LLVM  16.0.0git
LoopVectorizationLegality.h
Go to the documentation of this file.
1 //===- llvm/Transforms/Vectorize/LoopVectorizationLegality.h ----*- 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 /// \file
10 /// This file defines the LoopVectorizationLegality class. Original code
11 /// in Loop Vectorizer has been moved out to its own file for modularity
12 /// and reusability.
13 ///
14 /// Currently, it works for innermost loop vectorization. Extending this to
15 /// outer loop vectorization is a TODO item.
16 ///
17 /// Also provides:
18 /// 1) LoopVectorizeHints class which keeps a number of loop annotations
19 /// locally for easy look up. It has the ability to write them back as
20 /// loop metadata, upon request.
21 /// 2) LoopVectorizationRequirements class for lazy bail out for the purpose
22 /// of reporting useful failure to vectorize message.
23 //
24 //===----------------------------------------------------------------------===//
25 
26 #ifndef LLVM_TRANSFORMS_VECTORIZE_LOOPVECTORIZATIONLEGALITY_H
27 #define LLVM_TRANSFORMS_VECTORIZE_LOOPVECTORIZATIONLEGALITY_H
28 
29 #include "llvm/ADT/MapVector.h"
31 #include "llvm/Support/TypeSize.h"
33 
34 namespace llvm {
35 class AAResults;
36 class AssumptionCache;
37 class BasicBlock;
38 class BlockFrequencyInfo;
39 class DemandedBits;
40 class DominatorTree;
41 class Function;
42 class Loop;
43 class LoopInfo;
44 class Metadata;
45 class OptimizationRemarkEmitter;
46 class PredicatedScalarEvolution;
47 class ProfileSummaryInfo;
48 class TargetLibraryInfo;
49 class TargetTransformInfo;
50 class Type;
51 
52 /// Utility class for getting and setting loop vectorizer hints in the form
53 /// of loop metadata.
54 /// This class keeps a number of loop annotations locally (as member variables)
55 /// and can, upon request, write them back as metadata on the loop. It will
56 /// initially scan the loop for existing metadata, and will update the local
57 /// values based on information in the loop.
58 /// We cannot write all values to metadata, as the mere presence of some info,
59 /// for example 'force', means a decision has been made. So, we need to be
60 /// careful NOT to add them if the user hasn't specifically asked so.
62  enum HintKind {
63  HK_WIDTH,
64  HK_INTERLEAVE,
65  HK_FORCE,
66  HK_ISVECTORIZED,
67  HK_PREDICATE,
68  HK_SCALABLE
69  };
70 
71  /// Hint - associates name and validation with the hint value.
72  struct Hint {
73  const char *Name;
74  unsigned Value; // This may have to change for non-numeric values.
75  HintKind Kind;
76 
77  Hint(const char *Name, unsigned Value, HintKind Kind)
78  : Name(Name), Value(Value), Kind(Kind) {}
79 
80  bool validate(unsigned Val);
81  };
82 
83  /// Vectorization width.
84  Hint Width;
85 
86  /// Vectorization interleave factor.
87  Hint Interleave;
88 
89  /// Vectorization forced
90  Hint Force;
91 
92  /// Already Vectorized
93  Hint IsVectorized;
94 
95  /// Vector Predicate
96  Hint Predicate;
97 
98  /// Says whether we should use fixed width or scalable vectorization.
99  Hint Scalable;
100 
101  /// Return the loop metadata prefix.
102  static StringRef Prefix() { return "llvm.loop."; }
103 
104  /// True if there is any unsafe math in the loop.
105  bool PotentiallyUnsafe = false;
106 
107 public:
108  enum ForceKind {
109  FK_Undefined = -1, ///< Not selected.
110  FK_Disabled = 0, ///< Forcing disabled.
111  FK_Enabled = 1, ///< Forcing enabled.
112  };
113 
115  /// Not selected.
117  /// Disables vectorization with scalable vectors.
119  /// Vectorize loops using scalable vectors or fixed-width vectors, but favor
120  /// scalable vectors when the cost-model is inconclusive. This is the
121  /// default when the scalable.enable hint is enabled through a pragma.
123  };
124 
125  LoopVectorizeHints(const Loop *L, bool InterleaveOnlyWhenForced,
127  const TargetTransformInfo *TTI = nullptr);
128 
129  /// Mark the loop L as already vectorized by setting the width to 1.
130  void setAlreadyVectorized();
131 
132  bool allowVectorization(Function *F, Loop *L,
133  bool VectorizeOnlyWhenForced) const;
134 
135  /// Dumps all the hint information.
136  void emitRemarkWithHints() const;
137 
139  return ElementCount::get(Width.Value, (ScalableForceKind)Scalable.Value ==
141  }
142 
143  unsigned getInterleave() const {
144  if (Interleave.Value)
145  return Interleave.Value;
146  // If interleaving is not explicitly set, assume that if we do not want
147  // unrolling, we also don't want any interleaving.
149  return 1;
150  return 0;
151  }
152  unsigned getIsVectorized() const { return IsVectorized.Value; }
153  unsigned getPredicate() const { return Predicate.Value; }
154  enum ForceKind getForce() const {
155  if ((ForceKind)Force.Value == FK_Undefined &&
157  return FK_Disabled;
158  return (ForceKind)Force.Value;
159  }
160 
161  /// \return true if scalable vectorization has been explicitly disabled.
163  return (ScalableForceKind)Scalable.Value == SK_FixedWidthOnly;
164  }
165 
166  /// If hints are provided that force vectorization, use the AlwaysPrint
167  /// pass name to force the frontend to print the diagnostic.
168  const char *vectorizeAnalysisPassName() const;
169 
170  /// When enabling loop hints are provided we allow the vectorizer to change
171  /// the order of operations that is given by the scalar loop. This is not
172  /// enabled by default because can be unsafe or inefficient. For example,
173  /// reordering floating-point operations will change the way round-off
174  /// error accumulates in the loop.
175  bool allowReordering() const;
176 
177  bool isPotentiallyUnsafe() const {
178  // Avoid FP vectorization if the target is unsure about proper support.
179  // This may be related to the SIMD unit in the target not handling
180  // IEEE 754 FP ops properly, or bad single-to-double promotions.
181  // Otherwise, a sequence of vectorized loops, even without reduction,
182  // could lead to different end results on the destination vectors.
183  return getForce() != LoopVectorizeHints::FK_Enabled && PotentiallyUnsafe;
184  }
185 
186  void setPotentiallyUnsafe() { PotentiallyUnsafe = true; }
187 
188 private:
189  /// Find hints specified in the loop metadata and update local values.
190  void getHintsFromMetadata();
191 
192  /// Checks string hint with one operand and set value if valid.
193  void setHint(StringRef Name, Metadata *Arg);
194 
195  /// The loop these hints belong to.
196  const Loop *TheLoop;
197 
198  /// Interface to emit optimization remarks.
200 };
201 
202 /// This holds vectorization requirements that must be verified late in
203 /// the process. The requirements are set by legalize and costmodel. Once
204 /// vectorization has been determined to be possible and profitable the
205 /// requirements can be verified by looking for metadata or compiler options.
206 /// For example, some loops require FP commutativity which is only allowed if
207 /// vectorization is explicitly specified or if the fast-math compiler option
208 /// has been provided.
209 /// Late evaluation of these requirements allows helpful diagnostics to be
210 /// composed that tells the user what need to be done to vectorize the loop. For
211 /// example, by specifying #pragma clang loop vectorize or -ffast-math. Late
212 /// evaluation should be used only when diagnostics can generated that can be
213 /// followed by a non-expert user.
215 public:
216  /// Track the 1st floating-point instruction that can not be reassociated.
218  if (I && !ExactFPMathInst)
219  ExactFPMathInst = I;
220  }
221 
222  Instruction *getExactFPInst() { return ExactFPMathInst; }
223 
224 private:
225  Instruction *ExactFPMathInst = nullptr;
226 };
227 
228 /// LoopVectorizationLegality checks if it is legal to vectorize a loop, and
229 /// to what vectorization factor.
230 /// This class does not look at the profitability of vectorization, only the
231 /// legality. This class has two main kinds of checks:
232 /// * Memory checks - The code in canVectorizeMemory checks if vectorization
233 /// will change the order of memory accesses in a way that will change the
234 /// correctness of the program.
235 /// * Scalars checks - The code in canVectorizeInstrs and canVectorizeMemory
236 /// checks for a number of different conditions, such as the availability of a
237 /// single induction variable, that all types are supported and vectorize-able,
238 /// etc. This code reflects the capabilities of InnerLoopVectorizer.
239 /// This class is also used by InnerLoopVectorizer for identifying
240 /// induction variable and the different reduction variables.
242 public:
246  LoopAccessInfoManager &LAIs, LoopInfo *LI,
251  ProfileSummaryInfo *PSI)
252  : TheLoop(L), LI(LI), PSE(PSE), TTI(TTI), TLI(TLI), DT(DT), LAIs(LAIs),
253  ORE(ORE), Requirements(R), Hints(H), DB(DB), AC(AC), BFI(BFI),
254  PSI(PSI) {}
255 
256  /// ReductionList contains the reduction descriptors for all
257  /// of the reductions that were found in the loop.
259 
260  /// InductionList saves induction variables and maps them to the
261  /// induction descriptor.
263 
264  /// RecurrenceSet contains the phi nodes that are recurrences other than
265  /// inductions and reductions.
267 
268  /// Returns true if it is legal to vectorize this loop.
269  /// This does not mean that it is profitable to vectorize this
270  /// loop, only that it is legal to do so.
271  /// Temporarily taking UseVPlanNativePath parameter. If true, take
272  /// the new code path being implemented for outer loop vectorization
273  /// (should be functional for inner loop vectorization) based on VPlan.
274  /// If false, good old LV code.
275  bool canVectorize(bool UseVPlanNativePath);
276 
277  /// Returns true if it is legal to vectorize the FP math operations in this
278  /// loop. Vectorizing is legal if we allow reordering of FP operations, or if
279  /// we can use in-order reductions.
280  bool canVectorizeFPMath(bool EnableStrictReductions);
281 
282  /// Return true if we can vectorize this loop while folding its tail by
283  /// masking, and mark all respective loads/stores for masking.
284  /// This object's state is only modified iff this function returns true.
286 
287  /// Returns the primary induction variable.
288  PHINode *getPrimaryInduction() { return PrimaryInduction; }
289 
290  /// Returns the reduction variables found in the loop.
291  const ReductionList &getReductionVars() const { return Reductions; }
292 
293  /// Returns the induction variables found in the loop.
294  const InductionList &getInductionVars() const { return Inductions; }
295 
296  /// Return the fixed-order recurrences found in the loop.
297  RecurrenceSet &getFixedOrderRecurrences() { return FixedOrderRecurrences; }
298 
299  /// Return the set of instructions to sink to handle fixed-order recurrences.
301 
302  /// Returns the widest induction type.
303  Type *getWidestInductionType() { return WidestIndTy; }
304 
305  /// Returns True if given store is a final invariant store of one of the
306  /// reductions found in the loop.
308 
309  /// Returns True if given address is invariant and is used to store recurrent
310  /// expression
312 
313  /// Returns True if V is a Phi node of an induction variable in this loop.
314  bool isInductionPhi(const Value *V) const;
315 
316  /// Returns a pointer to the induction descriptor, if \p Phi is an integer or
317  /// floating point induction.
319 
320  /// Returns a pointer to the induction descriptor, if \p Phi is pointer
321  /// induction.
323 
324  /// Returns True if V is a cast that is part of an induction def-use chain,
325  /// and had been proven to be redundant under a runtime guard (in other
326  /// words, the cast has the same SCEV expression as the induction phi).
327  bool isCastedInductionVariable(const Value *V) const;
328 
329  /// Returns True if V can be considered as an induction variable in this
330  /// loop. V can be the induction phi, or some redundant cast in the def-use
331  /// chain of the inducion phi.
332  bool isInductionVariable(const Value *V) const;
333 
334  /// Returns True if PN is a reduction variable in this loop.
335  bool isReductionVariable(PHINode *PN) const { return Reductions.count(PN); }
336 
337  /// Returns True if Phi is a fixed-order recurrence in this loop.
338  bool isFixedOrderRecurrence(const PHINode *Phi) const;
339 
340  /// Return true if the block BB needs to be predicated in order for the loop
341  /// to be vectorized.
342  bool blockNeedsPredication(BasicBlock *BB) const;
343 
344  /// Check if this pointer is consecutive when vectorizing. This happens
345  /// when the last index of the GEP is the induction variable, or that the
346  /// pointer itself is an induction variable.
347  /// This check allows us to vectorize A[idx] into a wide load/store.
348  /// Returns:
349  /// 0 - Stride is unknown or non-consecutive.
350  /// 1 - Address is consecutive.
351  /// -1 - Address is consecutive, and decreasing.
352  /// NOTE: This method must only be used before modifying the original scalar
353  /// loop. Do not use after invoking 'createVectorizedLoopSkeleton' (PR34965).
354  int isConsecutivePtr(Type *AccessTy, Value *Ptr) const;
355 
356  /// Returns true if the value V is uniform within the loop.
357  bool isUniform(Value *V) const;
358 
359  /// A uniform memory op is a load or store which accesses the same memory
360  /// location on all lanes.
361  bool isUniformMemOp(Instruction &I) const;
362 
363  /// Returns the information that we collected about runtime memory check.
365  return LAI->getRuntimePointerChecking();
366  }
367 
368  const LoopAccessInfo *getLAI() const { return LAI; }
369 
370  bool isSafeForAnyVectorWidth() const {
371  return LAI->getDepChecker().isSafeForAnyVectorWidth();
372  }
373 
374  unsigned getMaxSafeDepDistBytes() { return LAI->getMaxSafeDepDistBytes(); }
375 
378  }
379 
380  bool hasStride(Value *V) { return LAI->hasStride(V); }
381 
382  /// Returns true if vector representation of the instruction \p I
383  /// requires mask.
384  bool isMaskRequired(const Instruction *I) const {
385  return MaskedOp.contains(I);
386  }
387 
388  unsigned getNumStores() const { return LAI->getNumStores(); }
389  unsigned getNumLoads() const { return LAI->getNumLoads(); }
390 
391  /// Returns all assume calls in predicated blocks. They need to be dropped
392  /// when flattening the CFG.
394  return ConditionalAssumes;
395  }
396 
397 private:
398  /// Return true if the pre-header, exiting and latch blocks of \p Lp and all
399  /// its nested loops are considered legal for vectorization. These legal
400  /// checks are common for inner and outer loop vectorization.
401  /// Temporarily taking UseVPlanNativePath parameter. If true, take
402  /// the new code path being implemented for outer loop vectorization
403  /// (should be functional for inner loop vectorization) based on VPlan.
404  /// If false, good old LV code.
405  bool canVectorizeLoopNestCFG(Loop *Lp, bool UseVPlanNativePath);
406 
407  /// Set up outer loop inductions by checking Phis in outer loop header for
408  /// supported inductions (int inductions). Return false if any of these Phis
409  /// is not a supported induction or if we fail to find an induction.
410  bool setupOuterLoopInductions();
411 
412  /// Return true if the pre-header, exiting and latch blocks of \p Lp
413  /// (non-recursive) are considered legal for vectorization.
414  /// Temporarily taking UseVPlanNativePath parameter. If true, take
415  /// the new code path being implemented for outer loop vectorization
416  /// (should be functional for inner loop vectorization) based on VPlan.
417  /// If false, good old LV code.
418  bool canVectorizeLoopCFG(Loop *Lp, bool UseVPlanNativePath);
419 
420  /// Check if a single basic block loop is vectorizable.
421  /// At this point we know that this is a loop with a constant trip count
422  /// and we only need to check individual instructions.
423  bool canVectorizeInstrs();
424 
425  /// When we vectorize loops we may change the order in which
426  /// we read and write from memory. This method checks if it is
427  /// legal to vectorize the code, considering only memory constrains.
428  /// Returns true if the loop is vectorizable
429  bool canVectorizeMemory();
430 
431  /// Return true if we can vectorize this loop using the IF-conversion
432  /// transformation.
433  bool canVectorizeWithIfConvert();
434 
435  /// Return true if we can vectorize this outer loop. The method performs
436  /// specific checks for outer loop vectorization.
437  bool canVectorizeOuterLoop();
438 
439  /// Return true if all of the instructions in the block can be speculatively
440  /// executed, and record the loads/stores that require masking.
441  /// \p SafePtrs is a list of addresses that are known to be legal and we know
442  /// that we can read from them without segfault.
443  /// \p MaskedOp is a list of instructions that have to be transformed into
444  /// calls to the appropriate masked intrinsic when the loop is vectorized.
445  /// \p ConditionalAssumes is a list of assume instructions in predicated
446  /// blocks that must be dropped if the CFG gets flattened.
447  bool blockCanBePredicated(
450  SmallPtrSetImpl<Instruction *> &ConditionalAssumes) const;
451 
452  /// Updates the vectorization state by adding \p Phi to the inductions list.
453  /// This can set \p Phi as the main induction of the loop if \p Phi is a
454  /// better choice for the main induction than the existing one.
455  void addInductionPhi(PHINode *Phi, const InductionDescriptor &ID,
456  SmallPtrSetImpl<Value *> &AllowedExit);
457 
458  /// If an access has a symbolic strides, this maps the pointer value to
459  /// the stride symbol.
460  const ValueToValueMap *getSymbolicStrides() const {
461  // FIXME: Currently, the set of symbolic strides is sometimes queried before
462  // it's collected. This happens from canVectorizeWithIfConvert, when the
463  // pointer is checked to reference consecutive elements suitable for a
464  // masked access.
465  return LAI ? &LAI->getSymbolicStrides() : nullptr;
466  }
467 
468  /// The loop that we evaluate.
469  Loop *TheLoop;
470 
471  /// Loop Info analysis.
472  LoopInfo *LI;
473 
474  /// A wrapper around ScalarEvolution used to add runtime SCEV checks.
475  /// Applies dynamic knowledge to simplify SCEV expressions in the context
476  /// of existing SCEV assumptions. The analysis will also add a minimal set
477  /// of new predicates if this is required to enable vectorization and
478  /// unrolling.
479  PredicatedScalarEvolution &PSE;
480 
481  /// Target Transform Info.
482  TargetTransformInfo *TTI;
483 
484  /// Target Library Info.
485  TargetLibraryInfo *TLI;
486 
487  /// Dominator Tree.
488  DominatorTree *DT;
489 
490  // LoopAccess analysis.
491  LoopAccessInfoManager &LAIs;
492 
493  const LoopAccessInfo *LAI = nullptr;
494 
495  /// Interface to emit optimization remarks.
496  OptimizationRemarkEmitter *ORE;
497 
498  // --- vectorization state --- //
499 
500  /// Holds the primary induction variable. This is the counter of the
501  /// loop.
502  PHINode *PrimaryInduction = nullptr;
503 
504  /// Holds the reduction variables.
505  ReductionList Reductions;
506 
507  /// Holds all of the induction variables that we found in the loop.
508  /// Notice that inductions don't need to start at zero and that induction
509  /// variables can be pointers.
510  InductionList Inductions;
511 
512  /// Holds all the casts that participate in the update chain of the induction
513  /// variables, and that have been proven to be redundant (possibly under a
514  /// runtime guard). These casts can be ignored when creating the vectorized
515  /// loop body.
516  SmallPtrSet<Instruction *, 4> InductionCastsToIgnore;
517 
518  /// Holds the phi nodes that are fixed-order recurrences.
519  RecurrenceSet FixedOrderRecurrences;
520 
521  /// Holds instructions that need to sink past other instructions to handle
522  /// fixed-order recurrences.
523  MapVector<Instruction *, Instruction *> SinkAfter;
524 
525  /// Holds the widest induction type encountered.
526  Type *WidestIndTy = nullptr;
527 
528  /// Allowed outside users. This holds the variables that can be accessed from
529  /// outside the loop.
530  SmallPtrSet<Value *, 4> AllowedExit;
531 
532  /// Vectorization requirements that will go through late-evaluation.
533  LoopVectorizationRequirements *Requirements;
534 
535  /// Used to emit an analysis of any legality issues.
536  LoopVectorizeHints *Hints;
537 
538  /// The demanded bits analysis is used to compute the minimum type size in
539  /// which a reduction can be computed.
540  DemandedBits *DB;
541 
542  /// The assumption cache analysis is used to compute the minimum type size in
543  /// which a reduction can be computed.
544  AssumptionCache *AC;
545 
546  /// While vectorizing these instructions we have to generate a
547  /// call to the appropriate masked intrinsic
548  SmallPtrSet<const Instruction *, 8> MaskedOp;
549 
550  /// Assume instructions in predicated blocks must be dropped if the CFG gets
551  /// flattened.
552  SmallPtrSet<Instruction *, 8> ConditionalAssumes;
553 
554  /// BFI and PSI are used to check for profile guided size optimizations.
555  BlockFrequencyInfo *BFI;
556  ProfileSummaryInfo *PSI;
557 };
558 
559 } // namespace llvm
560 
561 #endif // LLVM_TRANSFORMS_VECTORIZE_LOOPVECTORIZATIONLEGALITY_H
llvm::LoopVectorizationLegality::RecurrenceSet
SmallPtrSet< const PHINode *, 8 > RecurrenceSet
RecurrenceSet contains the phi nodes that are recurrences other than inductions and reductions.
Definition: LoopVectorizationLegality.h:266
llvm::hasUnrollTransformation
TransformationMode hasUnrollTransformation(const Loop *L)
Definition: LoopUtils.cpp:353
llvm::LoopVectorizeHints::setPotentiallyUnsafe
void setPotentiallyUnsafe()
Definition: LoopVectorizationLegality.h:186
llvm::LoopVectorizationLegality::getReductionVars
const ReductionList & getReductionVars() const
Returns the reduction variables found in the loop.
Definition: LoopVectorizationLegality.h:291
llvm
This is an optimization pass for GlobalISel generic memory operations.
Definition: AddressRanges.h:18
llvm::LoopVectorizationLegality::getSinkAfter
MapVector< Instruction *, Instruction * > & getSinkAfter()
Return the set of instructions to sink to handle fixed-order recurrences.
Definition: LoopVectorizationLegality.h:300
llvm::LoopVectorizationLegality::getNumLoads
unsigned getNumLoads() const
Definition: LoopVectorizationLegality.h:389
llvm::LoopAccessInfoManager
Definition: LoopAccessAnalysis.h:767
llvm::ElementCount
Definition: TypeSize.h:404
llvm::hasDisableAllTransformsHint
bool hasDisableAllTransformsHint(const Loop *L)
Look for the loop attribute that disables all transformation heuristic.
Definition: LoopUtils.cpp:345
llvm::LoopVectorizationLegality::isFixedOrderRecurrence
bool isFixedOrderRecurrence(const PHINode *Phi) const
Returns True if Phi is a fixed-order recurrence in this loop.
Definition: LoopVectorizationLegality.cpp:1100
llvm::LoopVectorizationLegality::LoopVectorizationLegality
LoopVectorizationLegality(Loop *L, PredicatedScalarEvolution &PSE, DominatorTree *DT, TargetTransformInfo *TTI, TargetLibraryInfo *TLI, AAResults *AA, Function *F, LoopAccessInfoManager &LAIs, LoopInfo *LI, OptimizationRemarkEmitter *ORE, LoopVectorizationRequirements *R, LoopVectorizeHints *H, DemandedBits *DB, AssumptionCache *AC, BlockFrequencyInfo *BFI, ProfileSummaryInfo *PSI)
Definition: LoopVectorizationLegality.h:243
llvm::Function
Definition: Function.h:60
llvm::Loop
Represents a single loop in the control flow graph.
Definition: LoopInfo.h:547
llvm::LoopVectorizationLegality::getLAI
const LoopAccessInfo * getLAI() const
Definition: LoopVectorizationLegality.h:368
llvm::LoopVectorizeHints::getPredicate
unsigned getPredicate() const
Definition: LoopVectorizationLegality.h:153
TypeSize.h
llvm::PredicatedScalarEvolution
An interface layer with SCEV used to manage how we see SCEV expressions for values in the context of ...
Definition: ScalarEvolution.h:2216
llvm::TargetTransformInfo
This pass provides access to the codegen interfaces that are needed for IR-level transformations.
Definition: TargetTransformInfo.h:172
LoopAccessAnalysis.h
MapVector.h
llvm::LoopVectorizeHints::SK_Unspecified
@ SK_Unspecified
Not selected.
Definition: LoopVectorizationLegality.h:116
llvm::LoopVectorizationLegality
LoopVectorizationLegality checks if it is legal to vectorize a loop, and to what vectorization factor...
Definition: LoopVectorizationLegality.h:241
llvm::DominatorTree
Concrete subclass of DominatorTreeBase that is used to compute a normal dominator tree.
Definition: Dominators.h:166
llvm::MemoryDepChecker::getMaxSafeVectorWidthInBits
uint64_t getMaxSafeVectorWidthInBits() const
Return the number of elements that are safe to operate on simultaneously, multiplied by the size of t...
Definition: LoopAccessAnalysis.h:206
llvm::LoopVectorizeHints::SK_FixedWidthOnly
@ SK_FixedWidthOnly
Disables vectorization with scalable vectors.
Definition: LoopVectorizationLegality.h:118
llvm::Type
The instances of the Type class are immutable: once they are created, they are never changed.
Definition: Type.h:45
llvm::LoopVectorizeHints::getWidth
ElementCount getWidth() const
Definition: LoopVectorizationLegality.h:138
llvm::LoopVectorizationLegality::canVectorizeFPMath
bool canVectorizeFPMath(bool EnableStrictReductions)
Returns true if it is legal to vectorize the FP math operations in this loop.
Definition: LoopVectorizationLegality.cpp:1014
llvm::LoopVectorizeHints::getIsVectorized
unsigned getIsVectorized() const
Definition: LoopVectorizationLegality.h:152
llvm::LoopVectorizeHints::LoopVectorizeHints
LoopVectorizeHints(const Loop *L, bool InterleaveOnlyWhenForced, OptimizationRemarkEmitter &ORE, const TargetTransformInfo *TTI=nullptr)
Definition: LoopVectorizationLegality.cpp:97
llvm::MapVector< PHINode *, RecurrenceDescriptor >
llvm::SmallPtrSet< const PHINode *, 8 >
llvm::RISCVFeatures::validate
void validate(const Triple &TT, const FeatureBitset &FeatureBits)
Definition: RISCVBaseInfo.cpp:97
llvm::LoopVectorizationLegality::prepareToFoldTailByMasking
bool prepareToFoldTailByMasking()
Return true if we can vectorize this loop while folding its tail by masking, and mark all respective ...
Definition: LoopVectorizationLegality.cpp:1383
F
#define F(x, y, z)
Definition: MD5.cpp:55
llvm::BasicBlock
LLVM Basic Block Representation.
Definition: BasicBlock.h:55
llvm::LoopVectorizationRequirements
This holds vectorization requirements that must be verified late in the process.
Definition: LoopVectorizationLegality.h:214
llvm::LoopVectorizeHints::FK_Undefined
@ FK_Undefined
Not selected.
Definition: LoopVectorizationLegality.h:109
Arg
amdgpu Simplify well known AMD library false FunctionCallee Value * Arg
Definition: AMDGPULibCalls.cpp:187
llvm::BlockFrequencyInfo
BlockFrequencyInfo pass uses BlockFrequencyInfoImpl implementation to estimate IR basic block frequen...
Definition: BlockFrequencyInfo.h:37
llvm::InductionDescriptor
A struct for saving information about induction variables.
Definition: IVDescriptors.h:306
llvm::LinearPolySize< ElementCount >::get
static ElementCount get(ScalarTy MinVal, bool Scalable)
Definition: TypeSize.h:289
llvm::LoopVectorizationLegality::ReductionList
MapVector< PHINode *, RecurrenceDescriptor > ReductionList
ReductionList contains the reduction descriptors for all of the reductions that were found in the loo...
Definition: LoopVectorizationLegality.h:258
llvm::AAResults
Definition: AliasAnalysis.h:294
llvm::LoopVectorizationLegality::isInvariantStoreOfReduction
bool isInvariantStoreOfReduction(StoreInst *SI)
Returns True if given store is a final invariant store of one of the reductions found in the loop.
Definition: LoopVectorizationLegality.cpp:1040
llvm::LoopAccessInfo::getNumLoads
unsigned getNumLoads() const
Definition: LoopAccessAnalysis.h:595
SI
@ SI
Definition: SIInstrInfo.cpp:7882
llvm::LoopVectorizeHints::vectorizeAnalysisPassName
const char * vectorizeAnalysisPassName() const
If hints are provided that force vectorization, use the AlwaysPrint pass name to force the frontend t...
Definition: LoopVectorizationLegality.cpp:231
llvm::LoopVectorizationRequirements::addExactFPMathInst
void addExactFPMathInst(Instruction *I)
Track the 1st floating-point instruction that can not be reassociated.
Definition: LoopVectorizationLegality.h:217
llvm::LoopVectorizationLegality::blockNeedsPredication
bool blockNeedsPredication(BasicBlock *BB) const
Return true if the block BB needs to be predicated in order for the loop to be vectorized.
Definition: LoopVectorizationLegality.cpp:1105
llvm::LoopVectorizationLegality::getPointerInductionDescriptor
const InductionDescriptor * getPointerInductionDescriptor(PHINode *Phi) const
Returns a pointer to the induction descriptor, if Phi is pointer induction.
Definition: LoopVectorizationLegality.cpp:1081
llvm::LoopVectorizationLegality::isUniformMemOp
bool isUniformMemOp(Instruction &I) const
A uniform memory op is a load or store which accesses the same memory location on all lanes.
Definition: LoopVectorizationLegality.cpp:469
llvm::Instruction
Definition: Instruction.h:42
llvm::LoopVectorizeHints::getForce
enum ForceKind getForce() const
Definition: LoopVectorizationLegality.h:154
llvm::LoopAccessInfo::hasStride
bool hasStride(Value *V) const
Pointer has a symbolic stride.
Definition: LoopAccessAnalysis.h:617
LoopUtils.h
llvm::Metadata
Root of the metadata hierarchy.
Definition: Metadata.h:62
llvm::CallingConv::ID
unsigned ID
LLVM IR allows to use arbitrary numbers as calling convention identifiers.
Definition: CallingConv.h:24
llvm::ProfileSummaryInfo
Analysis providing profile information.
Definition: ProfileSummaryInfo.h:39
llvm::LoopVectorizationLegality::hasStride
bool hasStride(Value *V)
Definition: LoopVectorizationLegality.h:380
llvm::LoopVectorizationLegality::getConditionalAssumes
const SmallPtrSetImpl< Instruction * > & getConditionalAssumes() const
Returns all assume calls in predicated blocks.
Definition: LoopVectorizationLegality.h:393
llvm::StoreInst
An instruction for storing to memory.
Definition: Instructions.h:297
llvm::LoopAccessInfo::getSymbolicStrides
const ValueToValueMap & getSymbolicStrides() const
If an access has a symbolic strides, this maps the pointer value to the stride symbol.
Definition: LoopAccessAnalysis.h:614
llvm::LoopVectorizeHints
Utility class for getting and setting loop vectorizer hints in the form of loop metadata.
Definition: LoopVectorizationLegality.h:61
llvm::LoopVectorizationLegality::isUniform
bool isUniform(Value *V) const
Returns true if the value V is uniform within the loop.
Definition: LoopVectorizationLegality.cpp:465
llvm::LoopVectorizeHints::SK_PreferScalable
@ SK_PreferScalable
Vectorize loops using scalable vectors or fixed-width vectors, but favor scalable vectors when the co...
Definition: LoopVectorizationLegality.h:122
llvm::RuntimePointerChecking
Holds information about the memory runtime legality checks to verify that a group of pointers do not ...
Definition: LoopAccessAnalysis.h:384
llvm::LoopVectorizationLegality::isMaskRequired
bool isMaskRequired(const Instruction *I) const
Returns true if vector representation of the instruction I requires mask.
Definition: LoopVectorizationLegality.h:384
uint64_t
llvm::LoopVectorizeHints::FK_Enabled
@ FK_Enabled
Forcing enabled.
Definition: LoopVectorizationLegality.h:111
llvm::LoopVectorizeHints::getInterleave
unsigned getInterleave() const
Definition: LoopVectorizationLegality.h:143
llvm::LoopVectorizeHints::allowVectorization
bool allowVectorization(Function *F, Loop *L, bool VectorizeOnlyWhenForced) const
Definition: LoopVectorizationLegality.cpp:172
llvm::DenseMap< const Value *, Value * >
llvm::DemandedBits
Definition: DemandedBits.h:40
llvm::LoopAccessInfo::getRuntimePointerChecking
const RuntimePointerChecking * getRuntimePointerChecking() const
Definition: LoopAccessAnalysis.h:575
I
#define I(x, y, z)
Definition: MD5.cpp:58
llvm::LoopAccessInfo
Drive the analysis of memory accesses in the loop.
Definition: LoopAccessAnalysis.h:561
llvm::LoopVectorizationLegality::isInvariantAddressOfReduction
bool isInvariantAddressOfReduction(Value *V)
Returns True if given address is invariant and is used to store recurrent expression.
Definition: LoopVectorizationLegality.cpp:1047
llvm::LoopVectorizationLegality::getFixedOrderRecurrences
RecurrenceSet & getFixedOrderRecurrences()
Return the fixed-order recurrences found in the loop.
Definition: LoopVectorizationLegality.h:297
TemplateParamKind::Type
@ Type
llvm::TM_Disable
@ TM_Disable
The transformation should not be applied.
Definition: LoopUtils.h:277
llvm::LoopVectorizationLegality::isCastedInductionVariable
bool isCastedInductionVariable(const Value *V) const
Returns True if V is a cast that is part of an induction def-use chain, and had been proven to be red...
Definition: LoopVectorizationLegality.cpp:1090
llvm::LoopVectorizationLegality::getNumStores
unsigned getNumStores() const
Definition: LoopVectorizationLegality.h:388
llvm::LoopVectorizationLegality::isInductionVariable
bool isInductionVariable(const Value *V) const
Returns True if V can be considered as an induction variable in this loop.
Definition: LoopVectorizationLegality.cpp:1096
Ptr
@ Ptr
Definition: TargetLibraryInfo.cpp:60
llvm::ISD::BasicBlock
@ BasicBlock
Various leaf nodes.
Definition: ISDOpcodes.h:71
llvm::LoopVectorizationRequirements::getExactFPInst
Instruction * getExactFPInst()
Definition: LoopVectorizationLegality.h:222
llvm::LoopAccessInfo::getDepChecker
const MemoryDepChecker & getDepChecker() const
the Memory Dependence Checker which can determine the loop-independent and loop-carried dependences b...
Definition: LoopAccessAnalysis.h:603
llvm::LoopInfo
Definition: LoopInfo.h:1108
llvm::OptimizationRemarkEmitter
The optimization diagnostic interface.
Definition: OptimizationRemarkEmitter.h:33
llvm::LoopVectorizationLegality::isConsecutivePtr
int isConsecutivePtr(Type *AccessTy, Value *Ptr) const
Check if this pointer is consecutive when vectorizing.
Definition: LoopVectorizationLegality.cpp:448
llvm::StringRef
StringRef - Represent a constant reference to a string, i.e.
Definition: StringRef.h:50
llvm::AssumptionCache
A cache of @llvm.assume calls within a function.
Definition: AssumptionCache.h:42
llvm::LoopVectorizeHints::ScalableForceKind
ScalableForceKind
Definition: LoopVectorizationLegality.h:114
llvm::LoopVectorizeHints::emitRemarkWithHints
void emitRemarkWithHints() const
Dumps all the hint information.
Definition: LoopVectorizationLegality.cpp:205
llvm::LoopVectorizationLegality::getRuntimePointerChecking
const RuntimePointerChecking * getRuntimePointerChecking() const
Returns the information that we collected about runtime memory check.
Definition: LoopVectorizationLegality.h:364
llvm::LoopVectorizationLegality::getIntOrFpInductionDescriptor
const InductionDescriptor * getIntOrFpInductionDescriptor(PHINode *Phi) const
Returns a pointer to the induction descriptor, if Phi is an integer or floating point induction.
Definition: LoopVectorizationLegality.cpp:1070
llvm::LoopAccessInfo::getMaxSafeDepDistBytes
uint64_t getMaxSafeDepDistBytes() const
Definition: LoopAccessAnalysis.h:593
llvm::LoopVectorizationLegality::isReductionVariable
bool isReductionVariable(PHINode *PN) const
Returns True if PN is a reduction variable in this loop.
Definition: LoopVectorizationLegality.h:335
llvm::LoopAccessInfo::getNumStores
unsigned getNumStores() const
Definition: LoopAccessAnalysis.h:594
llvm::MapVector::count
size_type count(const KeyT &Key) const
Definition: MapVector.h:143
llvm::MemoryDepChecker::isSafeForAnyVectorWidth
bool isSafeForAnyVectorWidth() const
Return true if the number of elements that are safe to operate on simultaneously is not bounded.
Definition: LoopAccessAnalysis.h:196
H
#define H(x, y, z)
Definition: MD5.cpp:57
llvm::LoopVectorizeHints::ForceKind
ForceKind
Definition: LoopVectorizationLegality.h:108
llvm::TargetStackID::Value
Value
Definition: TargetFrameLowering.h:27
llvm::LoopVectorizationLegality::getMaxSafeVectorWidthInBits
uint64_t getMaxSafeVectorWidthInBits() const
Definition: LoopVectorizationLegality.h:376
llvm::TargetLibraryInfo
Provides information about what library functions are available for the current target.
Definition: TargetLibraryInfo.h:225
llvm::LoopVectorizationLegality::isInductionPhi
bool isInductionPhi(const Value *V) const
Returns True if V is a Phi node of an induction variable in this loop.
Definition: LoopVectorizationLegality.cpp:1060
llvm::LoopVectorizeHints::setAlreadyVectorized
void setAlreadyVectorized()
Mark the loop L as already vectorized by setting the width to 1.
Definition: LoopVectorizationLegality.cpp:153
LineType::Metadata
@ Metadata
llvm::LoopVectorizeHints::isScalableVectorizationDisabled
bool isScalableVectorizationDisabled() const
Definition: LoopVectorizationLegality.h:162
llvm::LoopVectorizationLegality::isSafeForAnyVectorWidth
bool isSafeForAnyVectorWidth() const
Definition: LoopVectorizationLegality.h:370
llvm::PHINode
Definition: Instructions.h:2698
llvm::LoopVectorizeHints::allowReordering
bool allowReordering() const
When enabling loop hints are provided we allow the vectorizer to change the order of operations that ...
Definition: LoopVectorizationLegality.cpp:241
llvm::SmallPtrSetImpl< Instruction * >
llvm::LoopVectorizationLegality::getInductionVars
const InductionList & getInductionVars() const
Returns the induction variables found in the loop.
Definition: LoopVectorizationLegality.h:294
BB
Common register allocation spilling lr str ldr sxth r3 ldr mla r4 can lr mov lr str ldr sxth r3 mla r4 and then merge mul and lr str ldr sxth r3 mla r4 It also increase the likelihood the store may become dead bb27 Successors according to LLVM BB
Definition: README.txt:39
llvm::LoopVectorizationLegality::getPrimaryInduction
PHINode * getPrimaryInduction()
Returns the primary induction variable.
Definition: LoopVectorizationLegality.h:288
llvm::LoopVectorizationLegality::getMaxSafeDepDistBytes
unsigned getMaxSafeDepDistBytes()
Definition: LoopVectorizationLegality.h:374
llvm::LoopVectorizeHints::FK_Disabled
@ FK_Disabled
Forcing disabled.
Definition: LoopVectorizationLegality.h:110
llvm::LoopVectorizationLegality::canVectorize
bool canVectorize(bool UseVPlanNativePath)
Returns true if it is legal to vectorize this loop.
Definition: LoopVectorizationLegality.cpp:1289
llvm::Value
LLVM Value Representation.
Definition: Value.h:74
llvm::LoopVectorizationLegality::getWidestInductionType
Type * getWidestInductionType()
Returns the widest induction type.
Definition: LoopVectorizationLegality.h:303
llvm::LoopVectorizeHints::isPotentiallyUnsafe
bool isPotentiallyUnsafe() const
Definition: LoopVectorizationLegality.h:177
llvm::codeview::PublicSymFlags::Function
@ Function
llvm::LoopVectorizationLegality::InductionList
MapVector< PHINode *, InductionDescriptor > InductionList
InductionList saves induction variables and maps them to the induction descriptor.
Definition: LoopVectorizationLegality.h:262