1 //===- llvm/Analysis/ScalarEvolution.h - Scalar Evolution -------*- 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 // The ScalarEvolution class is an LLVM pass which can be used to analyze and
10 // categorize scalar expressions in loops.  It specializes in recognizing
11 // general induction variables, representing them with the abstract and opaque
12 // SCEV class.  Given this analysis, trip counts of loops and other important
13 // properties can be obtained.
14 //
15 // This analysis is primarily useful for induction variable substitution and
16 // strength reduction.
17 //
18 //===----------------------------------------------------------------------===//
19 
20 #ifndef LLVM_ANALYSIS_SCALAREVOLUTION_H
21 #define LLVM_ANALYSIS_SCALAREVOLUTION_H
22 
23 #include "llvm/ADT/APInt.h"
24 #include "llvm/ADT/ArrayRef.h"
25 #include "llvm/ADT/DenseMap.h"
26 #include "llvm/ADT/DenseMapInfo.h"
27 #include "llvm/ADT/FoldingSet.h"
28 #include "llvm/ADT/Hashing.h"
29 #include "llvm/ADT/Optional.h"
30 #include "llvm/ADT/PointerIntPair.h"
31 #include "llvm/ADT/SetVector.h"
32 #include "llvm/ADT/SmallPtrSet.h"
33 #include "llvm/ADT/SmallVector.h"
34 #include "llvm/IR/ConstantRange.h"
35 #include "llvm/IR/Function.h"
36 #include "llvm/IR/InstrTypes.h"
37 #include "llvm/IR/Instructions.h"
38 #include "llvm/IR/Operator.h"
39 #include "llvm/IR/PassManager.h"
40 #include "llvm/IR/ValueHandle.h"
41 #include "llvm/IR/ValueMap.h"
42 #include "llvm/Pass.h"
43 #include "llvm/Support/Allocator.h"
44 #include "llvm/Support/Casting.h"
45 #include "llvm/Support/Compiler.h"
46 #include <algorithm>
47 #include <cassert>
48 #include <cstdint>
49 #include <memory>
50 #include <utility>
51 
52 namespace llvm {
53 
54 class AssumptionCache;
55 class BasicBlock;
56 class Constant;
57 class ConstantInt;
58 class DataLayout;
59 class DominatorTree;
60 class GEPOperator;
61 class Instruction;
62 class LLVMContext;
63 class Loop;
64 class LoopInfo;
65 class raw_ostream;
66 class ScalarEvolution;
67 class SCEVAddRecExpr;
68 class SCEVUnknown;
69 class StructType;
70 class TargetLibraryInfo;
71 class Type;
72 class Value;
73 enum SCEVTypes : unsigned short;
74 
75 /// This class represents an analyzed expression in the program.  These are
76 /// opaque objects that the client is not allowed to do much with directly.
77 ///
78 class SCEV : public FoldingSetNode {
79   friend struct FoldingSetTrait<SCEV>;
80 
81   /// A reference to an Interned FoldingSetNodeID for this node.  The
82   /// ScalarEvolution's BumpPtrAllocator holds the data.
83   FoldingSetNodeIDRef FastID;
84 
85   // The SCEV baseclass this node corresponds to
86   const SCEVTypes SCEVType;
87 
88 protected:
89   // Estimated complexity of this node's expression tree size.
90   const unsigned short ExpressionSize;
91 
92   /// This field is initialized to zero and may be used in subclasses to store
93   /// miscellaneous information.
94   unsigned short SubclassData = 0;
95 
96 public:
97   /// NoWrapFlags are bitfield indices into SubclassData.
98   ///
99   /// Add and Mul expressions may have no-unsigned-wrap <NUW> or
100   /// no-signed-wrap <NSW> properties, which are derived from the IR
101   /// operator. NSW is a misnomer that we use to mean no signed overflow or
102   /// underflow.
103   ///
104   /// AddRec expressions may have a no-self-wraparound <NW> property if, in
105   /// the integer domain, abs(step) * max-iteration(loop) <=
106   /// unsigned-max(bitwidth).  This means that the recurrence will never reach
107   /// its start value if the step is non-zero.  Computing the same value on
108   /// each iteration is not considered wrapping, and recurrences with step = 0
109   /// are trivially <NW>.  <NW> is independent of the sign of step and the
110   /// value the add recurrence starts with.
111   ///
112   /// Note that NUW and NSW are also valid properties of a recurrence, and
113   /// either implies NW. For convenience, NW will be set for a recurrence
114   /// whenever either NUW or NSW are set.
115   enum NoWrapFlags {
116     FlagAnyWrap = 0,    // No guarantee.
117     FlagNW = (1 << 0),  // No self-wrap.
118     FlagNUW = (1 << 1), // No unsigned wrap.
119     FlagNSW = (1 << 2), // No signed wrap.
120     NoWrapMask = (1 << 3) - 1
121   };
122 
123   explicit SCEV(const FoldingSetNodeIDRef ID, SCEVTypes SCEVTy,
124                 unsigned short ExpressionSize)
125       : FastID(ID), SCEVType(SCEVTy), ExpressionSize(ExpressionSize) {}
126   SCEV(const SCEV &) = delete;
127   SCEV &operator=(const SCEV &) = delete;
128 
129   SCEVTypes getSCEVType() const { return SCEVType; }
130 
131   /// Return the LLVM type of this SCEV expression.
132   Type *getType() const;
133 
134   /// Return true if the expression is a constant zero.
135   bool isZero() const;
136 
137   /// Return true if the expression is a constant one.
138   bool isOne() const;
139 
140   /// Return true if the expression is a constant all-ones value.
141   bool isAllOnesValue() const;
142 
143   /// Return true if the specified scev is negated, but not a constant.
144   bool isNonConstantNegative() const;
145 
146   // Returns estimated size of the mathematical expression represented by this
147   // SCEV. The rules of its calculation are following:
148   // 1) Size of a SCEV without operands (like constants and SCEVUnknown) is 1;
149   // 2) Size SCEV with operands Op1, Op2, ..., OpN is calculated by formula:
150   //    (1 + Size(Op1) + ... + Size(OpN)).
151   // This value gives us an estimation of time we need to traverse through this
152   // SCEV and all its operands recursively. We may use it to avoid performing
153   // heavy transformations on SCEVs of excessive size for sake of saving the
154   // compilation time.
155   unsigned short getExpressionSize() const {
156     return ExpressionSize;
157   }
158 
159   /// Print out the internal representation of this scalar to the specified
160   /// stream.  This should really only be used for debugging purposes.
161   void print(raw_ostream &OS) const;
162 
163   /// This method is used for debugging.
164   void dump() const;
165 };
166 
167 // Specialize FoldingSetTrait for SCEV to avoid needing to compute
168 // temporary FoldingSetNodeID values.
169 template <> struct FoldingSetTrait<SCEV> : DefaultFoldingSetTrait<SCEV> {
170   static void Profile(const SCEV &X, FoldingSetNodeID &ID) { ID = X.FastID; }
171 
172   static bool Equals(const SCEV &X, const FoldingSetNodeID &ID, unsigned IDHash,
173                      FoldingSetNodeID &TempID) {
174     return ID == X.FastID;
175   }
176 
177   static unsigned ComputeHash(const SCEV &X, FoldingSetNodeID &TempID) {
178     return X.FastID.ComputeHash();
179   }
180 };
181 
182 inline raw_ostream &operator<<(raw_ostream &OS, const SCEV &S) {
183   S.print(OS);
184   return OS;
185 }
186 
187 /// An object of this class is returned by queries that could not be answered.
188 /// For example, if you ask for the number of iterations of a linked-list
189 /// traversal loop, you will get one of these.  None of the standard SCEV
190 /// operations are valid on this class, it is just a marker.
191 struct SCEVCouldNotCompute : public SCEV {
192   SCEVCouldNotCompute();
193 
194   /// Methods for support type inquiry through isa, cast, and dyn_cast:
195   static bool classof(const SCEV *S);
196 };
197 
198 /// This class represents an assumption made using SCEV expressions which can
199 /// be checked at run-time.
200 class SCEVPredicate : public FoldingSetNode {
201   friend struct FoldingSetTrait<SCEVPredicate>;
202 
203   /// A reference to an Interned FoldingSetNodeID for this node.  The
204   /// ScalarEvolution's BumpPtrAllocator holds the data.
205   FoldingSetNodeIDRef FastID;
206 
207 public:
208   enum SCEVPredicateKind { P_Union, P_Equal, P_Wrap };
209 
210 protected:
211   SCEVPredicateKind Kind;
212   ~SCEVPredicate() = default;
213   SCEVPredicate(const SCEVPredicate &) = default;
214   SCEVPredicate &operator=(const SCEVPredicate &) = default;
215 
216 public:
217   SCEVPredicate(const FoldingSetNodeIDRef ID, SCEVPredicateKind Kind);
218 
219   SCEVPredicateKind getKind() const { return Kind; }
220 
221   /// Returns the estimated complexity of this predicate.  This is roughly
222   /// measured in the number of run-time checks required.
223   virtual unsigned getComplexity() const { return 1; }
224 
225   /// Returns true if the predicate is always true. This means that no
226   /// assumptions were made and nothing needs to be checked at run-time.
227   virtual bool isAlwaysTrue() const = 0;
228 
229   /// Returns true if this predicate implies \p N.
230   virtual bool implies(const SCEVPredicate *N) const = 0;
231 
232   /// Prints a textual representation of this predicate with an indentation of
233   /// \p Depth.
234   virtual void print(raw_ostream &OS, unsigned Depth = 0) const = 0;
235 
236   /// Returns the SCEV to which this predicate applies, or nullptr if this is
237   /// a SCEVUnionPredicate.
238   virtual const SCEV *getExpr() const = 0;
239 };
240 
241 inline raw_ostream &operator<<(raw_ostream &OS, const SCEVPredicate &P) {
242   P.print(OS);
243   return OS;
244 }
245 
246 // Specialize FoldingSetTrait for SCEVPredicate to avoid needing to compute
247 // temporary FoldingSetNodeID values.
248 template <>
249 struct FoldingSetTrait<SCEVPredicate> : DefaultFoldingSetTrait<SCEVPredicate> {
250   static void Profile(const SCEVPredicate &X, FoldingSetNodeID &ID) {
251     ID = X.FastID;
252   }
253 
254   static bool Equals(const SCEVPredicate &X, const FoldingSetNodeID &ID,
255                      unsigned IDHash, FoldingSetNodeID &TempID) {
256     return ID == X.FastID;
257   }
258 
259   static unsigned ComputeHash(const SCEVPredicate &X,
260                               FoldingSetNodeID &TempID) {
261     return X.FastID.ComputeHash();
262   }
263 };
264 
265 /// This class represents an assumption that two SCEV expressions are equal,
266 /// and this can be checked at run-time.
267 class SCEVEqualPredicate final : public SCEVPredicate {
268   /// We assume that LHS == RHS.
269   const SCEV *LHS;
270   const SCEV *RHS;
271 
272 public:
273   SCEVEqualPredicate(const FoldingSetNodeIDRef ID, const SCEV *LHS,
274                      const SCEV *RHS);
275 
276   /// Implementation of the SCEVPredicate interface
277   bool implies(const SCEVPredicate *N) const override;
278   void print(raw_ostream &OS, unsigned Depth = 0) const override;
279   bool isAlwaysTrue() const override;
280   const SCEV *getExpr() const override;
281 
282   /// Returns the left hand side of the equality.
283   const SCEV *getLHS() const { return LHS; }
284 
285   /// Returns the right hand side of the equality.
286   const SCEV *getRHS() const { return RHS; }
287 
288   /// Methods for support type inquiry through isa, cast, and dyn_cast:
289   static bool classof(const SCEVPredicate *P) {
290     return P->getKind() == P_Equal;
291   }
292 };
293 
294 /// This class represents an assumption made on an AddRec expression. Given an
295 /// affine AddRec expression {a,+,b}, we assume that it has the nssw or nusw
296 /// flags (defined below) in the first X iterations of the loop, where X is a
297 /// SCEV expression returned by getPredicatedBackedgeTakenCount).
298 ///
299 /// Note that this does not imply that X is equal to the backedge taken
300 /// count. This means that if we have a nusw predicate for i32 {0,+,1} with a
301 /// predicated backedge taken count of X, we only guarantee that {0,+,1} has
302 /// nusw in the first X iterations. {0,+,1} may still wrap in the loop if we
303 /// have more than X iterations.
304 class SCEVWrapPredicate final : public SCEVPredicate {
305 public:
306   /// Similar to SCEV::NoWrapFlags, but with slightly different semantics
307   /// for FlagNUSW. The increment is considered to be signed, and a + b
308   /// (where b is the increment) is considered to wrap if:
309   ///    zext(a + b) != zext(a) + sext(b)
310   ///
311   /// If Signed is a function that takes an n-bit tuple and maps to the
312   /// integer domain as the tuples value interpreted as twos complement,
313   /// and Unsigned a function that takes an n-bit tuple and maps to the
314   /// integer domain as as the base two value of input tuple, then a + b
315   /// has IncrementNUSW iff:
316   ///
317   /// 0 <= Unsigned(a) + Signed(b) < 2^n
318   ///
319   /// The IncrementNSSW flag has identical semantics with SCEV::FlagNSW.
320   ///
321   /// Note that the IncrementNUSW flag is not commutative: if base + inc
322   /// has IncrementNUSW, then inc + base doesn't neccessarily have this
323   /// property. The reason for this is that this is used for sign/zero
324   /// extending affine AddRec SCEV expressions when a SCEVWrapPredicate is
325   /// assumed. A {base,+,inc} expression is already non-commutative with
326   /// regards to base and inc, since it is interpreted as:
327   ///     (((base + inc) + inc) + inc) ...
328   enum IncrementWrapFlags {
329     IncrementAnyWrap = 0,     // No guarantee.
330     IncrementNUSW = (1 << 0), // No unsigned with signed increment wrap.
331     IncrementNSSW = (1 << 1), // No signed with signed increment wrap
332                               // (equivalent with SCEV::NSW)
333     IncrementNoWrapMask = (1 << 2) - 1
334   };
335 
336   /// Convenient IncrementWrapFlags manipulation methods.
337   LLVM_NODISCARD static SCEVWrapPredicate::IncrementWrapFlags
338   clearFlags(SCEVWrapPredicate::IncrementWrapFlags Flags,
339              SCEVWrapPredicate::IncrementWrapFlags OffFlags) {
340     assert((Flags & IncrementNoWrapMask) == Flags && "Invalid flags value!");
341     assert((OffFlags & IncrementNoWrapMask) == OffFlags &&
342            "Invalid flags value!");
343     return (SCEVWrapPredicate::IncrementWrapFlags)(Flags & ~OffFlags);
344   }
345 
346   LLVM_NODISCARD static SCEVWrapPredicate::IncrementWrapFlags
347   maskFlags(SCEVWrapPredicate::IncrementWrapFlags Flags, int Mask) {
348     assert((Flags & IncrementNoWrapMask) == Flags && "Invalid flags value!");
349     assert((Mask & IncrementNoWrapMask) == Mask && "Invalid mask value!");
350 
351     return (SCEVWrapPredicate::IncrementWrapFlags)(Flags & Mask);
352   }
353 
354   LLVM_NODISCARD static SCEVWrapPredicate::IncrementWrapFlags
355   setFlags(SCEVWrapPredicate::IncrementWrapFlags Flags,
356            SCEVWrapPredicate::IncrementWrapFlags OnFlags) {
357     assert((Flags & IncrementNoWrapMask) == Flags && "Invalid flags value!");
358     assert((OnFlags & IncrementNoWrapMask) == OnFlags &&
359            "Invalid flags value!");
360 
361     return (SCEVWrapPredicate::IncrementWrapFlags)(Flags | OnFlags);
362   }
363 
364   /// Returns the set of SCEVWrapPredicate no wrap flags implied by a
365   /// SCEVAddRecExpr.
366   LLVM_NODISCARD static SCEVWrapPredicate::IncrementWrapFlags
367   getImpliedFlags(const SCEVAddRecExpr *AR, ScalarEvolution &SE);
368 
369 private:
370   const SCEVAddRecExpr *AR;
371   IncrementWrapFlags Flags;
372 
373 public:
374   explicit SCEVWrapPredicate(const FoldingSetNodeIDRef ID,
375                              const SCEVAddRecExpr *AR,
376                              IncrementWrapFlags Flags);
377 
378   /// Returns the set assumed no overflow flags.
379   IncrementWrapFlags getFlags() const { return Flags; }
380 
381   /// Implementation of the SCEVPredicate interface
382   const SCEV *getExpr() const override;
383   bool implies(const SCEVPredicate *N) const override;
384   void print(raw_ostream &OS, unsigned Depth = 0) const override;
385   bool isAlwaysTrue() const override;
386 
387   /// Methods for support type inquiry through isa, cast, and dyn_cast:
388   static bool classof(const SCEVPredicate *P) {
389     return P->getKind() == P_Wrap;
390   }
391 };
392 
393 /// This class represents a composition of other SCEV predicates, and is the
394 /// class that most clients will interact with.  This is equivalent to a
395 /// logical "AND" of all the predicates in the union.
396 ///
397 /// NB! Unlike other SCEVPredicate sub-classes this class does not live in the
398 /// ScalarEvolution::Preds folding set.  This is why the \c add function is sound.
399 class SCEVUnionPredicate final : public SCEVPredicate {
400 private:
401   using PredicateMap =
402       DenseMap<const SCEV *, SmallVector<const SCEVPredicate *, 4>>;
403 
404   /// Vector with references to all predicates in this union.
405   SmallVector<const SCEVPredicate *, 16> Preds;
406 
407   /// Maps SCEVs to predicates for quick look-ups.
408   PredicateMap SCEVToPreds;
409 
410 public:
411   SCEVUnionPredicate();
412 
413   const SmallVectorImpl<const SCEVPredicate *> &getPredicates() const {
414     return Preds;
415   }
416 
417   /// Adds a predicate to this union.
418   void add(const SCEVPredicate *N);
419 
420   /// Returns a reference to a vector containing all predicates which apply to
421   /// \p Expr.
422   ArrayRef<const SCEVPredicate *> getPredicatesForExpr(const SCEV *Expr);
423 
424   /// Implementation of the SCEVPredicate interface
425   bool isAlwaysTrue() const override;
426   bool implies(const SCEVPredicate *N) const override;
427   void print(raw_ostream &OS, unsigned Depth) const override;
428   const SCEV *getExpr() const override;
429 
430   /// We estimate the complexity of a union predicate as the size number of
431   /// predicates in the union.
432   unsigned getComplexity() const override { return Preds.size(); }
433 
434   /// Methods for support type inquiry through isa, cast, and dyn_cast:
435   static bool classof(const SCEVPredicate *P) {
436     return P->getKind() == P_Union;
437   }
438 };
439 
440 /// The main scalar evolution driver. Because client code (intentionally)
441 /// can't do much with the SCEV objects directly, they must ask this class
442 /// for services.
443 class ScalarEvolution {
444   friend class ScalarEvolutionsTest;
445 
446 public:
447   /// An enum describing the relationship between a SCEV and a loop.
448   enum LoopDisposition {
449     LoopVariant,   ///< The SCEV is loop-variant (unknown).
450     LoopInvariant, ///< The SCEV is loop-invariant.
451     LoopComputable ///< The SCEV varies predictably with the loop.
452   };
453 
454   /// An enum describing the relationship between a SCEV and a basic block.
455   enum BlockDisposition {
456     DoesNotDominateBlock,  ///< The SCEV does not dominate the block.
457     DominatesBlock,        ///< The SCEV dominates the block.
458     ProperlyDominatesBlock ///< The SCEV properly dominates the block.
459   };
460 
461   /// Convenient NoWrapFlags manipulation that hides enum casts and is
462   /// visible in the ScalarEvolution name space.
463   LLVM_NODISCARD static SCEV::NoWrapFlags maskFlags(SCEV::NoWrapFlags Flags,
464                                                     int Mask) {
465     return (SCEV::NoWrapFlags)(Flags & Mask);
466   }
467   LLVM_NODISCARD static SCEV::NoWrapFlags setFlags(SCEV::NoWrapFlags Flags,
468                                                    SCEV::NoWrapFlags OnFlags) {
469     return (SCEV::NoWrapFlags)(Flags | OnFlags);
470   }
471   LLVM_NODISCARD static SCEV::NoWrapFlags
472   clearFlags(SCEV::NoWrapFlags Flags, SCEV::NoWrapFlags OffFlags) {
473     return (SCEV::NoWrapFlags)(Flags & ~OffFlags);
474   }
475 
476   ScalarEvolution(Function &F, TargetLibraryInfo &TLI, AssumptionCache &AC,
477                   DominatorTree &DT, LoopInfo &LI);
478   ScalarEvolution(ScalarEvolution &&Arg);
479   ~ScalarEvolution();
480 
481   LLVMContext &getContext() const { return F.getContext(); }
482 
483   /// Test if values of the given type are analyzable within the SCEV
484   /// framework. This primarily includes integer types, and it can optionally
485   /// include pointer types if the ScalarEvolution class has access to
486   /// target-specific information.
487   bool isSCEVable(Type *Ty) const;
488 
489   /// Return the size in bits of the specified type, for which isSCEVable must
490   /// return true.
491   uint64_t getTypeSizeInBits(Type *Ty) const;
492 
493   /// Return a type with the same bitwidth as the given type and which
494   /// represents how SCEV will treat the given type, for which isSCEVable must
495   /// return true. For pointer types, this is the pointer-sized integer type.
496   Type *getEffectiveSCEVType(Type *Ty) const;
497 
498   // Returns a wider type among {Ty1, Ty2}.
499   Type *getWiderType(Type *Ty1, Type *Ty2) const;
500 
501   /// Return true if the SCEV is a scAddRecExpr or it contains
502   /// scAddRecExpr. The result will be cached in HasRecMap.
503   bool containsAddRecurrence(const SCEV *S);
504 
505   /// Erase Value from ValueExprMap and ExprValueMap.
506   void eraseValueFromMap(Value *V);
507 
508   /// Return a SCEV expression for the full generality of the specified
509   /// expression.
510   const SCEV *getSCEV(Value *V);
511 
512   const SCEV *getConstant(ConstantInt *V);
513   const SCEV *getConstant(const APInt &Val);
514   const SCEV *getConstant(Type *Ty, uint64_t V, bool isSigned = false);
515   const SCEV *getPtrToIntExpr(const SCEV *Op, Type *Ty, unsigned Depth = 0);
516   const SCEV *getTruncateExpr(const SCEV *Op, Type *Ty, unsigned Depth = 0);
517   const SCEV *getZeroExtendExpr(const SCEV *Op, Type *Ty, unsigned Depth = 0);
518   const SCEV *getSignExtendExpr(const SCEV *Op, Type *Ty, unsigned Depth = 0);
519   const SCEV *getAnyExtendExpr(const SCEV *Op, Type *Ty);
520   const SCEV *getAddExpr(SmallVectorImpl<const SCEV *> &Ops,
521                          SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap,
522                          unsigned Depth = 0);
523   const SCEV *getAddExpr(const SCEV *LHS, const SCEV *RHS,
524                          SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap,
525                          unsigned Depth = 0) {
526     SmallVector<const SCEV *, 2> Ops = {LHS, RHS};
527     return getAddExpr(Ops, Flags, Depth);
528   }
529   const SCEV *getAddExpr(const SCEV *Op0, const SCEV *Op1, const SCEV *Op2,
530                          SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap,
531                          unsigned Depth = 0) {
532     SmallVector<const SCEV *, 3> Ops = {Op0, Op1, Op2};
533     return getAddExpr(Ops, Flags, Depth);
534   }
535   const SCEV *getMulExpr(SmallVectorImpl<const SCEV *> &Ops,
536                          SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap,
537                          unsigned Depth = 0);
538   const SCEV *getMulExpr(const SCEV *LHS, const SCEV *RHS,
539                          SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap,
540                          unsigned Depth = 0) {
541     SmallVector<const SCEV *, 2> Ops = {LHS, RHS};
542     return getMulExpr(Ops, Flags, Depth);
543   }
544   const SCEV *getMulExpr(const SCEV *Op0, const SCEV *Op1, const SCEV *Op2,
545                          SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap,
546                          unsigned Depth = 0) {
547     SmallVector<const SCEV *, 3> Ops = {Op0, Op1, Op2};
548     return getMulExpr(Ops, Flags, Depth);
549   }
550   const SCEV *getUDivExpr(const SCEV *LHS, const SCEV *RHS);
551   const SCEV *getUDivExactExpr(const SCEV *LHS, const SCEV *RHS);
552   const SCEV *getURemExpr(const SCEV *LHS, const SCEV *RHS);
553   const SCEV *getAddRecExpr(const SCEV *Start, const SCEV *Step, const Loop *L,
554                             SCEV::NoWrapFlags Flags);
555   const SCEV *getAddRecExpr(SmallVectorImpl<const SCEV *> &Operands,
556                             const Loop *L, SCEV::NoWrapFlags Flags);
557   const SCEV *getAddRecExpr(const SmallVectorImpl<const SCEV *> &Operands,
558                             const Loop *L, SCEV::NoWrapFlags Flags) {
559     SmallVector<const SCEV *, 4> NewOp(Operands.begin(), Operands.end());
560     return getAddRecExpr(NewOp, L, Flags);
561   }
562 
563   /// Checks if \p SymbolicPHI can be rewritten as an AddRecExpr under some
564   /// Predicates. If successful return these <AddRecExpr, Predicates>;
565   /// The function is intended to be called from PSCEV (the caller will decide
566   /// whether to actually add the predicates and carry out the rewrites).
567   Optional<std::pair<const SCEV *, SmallVector<const SCEVPredicate *, 3>>>
568   createAddRecFromPHIWithCasts(const SCEVUnknown *SymbolicPHI);
569 
570   /// Returns an expression for a GEP
571   ///
572   /// \p GEP The GEP. The indices contained in the GEP itself are ignored,
573   /// instead we use IndexExprs.
574   /// \p IndexExprs The expressions for the indices.
575   const SCEV *getGEPExpr(GEPOperator *GEP,
576                          const SmallVectorImpl<const SCEV *> &IndexExprs);
577   const SCEV *getAbsExpr(const SCEV *Op, bool IsNSW);
578   const SCEV *getSignumExpr(const SCEV *Op);
579   const SCEV *getMinMaxExpr(SCEVTypes Kind,
580                             SmallVectorImpl<const SCEV *> &Operands);
581   const SCEV *getSMaxExpr(const SCEV *LHS, const SCEV *RHS);
582   const SCEV *getSMaxExpr(SmallVectorImpl<const SCEV *> &Operands);
583   const SCEV *getUMaxExpr(const SCEV *LHS, const SCEV *RHS);
584   const SCEV *getUMaxExpr(SmallVectorImpl<const SCEV *> &Operands);
585   const SCEV *getSMinExpr(const SCEV *LHS, const SCEV *RHS);
586   const SCEV *getSMinExpr(SmallVectorImpl<const SCEV *> &Operands);
587   const SCEV *getUMinExpr(const SCEV *LHS, const SCEV *RHS);
588   const SCEV *getUMinExpr(SmallVectorImpl<const SCEV *> &Operands);
589   const SCEV *getUnknown(Value *V);
590   const SCEV *getCouldNotCompute();
591 
592   /// Return a SCEV for the constant 0 of a specific type.
593   const SCEV *getZero(Type *Ty) { return getConstant(Ty, 0); }
594 
595   /// Return a SCEV for the constant 1 of a specific type.
596   const SCEV *getOne(Type *Ty) { return getConstant(Ty, 1); }
597 
598   /// Return a SCEV for the constant -1 of a specific type.
599   const SCEV *getMinusOne(Type *Ty) {
600     return getConstant(Ty, -1, /*isSigned=*/true);
601   }
602 
603   /// Return an expression for sizeof ScalableTy that is type IntTy, where
604   /// ScalableTy is a scalable vector type.
605   const SCEV *getSizeOfScalableVectorExpr(Type *IntTy,
606                                           ScalableVectorType *ScalableTy);
607 
608   /// Return an expression for the alloc size of AllocTy that is type IntTy
609   const SCEV *getSizeOfExpr(Type *IntTy, Type *AllocTy);
610 
611   /// Return an expression for the store size of StoreTy that is type IntTy
612   const SCEV *getStoreSizeOfExpr(Type *IntTy, Type *StoreTy);
613 
614   /// Return an expression for offsetof on the given field with type IntTy
615   const SCEV *getOffsetOfExpr(Type *IntTy, StructType *STy, unsigned FieldNo);
616 
617   /// Return the SCEV object corresponding to -V.
618   const SCEV *getNegativeSCEV(const SCEV *V,
619                               SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap);
620 
621   /// Return the SCEV object corresponding to ~V.
622   const SCEV *getNotSCEV(const SCEV *V);
623 
624   /// Return LHS-RHS.  Minus is represented in SCEV as A+B*-1.
625   const SCEV *getMinusSCEV(const SCEV *LHS, const SCEV *RHS,
626                            SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap,
627                            unsigned Depth = 0);
628 
629   /// Return a SCEV corresponding to a conversion of the input value to the
630   /// specified type.  If the type must be extended, it is zero extended.
631   const SCEV *getTruncateOrZeroExtend(const SCEV *V, Type *Ty,
632                                       unsigned Depth = 0);
633 
634   /// Return a SCEV corresponding to a conversion of the input value to the
635   /// specified type.  If the type must be extended, it is sign extended.
636   const SCEV *getTruncateOrSignExtend(const SCEV *V, Type *Ty,
637                                       unsigned Depth = 0);
638 
639   /// Return a SCEV corresponding to a conversion of the input value to the
640   /// specified type.  If the type must be extended, it is zero extended.  The
641   /// conversion must not be narrowing.
642   const SCEV *getNoopOrZeroExtend(const SCEV *V, Type *Ty);
643 
644   /// Return a SCEV corresponding to a conversion of the input value to the
645   /// specified type.  If the type must be extended, it is sign extended.  The
646   /// conversion must not be narrowing.
647   const SCEV *getNoopOrSignExtend(const SCEV *V, Type *Ty);
648 
649   /// Return a SCEV corresponding to a conversion of the input value to the
650   /// specified type. If the type must be extended, it is extended with
651   /// unspecified bits. The conversion must not be narrowing.
652   const SCEV *getNoopOrAnyExtend(const SCEV *V, Type *Ty);
653 
654   /// Return a SCEV corresponding to a conversion of the input value to the
655   /// specified type.  The conversion must not be widening.
656   const SCEV *getTruncateOrNoop(const SCEV *V, Type *Ty);
657 
658   /// Promote the operands to the wider of the types using zero-extension, and
659   /// then perform a umax operation with them.
660   const SCEV *getUMaxFromMismatchedTypes(const SCEV *LHS, const SCEV *RHS);
661 
662   /// Promote the operands to the wider of the types using zero-extension, and
663   /// then perform a umin operation with them.
664   const SCEV *getUMinFromMismatchedTypes(const SCEV *LHS, const SCEV *RHS);
665 
666   /// Promote the operands to the wider of the types using zero-extension, and
667   /// then perform a umin operation with them. N-ary function.
668   const SCEV *getUMinFromMismatchedTypes(SmallVectorImpl<const SCEV *> &Ops);
669 
670   /// Transitively follow the chain of pointer-type operands until reaching a
671   /// SCEV that does not have a single pointer operand. This returns a
672   /// SCEVUnknown pointer for well-formed pointer-type expressions, but corner
673   /// cases do exist.
674   const SCEV *getPointerBase(const SCEV *V);
675 
676   /// Return a SCEV expression for the specified value at the specified scope
677   /// in the program.  The L value specifies a loop nest to evaluate the
678   /// expression at, where null is the top-level or a specified loop is
679   /// immediately inside of the loop.
680   ///
681   /// This method can be used to compute the exit value for a variable defined
682   /// in a loop by querying what the value will hold in the parent loop.
683   ///
684   /// In the case that a relevant loop exit value cannot be computed, the
685   /// original value V is returned.
686   const SCEV *getSCEVAtScope(const SCEV *S, const Loop *L);
687 
688   /// This is a convenience function which does getSCEVAtScope(getSCEV(V), L).
689   const SCEV *getSCEVAtScope(Value *V, const Loop *L);
690 
691   /// Test whether entry to the loop is protected by a conditional between LHS
692   /// and RHS.  This is used to help avoid max expressions in loop trip
693   /// counts, and to eliminate casts.
694   bool isLoopEntryGuardedByCond(const Loop *L, ICmpInst::Predicate Pred,
695                                 const SCEV *LHS, const SCEV *RHS);
696 
697   /// Test whether entry to the basic block is protected by a conditional
698   /// between LHS and RHS.
699   bool isBasicBlockEntryGuardedByCond(const BasicBlock *BB,
700                                       ICmpInst::Predicate Pred, const SCEV *LHS,
701                                       const SCEV *RHS);
702 
703   /// Test whether the backedge of the loop is protected by a conditional
704   /// between LHS and RHS.  This is used to eliminate casts.
705   bool isLoopBackedgeGuardedByCond(const Loop *L, ICmpInst::Predicate Pred,
706                                    const SCEV *LHS, const SCEV *RHS);
707 
708   /// Returns the maximum trip count of the loop if it is a single-exit
709   /// loop and we can compute a small maximum for that loop.
710   ///
711   /// Implemented in terms of the \c getSmallConstantTripCount overload with
712   /// the single exiting block passed to it. See that routine for details.
713   unsigned getSmallConstantTripCount(const Loop *L);
714 
715   /// Returns the maximum trip count of this loop as a normal unsigned
716   /// value. Returns 0 if the trip count is unknown or not constant. This
717   /// "trip count" assumes that control exits via ExitingBlock. More
718   /// precisely, it is the number of times that control may reach ExitingBlock
719   /// before taking the branch. For loops with multiple exits, it may not be
720   /// the number times that the loop header executes if the loop exits
721   /// prematurely via another branch.
722   unsigned getSmallConstantTripCount(const Loop *L,
723                                      const BasicBlock *ExitingBlock);
724 
725   /// Returns the upper bound of the loop trip count as a normal unsigned
726   /// value.
727   /// Returns 0 if the trip count is unknown or not constant.
728   unsigned getSmallConstantMaxTripCount(const Loop *L);
729 
730   /// Returns the largest constant divisor of the trip count of the
731   /// loop if it is a single-exit loop and we can compute a small maximum for
732   /// that loop.
733   ///
734   /// Implemented in terms of the \c getSmallConstantTripMultiple overload with
735   /// the single exiting block passed to it. See that routine for details.
736   unsigned getSmallConstantTripMultiple(const Loop *L);
737 
738   /// Returns the largest constant divisor of the trip count of this loop as a
739   /// normal unsigned value, if possible. This means that the actual trip
740   /// count is always a multiple of the returned value (don't forget the trip
741   /// count could very well be zero as well!). As explained in the comments
742   /// for getSmallConstantTripCount, this assumes that control exits the loop
743   /// via ExitingBlock.
744   unsigned getSmallConstantTripMultiple(const Loop *L,
745                                         const BasicBlock *ExitingBlock);
746 
747   /// The terms "backedge taken count" and "exit count" are used
748   /// interchangeably to refer to the number of times the backedge of a loop
749   /// has executed before the loop is exited.
750   enum ExitCountKind {
751     /// An expression exactly describing the number of times the backedge has
752     /// executed when a loop is exited.
753     Exact,
754     /// A constant which provides an upper bound on the exact trip count.
755     ConstantMaximum,
756     /// An expression which provides an upper bound on the exact trip count.
757     SymbolicMaximum,
758   };
759 
760   /// Return the number of times the backedge executes before the given exit
761   /// would be taken; if not exactly computable, return SCEVCouldNotCompute.
762   /// For a single exit loop, this value is equivelent to the result of
763   /// getBackedgeTakenCount.  The loop is guaranteed to exit (via *some* exit)
764   /// before the backedge is executed (ExitCount + 1) times.  Note that there
765   /// is no guarantee about *which* exit is taken on the exiting iteration.
766   const SCEV *getExitCount(const Loop *L, const BasicBlock *ExitingBlock,
767                            ExitCountKind Kind = Exact);
768 
769   /// If the specified loop has a predictable backedge-taken count, return it,
770   /// otherwise return a SCEVCouldNotCompute object. The backedge-taken count is
771   /// the number of times the loop header will be branched to from within the
772   /// loop, assuming there are no abnormal exists like exception throws. This is
773   /// one less than the trip count of the loop, since it doesn't count the first
774   /// iteration, when the header is branched to from outside the loop.
775   ///
776   /// Note that it is not valid to call this method on a loop without a
777   /// loop-invariant backedge-taken count (see
778   /// hasLoopInvariantBackedgeTakenCount).
779   const SCEV *getBackedgeTakenCount(const Loop *L, ExitCountKind Kind = Exact);
780 
781   /// Similar to getBackedgeTakenCount, except it will add a set of
782   /// SCEV predicates to Predicates that are required to be true in order for
783   /// the answer to be correct. Predicates can be checked with run-time
784   /// checks and can be used to perform loop versioning.
785   const SCEV *getPredicatedBackedgeTakenCount(const Loop *L,
786                                               SCEVUnionPredicate &Predicates);
787 
788   /// When successful, this returns a SCEVConstant that is greater than or equal
789   /// to (i.e. a "conservative over-approximation") of the value returend by
790   /// getBackedgeTakenCount.  If such a value cannot be computed, it returns the
791   /// SCEVCouldNotCompute object.
792   const SCEV *getConstantMaxBackedgeTakenCount(const Loop *L) {
793     return getBackedgeTakenCount(L, ConstantMaximum);
794   }
795 
796   /// When successful, this returns a SCEV that is greater than or equal
797   /// to (i.e. a "conservative over-approximation") of the value returend by
798   /// getBackedgeTakenCount.  If such a value cannot be computed, it returns the
799   /// SCEVCouldNotCompute object.
800   const SCEV *getSymbolicMaxBackedgeTakenCount(const Loop *L) {
801     return getBackedgeTakenCount(L, SymbolicMaximum);
802   }
803 
804   /// Return true if the backedge taken count is either the value returned by
805   /// getConstantMaxBackedgeTakenCount or zero.
806   bool isBackedgeTakenCountMaxOrZero(const Loop *L);
807 
808   /// Return true if the specified loop has an analyzable loop-invariant
809   /// backedge-taken count.
810   bool hasLoopInvariantBackedgeTakenCount(const Loop *L);
811 
812   // This method should be called by the client when it made any change that
813   // would invalidate SCEV's answers, and the client wants to remove all loop
814   // information held internally by ScalarEvolution. This is intended to be used
815   // when the alternative to forget a loop is too expensive (i.e. large loop
816   // bodies).
817   void forgetAllLoops();
818 
819   /// This method should be called by the client when it has changed a loop in
820   /// a way that may effect ScalarEvolution's ability to compute a trip count,
821   /// or if the loop is deleted.  This call is potentially expensive for large
822   /// loop bodies.
823   void forgetLoop(const Loop *L);
824 
825   // This method invokes forgetLoop for the outermost loop of the given loop
826   // \p L, making ScalarEvolution forget about all this subtree. This needs to
827   // be done whenever we make a transform that may affect the parameters of the
828   // outer loop, such as exit counts for branches.
829   void forgetTopmostLoop(const Loop *L);
830 
831   /// This method should be called by the client when it has changed a value
832   /// in a way that may effect its value, or which may disconnect it from a
833   /// def-use chain linking it to a loop.
834   void forgetValue(Value *V);
835 
836   /// Called when the client has changed the disposition of values in
837   /// this loop.
838   ///
839   /// We don't have a way to invalidate per-loop dispositions. Clear and
840   /// recompute is simpler.
841   void forgetLoopDispositions(const Loop *L);
842 
843   /// Determine the minimum number of zero bits that S is guaranteed to end in
844   /// (at every loop iteration).  It is, at the same time, the minimum number
845   /// of times S is divisible by 2.  For example, given {4,+,8} it returns 2.
846   /// If S is guaranteed to be 0, it returns the bitwidth of S.
847   uint32_t GetMinTrailingZeros(const SCEV *S);
848 
849   /// Determine the unsigned range for a particular SCEV.
850   /// NOTE: This returns a copy of the reference returned by getRangeRef.
851   ConstantRange getUnsignedRange(const SCEV *S) {
852     return getRangeRef(S, HINT_RANGE_UNSIGNED);
853   }
854 
855   /// Determine the min of the unsigned range for a particular SCEV.
856   APInt getUnsignedRangeMin(const SCEV *S) {
857     return getRangeRef(S, HINT_RANGE_UNSIGNED).getUnsignedMin();
858   }
859 
860   /// Determine the max of the unsigned range for a particular SCEV.
861   APInt getUnsignedRangeMax(const SCEV *S) {
862     return getRangeRef(S, HINT_RANGE_UNSIGNED).getUnsignedMax();
863   }
864 
865   /// Determine the signed range for a particular SCEV.
866   /// NOTE: This returns a copy of the reference returned by getRangeRef.
867   ConstantRange getSignedRange(const SCEV *S) {
868     return getRangeRef(S, HINT_RANGE_SIGNED);
869   }
870 
871   /// Determine the min of the signed range for a particular SCEV.
872   APInt getSignedRangeMin(const SCEV *S) {
873     return getRangeRef(S, HINT_RANGE_SIGNED).getSignedMin();
874   }
875 
876   /// Determine the max of the signed range for a particular SCEV.
877   APInt getSignedRangeMax(const SCEV *S) {
878     return getRangeRef(S, HINT_RANGE_SIGNED).getSignedMax();
879   }
880 
881   /// Test if the given expression is known to be negative.
882   bool isKnownNegative(const SCEV *S);
883 
884   /// Test if the given expression is known to be positive.
885   bool isKnownPositive(const SCEV *S);
886 
887   /// Test if the given expression is known to be non-negative.
888   bool isKnownNonNegative(const SCEV *S);
889 
890   /// Test if the given expression is known to be non-positive.
891   bool isKnownNonPositive(const SCEV *S);
892 
893   /// Test if the given expression is known to be non-zero.
894   bool isKnownNonZero(const SCEV *S);
895 
896   /// Splits SCEV expression \p S into two SCEVs. One of them is obtained from
897   /// \p S by substitution of all AddRec sub-expression related to loop \p L
898   /// with initial value of that SCEV. The second is obtained from \p S by
899   /// substitution of all AddRec sub-expressions related to loop \p L with post
900   /// increment of this AddRec in the loop \p L. In both cases all other AddRec
901   /// sub-expressions (not related to \p L) remain the same.
902   /// If the \p S contains non-invariant unknown SCEV the function returns
903   /// CouldNotCompute SCEV in both values of std::pair.
904   /// For example, for SCEV S={0, +, 1}<L1> + {0, +, 1}<L2> and loop L=L1
905   /// the function returns pair:
906   /// first = {0, +, 1}<L2>
907   /// second = {1, +, 1}<L1> + {0, +, 1}<L2>
908   /// We can see that for the first AddRec sub-expression it was replaced with
909   /// 0 (initial value) for the first element and to {1, +, 1}<L1> (post
910   /// increment value) for the second one. In both cases AddRec expression
911   /// related to L2 remains the same.
912   std::pair<const SCEV *, const SCEV *> SplitIntoInitAndPostInc(const Loop *L,
913                                                                 const SCEV *S);
914 
915   /// We'd like to check the predicate on every iteration of the most dominated
916   /// loop between loops used in LHS and RHS.
917   /// To do this we use the following list of steps:
918   /// 1. Collect set S all loops on which either LHS or RHS depend.
919   /// 2. If S is non-empty
920   /// a. Let PD be the element of S which is dominated by all other elements.
921   /// b. Let E(LHS) be value of LHS on entry of PD.
922   ///    To get E(LHS), we should just take LHS and replace all AddRecs that are
923   ///    attached to PD on with their entry values.
924   ///    Define E(RHS) in the same way.
925   /// c. Let B(LHS) be value of L on backedge of PD.
926   ///    To get B(LHS), we should just take LHS and replace all AddRecs that are
927   ///    attached to PD on with their backedge values.
928   ///    Define B(RHS) in the same way.
929   /// d. Note that E(LHS) and E(RHS) are automatically available on entry of PD,
930   ///    so we can assert on that.
931   /// e. Return true if isLoopEntryGuardedByCond(Pred, E(LHS), E(RHS)) &&
932   ///                   isLoopBackedgeGuardedByCond(Pred, B(LHS), B(RHS))
933   bool isKnownViaInduction(ICmpInst::Predicate Pred, const SCEV *LHS,
934                            const SCEV *RHS);
935 
936   /// Test if the given expression is known to satisfy the condition described
937   /// by Pred, LHS, and RHS.
938   bool isKnownPredicate(ICmpInst::Predicate Pred, const SCEV *LHS,
939                         const SCEV *RHS);
940 
941   /// Test if the given expression is known to satisfy the condition described
942   /// by Pred, LHS, and RHS in the given Context.
943   bool isKnownPredicateAt(ICmpInst::Predicate Pred, const SCEV *LHS,
944                         const SCEV *RHS, const Instruction *Context);
945 
946   /// Test if the condition described by Pred, LHS, RHS is known to be true on
947   /// every iteration of the loop of the recurrency LHS.
948   bool isKnownOnEveryIteration(ICmpInst::Predicate Pred,
949                                const SCEVAddRecExpr *LHS, const SCEV *RHS);
950 
951   /// A predicate is said to be monotonically increasing if may go from being
952   /// false to being true as the loop iterates, but never the other way
953   /// around.  A predicate is said to be monotonically decreasing if may go
954   /// from being true to being false as the loop iterates, but never the other
955   /// way around.
956   enum MonotonicPredicateType {
957     MonotonicallyIncreasing,
958     MonotonicallyDecreasing
959   };
960 
961   /// If, for all loop invariant X, the predicate "LHS `Pred` X" is
962   /// monotonically increasing or decreasing, returns
963   /// Some(MonotonicallyIncreasing) and Some(MonotonicallyDecreasing)
964   /// respectively. If we could not prove either of these facts, returns None.
965   Optional<MonotonicPredicateType>
966   getMonotonicPredicateType(const SCEVAddRecExpr *LHS,
967                             ICmpInst::Predicate Pred);
968 
969   struct LoopInvariantPredicate {
970     ICmpInst::Predicate Pred;
971     const SCEV *LHS;
972     const SCEV *RHS;
973 
974     LoopInvariantPredicate(ICmpInst::Predicate Pred, const SCEV *LHS,
975                            const SCEV *RHS)
976         : Pred(Pred), LHS(LHS), RHS(RHS) {}
977   };
978   /// If the result of the predicate LHS `Pred` RHS is loop invariant with
979   /// respect to L, return a LoopInvariantPredicate with LHS and RHS being
980   /// invariants, available at L's entry. Otherwise, return None.
981   Optional<LoopInvariantPredicate>
982   getLoopInvariantPredicate(ICmpInst::Predicate Pred, const SCEV *LHS,
983                             const SCEV *RHS, const Loop *L);
984 
985   /// If the result of the predicate LHS `Pred` RHS is loop invariant with
986   /// respect to L at given Context during at least first MaxIter iterations,
987   /// return a LoopInvariantPredicate with LHS and RHS being invariants,
988   /// available at L's entry. Otherwise, return None. The predicate should be
989   /// the loop's exit condition.
990   Optional<LoopInvariantPredicate>
991   getLoopInvariantExitCondDuringFirstIterations(ICmpInst::Predicate Pred,
992                                                 const SCEV *LHS,
993                                                 const SCEV *RHS, const Loop *L,
994                                                 const Instruction *Context,
995                                                 const SCEV *MaxIter);
996 
997   /// Simplify LHS and RHS in a comparison with predicate Pred. Return true
998   /// iff any changes were made. If the operands are provably equal or
999   /// unequal, LHS and RHS are set to the same value and Pred is set to either
1000   /// ICMP_EQ or ICMP_NE.
1001   bool SimplifyICmpOperands(ICmpInst::Predicate &Pred, const SCEV *&LHS,
1002                             const SCEV *&RHS, unsigned Depth = 0);
1003 
1004   /// Return the "disposition" of the given SCEV with respect to the given
1005   /// loop.
1006   LoopDisposition getLoopDisposition(const SCEV *S, const Loop *L);
1007 
1008   /// Return true if the value of the given SCEV is unchanging in the
1009   /// specified loop.
1010   bool isLoopInvariant(const SCEV *S, const Loop *L);
1011 
1012   /// Determine if the SCEV can be evaluated at loop's entry. It is true if it
1013   /// doesn't depend on a SCEVUnknown of an instruction which is dominated by
1014   /// the header of loop L.
1015   bool isAvailableAtLoopEntry(const SCEV *S, const Loop *L);
1016 
1017   /// Return true if the given SCEV changes value in a known way in the
1018   /// specified loop.  This property being true implies that the value is
1019   /// variant in the loop AND that we can emit an expression to compute the
1020   /// value of the expression at any particular loop iteration.
1021   bool hasComputableLoopEvolution(const SCEV *S, const Loop *L);
1022 
1023   /// Return the "disposition" of the given SCEV with respect to the given
1024   /// block.
1025   BlockDisposition getBlockDisposition(const SCEV *S, const BasicBlock *BB);
1026 
1027   /// Return true if elements that makes up the given SCEV dominate the
1028   /// specified basic block.
1029   bool dominates(const SCEV *S, const BasicBlock *BB);
1030 
1031   /// Return true if elements that makes up the given SCEV properly dominate
1032   /// the specified basic block.
1033   bool properlyDominates(const SCEV *S, const BasicBlock *BB);
1034 
1035   /// Test whether the given SCEV has Op as a direct or indirect operand.
1036   bool hasOperand(const SCEV *S, const SCEV *Op) const;
1037 
1038   /// Return the size of an element read or written by Inst.
1039   const SCEV *getElementSize(Instruction *Inst);
1040 
1041   /// Compute the array dimensions Sizes from the set of Terms extracted from
1042   /// the memory access function of this SCEVAddRecExpr (second step of
1043   /// delinearization).
1044   void findArrayDimensions(SmallVectorImpl<const SCEV *> &Terms,
1045                            SmallVectorImpl<const SCEV *> &Sizes,
1046                            const SCEV *ElementSize);
1047 
1048   void print(raw_ostream &OS) const;
1049   void verify() const;
1050   bool invalidate(Function &F, const PreservedAnalyses &PA,
1051                   FunctionAnalysisManager::Invalidator &Inv);
1052 
1053   /// Collect parametric terms occurring in step expressions (first step of
1054   /// delinearization).
1055   void collectParametricTerms(const SCEV *Expr,
1056                               SmallVectorImpl<const SCEV *> &Terms);
1057 
1058   /// Return in Subscripts the access functions for each dimension in Sizes
1059   /// (third step of delinearization).
1060   void computeAccessFunctions(const SCEV *Expr,
1061                               SmallVectorImpl<const SCEV *> &Subscripts,
1062                               SmallVectorImpl<const SCEV *> &Sizes);
1063 
1064   /// Gathers the individual index expressions from a GEP instruction.
1065   ///
1066   /// This function optimistically assumes the GEP references into a fixed size
1067   /// array. If this is actually true, this function returns a list of array
1068   /// subscript expressions in \p Subscripts and a list of integers describing
1069   /// the size of the individual array dimensions in \p Sizes. Both lists have
1070   /// either equal length or the size list is one element shorter in case there
1071   /// is no known size available for the outermost array dimension. Returns true
1072   /// if successful and false otherwise.
1073   bool getIndexExpressionsFromGEP(const GetElementPtrInst *GEP,
1074                                   SmallVectorImpl<const SCEV *> &Subscripts,
1075                                   SmallVectorImpl<int> &Sizes);
1076 
1077   /// Split this SCEVAddRecExpr into two vectors of SCEVs representing the
1078   /// subscripts and sizes of an array access.
1079   ///
1080   /// The delinearization is a 3 step process: the first two steps compute the
1081   /// sizes of each subscript and the third step computes the access functions
1082   /// for the delinearized array:
1083   ///
1084   /// 1. Find the terms in the step functions
1085   /// 2. Compute the array size
1086   /// 3. Compute the access function: divide the SCEV by the array size
1087   ///    starting with the innermost dimensions found in step 2. The Quotient
1088   ///    is the SCEV to be divided in the next step of the recursion. The
1089   ///    Remainder is the subscript of the innermost dimension. Loop over all
1090   ///    array dimensions computed in step 2.
1091   ///
1092   /// To compute a uniform array size for several memory accesses to the same
1093   /// object, one can collect in step 1 all the step terms for all the memory
1094   /// accesses, and compute in step 2 a unique array shape. This guarantees
1095   /// that the array shape will be the same across all memory accesses.
1096   ///
1097   /// FIXME: We could derive the result of steps 1 and 2 from a description of
1098   /// the array shape given in metadata.
1099   ///
1100   /// Example:
1101   ///
1102   /// A[][n][m]
1103   ///
1104   /// for i
1105   ///   for j
1106   ///     for k
1107   ///       A[j+k][2i][5i] =
1108   ///
1109   /// The initial SCEV:
1110   ///
1111   /// A[{{{0,+,2*m+5}_i, +, n*m}_j, +, n*m}_k]
1112   ///
1113   /// 1. Find the different terms in the step functions:
1114   /// -> [2*m, 5, n*m, n*m]
1115   ///
1116   /// 2. Compute the array size: sort and unique them
1117   /// -> [n*m, 2*m, 5]
1118   /// find the GCD of all the terms = 1
1119   /// divide by the GCD and erase constant terms
1120   /// -> [n*m, 2*m]
1121   /// GCD = m
1122   /// divide by GCD -> [n, 2]
1123   /// remove constant terms
1124   /// -> [n]
1125   /// size of the array is A[unknown][n][m]
1126   ///
1127   /// 3. Compute the access function
1128   /// a. Divide {{{0,+,2*m+5}_i, +, n*m}_j, +, n*m}_k by the innermost size m
1129   /// Quotient: {{{0,+,2}_i, +, n}_j, +, n}_k
1130   /// Remainder: {{{0,+,5}_i, +, 0}_j, +, 0}_k
1131   /// The remainder is the subscript of the innermost array dimension: [5i].
1132   ///
1133   /// b. Divide Quotient: {{{0,+,2}_i, +, n}_j, +, n}_k by next outer size n
1134   /// Quotient: {{{0,+,0}_i, +, 1}_j, +, 1}_k
1135   /// Remainder: {{{0,+,2}_i, +, 0}_j, +, 0}_k
1136   /// The Remainder is the subscript of the next array dimension: [2i].
1137   ///
1138   /// The subscript of the outermost dimension is the Quotient: [j+k].
1139   ///
1140   /// Overall, we have: A[][n][m], and the access function: A[j+k][2i][5i].
1141   void delinearize(const SCEV *Expr, SmallVectorImpl<const SCEV *> &Subscripts,
1142                    SmallVectorImpl<const SCEV *> &Sizes,
1143                    const SCEV *ElementSize);
1144 
1145   /// Return the DataLayout associated with the module this SCEV instance is
1146   /// operating on.
1147   const DataLayout &getDataLayout() const {
1148     return F.getParent()->getDataLayout();
1149   }
1150 
1151   const SCEVPredicate *getEqualPredicate(const SCEV *LHS, const SCEV *RHS);
1152 
1153   const SCEVPredicate *
1154   getWrapPredicate(const SCEVAddRecExpr *AR,
1155                    SCEVWrapPredicate::IncrementWrapFlags AddedFlags);
1156 
1157   /// Re-writes the SCEV according to the Predicates in \p A.
1158   const SCEV *rewriteUsingPredicate(const SCEV *S, const Loop *L,
1159                                     SCEVUnionPredicate &A);
1160   /// Tries to convert the \p S expression to an AddRec expression,
1161   /// adding additional predicates to \p Preds as required.
1162   const SCEVAddRecExpr *convertSCEVToAddRecWithPredicates(
1163       const SCEV *S, const Loop *L,
1164       SmallPtrSetImpl<const SCEVPredicate *> &Preds);
1165 
1166   /// Compute \p LHS - \p RHS and returns the result as an APInt if it is a
1167   /// constant, and None if it isn't.
1168   ///
1169   /// This is intended to be a cheaper version of getMinusSCEV.  We can be
1170   /// frugal here since we just bail out of actually constructing and
1171   /// canonicalizing an expression in the cases where the result isn't going
1172   /// to be a constant.
1173   Optional<APInt> computeConstantDifference(const SCEV *LHS, const SCEV *RHS);
1174 
1175   /// Update no-wrap flags of an AddRec. This may drop the cached info about
1176   /// this AddRec (such as range info) in case if new flags may potentially
1177   /// sharpen it.
1178   void setNoWrapFlags(SCEVAddRecExpr *AddRec, SCEV::NoWrapFlags Flags);
1179 
1180 private:
1181   /// A CallbackVH to arrange for ScalarEvolution to be notified whenever a
1182   /// Value is deleted.
1183   class SCEVCallbackVH final : public CallbackVH {
1184     ScalarEvolution *SE;
1185 
1186     void deleted() override;
1187     void allUsesReplacedWith(Value *New) override;
1188 
1189   public:
1190     SCEVCallbackVH(Value *V, ScalarEvolution *SE = nullptr);
1191   };
1192 
1193   friend class SCEVCallbackVH;
1194   friend class SCEVExpander;
1195   friend class SCEVUnknown;
1196 
1197   /// The function we are analyzing.
1198   Function &F;
1199 
1200   /// Does the module have any calls to the llvm.experimental.guard intrinsic
1201   /// at all?  If this is false, we avoid doing work that will only help if
1202   /// thare are guards present in the IR.
1203   bool HasGuards;
1204 
1205   /// The target library information for the target we are targeting.
1206   TargetLibraryInfo &TLI;
1207 
1208   /// The tracker for \@llvm.assume intrinsics in this function.
1209   AssumptionCache &AC;
1210 
1211   /// The dominator tree.
1212   DominatorTree &DT;
1213 
1214   /// The loop information for the function we are currently analyzing.
1215   LoopInfo &LI;
1216 
1217   /// This SCEV is used to represent unknown trip counts and things.
1218   std::unique_ptr<SCEVCouldNotCompute> CouldNotCompute;
1219 
1220   /// The type for HasRecMap.
1221   using HasRecMapType = DenseMap<const SCEV *, bool>;
1222 
1223   /// This is a cache to record whether a SCEV contains any scAddRecExpr.
1224   HasRecMapType HasRecMap;
1225 
1226   /// The type for ExprValueMap.
1227   using ValueOffsetPair = std::pair<Value *, ConstantInt *>;
1228   using ExprValueMapType = DenseMap<const SCEV *, SetVector<ValueOffsetPair>>;
1229 
1230   /// ExprValueMap -- This map records the original values from which
1231   /// the SCEV expr is generated from.
1232   ///
1233   /// We want to represent the mapping as SCEV -> ValueOffsetPair instead
1234   /// of SCEV -> Value:
1235   /// Suppose we know S1 expands to V1, and
1236   ///  S1 = S2 + C_a
1237   ///  S3 = S2 + C_b
1238   /// where C_a and C_b are different SCEVConstants. Then we'd like to
1239   /// expand S3 as V1 - C_a + C_b instead of expanding S2 literally.
1240   /// It is helpful when S2 is a complex SCEV expr.
1241   ///
1242   /// In order to do that, we represent ExprValueMap as a mapping from
1243   /// SCEV to ValueOffsetPair. We will save both S1->{V1, 0} and
1244   /// S2->{V1, C_a} into the map when we create SCEV for V1. When S3
1245   /// is expanded, it will first expand S2 to V1 - C_a because of
1246   /// S2->{V1, C_a} in the map, then expand S3 to V1 - C_a + C_b.
1247   ///
1248   /// Note: S->{V, Offset} in the ExprValueMap means S can be expanded
1249   /// to V - Offset.
1250   ExprValueMapType ExprValueMap;
1251 
1252   /// The type for ValueExprMap.
1253   using ValueExprMapType =
1254       DenseMap<SCEVCallbackVH, const SCEV *, DenseMapInfo<Value *>>;
1255 
1256   /// This is a cache of the values we have analyzed so far.
1257   ValueExprMapType ValueExprMap;
1258 
1259   /// Mark predicate values currently being processed by isImpliedCond.
1260   SmallPtrSet<const Value *, 6> PendingLoopPredicates;
1261 
1262   /// Mark SCEVUnknown Phis currently being processed by getRangeRef.
1263   SmallPtrSet<const PHINode *, 6> PendingPhiRanges;
1264 
1265   // Mark SCEVUnknown Phis currently being processed by isImpliedViaMerge.
1266   SmallPtrSet<const PHINode *, 6> PendingMerges;
1267 
1268   /// Set to true by isLoopBackedgeGuardedByCond when we're walking the set of
1269   /// conditions dominating the backedge of a loop.
1270   bool WalkingBEDominatingConds = false;
1271 
1272   /// Set to true by isKnownPredicateViaSplitting when we're trying to prove a
1273   /// predicate by splitting it into a set of independent predicates.
1274   bool ProvingSplitPredicate = false;
1275 
1276   /// Memoized values for the GetMinTrailingZeros
1277   DenseMap<const SCEV *, uint32_t> MinTrailingZerosCache;
1278 
1279   /// Return the Value set from which the SCEV expr is generated.
1280   SetVector<ValueOffsetPair> *getSCEVValues(const SCEV *S);
1281 
1282   /// Private helper method for the GetMinTrailingZeros method
1283   uint32_t GetMinTrailingZerosImpl(const SCEV *S);
1284 
1285   /// Information about the number of loop iterations for which a loop exit's
1286   /// branch condition evaluates to the not-taken path.  This is a temporary
1287   /// pair of exact and max expressions that are eventually summarized in
1288   /// ExitNotTakenInfo and BackedgeTakenInfo.
1289   struct ExitLimit {
1290     const SCEV *ExactNotTaken; // The exit is not taken exactly this many times
1291     const SCEV *MaxNotTaken; // The exit is not taken at most this many times
1292 
1293     // Not taken either exactly MaxNotTaken or zero times
1294     bool MaxOrZero = false;
1295 
1296     /// A set of predicate guards for this ExitLimit. The result is only valid
1297     /// if all of the predicates in \c Predicates evaluate to 'true' at
1298     /// run-time.
1299     SmallPtrSet<const SCEVPredicate *, 4> Predicates;
1300 
1301     void addPredicate(const SCEVPredicate *P) {
1302       assert(!isa<SCEVUnionPredicate>(P) && "Only add leaf predicates here!");
1303       Predicates.insert(P);
1304     }
1305 
1306     /// Construct either an exact exit limit from a constant, or an unknown
1307     /// one from a SCEVCouldNotCompute.  No other types of SCEVs are allowed
1308     /// as arguments and asserts enforce that internally.
1309     /*implicit*/ ExitLimit(const SCEV *E);
1310 
1311     ExitLimit(
1312         const SCEV *E, const SCEV *M, bool MaxOrZero,
1313         ArrayRef<const SmallPtrSetImpl<const SCEVPredicate *> *> PredSetList);
1314 
1315     ExitLimit(const SCEV *E, const SCEV *M, bool MaxOrZero,
1316               const SmallPtrSetImpl<const SCEVPredicate *> &PredSet);
1317 
1318     ExitLimit(const SCEV *E, const SCEV *M, bool MaxOrZero);
1319 
1320     /// Test whether this ExitLimit contains any computed information, or
1321     /// whether it's all SCEVCouldNotCompute values.
1322     bool hasAnyInfo() const {
1323       return !isa<SCEVCouldNotCompute>(ExactNotTaken) ||
1324              !isa<SCEVCouldNotCompute>(MaxNotTaken);
1325     }
1326 
1327     bool hasOperand(const SCEV *S) const;
1328 
1329     /// Test whether this ExitLimit contains all information.
1330     bool hasFullInfo() const {
1331       return !isa<SCEVCouldNotCompute>(ExactNotTaken);
1332     }
1333   };
1334 
1335   /// Information about the number of times a particular loop exit may be
1336   /// reached before exiting the loop.
1337   struct ExitNotTakenInfo {
1338     PoisoningVH<BasicBlock> ExitingBlock;
1339     const SCEV *ExactNotTaken;
1340     const SCEV *MaxNotTaken;
1341     std::unique_ptr<SCEVUnionPredicate> Predicate;
1342 
1343     explicit ExitNotTakenInfo(PoisoningVH<BasicBlock> ExitingBlock,
1344                               const SCEV *ExactNotTaken,
1345                               const SCEV *MaxNotTaken,
1346                               std::unique_ptr<SCEVUnionPredicate> Predicate)
1347       : ExitingBlock(ExitingBlock), ExactNotTaken(ExactNotTaken),
1348         MaxNotTaken(ExactNotTaken), Predicate(std::move(Predicate)) {}
1349 
1350     bool hasAlwaysTruePredicate() const {
1351       return !Predicate || Predicate->isAlwaysTrue();
1352     }
1353   };
1354 
1355   /// Information about the backedge-taken count of a loop. This currently
1356   /// includes an exact count and a maximum count.
1357   ///
1358   class BackedgeTakenInfo {
1359     /// A list of computable exits and their not-taken counts.  Loops almost
1360     /// never have more than one computable exit.
1361     SmallVector<ExitNotTakenInfo, 1> ExitNotTaken;
1362 
1363     /// Expression indicating the least constant maximum backedge-taken count of
1364     /// the loop that is known, or a SCEVCouldNotCompute. This expression is
1365     /// only valid if the redicates associated with all loop exits are true.
1366     const SCEV *ConstantMax;
1367 
1368     /// Indicating if \c ExitNotTaken has an element for every exiting block in
1369     /// the loop.
1370     bool IsComplete;
1371 
1372     /// Expression indicating the least maximum backedge-taken count of the loop
1373     /// that is known, or a SCEVCouldNotCompute. Lazily computed on first query.
1374     const SCEV *SymbolicMax = nullptr;
1375 
1376     /// True iff the backedge is taken either exactly Max or zero times.
1377     bool MaxOrZero = false;
1378 
1379     bool isComplete() const { return IsComplete; }
1380     const SCEV *getConstantMax() const { return ConstantMax; }
1381 
1382   public:
1383     BackedgeTakenInfo() : ConstantMax(nullptr), IsComplete(false) {}
1384     BackedgeTakenInfo(BackedgeTakenInfo &&) = default;
1385     BackedgeTakenInfo &operator=(BackedgeTakenInfo &&) = default;
1386 
1387     using EdgeExitInfo = std::pair<BasicBlock *, ExitLimit>;
1388 
1389     /// Initialize BackedgeTakenInfo from a list of exact exit counts.
1390     BackedgeTakenInfo(ArrayRef<EdgeExitInfo> ExitCounts, bool IsComplete,
1391                       const SCEV *ConstantMax, bool MaxOrZero);
1392 
1393     /// Test whether this BackedgeTakenInfo contains any computed information,
1394     /// or whether it's all SCEVCouldNotCompute values.
1395     bool hasAnyInfo() const {
1396       return !ExitNotTaken.empty() ||
1397              !isa<SCEVCouldNotCompute>(getConstantMax());
1398     }
1399 
1400     /// Test whether this BackedgeTakenInfo contains complete information.
1401     bool hasFullInfo() const { return isComplete(); }
1402 
1403     /// Return an expression indicating the exact *backedge-taken*
1404     /// count of the loop if it is known or SCEVCouldNotCompute
1405     /// otherwise.  If execution makes it to the backedge on every
1406     /// iteration (i.e. there are no abnormal exists like exception
1407     /// throws and thread exits) then this is the number of times the
1408     /// loop header will execute minus one.
1409     ///
1410     /// If the SCEV predicate associated with the answer can be different
1411     /// from AlwaysTrue, we must add a (non null) Predicates argument.
1412     /// The SCEV predicate associated with the answer will be added to
1413     /// Predicates. A run-time check needs to be emitted for the SCEV
1414     /// predicate in order for the answer to be valid.
1415     ///
1416     /// Note that we should always know if we need to pass a predicate
1417     /// argument or not from the way the ExitCounts vector was computed.
1418     /// If we allowed SCEV predicates to be generated when populating this
1419     /// vector, this information can contain them and therefore a
1420     /// SCEVPredicate argument should be added to getExact.
1421     const SCEV *getExact(const Loop *L, ScalarEvolution *SE,
1422                          SCEVUnionPredicate *Predicates = nullptr) const;
1423 
1424     /// Return the number of times this loop exit may fall through to the back
1425     /// edge, or SCEVCouldNotCompute. The loop is guaranteed not to exit via
1426     /// this block before this number of iterations, but may exit via another
1427     /// block.
1428     const SCEV *getExact(const BasicBlock *ExitingBlock,
1429                          ScalarEvolution *SE) const;
1430 
1431     /// Get the constant max backedge taken count for the loop.
1432     const SCEV *getConstantMax(ScalarEvolution *SE) const;
1433 
1434     /// Get the constant max backedge taken count for the particular loop exit.
1435     const SCEV *getConstantMax(const BasicBlock *ExitingBlock,
1436                                ScalarEvolution *SE) const;
1437 
1438     /// Get the symbolic max backedge taken count for the loop.
1439     const SCEV *getSymbolicMax(const Loop *L, ScalarEvolution *SE);
1440 
1441     /// Return true if the number of times this backedge is taken is either the
1442     /// value returned by getConstantMax or zero.
1443     bool isConstantMaxOrZero(ScalarEvolution *SE) const;
1444 
1445     /// Return true if any backedge taken count expressions refer to the given
1446     /// subexpression.
1447     bool hasOperand(const SCEV *S, ScalarEvolution *SE) const;
1448 
1449     /// Invalidate this result and free associated memory.
1450     void clear();
1451   };
1452 
1453   /// Cache the backedge-taken count of the loops for this function as they
1454   /// are computed.
1455   DenseMap<const Loop *, BackedgeTakenInfo> BackedgeTakenCounts;
1456 
1457   /// Cache the predicated backedge-taken count of the loops for this
1458   /// function as they are computed.
1459   DenseMap<const Loop *, BackedgeTakenInfo> PredicatedBackedgeTakenCounts;
1460 
1461   /// This map contains entries for all of the PHI instructions that we
1462   /// attempt to compute constant evolutions for.  This allows us to avoid
1463   /// potentially expensive recomputation of these properties.  An instruction
1464   /// maps to null if we are unable to compute its exit value.
1465   DenseMap<PHINode *, Constant *> ConstantEvolutionLoopExitValue;
1466 
1467   /// This map contains entries for all the expressions that we attempt to
1468   /// compute getSCEVAtScope information for, which can be expensive in
1469   /// extreme cases.
1470   DenseMap<const SCEV *, SmallVector<std::pair<const Loop *, const SCEV *>, 2>>
1471       ValuesAtScopes;
1472 
1473   /// Memoized computeLoopDisposition results.
1474   DenseMap<const SCEV *,
1475            SmallVector<PointerIntPair<const Loop *, 2, LoopDisposition>, 2>>
1476       LoopDispositions;
1477 
1478   struct LoopProperties {
1479     /// Set to true if the loop contains no instruction that can have side
1480     /// effects (i.e. via throwing an exception, volatile or atomic access).
1481     bool HasNoAbnormalExits;
1482 
1483     /// Set to true if the loop contains no instruction that can abnormally exit
1484     /// the loop (i.e. via throwing an exception, by terminating the thread
1485     /// cleanly or by infinite looping in a called function).  Strictly
1486     /// speaking, the last one is not leaving the loop, but is identical to
1487     /// leaving the loop for reasoning about undefined behavior.
1488     bool HasNoSideEffects;
1489   };
1490 
1491   /// Cache for \c getLoopProperties.
1492   DenseMap<const Loop *, LoopProperties> LoopPropertiesCache;
1493 
1494   /// Return a \c LoopProperties instance for \p L, creating one if necessary.
1495   LoopProperties getLoopProperties(const Loop *L);
1496 
1497   bool loopHasNoSideEffects(const Loop *L) {
1498     return getLoopProperties(L).HasNoSideEffects;
1499   }
1500 
1501   bool loopHasNoAbnormalExits(const Loop *L) {
1502     return getLoopProperties(L).HasNoAbnormalExits;
1503   }
1504 
1505   /// Compute a LoopDisposition value.
1506   LoopDisposition computeLoopDisposition(const SCEV *S, const Loop *L);
1507 
1508   /// Memoized computeBlockDisposition results.
1509   DenseMap<
1510       const SCEV *,
1511       SmallVector<PointerIntPair<const BasicBlock *, 2, BlockDisposition>, 2>>
1512       BlockDispositions;
1513 
1514   /// Compute a BlockDisposition value.
1515   BlockDisposition computeBlockDisposition(const SCEV *S, const BasicBlock *BB);
1516 
1517   /// Memoized results from getRange
1518   DenseMap<const SCEV *, ConstantRange> UnsignedRanges;
1519 
1520   /// Memoized results from getRange
1521   DenseMap<const SCEV *, ConstantRange> SignedRanges;
1522 
1523   /// Used to parameterize getRange
1524   enum RangeSignHint { HINT_RANGE_UNSIGNED, HINT_RANGE_SIGNED };
1525 
1526   /// Set the memoized range for the given SCEV.
1527   const ConstantRange &setRange(const SCEV *S, RangeSignHint Hint,
1528                                 ConstantRange CR) {
1529     DenseMap<const SCEV *, ConstantRange> &Cache =
1530         Hint == HINT_RANGE_UNSIGNED ? UnsignedRanges : SignedRanges;
1531 
1532     auto Pair = Cache.try_emplace(S, std::move(CR));
1533     if (!Pair.second)
1534       Pair.first->second = std::move(CR);
1535     return Pair.first->second;
1536   }
1537 
1538   /// Determine the range for a particular SCEV.
1539   /// NOTE: This returns a reference to an entry in a cache. It must be
1540   /// copied if its needed for longer.
1541   const ConstantRange &getRangeRef(const SCEV *S, RangeSignHint Hint);
1542 
1543   /// Determines the range for the affine SCEVAddRecExpr {\p Start,+,\p Stop}.
1544   /// Helper for \c getRange.
1545   ConstantRange getRangeForAffineAR(const SCEV *Start, const SCEV *Stop,
1546                                     const SCEV *MaxBECount, unsigned BitWidth);
1547 
1548   /// Determines the range for the affine non-self-wrapping SCEVAddRecExpr {\p
1549   /// Start,+,\p Stop}<nw>.
1550   ConstantRange getRangeForAffineNoSelfWrappingAR(const SCEVAddRecExpr *AddRec,
1551                                                   const SCEV *MaxBECount,
1552                                                   unsigned BitWidth,
1553                                                   RangeSignHint SignHint);
1554 
1555   /// Try to compute a range for the affine SCEVAddRecExpr {\p Start,+,\p
1556   /// Stop} by "factoring out" a ternary expression from the add recurrence.
1557   /// Helper called by \c getRange.
1558   ConstantRange getRangeViaFactoring(const SCEV *Start, const SCEV *Stop,
1559                                      const SCEV *MaxBECount, unsigned BitWidth);
1560 
1561   /// We know that there is no SCEV for the specified value.  Analyze the
1562   /// expression.
1563   const SCEV *createSCEV(Value *V);
1564 
1565   /// Provide the special handling we need to analyze PHI SCEVs.
1566   const SCEV *createNodeForPHI(PHINode *PN);
1567 
1568   /// Helper function called from createNodeForPHI.
1569   const SCEV *createAddRecFromPHI(PHINode *PN);
1570 
1571   /// A helper function for createAddRecFromPHI to handle simple cases.
1572   const SCEV *createSimpleAffineAddRec(PHINode *PN, Value *BEValueV,
1573                                             Value *StartValueV);
1574 
1575   /// Helper function called from createNodeForPHI.
1576   const SCEV *createNodeFromSelectLikePHI(PHINode *PN);
1577 
1578   /// Provide special handling for a select-like instruction (currently this
1579   /// is either a select instruction or a phi node).  \p I is the instruction
1580   /// being processed, and it is assumed equivalent to "Cond ? TrueVal :
1581   /// FalseVal".
1582   const SCEV *createNodeForSelectOrPHI(Instruction *I, Value *Cond,
1583                                        Value *TrueVal, Value *FalseVal);
1584 
1585   /// Provide the special handling we need to analyze GEP SCEVs.
1586   const SCEV *createNodeForGEP(GEPOperator *GEP);
1587 
1588   /// Implementation code for getSCEVAtScope; called at most once for each
1589   /// SCEV+Loop pair.
1590   const SCEV *computeSCEVAtScope(const SCEV *S, const Loop *L);
1591 
1592   /// This looks up computed SCEV values for all instructions that depend on
1593   /// the given instruction and removes them from the ValueExprMap map if they
1594   /// reference SymName. This is used during PHI resolution.
1595   void forgetSymbolicName(Instruction *I, const SCEV *SymName);
1596 
1597   /// Return the BackedgeTakenInfo for the given loop, lazily computing new
1598   /// values if the loop hasn't been analyzed yet. The returned result is
1599   /// guaranteed not to be predicated.
1600   BackedgeTakenInfo &getBackedgeTakenInfo(const Loop *L);
1601 
1602   /// Similar to getBackedgeTakenInfo, but will add predicates as required
1603   /// with the purpose of returning complete information.
1604   const BackedgeTakenInfo &getPredicatedBackedgeTakenInfo(const Loop *L);
1605 
1606   /// Compute the number of times the specified loop will iterate.
1607   /// If AllowPredicates is set, we will create new SCEV predicates as
1608   /// necessary in order to return an exact answer.
1609   BackedgeTakenInfo computeBackedgeTakenCount(const Loop *L,
1610                                               bool AllowPredicates = false);
1611 
1612   /// Compute the number of times the backedge of the specified loop will
1613   /// execute if it exits via the specified block. If AllowPredicates is set,
1614   /// this call will try to use a minimal set of SCEV predicates in order to
1615   /// return an exact answer.
1616   ExitLimit computeExitLimit(const Loop *L, BasicBlock *ExitingBlock,
1617                              bool AllowPredicates = false);
1618 
1619   /// Compute the number of times the backedge of the specified loop will
1620   /// execute if its exit condition were a conditional branch of ExitCond.
1621   ///
1622   /// \p ControlsExit is true if ExitCond directly controls the exit
1623   /// branch. In this case, we can assume that the loop exits only if the
1624   /// condition is true and can infer that failing to meet the condition prior
1625   /// to integer wraparound results in undefined behavior.
1626   ///
1627   /// If \p AllowPredicates is set, this call will try to use a minimal set of
1628   /// SCEV predicates in order to return an exact answer.
1629   ExitLimit computeExitLimitFromCond(const Loop *L, Value *ExitCond,
1630                                      bool ExitIfTrue, bool ControlsExit,
1631                                      bool AllowPredicates = false);
1632 
1633   /// Return a symbolic upper bound for the backedge taken count of the loop.
1634   /// This is more general than getConstantMaxBackedgeTakenCount as it returns
1635   /// an arbitrary expression as opposed to only constants.
1636   const SCEV *computeSymbolicMaxBackedgeTakenCount(const Loop *L);
1637 
1638   // Helper functions for computeExitLimitFromCond to avoid exponential time
1639   // complexity.
1640 
1641   class ExitLimitCache {
1642     // It may look like we need key on the whole (L, ExitIfTrue, ControlsExit,
1643     // AllowPredicates) tuple, but recursive calls to
1644     // computeExitLimitFromCondCached from computeExitLimitFromCondImpl only
1645     // vary the in \c ExitCond and \c ControlsExit parameters.  We remember the
1646     // initial values of the other values to assert our assumption.
1647     SmallDenseMap<PointerIntPair<Value *, 1>, ExitLimit> TripCountMap;
1648 
1649     const Loop *L;
1650     bool ExitIfTrue;
1651     bool AllowPredicates;
1652 
1653   public:
1654     ExitLimitCache(const Loop *L, bool ExitIfTrue, bool AllowPredicates)
1655         : L(L), ExitIfTrue(ExitIfTrue), AllowPredicates(AllowPredicates) {}
1656 
1657     Optional<ExitLimit> find(const Loop *L, Value *ExitCond, bool ExitIfTrue,
1658                              bool ControlsExit, bool AllowPredicates);
1659 
1660     void insert(const Loop *L, Value *ExitCond, bool ExitIfTrue,
1661                 bool ControlsExit, bool AllowPredicates, const ExitLimit &EL);
1662   };
1663 
1664   using ExitLimitCacheTy = ExitLimitCache;
1665 
1666   ExitLimit computeExitLimitFromCondCached(ExitLimitCacheTy &Cache,
1667                                            const Loop *L, Value *ExitCond,
1668                                            bool ExitIfTrue,
1669                                            bool ControlsExit,
1670                                            bool AllowPredicates);
1671   ExitLimit computeExitLimitFromCondImpl(ExitLimitCacheTy &Cache, const Loop *L,
1672                                          Value *ExitCond, bool ExitIfTrue,
1673                                          bool ControlsExit,
1674                                          bool AllowPredicates);
1675   Optional<ScalarEvolution::ExitLimit>
1676   computeExitLimitFromCondFromBinOp(ExitLimitCacheTy &Cache, const Loop *L,
1677                                     Value *ExitCond, bool ExitIfTrue,
1678                                     bool ControlsExit, bool AllowPredicates);
1679 
1680   /// Compute the number of times the backedge of the specified loop will
1681   /// execute if its exit condition were a conditional branch of the ICmpInst
1682   /// ExitCond and ExitIfTrue. If AllowPredicates is set, this call will try
1683   /// to use a minimal set of SCEV predicates in order to return an exact
1684   /// answer.
1685   ExitLimit computeExitLimitFromICmp(const Loop *L, ICmpInst *ExitCond,
1686                                      bool ExitIfTrue,
1687                                      bool IsSubExpr,
1688                                      bool AllowPredicates = false);
1689 
1690   /// Compute the number of times the backedge of the specified loop will
1691   /// execute if its exit condition were a switch with a single exiting case
1692   /// to ExitingBB.
1693   ExitLimit computeExitLimitFromSingleExitSwitch(const Loop *L,
1694                                                  SwitchInst *Switch,
1695                                                  BasicBlock *ExitingBB,
1696                                                  bool IsSubExpr);
1697 
1698   /// Given an exit condition of 'icmp op load X, cst', try to see if we can
1699   /// compute the backedge-taken count.
1700   ExitLimit computeLoadConstantCompareExitLimit(LoadInst *LI, Constant *RHS,
1701                                                 const Loop *L,
1702                                                 ICmpInst::Predicate p);
1703 
1704   /// Compute the exit limit of a loop that is controlled by a
1705   /// "(IV >> 1) != 0" type comparison.  We cannot compute the exact trip
1706   /// count in these cases (since SCEV has no way of expressing them), but we
1707   /// can still sometimes compute an upper bound.
1708   ///
1709   /// Return an ExitLimit for a loop whose backedge is guarded by `LHS Pred
1710   /// RHS`.
1711   ExitLimit computeShiftCompareExitLimit(Value *LHS, Value *RHS, const Loop *L,
1712                                          ICmpInst::Predicate Pred);
1713 
1714   /// If the loop is known to execute a constant number of times (the
1715   /// condition evolves only from constants), try to evaluate a few iterations
1716   /// of the loop until we get the exit condition gets a value of ExitWhen
1717   /// (true or false).  If we cannot evaluate the exit count of the loop,
1718   /// return CouldNotCompute.
1719   const SCEV *computeExitCountExhaustively(const Loop *L, Value *Cond,
1720                                            bool ExitWhen);
1721 
1722   /// Return the number of times an exit condition comparing the specified
1723   /// value to zero will execute.  If not computable, return CouldNotCompute.
1724   /// If AllowPredicates is set, this call will try to use a minimal set of
1725   /// SCEV predicates in order to return an exact answer.
1726   ExitLimit howFarToZero(const SCEV *V, const Loop *L, bool IsSubExpr,
1727                          bool AllowPredicates = false);
1728 
1729   /// Return the number of times an exit condition checking the specified
1730   /// value for nonzero will execute.  If not computable, return
1731   /// CouldNotCompute.
1732   ExitLimit howFarToNonZero(const SCEV *V, const Loop *L);
1733 
1734   /// Return the number of times an exit condition containing the specified
1735   /// less-than comparison will execute.  If not computable, return
1736   /// CouldNotCompute.
1737   ///
1738   /// \p isSigned specifies whether the less-than is signed.
1739   ///
1740   /// \p ControlsExit is true when the LHS < RHS condition directly controls
1741   /// the branch (loops exits only if condition is true). In this case, we can
1742   /// use NoWrapFlags to skip overflow checks.
1743   ///
1744   /// If \p AllowPredicates is set, this call will try to use a minimal set of
1745   /// SCEV predicates in order to return an exact answer.
1746   ExitLimit howManyLessThans(const SCEV *LHS, const SCEV *RHS, const Loop *L,
1747                              bool isSigned, bool ControlsExit,
1748                              bool AllowPredicates = false);
1749 
1750   ExitLimit howManyGreaterThans(const SCEV *LHS, const SCEV *RHS, const Loop *L,
1751                                 bool isSigned, bool IsSubExpr,
1752                                 bool AllowPredicates = false);
1753 
1754   /// Return a predecessor of BB (which may not be an immediate predecessor)
1755   /// which has exactly one successor from which BB is reachable, or null if
1756   /// no such block is found.
1757   std::pair<const BasicBlock *, const BasicBlock *>
1758   getPredecessorWithUniqueSuccessorForBB(const BasicBlock *BB) const;
1759 
1760   /// Test whether the condition described by Pred, LHS, and RHS is true
1761   /// whenever the given FoundCondValue value evaluates to true in given
1762   /// Context. If Context is nullptr, then the found predicate is true
1763   /// everywhere. LHS and FoundLHS may have different type width.
1764   bool isImpliedCond(ICmpInst::Predicate Pred, const SCEV *LHS, const SCEV *RHS,
1765                      const Value *FoundCondValue, bool Inverse,
1766                      const Instruction *Context = nullptr);
1767 
1768   /// Test whether the condition described by Pred, LHS, and RHS is true
1769   /// whenever the given FoundCondValue value evaluates to true in given
1770   /// Context. If Context is nullptr, then the found predicate is true
1771   /// everywhere. LHS and FoundLHS must have same type width.
1772   bool isImpliedCondBalancedTypes(ICmpInst::Predicate Pred, const SCEV *LHS,
1773                                   const SCEV *RHS,
1774                                   ICmpInst::Predicate FoundPred,
1775                                   const SCEV *FoundLHS, const SCEV *FoundRHS,
1776                                   const Instruction *Context);
1777 
1778   /// Test whether the condition described by Pred, LHS, and RHS is true
1779   /// whenever the condition described by FoundPred, FoundLHS, FoundRHS is
1780   /// true in given Context. If Context is nullptr, then the found predicate is
1781   /// true everywhere.
1782   bool isImpliedCond(ICmpInst::Predicate Pred, const SCEV *LHS, const SCEV *RHS,
1783                      ICmpInst::Predicate FoundPred, const SCEV *FoundLHS,
1784                      const SCEV *FoundRHS,
1785                      const Instruction *Context = nullptr);
1786 
1787   /// Test whether the condition described by Pred, LHS, and RHS is true
1788   /// whenever the condition described by Pred, FoundLHS, and FoundRHS is
1789   /// true in given Context. If Context is nullptr, then the found predicate is
1790   /// true everywhere.
1791   bool isImpliedCondOperands(ICmpInst::Predicate Pred, const SCEV *LHS,
1792                              const SCEV *RHS, const SCEV *FoundLHS,
1793                              const SCEV *FoundRHS,
1794                              const Instruction *Context = nullptr);
1795 
1796   /// Test whether the condition described by Pred, LHS, and RHS is true
1797   /// whenever the condition described by Pred, FoundLHS, and FoundRHS is
1798   /// true. Here LHS is an operation that includes FoundLHS as one of its
1799   /// arguments.
1800   bool isImpliedViaOperations(ICmpInst::Predicate Pred,
1801                               const SCEV *LHS, const SCEV *RHS,
1802                               const SCEV *FoundLHS, const SCEV *FoundRHS,
1803                               unsigned Depth = 0);
1804 
1805   /// Test whether the condition described by Pred, LHS, and RHS is true.
1806   /// Use only simple non-recursive types of checks, such as range analysis etc.
1807   bool isKnownViaNonRecursiveReasoning(ICmpInst::Predicate Pred,
1808                                        const SCEV *LHS, const SCEV *RHS);
1809 
1810   /// Test whether the condition described by Pred, LHS, and RHS is true
1811   /// whenever the condition described by Pred, FoundLHS, and FoundRHS is
1812   /// true.
1813   bool isImpliedCondOperandsHelper(ICmpInst::Predicate Pred, const SCEV *LHS,
1814                                    const SCEV *RHS, const SCEV *FoundLHS,
1815                                    const SCEV *FoundRHS);
1816 
1817   /// Test whether the condition described by Pred, LHS, and RHS is true
1818   /// whenever the condition described by Pred, FoundLHS, and FoundRHS is
1819   /// true.  Utility function used by isImpliedCondOperands.  Tries to get
1820   /// cases like "X `sgt` 0 => X - 1 `sgt` -1".
1821   bool isImpliedCondOperandsViaRanges(ICmpInst::Predicate Pred, const SCEV *LHS,
1822                                       const SCEV *RHS, const SCEV *FoundLHS,
1823                                       const SCEV *FoundRHS);
1824 
1825   /// Return true if the condition denoted by \p LHS \p Pred \p RHS is implied
1826   /// by a call to @llvm.experimental.guard in \p BB.
1827   bool isImpliedViaGuard(const BasicBlock *BB, ICmpInst::Predicate Pred,
1828                          const SCEV *LHS, const SCEV *RHS);
1829 
1830   /// Test whether the condition described by Pred, LHS, and RHS is true
1831   /// whenever the condition described by Pred, FoundLHS, and FoundRHS is
1832   /// true.
1833   ///
1834   /// This routine tries to rule out certain kinds of integer overflow, and
1835   /// then tries to reason about arithmetic properties of the predicates.
1836   bool isImpliedCondOperandsViaNoOverflow(ICmpInst::Predicate Pred,
1837                                           const SCEV *LHS, const SCEV *RHS,
1838                                           const SCEV *FoundLHS,
1839                                           const SCEV *FoundRHS);
1840 
1841   /// Test whether the condition described by Pred, LHS, and RHS is true
1842   /// whenever the condition described by Pred, FoundLHS, and FoundRHS is
1843   /// true.
1844   ///
1845   /// This routine tries to weaken the known condition basing on fact that
1846   /// FoundLHS is an AddRec.
1847   bool isImpliedCondOperandsViaAddRecStart(ICmpInst::Predicate Pred,
1848                                            const SCEV *LHS, const SCEV *RHS,
1849                                            const SCEV *FoundLHS,
1850                                            const SCEV *FoundRHS,
1851                                            const Instruction *Context);
1852 
1853   /// Test whether the condition described by Pred, LHS, and RHS is true
1854   /// whenever the condition described by Pred, FoundLHS, and FoundRHS is
1855   /// true.
1856   ///
1857   /// This routine tries to figure out predicate for Phis which are SCEVUnknown
1858   /// if it is true for every possible incoming value from their respective
1859   /// basic blocks.
1860   bool isImpliedViaMerge(ICmpInst::Predicate Pred,
1861                          const SCEV *LHS, const SCEV *RHS,
1862                          const SCEV *FoundLHS, const SCEV *FoundRHS,
1863                          unsigned Depth);
1864 
1865   /// If we know that the specified Phi is in the header of its containing
1866   /// loop, we know the loop executes a constant number of times, and the PHI
1867   /// node is just a recurrence involving constants, fold it.
1868   Constant *getConstantEvolutionLoopExitValue(PHINode *PN, const APInt &BEs,
1869                                               const Loop *L);
1870 
1871   /// Test if the given expression is known to satisfy the condition described
1872   /// by Pred and the known constant ranges of LHS and RHS.
1873   bool isKnownPredicateViaConstantRanges(ICmpInst::Predicate Pred,
1874                                          const SCEV *LHS, const SCEV *RHS);
1875 
1876   /// Try to prove the condition described by "LHS Pred RHS" by ruling out
1877   /// integer overflow.
1878   ///
1879   /// For instance, this will return true for "A s< (A + C)<nsw>" if C is
1880   /// positive.
1881   bool isKnownPredicateViaNoOverflow(ICmpInst::Predicate Pred, const SCEV *LHS,
1882                                      const SCEV *RHS);
1883 
1884   /// Try to split Pred LHS RHS into logical conjunctions (and's) and try to
1885   /// prove them individually.
1886   bool isKnownPredicateViaSplitting(ICmpInst::Predicate Pred, const SCEV *LHS,
1887                                     const SCEV *RHS);
1888 
1889   /// Try to match the Expr as "(L + R)<Flags>".
1890   bool splitBinaryAdd(const SCEV *Expr, const SCEV *&L, const SCEV *&R,
1891                       SCEV::NoWrapFlags &Flags);
1892 
1893   /// Drop memoized information computed for S.
1894   void forgetMemoizedResults(const SCEV *S);
1895 
1896   /// Return an existing SCEV for V if there is one, otherwise return nullptr.
1897   const SCEV *getExistingSCEV(Value *V);
1898 
1899   /// Return false iff given SCEV contains a SCEVUnknown with NULL value-
1900   /// pointer.
1901   bool checkValidity(const SCEV *S) const;
1902 
1903   /// Return true if `ExtendOpTy`({`Start`,+,`Step`}) can be proved to be
1904   /// equal to {`ExtendOpTy`(`Start`),+,`ExtendOpTy`(`Step`)}.  This is
1905   /// equivalent to proving no signed (resp. unsigned) wrap in
1906   /// {`Start`,+,`Step`} if `ExtendOpTy` is `SCEVSignExtendExpr`
1907   /// (resp. `SCEVZeroExtendExpr`).
1908   template <typename ExtendOpTy>
1909   bool proveNoWrapByVaryingStart(const SCEV *Start, const SCEV *Step,
1910                                  const Loop *L);
1911 
1912   /// Try to prove NSW or NUW on \p AR relying on ConstantRange manipulation.
1913   SCEV::NoWrapFlags proveNoWrapViaConstantRanges(const SCEVAddRecExpr *AR);
1914 
1915   /// Try to prove NSW on \p AR by proving facts about conditions known  on
1916   /// entry and backedge.
1917   SCEV::NoWrapFlags proveNoSignedWrapViaInduction(const SCEVAddRecExpr *AR);
1918 
1919   /// Try to prove NUW on \p AR by proving facts about conditions known on
1920   /// entry and backedge.
1921   SCEV::NoWrapFlags proveNoUnsignedWrapViaInduction(const SCEVAddRecExpr *AR);
1922 
1923   Optional<MonotonicPredicateType>
1924   getMonotonicPredicateTypeImpl(const SCEVAddRecExpr *LHS,
1925                                 ICmpInst::Predicate Pred);
1926 
1927   /// Return SCEV no-wrap flags that can be proven based on reasoning about
1928   /// how poison produced from no-wrap flags on this value (e.g. a nuw add)
1929   /// would trigger undefined behavior on overflow.
1930   SCEV::NoWrapFlags getNoWrapFlagsFromUB(const Value *V);
1931 
1932   /// Return true if the SCEV corresponding to \p I is never poison.  Proving
1933   /// this is more complex than proving that just \p I is never poison, since
1934   /// SCEV commons expressions across control flow, and you can have cases
1935   /// like:
1936   ///
1937   ///   idx0 = a + b;
1938   ///   ptr[idx0] = 100;
1939   ///   if (<condition>) {
1940   ///     idx1 = a +nsw b;
1941   ///     ptr[idx1] = 200;
1942   ///   }
1943   ///
1944   /// where the SCEV expression (+ a b) is guaranteed to not be poison (and
1945   /// hence not sign-overflow) only if "<condition>" is true.  Since both
1946   /// `idx0` and `idx1` will be mapped to the same SCEV expression, (+ a b),
1947   /// it is not okay to annotate (+ a b) with <nsw> in the above example.
1948   bool isSCEVExprNeverPoison(const Instruction *I);
1949 
1950   /// This is like \c isSCEVExprNeverPoison but it specifically works for
1951   /// instructions that will get mapped to SCEV add recurrences.  Return true
1952   /// if \p I will never generate poison under the assumption that \p I is an
1953   /// add recurrence on the loop \p L.
1954   bool isAddRecNeverPoison(const Instruction *I, const Loop *L);
1955 
1956   /// Similar to createAddRecFromPHI, but with the additional flexibility of
1957   /// suggesting runtime overflow checks in case casts are encountered.
1958   /// If successful, the analysis records that for this loop, \p SymbolicPHI,
1959   /// which is the UnknownSCEV currently representing the PHI, can be rewritten
1960   /// into an AddRec, assuming some predicates; The function then returns the
1961   /// AddRec and the predicates as a pair, and caches this pair in
1962   /// PredicatedSCEVRewrites.
1963   /// If the analysis is not successful, a mapping from the \p SymbolicPHI to
1964   /// itself (with no predicates) is recorded, and a nullptr with an empty
1965   /// predicates vector is returned as a pair.
1966   Optional<std::pair<const SCEV *, SmallVector<const SCEVPredicate *, 3>>>
1967   createAddRecFromPHIWithCastsImpl(const SCEVUnknown *SymbolicPHI);
1968 
1969   /// Compute the backedge taken count knowing the interval difference, the
1970   /// stride and presence of the equality in the comparison.
1971   const SCEV *computeBECount(const SCEV *Delta, const SCEV *Stride,
1972                              bool Equality);
1973 
1974   /// Compute the maximum backedge count based on the range of values
1975   /// permitted by Start, End, and Stride. This is for loops of the form
1976   /// {Start, +, Stride} LT End.
1977   ///
1978   /// Precondition: the induction variable is known to be positive.  We *don't*
1979   /// assert these preconditions so please be careful.
1980   const SCEV *computeMaxBECountForLT(const SCEV *Start, const SCEV *Stride,
1981                                      const SCEV *End, unsigned BitWidth,
1982                                      bool IsSigned);
1983 
1984   /// Verify if an linear IV with positive stride can overflow when in a
1985   /// less-than comparison, knowing the invariant term of the comparison,
1986   /// the stride and the knowledge of NSW/NUW flags on the recurrence.
1987   bool doesIVOverflowOnLT(const SCEV *RHS, const SCEV *Stride, bool IsSigned,
1988                           bool NoWrap);
1989 
1990   /// Verify if an linear IV with negative stride can overflow when in a
1991   /// greater-than comparison, knowing the invariant term of the comparison,
1992   /// the stride and the knowledge of NSW/NUW flags on the recurrence.
1993   bool doesIVOverflowOnGT(const SCEV *RHS, const SCEV *Stride, bool IsSigned,
1994                           bool NoWrap);
1995 
1996   /// Get add expr already created or create a new one.
1997   const SCEV *getOrCreateAddExpr(ArrayRef<const SCEV *> Ops,
1998                                  SCEV::NoWrapFlags Flags);
1999 
2000   /// Get mul expr already created or create a new one.
2001   const SCEV *getOrCreateMulExpr(ArrayRef<const SCEV *> Ops,
2002                                  SCEV::NoWrapFlags Flags);
2003 
2004   // Get addrec expr already created or create a new one.
2005   const SCEV *getOrCreateAddRecExpr(ArrayRef<const SCEV *> Ops,
2006                                     const Loop *L, SCEV::NoWrapFlags Flags);
2007 
2008   /// Return x if \p Val is f(x) where f is a 1-1 function.
2009   const SCEV *stripInjectiveFunctions(const SCEV *Val) const;
2010 
2011   /// Find all of the loops transitively used in \p S, and fill \p LoopsUsed.
2012   /// A loop is considered "used" by an expression if it contains
2013   /// an add rec on said loop.
2014   void getUsedLoops(const SCEV *S, SmallPtrSetImpl<const Loop *> &LoopsUsed);
2015 
2016   /// Find all of the loops transitively used in \p S, and update \c LoopUsers
2017   /// accordingly.
2018   void addToLoopUseLists(const SCEV *S);
2019 
2020   /// Try to match the pattern generated by getURemExpr(A, B). If successful,
2021   /// Assign A and B to LHS and RHS, respectively.
2022   bool matchURem(const SCEV *Expr, const SCEV *&LHS, const SCEV *&RHS);
2023 
2024   /// Try to apply information from loop guards for \p L to \p Expr.
2025   const SCEV *applyLoopGuards(const SCEV *Expr, const Loop *L);
2026 
2027   /// Look for a SCEV expression with type `SCEVType` and operands `Ops` in
2028   /// `UniqueSCEVs`.
2029   ///
2030   /// The first component of the returned tuple is the SCEV if found and null
2031   /// otherwise.  The second component is the `FoldingSetNodeID` that was
2032   /// constructed to look up the SCEV and the third component is the insertion
2033   /// point.
2034   std::tuple<SCEV *, FoldingSetNodeID, void *>
2035   findExistingSCEVInCache(SCEVTypes SCEVType, ArrayRef<const SCEV *> Ops);
2036 
2037   FoldingSet<SCEV> UniqueSCEVs;
2038   FoldingSet<SCEVPredicate> UniquePreds;
2039   BumpPtrAllocator SCEVAllocator;
2040 
2041   /// This maps loops to a list of SCEV expressions that (transitively) use said
2042   /// loop.
2043   DenseMap<const Loop *, SmallVector<const SCEV *, 4>> LoopUsers;
2044 
2045   /// Cache tentative mappings from UnknownSCEVs in a Loop, to a SCEV expression
2046   /// they can be rewritten into under certain predicates.
2047   DenseMap<std::pair<const SCEVUnknown *, const Loop *>,
2048            std::pair<const SCEV *, SmallVector<const SCEVPredicate *, 3>>>
2049       PredicatedSCEVRewrites;
2050 
2051   /// The head of a linked list of all SCEVUnknown values that have been
2052   /// allocated. This is used by releaseMemory to locate them all and call
2053   /// their destructors.
2054   SCEVUnknown *FirstUnknown = nullptr;
2055 };
2056 
2057 /// Analysis pass that exposes the \c ScalarEvolution for a function.
2058 class ScalarEvolutionAnalysis
2059     : public AnalysisInfoMixin<ScalarEvolutionAnalysis> {
2060   friend AnalysisInfoMixin<ScalarEvolutionAnalysis>;
2061 
2062   static AnalysisKey Key;
2063 
2064 public:
2065   using Result = ScalarEvolution;
2066 
2067   ScalarEvolution run(Function &F, FunctionAnalysisManager &AM);
2068 };
2069 
2070 /// Verifier pass for the \c ScalarEvolutionAnalysis results.
2071 class ScalarEvolutionVerifierPass
2072     : public PassInfoMixin<ScalarEvolutionVerifierPass> {
2073 public:
2074   PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);
2075 };
2076 
2077 /// Printer pass for the \c ScalarEvolutionAnalysis results.
2078 class ScalarEvolutionPrinterPass
2079     : public PassInfoMixin<ScalarEvolutionPrinterPass> {
2080   raw_ostream &OS;
2081 
2082 public:
2083   explicit ScalarEvolutionPrinterPass(raw_ostream &OS) : OS(OS) {}
2084 
2085   PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);
2086 };
2087 
2088 class ScalarEvolutionWrapperPass : public FunctionPass {
2089   std::unique_ptr<ScalarEvolution> SE;
2090 
2091 public:
2092   static char ID;
2093 
2094   ScalarEvolutionWrapperPass();
2095 
2096   ScalarEvolution &getSE() { return *SE; }
2097   const ScalarEvolution &getSE() const { return *SE; }
2098 
2099   bool runOnFunction(Function &F) override;
2100   void releaseMemory() override;
2101   void getAnalysisUsage(AnalysisUsage &AU) const override;
2102   void print(raw_ostream &OS, const Module * = nullptr) const override;
2103   void verifyAnalysis() const override;
2104 };
2105 
2106 /// An interface layer with SCEV used to manage how we see SCEV expressions
2107 /// for values in the context of existing predicates. We can add new
2108 /// predicates, but we cannot remove them.
2109 ///
2110 /// This layer has multiple purposes:
2111 ///   - provides a simple interface for SCEV versioning.
2112 ///   - guarantees that the order of transformations applied on a SCEV
2113 ///     expression for a single Value is consistent across two different
2114 ///     getSCEV calls. This means that, for example, once we've obtained
2115 ///     an AddRec expression for a certain value through expression
2116 ///     rewriting, we will continue to get an AddRec expression for that
2117 ///     Value.
2118 ///   - lowers the number of expression rewrites.
2119 class PredicatedScalarEvolution {
2120 public:
2121   PredicatedScalarEvolution(ScalarEvolution &SE, Loop &L);
2122 
2123   const SCEVUnionPredicate &getUnionPredicate() const;
2124 
2125   /// Returns the SCEV expression of V, in the context of the current SCEV
2126   /// predicate.  The order of transformations applied on the expression of V
2127   /// returned by ScalarEvolution is guaranteed to be preserved, even when
2128   /// adding new predicates.
2129   const SCEV *getSCEV(Value *V);
2130 
2131   /// Get the (predicated) backedge count for the analyzed loop.
2132   const SCEV *getBackedgeTakenCount();
2133 
2134   /// Adds a new predicate.
2135   void addPredicate(const SCEVPredicate &Pred);
2136 
2137   /// Attempts to produce an AddRecExpr for V by adding additional SCEV
2138   /// predicates. If we can't transform the expression into an AddRecExpr we
2139   /// return nullptr and not add additional SCEV predicates to the current
2140   /// context.
2141   const SCEVAddRecExpr *getAsAddRec(Value *V);
2142 
2143   /// Proves that V doesn't overflow by adding SCEV predicate.
2144   void setNoOverflow(Value *V, SCEVWrapPredicate::IncrementWrapFlags Flags);
2145 
2146   /// Returns true if we've proved that V doesn't wrap by means of a SCEV
2147   /// predicate.
2148   bool hasNoOverflow(Value *V, SCEVWrapPredicate::IncrementWrapFlags Flags);
2149 
2150   /// Returns the ScalarEvolution analysis used.
2151   ScalarEvolution *getSE() const { return &SE; }
2152 
2153   /// We need to explicitly define the copy constructor because of FlagsMap.
2154   PredicatedScalarEvolution(const PredicatedScalarEvolution &);
2155 
2156   /// Print the SCEV mappings done by the Predicated Scalar Evolution.
2157   /// The printed text is indented by \p Depth.
2158   void print(raw_ostream &OS, unsigned Depth) const;
2159 
2160   /// Check if \p AR1 and \p AR2 are equal, while taking into account
2161   /// Equal predicates in Preds.
2162   bool areAddRecsEqualWithPreds(const SCEVAddRecExpr *AR1,
2163                                 const SCEVAddRecExpr *AR2) const;
2164 
2165 private:
2166   /// Increments the version number of the predicate.  This needs to be called
2167   /// every time the SCEV predicate changes.
2168   void updateGeneration();
2169 
2170   /// Holds a SCEV and the version number of the SCEV predicate used to
2171   /// perform the rewrite of the expression.
2172   using RewriteEntry = std::pair<unsigned, const SCEV *>;
2173 
2174   /// Maps a SCEV to the rewrite result of that SCEV at a certain version
2175   /// number. If this number doesn't match the current Generation, we will
2176   /// need to do a rewrite. To preserve the transformation order of previous
2177   /// rewrites, we will rewrite the previous result instead of the original
2178   /// SCEV.
2179   DenseMap<const SCEV *, RewriteEntry> RewriteMap;
2180 
2181   /// Records what NoWrap flags we've added to a Value *.
2182   ValueMap<Value *, SCEVWrapPredicate::IncrementWrapFlags> FlagsMap;
2183 
2184   /// The ScalarEvolution analysis.
2185   ScalarEvolution &SE;
2186 
2187   /// The analyzed Loop.
2188   const Loop &L;
2189 
2190   /// The SCEVPredicate that forms our context. We will rewrite all
2191   /// expressions assuming that this predicate true.
2192   SCEVUnionPredicate Preds;
2193 
2194   /// Marks the version of the SCEV predicate used. When rewriting a SCEV
2195   /// expression we mark it with the version of the predicate. We use this to
2196   /// figure out if the predicate has changed from the last rewrite of the
2197   /// SCEV. If so, we need to perform a new rewrite.
2198   unsigned Generation = 0;
2199 
2200   /// The backedge taken count.
2201   const SCEV *BackedgeCount = nullptr;
2202 };
2203 
2204 } // end namespace llvm
2205 
2206 #endif // LLVM_ANALYSIS_SCALAREVOLUTION_H
2207