dyn_cast: 266 static inline bool classof(const SCEVPredicate *P) { 267 return P->getKind() == P_Equal; 268 } 269 }; 270 271 /// This class represents an assumption made on an AddRec expression. Given an 272 /// affine AddRec expression {a,+,b}, we assume that it has the nssw or nusw 273 /// flags (defined below) in the first X iterations of the loop, where X is a 274 /// SCEV expression returned by getPredicatedBackedgeTakenCount). 275 /// 276 /// Note that this does not imply that X is equal to the backedge taken 277 /// count. This means that if we have a nusw predicate for i32 {0,+,1} with a 278 /// predicated backedge taken count of X, we only guarantee that {0,+,1} has 279 /// nusw in the first X iterations. {0,+,1} may still wrap in the loop if we 280 /// have more than X iterations. 281 class SCEVWrapPredicate final : public SCEVPredicate { 282 public: 283 /// Similar to SCEV::NoWrapFlags, but with slightly different semantics 284 /// for FlagNUSW. The increment is considered to be signed, and a + b 285 /// (where b is the increment) is considered to wrap if: 286 /// zext(a + b) != zext(a) + sext(b) 287 /// 288 /// If Signed is a function that takes an n-bit tuple and maps to the 289 /// integer domain as the tuples value interpreted as twos complement, 290 /// and Unsigned a function that takes an n-bit tuple and maps to the 291 /// integer domain as as the base two value of input tuple, then a + b 292 /// has IncrementNUSW iff: 293 /// 294 /// 0 <= Unsigned(a) + Signed(b) < 2^n 295 /// 296 /// The IncrementNSSW flag has identical semantics with SCEV::FlagNSW. 297 /// 298 /// Note that the IncrementNUSW flag is not commutative: if base + inc 299 /// has IncrementNUSW, then inc + base doesn't neccessarily have this 300 /// property. The reason for this is that this is used for sign/zero 301 /// extending affine AddRec SCEV expressions when a SCEVWrapPredicate is 302 /// assumed. A {base,+,inc} expression is already non-commutative with 303 /// regards to base and inc, since it is interpreted as: 304 /// (((base + inc) + inc) + inc) ... 305 enum IncrementWrapFlags { 306 IncrementAnyWrap = 0, // No guarantee. 307 IncrementNUSW = (1 << 0), // No unsigned with signed increment wrap. 308 IncrementNSSW = (1 << 1), // No signed with signed increment wrap 309 // (equivalent with SCEV::NSW) 310 IncrementNoWrapMask = (1 << 2) - 1 311 }; 312 313 /// Convenient IncrementWrapFlags manipulation methods. 314 static SCEVWrapPredicate::IncrementWrapFlags LLVM_ATTRIBUTE_UNUSED_RESULT 315 clearFlags(SCEVWrapPredicate::IncrementWrapFlags Flags, 316 SCEVWrapPredicate::IncrementWrapFlags OffFlags) { 317 assert((Flags & IncrementNoWrapMask) == Flags && "Invalid flags value!"); 318 assert((OffFlags & IncrementNoWrapMask) == OffFlags && 319 "Invalid flags value!"); 320 return (SCEVWrapPredicate::IncrementWrapFlags)(Flags & ~OffFlags); 321 } 322 323 static SCEVWrapPredicate::IncrementWrapFlags LLVM_ATTRIBUTE_UNUSED_RESULT 324 maskFlags(SCEVWrapPredicate::IncrementWrapFlags Flags, int Mask) { 325 assert((Flags & IncrementNoWrapMask) == Flags && "Invalid flags value!"); 326 assert((Mask & IncrementNoWrapMask) == Mask && "Invalid mask value!"); 327 328 return (SCEVWrapPredicate::IncrementWrapFlags)(Flags & Mask); 329 } 330 331 static SCEVWrapPredicate::IncrementWrapFlags LLVM_ATTRIBUTE_UNUSED_RESULT 332 setFlags(SCEVWrapPredicate::IncrementWrapFlags Flags, 333 SCEVWrapPredicate::IncrementWrapFlags OnFlags) { 334 assert((Flags & IncrementNoWrapMask) == Flags && "Invalid flags value!"); 335 assert((OnFlags & IncrementNoWrapMask) == OnFlags && 336 "Invalid flags value!"); 337 338 return (SCEVWrapPredicate::IncrementWrapFlags)(Flags | OnFlags); 339 } 340 341 /// Returns the set of SCEVWrapPredicate no wrap flags implied by a 342 /// SCEVAddRecExpr. 343 static SCEVWrapPredicate::IncrementWrapFlags 344 getImpliedFlags(const SCEVAddRecExpr *AR, ScalarEvolution &SE); 345 346 private: 347 const SCEVAddRecExpr *AR; 348 IncrementWrapFlags Flags; 349 350 public: 351 explicit SCEVWrapPredicate(const FoldingSetNodeIDRef ID, 352 const SCEVAddRecExpr *AR, 353 IncrementWrapFlags Flags); 354 355 /// Returns the set assumed no overflow flags. 356 IncrementWrapFlags getFlags() const { return Flags; } 357 /// Implementation of the SCEVPredicate interface 358 const SCEV *getExpr() const override; 359 bool implies(const SCEVPredicate *N) const override; 360 void print(raw_ostream &OS, unsigned Depth = 0) const override; 361 bool isAlwaysTrue() const override; 362 363 /// Methods for support type inquiry through isa, cast, and dyn_cast: 364 static inline bool classof(const SCEVPredicate *P) { 365 return P->getKind() == P_Wrap; 366 } 367 }; 368 369 /// This class represents a composition of other SCEV predicates, and is the 370 /// class that most clients will interact with. This is equivalent to a 371 /// logical "AND" of all the predicates in the union. 372 class SCEVUnionPredicate final : public SCEVPredicate { 373 private: 374 typedef DenseMap<const SCEV *, SmallVector<const SCEVPredicate *, 4>> 375 PredicateMap; 376 377 /// Vector with references to all predicates in this union. 378 SmallVector<const SCEVPredicate *, 16> Preds; 379 /// Maps SCEVs to predicates for quick look-ups. 380 PredicateMap SCEVToPreds; 381 382 public: 383 SCEVUnionPredicate(); 384 385 const SmallVectorImpl<const SCEVPredicate *> &getPredicates() const { 386 return Preds; 387 } 388 389 /// Adds a predicate to this union. 390 void add(const SCEVPredicate *N); 391 392 /// Returns a reference to a vector containing all predicates which apply to 393 /// \p Expr. 394 ArrayRef<const SCEVPredicate *> getPredicatesForExpr(const SCEV *Expr); 395 396 /// Implementation of the SCEVPredicate interface 397 bool isAlwaysTrue() const override; 398 bool implies(const SCEVPredicate *N) const override; 399 void print(raw_ostream &OS, unsigned Depth) const override; 400 const SCEV *getExpr() const override; 401 402 /// We estimate the complexity of a union predicate as the size number of 403 /// predicates in the union. 404 unsigned getComplexity() const override { return Preds.size(); } 405 406 /// Methods for support type inquiry through isa, cast, and dyn_cast: 407 static inline bool classof(const SCEVPredicate *P) { 408 return P->getKind() == P_Union; 409 } 410 }; 411 412 /// The main scalar evolution driver. Because client code (intentionally) 413 /// can't do much with the SCEV objects directly, they must ask this class 414 /// for services. 415 class ScalarEvolution { 416 public: 417 /// An enum describing the relationship between a SCEV and a loop. 418 enum LoopDisposition { 419 LoopVariant, ///< The SCEV is loop-variant (unknown). 420 LoopInvariant, ///< The SCEV is loop-invariant. 421 LoopComputable ///< The SCEV varies predictably with the loop. 422 }; 423 424 /// An enum describing the relationship between a SCEV and a basic block. 425 enum BlockDisposition { 426 DoesNotDominateBlock, ///< The SCEV does not dominate the block. 427 DominatesBlock, ///< The SCEV dominates the block. 428 ProperlyDominatesBlock ///< The SCEV properly dominates the block. 429 }; 430 431 /// Convenient NoWrapFlags manipulation that hides enum casts and is 432 /// visible in the ScalarEvolution name space. 433 static SCEV::NoWrapFlags LLVM_ATTRIBUTE_UNUSED_RESULT 434 maskFlags(SCEV::NoWrapFlags Flags, int Mask) { 435 return (SCEV::NoWrapFlags)(Flags & Mask); 436 } 437 static SCEV::NoWrapFlags LLVM_ATTRIBUTE_UNUSED_RESULT 438 setFlags(SCEV::NoWrapFlags Flags, SCEV::NoWrapFlags OnFlags) { 439 return (SCEV::NoWrapFlags)(Flags | OnFlags); 440 } 441 static SCEV::NoWrapFlags LLVM_ATTRIBUTE_UNUSED_RESULT 442 clearFlags(SCEV::NoWrapFlags Flags, SCEV::NoWrapFlags OffFlags) { 443 return (SCEV::NoWrapFlags)(Flags & ~OffFlags); 444 } 445 446 private: 447 /// A CallbackVH to arrange for ScalarEvolution to be notified whenever a 448 /// Value is deleted. 449 class SCEVCallbackVH final : public CallbackVH { 450 ScalarEvolution *SE; 451 void deleted() override; 452 void allUsesReplacedWith(Value *New) override; 453 public: 454 SCEVCallbackVH(Value *V, ScalarEvolution *SE = nullptr); 455 }; 456 457 friend class SCEVCallbackVH; 458 friend class SCEVExpander; 459 friend class SCEVUnknown; 460 461 /// The function we are analyzing. 462 /// 463 Function &F; 464 465 /// Does the module have any calls to the llvm.experimental.guard intrinsic 466 /// at all? If this is false, we avoid doing work that will only help if 467 /// thare are guards present in the IR. 468 /// 469 bool HasGuards; 470 471 /// The target library information for the target we are targeting. 472 /// 473 TargetLibraryInfo &TLI; 474 475 /// The tracker for @llvm.assume intrinsics in this function. 476 AssumptionCache &AC; 477 478 /// The dominator tree. 479 /// 480 DominatorTree &DT; 481 482 /// The loop information for the function we are currently analyzing. 483 /// 484 LoopInfo &LI; 485 486 /// This SCEV is used to represent unknown trip counts and things. 487 std::unique_ptr<SCEVCouldNotCompute> CouldNotCompute; 488 489 /// The typedef for HasRecMap. 490 /// 491 typedef DenseMap<const SCEV *, bool> HasRecMapType; 492 493 /// This is a cache to record whether a SCEV contains any scAddRecExpr. 494 HasRecMapType HasRecMap; 495 496 /// The typedef for ExprValueMap. 497 /// 498 typedef DenseMap<const SCEV *, SetVector<Value *>> ExprValueMapType; 499 500 /// ExprValueMap -- This map records the original values from which 501 /// the SCEV expr is generated from. 502 ExprValueMapType ExprValueMap; 503 504 /// The typedef for ValueExprMap. 505 /// 506 typedef DenseMap<SCEVCallbackVH, const SCEV *, DenseMapInfo<Value *> > 507 ValueExprMapType; 508 509 /// This is a cache of the values we have analyzed so far. 510 /// 511 ValueExprMapType ValueExprMap; 512 513 /// Mark predicate values currently being processed by isImpliedCond. 514 DenseSet<Value*> PendingLoopPredicates; 515 516 /// Set to true by isLoopBackedgeGuardedByCond when we're walking the set of 517 /// conditions dominating the backedge of a loop. 518 bool WalkingBEDominatingConds; 519 520 /// Set to true by isKnownPredicateViaSplitting when we're trying to prove a 521 /// predicate by splitting it into a set of independent predicates. 522 bool ProvingSplitPredicate; 523 524 /// Information about the number of loop iterations for which a loop exit's 525 /// branch condition evaluates to the not-taken path. This is a temporary 526 /// pair of exact and max expressions that are eventually summarized in 527 /// ExitNotTakenInfo and BackedgeTakenInfo. 528 struct ExitLimit { 529 const SCEV *Exact; 530 const SCEV *Max; 531 532 /// A predicate union guard for this ExitLimit. The result is only 533 /// valid if this predicate evaluates to 'true' at run-time. 534 SCEVUnionPredicate Pred; 535 536 /*implicit*/ ExitLimit(const SCEV *E) : Exact(E), Max(E) {} 537 538 ExitLimit(const SCEV *E, const SCEV *M, SCEVUnionPredicate &P) 539 : Exact(E), Max(M), Pred(P) { 540 assert((isa<SCEVCouldNotCompute>(Exact) || 541 !isa<SCEVCouldNotCompute>(Max)) && 542 "Exact is not allowed to be less precise than Max"); 543 } 544 545 /// Test whether this ExitLimit contains any computed information, or 546 /// whether it's all SCEVCouldNotCompute values. 547 bool hasAnyInfo() const { 548 return !isa<SCEVCouldNotCompute>(Exact) || 549 !isa<SCEVCouldNotCompute>(Max); 550 } 551 552 /// Test whether this ExitLimit contains all information. 553 bool hasFullInfo() const { return !isa<SCEVCouldNotCompute>(Exact); } 554 }; 555 556 /// Forward declaration of ExitNotTakenExtras 557 struct ExitNotTakenExtras; 558 559 /// Information about the number of times a particular loop exit may be 560 /// reached before exiting the loop. 561 struct ExitNotTakenInfo { 562 AssertingVH<BasicBlock> ExitingBlock; 563 const SCEV *ExactNotTaken; 564 565 ExitNotTakenExtras *ExtraInfo; 566 bool Complete; 567 568 ExitNotTakenInfo() 569 : ExitingBlock(nullptr), ExactNotTaken(nullptr), ExtraInfo(nullptr), 570 Complete(true) {} 571 572 ExitNotTakenInfo(BasicBlock *ExitBlock, const SCEV *Expr, 573 ExitNotTakenExtras *Ptr) 574 : ExitingBlock(ExitBlock), ExactNotTaken(Expr), ExtraInfo(Ptr), 575 Complete(true) {} 576 577 /// Return true if all loop exits are computable. 578 bool isCompleteList() const { return Complete; } 579 580 /// Sets the incomplete property, indicating that one of the loop exits 581 /// doesn't have a corresponding ExitNotTakenInfo entry. 582 void setIncomplete() { Complete = false; } 583 584 /// Returns a pointer to the predicate associated with this information, 585 /// or nullptr if this doesn't exist (meaning always true). 586 SCEVUnionPredicate *getPred() const { 587 if (ExtraInfo) 588 return &ExtraInfo->Pred; 589 590 return nullptr; 591 } 592 593 /// Return true if the SCEV predicate associated with this information 594 /// is always true. 595 bool hasAlwaysTruePred() const { 596 return !getPred() || getPred()->isAlwaysTrue(); 597 } 598 599 /// Defines a simple forward iterator for ExitNotTakenInfo. 600 class ExitNotTakenInfoIterator 601 : public std::iterator<std::forward_iterator_tag, ExitNotTakenInfo> { 602 const ExitNotTakenInfo *Start; 603 unsigned Position; 604 605 public: 606 ExitNotTakenInfoIterator(const ExitNotTakenInfo *Start, 607 unsigned Position) 608 : Start(Start), Position(Position) {} 609 610 const ExitNotTakenInfo &operator*() const { 611 if (Position == 0) 612 return *Start; 613 614 return Start->ExtraInfo->Exits[Position - 1]; 615 } 616 617 const ExitNotTakenInfo *operator->() const { 618 if (Position == 0) 619 return Start; 620 621 return &Start->ExtraInfo->Exits[Position - 1]; 622 } 623 624 bool operator==(const ExitNotTakenInfoIterator &RHS) const { 625 return Start == RHS.Start && Position == RHS.Position; 626 } 627 628 bool operator!=(const ExitNotTakenInfoIterator &RHS) const { 629 return Start != RHS.Start || Position != RHS.Position; 630 } 631 632 ExitNotTakenInfoIterator &operator++() { // Preincrement 633 if (!Start) 634 return *this; 635 636 unsigned Elements = 637 Start->ExtraInfo ? Start->ExtraInfo->Exits.size() + 1 : 1; 638 639 ++Position; 640 641 // We've run out of elements. 642 if (Position == Elements) { 643 Start = nullptr; 644 Position = 0; 645 } 646 647 return *this; 648 } 649 ExitNotTakenInfoIterator operator++(int) { // Postincrement 650 ExitNotTakenInfoIterator Tmp = *this; 651 ++*this; 652 return Tmp; 653 } 654 }; 655 656 /// Iterators 657 ExitNotTakenInfoIterator begin() const { 658 return ExitNotTakenInfoIterator(this, 0); 659 } 660 ExitNotTakenInfoIterator end() const { 661 return ExitNotTakenInfoIterator(nullptr, 0); 662 } 663 }; 664 665 /// Describes the extra information that a ExitNotTakenInfo can have. 666 struct ExitNotTakenExtras { 667 /// The predicate associated with the ExitNotTakenInfo struct. 668 SCEVUnionPredicate Pred; 669 670 /// The extra exits in the loop. Only the ExitNotTakenExtras structure 671 /// pointed to by the first ExitNotTakenInfo struct (associated with the 672 /// first loop exit) will populate this vector to prevent having 673 /// redundant information. 674 SmallVector<ExitNotTakenInfo, 4> Exits; 675 }; 676 677 /// A struct containing the information attached to a backedge. 678 struct EdgeInfo { 679 EdgeInfo(BasicBlock *Block, const SCEV *Taken, SCEVUnionPredicate &P) : 680 ExitBlock(Block), Taken(Taken), Pred(std::move(P)) {} 681 682 /// The exit basic block. 683 BasicBlock *ExitBlock; 684 685 /// The (exact) number of time we take the edge back. 686 const SCEV *Taken; 687 688 /// The SCEV predicated associated with Taken. If Pred doesn't evaluate 689 /// to true, the information in Taken is not valid (or equivalent with 690 /// a CouldNotCompute. 691 SCEVUnionPredicate Pred; 692 }; 693 694 /// Information about the backedge-taken count of a loop. This currently 695 /// includes an exact count and a maximum count. 696 /// 697 class BackedgeTakenInfo { 698 /// A list of computable exits and their not-taken counts. Loops almost 699 /// never have more than one computable exit. 700 ExitNotTakenInfo ExitNotTaken; 701 702 /// An expression indicating the least maximum backedge-taken count of the 703 /// loop that is known, or a SCEVCouldNotCompute. This expression is only 704 /// valid if the predicates associated with all loop exits are true. 705 const SCEV *Max; 706 707 public: 708 BackedgeTakenInfo() : Max(nullptr) {} 709 710 /// Initialize BackedgeTakenInfo from a list of exact exit counts. 711 BackedgeTakenInfo(SmallVectorImpl<EdgeInfo> &ExitCounts, bool Complete, 712 const SCEV *MaxCount); 713 714 /// Test whether this BackedgeTakenInfo contains any computed information, 715 /// or whether it's all SCEVCouldNotCompute values. 716 bool hasAnyInfo() const { 717 return ExitNotTaken.ExitingBlock || !isa<SCEVCouldNotCompute>(Max); 718 } 719 720 /// Test whether this BackedgeTakenInfo contains complete information. 721 bool hasFullInfo() const { return ExitNotTaken.isCompleteList(); } 722 723 /// Return an expression indicating the exact backedge-taken count of the 724 /// loop if it is known or SCEVCouldNotCompute otherwise. This is the 725 /// number of times the loop header can be guaranteed to execute, minus 726 /// one. 727 /// 728 /// If the SCEV predicate associated with the answer can be different 729 /// from AlwaysTrue, we must add a (non null) Predicates argument. 730 /// The SCEV predicate associated with the answer will be added to 731 /// Predicates. A run-time check needs to be emitted for the SCEV 732 /// predicate in order for the answer to be valid. 733 /// 734 /// Note that we should always know if we need to pass a predicate 735 /// argument or not from the way the ExitCounts vector was computed. 736 /// If we allowed SCEV predicates to be generated when populating this 737 /// vector, this information can contain them and therefore a 738 /// SCEVPredicate argument should be added to getExact. 739 const SCEV *getExact(ScalarEvolution *SE, 740 SCEVUnionPredicate *Predicates = nullptr) const; 741 742 /// Return the number of times this loop exit may fall through to the back 743 /// edge, or SCEVCouldNotCompute. The loop is guaranteed not to exit via 744 /// this block before this number of iterations, but may exit via another 745 /// block. 746 const SCEV *getExact(BasicBlock *ExitingBlock, ScalarEvolution *SE) const; 747 748 /// Get the max backedge taken count for the loop. 749 const SCEV *getMax(ScalarEvolution *SE) const; 750 751 /// Return true if any backedge taken count expressions refer to the given 752 /// subexpression. 753 bool hasOperand(const SCEV *S, ScalarEvolution *SE) const; 754 755 /// Invalidate this result and free associated memory. 756 void clear(); 757 }; 758 759 /// Cache the backedge-taken count of the loops for this function as they 760 /// are computed. 761 DenseMap<const Loop *, BackedgeTakenInfo> BackedgeTakenCounts; 762 763 /// Cache the predicated backedge-taken count of the loops for this 764 /// function as they are computed. 765 DenseMap<const Loop *, BackedgeTakenInfo> PredicatedBackedgeTakenCounts; 766 767 /// This map contains entries for all of the PHI instructions that we 768 /// attempt to compute constant evolutions for. This allows us to avoid 769 /// potentially expensive recomputation of these properties. An instruction 770 /// maps to null if we are unable to compute its exit value. 771 DenseMap<PHINode*, Constant*> ConstantEvolutionLoopExitValue; 772 773 /// This map contains entries for all the expressions that we attempt to 774 /// compute getSCEVAtScope information for, which can be expensive in 775 /// extreme cases. 776 DenseMap<const SCEV *, 777 SmallVector<std::pair<const Loop *, const SCEV *>, 2> > ValuesAtScopes; 778 779 /// Memoized computeLoopDisposition results. 780 DenseMap<const SCEV *, 781 SmallVector<PointerIntPair<const Loop *, 2, LoopDisposition>, 2>> 782 LoopDispositions; 783 784 /// Cache for \c loopHasNoAbnormalExits. 785 DenseMap<const Loop *, bool> LoopHasNoAbnormalExits; 786 787 /// Returns true if \p L contains no instruction that can abnormally exit 788 /// the loop (i.e. via throwing an exception, by terminating the thread 789 /// cleanly or by infinite looping in a called function). Strictly 790 /// speaking, the last one is not leaving the loop, but is identical to 791 /// leaving the loop for reasoning about undefined behavior. 792 bool loopHasNoAbnormalExits(const Loop *L); 793 794 /// Compute a LoopDisposition value. 795 LoopDisposition computeLoopDisposition(const SCEV *S, const Loop *L); 796 797 /// Memoized computeBlockDisposition results. 798 DenseMap< 799 const SCEV *, 800 SmallVector<PointerIntPair<const BasicBlock *, 2, BlockDisposition>, 2>> 801 BlockDispositions; 802 803 /// Compute a BlockDisposition value. 804 BlockDisposition computeBlockDisposition(const SCEV *S, const BasicBlock *BB); 805 806 /// Memoized results from getRange 807 DenseMap<const SCEV *, ConstantRange> UnsignedRanges; 808 809 /// Memoized results from getRange 810 DenseMap<const SCEV *, ConstantRange> SignedRanges; 811 812 /// Used to parameterize getRange 813 enum RangeSignHint { HINT_RANGE_UNSIGNED, HINT_RANGE_SIGNED }; 814 815 /// Set the memoized range for the given SCEV. 816 const ConstantRange &setRange(const SCEV *S, RangeSignHint Hint, 817 const ConstantRange &CR) { 818 DenseMap<const SCEV *, ConstantRange> &Cache = 819 Hint == HINT_RANGE_UNSIGNED ? UnsignedRanges : SignedRanges; 820 821 auto Pair = Cache.insert({S, CR}); 822 if (!Pair.second) 823 Pair.first->second = CR; 824 return Pair.first->second; 825 } 826 827 /// Determine the range for a particular SCEV. 828 ConstantRange getRange(const SCEV *S, RangeSignHint Hint); 829 830 /// Determines the range for the affine SCEVAddRecExpr {\p Start,+,\p Stop}. 831 /// Helper for \c getRange. 832 ConstantRange getRangeForAffineAR(const SCEV *Start, const SCEV *Stop, 833 const SCEV *MaxBECount, 834 unsigned BitWidth); 835 836 /// Try to compute a range for the affine SCEVAddRecExpr {\p Start,+,\p 837 /// Stop} by "factoring out" a ternary expression from the add recurrence. 838 /// Helper called by \c getRange. 839 ConstantRange getRangeViaFactoring(const SCEV *Start, const SCEV *Stop, 840 const SCEV *MaxBECount, 841 unsigned BitWidth); 842 843 /// We know that there is no SCEV for the specified value. Analyze the 844 /// expression. 845 const SCEV *createSCEV(Value *V); 846 847 /// Provide the special handling we need to analyze PHI SCEVs. 848 const SCEV *createNodeForPHI(PHINode *PN); 849 850 /// Helper function called from createNodeForPHI. 851 const SCEV *createAddRecFromPHI(PHINode *PN); 852 853 /// Helper function called from createNodeForPHI. 854 const SCEV *createNodeFromSelectLikePHI(PHINode *PN); 855 856 /// Provide special handling for a select-like instruction (currently this 857 /// is either a select instruction or a phi node). \p I is the instruction 858 /// being processed, and it is assumed equivalent to "Cond ? TrueVal : 859 /// FalseVal". 860 const SCEV *createNodeForSelectOrPHI(Instruction *I, Value *Cond, 861 Value *TrueVal, Value *FalseVal); 862 863 /// Provide the special handling we need to analyze GEP SCEVs. 864 const SCEV *createNodeForGEP(GEPOperator *GEP); 865 866 /// Implementation code for getSCEVAtScope; called at most once for each 867 /// SCEV+Loop pair. 868 /// 869 const SCEV *computeSCEVAtScope(const SCEV *S, const Loop *L); 870 871 /// This looks up computed SCEV values for all instructions that depend on 872 /// the given instruction and removes them from the ValueExprMap map if they 873 /// reference SymName. This is used during PHI resolution. 874 void forgetSymbolicName(Instruction *I, const SCEV *SymName); 875 876 /// Return the BackedgeTakenInfo for the given loop, lazily computing new 877 /// values if the loop hasn't been analyzed yet. The returned result is 878 /// guaranteed not to be predicated. 879 const BackedgeTakenInfo &getBackedgeTakenInfo(const Loop *L); 880 881 /// Similar to getBackedgeTakenInfo, but will add predicates as required 882 /// with the purpose of returning complete information. 883 const BackedgeTakenInfo &getPredicatedBackedgeTakenInfo(const Loop *L); 884 885 /// Compute the number of times the specified loop will iterate. 886 /// If AllowPredicates is set, we will create new SCEV predicates as 887 /// necessary in order to return an exact answer. 888 BackedgeTakenInfo computeBackedgeTakenCount(const Loop *L, 889 bool AllowPredicates = false); 890 891 /// Compute the number of times the backedge of the specified loop will 892 /// execute if it exits via the specified block. If AllowPredicates is set, 893 /// this call will try to use a minimal set of SCEV predicates in order to 894 /// return an exact answer. 895 ExitLimit computeExitLimit(const Loop *L, BasicBlock *ExitingBlock, 896 bool AllowPredicates = false); 897 898 /// Compute the number of times the backedge of the specified loop will 899 /// execute if its exit condition were a conditional branch of ExitCond, 900 /// TBB, and FBB. 901 /// 902 /// \p ControlsExit is true if ExitCond directly controls the exit 903 /// branch. In this case, we can assume that the loop exits only if the 904 /// condition is true and can infer that failing to meet the condition prior 905 /// to integer wraparound results in undefined behavior. 906 /// 907 /// If \p AllowPredicates is set, this call will try to use a minimal set of 908 /// SCEV predicates in order to return an exact answer. 909 ExitLimit computeExitLimitFromCond(const Loop *L, 910 Value *ExitCond, 911 BasicBlock *TBB, 912 BasicBlock *FBB, 913 bool ControlsExit, 914 bool AllowPredicates = false); 915 916 /// Compute the number of times the backedge of the specified loop will 917 /// execute if its exit condition were a conditional branch of the ICmpInst 918 /// ExitCond, TBB, and FBB. If AllowPredicates is set, this call will try 919 /// to use a minimal set of SCEV predicates in order to return an exact 920 /// answer. 921 ExitLimit computeExitLimitFromICmp(const Loop *L, 922 ICmpInst *ExitCond, 923 BasicBlock *TBB, 924 BasicBlock *FBB, 925 bool IsSubExpr, 926 bool AllowPredicates = false); 927 928 /// Compute the number of times the backedge of the specified loop will 929 /// execute if its exit condition were a switch with a single exiting case 930 /// to ExitingBB. 931 ExitLimit 932 computeExitLimitFromSingleExitSwitch(const Loop *L, SwitchInst *Switch, 933 BasicBlock *ExitingBB, bool IsSubExpr); 934 935 /// Given an exit condition of 'icmp op load X, cst', try to see if we can 936 /// compute the backedge-taken count. 937 ExitLimit computeLoadConstantCompareExitLimit(LoadInst *LI, 938 Constant *RHS, 939 const Loop *L, 940 ICmpInst::Predicate p); 941 942 /// Compute the exit limit of a loop that is controlled by a 943 /// "(IV >> 1) != 0" type comparison. We cannot compute the exact trip 944 /// count in these cases (since SCEV has no way of expressing them), but we 945 /// can still sometimes compute an upper bound. 946 /// 947 /// Return an ExitLimit for a loop whose backedge is guarded by `LHS Pred 948 /// RHS`. 949 ExitLimit computeShiftCompareExitLimit(Value *LHS, Value *RHS, 950 const Loop *L, 951 ICmpInst::Predicate Pred); 952 953 /// If the loop is known to execute a constant number of times (the 954 /// condition evolves only from constants), try to evaluate a few iterations 955 /// of the loop until we get the exit condition gets a value of ExitWhen 956 /// (true or false). If we cannot evaluate the exit count of the loop, 957 /// return CouldNotCompute. 958 const SCEV *computeExitCountExhaustively(const Loop *L, 959 Value *Cond, 960 bool ExitWhen); 961 962 /// Return the number of times an exit condition comparing the specified 963 /// value to zero will execute. If not computable, return CouldNotCompute. 964 /// If AllowPredicates is set, this call will try to use a minimal set of 965 /// SCEV predicates in order to return an exact answer. 966 ExitLimit howFarToZero(const SCEV *V, const Loop *L, bool IsSubExpr, 967 bool AllowPredicates = false); 968 969 /// Return the number of times an exit condition checking the specified 970 /// value for nonzero will execute. If not computable, return 971 /// CouldNotCompute. 972 ExitLimit howFarToNonZero(const SCEV *V, const Loop *L); 973 974 /// Return the number of times an exit condition containing the specified 975 /// less-than comparison will execute. If not computable, return 976 /// CouldNotCompute. 977 /// 978 /// \p isSigned specifies whether the less-than is signed. 979 /// 980 /// \p ControlsExit is true when the LHS < RHS condition directly controls 981 /// the branch (loops exits only if condition is true). In this case, we can 982 /// use NoWrapFlags to skip overflow checks. 983 /// 984 /// If \p AllowPredicates is set, this call will try to use a minimal set of 985 /// SCEV predicates in order to return an exact answer. 986 ExitLimit howManyLessThans(const SCEV *LHS, const SCEV *RHS, const Loop *L, 987 bool isSigned, bool ControlsExit, 988 bool AllowPredicates = false); 989 990 ExitLimit howManyGreaterThans(const SCEV *LHS, const SCEV *RHS, 991 const Loop *L, bool isSigned, bool IsSubExpr, 992 bool AllowPredicates = false); 993 994 /// Return a predecessor of BB (which may not be an immediate predecessor) 995 /// which has exactly one successor from which BB is reachable, or null if 996 /// no such block is found. 997 std::pair<BasicBlock *, BasicBlock *> 998 getPredecessorWithUniqueSuccessorForBB(BasicBlock *BB); 999 1000 /// Test whether the condition described by Pred, LHS, and RHS is true 1001 /// whenever the given FoundCondValue value evaluates to true. 1002 bool isImpliedCond(ICmpInst::Predicate Pred, 1003 const SCEV *LHS, const SCEV *RHS, 1004 Value *FoundCondValue, 1005 bool Inverse); 1006 1007 /// Test whether the condition described by Pred, LHS, and RHS is true 1008 /// whenever the condition described by FoundPred, FoundLHS, FoundRHS is 1009 /// true. 1010 bool isImpliedCond(ICmpInst::Predicate Pred, const SCEV *LHS, 1011 const SCEV *RHS, ICmpInst::Predicate FoundPred, 1012 const SCEV *FoundLHS, const SCEV *FoundRHS); 1013 1014 /// Test whether the condition described by Pred, LHS, and RHS is true 1015 /// whenever the condition described by Pred, FoundLHS, and FoundRHS is 1016 /// true. 1017 bool isImpliedCondOperands(ICmpInst::Predicate Pred, 1018 const SCEV *LHS, const SCEV *RHS, 1019 const SCEV *FoundLHS, const SCEV *FoundRHS); 1020 1021 /// Test whether the condition described by Pred, LHS, and RHS is true 1022 /// whenever the condition described by Pred, FoundLHS, and FoundRHS is 1023 /// true. 1024 bool isImpliedCondOperandsHelper(ICmpInst::Predicate Pred, 1025 const SCEV *LHS, const SCEV *RHS, 1026 const SCEV *FoundLHS, 1027 const SCEV *FoundRHS); 1028 1029 /// Test whether the condition described by Pred, LHS, and RHS is true 1030 /// whenever the condition described by Pred, FoundLHS, and FoundRHS is 1031 /// true. Utility function used by isImpliedCondOperands. Tries to get 1032 /// cases like "X `sgt` 0 => X - 1 `sgt` -1". 1033 bool isImpliedCondOperandsViaRanges(ICmpInst::Predicate Pred, 1034 const SCEV *LHS, const SCEV *RHS, 1035 const SCEV *FoundLHS, 1036 const SCEV *FoundRHS); 1037 1038 /// Return true if the condition denoted by \p LHS \p Pred \p RHS is implied 1039 /// by a call to \c @llvm.experimental.guard in \p BB. 1040 bool isImpliedViaGuard(BasicBlock *BB, ICmpInst::Predicate Pred, 1041 const SCEV *LHS, const SCEV *RHS); 1042 1043 /// Test whether the condition described by Pred, LHS, and RHS is true 1044 /// whenever the condition described by Pred, FoundLHS, and FoundRHS is 1045 /// true. 1046 /// 1047 /// This routine tries to rule out certain kinds of integer overflow, and 1048 /// then tries to reason about arithmetic properties of the predicates. 1049 bool isImpliedCondOperandsViaNoOverflow(ICmpInst::Predicate Pred, 1050 const SCEV *LHS, const SCEV *RHS, 1051 const SCEV *FoundLHS, 1052 const SCEV *FoundRHS); 1053 1054 /// If we know that the specified Phi is in the header of its containing 1055 /// loop, we know the loop executes a constant number of times, and the PHI 1056 /// node is just a recurrence involving constants, fold it. 1057 Constant *getConstantEvolutionLoopExitValue(PHINode *PN, const APInt& BEs, 1058 const Loop *L); 1059 1060 /// Test if the given expression is known to satisfy the condition described 1061 /// by Pred and the known constant ranges of LHS and RHS. 1062 /// 1063 bool isKnownPredicateViaConstantRanges(ICmpInst::Predicate Pred, 1064 const SCEV *LHS, const SCEV *RHS); 1065 1066 /// Try to prove the condition described by "LHS Pred RHS" by ruling out 1067 /// integer overflow. 1068 /// 1069 /// For instance, this will return true for "A s< (A + C)<nsw>" if C is 1070 /// positive. 1071 bool isKnownPredicateViaNoOverflow(ICmpInst::Predicate Pred, 1072 const SCEV *LHS, const SCEV *RHS); 1073 1074 /// Try to split Pred LHS RHS into logical conjunctions (and's) and try to 1075 /// prove them individually. 1076 bool isKnownPredicateViaSplitting(ICmpInst::Predicate Pred, const SCEV *LHS, 1077 const SCEV *RHS); 1078 1079 /// Try to match the Expr as "(L + R)<Flags>". 1080 bool splitBinaryAdd(const SCEV *Expr, const SCEV *&L, const SCEV *&R, 1081 SCEV::NoWrapFlags &Flags); 1082 1083 /// Return true if More == (Less + C), where C is a constant. This is 1084 /// intended to be used as a cheaper substitute for full SCEV subtraction. 1085 bool computeConstantDifference(const SCEV *Less, const SCEV *More, 1086 APInt &C); 1087 1088 /// Drop memoized information computed for S. 1089 void forgetMemoizedResults(const SCEV *S); 1090 1091 /// Return an existing SCEV for V if there is one, otherwise return nullptr. 1092 const SCEV *getExistingSCEV(Value *V); 1093 1094 /// Return false iff given SCEV contains a SCEVUnknown with NULL value- 1095 /// pointer. 1096 bool checkValidity(const SCEV *S) const; 1097 1098 /// Return true if `ExtendOpTy`({`Start`,+,`Step`}) can be proved to be 1099 /// equal to {`ExtendOpTy`(`Start`),+,`ExtendOpTy`(`Step`)}. This is 1100 /// equivalent to proving no signed (resp. unsigned) wrap in 1101 /// {`Start`,+,`Step`} if `ExtendOpTy` is `SCEVSignExtendExpr` 1102 /// (resp. `SCEVZeroExtendExpr`). 1103 /// 1104 template<typename ExtendOpTy> 1105 bool proveNoWrapByVaryingStart(const SCEV *Start, const SCEV *Step, 1106 const Loop *L); 1107 1108 /// Try to prove NSW or NUW on \p AR relying on ConstantRange manipulation. 1109 SCEV::NoWrapFlags proveNoWrapViaConstantRanges(const SCEVAddRecExpr *AR); 1110 1111 bool isMonotonicPredicateImpl(const SCEVAddRecExpr *LHS, 1112 ICmpInst::Predicate Pred, bool &Increasing); 1113 1114 /// Return true if, for all loop invariant X, the predicate "LHS `Pred` X" 1115 /// is monotonically increasing or decreasing. In the former case set 1116 /// `Increasing` to true and in the latter case set `Increasing` to false. 1117 /// 1118 /// A predicate is said to be monotonically increasing if may go from being 1119 /// false to being true as the loop iterates, but never the other way 1120 /// around. A predicate is said to be monotonically decreasing if may go 1121 /// from being true to being false as the loop iterates, but never the other 1122 /// way around. 1123 bool isMonotonicPredicate(const SCEVAddRecExpr *LHS, 1124 ICmpInst::Predicate Pred, bool &Increasing); 1125 1126 /// Return SCEV no-wrap flags that can be proven based on reasoning about 1127 /// how poison produced from no-wrap flags on this value (e.g. a nuw add) 1128 /// would trigger undefined behavior on overflow. 1129 SCEV::NoWrapFlags getNoWrapFlagsFromUB(const Value *V); 1130 1131 /// Return true if the SCEV corresponding to \p I is never poison. Proving 1132 /// this is more complex than proving that just \p I is never poison, since 1133 /// SCEV commons expressions across control flow, and you can have cases 1134 /// like: 1135 /// 1136 /// idx0 = a + b; 1137 /// ptr[idx0] = 100; 1138 /// if (<condition>) { 1139 /// idx1 = a +nsw b; 1140 /// ptr[idx1] = 200; 1141 /// } 1142 /// 1143 /// where the SCEV expression (+ a b) is guaranteed to not be poison (and 1144 /// hence not sign-overflow) only if "<condition>" is true. Since both 1145 /// `idx0` and `idx1` will be mapped to the same SCEV expression, (+ a b), 1146 /// it is not okay to annotate (+ a b) with <nsw> in the above example. 1147 bool isSCEVExprNeverPoison(const Instruction *I); 1148 1149 /// This is like \c isSCEVExprNeverPoison but it specifically works for 1150 /// instructions that will get mapped to SCEV add recurrences. Return true 1151 /// if \p I will never generate poison under the assumption that \p I is an 1152 /// add recurrence on the loop \p L. 1153 bool isAddRecNeverPoison(const Instruction *I, const Loop *L); 1154 1155 public: 1156 ScalarEvolution(Function &F, TargetLibraryInfo &TLI, AssumptionCache &AC, 1157 DominatorTree &DT, LoopInfo &LI); 1158 ~ScalarEvolution(); 1159 ScalarEvolution(ScalarEvolution &&Arg); 1160 1161 LLVMContext &getContext() const { return F.getContext(); } 1162 1163 /// Test if values of the given type are analyzable within the SCEV 1164 /// framework. This primarily includes integer types, and it can optionally 1165 /// include pointer types if the ScalarEvolution class has access to 1166 /// target-specific information. 1167 bool isSCEVable(Type *Ty) const; 1168 1169 /// Return the size in bits of the specified type, for which isSCEVable must 1170 /// return true. 1171 uint64_t getTypeSizeInBits(Type *Ty) const; 1172 1173 /// Return a type with the same bitwidth as the given type and which 1174 /// represents how SCEV will treat the given type, for which isSCEVable must 1175 /// return true. For pointer types, this is the pointer-sized integer type. 1176 Type *getEffectiveSCEVType(Type *Ty) const; 1177 1178 /// Return true if the SCEV is a scAddRecExpr or it contains 1179 /// scAddRecExpr. The result will be cached in HasRecMap. 1180 /// 1181 bool containsAddRecurrence(const SCEV *S); 1182 1183 /// Return the Value set from which the SCEV expr is generated. 1184 SetVector<Value *> *getSCEVValues(const SCEV *S); 1185 1186 /// Erase Value from ValueExprMap and ExprValueMap. 1187 void eraseValueFromMap(Value *V); 1188 1189 /// Return a SCEV expression for the full generality of the specified 1190 /// expression. 1191 const SCEV *getSCEV(Value *V); 1192 1193 const SCEV *getConstant(ConstantInt *V); 1194 const SCEV *getConstant(const APInt& Val); 1195 const SCEV *getConstant(Type *Ty, uint64_t V, bool isSigned = false); 1196 const SCEV *getTruncateExpr(const SCEV *Op, Type *Ty); 1197 const SCEV *getZeroExtendExpr(const SCEV *Op, Type *Ty); 1198 const SCEV *getSignExtendExpr(const SCEV *Op, Type *Ty); 1199 const SCEV *getAnyExtendExpr(const SCEV *Op, Type *Ty); 1200 const SCEV *getAddExpr(SmallVectorImpl<const SCEV *> &Ops, 1201 SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap); 1202 const SCEV *getAddExpr(const SCEV *LHS, const SCEV *RHS, 1203 SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap) { 1204 SmallVector<const SCEV *, 2> Ops = {LHS, RHS}; 1205 return getAddExpr(Ops, Flags); 1206 } 1207 const SCEV *getAddExpr(const SCEV *Op0, const SCEV *Op1, const SCEV *Op2, 1208 SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap) { 1209 SmallVector<const SCEV *, 3> Ops = {Op0, Op1, Op2}; 1210 return getAddExpr(Ops, Flags); 1211 } 1212 const SCEV *getMulExpr(SmallVectorImpl<const SCEV *> &Ops, 1213 SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap); 1214 const SCEV *getMulExpr(const SCEV *LHS, const SCEV *RHS, 1215 SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap) { 1216 SmallVector<const SCEV *, 2> Ops = {LHS, RHS}; 1217 return getMulExpr(Ops, Flags); 1218 } 1219 const SCEV *getMulExpr(const SCEV *Op0, const SCEV *Op1, const SCEV *Op2, 1220 SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap) { 1221 SmallVector<const SCEV *, 3> Ops = {Op0, Op1, Op2}; 1222 return getMulExpr(Ops, Flags); 1223 } 1224 const SCEV *getUDivExpr(const SCEV *LHS, const SCEV *RHS); 1225 const SCEV *getUDivExactExpr(const SCEV *LHS, const SCEV *RHS); 1226 const SCEV *getAddRecExpr(const SCEV *Start, const SCEV *Step, 1227 const Loop *L, SCEV::NoWrapFlags Flags); 1228 const SCEV *getAddRecExpr(SmallVectorImpl<const SCEV *> &Operands, 1229 const Loop *L, SCEV::NoWrapFlags Flags); 1230 const SCEV *getAddRecExpr(const SmallVectorImpl<const SCEV *> &Operands, 1231 const Loop *L, SCEV::NoWrapFlags Flags) { 1232 SmallVector<const SCEV *, 4> NewOp(Operands.begin(), Operands.end()); 1233 return getAddRecExpr(NewOp, L, Flags); 1234 } 1235 /// Returns an expression for a GEP 1236 /// 1237 /// \p PointeeType The type used as the basis for the pointer arithmetics 1238 /// \p BaseExpr The expression for the pointer operand. 1239 /// \p IndexExprs The expressions for the indices. 1240 /// \p InBounds Whether the GEP is in bounds. 1241 const SCEV *getGEPExpr(Type *PointeeType, const SCEV *BaseExpr, 1242 const SmallVectorImpl<const SCEV *> &IndexExprs, 1243 bool InBounds = false); 1244 const SCEV *getSMaxExpr(const SCEV *LHS, const SCEV *RHS); 1245 const SCEV *getSMaxExpr(SmallVectorImpl<const SCEV *> &Operands); 1246 const SCEV *getUMaxExpr(const SCEV *LHS, const SCEV *RHS); 1247 const SCEV *getUMaxExpr(SmallVectorImpl<const SCEV *> &Operands); 1248 const SCEV *getSMinExpr(const SCEV *LHS, const SCEV *RHS); 1249 const SCEV *getUMinExpr(const SCEV *LHS, const SCEV *RHS); 1250 const SCEV *getUnknown(Value *V); 1251 const SCEV *getCouldNotCompute(); 1252 1253 /// Return a SCEV for the constant 0 of a specific type. 1254 const SCEV *getZero(Type *Ty) { return getConstant(Ty, 0); } 1255 1256 /// Return a SCEV for the constant 1 of a specific type. 1257 const SCEV *getOne(Type *Ty) { return getConstant(Ty, 1); } 1258 1259 /// Return an expression for sizeof AllocTy that is type IntTy 1260 /// 1261 const SCEV *getSizeOfExpr(Type *IntTy, Type *AllocTy); 1262 1263 /// Return an expression for offsetof on the given field with type IntTy 1264 /// 1265 const SCEV *getOffsetOfExpr(Type *IntTy, StructType *STy, unsigned FieldNo); 1266 1267 /// Return the SCEV object corresponding to -V. 1268 /// 1269 const SCEV *getNegativeSCEV(const SCEV *V, 1270 SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap); 1271 1272 /// Return the SCEV object corresponding to ~V. 1273 /// 1274 const SCEV *getNotSCEV(const SCEV *V); 1275 1276 /// Return LHS-RHS. Minus is represented in SCEV as A+B*-1. 1277 const SCEV *getMinusSCEV(const SCEV *LHS, const SCEV *RHS, 1278 SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap); 1279 1280 /// Return a SCEV corresponding to a conversion of the input value to the 1281 /// specified type. If the type must be extended, it is zero extended. 1282 const SCEV *getTruncateOrZeroExtend(const SCEV *V, Type *Ty); 1283 1284 /// Return a SCEV corresponding to a conversion of the input value to the 1285 /// specified type. If the type must be extended, it is sign extended. 1286 const SCEV *getTruncateOrSignExtend(const SCEV *V, Type *Ty); 1287 1288 /// Return a SCEV corresponding to a conversion of the input value to the 1289 /// specified type. If the type must be extended, it is zero extended. The 1290 /// conversion must not be narrowing. 1291 const SCEV *getNoopOrZeroExtend(const SCEV *V, Type *Ty); 1292 1293 /// Return a SCEV corresponding to a conversion of the input value to the 1294 /// specified type. If the type must be extended, it is sign extended. The 1295 /// conversion must not be narrowing. 1296 const SCEV *getNoopOrSignExtend(const SCEV *V, Type *Ty); 1297 1298 /// Return a SCEV corresponding to a conversion of the input value to the 1299 /// specified type. If the type must be extended, it is extended with 1300 /// unspecified bits. The conversion must not be narrowing. 1301 const SCEV *getNoopOrAnyExtend(const SCEV *V, Type *Ty); 1302 1303 /// Return a SCEV corresponding to a conversion of the input value to the 1304 /// specified type. The conversion must not be widening. 1305 const SCEV *getTruncateOrNoop(const SCEV *V, Type *Ty); 1306 1307 /// Promote the operands to the wider of the types using zero-extension, and 1308 /// then perform a umax operation with them. 1309 const SCEV *getUMaxFromMismatchedTypes(const SCEV *LHS, 1310 const SCEV *RHS); 1311 1312 /// Promote the operands to the wider of the types using zero-extension, and 1313 /// then perform a umin operation with them. 1314 const SCEV *getUMinFromMismatchedTypes(const SCEV *LHS, 1315 const SCEV *RHS); 1316 1317 /// Transitively follow the chain of pointer-type operands until reaching a 1318 /// SCEV that does not have a single pointer operand. This returns a 1319 /// SCEVUnknown pointer for well-formed pointer-type expressions, but corner 1320 /// cases do exist. 1321 const SCEV *getPointerBase(const SCEV *V); 1322 1323 /// Return a SCEV expression for the specified value at the specified scope 1324 /// in the program. The L value specifies a loop nest to evaluate the 1325 /// expression at, where null is the top-level or a specified loop is 1326 /// immediately inside of the loop. 1327 /// 1328 /// This method can be used to compute the exit value for a variable defined 1329 /// in a loop by querying what the value will hold in the parent loop. 1330 /// 1331 /// In the case that a relevant loop exit value cannot be computed, the 1332 /// original value V is returned. 1333 const SCEV *getSCEVAtScope(const SCEV *S, const Loop *L); 1334 1335 /// This is a convenience function which does getSCEVAtScope(getSCEV(V), L). 1336 const SCEV *getSCEVAtScope(Value *V, const Loop *L); 1337 1338 /// Test whether entry to the loop is protected by a conditional between LHS 1339 /// and RHS. This is used to help avoid max expressions in loop trip 1340 /// counts, and to eliminate casts. 1341 bool isLoopEntryGuardedByCond(const Loop *L, ICmpInst::Predicate Pred, 1342 const SCEV *LHS, const SCEV *RHS); 1343 1344 /// Test whether the backedge of the loop is protected by a conditional 1345 /// between LHS and RHS. This is used to to eliminate casts. 1346 bool isLoopBackedgeGuardedByCond(const Loop *L, ICmpInst::Predicate Pred, 1347 const SCEV *LHS, const SCEV *RHS); 1348 1349 /// Returns the maximum trip count of the loop if it is a single-exit 1350 /// loop and we can compute a small maximum for that loop. 1351 /// 1352 /// Implemented in terms of the \c getSmallConstantTripCount overload with 1353 /// the single exiting block passed to it. See that routine for details. 1354 unsigned getSmallConstantTripCount(Loop *L); 1355 1356 /// Returns the maximum trip count of this loop as a normal unsigned 1357 /// value. Returns 0 if the trip count is unknown or not constant. This 1358 /// "trip count" assumes that control exits via ExitingBlock. More 1359 /// precisely, it is the number of times that control may reach ExitingBlock 1360 /// before taking the branch. For loops with multiple exits, it may not be 1361 /// the number times that the loop header executes if the loop exits 1362 /// prematurely via another branch. 1363 unsigned getSmallConstantTripCount(Loop *L, BasicBlock *ExitingBlock); 1364 1365 /// Returns the largest constant divisor of the trip count of the 1366 /// loop if it is a single-exit loop and we can compute a small maximum for 1367 /// that loop. 1368 /// 1369 /// Implemented in terms of the \c getSmallConstantTripMultiple overload with 1370 /// the single exiting block passed to it. See that routine for details. 1371 unsigned getSmallConstantTripMultiple(Loop *L); 1372 1373 /// Returns the largest constant divisor of the trip count of this loop as a 1374 /// normal unsigned value, if possible. This means that the actual trip 1375 /// count is always a multiple of the returned value (don't forget the trip 1376 /// count could very well be zero as well!). As explained in the comments 1377 /// for getSmallConstantTripCount, this assumes that control exits the loop 1378 /// via ExitingBlock. 1379 unsigned getSmallConstantTripMultiple(Loop *L, BasicBlock *ExitingBlock); 1380 1381 /// Get the expression for the number of loop iterations for which this loop 1382 /// is guaranteed not to exit via ExitingBlock. Otherwise return 1383 /// SCEVCouldNotCompute. 1384 const SCEV *getExitCount(Loop *L, BasicBlock *ExitingBlock); 1385 1386 /// If the specified loop has a predictable backedge-taken count, return it, 1387 /// otherwise return a SCEVCouldNotCompute object. The backedge-taken count 1388 /// is the number of times the loop header will be branched to from within 1389 /// the loop. This is one less than the trip count of the loop, since it 1390 /// doesn't count the first iteration, when the header is branched to from 1391 /// outside the loop. 1392 /// 1393 /// Note that it is not valid to call this method on a loop without a 1394 /// loop-invariant backedge-taken count (see 1395 /// hasLoopInvariantBackedgeTakenCount). 1396 /// 1397 const SCEV *getBackedgeTakenCount(const Loop *L); 1398 1399 /// Similar to getBackedgeTakenCount, except it will add a set of 1400 /// SCEV predicates to Predicates that are required to be true in order for 1401 /// the answer to be correct. Predicates can be checked with run-time 1402 /// checks and can be used to perform loop versioning. 1403 const SCEV *getPredicatedBackedgeTakenCount(const Loop *L, 1404 SCEVUnionPredicate &Predicates); 1405 1406 /// Similar to getBackedgeTakenCount, except return the least SCEV value 1407 /// that is known never to be less than the actual backedge taken count. 1408 const SCEV *getMaxBackedgeTakenCount(const Loop *L); 1409 1410 /// Return true if the specified loop has an analyzable loop-invariant 1411 /// backedge-taken count. 1412 bool hasLoopInvariantBackedgeTakenCount(const Loop *L); 1413 1414 /// This method should be called by the client when it has changed a loop in 1415 /// a way that may effect ScalarEvolution's ability to compute a trip count, 1416 /// or if the loop is deleted. This call is potentially expensive for large 1417 /// loop bodies. 1418 void forgetLoop(const Loop *L); 1419 1420 /// This method should be called by the client when it has changed a value 1421 /// in a way that may effect its value, or which may disconnect it from a 1422 /// def-use chain linking it to a loop. 1423 void forgetValue(Value *V); 1424 1425 /// Called when the client has changed the disposition of values in 1426 /// this loop. 1427 /// 1428 /// We don't have a way to invalidate per-loop dispositions. Clear and 1429 /// recompute is simpler. 1430 void forgetLoopDispositions(const Loop *L) { LoopDispositions.clear(); } 1431 1432 /// Determine the minimum number of zero bits that S is guaranteed to end in 1433 /// (at every loop iteration). It is, at the same time, the minimum number 1434 /// of times S is divisible by 2. For example, given {4,+,8} it returns 2. 1435 /// If S is guaranteed to be 0, it returns the bitwidth of S. 1436 uint32_t GetMinTrailingZeros(const SCEV *S); 1437 1438 /// Determine the unsigned range for a particular SCEV. 1439 /// 1440 ConstantRange getUnsignedRange(const SCEV *S) { 1441 return getRange(S, HINT_RANGE_UNSIGNED); 1442 } 1443 1444 /// Determine the signed range for a particular SCEV. 1445 /// 1446 ConstantRange getSignedRange(const SCEV *S) { 1447 return getRange(S, HINT_RANGE_SIGNED); 1448 } 1449 1450 /// Test if the given expression is known to be negative. 1451 /// 1452 bool isKnownNegative(const SCEV *S); 1453 1454 /// Test if the given expression is known to be positive. 1455 /// 1456 bool isKnownPositive(const SCEV *S); 1457 1458 /// Test if the given expression is known to be non-negative. 1459 /// 1460 bool isKnownNonNegative(const SCEV *S); 1461 1462 /// Test if the given expression is known to be non-positive. 1463 /// 1464 bool isKnownNonPositive(const SCEV *S); 1465 1466 /// Test if the given expression is known to be non-zero. 1467 /// 1468 bool isKnownNonZero(const SCEV *S); 1469 1470 /// Test if the given expression is known to satisfy the condition described 1471 /// by Pred, LHS, and RHS. 1472 /// 1473 bool isKnownPredicate(ICmpInst::Predicate Pred, 1474 const SCEV *LHS, const SCEV *RHS); 1475 1476 /// Return true if the result of the predicate LHS `Pred` RHS is loop 1477 /// invariant with respect to L. Set InvariantPred, InvariantLHS and 1478 /// InvariantLHS so that InvariantLHS `InvariantPred` InvariantRHS is the 1479 /// loop invariant form of LHS `Pred` RHS. 1480 bool isLoopInvariantPredicate(ICmpInst::Predicate Pred, const SCEV *LHS, 1481 const SCEV *RHS, const Loop *L, 1482 ICmpInst::Predicate &InvariantPred, 1483 const SCEV *&InvariantLHS, 1484 const SCEV *&InvariantRHS); 1485 1486 /// Simplify LHS and RHS in a comparison with predicate Pred. Return true 1487 /// iff any changes were made. If the operands are provably equal or 1488 /// unequal, LHS and RHS are set to the same value and Pred is set to either 1489 /// ICMP_EQ or ICMP_NE. 1490 /// 1491 bool SimplifyICmpOperands(ICmpInst::Predicate &Pred, 1492 const SCEV *&LHS, 1493 const SCEV *&RHS, 1494 unsigned Depth = 0); 1495 1496 /// Return the "disposition" of the given SCEV with respect to the given 1497 /// loop. 1498 LoopDisposition getLoopDisposition(const SCEV *S, const Loop *L); 1499 1500 /// Return true if the value of the given SCEV is unchanging in the 1501 /// specified loop. 1502 bool isLoopInvariant(const SCEV *S, const Loop *L); 1503 1504 /// Return true if the given SCEV changes value in a known way in the 1505 /// specified loop. This property being true implies that the value is 1506 /// variant in the loop AND that we can emit an expression to compute the 1507 /// value of the expression at any particular loop iteration. 1508 bool hasComputableLoopEvolution(const SCEV *S, const Loop *L); 1509 1510 /// Return the "disposition" of the given SCEV with respect to the given 1511 /// block. 1512 BlockDisposition getBlockDisposition(const SCEV *S, const BasicBlock *BB); 1513 1514 /// Return true if elements that makes up the given SCEV dominate the 1515 /// specified basic block. 1516 bool dominates(const SCEV *S, const BasicBlock *BB); 1517 1518 /// Return true if elements that makes up the given SCEV properly dominate 1519 /// the specified basic block. 1520 bool properlyDominates(const SCEV *S, const BasicBlock *BB); 1521 1522 /// Test whether the given SCEV has Op as a direct or indirect operand. 1523 bool hasOperand(const SCEV *S, const SCEV *Op) const; 1524 1525 /// Return the size of an element read or written by Inst. 1526 const SCEV *getElementSize(Instruction *Inst); 1527 1528 /// Compute the array dimensions Sizes from the set of Terms extracted from 1529 /// the memory access function of this SCEVAddRecExpr (second step of 1530 /// delinearization). 1531 void findArrayDimensions(SmallVectorImpl<const SCEV *> &Terms, 1532 SmallVectorImpl<const SCEV *> &Sizes, 1533 const SCEV *ElementSize) const; 1534 1535 void print(raw_ostream &OS) const; 1536 void verify() const; 1537 1538 /// Collect parametric terms occurring in step expressions (first step of 1539 /// delinearization). 1540 void collectParametricTerms(const SCEV *Expr, 1541 SmallVectorImpl<const SCEV *> &Terms); 1542 1543 1544 1545 /// Return in Subscripts the access functions for each dimension in Sizes 1546 /// (third step of delinearization). 1547 void computeAccessFunctions(const SCEV *Expr, 1548 SmallVectorImpl<const SCEV *> &Subscripts, 1549 SmallVectorImpl<const SCEV *> &Sizes); 1550 1551 /// Split this SCEVAddRecExpr into two vectors of SCEVs representing the 1552 /// subscripts and sizes of an array access. 1553 /// 1554 /// The delinearization is a 3 step process: the first two steps compute the 1555 /// sizes of each subscript and the third step computes the access functions 1556 /// for the delinearized array: 1557 /// 1558 /// 1. Find the terms in the step functions 1559 /// 2. Compute the array size 1560 /// 3. Compute the access function: divide the SCEV by the array size 1561 /// starting with the innermost dimensions found in step 2. The Quotient 1562 /// is the SCEV to be divided in the next step of the recursion. The 1563 /// Remainder is the subscript of the innermost dimension. Loop over all 1564 /// array dimensions computed in step 2. 1565 /// 1566 /// To compute a uniform array size for several memory accesses to the same 1567 /// object, one can collect in step 1 all the step terms for all the memory 1568 /// accesses, and compute in step 2 a unique array shape. This guarantees 1569 /// that the array shape will be the same across all memory accesses. 1570 /// 1571 /// FIXME: We could derive the result of steps 1 and 2 from a description of 1572 /// the array shape given in metadata. 1573 /// 1574 /// Example: 1575 /// 1576 /// A[][n][m] 1577 /// 1578 /// for i 1579 /// for j 1580 /// for k 1581 /// A[j+k][2i][5i] = 1582 /// 1583 /// The initial SCEV: 1584 /// 1585 /// A[{{{0,+,2*m+5}_i, +, n*m}_j, +, n*m}_k] 1586 /// 1587 /// 1. Find the different terms in the step functions: 1588 /// -> [2*m, 5, n*m, n*m] 1589 /// 1590 /// 2. Compute the array size: sort and unique them 1591 /// -> [n*m, 2*m, 5] 1592 /// find the GCD of all the terms = 1 1593 /// divide by the GCD and erase constant terms 1594 /// -> [n*m, 2*m] 1595 /// GCD = m 1596 /// divide by GCD -> [n, 2] 1597 /// remove constant terms 1598 /// -> [n] 1599 /// size of the array is A[unknown][n][m] 1600 /// 1601 /// 3. Compute the access function 1602 /// a. Divide {{{0,+,2*m+5}_i, +, n*m}_j, +, n*m}_k by the innermost size m 1603 /// Quotient: {{{0,+,2}_i, +, n}_j, +, n}_k 1604 /// Remainder: {{{0,+,5}_i, +, 0}_j, +, 0}_k 1605 /// The remainder is the subscript of the innermost array dimension: [5i]. 1606 /// 1607 /// b. Divide Quotient: {{{0,+,2}_i, +, n}_j, +, n}_k by next outer size n 1608 /// Quotient: {{{0,+,0}_i, +, 1}_j, +, 1}_k 1609 /// Remainder: {{{0,+,2}_i, +, 0}_j, +, 0}_k 1610 /// The Remainder is the subscript of the next array dimension: [2i]. 1611 /// 1612 /// The subscript of the outermost dimension is the Quotient: [j+k]. 1613 /// 1614 /// Overall, we have: A[][n][m], and the access function: A[j+k][2i][5i]. 1615 void delinearize(const SCEV *Expr, 1616 SmallVectorImpl<const SCEV *> &Subscripts, 1617 SmallVectorImpl<const SCEV *> &Sizes, 1618 const SCEV *ElementSize); 1619 1620 /// Return the DataLayout associated with the module this SCEV instance is 1621 /// operating on. 1622 const DataLayout &getDataLayout() const { 1623 return F.getParent()->getDataLayout(); 1624 } 1625 1626 const SCEVPredicate *getEqualPredicate(const SCEVUnknown *LHS, 1627 const SCEVConstant *RHS); 1628 1629 const SCEVPredicate * 1630 getWrapPredicate(const SCEVAddRecExpr *AR, 1631 SCEVWrapPredicate::IncrementWrapFlags AddedFlags); 1632 1633 /// Re-writes the SCEV according to the Predicates in \p A. 1634 const SCEV *rewriteUsingPredicate(const SCEV *S, const Loop *L, 1635 SCEVUnionPredicate &A); 1636 /// Tries to convert the \p S expression to an AddRec expression, 1637 /// adding additional predicates to \p Preds as required. 1638 const SCEVAddRecExpr * 1639 convertSCEVToAddRecWithPredicates(const SCEV *S, const Loop *L, 1640 SCEVUnionPredicate &Preds); 1641 1642 private: 1643 /// Compute the backedge taken count knowing the interval difference, the 1644 /// stride and presence of the equality in the comparison. 1645 const SCEV *computeBECount(const SCEV *Delta, const SCEV *Stride, 1646 bool Equality); 1647 1648 /// Verify if an linear IV with positive stride can overflow when in a 1649 /// less-than comparison, knowing the invariant term of the comparison, 1650 /// the stride and the knowledge of NSW/NUW flags on the recurrence. 1651 bool doesIVOverflowOnLT(const SCEV *RHS, const SCEV *Stride, 1652 bool IsSigned, bool NoWrap); 1653 1654 /// Verify if an linear IV with negative stride can overflow when in a 1655 /// greater-than comparison, knowing the invariant term of the comparison, 1656 /// the stride and the knowledge of NSW/NUW flags on the recurrence. 1657 bool doesIVOverflowOnGT(const SCEV *RHS, const SCEV *Stride, 1658 bool IsSigned, bool NoWrap); 1659 1660 private: 1661 FoldingSet<SCEV> UniqueSCEVs; 1662 FoldingSet<SCEVPredicate> UniquePreds; 1663 BumpPtrAllocator SCEVAllocator; 1664 1665 /// The head of a linked list of all SCEVUnknown values that have been 1666 /// allocated. This is used by releaseMemory to locate them all and call 1667 /// their destructors. 1668 SCEVUnknown *FirstUnknown; 1669 }; 1670 1671 /// Analysis pass that exposes the \c ScalarEvolution for a function. 1672 class ScalarEvolutionAnalysis 1673 : public AnalysisInfoMixin<ScalarEvolutionAnalysis> { 1674 friend AnalysisInfoMixin<ScalarEvolutionAnalysis>; 1675 static char PassID; 1676 1677 public: 1678 typedef ScalarEvolution Result; 1679 1680 ScalarEvolution run(Function &F, AnalysisManager<Function> &AM); 1681 }; 1682 1683 /// Printer pass for the \c ScalarEvolutionAnalysis results. 1684 class ScalarEvolutionPrinterPass 1685 : public PassInfoMixin<ScalarEvolutionPrinterPass> { 1686 raw_ostream &OS; 1687 1688 public: 1689 explicit ScalarEvolutionPrinterPass(raw_ostream &OS) : OS(OS) {} 1690 PreservedAnalyses run(Function &F, AnalysisManager<Function> &AM); 1691 }; 1692 1693 class ScalarEvolutionWrapperPass : public FunctionPass { 1694 std::unique_ptr<ScalarEvolution> SE; 1695 1696 public: 1697 static char ID; 1698 1699 ScalarEvolutionWrapperPass(); 1700 1701 ScalarEvolution &getSE() { return *SE; } 1702 const ScalarEvolution &getSE() const { return *SE; } 1703 1704 bool runOnFunction(Function &F) override; 1705 void releaseMemory() override; 1706 void getAnalysisUsage(AnalysisUsage &AU) const override; 1707 void print(raw_ostream &OS, const Module * = nullptr) const override; 1708 void verifyAnalysis() const override; 1709 }; 1710 1711 /// An interface layer with SCEV used to manage how we see SCEV expressions 1712 /// for values in the context of existing predicates. We can add new 1713 /// predicates, but we cannot remove them. 1714 /// 1715 /// This layer has multiple purposes: 1716 /// - provides a simple interface for SCEV versioning. 1717 /// - guarantees that the order of transformations applied on a SCEV 1718 /// expression for a single Value is consistent across two different 1719 /// getSCEV calls. This means that, for example, once we've obtained 1720 /// an AddRec expression for a certain value through expression 1721 /// rewriting, we will continue to get an AddRec expression for that 1722 /// Value. 1723 /// - lowers the number of expression rewrites. 1724 class PredicatedScalarEvolution { 1725 public: 1726 PredicatedScalarEvolution(ScalarEvolution &SE, Loop &L); 1727 const SCEVUnionPredicate &getUnionPredicate() const; 1728 1729 /// Returns the SCEV expression of V, in the context of the current SCEV 1730 /// predicate. The order of transformations applied on the expression of V 1731 /// returned by ScalarEvolution is guaranteed to be preserved, even when 1732 /// adding new predicates. 1733 const SCEV *getSCEV(Value *V); 1734 1735 /// Get the (predicated) backedge count for the analyzed loop. 1736 const SCEV *getBackedgeTakenCount(); 1737 1738 /// Adds a new predicate. 1739 void addPredicate(const SCEVPredicate &Pred); 1740 1741 /// Attempts to produce an AddRecExpr for V by adding additional SCEV 1742 /// predicates. If we can't transform the expression into an AddRecExpr we 1743 /// return nullptr and not add additional SCEV predicates to the current 1744 /// context. 1745 const SCEVAddRecExpr *getAsAddRec(Value *V); 1746 1747 /// Proves that V doesn't overflow by adding SCEV predicate. 1748 void setNoOverflow(Value *V, SCEVWrapPredicate::IncrementWrapFlags Flags); 1749 1750 /// Returns true if we've proved that V doesn't wrap by means of a SCEV 1751 /// predicate. 1752 bool hasNoOverflow(Value *V, SCEVWrapPredicate::IncrementWrapFlags Flags); 1753 1754 /// Returns the ScalarEvolution analysis used. 1755 ScalarEvolution *getSE() const { return &SE; } 1756 1757 /// We need to explicitly define the copy constructor because of FlagsMap. 1758 PredicatedScalarEvolution(const PredicatedScalarEvolution&); 1759 1760 /// Print the SCEV mappings done by the Predicated Scalar Evolution. 1761 /// The printed text is indented by \p Depth. 1762 void print(raw_ostream &OS, unsigned Depth) const; 1763 1764 private: 1765 /// Increments the version number of the predicate. This needs to be called 1766 /// every time the SCEV predicate changes. 1767 void updateGeneration(); 1768 1769 /// Holds a SCEV and the version number of the SCEV predicate used to 1770 /// perform the rewrite of the expression. 1771 typedef std::pair<unsigned, const SCEV *> RewriteEntry; 1772 1773 /// Maps a SCEV to the rewrite result of that SCEV at a certain version 1774 /// number. If this number doesn't match the current Generation, we will 1775 /// need to do a rewrite. To preserve the transformation order of previous 1776 /// rewrites, we will rewrite the previous result instead of the original 1777 /// SCEV. 1778 DenseMap<const SCEV *, RewriteEntry> RewriteMap; 1779 1780 /// Records what NoWrap flags we've added to a Value *. 1781 ValueMap<Value *, SCEVWrapPredicate::IncrementWrapFlags> FlagsMap; 1782 1783 /// The ScalarEvolution analysis. 1784 ScalarEvolution &SE; 1785 1786 /// The analyzed Loop. 1787 const Loop &L; 1788 1789 /// The SCEVPredicate that forms our context. We will rewrite all 1790 /// expressions assuming that this predicate true. 1791 SCEVUnionPredicate Preds; 1792 1793 /// Marks the version of the SCEV predicate used. When rewriting a SCEV 1794 /// expression we mark it with the version of the predicate. We use this to 1795 /// figure out if the predicate has changed from the last rewrite of the 1796 /// SCEV. If so, we need to perform a new rewrite. 1797 unsigned Generation; 1798 1799 /// The backedge taken count. 1800 const SCEV *BackedgeCount; 1801 }; 1802} 1803 1804#endif 1805