LLVM  16.0.0git
NewGVN.cpp
Go to the documentation of this file.
1 //===- NewGVN.cpp - Global Value Numbering Pass ---------------------------===//
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 implements the new LLVM's Global Value Numbering pass.
11 /// GVN partitions values computed by a function into congruence classes.
12 /// Values ending up in the same congruence class are guaranteed to be the same
13 /// for every execution of the program. In that respect, congruency is a
14 /// compile-time approximation of equivalence of values at runtime.
15 /// The algorithm implemented here uses a sparse formulation and it's based
16 /// on the ideas described in the paper:
17 /// "A Sparse Algorithm for Predicated Global Value Numbering" from
18 /// Karthik Gargi.
19 ///
20 /// A brief overview of the algorithm: The algorithm is essentially the same as
21 /// the standard RPO value numbering algorithm (a good reference is the paper
22 /// "SCC based value numbering" by L. Taylor Simpson) with one major difference:
23 /// The RPO algorithm proceeds, on every iteration, to process every reachable
24 /// block and every instruction in that block. This is because the standard RPO
25 /// algorithm does not track what things have the same value number, it only
26 /// tracks what the value number of a given operation is (the mapping is
27 /// operation -> value number). Thus, when a value number of an operation
28 /// changes, it must reprocess everything to ensure all uses of a value number
29 /// get updated properly. In constrast, the sparse algorithm we use *also*
30 /// tracks what operations have a given value number (IE it also tracks the
31 /// reverse mapping from value number -> operations with that value number), so
32 /// that it only needs to reprocess the instructions that are affected when
33 /// something's value number changes. The vast majority of complexity and code
34 /// in this file is devoted to tracking what value numbers could change for what
35 /// instructions when various things happen. The rest of the algorithm is
36 /// devoted to performing symbolic evaluation, forward propagation, and
37 /// simplification of operations based on the value numbers deduced so far
38 ///
39 /// In order to make the GVN mostly-complete, we use a technique derived from
40 /// "Detection of Redundant Expressions: A Complete and Polynomial-time
41 /// Algorithm in SSA" by R.R. Pai. The source of incompleteness in most SSA
42 /// based GVN algorithms is related to their inability to detect equivalence
43 /// between phi of ops (IE phi(a+b, c+d)) and op of phis (phi(a,c) + phi(b, d)).
44 /// We resolve this issue by generating the equivalent "phi of ops" form for
45 /// each op of phis we see, in a way that only takes polynomial time to resolve.
46 ///
47 /// We also do not perform elimination by using any published algorithm. All
48 /// published algorithms are O(Instructions). Instead, we use a technique that
49 /// is O(number of operations with the same value number), enabling us to skip
50 /// trying to eliminate things that have unique value numbers.
51 //
52 //===----------------------------------------------------------------------===//
53 
55 #include "llvm/ADT/ArrayRef.h"
56 #include "llvm/ADT/BitVector.h"
57 #include "llvm/ADT/DenseMap.h"
58 #include "llvm/ADT/DenseMapInfo.h"
59 #include "llvm/ADT/DenseSet.h"
61 #include "llvm/ADT/GraphTraits.h"
62 #include "llvm/ADT/Hashing.h"
65 #include "llvm/ADT/SetOperations.h"
66 #include "llvm/ADT/SmallPtrSet.h"
67 #include "llvm/ADT/SmallVector.h"
69 #include "llvm/ADT/Statistic.h"
81 #include "llvm/IR/Argument.h"
82 #include "llvm/IR/BasicBlock.h"
83 #include "llvm/IR/Constant.h"
84 #include "llvm/IR/Constants.h"
85 #include "llvm/IR/Dominators.h"
86 #include "llvm/IR/Function.h"
87 #include "llvm/IR/InstrTypes.h"
88 #include "llvm/IR/Instruction.h"
89 #include "llvm/IR/Instructions.h"
90 #include "llvm/IR/IntrinsicInst.h"
91 #include "llvm/IR/PatternMatch.h"
92 #include "llvm/IR/Type.h"
93 #include "llvm/IR/Use.h"
94 #include "llvm/IR/User.h"
95 #include "llvm/IR/Value.h"
96 #include "llvm/InitializePasses.h"
97 #include "llvm/Pass.h"
98 #include "llvm/Support/Allocator.h"
100 #include "llvm/Support/Casting.h"
102 #include "llvm/Support/Debug.h"
107 #include "llvm/Transforms/Scalar.h"
113 #include <algorithm>
114 #include <cassert>
115 #include <cstdint>
116 #include <iterator>
117 #include <map>
118 #include <memory>
119 #include <set>
120 #include <string>
121 #include <tuple>
122 #include <utility>
123 #include <vector>
124 
125 using namespace llvm;
126 using namespace llvm::GVNExpression;
127 using namespace llvm::VNCoercion;
128 using namespace llvm::PatternMatch;
129 
130 #define DEBUG_TYPE "newgvn"
131 
132 STATISTIC(NumGVNInstrDeleted, "Number of instructions deleted");
133 STATISTIC(NumGVNBlocksDeleted, "Number of blocks deleted");
134 STATISTIC(NumGVNOpsSimplified, "Number of Expressions simplified");
135 STATISTIC(NumGVNPhisAllSame, "Number of PHIs whos arguments are all the same");
136 STATISTIC(NumGVNMaxIterations,
137  "Maximum Number of iterations it took to converge GVN");
138 STATISTIC(NumGVNLeaderChanges, "Number of leader changes");
139 STATISTIC(NumGVNSortedLeaderChanges, "Number of sorted leader changes");
140 STATISTIC(NumGVNAvoidedSortedLeaderChanges,
141  "Number of avoided sorted leader changes");
142 STATISTIC(NumGVNDeadStores, "Number of redundant/dead stores eliminated");
143 STATISTIC(NumGVNPHIOfOpsCreated, "Number of PHI of ops created");
144 STATISTIC(NumGVNPHIOfOpsEliminations,
145  "Number of things eliminated using PHI of ops");
146 DEBUG_COUNTER(VNCounter, "newgvn-vn",
147  "Controls which instructions are value numbered");
148 DEBUG_COUNTER(PHIOfOpsCounter, "newgvn-phi",
149  "Controls which instructions we create phi of ops for");
150 // Currently store defining access refinement is too slow due to basicaa being
151 // egregiously slow. This flag lets us keep it working while we work on this
152 // issue.
153 static cl::opt<bool> EnableStoreRefinement("enable-store-refinement",
154  cl::init(false), cl::Hidden);
155 
156 /// Currently, the generation "phi of ops" can result in correctness issues.
157 static cl::opt<bool> EnablePhiOfOps("enable-phi-of-ops", cl::init(true),
158  cl::Hidden);
159 
160 //===----------------------------------------------------------------------===//
161 // GVN Pass
162 //===----------------------------------------------------------------------===//
163 
164 // Anchor methods.
165 namespace llvm {
166 namespace GVNExpression {
167 
168 Expression::~Expression() = default;
175 
176 } // end namespace GVNExpression
177 } // end namespace llvm
178 
179 namespace {
180 
181 // Tarjan's SCC finding algorithm with Nuutila's improvements
182 // SCCIterator is actually fairly complex for the simple thing we want.
183 // It also wants to hand us SCC's that are unrelated to the phi node we ask
184 // about, and have us process them there or risk redoing work.
185 // Graph traits over a filter iterator also doesn't work that well here.
186 // This SCC finder is specialized to walk use-def chains, and only follows
187 // instructions,
188 // not generic values (arguments, etc).
189 struct TarjanSCC {
190  TarjanSCC() : Components(1) {}
191 
192  void Start(const Instruction *Start) {
193  if (Root.lookup(Start) == 0)
194  FindSCC(Start);
195  }
196 
197  const SmallPtrSetImpl<const Value *> &getComponentFor(const Value *V) const {
198  unsigned ComponentID = ValueToComponent.lookup(V);
199 
200  assert(ComponentID > 0 &&
201  "Asking for a component for a value we never processed");
202  return Components[ComponentID];
203  }
204 
205 private:
206  void FindSCC(const Instruction *I) {
207  Root[I] = ++DFSNum;
208  // Store the DFS Number we had before it possibly gets incremented.
209  unsigned int OurDFS = DFSNum;
210  for (const auto &Op : I->operands()) {
211  if (auto *InstOp = dyn_cast<Instruction>(Op)) {
212  if (Root.lookup(Op) == 0)
213  FindSCC(InstOp);
214  if (!InComponent.count(Op))
215  Root[I] = std::min(Root.lookup(I), Root.lookup(Op));
216  }
217  }
218  // See if we really were the root of a component, by seeing if we still have
219  // our DFSNumber. If we do, we are the root of the component, and we have
220  // completed a component. If we do not, we are not the root of a component,
221  // and belong on the component stack.
222  if (Root.lookup(I) == OurDFS) {
223  unsigned ComponentID = Components.size();
224  Components.resize(Components.size() + 1);
225  auto &Component = Components.back();
226  Component.insert(I);
227  LLVM_DEBUG(dbgs() << "Component root is " << *I << "\n");
228  InComponent.insert(I);
229  ValueToComponent[I] = ComponentID;
230  // Pop a component off the stack and label it.
231  while (!Stack.empty() && Root.lookup(Stack.back()) >= OurDFS) {
232  auto *Member = Stack.back();
233  LLVM_DEBUG(dbgs() << "Component member is " << *Member << "\n");
234  Component.insert(Member);
235  InComponent.insert(Member);
236  ValueToComponent[Member] = ComponentID;
237  Stack.pop_back();
238  }
239  } else {
240  // Part of a component, push to stack
241  Stack.push_back(I);
242  }
243  }
244 
245  unsigned int DFSNum = 1;
246  SmallPtrSet<const Value *, 8> InComponent;
249 
250  // Store the components as vector of ptr sets, because we need the topo order
251  // of SCC's, but not individual member order
253 
254  DenseMap<const Value *, unsigned> ValueToComponent;
255 };
256 
257 // Congruence classes represent the set of expressions/instructions
258 // that are all the same *during some scope in the function*.
259 // That is, because of the way we perform equality propagation, and
260 // because of memory value numbering, it is not correct to assume
261 // you can willy-nilly replace any member with any other at any
262 // point in the function.
263 //
264 // For any Value in the Member set, it is valid to replace any dominated member
265 // with that Value.
266 //
267 // Every congruence class has a leader, and the leader is used to symbolize
268 // instructions in a canonical way (IE every operand of an instruction that is a
269 // member of the same congruence class will always be replaced with leader
270 // during symbolization). To simplify symbolization, we keep the leader as a
271 // constant if class can be proved to be a constant value. Otherwise, the
272 // leader is the member of the value set with the smallest DFS number. Each
273 // congruence class also has a defining expression, though the expression may be
274 // null. If it exists, it can be used for forward propagation and reassociation
275 // of values.
276 
277 // For memory, we also track a representative MemoryAccess, and a set of memory
278 // members for MemoryPhis (which have no real instructions). Note that for
279 // memory, it seems tempting to try to split the memory members into a
280 // MemoryCongruenceClass or something. Unfortunately, this does not work
281 // easily. The value numbering of a given memory expression depends on the
282 // leader of the memory congruence class, and the leader of memory congruence
283 // class depends on the value numbering of a given memory expression. This
284 // leads to wasted propagation, and in some cases, missed optimization. For
285 // example: If we had value numbered two stores together before, but now do not,
286 // we move them to a new value congruence class. This in turn will move at one
287 // of the memorydefs to a new memory congruence class. Which in turn, affects
288 // the value numbering of the stores we just value numbered (because the memory
289 // congruence class is part of the value number). So while theoretically
290 // possible to split them up, it turns out to be *incredibly* complicated to get
291 // it to work right, because of the interdependency. While structurally
292 // slightly messier, it is algorithmically much simpler and faster to do what we
293 // do here, and track them both at once in the same class.
294 // Note: The default iterators for this class iterate over values
295 class CongruenceClass {
296 public:
297  using MemberType = Value;
298  using MemberSet = SmallPtrSet<MemberType *, 4>;
299  using MemoryMemberType = MemoryPhi;
300  using MemoryMemberSet = SmallPtrSet<const MemoryMemberType *, 2>;
301 
302  explicit CongruenceClass(unsigned ID) : ID(ID) {}
303  CongruenceClass(unsigned ID, Value *Leader, const Expression *E)
304  : ID(ID), RepLeader(Leader), DefiningExpr(E) {}
305 
306  unsigned getID() const { return ID; }
307 
308  // True if this class has no members left. This is mainly used for assertion
309  // purposes, and for skipping empty classes.
310  bool isDead() const {
311  // If it's both dead from a value perspective, and dead from a memory
312  // perspective, it's really dead.
313  return empty() && memory_empty();
314  }
315 
316  // Leader functions
317  Value *getLeader() const { return RepLeader; }
318  void setLeader(Value *Leader) { RepLeader = Leader; }
319  const std::pair<Value *, unsigned int> &getNextLeader() const {
320  return NextLeader;
321  }
322  void resetNextLeader() { NextLeader = {nullptr, ~0}; }
323  void addPossibleNextLeader(std::pair<Value *, unsigned int> LeaderPair) {
324  if (LeaderPair.second < NextLeader.second)
325  NextLeader = LeaderPair;
326  }
327 
328  Value *getStoredValue() const { return RepStoredValue; }
329  void setStoredValue(Value *Leader) { RepStoredValue = Leader; }
330  const MemoryAccess *getMemoryLeader() const { return RepMemoryAccess; }
331  void setMemoryLeader(const MemoryAccess *Leader) { RepMemoryAccess = Leader; }
332 
333  // Forward propagation info
334  const Expression *getDefiningExpr() const { return DefiningExpr; }
335 
336  // Value member set
337  bool empty() const { return Members.empty(); }
338  unsigned size() const { return Members.size(); }
339  MemberSet::const_iterator begin() const { return Members.begin(); }
340  MemberSet::const_iterator end() const { return Members.end(); }
341  void insert(MemberType *M) { Members.insert(M); }
342  void erase(MemberType *M) { Members.erase(M); }
343  void swap(MemberSet &Other) { Members.swap(Other); }
344 
345  // Memory member set
346  bool memory_empty() const { return MemoryMembers.empty(); }
347  unsigned memory_size() const { return MemoryMembers.size(); }
348  MemoryMemberSet::const_iterator memory_begin() const {
349  return MemoryMembers.begin();
350  }
351  MemoryMemberSet::const_iterator memory_end() const {
352  return MemoryMembers.end();
353  }
355  return make_range(memory_begin(), memory_end());
356  }
357 
358  void memory_insert(const MemoryMemberType *M) { MemoryMembers.insert(M); }
359  void memory_erase(const MemoryMemberType *M) { MemoryMembers.erase(M); }
360 
361  // Store count
362  unsigned getStoreCount() const { return StoreCount; }
363  void incStoreCount() { ++StoreCount; }
364  void decStoreCount() {
365  assert(StoreCount != 0 && "Store count went negative");
366  --StoreCount;
367  }
368 
369  // True if this class has no memory members.
370  bool definesNoMemory() const { return StoreCount == 0 && memory_empty(); }
371 
372  // Return true if two congruence classes are equivalent to each other. This
373  // means that every field but the ID number and the dead field are equivalent.
374  bool isEquivalentTo(const CongruenceClass *Other) const {
375  if (!Other)
376  return false;
377  if (this == Other)
378  return true;
379 
380  if (std::tie(StoreCount, RepLeader, RepStoredValue, RepMemoryAccess) !=
381  std::tie(Other->StoreCount, Other->RepLeader, Other->RepStoredValue,
382  Other->RepMemoryAccess))
383  return false;
384  if (DefiningExpr != Other->DefiningExpr)
385  if (!DefiningExpr || !Other->DefiningExpr ||
386  *DefiningExpr != *Other->DefiningExpr)
387  return false;
388 
389  if (Members.size() != Other->Members.size())
390  return false;
391 
392  return llvm::set_is_subset(Members, Other->Members);
393  }
394 
395 private:
396  unsigned ID;
397 
398  // Representative leader.
399  Value *RepLeader = nullptr;
400 
401  // The most dominating leader after our current leader, because the member set
402  // is not sorted and is expensive to keep sorted all the time.
403  std::pair<Value *, unsigned int> NextLeader = {nullptr, ~0U};
404 
405  // If this is represented by a store, the value of the store.
406  Value *RepStoredValue = nullptr;
407 
408  // If this class contains MemoryDefs or MemoryPhis, this is the leading memory
409  // access.
410  const MemoryAccess *RepMemoryAccess = nullptr;
411 
412  // Defining Expression.
413  const Expression *DefiningExpr = nullptr;
414 
415  // Actual members of this class.
416  MemberSet Members;
417 
418  // This is the set of MemoryPhis that exist in the class. MemoryDefs and
419  // MemoryUses have real instructions representing them, so we only need to
420  // track MemoryPhis here.
421  MemoryMemberSet MemoryMembers;
422 
423  // Number of stores in this congruence class.
424  // This is used so we can detect store equivalence changes properly.
425  int StoreCount = 0;
426 };
427 
428 } // end anonymous namespace
429 
430 namespace llvm {
431 
433  const Expression &E;
434 
435  explicit ExactEqualsExpression(const Expression &E) : E(E) {}
436 
437  hash_code getComputedHash() const { return E.getComputedHash(); }
438 
439  bool operator==(const Expression &Other) const {
440  return E.exactlyEquals(Other);
441  }
442 };
443 
444 template <> struct DenseMapInfo<const Expression *> {
445  static const Expression *getEmptyKey() {
446  auto Val = static_cast<uintptr_t>(-1);
447  Val <<= PointerLikeTypeTraits<const Expression *>::NumLowBitsAvailable;
448  return reinterpret_cast<const Expression *>(Val);
449  }
450 
451  static const Expression *getTombstoneKey() {
452  auto Val = static_cast<uintptr_t>(~1U);
453  Val <<= PointerLikeTypeTraits<const Expression *>::NumLowBitsAvailable;
454  return reinterpret_cast<const Expression *>(Val);
455  }
456 
457  static unsigned getHashValue(const Expression *E) {
458  return E->getComputedHash();
459  }
460 
461  static unsigned getHashValue(const ExactEqualsExpression &E) {
462  return E.getComputedHash();
463  }
464 
465  static bool isEqual(const ExactEqualsExpression &LHS, const Expression *RHS) {
466  if (RHS == getTombstoneKey() || RHS == getEmptyKey())
467  return false;
468  return LHS == *RHS;
469  }
470 
471  static bool isEqual(const Expression *LHS, const Expression *RHS) {
472  if (LHS == RHS)
473  return true;
474  if (LHS == getTombstoneKey() || RHS == getTombstoneKey() ||
475  LHS == getEmptyKey() || RHS == getEmptyKey())
476  return false;
477  // Compare hashes before equality. This is *not* what the hashtable does,
478  // since it is computing it modulo the number of buckets, whereas we are
479  // using the full hash keyspace. Since the hashes are precomputed, this
480  // check is *much* faster than equality.
481  if (LHS->getComputedHash() != RHS->getComputedHash())
482  return false;
483  return *LHS == *RHS;
484  }
485 };
486 
487 } // end namespace llvm
488 
489 namespace {
490 
491 class NewGVN {
492  Function &F;
493  DominatorTree *DT = nullptr;
494  const TargetLibraryInfo *TLI = nullptr;
495  AliasAnalysis *AA = nullptr;
496  MemorySSA *MSSA = nullptr;
497  MemorySSAWalker *MSSAWalker = nullptr;
498  AssumptionCache *AC = nullptr;
499  const DataLayout &DL;
500  std::unique_ptr<PredicateInfo> PredInfo;
501 
502  // These are the only two things the create* functions should have
503  // side-effects on due to allocating memory.
504  mutable BumpPtrAllocator ExpressionAllocator;
505  mutable ArrayRecycler<Value *> ArgRecycler;
506  mutable TarjanSCC SCCFinder;
507  const SimplifyQuery SQ;
508 
509  // Number of function arguments, used by ranking
510  unsigned int NumFuncArgs = 0;
511 
512  // RPOOrdering of basic blocks
514 
515  // Congruence class info.
516 
517  // This class is called INITIAL in the paper. It is the class everything
518  // startsout in, and represents any value. Being an optimistic analysis,
519  // anything in the TOP class has the value TOP, which is indeterminate and
520  // equivalent to everything.
521  CongruenceClass *TOPClass = nullptr;
522  std::vector<CongruenceClass *> CongruenceClasses;
523  unsigned NextCongruenceNum = 0;
524 
525  // Value Mappings.
527  DenseMap<Value *, const Expression *> ValueToExpression;
528 
529  // Value PHI handling, used to make equivalence between phi(op, op) and
530  // op(phi, phi).
531  // These mappings just store various data that would normally be part of the
532  // IR.
534 
535  DenseMap<const Value *, bool> OpSafeForPHIOfOps;
536 
537  // Map a temporary instruction we created to a parent block.
539 
540  // Map between the already in-program instructions and the temporary phis we
541  // created that they are known equivalent to.
543 
544  // In order to know when we should re-process instructions that have
545  // phi-of-ops, we track the set of expressions that they needed as
546  // leaders. When we discover new leaders for those expressions, we process the
547  // associated phi-of-op instructions again in case they have changed. The
548  // other way they may change is if they had leaders, and those leaders
549  // disappear. However, at the point they have leaders, there are uses of the
550  // relevant operands in the created phi node, and so they will get reprocessed
551  // through the normal user marking we perform.
552  mutable DenseMap<const Value *, SmallPtrSet<Value *, 2>> AdditionalUsers;
554  ExpressionToPhiOfOps;
555 
556  // Map from temporary operation to MemoryAccess.
558 
559  // Set of all temporary instructions we created.
560  // Note: This will include instructions that were just created during value
561  // numbering. The way to test if something is using them is to check
562  // RealToTemp.
563  DenseSet<Instruction *> AllTempInstructions;
564 
565  // This is the set of instructions to revisit on a reachability change. At
566  // the end of the main iteration loop it will contain at least all the phi of
567  // ops instructions that will be changed to phis, as well as regular phis.
568  // During the iteration loop, it may contain other things, such as phi of ops
569  // instructions that used edge reachability to reach a result, and so need to
570  // be revisited when the edge changes, independent of whether the phi they
571  // depended on changes.
572  DenseMap<BasicBlock *, SparseBitVector<>> RevisitOnReachabilityChange;
573 
574  // Mapping from predicate info we used to the instructions we used it with.
575  // In order to correctly ensure propagation, we must keep track of what
576  // comparisons we used, so that when the values of the comparisons change, we
577  // propagate the information to the places we used the comparison.
579  PredicateToUsers;
580 
581  // the same reasoning as PredicateToUsers. When we skip MemoryAccesses for
582  // stores, we no longer can rely solely on the def-use chains of MemorySSA.
584  MemoryToUsers;
585 
586  // A table storing which memorydefs/phis represent a memory state provably
587  // equivalent to another memory state.
588  // We could use the congruence class machinery, but the MemoryAccess's are
589  // abstract memory states, so they can only ever be equivalent to each other,
590  // and not to constants, etc.
592 
593  // We could, if we wanted, build MemoryPhiExpressions and
594  // MemoryVariableExpressions, etc, and value number them the same way we value
595  // number phi expressions. For the moment, this seems like overkill. They
596  // can only exist in one of three states: they can be TOP (equal to
597  // everything), Equivalent to something else, or unique. Because we do not
598  // create expressions for them, we need to simulate leader change not just
599  // when they change class, but when they change state. Note: We can do the
600  // same thing for phis, and avoid having phi expressions if we wanted, We
601  // should eventually unify in one direction or the other, so this is a little
602  // bit of an experiment in which turns out easier to maintain.
603  enum MemoryPhiState { MPS_Invalid, MPS_TOP, MPS_Equivalent, MPS_Unique };
605 
606  enum InstCycleState { ICS_Unknown, ICS_CycleFree, ICS_Cycle };
607  mutable DenseMap<const Instruction *, InstCycleState> InstCycleState;
608 
609  // Expression to class mapping.
610  using ExpressionClassMap = DenseMap<const Expression *, CongruenceClass *>;
611  ExpressionClassMap ExpressionToClass;
612 
613  // We have a single expression that represents currently DeadExpressions.
614  // For dead expressions we can prove will stay dead, we mark them with
615  // DFS number zero. However, it's possible in the case of phi nodes
616  // for us to assume/prove all arguments are dead during fixpointing.
617  // We use DeadExpression for that case.
618  DeadExpression *SingletonDeadExpression = nullptr;
619 
620  // Which values have changed as a result of leader changes.
621  SmallPtrSet<Value *, 8> LeaderChanges;
622 
623  // Reachability info.
624  using BlockEdge = BasicBlockEdge;
625  DenseSet<BlockEdge> ReachableEdges;
626  SmallPtrSet<const BasicBlock *, 8> ReachableBlocks;
627 
628  // This is a bitvector because, on larger functions, we may have
629  // thousands of touched instructions at once (entire blocks,
630  // instructions with hundreds of uses, etc). Even with optimization
631  // for when we mark whole blocks as touched, when this was a
632  // SmallPtrSet or DenseSet, for some functions, we spent >20% of all
633  // the time in GVN just managing this list. The bitvector, on the
634  // other hand, efficiently supports test/set/clear of both
635  // individual and ranges, as well as "find next element" This
636  // enables us to use it as a worklist with essentially 0 cost.
637  BitVector TouchedInstructions;
638 
640  mutable DenseMap<const IntrinsicInst *, const Value *> IntrinsicInstPred;
641 
642 #ifndef NDEBUG
643  // Debugging for how many times each block and instruction got processed.
644  DenseMap<const Value *, unsigned> ProcessedCount;
645 #endif
646 
647  // DFS info.
648  // This contains a mapping from Instructions to DFS numbers.
649  // The numbering starts at 1. An instruction with DFS number zero
650  // means that the instruction is dead.
652 
653  // This contains the mapping DFS numbers to instructions.
654  SmallVector<Value *, 32> DFSToInstr;
655 
656  // Deletion info.
657  SmallPtrSet<Instruction *, 8> InstructionsToErase;
658 
659 public:
660  NewGVN(Function &F, DominatorTree *DT, AssumptionCache *AC,
662  const DataLayout &DL)
663  : F(F), DT(DT), TLI(TLI), AA(AA), MSSA(MSSA), AC(AC), DL(DL),
664  PredInfo(std::make_unique<PredicateInfo>(F, *DT, *AC)),
665  SQ(DL, TLI, DT, AC, /*CtxI=*/nullptr, /*UseInstrInfo=*/false,
666  /*CanUseUndef=*/false) {}
667 
668  bool runGVN();
669 
670 private:
671  /// Helper struct return a Expression with an optional extra dependency.
672  struct ExprResult {
673  const Expression *Expr;
674  Value *ExtraDep;
675  const PredicateBase *PredDep;
676 
677  ExprResult(const Expression *Expr, Value *ExtraDep = nullptr,
678  const PredicateBase *PredDep = nullptr)
679  : Expr(Expr), ExtraDep(ExtraDep), PredDep(PredDep) {}
680  ExprResult(const ExprResult &) = delete;
681  ExprResult(ExprResult &&Other)
682  : Expr(Other.Expr), ExtraDep(Other.ExtraDep), PredDep(Other.PredDep) {
683  Other.Expr = nullptr;
684  Other.ExtraDep = nullptr;
685  Other.PredDep = nullptr;
686  }
687  ExprResult &operator=(const ExprResult &Other) = delete;
688  ExprResult &operator=(ExprResult &&Other) = delete;
689 
690  ~ExprResult() { assert(!ExtraDep && "unhandled ExtraDep"); }
691 
692  operator bool() const { return Expr; }
693 
694  static ExprResult none() { return {nullptr, nullptr, nullptr}; }
695  static ExprResult some(const Expression *Expr, Value *ExtraDep = nullptr) {
696  return {Expr, ExtraDep, nullptr};
697  }
698  static ExprResult some(const Expression *Expr,
699  const PredicateBase *PredDep) {
700  return {Expr, nullptr, PredDep};
701  }
702  static ExprResult some(const Expression *Expr, Value *ExtraDep,
703  const PredicateBase *PredDep) {
704  return {Expr, ExtraDep, PredDep};
705  }
706  };
707 
708  // Expression handling.
709  ExprResult createExpression(Instruction *) const;
710  const Expression *createBinaryExpression(unsigned, Type *, Value *, Value *,
711  Instruction *) const;
712 
713  // Our canonical form for phi arguments is a pair of incoming value, incoming
714  // basic block.
715  using ValPair = std::pair<Value *, BasicBlock *>;
716 
717  PHIExpression *createPHIExpression(ArrayRef<ValPair>, const Instruction *,
718  BasicBlock *, bool &HasBackEdge,
719  bool &OriginalOpsConstant) const;
720  const DeadExpression *createDeadExpression() const;
721  const VariableExpression *createVariableExpression(Value *) const;
722  const ConstantExpression *createConstantExpression(Constant *) const;
723  const Expression *createVariableOrConstant(Value *V) const;
724  const UnknownExpression *createUnknownExpression(Instruction *) const;
725  const StoreExpression *createStoreExpression(StoreInst *,
726  const MemoryAccess *) const;
727  LoadExpression *createLoadExpression(Type *, Value *, LoadInst *,
728  const MemoryAccess *) const;
729  const CallExpression *createCallExpression(CallInst *,
730  const MemoryAccess *) const;
732  createAggregateValueExpression(Instruction *) const;
733  bool setBasicExpressionInfo(Instruction *, BasicExpression *) const;
734 
735  // Congruence class handling.
736  CongruenceClass *createCongruenceClass(Value *Leader, const Expression *E) {
737  auto *result = new CongruenceClass(NextCongruenceNum++, Leader, E);
738  CongruenceClasses.emplace_back(result);
739  return result;
740  }
741 
742  CongruenceClass *createMemoryClass(MemoryAccess *MA) {
743  auto *CC = createCongruenceClass(nullptr, nullptr);
744  CC->setMemoryLeader(MA);
745  return CC;
746  }
747 
748  CongruenceClass *ensureLeaderOfMemoryClass(MemoryAccess *MA) {
749  auto *CC = getMemoryClass(MA);
750  if (CC->getMemoryLeader() != MA)
751  CC = createMemoryClass(MA);
752  return CC;
753  }
754 
755  CongruenceClass *createSingletonCongruenceClass(Value *Member) {
756  CongruenceClass *CClass = createCongruenceClass(Member, nullptr);
757  CClass->insert(Member);
758  ValueToClass[Member] = CClass;
759  return CClass;
760  }
761 
762  void initializeCongruenceClasses(Function &F);
763  const Expression *makePossiblePHIOfOps(Instruction *,
765  Value *findLeaderForInst(Instruction *ValueOp,
766  SmallPtrSetImpl<Value *> &Visited,
767  MemoryAccess *MemAccess, Instruction *OrigInst,
768  BasicBlock *PredBB);
769  bool OpIsSafeForPHIOfOps(Value *Op, const BasicBlock *PHIBlock,
771  void addPhiOfOps(PHINode *Op, BasicBlock *BB, Instruction *ExistingValue);
772  void removePhiOfOps(Instruction *I, PHINode *PHITemp);
773 
774  // Value number an Instruction or MemoryPhi.
775  void valueNumberMemoryPhi(MemoryPhi *);
776  void valueNumberInstruction(Instruction *);
777 
778  // Symbolic evaluation.
779  ExprResult checkExprResults(Expression *, Instruction *, Value *) const;
780  ExprResult performSymbolicEvaluation(Value *,
781  SmallPtrSetImpl<Value *> &) const;
782  const Expression *performSymbolicLoadCoercion(Type *, Value *, LoadInst *,
783  Instruction *,
784  MemoryAccess *) const;
785  const Expression *performSymbolicLoadEvaluation(Instruction *) const;
786  const Expression *performSymbolicStoreEvaluation(Instruction *) const;
787  ExprResult performSymbolicCallEvaluation(Instruction *) const;
788  void sortPHIOps(MutableArrayRef<ValPair> Ops) const;
789  const Expression *performSymbolicPHIEvaluation(ArrayRef<ValPair>,
790  Instruction *I,
791  BasicBlock *PHIBlock) const;
792  const Expression *performSymbolicAggrValueEvaluation(Instruction *) const;
793  ExprResult performSymbolicCmpEvaluation(Instruction *) const;
794  ExprResult performSymbolicPredicateInfoEvaluation(IntrinsicInst *) const;
795 
796  // Congruence finding.
797  bool someEquivalentDominates(const Instruction *, const Instruction *) const;
798  Value *lookupOperandLeader(Value *) const;
799  CongruenceClass *getClassForExpression(const Expression *E) const;
800  void performCongruenceFinding(Instruction *, const Expression *);
801  void moveValueToNewCongruenceClass(Instruction *, const Expression *,
802  CongruenceClass *, CongruenceClass *);
803  void moveMemoryToNewCongruenceClass(Instruction *, MemoryAccess *,
804  CongruenceClass *, CongruenceClass *);
805  Value *getNextValueLeader(CongruenceClass *) const;
806  const MemoryAccess *getNextMemoryLeader(CongruenceClass *) const;
807  bool setMemoryClass(const MemoryAccess *From, CongruenceClass *To);
808  CongruenceClass *getMemoryClass(const MemoryAccess *MA) const;
809  const MemoryAccess *lookupMemoryLeader(const MemoryAccess *) const;
810  bool isMemoryAccessTOP(const MemoryAccess *) const;
811 
812  // Ranking
813  unsigned int getRank(const Value *) const;
814  bool shouldSwapOperands(const Value *, const Value *) const;
815  bool shouldSwapOperandsForIntrinsic(const Value *, const Value *,
816  const IntrinsicInst *I) const;
817 
818  // Reachability handling.
819  void updateReachableEdge(BasicBlock *, BasicBlock *);
820  void processOutgoingEdges(Instruction *, BasicBlock *);
821  Value *findConditionEquivalence(Value *) const;
822 
823  // Elimination.
824  struct ValueDFS;
825  void convertClassToDFSOrdered(const CongruenceClass &,
829  void convertClassToLoadsAndStores(const CongruenceClass &,
830  SmallVectorImpl<ValueDFS> &) const;
831 
832  bool eliminateInstructions(Function &);
833  void replaceInstruction(Instruction *, Value *);
834  void markInstructionForDeletion(Instruction *);
835  void deleteInstructionsInBlock(BasicBlock *);
836  Value *findPHIOfOpsLeader(const Expression *, const Instruction *,
837  const BasicBlock *) const;
838 
839  // Various instruction touch utilities
840  template <typename Map, typename KeyType>
841  void touchAndErase(Map &, const KeyType &);
842  void markUsersTouched(Value *);
843  void markMemoryUsersTouched(const MemoryAccess *);
844  void markMemoryDefTouched(const MemoryAccess *);
845  void markPredicateUsersTouched(Instruction *);
846  void markValueLeaderChangeTouched(CongruenceClass *CC);
847  void markMemoryLeaderChangeTouched(CongruenceClass *CC);
848  void markPhiOfOpsChanged(const Expression *E);
849  void addMemoryUsers(const MemoryAccess *To, MemoryAccess *U) const;
850  void addAdditionalUsers(Value *To, Value *User) const;
851  void addAdditionalUsers(ExprResult &Res, Instruction *User) const;
852 
853  // Main loop of value numbering
854  void iterateTouchedInstructions();
855 
856  // Utilities.
857  void cleanupTables();
858  std::pair<unsigned, unsigned> assignDFSNumbers(BasicBlock *, unsigned);
859  void updateProcessedCount(const Value *V);
860  void verifyMemoryCongruency() const;
861  void verifyIterationSettled(Function &F);
862  void verifyStoreExpressions() const;
863  bool singleReachablePHIPath(SmallPtrSet<const MemoryAccess *, 8> &,
864  const MemoryAccess *, const MemoryAccess *) const;
865  BasicBlock *getBlockForValue(Value *V) const;
866  void deleteExpression(const Expression *E) const;
867  MemoryUseOrDef *getMemoryAccess(const Instruction *) const;
868  MemoryPhi *getMemoryAccess(const BasicBlock *) const;
869  template <class T, class Range> T *getMinDFSOfRange(const Range &) const;
870 
871  unsigned InstrToDFSNum(const Value *V) const {
872  assert(isa<Instruction>(V) && "This should not be used for MemoryAccesses");
873  return InstrDFS.lookup(V);
874  }
875 
876  unsigned InstrToDFSNum(const MemoryAccess *MA) const {
877  return MemoryToDFSNum(MA);
878  }
879 
880  Value *InstrFromDFSNum(unsigned DFSNum) { return DFSToInstr[DFSNum]; }
881 
882  // Given a MemoryAccess, return the relevant instruction DFS number. Note:
883  // This deliberately takes a value so it can be used with Use's, which will
884  // auto-convert to Value's but not to MemoryAccess's.
885  unsigned MemoryToDFSNum(const Value *MA) const {
886  assert(isa<MemoryAccess>(MA) &&
887  "This should not be used with instructions");
888  return isa<MemoryUseOrDef>(MA)
889  ? InstrToDFSNum(cast<MemoryUseOrDef>(MA)->getMemoryInst())
890  : InstrDFS.lookup(MA);
891  }
892 
893  bool isCycleFree(const Instruction *) const;
894  bool isBackedge(BasicBlock *From, BasicBlock *To) const;
895 
896  // Debug counter info. When verifying, we have to reset the value numbering
897  // debug counter to the same state it started in to get the same results.
898  int64_t StartingVNCounter = 0;
899 };
900 
901 } // end anonymous namespace
902 
903 template <typename T>
904 static bool equalsLoadStoreHelper(const T &LHS, const Expression &RHS) {
905  if (!isa<LoadExpression>(RHS) && !isa<StoreExpression>(RHS))
906  return false;
907  return LHS.MemoryExpression::equals(RHS);
908 }
909 
911  return equalsLoadStoreHelper(*this, Other);
912 }
913 
915  if (!equalsLoadStoreHelper(*this, Other))
916  return false;
917  // Make sure that store vs store includes the value operand.
918  if (const auto *S = dyn_cast<StoreExpression>(&Other))
919  if (getStoredValue() != S->getStoredValue())
920  return false;
921  return true;
922 }
923 
924 // Determine if the edge From->To is a backedge
925 bool NewGVN::isBackedge(BasicBlock *From, BasicBlock *To) const {
926  return From == To ||
927  RPOOrdering.lookup(DT->getNode(From)) >=
928  RPOOrdering.lookup(DT->getNode(To));
929 }
930 
931 #ifndef NDEBUG
932 static std::string getBlockName(const BasicBlock *B) {
934 }
935 #endif
936 
937 // Get a MemoryAccess for an instruction, fake or real.
938 MemoryUseOrDef *NewGVN::getMemoryAccess(const Instruction *I) const {
939  auto *Result = MSSA->getMemoryAccess(I);
940  return Result ? Result : TempToMemory.lookup(I);
941 }
942 
943 // Get a MemoryPhi for a basic block. These are all real.
944 MemoryPhi *NewGVN::getMemoryAccess(const BasicBlock *BB) const {
945  return MSSA->getMemoryAccess(BB);
946 }
947 
948 // Get the basic block from an instruction/memory value.
949 BasicBlock *NewGVN::getBlockForValue(Value *V) const {
950  if (auto *I = dyn_cast<Instruction>(V)) {
951  auto *Parent = I->getParent();
952  if (Parent)
953  return Parent;
954  Parent = TempToBlock.lookup(V);
955  assert(Parent && "Every fake instruction should have a block");
956  return Parent;
957  }
958 
959  auto *MP = dyn_cast<MemoryPhi>(V);
960  assert(MP && "Should have been an instruction or a MemoryPhi");
961  return MP->getBlock();
962 }
963 
964 // Delete a definitely dead expression, so it can be reused by the expression
965 // allocator. Some of these are not in creation functions, so we have to accept
966 // const versions.
967 void NewGVN::deleteExpression(const Expression *E) const {
968  assert(isa<BasicExpression>(E));
969  auto *BE = cast<BasicExpression>(E);
970  const_cast<BasicExpression *>(BE)->deallocateOperands(ArgRecycler);
971  ExpressionAllocator.Deallocate(E);
972 }
973 
974 // If V is a predicateinfo copy, get the thing it is a copy of.
975 static Value *getCopyOf(const Value *V) {
976  if (auto *II = dyn_cast<IntrinsicInst>(V))
977  if (II->getIntrinsicID() == Intrinsic::ssa_copy)
978  return II->getOperand(0);
979  return nullptr;
980 }
981 
982 // Return true if V is really PN, even accounting for predicateinfo copies.
983 static bool isCopyOfPHI(const Value *V, const PHINode *PN) {
984  return V == PN || getCopyOf(V) == PN;
985 }
986 
987 static bool isCopyOfAPHI(const Value *V) {
988  auto *CO = getCopyOf(V);
989  return CO && isa<PHINode>(CO);
990 }
991 
992 // Sort PHI Operands into a canonical order. What we use here is an RPO
993 // order. The BlockInstRange numbers are generated in an RPO walk of the basic
994 // blocks.
995 void NewGVN::sortPHIOps(MutableArrayRef<ValPair> Ops) const {
996  llvm::sort(Ops, [&](const ValPair &P1, const ValPair &P2) {
997  return BlockInstRange.lookup(P1.second).first <
998  BlockInstRange.lookup(P2.second).first;
999  });
1000 }
1001 
1002 // Return true if V is a value that will always be available (IE can
1003 // be placed anywhere) in the function. We don't do globals here
1004 // because they are often worse to put in place.
1005 static bool alwaysAvailable(Value *V) {
1006  return isa<Constant>(V) || isa<Argument>(V);
1007 }
1008 
1009 // Create a PHIExpression from an array of {incoming edge, value} pairs. I is
1010 // the original instruction we are creating a PHIExpression for (but may not be
1011 // a phi node). We require, as an invariant, that all the PHIOperands in the
1012 // same block are sorted the same way. sortPHIOps will sort them into a
1013 // canonical order.
1014 PHIExpression *NewGVN::createPHIExpression(ArrayRef<ValPair> PHIOperands,
1015  const Instruction *I,
1016  BasicBlock *PHIBlock,
1017  bool &HasBackedge,
1018  bool &OriginalOpsConstant) const {
1019  unsigned NumOps = PHIOperands.size();
1020  auto *E = new (ExpressionAllocator) PHIExpression(NumOps, PHIBlock);
1021 
1022  E->allocateOperands(ArgRecycler, ExpressionAllocator);
1023  E->setType(PHIOperands.begin()->first->getType());
1024  E->setOpcode(Instruction::PHI);
1025 
1026  // Filter out unreachable phi operands.
1027  auto Filtered = make_filter_range(PHIOperands, [&](const ValPair &P) {
1028  auto *BB = P.second;
1029  if (auto *PHIOp = dyn_cast<PHINode>(I))
1030  if (isCopyOfPHI(P.first, PHIOp))
1031  return false;
1032  if (!ReachableEdges.count({BB, PHIBlock}))
1033  return false;
1034  // Things in TOPClass are equivalent to everything.
1035  if (ValueToClass.lookup(P.first) == TOPClass)
1036  return false;
1037  OriginalOpsConstant = OriginalOpsConstant && isa<Constant>(P.first);
1038  HasBackedge = HasBackedge || isBackedge(BB, PHIBlock);
1039  return lookupOperandLeader(P.first) != I;
1040  });
1041  std::transform(Filtered.begin(), Filtered.end(), op_inserter(E),
1042  [&](const ValPair &P) -> Value * {
1043  return lookupOperandLeader(P.first);
1044  });
1045  return E;
1046 }
1047 
1048 // Set basic expression info (Arguments, type, opcode) for Expression
1049 // E from Instruction I in block B.
1050 bool NewGVN::setBasicExpressionInfo(Instruction *I, BasicExpression *E) const {
1051  bool AllConstant = true;
1052  if (auto *GEP = dyn_cast<GetElementPtrInst>(I))
1053  E->setType(GEP->getSourceElementType());
1054  else
1055  E->setType(I->getType());
1056  E->setOpcode(I->getOpcode());
1057  E->allocateOperands(ArgRecycler, ExpressionAllocator);
1058 
1059  // Transform the operand array into an operand leader array, and keep track of
1060  // whether all members are constant.
1061  std::transform(I->op_begin(), I->op_end(), op_inserter(E), [&](Value *O) {
1062  auto Operand = lookupOperandLeader(O);
1063  AllConstant = AllConstant && isa<Constant>(Operand);
1064  return Operand;
1065  });
1066 
1067  return AllConstant;
1068 }
1069 
1070 const Expression *NewGVN::createBinaryExpression(unsigned Opcode, Type *T,
1071  Value *Arg1, Value *Arg2,
1072  Instruction *I) const {
1073  auto *E = new (ExpressionAllocator) BasicExpression(2);
1074  // TODO: we need to remove context instruction after Value Tracking
1075  // can run without context instruction
1076  const SimplifyQuery Q = SQ.getWithInstruction(I);
1077 
1078  E->setType(T);
1079  E->setOpcode(Opcode);
1080  E->allocateOperands(ArgRecycler, ExpressionAllocator);
1081  if (Instruction::isCommutative(Opcode)) {
1082  // Ensure that commutative instructions that only differ by a permutation
1083  // of their operands get the same value number by sorting the operand value
1084  // numbers. Since all commutative instructions have two operands it is more
1085  // efficient to sort by hand rather than using, say, std::sort.
1086  if (shouldSwapOperands(Arg1, Arg2))
1087  std::swap(Arg1, Arg2);
1088  }
1089  E->op_push_back(lookupOperandLeader(Arg1));
1090  E->op_push_back(lookupOperandLeader(Arg2));
1091 
1092  Value *V = simplifyBinOp(Opcode, E->getOperand(0), E->getOperand(1), Q);
1093  if (auto Simplified = checkExprResults(E, I, V)) {
1094  addAdditionalUsers(Simplified, I);
1095  return Simplified.Expr;
1096  }
1097  return E;
1098 }
1099 
1100 // Take a Value returned by simplification of Expression E/Instruction
1101 // I, and see if it resulted in a simpler expression. If so, return
1102 // that expression.
1103 NewGVN::ExprResult NewGVN::checkExprResults(Expression *E, Instruction *I,
1104  Value *V) const {
1105  if (!V)
1106  return ExprResult::none();
1107 
1108  if (auto *C = dyn_cast<Constant>(V)) {
1109  if (I)
1110  LLVM_DEBUG(dbgs() << "Simplified " << *I << " to "
1111  << " constant " << *C << "\n");
1112  NumGVNOpsSimplified++;
1113  assert(isa<BasicExpression>(E) &&
1114  "We should always have had a basic expression here");
1115  deleteExpression(E);
1116  return ExprResult::some(createConstantExpression(C));
1117  } else if (isa<Argument>(V) || isa<GlobalVariable>(V)) {
1118  if (I)
1119  LLVM_DEBUG(dbgs() << "Simplified " << *I << " to "
1120  << " variable " << *V << "\n");
1121  deleteExpression(E);
1122  return ExprResult::some(createVariableExpression(V));
1123  }
1124 
1125  CongruenceClass *CC = ValueToClass.lookup(V);
1126  if (CC) {
1127  if (CC->getLeader() && CC->getLeader() != I) {
1128  return ExprResult::some(createVariableOrConstant(CC->getLeader()), V);
1129  }
1130  if (CC->getDefiningExpr()) {
1131  if (I)
1132  LLVM_DEBUG(dbgs() << "Simplified " << *I << " to "
1133  << " expression " << *CC->getDefiningExpr() << "\n");
1134  NumGVNOpsSimplified++;
1135  deleteExpression(E);
1136  return ExprResult::some(CC->getDefiningExpr(), V);
1137  }
1138  }
1139 
1140  return ExprResult::none();
1141 }
1142 
1143 // Create a value expression from the instruction I, replacing operands with
1144 // their leaders.
1145 
1146 NewGVN::ExprResult NewGVN::createExpression(Instruction *I) const {
1147  auto *E = new (ExpressionAllocator) BasicExpression(I->getNumOperands());
1148  // TODO: we need to remove context instruction after Value Tracking
1149  // can run without context instruction
1150  const SimplifyQuery Q = SQ.getWithInstruction(I);
1151 
1152  bool AllConstant = setBasicExpressionInfo(I, E);
1153 
1154  if (I->isCommutative()) {
1155  // Ensure that commutative instructions that only differ by a permutation
1156  // of their operands get the same value number by sorting the operand value
1157  // numbers. Since all commutative instructions have two operands it is more
1158  // efficient to sort by hand rather than using, say, std::sort.
1159  assert(I->getNumOperands() == 2 && "Unsupported commutative instruction!");
1160  if (shouldSwapOperands(E->getOperand(0), E->getOperand(1)))
1161  E->swapOperands(0, 1);
1162  }
1163  // Perform simplification.
1164  if (auto *CI = dyn_cast<CmpInst>(I)) {
1165  // Sort the operand value numbers so x<y and y>x get the same value
1166  // number.
1167  CmpInst::Predicate Predicate = CI->getPredicate();
1168  if (shouldSwapOperands(E->getOperand(0), E->getOperand(1))) {
1169  E->swapOperands(0, 1);
1171  }
1172  E->setOpcode((CI->getOpcode() << 8) | Predicate);
1173  // TODO: 25% of our time is spent in simplifyCmpInst with pointer operands
1174  assert(I->getOperand(0)->getType() == I->getOperand(1)->getType() &&
1175  "Wrong types on cmp instruction");
1176  assert((E->getOperand(0)->getType() == I->getOperand(0)->getType() &&
1177  E->getOperand(1)->getType() == I->getOperand(1)->getType()));
1178  Value *V =
1179  simplifyCmpInst(Predicate, E->getOperand(0), E->getOperand(1), Q);
1180  if (auto Simplified = checkExprResults(E, I, V))
1181  return Simplified;
1182  } else if (isa<SelectInst>(I)) {
1183  if (isa<Constant>(E->getOperand(0)) ||
1184  E->getOperand(1) == E->getOperand(2)) {
1185  assert(E->getOperand(1)->getType() == I->getOperand(1)->getType() &&
1186  E->getOperand(2)->getType() == I->getOperand(2)->getType());
1187  Value *V = simplifySelectInst(E->getOperand(0), E->getOperand(1),
1188  E->getOperand(2), Q);
1189  if (auto Simplified = checkExprResults(E, I, V))
1190  return Simplified;
1191  }
1192  } else if (I->isBinaryOp()) {
1193  Value *V =
1194  simplifyBinOp(E->getOpcode(), E->getOperand(0), E->getOperand(1), Q);
1195  if (auto Simplified = checkExprResults(E, I, V))
1196  return Simplified;
1197  } else if (auto *CI = dyn_cast<CastInst>(I)) {
1198  Value *V =
1199  simplifyCastInst(CI->getOpcode(), E->getOperand(0), CI->getType(), Q);
1200  if (auto Simplified = checkExprResults(E, I, V))
1201  return Simplified;
1202  } else if (auto *GEPI = dyn_cast<GetElementPtrInst>(I)) {
1203  Value *V =
1204  simplifyGEPInst(GEPI->getSourceElementType(), *E->op_begin(),
1205  makeArrayRef(std::next(E->op_begin()), E->op_end()),
1206  GEPI->isInBounds(), Q);
1207  if (auto Simplified = checkExprResults(E, I, V))
1208  return Simplified;
1209  } else if (AllConstant) {
1210  // We don't bother trying to simplify unless all of the operands
1211  // were constant.
1212  // TODO: There are a lot of Simplify*'s we could call here, if we
1213  // wanted to. The original motivating case for this code was a
1214  // zext i1 false to i8, which we don't have an interface to
1215  // simplify (IE there is no SimplifyZExt).
1216 
1218  for (Value *Arg : E->operands())
1219  C.emplace_back(cast<Constant>(Arg));
1220 
1221  if (Value *V = ConstantFoldInstOperands(I, C, DL, TLI))
1222  if (auto Simplified = checkExprResults(E, I, V))
1223  return Simplified;
1224  }
1225  return ExprResult::some(E);
1226 }
1227 
1229 NewGVN::createAggregateValueExpression(Instruction *I) const {
1230  if (auto *II = dyn_cast<InsertValueInst>(I)) {
1231  auto *E = new (ExpressionAllocator)
1232  AggregateValueExpression(I->getNumOperands(), II->getNumIndices());
1233  setBasicExpressionInfo(I, E);
1234  E->allocateIntOperands(ExpressionAllocator);
1235  std::copy(II->idx_begin(), II->idx_end(), int_op_inserter(E));
1236  return E;
1237  } else if (auto *EI = dyn_cast<ExtractValueInst>(I)) {
1238  auto *E = new (ExpressionAllocator)
1239  AggregateValueExpression(I->getNumOperands(), EI->getNumIndices());
1240  setBasicExpressionInfo(EI, E);
1241  E->allocateIntOperands(ExpressionAllocator);
1242  std::copy(EI->idx_begin(), EI->idx_end(), int_op_inserter(E));
1243  return E;
1244  }
1245  llvm_unreachable("Unhandled type of aggregate value operation");
1246 }
1247 
1248 const DeadExpression *NewGVN::createDeadExpression() const {
1249  // DeadExpression has no arguments and all DeadExpression's are the same,
1250  // so we only need one of them.
1251  return SingletonDeadExpression;
1252 }
1253 
1254 const VariableExpression *NewGVN::createVariableExpression(Value *V) const {
1255  auto *E = new (ExpressionAllocator) VariableExpression(V);
1256  E->setOpcode(V->getValueID());
1257  return E;
1258 }
1259 
1260 const Expression *NewGVN::createVariableOrConstant(Value *V) const {
1261  if (auto *C = dyn_cast<Constant>(V))
1262  return createConstantExpression(C);
1263  return createVariableExpression(V);
1264 }
1265 
1266 const ConstantExpression *NewGVN::createConstantExpression(Constant *C) const {
1267  auto *E = new (ExpressionAllocator) ConstantExpression(C);
1268  E->setOpcode(C->getValueID());
1269  return E;
1270 }
1271 
1272 const UnknownExpression *NewGVN::createUnknownExpression(Instruction *I) const {
1273  auto *E = new (ExpressionAllocator) UnknownExpression(I);
1274  E->setOpcode(I->getOpcode());
1275  return E;
1276 }
1277 
1278 const CallExpression *
1279 NewGVN::createCallExpression(CallInst *CI, const MemoryAccess *MA) const {
1280  // FIXME: Add operand bundles for calls.
1281  // FIXME: Allow commutative matching for intrinsics.
1282  auto *E =
1283  new (ExpressionAllocator) CallExpression(CI->getNumOperands(), CI, MA);
1284  setBasicExpressionInfo(CI, E);
1285  return E;
1286 }
1287 
1288 // Return true if some equivalent of instruction Inst dominates instruction U.
1289 bool NewGVN::someEquivalentDominates(const Instruction *Inst,
1290  const Instruction *U) const {
1291  auto *CC = ValueToClass.lookup(Inst);
1292  // This must be an instruction because we are only called from phi nodes
1293  // in the case that the value it needs to check against is an instruction.
1294 
1295  // The most likely candidates for dominance are the leader and the next leader.
1296  // The leader or nextleader will dominate in all cases where there is an
1297  // equivalent that is higher up in the dom tree.
1298  // We can't *only* check them, however, because the
1299  // dominator tree could have an infinite number of non-dominating siblings
1300  // with instructions that are in the right congruence class.
1301  // A
1302  // B C D E F G
1303  // |
1304  // H
1305  // Instruction U could be in H, with equivalents in every other sibling.
1306  // Depending on the rpo order picked, the leader could be the equivalent in
1307  // any of these siblings.
1308  if (!CC)
1309  return false;
1310  if (alwaysAvailable(CC->getLeader()))
1311  return true;
1312  if (DT->dominates(cast<Instruction>(CC->getLeader()), U))
1313  return true;
1314  if (CC->getNextLeader().first &&
1315  DT->dominates(cast<Instruction>(CC->getNextLeader().first), U))
1316  return true;
1317  return llvm::any_of(*CC, [&](const Value *Member) {
1318  return Member != CC->getLeader() &&
1319  DT->dominates(cast<Instruction>(Member), U);
1320  });
1321 }
1322 
1323 // See if we have a congruence class and leader for this operand, and if so,
1324 // return it. Otherwise, return the operand itself.
1325 Value *NewGVN::lookupOperandLeader(Value *V) const {
1326  CongruenceClass *CC = ValueToClass.lookup(V);
1327  if (CC) {
1328  // Everything in TOP is represented by poison, as it can be any value.
1329  // We do have to make sure we get the type right though, so we can't set the
1330  // RepLeader to poison.
1331  if (CC == TOPClass)
1332  return PoisonValue::get(V->getType());
1333  return CC->getStoredValue() ? CC->getStoredValue() : CC->getLeader();
1334  }
1335 
1336  return V;
1337 }
1338 
1339 const MemoryAccess *NewGVN::lookupMemoryLeader(const MemoryAccess *MA) const {
1340  auto *CC = getMemoryClass(MA);
1341  assert(CC->getMemoryLeader() &&
1342  "Every MemoryAccess should be mapped to a congruence class with a "
1343  "representative memory access");
1344  return CC->getMemoryLeader();
1345 }
1346 
1347 // Return true if the MemoryAccess is really equivalent to everything. This is
1348 // equivalent to the lattice value "TOP" in most lattices. This is the initial
1349 // state of all MemoryAccesses.
1350 bool NewGVN::isMemoryAccessTOP(const MemoryAccess *MA) const {
1351  return getMemoryClass(MA) == TOPClass;
1352 }
1353 
1354 LoadExpression *NewGVN::createLoadExpression(Type *LoadType, Value *PointerOp,
1355  LoadInst *LI,
1356  const MemoryAccess *MA) const {
1357  auto *E =
1358  new (ExpressionAllocator) LoadExpression(1, LI, lookupMemoryLeader(MA));
1359  E->allocateOperands(ArgRecycler, ExpressionAllocator);
1360  E->setType(LoadType);
1361 
1362  // Give store and loads same opcode so they value number together.
1363  E->setOpcode(0);
1364  E->op_push_back(PointerOp);
1365 
1366  // TODO: Value number heap versions. We may be able to discover
1367  // things alias analysis can't on it's own (IE that a store and a
1368  // load have the same value, and thus, it isn't clobbering the load).
1369  return E;
1370 }
1371 
1372 const StoreExpression *
1373 NewGVN::createStoreExpression(StoreInst *SI, const MemoryAccess *MA) const {
1374  auto *StoredValueLeader = lookupOperandLeader(SI->getValueOperand());
1375  auto *E = new (ExpressionAllocator)
1376  StoreExpression(SI->getNumOperands(), SI, StoredValueLeader, MA);
1377  E->allocateOperands(ArgRecycler, ExpressionAllocator);
1378  E->setType(SI->getValueOperand()->getType());
1379 
1380  // Give store and loads same opcode so they value number together.
1381  E->setOpcode(0);
1382  E->op_push_back(lookupOperandLeader(SI->getPointerOperand()));
1383 
1384  // TODO: Value number heap versions. We may be able to discover
1385  // things alias analysis can't on it's own (IE that a store and a
1386  // load have the same value, and thus, it isn't clobbering the load).
1387  return E;
1388 }
1389 
1390 const Expression *NewGVN::performSymbolicStoreEvaluation(Instruction *I) const {
1391  // Unlike loads, we never try to eliminate stores, so we do not check if they
1392  // are simple and avoid value numbering them.
1393  auto *SI = cast<StoreInst>(I);
1394  auto *StoreAccess = getMemoryAccess(SI);
1395  // Get the expression, if any, for the RHS of the MemoryDef.
1396  const MemoryAccess *StoreRHS = StoreAccess->getDefiningAccess();
1398  StoreRHS = MSSAWalker->getClobberingMemoryAccess(StoreAccess);
1399  // If we bypassed the use-def chains, make sure we add a use.
1400  StoreRHS = lookupMemoryLeader(StoreRHS);
1401  if (StoreRHS != StoreAccess->getDefiningAccess())
1402  addMemoryUsers(StoreRHS, StoreAccess);
1403  // If we are defined by ourselves, use the live on entry def.
1404  if (StoreRHS == StoreAccess)
1405  StoreRHS = MSSA->getLiveOnEntryDef();
1406 
1407  if (SI->isSimple()) {
1408  // See if we are defined by a previous store expression, it already has a
1409  // value, and it's the same value as our current store. FIXME: Right now, we
1410  // only do this for simple stores, we should expand to cover memcpys, etc.
1411  const auto *LastStore = createStoreExpression(SI, StoreRHS);
1412  const auto *LastCC = ExpressionToClass.lookup(LastStore);
1413  // We really want to check whether the expression we matched was a store. No
1414  // easy way to do that. However, we can check that the class we found has a
1415  // store, which, assuming the value numbering state is not corrupt, is
1416  // sufficient, because we must also be equivalent to that store's expression
1417  // for it to be in the same class as the load.
1418  if (LastCC && LastCC->getStoredValue() == LastStore->getStoredValue())
1419  return LastStore;
1420  // Also check if our value operand is defined by a load of the same memory
1421  // location, and the memory state is the same as it was then (otherwise, it
1422  // could have been overwritten later. See test32 in
1423  // transforms/DeadStoreElimination/simple.ll).
1424  if (auto *LI = dyn_cast<LoadInst>(LastStore->getStoredValue()))
1425  if ((lookupOperandLeader(LI->getPointerOperand()) ==
1426  LastStore->getOperand(0)) &&
1427  (lookupMemoryLeader(getMemoryAccess(LI)->getDefiningAccess()) ==
1428  StoreRHS))
1429  return LastStore;
1430  deleteExpression(LastStore);
1431  }
1432 
1433  // If the store is not equivalent to anything, value number it as a store that
1434  // produces a unique memory state (instead of using it's MemoryUse, we use
1435  // it's MemoryDef).
1436  return createStoreExpression(SI, StoreAccess);
1437 }
1438 
1439 // See if we can extract the value of a loaded pointer from a load, a store, or
1440 // a memory instruction.
1441 const Expression *
1442 NewGVN::performSymbolicLoadCoercion(Type *LoadType, Value *LoadPtr,
1443  LoadInst *LI, Instruction *DepInst,
1444  MemoryAccess *DefiningAccess) const {
1445  assert((!LI || LI->isSimple()) && "Not a simple load");
1446  if (auto *DepSI = dyn_cast<StoreInst>(DepInst)) {
1447  // Can't forward from non-atomic to atomic without violating memory model.
1448  // Also don't need to coerce if they are the same type, we will just
1449  // propagate.
1450  if (LI->isAtomic() > DepSI->isAtomic() ||
1451  LoadType == DepSI->getValueOperand()->getType())
1452  return nullptr;
1453  int Offset = analyzeLoadFromClobberingStore(LoadType, LoadPtr, DepSI, DL);
1454  if (Offset >= 0) {
1455  if (auto *C = dyn_cast<Constant>(
1456  lookupOperandLeader(DepSI->getValueOperand()))) {
1457  if (Constant *Res =
1458  getConstantStoreValueForLoad(C, Offset, LoadType, DL)) {
1459  LLVM_DEBUG(dbgs() << "Coercing load from store " << *DepSI
1460  << " to constant " << *Res << "\n");
1461  return createConstantExpression(Res);
1462  }
1463  }
1464  }
1465  } else if (auto *DepLI = dyn_cast<LoadInst>(DepInst)) {
1466  // Can't forward from non-atomic to atomic without violating memory model.
1467  if (LI->isAtomic() > DepLI->isAtomic())
1468  return nullptr;
1469  int Offset = analyzeLoadFromClobberingLoad(LoadType, LoadPtr, DepLI, DL);
1470  if (Offset >= 0) {
1471  // We can coerce a constant load into a load.
1472  if (auto *C = dyn_cast<Constant>(lookupOperandLeader(DepLI)))
1473  if (auto *PossibleConstant =
1474  getConstantLoadValueForLoad(C, Offset, LoadType, DL)) {
1475  LLVM_DEBUG(dbgs() << "Coercing load from load " << *LI
1476  << " to constant " << *PossibleConstant << "\n");
1477  return createConstantExpression(PossibleConstant);
1478  }
1479  }
1480  } else if (auto *DepMI = dyn_cast<MemIntrinsic>(DepInst)) {
1481  int Offset = analyzeLoadFromClobberingMemInst(LoadType, LoadPtr, DepMI, DL);
1482  if (Offset >= 0) {
1483  if (auto *PossibleConstant =
1484  getConstantMemInstValueForLoad(DepMI, Offset, LoadType, DL)) {
1485  LLVM_DEBUG(dbgs() << "Coercing load from meminst " << *DepMI
1486  << " to constant " << *PossibleConstant << "\n");
1487  return createConstantExpression(PossibleConstant);
1488  }
1489  }
1490  }
1491 
1492  // All of the below are only true if the loaded pointer is produced
1493  // by the dependent instruction.
1494  if (LoadPtr != lookupOperandLeader(DepInst) &&
1495  !AA->isMustAlias(LoadPtr, DepInst))
1496  return nullptr;
1497  // If this load really doesn't depend on anything, then we must be loading an
1498  // undef value. This can happen when loading for a fresh allocation with no
1499  // intervening stores, for example. Note that this is only true in the case
1500  // that the result of the allocation is pointer equal to the load ptr.
1501  if (isa<AllocaInst>(DepInst)) {
1502  return createConstantExpression(UndefValue::get(LoadType));
1503  }
1504  // If this load occurs either right after a lifetime begin,
1505  // then the loaded value is undefined.
1506  else if (auto *II = dyn_cast<IntrinsicInst>(DepInst)) {
1507  if (II->getIntrinsicID() == Intrinsic::lifetime_start)
1508  return createConstantExpression(UndefValue::get(LoadType));
1509  } else if (auto *InitVal =
1510  getInitialValueOfAllocation(DepInst, TLI, LoadType))
1511  return createConstantExpression(InitVal);
1512 
1513  return nullptr;
1514 }
1515 
1516 const Expression *NewGVN::performSymbolicLoadEvaluation(Instruction *I) const {
1517  auto *LI = cast<LoadInst>(I);
1518 
1519  // We can eliminate in favor of non-simple loads, but we won't be able to
1520  // eliminate the loads themselves.
1521  if (!LI->isSimple())
1522  return nullptr;
1523 
1524  Value *LoadAddressLeader = lookupOperandLeader(LI->getPointerOperand());
1525  // Load of undef is UB.
1526  if (isa<UndefValue>(LoadAddressLeader))
1527  return createConstantExpression(PoisonValue::get(LI->getType()));
1528  MemoryAccess *OriginalAccess = getMemoryAccess(I);
1529  MemoryAccess *DefiningAccess =
1530  MSSAWalker->getClobberingMemoryAccess(OriginalAccess);
1531 
1532  if (!MSSA->isLiveOnEntryDef(DefiningAccess)) {
1533  if (auto *MD = dyn_cast<MemoryDef>(DefiningAccess)) {
1534  Instruction *DefiningInst = MD->getMemoryInst();
1535  // If the defining instruction is not reachable, replace with poison.
1536  if (!ReachableBlocks.count(DefiningInst->getParent()))
1537  return createConstantExpression(PoisonValue::get(LI->getType()));
1538  // This will handle stores and memory insts. We only do if it the
1539  // defining access has a different type, or it is a pointer produced by
1540  // certain memory operations that cause the memory to have a fixed value
1541  // (IE things like calloc).
1542  if (const auto *CoercionResult =
1543  performSymbolicLoadCoercion(LI->getType(), LoadAddressLeader, LI,
1544  DefiningInst, DefiningAccess))
1545  return CoercionResult;
1546  }
1547  }
1548 
1549  const auto *LE = createLoadExpression(LI->getType(), LoadAddressLeader, LI,
1550  DefiningAccess);
1551  // If our MemoryLeader is not our defining access, add a use to the
1552  // MemoryLeader, so that we get reprocessed when it changes.
1553  if (LE->getMemoryLeader() != DefiningAccess)
1554  addMemoryUsers(LE->getMemoryLeader(), OriginalAccess);
1555  return LE;
1556 }
1557 
1558 NewGVN::ExprResult
1559 NewGVN::performSymbolicPredicateInfoEvaluation(IntrinsicInst *I) const {
1560  auto *PI = PredInfo->getPredicateInfoFor(I);
1561  if (!PI)
1562  return ExprResult::none();
1563 
1564  LLVM_DEBUG(dbgs() << "Found predicate info from instruction !\n");
1565 
1566  const Optional<PredicateConstraint> &Constraint = PI->getConstraint();
1567  if (!Constraint)
1568  return ExprResult::none();
1569 
1570  CmpInst::Predicate Predicate = Constraint->Predicate;
1571  Value *CmpOp0 = I->getOperand(0);
1572  Value *CmpOp1 = Constraint->OtherOp;
1573 
1574  Value *FirstOp = lookupOperandLeader(CmpOp0);
1575  Value *SecondOp = lookupOperandLeader(CmpOp1);
1576  Value *AdditionallyUsedValue = CmpOp0;
1577 
1578  // Sort the ops.
1579  if (shouldSwapOperandsForIntrinsic(FirstOp, SecondOp, I)) {
1580  std::swap(FirstOp, SecondOp);
1582  AdditionallyUsedValue = CmpOp1;
1583  }
1584 
1585  if (Predicate == CmpInst::ICMP_EQ)
1586  return ExprResult::some(createVariableOrConstant(FirstOp),
1587  AdditionallyUsedValue, PI);
1588 
1589  // Handle the special case of floating point.
1590  if (Predicate == CmpInst::FCMP_OEQ && isa<ConstantFP>(FirstOp) &&
1591  !cast<ConstantFP>(FirstOp)->isZero())
1592  return ExprResult::some(createConstantExpression(cast<Constant>(FirstOp)),
1593  AdditionallyUsedValue, PI);
1594 
1595  return ExprResult::none();
1596 }
1597 
1598 // Evaluate read only and pure calls, and create an expression result.
1599 NewGVN::ExprResult NewGVN::performSymbolicCallEvaluation(Instruction *I) const {
1600  auto *CI = cast<CallInst>(I);
1601  if (auto *II = dyn_cast<IntrinsicInst>(I)) {
1602  // Intrinsics with the returned attribute are copies of arguments.
1603  if (auto *ReturnedValue = II->getReturnedArgOperand()) {
1604  if (II->getIntrinsicID() == Intrinsic::ssa_copy)
1605  if (auto Res = performSymbolicPredicateInfoEvaluation(II))
1606  return Res;
1607  return ExprResult::some(createVariableOrConstant(ReturnedValue));
1608  }
1609  }
1610  if (AA->doesNotAccessMemory(CI)) {
1611  return ExprResult::some(
1612  createCallExpression(CI, TOPClass->getMemoryLeader()));
1613  } else if (AA->onlyReadsMemory(CI)) {
1614  if (auto *MA = MSSA->getMemoryAccess(CI)) {
1615  auto *DefiningAccess = MSSAWalker->getClobberingMemoryAccess(MA);
1616  return ExprResult::some(createCallExpression(CI, DefiningAccess));
1617  } else // MSSA determined that CI does not access memory.
1618  return ExprResult::some(
1619  createCallExpression(CI, TOPClass->getMemoryLeader()));
1620  }
1621  return ExprResult::none();
1622 }
1623 
1624 // Retrieve the memory class for a given MemoryAccess.
1625 CongruenceClass *NewGVN::getMemoryClass(const MemoryAccess *MA) const {
1626  auto *Result = MemoryAccessToClass.lookup(MA);
1627  assert(Result && "Should have found memory class");
1628  return Result;
1629 }
1630 
1631 // Update the MemoryAccess equivalence table to say that From is equal to To,
1632 // and return true if this is different from what already existed in the table.
1633 bool NewGVN::setMemoryClass(const MemoryAccess *From,
1634  CongruenceClass *NewClass) {
1635  assert(NewClass &&
1636  "Every MemoryAccess should be getting mapped to a non-null class");
1637  LLVM_DEBUG(dbgs() << "Setting " << *From);
1638  LLVM_DEBUG(dbgs() << " equivalent to congruence class ");
1639  LLVM_DEBUG(dbgs() << NewClass->getID()
1640  << " with current MemoryAccess leader ");
1641  LLVM_DEBUG(dbgs() << *NewClass->getMemoryLeader() << "\n");
1642 
1643  auto LookupResult = MemoryAccessToClass.find(From);
1644  bool Changed = false;
1645  // If it's already in the table, see if the value changed.
1646  if (LookupResult != MemoryAccessToClass.end()) {
1647  auto *OldClass = LookupResult->second;
1648  if (OldClass != NewClass) {
1649  // If this is a phi, we have to handle memory member updates.
1650  if (auto *MP = dyn_cast<MemoryPhi>(From)) {
1651  OldClass->memory_erase(MP);
1652  NewClass->memory_insert(MP);
1653  // This may have killed the class if it had no non-memory members
1654  if (OldClass->getMemoryLeader() == From) {
1655  if (OldClass->definesNoMemory()) {
1656  OldClass->setMemoryLeader(nullptr);
1657  } else {
1658  OldClass->setMemoryLeader(getNextMemoryLeader(OldClass));
1659  LLVM_DEBUG(dbgs() << "Memory class leader change for class "
1660  << OldClass->getID() << " to "
1661  << *OldClass->getMemoryLeader()
1662  << " due to removal of a memory member " << *From
1663  << "\n");
1664  markMemoryLeaderChangeTouched(OldClass);
1665  }
1666  }
1667  }
1668  // It wasn't equivalent before, and now it is.
1669  LookupResult->second = NewClass;
1670  Changed = true;
1671  }
1672  }
1673 
1674  return Changed;
1675 }
1676 
1677 // Determine if a instruction is cycle-free. That means the values in the
1678 // instruction don't depend on any expressions that can change value as a result
1679 // of the instruction. For example, a non-cycle free instruction would be v =
1680 // phi(0, v+1).
1681 bool NewGVN::isCycleFree(const Instruction *I) const {
1682  // In order to compute cycle-freeness, we do SCC finding on the instruction,
1683  // and see what kind of SCC it ends up in. If it is a singleton, it is
1684  // cycle-free. If it is not in a singleton, it is only cycle free if the
1685  // other members are all phi nodes (as they do not compute anything, they are
1686  // copies).
1687  auto ICS = InstCycleState.lookup(I);
1688  if (ICS == ICS_Unknown) {
1689  SCCFinder.Start(I);
1690  auto &SCC = SCCFinder.getComponentFor(I);
1691  // It's cycle free if it's size 1 or the SCC is *only* phi nodes.
1692  if (SCC.size() == 1)
1693  InstCycleState.insert({I, ICS_CycleFree});
1694  else {
1695  bool AllPhis = llvm::all_of(SCC, [](const Value *V) {
1696  return isa<PHINode>(V) || isCopyOfAPHI(V);
1697  });
1698  ICS = AllPhis ? ICS_CycleFree : ICS_Cycle;
1699  for (const auto *Member : SCC)
1700  if (auto *MemberPhi = dyn_cast<PHINode>(Member))
1701  InstCycleState.insert({MemberPhi, ICS});
1702  }
1703  }
1704  if (ICS == ICS_Cycle)
1705  return false;
1706  return true;
1707 }
1708 
1709 // Evaluate PHI nodes symbolically and create an expression result.
1710 const Expression *
1711 NewGVN::performSymbolicPHIEvaluation(ArrayRef<ValPair> PHIOps,
1712  Instruction *I,
1713  BasicBlock *PHIBlock) const {
1714  // True if one of the incoming phi edges is a backedge.
1715  bool HasBackedge = false;
1716  // All constant tracks the state of whether all the *original* phi operands
1717  // This is really shorthand for "this phi cannot cycle due to forward
1718  // change in value of the phi is guaranteed not to later change the value of
1719  // the phi. IE it can't be v = phi(undef, v+1)
1720  bool OriginalOpsConstant = true;
1721  auto *E = cast<PHIExpression>(createPHIExpression(
1722  PHIOps, I, PHIBlock, HasBackedge, OriginalOpsConstant));
1723  // We match the semantics of SimplifyPhiNode from InstructionSimplify here.
1724  // See if all arguments are the same.
1725  // We track if any were undef because they need special handling.
1726  bool HasUndef = false, HasPoison = false;
1727  auto Filtered = make_filter_range(E->operands(), [&](Value *Arg) {
1728  if (isa<PoisonValue>(Arg)) {
1729  HasPoison = true;
1730  return false;
1731  }
1732  if (isa<UndefValue>(Arg)) {
1733  HasUndef = true;
1734  return false;
1735  }
1736  return true;
1737  });
1738  // If we are left with no operands, it's dead.
1739  if (Filtered.empty()) {
1740  // If it has undef or poison at this point, it means there are no-non-undef
1741  // arguments, and thus, the value of the phi node must be undef.
1742  if (HasUndef) {
1743  LLVM_DEBUG(
1744  dbgs() << "PHI Node " << *I
1745  << " has no non-undef arguments, valuing it as undef\n");
1746  return createConstantExpression(UndefValue::get(I->getType()));
1747  }
1748  if (HasPoison) {
1749  LLVM_DEBUG(
1750  dbgs() << "PHI Node " << *I
1751  << " has no non-poison arguments, valuing it as poison\n");
1752  return createConstantExpression(PoisonValue::get(I->getType()));
1753  }
1754 
1755  LLVM_DEBUG(dbgs() << "No arguments of PHI node " << *I << " are live\n");
1756  deleteExpression(E);
1757  return createDeadExpression();
1758  }
1759  Value *AllSameValue = *(Filtered.begin());
1760  ++Filtered.begin();
1761  // Can't use std::equal here, sadly, because filter.begin moves.
1762  if (llvm::all_of(Filtered, [&](Value *Arg) { return Arg == AllSameValue; })) {
1763  // Can't fold phi(undef, X) -> X unless X can't be poison (thus X is undef
1764  // in the worst case).
1765  if (HasUndef && !isGuaranteedNotToBePoison(AllSameValue, AC, nullptr, DT))
1766  return E;
1767 
1768  // In LLVM's non-standard representation of phi nodes, it's possible to have
1769  // phi nodes with cycles (IE dependent on other phis that are .... dependent
1770  // on the original phi node), especially in weird CFG's where some arguments
1771  // are unreachable, or uninitialized along certain paths. This can cause
1772  // infinite loops during evaluation. We work around this by not trying to
1773  // really evaluate them independently, but instead using a variable
1774  // expression to say if one is equivalent to the other.
1775  // We also special case undef/poison, so that if we have an undef, we can't
1776  // use the common value unless it dominates the phi block.
1777  if (HasPoison || HasUndef) {
1778  // If we have undef and at least one other value, this is really a
1779  // multivalued phi, and we need to know if it's cycle free in order to
1780  // evaluate whether we can ignore the undef. The other parts of this are
1781  // just shortcuts. If there is no backedge, or all operands are
1782  // constants, it also must be cycle free.
1783  if (HasBackedge && !OriginalOpsConstant &&
1784  !isa<UndefValue>(AllSameValue) && !isCycleFree(I))
1785  return E;
1786 
1787  // Only have to check for instructions
1788  if (auto *AllSameInst = dyn_cast<Instruction>(AllSameValue))
1789  if (!someEquivalentDominates(AllSameInst, I))
1790  return E;
1791  }
1792  // Can't simplify to something that comes later in the iteration.
1793  // Otherwise, when and if it changes congruence class, we will never catch
1794  // up. We will always be a class behind it.
1795  if (isa<Instruction>(AllSameValue) &&
1796  InstrToDFSNum(AllSameValue) > InstrToDFSNum(I))
1797  return E;
1798  NumGVNPhisAllSame++;
1799  LLVM_DEBUG(dbgs() << "Simplified PHI node " << *I << " to " << *AllSameValue
1800  << "\n");
1801  deleteExpression(E);
1802  return createVariableOrConstant(AllSameValue);
1803  }
1804  return E;
1805 }
1806 
1807 const Expression *
1808 NewGVN::performSymbolicAggrValueEvaluation(Instruction *I) const {
1809  if (auto *EI = dyn_cast<ExtractValueInst>(I)) {
1810  auto *WO = dyn_cast<WithOverflowInst>(EI->getAggregateOperand());
1811  if (WO && EI->getNumIndices() == 1 && *EI->idx_begin() == 0)
1812  // EI is an extract from one of our with.overflow intrinsics. Synthesize
1813  // a semantically equivalent expression instead of an extract value
1814  // expression.
1815  return createBinaryExpression(WO->getBinaryOp(), EI->getType(),
1816  WO->getLHS(), WO->getRHS(), I);
1817  }
1818 
1819  return createAggregateValueExpression(I);
1820 }
1821 
1822 NewGVN::ExprResult NewGVN::performSymbolicCmpEvaluation(Instruction *I) const {
1823  assert(isa<CmpInst>(I) && "Expected a cmp instruction.");
1824 
1825  auto *CI = cast<CmpInst>(I);
1826  // See if our operands are equal to those of a previous predicate, and if so,
1827  // if it implies true or false.
1828  auto Op0 = lookupOperandLeader(CI->getOperand(0));
1829  auto Op1 = lookupOperandLeader(CI->getOperand(1));
1830  auto OurPredicate = CI->getPredicate();
1831  if (shouldSwapOperands(Op0, Op1)) {
1832  std::swap(Op0, Op1);
1833  OurPredicate = CI->getSwappedPredicate();
1834  }
1835 
1836  // Avoid processing the same info twice.
1837  const PredicateBase *LastPredInfo = nullptr;
1838  // See if we know something about the comparison itself, like it is the target
1839  // of an assume.
1840  auto *CmpPI = PredInfo->getPredicateInfoFor(I);
1841  if (isa_and_nonnull<PredicateAssume>(CmpPI))
1842  return ExprResult::some(
1843  createConstantExpression(ConstantInt::getTrue(CI->getType())));
1844 
1845  if (Op0 == Op1) {
1846  // This condition does not depend on predicates, no need to add users
1847  if (CI->isTrueWhenEqual())
1848  return ExprResult::some(
1849  createConstantExpression(ConstantInt::getTrue(CI->getType())));
1850  else if (CI->isFalseWhenEqual())
1851  return ExprResult::some(
1852  createConstantExpression(ConstantInt::getFalse(CI->getType())));
1853  }
1854 
1855  // NOTE: Because we are comparing both operands here and below, and using
1856  // previous comparisons, we rely on fact that predicateinfo knows to mark
1857  // comparisons that use renamed operands as users of the earlier comparisons.
1858  // It is *not* enough to just mark predicateinfo renamed operands as users of
1859  // the earlier comparisons, because the *other* operand may have changed in a
1860  // previous iteration.
1861  // Example:
1862  // icmp slt %a, %b
1863  // %b.0 = ssa.copy(%b)
1864  // false branch:
1865  // icmp slt %c, %b.0
1866 
1867  // %c and %a may start out equal, and thus, the code below will say the second
1868  // %icmp is false. c may become equal to something else, and in that case the
1869  // %second icmp *must* be reexamined, but would not if only the renamed
1870  // %operands are considered users of the icmp.
1871 
1872  // *Currently* we only check one level of comparisons back, and only mark one
1873  // level back as touched when changes happen. If you modify this code to look
1874  // back farther through comparisons, you *must* mark the appropriate
1875  // comparisons as users in PredicateInfo.cpp, or you will cause bugs. See if
1876  // we know something just from the operands themselves
1877 
1878  // See if our operands have predicate info, so that we may be able to derive
1879  // something from a previous comparison.
1880  for (const auto &Op : CI->operands()) {
1881  auto *PI = PredInfo->getPredicateInfoFor(Op);
1882  if (const auto *PBranch = dyn_cast_or_null<PredicateBranch>(PI)) {
1883  if (PI == LastPredInfo)
1884  continue;
1885  LastPredInfo = PI;
1886  // In phi of ops cases, we may have predicate info that we are evaluating
1887  // in a different context.
1888  if (!DT->dominates(PBranch->To, getBlockForValue(I)))
1889  continue;
1890  // TODO: Along the false edge, we may know more things too, like
1891  // icmp of
1892  // same operands is false.
1893  // TODO: We only handle actual comparison conditions below, not
1894  // and/or.
1895  auto *BranchCond = dyn_cast<CmpInst>(PBranch->Condition);
1896  if (!BranchCond)
1897  continue;
1898  auto *BranchOp0 = lookupOperandLeader(BranchCond->getOperand(0));
1899  auto *BranchOp1 = lookupOperandLeader(BranchCond->getOperand(1));
1900  auto BranchPredicate = BranchCond->getPredicate();
1901  if (shouldSwapOperands(BranchOp0, BranchOp1)) {
1902  std::swap(BranchOp0, BranchOp1);
1903  BranchPredicate = BranchCond->getSwappedPredicate();
1904  }
1905  if (BranchOp0 == Op0 && BranchOp1 == Op1) {
1906  if (PBranch->TrueEdge) {
1907  // If we know the previous predicate is true and we are in the true
1908  // edge then we may be implied true or false.
1909  if (CmpInst::isImpliedTrueByMatchingCmp(BranchPredicate,
1910  OurPredicate)) {
1911  return ExprResult::some(
1912  createConstantExpression(ConstantInt::getTrue(CI->getType())),
1913  PI);
1914  }
1915 
1916  if (CmpInst::isImpliedFalseByMatchingCmp(BranchPredicate,
1917  OurPredicate)) {
1918  return ExprResult::some(
1919  createConstantExpression(ConstantInt::getFalse(CI->getType())),
1920  PI);
1921  }
1922  } else {
1923  // Just handle the ne and eq cases, where if we have the same
1924  // operands, we may know something.
1925  if (BranchPredicate == OurPredicate) {
1926  // Same predicate, same ops,we know it was false, so this is false.
1927  return ExprResult::some(
1928  createConstantExpression(ConstantInt::getFalse(CI->getType())),
1929  PI);
1930  } else if (BranchPredicate ==
1931  CmpInst::getInversePredicate(OurPredicate)) {
1932  // Inverse predicate, we know the other was false, so this is true.
1933  return ExprResult::some(
1934  createConstantExpression(ConstantInt::getTrue(CI->getType())),
1935  PI);
1936  }
1937  }
1938  }
1939  }
1940  }
1941  // Create expression will take care of simplifyCmpInst
1942  return createExpression(I);
1943 }
1944 
1945 // Substitute and symbolize the value before value numbering.
1946 NewGVN::ExprResult
1947 NewGVN::performSymbolicEvaluation(Value *V,
1948  SmallPtrSetImpl<Value *> &Visited) const {
1949 
1950  const Expression *E = nullptr;
1951  if (auto *C = dyn_cast<Constant>(V))
1952  E = createConstantExpression(C);
1953  else if (isa<Argument>(V) || isa<GlobalVariable>(V)) {
1954  E = createVariableExpression(V);
1955  } else {
1956  // TODO: memory intrinsics.
1957  // TODO: Some day, we should do the forward propagation and reassociation
1958  // parts of the algorithm.
1959  auto *I = cast<Instruction>(V);
1960  switch (I->getOpcode()) {
1961  case Instruction::ExtractValue:
1962  case Instruction::InsertValue:
1963  E = performSymbolicAggrValueEvaluation(I);
1964  break;
1965  case Instruction::PHI: {
1967  auto *PN = cast<PHINode>(I);
1968  for (unsigned i = 0; i < PN->getNumOperands(); ++i)
1969  Ops.push_back({PN->getIncomingValue(i), PN->getIncomingBlock(i)});
1970  // Sort to ensure the invariant createPHIExpression requires is met.
1971  sortPHIOps(Ops);
1972  E = performSymbolicPHIEvaluation(Ops, I, getBlockForValue(I));
1973  } break;
1974  case Instruction::Call:
1975  return performSymbolicCallEvaluation(I);
1976  break;
1977  case Instruction::Store:
1978  E = performSymbolicStoreEvaluation(I);
1979  break;
1980  case Instruction::Load:
1981  E = performSymbolicLoadEvaluation(I);
1982  break;
1983  case Instruction::BitCast:
1984  case Instruction::AddrSpaceCast:
1985  return createExpression(I);
1986  break;
1987  case Instruction::ICmp:
1988  case Instruction::FCmp:
1989  return performSymbolicCmpEvaluation(I);
1990  break;
1991  case Instruction::FNeg:
1992  case Instruction::Add:
1993  case Instruction::FAdd:
1994  case Instruction::Sub:
1995  case Instruction::FSub:
1996  case Instruction::Mul:
1997  case Instruction::FMul:
1998  case Instruction::UDiv:
1999  case Instruction::SDiv:
2000  case Instruction::FDiv:
2001  case Instruction::URem:
2002  case Instruction::SRem:
2003  case Instruction::FRem:
2004  case Instruction::Shl:
2005  case Instruction::LShr:
2006  case Instruction::AShr:
2007  case Instruction::And:
2008  case Instruction::Or:
2009  case Instruction::Xor:
2010  case Instruction::Trunc:
2011  case Instruction::ZExt:
2012  case Instruction::SExt:
2013  case Instruction::FPToUI:
2014  case Instruction::FPToSI:
2015  case Instruction::UIToFP:
2016  case Instruction::SIToFP:
2017  case Instruction::FPTrunc:
2018  case Instruction::FPExt:
2019  case Instruction::PtrToInt:
2020  case Instruction::IntToPtr:
2021  case Instruction::Select:
2022  case Instruction::ExtractElement:
2023  case Instruction::InsertElement:
2024  case Instruction::GetElementPtr:
2025  return createExpression(I);
2026  break;
2027  case Instruction::ShuffleVector:
2028  // FIXME: Add support for shufflevector to createExpression.
2029  return ExprResult::none();
2030  default:
2031  return ExprResult::none();
2032  }
2033  }
2034  return ExprResult::some(E);
2035 }
2036 
2037 // Look up a container of values/instructions in a map, and touch all the
2038 // instructions in the container. Then erase value from the map.
2039 template <typename Map, typename KeyType>
2040 void NewGVN::touchAndErase(Map &M, const KeyType &Key) {
2041  const auto Result = M.find_as(Key);
2042  if (Result != M.end()) {
2043  for (const typename Map::mapped_type::value_type Mapped : Result->second)
2044  TouchedInstructions.set(InstrToDFSNum(Mapped));
2045  M.erase(Result);
2046  }
2047 }
2048 
2049 void NewGVN::addAdditionalUsers(Value *To, Value *User) const {
2050  assert(User && To != User);
2051  if (isa<Instruction>(To))
2052  AdditionalUsers[To].insert(User);
2053 }
2054 
2055 void NewGVN::addAdditionalUsers(ExprResult &Res, Instruction *User) const {
2056  if (Res.ExtraDep && Res.ExtraDep != User)
2057  addAdditionalUsers(Res.ExtraDep, User);
2058  Res.ExtraDep = nullptr;
2059 
2060  if (Res.PredDep) {
2061  if (const auto *PBranch = dyn_cast<PredicateBranch>(Res.PredDep))
2062  PredicateToUsers[PBranch->Condition].insert(User);
2063  else if (const auto *PAssume = dyn_cast<PredicateAssume>(Res.PredDep))
2064  PredicateToUsers[PAssume->Condition].insert(User);
2065  }
2066  Res.PredDep = nullptr;
2067 }
2068 
2069 void NewGVN::markUsersTouched(Value *V) {
2070  // Now mark the users as touched.
2071  for (auto *User : V->users()) {
2072  assert(isa<Instruction>(User) && "Use of value not within an instruction?");
2073  TouchedInstructions.set(InstrToDFSNum(User));
2074  }
2075  touchAndErase(AdditionalUsers, V);
2076 }
2077 
2078 void NewGVN::addMemoryUsers(const MemoryAccess *To, MemoryAccess *U) const {
2079  LLVM_DEBUG(dbgs() << "Adding memory user " << *U << " to " << *To << "\n");
2080  MemoryToUsers[To].insert(U);
2081 }
2082 
2083 void NewGVN::markMemoryDefTouched(const MemoryAccess *MA) {
2084  TouchedInstructions.set(MemoryToDFSNum(MA));
2085 }
2086 
2087 void NewGVN::markMemoryUsersTouched(const MemoryAccess *MA) {
2088  if (isa<MemoryUse>(MA))
2089  return;
2090  for (const auto *U : MA->users())
2091  TouchedInstructions.set(MemoryToDFSNum(U));
2092  touchAndErase(MemoryToUsers, MA);
2093 }
2094 
2095 // Touch all the predicates that depend on this instruction.
2096 void NewGVN::markPredicateUsersTouched(Instruction *I) {
2097  touchAndErase(PredicateToUsers, I);
2098 }
2099 
2100 // Mark users affected by a memory leader change.
2101 void NewGVN::markMemoryLeaderChangeTouched(CongruenceClass *CC) {
2102  for (const auto *M : CC->memory())
2103  markMemoryDefTouched(M);
2104 }
2105 
2106 // Touch the instructions that need to be updated after a congruence class has a
2107 // leader change, and mark changed values.
2108 void NewGVN::markValueLeaderChangeTouched(CongruenceClass *CC) {
2109  for (auto *M : *CC) {
2110  if (auto *I = dyn_cast<Instruction>(M))
2111  TouchedInstructions.set(InstrToDFSNum(I));
2112  LeaderChanges.insert(M);
2113  }
2114 }
2115 
2116 // Give a range of things that have instruction DFS numbers, this will return
2117 // the member of the range with the smallest dfs number.
2118 template <class T, class Range>
2119 T *NewGVN::getMinDFSOfRange(const Range &R) const {
2120  std::pair<T *, unsigned> MinDFS = {nullptr, ~0U};
2121  for (const auto X : R) {
2122  auto DFSNum = InstrToDFSNum(X);
2123  if (DFSNum < MinDFS.second)
2124  MinDFS = {X, DFSNum};
2125  }
2126  return MinDFS.first;
2127 }
2128 
2129 // This function returns the MemoryAccess that should be the next leader of
2130 // congruence class CC, under the assumption that the current leader is going to
2131 // disappear.
2132 const MemoryAccess *NewGVN::getNextMemoryLeader(CongruenceClass *CC) const {
2133  // TODO: If this ends up to slow, we can maintain a next memory leader like we
2134  // do for regular leaders.
2135  // Make sure there will be a leader to find.
2136  assert(!CC->definesNoMemory() && "Can't get next leader if there is none");
2137  if (CC->getStoreCount() > 0) {
2138  if (auto *NL = dyn_cast_or_null<StoreInst>(CC->getNextLeader().first))
2139  return getMemoryAccess(NL);
2140  // Find the store with the minimum DFS number.
2141  auto *V = getMinDFSOfRange<Value>(make_filter_range(
2142  *CC, [&](const Value *V) { return isa<StoreInst>(V); }));
2143  return getMemoryAccess(cast<StoreInst>(V));
2144  }
2145  assert(CC->getStoreCount() == 0);
2146 
2147  // Given our assertion, hitting this part must mean
2148  // !OldClass->memory_empty()
2149  if (CC->memory_size() == 1)
2150  return *CC->memory_begin();
2151  return getMinDFSOfRange<const MemoryPhi>(CC->memory());
2152 }
2153 
2154 // This function returns the next value leader of a congruence class, under the
2155 // assumption that the current leader is going away. This should end up being
2156 // the next most dominating member.
2157 Value *NewGVN::getNextValueLeader(CongruenceClass *CC) const {
2158  // We don't need to sort members if there is only 1, and we don't care about
2159  // sorting the TOP class because everything either gets out of it or is
2160  // unreachable.
2161 
2162  if (CC->size() == 1 || CC == TOPClass) {
2163  return *(CC->begin());
2164  } else if (CC->getNextLeader().first) {
2165  ++NumGVNAvoidedSortedLeaderChanges;
2166  return CC->getNextLeader().first;
2167  } else {
2168  ++NumGVNSortedLeaderChanges;
2169  // NOTE: If this ends up to slow, we can maintain a dual structure for
2170  // member testing/insertion, or keep things mostly sorted, and sort only
2171  // here, or use SparseBitVector or ....
2172  return getMinDFSOfRange<Value>(*CC);
2173  }
2174 }
2175 
2176 // Move a MemoryAccess, currently in OldClass, to NewClass, including updates to
2177 // the memory members, etc for the move.
2178 //
2179 // The invariants of this function are:
2180 //
2181 // - I must be moving to NewClass from OldClass
2182 // - The StoreCount of OldClass and NewClass is expected to have been updated
2183 // for I already if it is a store.
2184 // - The OldClass memory leader has not been updated yet if I was the leader.
2185 void NewGVN::moveMemoryToNewCongruenceClass(Instruction *I,
2186  MemoryAccess *InstMA,
2187  CongruenceClass *OldClass,
2188  CongruenceClass *NewClass) {
2189  // If the leader is I, and we had a representative MemoryAccess, it should
2190  // be the MemoryAccess of OldClass.
2191  assert((!InstMA || !OldClass->getMemoryLeader() ||
2192  OldClass->getLeader() != I ||
2193  MemoryAccessToClass.lookup(OldClass->getMemoryLeader()) ==
2194  MemoryAccessToClass.lookup(InstMA)) &&
2195  "Representative MemoryAccess mismatch");
2196  // First, see what happens to the new class
2197  if (!NewClass->getMemoryLeader()) {
2198  // Should be a new class, or a store becoming a leader of a new class.
2199  assert(NewClass->size() == 1 ||
2200  (isa<StoreInst>(I) && NewClass->getStoreCount() == 1));
2201  NewClass->setMemoryLeader(InstMA);
2202  // Mark it touched if we didn't just create a singleton
2203  LLVM_DEBUG(dbgs() << "Memory class leader change for class "
2204  << NewClass->getID()
2205  << " due to new memory instruction becoming leader\n");
2206  markMemoryLeaderChangeTouched(NewClass);
2207  }
2208  setMemoryClass(InstMA, NewClass);
2209  // Now, fixup the old class if necessary
2210  if (OldClass->getMemoryLeader() == InstMA) {
2211  if (!OldClass->definesNoMemory()) {
2212  OldClass->setMemoryLeader(getNextMemoryLeader(OldClass));
2213  LLVM_DEBUG(dbgs() << "Memory class leader change for class "
2214  << OldClass->getID() << " to "
2215  << *OldClass->getMemoryLeader()
2216  << " due to removal of old leader " << *InstMA << "\n");
2217  markMemoryLeaderChangeTouched(OldClass);
2218  } else
2219  OldClass->setMemoryLeader(nullptr);
2220  }
2221 }
2222 
2223 // Move a value, currently in OldClass, to be part of NewClass
2224 // Update OldClass and NewClass for the move (including changing leaders, etc).
2225 void NewGVN::moveValueToNewCongruenceClass(Instruction *I, const Expression *E,
2226  CongruenceClass *OldClass,
2227  CongruenceClass *NewClass) {
2228  if (I == OldClass->getNextLeader().first)
2229  OldClass->resetNextLeader();
2230 
2231  OldClass->erase(I);
2232  NewClass->insert(I);
2233 
2234  if (NewClass->getLeader() != I)
2235  NewClass->addPossibleNextLeader({I, InstrToDFSNum(I)});
2236  // Handle our special casing of stores.
2237  if (auto *SI = dyn_cast<StoreInst>(I)) {
2238  OldClass->decStoreCount();
2239  // Okay, so when do we want to make a store a leader of a class?
2240  // If we have a store defined by an earlier load, we want the earlier load
2241  // to lead the class.
2242  // If we have a store defined by something else, we want the store to lead
2243  // the class so everything else gets the "something else" as a value.
2244  // If we have a store as the single member of the class, we want the store
2245  // as the leader
2246  if (NewClass->getStoreCount() == 0 && !NewClass->getStoredValue()) {
2247  // If it's a store expression we are using, it means we are not equivalent
2248  // to something earlier.
2249  if (auto *SE = dyn_cast<StoreExpression>(E)) {
2250  NewClass->setStoredValue(SE->getStoredValue());
2251  markValueLeaderChangeTouched(NewClass);
2252  // Shift the new class leader to be the store
2253  LLVM_DEBUG(dbgs() << "Changing leader of congruence class "
2254  << NewClass->getID() << " from "
2255  << *NewClass->getLeader() << " to " << *SI
2256  << " because store joined class\n");
2257  // If we changed the leader, we have to mark it changed because we don't
2258  // know what it will do to symbolic evaluation.
2259  NewClass->setLeader(SI);
2260  }
2261  // We rely on the code below handling the MemoryAccess change.
2262  }
2263  NewClass->incStoreCount();
2264  }
2265  // True if there is no memory instructions left in a class that had memory
2266  // instructions before.
2267 
2268  // If it's not a memory use, set the MemoryAccess equivalence
2269  auto *InstMA = dyn_cast_or_null<MemoryDef>(getMemoryAccess(I));
2270  if (InstMA)
2271  moveMemoryToNewCongruenceClass(I, InstMA, OldClass, NewClass);
2272  ValueToClass[I] = NewClass;
2273  // See if we destroyed the class or need to swap leaders.
2274  if (OldClass->empty() && OldClass != TOPClass) {
2275  if (OldClass->getDefiningExpr()) {
2276  LLVM_DEBUG(dbgs() << "Erasing expression " << *OldClass->getDefiningExpr()
2277  << " from table\n");
2278  // We erase it as an exact expression to make sure we don't just erase an
2279  // equivalent one.
2280  auto Iter = ExpressionToClass.find_as(
2281  ExactEqualsExpression(*OldClass->getDefiningExpr()));
2282  if (Iter != ExpressionToClass.end())
2283  ExpressionToClass.erase(Iter);
2284 #ifdef EXPENSIVE_CHECKS
2285  assert(
2286  (*OldClass->getDefiningExpr() != *E || ExpressionToClass.lookup(E)) &&
2287  "We erased the expression we just inserted, which should not happen");
2288 #endif
2289  }
2290  } else if (OldClass->getLeader() == I) {
2291  // When the leader changes, the value numbering of
2292  // everything may change due to symbolization changes, so we need to
2293  // reprocess.
2294  LLVM_DEBUG(dbgs() << "Value class leader change for class "
2295  << OldClass->getID() << "\n");
2296  ++NumGVNLeaderChanges;
2297  // Destroy the stored value if there are no more stores to represent it.
2298  // Note that this is basically clean up for the expression removal that
2299  // happens below. If we remove stores from a class, we may leave it as a
2300  // class of equivalent memory phis.
2301  if (OldClass->getStoreCount() == 0) {
2302  if (OldClass->getStoredValue())
2303  OldClass->setStoredValue(nullptr);
2304  }
2305  OldClass->setLeader(getNextValueLeader(OldClass));
2306  OldClass->resetNextLeader();
2307  markValueLeaderChangeTouched(OldClass);
2308  }
2309 }
2310 
2311 // For a given expression, mark the phi of ops instructions that could have
2312 // changed as a result.
2313 void NewGVN::markPhiOfOpsChanged(const Expression *E) {
2314  touchAndErase(ExpressionToPhiOfOps, E);
2315 }
2316 
2317 // Perform congruence finding on a given value numbering expression.
2318 void NewGVN::performCongruenceFinding(Instruction *I, const Expression *E) {
2319  // This is guaranteed to return something, since it will at least find
2320  // TOP.
2321 
2322  CongruenceClass *IClass = ValueToClass.lookup(I);
2323  assert(IClass && "Should have found a IClass");
2324  // Dead classes should have been eliminated from the mapping.
2325  assert(!IClass->isDead() && "Found a dead class");
2326 
2327  CongruenceClass *EClass = nullptr;
2328  if (const auto *VE = dyn_cast<VariableExpression>(E)) {
2329  EClass = ValueToClass.lookup(VE->getVariableValue());
2330  } else if (isa<DeadExpression>(E)) {
2331  EClass = TOPClass;
2332  }
2333  if (!EClass) {
2334  auto lookupResult = ExpressionToClass.insert({E, nullptr});
2335 
2336  // If it's not in the value table, create a new congruence class.
2337  if (lookupResult.second) {
2338  CongruenceClass *NewClass = createCongruenceClass(nullptr, E);
2339  auto place = lookupResult.first;
2340  place->second = NewClass;
2341 
2342  // Constants and variables should always be made the leader.
2343  if (const auto *CE = dyn_cast<ConstantExpression>(E)) {
2344  NewClass->setLeader(CE->getConstantValue());
2345  } else if (const auto *SE = dyn_cast<StoreExpression>(E)) {
2346  StoreInst *SI = SE->getStoreInst();
2347  NewClass->setLeader(SI);
2348  NewClass->setStoredValue(SE->getStoredValue());
2349  // The RepMemoryAccess field will be filled in properly by the
2350  // moveValueToNewCongruenceClass call.
2351  } else {
2352  NewClass->setLeader(I);
2353  }
2354  assert(!isa<VariableExpression>(E) &&
2355  "VariableExpression should have been handled already");
2356 
2357  EClass = NewClass;
2358  LLVM_DEBUG(dbgs() << "Created new congruence class for " << *I
2359  << " using expression " << *E << " at "
2360  << NewClass->getID() << " and leader "
2361  << *(NewClass->getLeader()));
2362  if (NewClass->getStoredValue())
2363  LLVM_DEBUG(dbgs() << " and stored value "
2364  << *(NewClass->getStoredValue()));
2365  LLVM_DEBUG(dbgs() << "\n");
2366  } else {
2367  EClass = lookupResult.first->second;
2368  if (isa<ConstantExpression>(E))
2369  assert((isa<Constant>(EClass->getLeader()) ||
2370  (EClass->getStoredValue() &&
2371  isa<Constant>(EClass->getStoredValue()))) &&
2372  "Any class with a constant expression should have a "
2373  "constant leader");
2374 
2375  assert(EClass && "Somehow don't have an eclass");
2376 
2377  assert(!EClass->isDead() && "We accidentally looked up a dead class");
2378  }
2379  }
2380  bool ClassChanged = IClass != EClass;
2381  bool LeaderChanged = LeaderChanges.erase(I);
2382  if (ClassChanged || LeaderChanged) {
2383  LLVM_DEBUG(dbgs() << "New class " << EClass->getID() << " for expression "
2384  << *E << "\n");
2385  if (ClassChanged) {
2386  moveValueToNewCongruenceClass(I, E, IClass, EClass);
2387  markPhiOfOpsChanged(E);
2388  }
2389 
2390  markUsersTouched(I);
2391  if (MemoryAccess *MA = getMemoryAccess(I))
2392  markMemoryUsersTouched(MA);
2393  if (auto *CI = dyn_cast<CmpInst>(I))
2394  markPredicateUsersTouched(CI);
2395  }
2396  // If we changed the class of the store, we want to ensure nothing finds the
2397  // old store expression. In particular, loads do not compare against stored
2398  // value, so they will find old store expressions (and associated class
2399  // mappings) if we leave them in the table.
2400  if (ClassChanged && isa<StoreInst>(I)) {
2401  auto *OldE = ValueToExpression.lookup(I);
2402  // It could just be that the old class died. We don't want to erase it if we
2403  // just moved classes.
2404  if (OldE && isa<StoreExpression>(OldE) && *E != *OldE) {
2405  // Erase this as an exact expression to ensure we don't erase expressions
2406  // equivalent to it.
2407  auto Iter = ExpressionToClass.find_as(ExactEqualsExpression(*OldE));
2408  if (Iter != ExpressionToClass.end())
2409  ExpressionToClass.erase(Iter);
2410  }
2411  }
2412  ValueToExpression[I] = E;
2413 }
2414 
2415 // Process the fact that Edge (from, to) is reachable, including marking
2416 // any newly reachable blocks and instructions for processing.
2417 void NewGVN::updateReachableEdge(BasicBlock *From, BasicBlock *To) {
2418  // Check if the Edge was reachable before.
2419  if (ReachableEdges.insert({From, To}).second) {
2420  // If this block wasn't reachable before, all instructions are touched.
2421  if (ReachableBlocks.insert(To).second) {
2422  LLVM_DEBUG(dbgs() << "Block " << getBlockName(To)
2423  << " marked reachable\n");
2424  const auto &InstRange = BlockInstRange.lookup(To);
2425  TouchedInstructions.set(InstRange.first, InstRange.second);
2426  } else {
2427  LLVM_DEBUG(dbgs() << "Block " << getBlockName(To)
2428  << " was reachable, but new edge {"
2429  << getBlockName(From) << "," << getBlockName(To)
2430  << "} to it found\n");
2431 
2432  // We've made an edge reachable to an existing block, which may
2433  // impact predicates. Otherwise, only mark the phi nodes as touched, as
2434  // they are the only thing that depend on new edges. Anything using their
2435  // values will get propagated to if necessary.
2436  if (MemoryAccess *MemPhi = getMemoryAccess(To))
2437  TouchedInstructions.set(InstrToDFSNum(MemPhi));
2438 
2439  // FIXME: We should just add a union op on a Bitvector and
2440  // SparseBitVector. We can do it word by word faster than we are doing it
2441  // here.
2442  for (auto InstNum : RevisitOnReachabilityChange[To])
2443  TouchedInstructions.set(InstNum);
2444  }
2445  }
2446 }
2447 
2448 // Given a predicate condition (from a switch, cmp, or whatever) and a block,
2449 // see if we know some constant value for it already.
2450 Value *NewGVN::findConditionEquivalence(Value *Cond) const {
2451  auto Result = lookupOperandLeader(Cond);
2452  return isa<Constant>(Result) ? Result : nullptr;
2453 }
2454 
2455 // Process the outgoing edges of a block for reachability.
2456 void NewGVN::processOutgoingEdges(Instruction *TI, BasicBlock *B) {
2457  // Evaluate reachability of terminator instruction.
2458  Value *Cond;
2459  BasicBlock *TrueSucc, *FalseSucc;
2460  if (match(TI, m_Br(m_Value(Cond), TrueSucc, FalseSucc))) {
2461  Value *CondEvaluated = findConditionEquivalence(Cond);
2462  if (!CondEvaluated) {
2463  if (auto *I = dyn_cast<Instruction>(Cond)) {
2464  SmallPtrSet<Value *, 4> Visited;
2465  auto Res = performSymbolicEvaluation(I, Visited);
2466  if (const auto *CE = dyn_cast_or_null<ConstantExpression>(Res.Expr)) {
2467  CondEvaluated = CE->getConstantValue();
2468  addAdditionalUsers(Res, I);
2469  } else {
2470  // Did not use simplification result, no need to add the extra
2471  // dependency.
2472  Res.ExtraDep = nullptr;
2473  }
2474  } else if (isa<ConstantInt>(Cond)) {
2475  CondEvaluated = Cond;
2476  }
2477  }
2478  ConstantInt *CI;
2479  if (CondEvaluated && (CI = dyn_cast<ConstantInt>(CondEvaluated))) {
2480  if (CI->isOne()) {
2481  LLVM_DEBUG(dbgs() << "Condition for Terminator " << *TI
2482  << " evaluated to true\n");
2483  updateReachableEdge(B, TrueSucc);
2484  } else if (CI->isZero()) {
2485  LLVM_DEBUG(dbgs() << "Condition for Terminator " << *TI
2486  << " evaluated to false\n");
2487  updateReachableEdge(B, FalseSucc);
2488  }
2489  } else {
2490  updateReachableEdge(B, TrueSucc);
2491  updateReachableEdge(B, FalseSucc);
2492  }
2493  } else if (auto *SI = dyn_cast<SwitchInst>(TI)) {
2494  // For switches, propagate the case values into the case
2495  // destinations.
2496 
2497  Value *SwitchCond = SI->getCondition();
2498  Value *CondEvaluated = findConditionEquivalence(SwitchCond);
2499  // See if we were able to turn this switch statement into a constant.
2500  if (CondEvaluated && isa<ConstantInt>(CondEvaluated)) {
2501  auto *CondVal = cast<ConstantInt>(CondEvaluated);
2502  // We should be able to get case value for this.
2503  auto Case = *SI->findCaseValue(CondVal);
2504  if (Case.getCaseSuccessor() == SI->getDefaultDest()) {
2505  // We proved the value is outside of the range of the case.
2506  // We can't do anything other than mark the default dest as reachable,
2507  // and go home.
2508  updateReachableEdge(B, SI->getDefaultDest());
2509  return;
2510  }
2511  // Now get where it goes and mark it reachable.
2512  BasicBlock *TargetBlock = Case.getCaseSuccessor();
2513  updateReachableEdge(B, TargetBlock);
2514  } else {
2515  for (unsigned i = 0, e = SI->getNumSuccessors(); i != e; ++i) {
2516  BasicBlock *TargetBlock = SI->getSuccessor(i);
2517  updateReachableEdge(B, TargetBlock);
2518  }
2519  }
2520  } else {
2521  // Otherwise this is either unconditional, or a type we have no
2522  // idea about. Just mark successors as reachable.
2523  for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i) {
2524  BasicBlock *TargetBlock = TI->getSuccessor(i);
2525  updateReachableEdge(B, TargetBlock);
2526  }
2527 
2528  // This also may be a memory defining terminator, in which case, set it
2529  // equivalent only to itself.
2530  //
2531  auto *MA = getMemoryAccess(TI);
2532  if (MA && !isa<MemoryUse>(MA)) {
2533  auto *CC = ensureLeaderOfMemoryClass(MA);
2534  if (setMemoryClass(MA, CC))
2535  markMemoryUsersTouched(MA);
2536  }
2537  }
2538 }
2539 
2540 // Remove the PHI of Ops PHI for I
2541 void NewGVN::removePhiOfOps(Instruction *I, PHINode *PHITemp) {
2542  InstrDFS.erase(PHITemp);
2543  // It's still a temp instruction. We keep it in the array so it gets erased.
2544  // However, it's no longer used by I, or in the block
2545  TempToBlock.erase(PHITemp);
2546  RealToTemp.erase(I);
2547  // We don't remove the users from the phi node uses. This wastes a little
2548  // time, but such is life. We could use two sets to track which were there
2549  // are the start of NewGVN, and which were added, but right nowt he cost of
2550  // tracking is more than the cost of checking for more phi of ops.
2551 }
2552 
2553 // Add PHI Op in BB as a PHI of operations version of ExistingValue.
2554 void NewGVN::addPhiOfOps(PHINode *Op, BasicBlock *BB,
2555  Instruction *ExistingValue) {
2556  InstrDFS[Op] = InstrToDFSNum(ExistingValue);
2557  AllTempInstructions.insert(Op);
2558  TempToBlock[Op] = BB;
2559  RealToTemp[ExistingValue] = Op;
2560  // Add all users to phi node use, as they are now uses of the phi of ops phis
2561  // and may themselves be phi of ops.
2562  for (auto *U : ExistingValue->users())
2563  if (auto *UI = dyn_cast<Instruction>(U))
2564  PHINodeUses.insert(UI);
2565 }
2566 
2567 static bool okayForPHIOfOps(const Instruction *I) {
2568  if (!EnablePhiOfOps)
2569  return false;
2570  return isa<BinaryOperator>(I) || isa<SelectInst>(I) || isa<CmpInst>(I) ||
2571  isa<LoadInst>(I);
2572 }
2573 
2574 // Return true if this operand will be safe to use for phi of ops.
2575 //
2576 // The reason some operands are unsafe is that we are not trying to recursively
2577 // translate everything back through phi nodes. We actually expect some lookups
2578 // of expressions to fail. In particular, a lookup where the expression cannot
2579 // exist in the predecessor. This is true even if the expression, as shown, can
2580 // be determined to be constant.
2581 bool NewGVN::OpIsSafeForPHIOfOps(Value *V, const BasicBlock *PHIBlock,
2582  SmallPtrSetImpl<const Value *> &Visited) {
2583  SmallVector<Value *, 4> Worklist;
2584  Worklist.push_back(V);
2585  while (!Worklist.empty()) {
2586  auto *I = Worklist.pop_back_val();
2587  if (!isa<Instruction>(I))
2588  continue;
2589 
2590  auto OISIt = OpSafeForPHIOfOps.find(I);
2591  if (OISIt != OpSafeForPHIOfOps.end())
2592  return OISIt->second;
2593 
2594  // Keep walking until we either dominate the phi block, or hit a phi, or run
2595  // out of things to check.
2596  if (DT->properlyDominates(getBlockForValue(I), PHIBlock)) {
2597  OpSafeForPHIOfOps.insert({I, true});
2598  continue;
2599  }
2600  // PHI in the same block.
2601  if (isa<PHINode>(I) && getBlockForValue(I) == PHIBlock) {
2602  OpSafeForPHIOfOps.insert({I, false});
2603  return false;
2604  }
2605 
2606  auto *OrigI = cast<Instruction>(I);
2607  // When we hit an instruction that reads memory (load, call, etc), we must
2608  // consider any store that may happen in the loop. For now, we assume the
2609  // worst: there is a store in the loop that alias with this read.
2610  // The case where the load is outside the loop is already covered by the
2611  // dominator check above.
2612  // TODO: relax this condition
2613  if (OrigI->mayReadFromMemory())
2614  return false;
2615 
2616  // Check the operands of the current instruction.
2617  for (auto *Op : OrigI->operand_values()) {
2618  if (!isa<Instruction>(Op))
2619  continue;
2620  // Stop now if we find an unsafe operand.
2621  auto OISIt = OpSafeForPHIOfOps.find(OrigI);
2622  if (OISIt != OpSafeForPHIOfOps.end()) {
2623  if (!OISIt->second) {
2624  OpSafeForPHIOfOps.insert({I, false});
2625  return false;
2626  }
2627  continue;
2628  }
2629  if (!Visited.insert(Op).second)
2630  continue;
2631  Worklist.push_back(cast<Instruction>(Op));
2632  }
2633  }
2634  OpSafeForPHIOfOps.insert({V, true});
2635  return true;
2636 }
2637 
2638 // Try to find a leader for instruction TransInst, which is a phi translated
2639 // version of something in our original program. Visited is used to ensure we
2640 // don't infinite loop during translations of cycles. OrigInst is the
2641 // instruction in the original program, and PredBB is the predecessor we
2642 // translated it through.
2643 Value *NewGVN::findLeaderForInst(Instruction *TransInst,
2644  SmallPtrSetImpl<Value *> &Visited,
2645  MemoryAccess *MemAccess, Instruction *OrigInst,
2646  BasicBlock *PredBB) {
2647  unsigned IDFSNum = InstrToDFSNum(OrigInst);
2648  // Make sure it's marked as a temporary instruction.
2649  AllTempInstructions.insert(TransInst);
2650  // and make sure anything that tries to add it's DFS number is
2651  // redirected to the instruction we are making a phi of ops
2652  // for.
2653  TempToBlock.insert({TransInst, PredBB});
2654  InstrDFS.insert({TransInst, IDFSNum});
2655 
2656  auto Res = performSymbolicEvaluation(TransInst, Visited);
2657  const Expression *E = Res.Expr;
2658  addAdditionalUsers(Res, OrigInst);
2659  InstrDFS.erase(TransInst);
2660  AllTempInstructions.erase(TransInst);
2661  TempToBlock.erase(TransInst);
2662  if (MemAccess)
2663  TempToMemory.erase(TransInst);
2664  if (!E)
2665  return nullptr;
2666  auto *FoundVal = findPHIOfOpsLeader(E, OrigInst, PredBB);
2667  if (!FoundVal) {
2668  ExpressionToPhiOfOps[E].insert(OrigInst);
2669  LLVM_DEBUG(dbgs() << "Cannot find phi of ops operand for " << *TransInst
2670  << " in block " << getBlockName(PredBB) << "\n");
2671  return nullptr;
2672  }
2673  if (auto *SI = dyn_cast<StoreInst>(FoundVal))
2674  FoundVal = SI->getValueOperand();
2675  return FoundVal;
2676 }
2677 
2678 // When we see an instruction that is an op of phis, generate the equivalent phi
2679 // of ops form.
2680 const Expression *
2681 NewGVN::makePossiblePHIOfOps(Instruction *I,
2682  SmallPtrSetImpl<Value *> &Visited) {
2683  if (!okayForPHIOfOps(I))
2684  return nullptr;
2685 
2686  if (!Visited.insert(I).second)
2687  return nullptr;
2688  // For now, we require the instruction be cycle free because we don't
2689  // *always* create a phi of ops for instructions that could be done as phi
2690  // of ops, we only do it if we think it is useful. If we did do it all the
2691  // time, we could remove the cycle free check.
2692  if (!isCycleFree(I))
2693  return nullptr;
2694 
2695  SmallPtrSet<const Value *, 8> ProcessedPHIs;
2696  // TODO: We don't do phi translation on memory accesses because it's
2697  // complicated. For a load, we'd need to be able to simulate a new memoryuse,
2698  // which we don't have a good way of doing ATM.
2699  auto *MemAccess = getMemoryAccess(I);
2700  // If the memory operation is defined by a memory operation this block that
2701  // isn't a MemoryPhi, transforming the pointer backwards through a scalar phi
2702  // can't help, as it would still be killed by that memory operation.
2703  if (MemAccess && !isa<MemoryPhi>(MemAccess->getDefiningAccess()) &&
2704  MemAccess->getDefiningAccess()->getBlock() == I->getParent())
2705  return nullptr;
2706 
2707  // Convert op of phis to phi of ops
2708  SmallPtrSet<const Value *, 10> VisitedOps;
2709  SmallVector<Value *, 4> Ops(I->operand_values());
2710  BasicBlock *SamePHIBlock = nullptr;
2711  PHINode *OpPHI = nullptr;
2712  if (!DebugCounter::shouldExecute(PHIOfOpsCounter))
2713  return nullptr;
2714  for (auto *Op : Ops) {
2715  if (!isa<PHINode>(Op)) {
2716  auto *ValuePHI = RealToTemp.lookup(Op);
2717  if (!ValuePHI)
2718  continue;
2719  LLVM_DEBUG(dbgs() << "Found possible dependent phi of ops\n");
2720  Op = ValuePHI;
2721  }
2722  OpPHI = cast<PHINode>(Op);
2723  if (!SamePHIBlock) {
2724  SamePHIBlock = getBlockForValue(OpPHI);
2725  } else if (SamePHIBlock != getBlockForValue(OpPHI)) {
2726  LLVM_DEBUG(
2727  dbgs()
2728  << "PHIs for operands are not all in the same block, aborting\n");
2729  return nullptr;
2730  }
2731  // No point in doing this for one-operand phis.
2732  if (OpPHI->getNumOperands() == 1) {
2733  OpPHI = nullptr;
2734  continue;
2735  }
2736  }
2737 
2738  if (!OpPHI)
2739  return nullptr;
2740 
2741  SmallVector<ValPair, 4> PHIOps;
2743  auto *PHIBlock = getBlockForValue(OpPHI);
2744  RevisitOnReachabilityChange[PHIBlock].reset(InstrToDFSNum(I));
2745  for (unsigned PredNum = 0; PredNum < OpPHI->getNumOperands(); ++PredNum) {
2746  auto *PredBB = OpPHI->getIncomingBlock(PredNum);
2747  Value *FoundVal = nullptr;
2748  SmallPtrSet<Value *, 4> CurrentDeps;
2749  // We could just skip unreachable edges entirely but it's tricky to do
2750  // with rewriting existing phi nodes.
2751  if (ReachableEdges.count({PredBB, PHIBlock})) {
2752  // Clone the instruction, create an expression from it that is
2753  // translated back into the predecessor, and see if we have a leader.
2754  Instruction *ValueOp = I->clone();
2755  if (MemAccess)
2756  TempToMemory.insert({ValueOp, MemAccess});
2757  bool SafeForPHIOfOps = true;
2758  VisitedOps.clear();
2759  for (auto &Op : ValueOp->operands()) {
2760  auto *OrigOp = &*Op;
2761  // When these operand changes, it could change whether there is a
2762  // leader for us or not, so we have to add additional users.
2763  if (isa<PHINode>(Op)) {
2764  Op = Op->DoPHITranslation(PHIBlock, PredBB);
2765  if (Op != OrigOp && Op != I)
2766  CurrentDeps.insert(Op);
2767  } else if (auto *ValuePHI = RealToTemp.lookup(Op)) {
2768  if (getBlockForValue(ValuePHI) == PHIBlock)
2769  Op = ValuePHI->getIncomingValueForBlock(PredBB);
2770  }
2771  // If we phi-translated the op, it must be safe.
2772  SafeForPHIOfOps =
2773  SafeForPHIOfOps &&
2774  (Op != OrigOp || OpIsSafeForPHIOfOps(Op, PHIBlock, VisitedOps));
2775  }
2776  // FIXME: For those things that are not safe we could generate
2777  // expressions all the way down, and see if this comes out to a
2778  // constant. For anything where that is true, and unsafe, we should
2779  // have made a phi-of-ops (or value numbered it equivalent to something)
2780  // for the pieces already.
2781  FoundVal = !SafeForPHIOfOps ? nullptr
2782  : findLeaderForInst(ValueOp, Visited,
2783  MemAccess, I, PredBB);
2784  ValueOp->deleteValue();
2785  if (!FoundVal) {
2786  // We failed to find a leader for the current ValueOp, but this might
2787  // change in case of the translated operands change.
2788  if (SafeForPHIOfOps)
2789  for (auto *Dep : CurrentDeps)
2790  addAdditionalUsers(Dep, I);
2791 
2792  return nullptr;
2793  }
2794  Deps.insert(CurrentDeps.begin(), CurrentDeps.end());
2795  } else {
2796  LLVM_DEBUG(dbgs() << "Skipping phi of ops operand for incoming block "
2797  << getBlockName(PredBB)
2798  << " because the block is unreachable\n");
2799  FoundVal = PoisonValue::get(I->getType());
2800  RevisitOnReachabilityChange[PHIBlock].set(InstrToDFSNum(I));
2801  }
2802 
2803  PHIOps.push_back({FoundVal, PredBB});
2804  LLVM_DEBUG(dbgs() << "Found phi of ops operand " << *FoundVal << " in "
2805  << getBlockName(PredBB) << "\n");
2806  }
2807  for (auto *Dep : Deps)
2808  addAdditionalUsers(Dep, I);
2809  sortPHIOps(PHIOps);
2810  auto *E = performSymbolicPHIEvaluation(PHIOps, I, PHIBlock);
2811  if (isa<ConstantExpression>(E) || isa<VariableExpression>(E)) {
2812  LLVM_DEBUG(
2813  dbgs()
2814  << "Not creating real PHI of ops because it simplified to existing "
2815  "value or constant\n");
2816  // We have leaders for all operands, but do not create a real PHI node with
2817  // those leaders as operands, so the link between the operands and the
2818  // PHI-of-ops is not materialized in the IR. If any of those leaders
2819  // changes, the PHI-of-op may change also, so we need to add the operands as
2820  // additional users.
2821  for (auto &O : PHIOps)
2822  addAdditionalUsers(O.first, I);
2823 
2824  return E;
2825  }
2826  auto *ValuePHI = RealToTemp.lookup(I);
2827  bool NewPHI = false;
2828  if (!ValuePHI) {
2829  ValuePHI =
2830  PHINode::Create(I->getType(), OpPHI->getNumOperands(), "phiofops");
2831  addPhiOfOps(ValuePHI, PHIBlock, I);
2832  NewPHI = true;
2833  NumGVNPHIOfOpsCreated++;
2834  }
2835  if (NewPHI) {
2836  for (auto PHIOp : PHIOps)
2837  ValuePHI->addIncoming(PHIOp.first, PHIOp.second);
2838  } else {
2839  TempToBlock[ValuePHI] = PHIBlock;
2840  unsigned int i = 0;
2841  for (auto PHIOp : PHIOps) {
2842  ValuePHI->setIncomingValue(i, PHIOp.first);
2843  ValuePHI->setIncomingBlock(i, PHIOp.second);
2844  ++i;
2845  }
2846  }
2847  RevisitOnReachabilityChange[PHIBlock].set(InstrToDFSNum(I));
2848  LLVM_DEBUG(dbgs() << "Created phi of ops " << *ValuePHI << " for " << *I
2849  << "\n");
2850 
2851  return E;
2852 }
2853 
2854 // The algorithm initially places the values of the routine in the TOP
2855 // congruence class. The leader of TOP is the undetermined value `poison`.
2856 // When the algorithm has finished, values still in TOP are unreachable.
2857 void NewGVN::initializeCongruenceClasses(Function &F) {
2858  NextCongruenceNum = 0;
2859 
2860  // Note that even though we use the live on entry def as a representative
2861  // MemoryAccess, it is *not* the same as the actual live on entry def. We
2862  // have no real equivalent to poison for MemoryAccesses, and so we really
2863  // should be checking whether the MemoryAccess is top if we want to know if it
2864  // is equivalent to everything. Otherwise, what this really signifies is that
2865  // the access "it reaches all the way back to the beginning of the function"
2866 
2867  // Initialize all other instructions to be in TOP class.
2868  TOPClass = createCongruenceClass(nullptr, nullptr);
2869  TOPClass->setMemoryLeader(MSSA->getLiveOnEntryDef());
2870  // The live on entry def gets put into it's own class
2871  MemoryAccessToClass[MSSA->getLiveOnEntryDef()] =
2872  createMemoryClass(MSSA->getLiveOnEntryDef());
2873 
2874  for (auto *DTN : nodes(DT)) {
2875  BasicBlock *BB = DTN->getBlock();
2876  // All MemoryAccesses are equivalent to live on entry to start. They must
2877  // be initialized to something so that initial changes are noticed. For
2878  // the maximal answer, we initialize them all to be the same as
2879  // liveOnEntry.
2880  auto *MemoryBlockDefs = MSSA->getBlockDefs(BB);
2881  if (MemoryBlockDefs)
2882  for (const auto &Def : *MemoryBlockDefs) {
2883  MemoryAccessToClass[&Def] = TOPClass;
2884  auto *MD = dyn_cast<MemoryDef>(&Def);
2885  // Insert the memory phis into the member list.
2886  if (!MD) {
2887  const MemoryPhi *MP = cast<MemoryPhi>(&Def);
2888  TOPClass->memory_insert(MP);
2889  MemoryPhiState.insert({MP, MPS_TOP});
2890  }
2891 
2892  if (MD && isa<StoreInst>(MD->getMemoryInst()))
2893  TOPClass->incStoreCount();
2894  }
2895 
2896  // FIXME: This is trying to discover which instructions are uses of phi
2897  // nodes. We should move this into one of the myriad of places that walk
2898  // all the operands already.
2899  for (auto &I : *BB) {
2900  if (isa<PHINode>(&I))
2901  for (auto *U : I.users())
2902  if (auto *UInst = dyn_cast<Instruction>(U))
2903  if (InstrToDFSNum(UInst) != 0 && okayForPHIOfOps(UInst))
2904  PHINodeUses.insert(UInst);
2905  // Don't insert void terminators into the class. We don't value number
2906  // them, and they just end up sitting in TOP.
2907  if (I.isTerminator() && I.getType()->isVoidTy())
2908  continue;
2909  TOPClass->insert(&I);
2910  ValueToClass[&I] = TOPClass;
2911  }
2912  }
2913 
2914  // Initialize arguments to be in their own unique congruence classes
2915  for (auto &FA : F.args())
2916  createSingletonCongruenceClass(&FA);
2917 }
2918 
2919 void NewGVN::cleanupTables() {
2920  for (CongruenceClass *&CC : CongruenceClasses) {
2921  LLVM_DEBUG(dbgs() << "Congruence class " << CC->getID() << " has "
2922  << CC->size() << " members\n");
2923  // Make sure we delete the congruence class (probably worth switching to
2924  // a unique_ptr at some point.
2925  delete CC;
2926  CC = nullptr;
2927  }
2928 
2929  // Destroy the value expressions
2930  SmallVector<Instruction *, 8> TempInst(AllTempInstructions.begin(),
2931  AllTempInstructions.end());
2932  AllTempInstructions.clear();
2933 
2934  // We have to drop all references for everything first, so there are no uses
2935  // left as we delete them.
2936  for (auto *I : TempInst) {
2937  I->dropAllReferences();
2938  }
2939 
2940  while (!TempInst.empty()) {
2941  auto *I = TempInst.pop_back_val();
2942  I->deleteValue();
2943  }
2944 
2945  ValueToClass.clear();
2946  ArgRecycler.clear(ExpressionAllocator);
2947  ExpressionAllocator.Reset();
2948  CongruenceClasses.clear();
2949  ExpressionToClass.clear();
2950  ValueToExpression.clear();
2951  RealToTemp.clear();
2952  AdditionalUsers.clear();
2953  ExpressionToPhiOfOps.clear();
2954  TempToBlock.clear();
2955  TempToMemory.clear();
2956  PHINodeUses.clear();
2957  OpSafeForPHIOfOps.clear();
2958  ReachableBlocks.clear();
2959  ReachableEdges.clear();
2960 #ifndef NDEBUG
2961  ProcessedCount.clear();
2962 #endif
2963  InstrDFS.clear();
2964  InstructionsToErase.clear();
2965  DFSToInstr.clear();
2966  BlockInstRange.clear();
2967  TouchedInstructions.clear();
2968  MemoryAccessToClass.clear();
2969  PredicateToUsers.clear();
2970  MemoryToUsers.clear();
2971  RevisitOnReachabilityChange.clear();
2972  IntrinsicInstPred.clear();
2973 }
2974 
2975 // Assign local DFS number mapping to instructions, and leave space for Value
2976 // PHI's.
2977 std::pair<unsigned, unsigned> NewGVN::assignDFSNumbers(BasicBlock *B,
2978  unsigned Start) {
2979  unsigned End = Start;
2980  if (MemoryAccess *MemPhi = getMemoryAccess(B)) {
2981  InstrDFS[MemPhi] = End++;
2982  DFSToInstr.emplace_back(MemPhi);
2983  }
2984 
2985  // Then the real block goes next.
2986  for (auto &I : *B) {
2987  // There's no need to call isInstructionTriviallyDead more than once on
2988  // an instruction. Therefore, once we know that an instruction is dead
2989  // we change its DFS number so that it doesn't get value numbered.
2990  if (isInstructionTriviallyDead(&I, TLI)) {
2991  InstrDFS[&I] = 0;
2992  LLVM_DEBUG(dbgs() << "Skipping trivially dead instruction " << I << "\n");
2993  markInstructionForDeletion(&I);
2994  continue;
2995  }
2996  if (isa<PHINode>(&I))
2997  RevisitOnReachabilityChange[B].set(End);
2998  InstrDFS[&I] = End++;
2999  DFSToInstr.emplace_back(&I);
3000  }
3001 
3002  // All of the range functions taken half-open ranges (open on the end side).
3003  // So we do not subtract one from count, because at this point it is one
3004  // greater than the last instruction.
3005  return std::make_pair(Start, End);
3006 }
3007 
3008 void NewGVN::updateProcessedCount(const Value *V) {
3009 #ifndef NDEBUG
3010  if (ProcessedCount.count(V) == 0) {
3011  ProcessedCount.insert({V, 1});
3012  } else {
3013  ++ProcessedCount[V];
3014  assert(ProcessedCount[V] < 100 &&
3015  "Seem to have processed the same Value a lot");
3016  }
3017 #endif
3018 }
3019 
3020 // Evaluate MemoryPhi nodes symbolically, just like PHI nodes
3021 void NewGVN::valueNumberMemoryPhi(MemoryPhi *MP) {
3022  // If all the arguments are the same, the MemoryPhi has the same value as the
3023  // argument. Filter out unreachable blocks and self phis from our operands.
3024  // TODO: We could do cycle-checking on the memory phis to allow valueizing for
3025  // self-phi checking.
3026  const BasicBlock *PHIBlock = MP->getBlock();
3027  auto Filtered = make_filter_range(MP->operands(), [&](const Use &U) {
3028  return cast<MemoryAccess>(U) != MP &&
3029  !isMemoryAccessTOP(cast<MemoryAccess>(U)) &&
3030  ReachableEdges.count({MP->getIncomingBlock(U), PHIBlock});
3031  });
3032  // If all that is left is nothing, our memoryphi is poison. We keep it as
3033  // InitialClass. Note: The only case this should happen is if we have at
3034  // least one self-argument.
3035  if (Filtered.begin() == Filtered.end()) {
3036  if (setMemoryClass(MP, TOPClass))
3037  markMemoryUsersTouched(MP);
3038  return;
3039  }
3040 
3041  // Transform the remaining operands into operand leaders.
3042  // FIXME: mapped_iterator should have a range version.
3043  auto LookupFunc = [&](const Use &U) {
3044  return lookupMemoryLeader(cast<MemoryAccess>(U));
3045  };
3046  auto MappedBegin = map_iterator(Filtered.begin(), LookupFunc);
3047  auto MappedEnd = map_iterator(Filtered.end(), LookupFunc);
3048 
3049  // and now check if all the elements are equal.
3050  // Sadly, we can't use std::equals since these are random access iterators.
3051  const auto *AllSameValue = *MappedBegin;
3052  ++MappedBegin;
3053  bool AllEqual = std::all_of(
3054  MappedBegin, MappedEnd,
3055  [&AllSameValue](const MemoryAccess *V) { return V == AllSameValue; });
3056 
3057  if (AllEqual)
3058  LLVM_DEBUG(dbgs() << "Memory Phi value numbered to " << *AllSameValue
3059  << "\n");
3060  else
3061  LLVM_DEBUG(dbgs() << "Memory Phi value numbered to itself\n");
3062  // If it's equal to something, it's in that class. Otherwise, it has to be in
3063  // a class where it is the leader (other things may be equivalent to it, but
3064  // it needs to start off in its own class, which means it must have been the
3065  // leader, and it can't have stopped being the leader because it was never
3066  // removed).
3067  CongruenceClass *CC =
3068  AllEqual ? getMemoryClass(AllSameValue) : ensureLeaderOfMemoryClass(MP);
3069  auto OldState = MemoryPhiState.lookup(MP);
3070  assert(OldState != MPS_Invalid && "Invalid memory phi state");
3071  auto NewState = AllEqual ? MPS_Equivalent : MPS_Unique;
3072  MemoryPhiState[MP] = NewState;
3073  if (setMemoryClass(MP, CC) || OldState != NewState)
3074  markMemoryUsersTouched(MP);
3075 }
3076 
3077 // Value number a single instruction, symbolically evaluating, performing
3078 // congruence finding, and updating mappings.
3079 void NewGVN::valueNumberInstruction(Instruction *I) {
3080  LLVM_DEBUG(dbgs() << "Processing instruction " << *I << "\n");
3081  if (!I->isTerminator()) {
3082  const Expression *Symbolized = nullptr;
3083  SmallPtrSet<Value *, 2> Visited;
3084  if (DebugCounter::shouldExecute(VNCounter)) {
3085  auto Res = performSymbolicEvaluation(I, Visited);
3086  Symbolized = Res.Expr;
3087  addAdditionalUsers(Res, I);
3088 
3089  // Make a phi of ops if necessary
3090  if (Symbolized && !isa<ConstantExpression>(Symbolized) &&
3091  !isa<VariableExpression>(Symbolized) && PHINodeUses.count(I)) {
3092  auto *PHIE = makePossiblePHIOfOps(I, Visited);
3093  // If we created a phi of ops, use it.
3094  // If we couldn't create one, make sure we don't leave one lying around
3095  if (PHIE) {
3096  Symbolized = PHIE;
3097  } else if (auto *Op = RealToTemp.lookup(I)) {
3098  removePhiOfOps(I, Op);
3099  }
3100  }
3101  } else {
3102  // Mark the instruction as unused so we don't value number it again.
3103  InstrDFS[I] = 0;
3104  }
3105  // If we couldn't come up with a symbolic expression, use the unknown
3106  // expression
3107  if (Symbolized == nullptr)
3108  Symbolized = createUnknownExpression(I);
3109  performCongruenceFinding(I, Symbolized);
3110  } else {
3111  // Handle terminators that return values. All of them produce values we
3112  // don't currently understand. We don't place non-value producing
3113  // terminators in a class.
3114  if (!I->getType()->isVoidTy()) {
3115  auto *Symbolized = createUnknownExpression(I);
3116  performCongruenceFinding(I, Symbolized);
3117  }
3118  processOutgoingEdges(I, I->getParent());
3119  }
3120 }
3121 
3122 // Check if there is a path, using single or equal argument phi nodes, from
3123 // First to Second.
3124 bool NewGVN::singleReachablePHIPath(
3126  const MemoryAccess *Second) const {
3127  if (First == Second)
3128  return true;
3129  if (MSSA->isLiveOnEntryDef(First))
3130  return false;
3131 
3132  // This is not perfect, but as we're just verifying here, we can live with
3133  // the loss of precision. The real solution would be that of doing strongly
3134  // connected component finding in this routine, and it's probably not worth
3135  // the complexity for the time being. So, we just keep a set of visited
3136  // MemoryAccess and return true when we hit a cycle.
3137  if (!Visited.insert(First).second)
3138  return true;
3139 
3140  const auto *EndDef = First;
3141  for (const auto *ChainDef : optimized_def_chain(First)) {
3142  if (ChainDef == Second)
3143  return true;
3144  if (MSSA->isLiveOnEntryDef(ChainDef))
3145  return false;
3146  EndDef = ChainDef;
3147  }
3148  auto *MP = cast<MemoryPhi>(EndDef);
3149  auto ReachableOperandPred = [&](const Use &U) {
3150  return ReachableEdges.count({MP->getIncomingBlock(U), MP->getBlock()});
3151  };
3152  auto FilteredPhiArgs =
3153  make_filter_range(MP->operands(), ReachableOperandPred);
3154  SmallVector<const Value *, 32> OperandList;
3155  llvm::copy(FilteredPhiArgs, std::back_inserter(OperandList));
3156  bool Okay = all_equal(OperandList);
3157  if (Okay)
3158  return singleReachablePHIPath(Visited, cast<MemoryAccess>(OperandList[0]),
3159  Second);
3160  return false;
3161 }
3162 
3163 // Verify the that the memory equivalence table makes sense relative to the
3164 // congruence classes. Note that this checking is not perfect, and is currently
3165 // subject to very rare false negatives. It is only useful for
3166 // testing/debugging.
3167 void NewGVN::verifyMemoryCongruency() const {
3168 #ifndef NDEBUG
3169  // Verify that the memory table equivalence and memory member set match
3170  for (const auto *CC : CongruenceClasses) {
3171  if (CC == TOPClass || CC->isDead())
3172  continue;
3173  if (CC->getStoreCount() != 0) {
3174  assert((CC->getStoredValue() || !isa<StoreInst>(CC->getLeader())) &&
3175  "Any class with a store as a leader should have a "
3176  "representative stored value");
3177  assert(CC->getMemoryLeader() &&
3178  "Any congruence class with a store should have a "
3179  "representative access");
3180  }
3181 
3182  if (CC->getMemoryLeader())
3183  assert(MemoryAccessToClass.lookup(CC->getMemoryLeader()) == CC &&
3184  "Representative MemoryAccess does not appear to be reverse "
3185  "mapped properly");
3186  for (const auto *M : CC->memory())
3187  assert(MemoryAccessToClass.lookup(M) == CC &&
3188  "Memory member does not appear to be reverse mapped properly");
3189  }
3190 
3191  // Anything equivalent in the MemoryAccess table should be in the same
3192  // congruence class.
3193 
3194  // Filter out the unreachable and trivially dead entries, because they may
3195  // never have been updated if the instructions were not processed.
3196  auto ReachableAccessPred =
3197  [&](const std::pair<const MemoryAccess *, CongruenceClass *> Pair) {
3198  bool Result = ReachableBlocks.count(Pair.first->getBlock());
3199  if (!Result || MSSA->isLiveOnEntryDef(Pair.first) ||
3200  MemoryToDFSNum(Pair.first) == 0)
3201  return false;
3202  if (auto *MemDef = dyn_cast<MemoryDef>(Pair.first))
3203  return !isInstructionTriviallyDead(MemDef->getMemoryInst());
3204 
3205  // We could have phi nodes which operands are all trivially dead,
3206  // so we don't process them.
3207  if (auto *MemPHI = dyn_cast<MemoryPhi>(Pair.first)) {
3208  for (const auto &U : MemPHI->incoming_values()) {
3209  if (auto *I = dyn_cast<Instruction>(&*U)) {
3211  return true;
3212  }
3213  }
3214  return false;
3215  }
3216 
3217  return true;
3218  };
3219 
3220  auto Filtered = make_filter_range(MemoryAccessToClass, ReachableAccessPred);
3221  for (auto KV : Filtered) {
3222  if (auto *FirstMUD = dyn_cast<MemoryUseOrDef>(KV.first)) {
3223  auto *SecondMUD = dyn_cast<MemoryUseOrDef>(KV.second->getMemoryLeader());
3224  if (FirstMUD && SecondMUD) {
3226  assert((singleReachablePHIPath(VisitedMAS, FirstMUD, SecondMUD) ||
3227  ValueToClass.lookup(FirstMUD->getMemoryInst()) ==
3228  ValueToClass.lookup(SecondMUD->getMemoryInst())) &&
3229  "The instructions for these memory operations should have "
3230  "been in the same congruence class or reachable through"
3231  "a single argument phi");
3232  }
3233  } else if (auto *FirstMP = dyn_cast<MemoryPhi>(KV.first)) {
3234  // We can only sanely verify that MemoryDefs in the operand list all have
3235  // the same class.
3236  auto ReachableOperandPred = [&](const Use &U) {
3237  return ReachableEdges.count(
3238  {FirstMP->getIncomingBlock(U), FirstMP->getBlock()}) &&
3239  isa<MemoryDef>(U);
3240 
3241  };
3242  // All arguments should in the same class, ignoring unreachable arguments
3243  auto FilteredPhiArgs =
3244  make_filter_range(FirstMP->operands(), ReachableOperandPred);
3246  std::transform(FilteredPhiArgs.begin(), FilteredPhiArgs.end(),
3247  std::back_inserter(PhiOpClasses), [&](const Use &U) {
3248  const MemoryDef *MD = cast<MemoryDef>(U);
3249  return ValueToClass.lookup(MD->getMemoryInst());
3250  });
3251  assert(all_equal(PhiOpClasses) &&
3252  "All MemoryPhi arguments should be in the same class");
3253  }
3254  }
3255 #endif
3256 }
3257 
3258 // Verify that the sparse propagation we did actually found the maximal fixpoint
3259 // We do this by storing the value to class mapping, touching all instructions,
3260 // and redoing the iteration to see if anything changed.
3261 void NewGVN::verifyIterationSettled(Function &F) {
3262 #ifndef NDEBUG
3263  LLVM_DEBUG(dbgs() << "Beginning iteration verification\n");
3264  if (DebugCounter::isCounterSet(VNCounter))
3265  DebugCounter::setCounterValue(VNCounter, StartingVNCounter);
3266 
3267  // Note that we have to store the actual classes, as we may change existing
3268  // classes during iteration. This is because our memory iteration propagation
3269  // is not perfect, and so may waste a little work. But it should generate
3270  // exactly the same congruence classes we have now, with different IDs.
3271  std::map<const Value *, CongruenceClass> BeforeIteration;
3272 
3273  for (auto &KV : ValueToClass) {
3274  if (auto *I = dyn_cast<Instruction>(KV.first))
3275  // Skip unused/dead instructions.
3276  if (InstrToDFSNum(I) == 0)
3277  continue;
3278  BeforeIteration.insert({KV.first, *KV.second});
3279  }
3280 
3281  TouchedInstructions.set();
3282  TouchedInstructions.reset(0);
3283  OpSafeForPHIOfOps.clear();
3284  iterateTouchedInstructions();
3286  EqualClasses;
3287  for (const auto &KV : ValueToClass) {
3288  if (auto *I = dyn_cast<Instruction>(KV.first))
3289  // Skip unused/dead instructions.
3290  if (InstrToDFSNum(I) == 0)
3291  continue;
3292  // We could sink these uses, but i think this adds a bit of clarity here as
3293  // to what we are comparing.
3294  auto *BeforeCC = &BeforeIteration.find(KV.first)->second;
3295  auto *AfterCC = KV.second;
3296  // Note that the classes can't change at this point, so we memoize the set
3297  // that are equal.
3298  if (!EqualClasses.count({BeforeCC, AfterCC})) {
3299  assert(BeforeCC->isEquivalentTo(AfterCC) &&
3300  "Value number changed after main loop completed!");
3301  EqualClasses.insert({BeforeCC, AfterCC});
3302  }
3303  }
3304 #endif
3305 }
3306 
3307 // Verify that for each store expression in the expression to class mapping,
3308 // only the latest appears, and multiple ones do not appear.
3309 // Because loads do not use the stored value when doing equality with stores,
3310 // if we don't erase the old store expressions from the table, a load can find
3311 // a no-longer valid StoreExpression.
3312 void NewGVN::verifyStoreExpressions() const {
3313 #ifndef NDEBUG
3314  // This is the only use of this, and it's not worth defining a complicated
3315  // densemapinfo hash/equality function for it.
3316  std::set<
3317  std::pair<const Value *,
3318  std::tuple<const Value *, const CongruenceClass *, Value *>>>
3319  StoreExpressionSet;
3320  for (const auto &KV : ExpressionToClass) {
3321  if (auto *SE = dyn_cast<StoreExpression>(KV.first)) {
3322  // Make sure a version that will conflict with loads is not already there
3323  auto Res = StoreExpressionSet.insert(
3324  {SE->getOperand(0), std::make_tuple(SE->getMemoryLeader(), KV.second,
3325  SE->getStoredValue())});
3326  bool Okay = Res.second;
3327  // It's okay to have the same expression already in there if it is
3328  // identical in nature.
3329  // This can happen when the leader of the stored value changes over time.
3330  if (!Okay)
3331  Okay = (std::get<1>(Res.first->second) == KV.second) &&
3332  (lookupOperandLeader(std::get<2>(Res.first->second)) ==
3333  lookupOperandLeader(SE->getStoredValue()));
3334  assert(Okay && "Stored expression conflict exists in expression table");
3335  auto *ValueExpr = ValueToExpression.lookup(SE->getStoreInst());
3336  assert(ValueExpr && ValueExpr->equals(*SE) &&
3337  "StoreExpression in ExpressionToClass is not latest "
3338  "StoreExpression for value");
3339  }
3340  }
3341 #endif
3342 }
3343 
3344 // This is the main value numbering loop, it iterates over the initial touched
3345 // instruction set, propagating value numbers, marking things touched, etc,
3346 // until the set of touched instructions is completely empty.
3347 void NewGVN::iterateTouchedInstructions() {
3348  uint64_t Iterations = 0;
3349  // Figure out where touchedinstructions starts
3350  int FirstInstr = TouchedInstructions.find_first();
3351  // Nothing set, nothing to iterate, just return.
3352  if (FirstInstr == -1)
3353  return;
3354  const BasicBlock *LastBlock = getBlockForValue(InstrFromDFSNum(FirstInstr));
3355  while (TouchedInstructions.any()) {
3356  ++Iterations;
3357  // Walk through all the instructions in all the blocks in RPO.
3358  // TODO: As we hit a new block, we should push and pop equalities into a
3359  // table lookupOperandLeader can use, to catch things PredicateInfo
3360  // might miss, like edge-only equivalences.
3361  for (unsigned InstrNum : TouchedInstructions.set_bits()) {
3362 
3363  // This instruction was found to be dead. We don't bother looking
3364  // at it again.
3365  if (InstrNum == 0) {
3366  TouchedInstructions.reset(InstrNum);
3367  continue;
3368  }
3369 
3370  Value *V = InstrFromDFSNum(InstrNum);
3371  const BasicBlock *CurrBlock = getBlockForValue(V);
3372 
3373  // If we hit a new block, do reachability processing.
3374  if (CurrBlock != LastBlock) {
3375  LastBlock = CurrBlock;
3376  bool BlockReachable = ReachableBlocks.count(CurrBlock);
3377  const auto &CurrInstRange = BlockInstRange.lookup(CurrBlock);
3378 
3379  // If it's not reachable, erase any touched instructions and move on.
3380  if (!BlockReachable) {
3381  TouchedInstructions.reset(CurrInstRange.first, CurrInstRange.second);
3382  LLVM_DEBUG(dbgs() << "Skipping instructions in block "
3383  << getBlockName(CurrBlock)
3384  << " because it is unreachable\n");
3385  continue;
3386  }
3387  updateProcessedCount(CurrBlock);
3388  }
3389  // Reset after processing (because we may mark ourselves as touched when
3390  // we propagate equalities).
3391  TouchedInstructions.reset(InstrNum);
3392 
3393  if (auto *MP = dyn_cast<MemoryPhi>(V)) {
3394  LLVM_DEBUG(dbgs() << "Processing MemoryPhi " << *MP << "\n");
3395  valueNumberMemoryPhi(MP);
3396  } else if (auto *I = dyn_cast<Instruction>(V)) {
3397  valueNumberInstruction(I);
3398  } else {
3399  llvm_unreachable("Should have been a MemoryPhi or Instruction");
3400  }
3401  updateProcessedCount(V);
3402  }
3403  }
3404  NumGVNMaxIterations = std::max(NumGVNMaxIterations.getValue(), Iterations);
3405 }
3406 
3407 // This is the main transformation entry point.
3408 bool NewGVN::runGVN() {
3409  if (DebugCounter::isCounterSet(VNCounter))
3410  StartingVNCounter = DebugCounter::getCounterValue(VNCounter);
3411  bool Changed = false;
3412  NumFuncArgs = F.arg_size();
3413  MSSAWalker = MSSA->getWalker();
3414  SingletonDeadExpression = new (ExpressionAllocator) DeadExpression();
3415 
3416  // Count number of instructions for sizing of hash tables, and come
3417  // up with a global dfs numbering for instructions.
3418  unsigned ICount = 1;
3419  // Add an empty instruction to account for the fact that we start at 1
3420  DFSToInstr.emplace_back(nullptr);
3421  // Note: We want ideal RPO traversal of the blocks, which is not quite the
3422  // same as dominator tree order, particularly with regard whether backedges
3423  // get visited first or second, given a block with multiple successors.
3424  // If we visit in the wrong order, we will end up performing N times as many
3425  // iterations.
3426  // The dominator tree does guarantee that, for a given dom tree node, it's
3427  // parent must occur before it in the RPO ordering. Thus, we only need to sort
3428  // the siblings.
3430  unsigned Counter = 0;
3431  for (auto &B : RPOT) {
3432  auto *Node = DT->getNode(B);
3433  assert(Node && "RPO and Dominator tree should have same reachability");
3434  RPOOrdering[Node] = ++Counter;
3435  }
3436  // Sort dominator tree children arrays into RPO.
3437  for (auto &B : RPOT) {
3438  auto *Node = DT->getNode(B);
3439  if (Node->getNumChildren() > 1)
3440  llvm::sort(*Node, [&](const DomTreeNode *A, const DomTreeNode *B) {
3441  return RPOOrdering[A] < RPOOrdering[B];
3442  });
3443  }
3444 
3445  // Now a standard depth first ordering of the domtree is equivalent to RPO.
3446  for (auto *DTN : depth_first(DT->getRootNode())) {
3447  BasicBlock *B = DTN->getBlock();
3448  const auto &BlockRange = assignDFSNumbers(B, ICount);
3449  BlockInstRange.insert({B, BlockRange});
3450  ICount += BlockRange.second - BlockRange.first;
3451  }
3452  initializeCongruenceClasses(F);
3453 
3454  TouchedInstructions.resize(ICount);
3455  // Ensure we don't end up resizing the expressionToClass map, as
3456  // that can be quite expensive. At most, we have one expression per
3457  // instruction.
3458  ExpressionToClass.reserve(ICount);
3459 
3460  // Initialize the touched instructions to include the entry block.
3461  const auto &InstRange = BlockInstRange.lookup(&F.getEntryBlock());
3462  TouchedInstructions.set(InstRange.first, InstRange.second);
3463  LLVM_DEBUG(dbgs() << "Block " << getBlockName(&F.getEntryBlock())
3464  << " marked reachable\n");
3465  ReachableBlocks.insert(&F.getEntryBlock());
3466 
3467  iterateTouchedInstructions();
3468  verifyMemoryCongruency();
3469  verifyIterationSettled(F);
3470  verifyStoreExpressions();
3471 
3472  Changed |= eliminateInstructions(F);
3473 
3474  // Delete all instructions marked for deletion.
3475  for (Instruction *ToErase : InstructionsToErase) {
3476  if (!ToErase->use_empty())
3477  ToErase->replaceAllUsesWith(PoisonValue::get(ToErase->getType()));
3478 
3479  assert(ToErase->getParent() &&
3480  "BB containing ToErase deleted unexpectedly!");
3481  ToErase->eraseFromParent();
3482  }
3483  Changed |= !InstructionsToErase.empty();
3484 
3485  // Delete all unreachable blocks.
3486  auto UnreachableBlockPred = [&](const BasicBlock &BB) {
3487  return !ReachableBlocks.count(&BB);
3488  };
3489 
3490  for (auto &BB : make_filter_range(F, UnreachableBlockPred)) {
3491  LLVM_DEBUG(dbgs() << "We believe block " << getBlockName(&BB)
3492  << " is unreachable\n");
3493  deleteInstructionsInBlock(&BB);
3494  Changed = true;
3495  }
3496 
3497  cleanupTables();
3498  return Changed;
3499 }
3500 
3502  int DFSIn = 0;
3503  int DFSOut = 0;
3504  int LocalNum = 0;
3505 
3506  // Only one of Def and U will be set.
3507  // The bool in the Def tells us whether the Def is the stored value of a
3508  // store.
3510  Use *U = nullptr;
3511 
3512  bool operator<(const ValueDFS &Other) const {
3513  // It's not enough that any given field be less than - we have sets
3514  // of fields that need to be evaluated together to give a proper ordering.
3515  // For example, if you have;
3516  // DFS (1, 3)
3517  // Val 0
3518  // DFS (1, 2)
3519  // Val 50
3520  // We want the second to be less than the first, but if we just go field
3521  // by field, we will get to Val 0 < Val 50 and say the first is less than
3522  // the second. We only want it to be less than if the DFS orders are equal.
3523  //
3524  // Each LLVM instruction only produces one value, and thus the lowest-level
3525  // differentiator that really matters for the stack (and what we use as as a
3526  // replacement) is the local dfs number.
3527  // Everything else in the structure is instruction level, and only affects
3528  // the order in which we will replace operands of a given instruction.
3529  //
3530  // For a given instruction (IE things with equal dfsin, dfsout, localnum),
3531  // the order of replacement of uses does not matter.
3532  // IE given,
3533  // a = 5
3534  // b = a + a
3535  // When you hit b, you will have two valuedfs with the same dfsin, out, and
3536  // localnum.
3537  // The .val will be the same as well.
3538  // The .u's will be different.
3539  // You will replace both, and it does not matter what order you replace them
3540  // in (IE whether you replace operand 2, then operand 1, or operand 1, then
3541  // operand 2).
3542  // Similarly for the case of same dfsin, dfsout, localnum, but different
3543  // .val's
3544  // a = 5
3545  // b = 6
3546  // c = a + b
3547  // in c, we will a valuedfs for a, and one for b,with everything the same
3548  // but .val and .u.
3549  // It does not matter what order we replace these operands in.
3550  // You will always end up with the same IR, and this is guaranteed.
3551  return std::tie(DFSIn, DFSOut, LocalNum, Def, U) <
3552  std::tie(Other.DFSIn, Other.DFSOut, Other.LocalNum, Other.Def,
3553  Other.U);
3554  }
3555 };
3556 
3557 // This function converts the set of members for a congruence class from values,
3558 // to sets of defs and uses with associated DFS info. The total number of
3559 // reachable uses for each value is stored in UseCount, and instructions that
3560 // seem
3561 // dead (have no non-dead uses) are stored in ProbablyDead.
3562 void NewGVN::convertClassToDFSOrdered(
3563  const CongruenceClass &Dense, SmallVectorImpl<ValueDFS> &DFSOrderedSet,
3565  SmallPtrSetImpl<Instruction *> &ProbablyDead) const {
3566  for (auto *D : Dense) {
3567  // First add the value.
3568  BasicBlock *BB = getBlockForValue(D);
3569  // Constants are handled prior to ever calling this function, so
3570  // we should only be left with instructions as members.
3571  assert(BB && "Should have figured out a basic block for value");
3572  ValueDFS VDDef;
3573  DomTreeNode *DomNode = DT->getNode(BB);
3574  VDDef.DFSIn = DomNode->getDFSNumIn();
3575  VDDef.DFSOut = DomNode->getDFSNumOut();
3576  // If it's a store, use the leader of the value operand, if it's always
3577  // available, or the value operand. TODO: We could do dominance checks to
3578  // find a dominating leader, but not worth it ATM.
3579  if (auto *SI = dyn_cast<StoreInst>(D)) {
3580  auto Leader = lookupOperandLeader(SI->getValueOperand());
3581  if (alwaysAvailable(Leader)) {
3582  VDDef.Def.setPointer(Leader);
3583  } else {
3584  VDDef.Def.setPointer(SI->getValueOperand());
3585  VDDef.Def.setInt(true);
3586  }
3587  } else {
3588  VDDef.Def.setPointer(D);
3589  }
3590  assert(isa<Instruction>(D) &&
3591  "The dense set member should always be an instruction");
3592  Instruction *Def = cast<Instruction>(D);
3593  VDDef.LocalNum = InstrToDFSNum(D);
3594  DFSOrderedSet.push_back(VDDef);
3595  // If there is a phi node equivalent, add it
3596  if (auto *PN = RealToTemp.lookup(Def)) {
3597  auto *PHIE =
3598  dyn_cast_or_null<PHIExpression>(ValueToExpression.lookup(Def));
3599  if (PHIE) {
3600  VDDef.Def.setInt(false);
3601  VDDef.Def.setPointer(PN);
3602  VDDef.LocalNum = 0;
3603  DFSOrderedSet.push_back(VDDef);
3604  }
3605  }
3606 
3607  unsigned int UseCount = 0;
3608  // Now add the uses.
3609  for (auto &U : Def->uses()) {
3610  if (auto *I = dyn_cast<Instruction>(U.getUser())) {
3611  // Don't try to replace into dead uses
3612  if (InstructionsToErase.count(I))
3613  continue;
3614  ValueDFS VDUse;
3615  // Put the phi node uses in the incoming block.
3616  BasicBlock *IBlock;
3617  if (auto *P = dyn_cast<PHINode>(I)) {
3618  IBlock = P->getIncomingBlock(U);
3619  // Make phi node users appear last in the incoming block
3620  // they are from.
3621  VDUse.LocalNum = InstrDFS.size() + 1;
3622  } else {
3623  IBlock = getBlockForValue(I);
3624  VDUse.LocalNum = InstrToDFSNum(I);
3625  }
3626 
3627  // Skip uses in unreachable blocks, as we're going
3628  // to delete them.
3629  if (!ReachableBlocks.contains(IBlock))
3630  continue;
3631 
3632  DomTreeNode *DomNode = DT->getNode(IBlock);
3633  VDUse.DFSIn = DomNode->getDFSNumIn();
3634  VDUse.DFSOut = DomNode->getDFSNumOut();
3635  VDUse.U = &U;
3636  ++UseCount;
3637  DFSOrderedSet.emplace_back(VDUse);
3638  }
3639  }
3640 
3641  // If there are no uses, it's probably dead (but it may have side-effects,
3642  // so not definitely dead. Otherwise, store the number of uses so we can
3643  // track if it becomes dead later).
3644  if (UseCount == 0)
3645  ProbablyDead.insert(Def);
3646  else
3647  UseCounts[Def] = UseCount;
3648  }
3649 }
3650 
3651 // This function converts the set of members for a congruence class from values,
3652 // to the set of defs for loads and stores, with associated DFS info.
3653 void NewGVN::convertClassToLoadsAndStores(
3654  const CongruenceClass &Dense,
3655  SmallVectorImpl<ValueDFS> &LoadsAndStores) const {
3656  for (auto *D : Dense) {
3657  if (!isa<LoadInst>(D) && !isa<StoreInst>(D))
3658  continue;
3659 
3660  BasicBlock *BB = getBlockForValue(D);
3661  ValueDFS VD;
3662  DomTreeNode *DomNode = DT->getNode(BB);
3663  VD.DFSIn = DomNode->getDFSNumIn();
3664  VD.DFSOut = DomNode->getDFSNumOut();
3665  VD.Def.setPointer(D);
3666 
3667  // If it's an instruction, use the real local dfs number.
3668  if (auto *I = dyn_cast<Instruction>(D))
3669  VD.LocalNum = InstrToDFSNum(I);
3670  else
3671  llvm_unreachable("Should have been an instruction");
3672 
3673  LoadsAndStores.emplace_back(VD);
3674  }
3675 }
3676 
3679  I->replaceAllUsesWith(Repl);
3680 }
3681 
3682 void NewGVN::deleteInstructionsInBlock(BasicBlock *BB) {
3683  LLVM_DEBUG(dbgs() << " BasicBlock Dead:" << *BB);
3684  ++NumGVNBlocksDeleted;
3685 
3686  // Delete the instructions backwards, as it has a reduced likelihood of having
3687  // to update as many def-use and use-def chains. Start after the terminator.
3688  auto StartPoint = BB->rbegin();
3689  ++StartPoint;
3690  // Note that we explicitly recalculate BB->rend() on each iteration,
3691  // as it may change when we remove the first instruction.
3692  for (BasicBlock::reverse_iterator I(StartPoint); I != BB->rend();) {
3693  Instruction &Inst = *I++;
3694  if (!Inst.use_empty())
3696  if (isa<LandingPadInst>(Inst))
3697  continue;
3698  salvageKnowledge(&Inst, AC);
3699 
3700  Inst.eraseFromParent();
3701  ++NumGVNInstrDeleted;
3702  }
3703  // Now insert something that simplifycfg will turn into an unreachable.
3704  Type *Int8Ty = Type::getInt8Ty(BB->getContext());
3705  new StoreInst(PoisonValue::get(Int8Ty),
3707  BB->getTerminator());
3708 }
3709 
3710 void NewGVN::markInstructionForDeletion(Instruction *I) {
3711  LLVM_DEBUG(dbgs() << "Marking " << *I << " for deletion\n");
3712  InstructionsToErase.insert(I);
3713 }
3714 
3715 void NewGVN::replaceInstruction(Instruction *I, Value *V) {
3716  LLVM_DEBUG(dbgs() << "Replacing " << *I << " with " << *V << "\n");
3718  // We save the actual erasing to avoid invalidating memory
3719  // dependencies until we are done with everything.
3720  markInstructionForDeletion(I);
3721 }
3722 
3723 namespace {
3724 
3725 // This is a stack that contains both the value and dfs info of where
3726 // that value is valid.
3727 class ValueDFSStack {
3728 public:
3729  Value *back() const { return ValueStack.back(); }
3730  std::pair<int, int> dfs_back() const { return DFSStack.back(); }
3731 
3732  void push_back(Value *V, int DFSIn, int DFSOut) {
3733  ValueStack.emplace_back(V);
3734  DFSStack.emplace_back(DFSIn, DFSOut);
3735  }
3736 
3737  bool empty() const { return DFSStack.empty(); }
3738 
3739  bool isInScope(int DFSIn, int DFSOut) const {
3740  if (empty())
3741  return false;
3742  return DFSIn >= DFSStack.back().first && DFSOut <= DFSStack.back().second;
3743  }
3744 
3745  void popUntilDFSScope(int DFSIn, int DFSOut) {
3746 
3747  // These two should always be in sync at this point.
3748  assert(ValueStack.size() == DFSStack.size() &&
3749  "Mismatch between ValueStack and DFSStack");
3750  while (
3751  !DFSStack.empty() &&
3752  !(DFSIn >= DFSStack.back().first && DFSOut <= DFSStack.back().second)) {
3753  DFSStack.pop_back();
3754  ValueStack.pop_back();
3755  }
3756  }
3757 
3758 private:
3759  SmallVector<Value *, 8> ValueStack;
3760  SmallVector<std::pair<int, int>, 8> DFSStack;
3761 };
3762 
3763 } // end anonymous namespace
3764 
3765 // Given an expression, get the congruence class for it.
3766 CongruenceClass *NewGVN::getClassForExpression(const Expression *E) const {
3767  if (auto *VE = dyn_cast<VariableExpression>(E))
3768  return ValueToClass.lookup(VE->getVariableValue());
3769  else if (isa<DeadExpression>(E))
3770  return TOPClass;
3771  return ExpressionToClass.lookup(E);
3772 }
3773 
3774 // Given a value and a basic block we are trying to see if it is available in,
3775 // see if the value has a leader available in that block.
3776 Value *NewGVN::findPHIOfOpsLeader(const Expression *E,
3777  const Instruction *OrigInst,
3778  const BasicBlock *BB) const {
3779  // It would already be constant if we could make it constant
3780  if (auto *CE = dyn_cast<ConstantExpression>(E))
3781  return CE->getConstantValue();
3782  if (auto *VE = dyn_cast<VariableExpression>(E)) {
3783  auto *V = VE->getVariableValue();
3784  if (alwaysAvailable(V) || DT->dominates(getBlockForValue(V), BB))
3785  return VE->getVariableValue();
3786  }
3787 
3788  auto *CC = getClassForExpression(E);
3789  if (!CC)
3790  return nullptr;
3791  if (alwaysAvailable(CC->getLeader()))
3792  return CC->getLeader();
3793 
3794  for (auto *Member : *CC) {
3795  auto *MemberInst = dyn_cast<Instruction>(Member);
3796  if (MemberInst == OrigInst)
3797  continue;
3798  // Anything that isn't an instruction is always available.
3799  if (!MemberInst)
3800  return Member;
3801  if (DT->dominates(getBlockForValue(MemberInst), BB))
3802  return Member;
3803  }
3804  return nullptr;
3805 }
3806 
3807 bool NewGVN::eliminateInstructions(Function &F) {
3808  // This is a non-standard eliminator. The normal way to eliminate is
3809  // to walk the dominator tree in order, keeping track of available
3810  // values, and eliminating them. However, this is mildly
3811  // pointless. It requires doing lookups on every instruction,
3812  // regardless of whether we will ever eliminate it. For
3813  // instructions part of most singleton congruence classes, we know we
3814  // will never eliminate them.
3815 
3816  // Instead, this eliminator looks at the congruence classes directly, sorts
3817  // them into a DFS ordering of the dominator tree, and then we just
3818  // perform elimination straight on the sets by walking the congruence
3819  // class member uses in order, and eliminate the ones dominated by the
3820  // last member. This is worst case O(E log E) where E = number of
3821  // instructions in a single congruence class. In theory, this is all
3822  // instructions. In practice, it is much faster, as most instructions are
3823  // either in singleton congruence classes or can't possibly be eliminated
3824  // anyway (if there are no overlapping DFS ranges in class).
3825  // When we find something not dominated, it becomes the new leader
3826  // for elimination purposes.
3827  // TODO: If we wanted to be faster, We could remove any members with no
3828  // overlapping ranges while sorting, as we will never eliminate anything
3829  // with those members, as they don't dominate anything else in our set.
3830 
3831  bool AnythingReplaced = false;
3832 
3833  // Since we are going to walk the domtree anyway, and we can't guarantee the
3834  // DFS numbers are updated, we compute some ourselves.
3835  DT->updateDFSNumbers();
3836 
3837  // Go through all of our phi nodes, and kill the arguments associated with
3838  // unreachable edges.
3839  auto ReplaceUnreachablePHIArgs = [&](PHINode *PHI, BasicBlock *BB) {
3840  for (auto &Operand : PHI->incoming_values())
3841  if (!ReachableEdges.count({PHI->getIncomingBlock(Operand), BB})) {
3842  LLVM_DEBUG(dbgs() << "Replacing incoming value of " << PHI
3843  << " for block "
3844  << getBlockName(PHI->getIncomingBlock(Operand))
3845  << " with poison due to it being unreachable\n");
3846  Operand.set(PoisonValue::get(PHI->getType()));
3847  }
3848  };
3849  // Replace unreachable phi arguments.
3850  // At this point, RevisitOnReachabilityChange only contains:
3851  //
3852  // 1. PHIs
3853  // 2. Temporaries that will convert to PHIs
3854  // 3. Operations that are affected by an unreachable edge but do not fit into
3855  // 1 or 2 (rare).
3856  // So it is a slight overshoot of what we want. We could make it exact by
3857  // using two SparseBitVectors per block.
3858  DenseMap<const BasicBlock *, unsigned> ReachablePredCount;
3859  for (auto &KV : ReachableEdges)
3860  ReachablePredCount[KV.getEnd()]++;
3861  for (auto &BBPair : RevisitOnReachabilityChange) {
3862  for (auto InstNum : BBPair.second) {
3863  auto *Inst = InstrFromDFSNum(InstNum);
3864  auto *PHI = dyn_cast<PHINode>(Inst);
3865  PHI = PHI ? PHI : dyn_cast_or_null<PHINode>(RealToTemp.lookup(Inst));
3866  if (!PHI)
3867  continue;
3868  auto *BB = BBPair.first;
3869  if (ReachablePredCount.lookup(BB) != PHI->getNumIncomingValues())
3870  ReplaceUnreachablePHIArgs(PHI, BB);
3871  }
3872  }
3873 
3874  // Map to store the use counts
3876  for (auto *CC : reverse(CongruenceClasses)) {
3877  LLVM_DEBUG(dbgs() << "Eliminating in congruence class " << CC->getID()
3878  << "\n");
3879  // Track the equivalent store info so we can decide whether to try
3880  // dead store elimination.
3881  SmallVector<ValueDFS, 8> PossibleDeadStores;
3882  SmallPtrSet<Instruction *, 8> ProbablyDead;
3883  if (CC->isDead() || CC->empty())
3884  continue;
3885  // Everything still in the TOP class is unreachable or dead.
3886  if (CC == TOPClass) {
3887  for (auto *M : *CC) {
3888  auto *VTE = ValueToExpression.lookup(M);
3889  if (VTE && isa<DeadExpression>(VTE))
3890  markInstructionForDeletion(cast<Instruction>(M));
3891  assert((!ReachableBlocks.count(cast<Instruction>(M)->getParent()) ||
3892  InstructionsToErase.count(cast<Instruction>(M))) &&
3893  "Everything in TOP should be unreachable or dead at this "
3894  "point");
3895  }
3896  continue;
3897  }
3898 
3899  assert(CC->getLeader() && "We should have had a leader");
3900  // If this is a leader that is always available, and it's a
3901  // constant or has no equivalences, just replace everything with
3902  // it. We then update the congruence class with whatever members
3903  // are left.
3904  Value *Leader =
3905  CC->getStoredValue() ? CC->getStoredValue() : CC->getLeader();
3906  if (alwaysAvailable(Leader)) {
3907  CongruenceClass::MemberSet MembersLeft;
3908  for (auto *M : *CC) {
3909  Value *Member = M;
3910  // Void things have no uses we can replace.
3911  if (Member == Leader || !isa<Instruction>(Member) ||
3912  Member->getType()->isVoidTy()) {
3913  MembersLeft.insert(Member);
3914  continue;
3915  }
3916  LLVM_DEBUG(dbgs() << "Found replacement " << *(Leader) << " for "
3917  << *Member << "\n");
3918  auto *I = cast<Instruction>(Member);
3919  assert(Leader != I && "About to accidentally remove our leader");
3920  replaceInstruction(I, Leader);
3921  AnythingReplaced = true;
3922  }
3923  CC->swap(MembersLeft);
3924  } else {
3925  // If this is a singleton, we can skip it.
3926  if (CC->size() != 1 || RealToTemp.count(Leader)) {
3927  // This is a stack because equality replacement/etc may place
3928  // constants in the middle of the member list, and we want to use
3929  // those constant values in preference to the current leader, over
3930  // the scope of those constants.
3931  ValueDFSStack EliminationStack;
3932 
3933  // Convert the members to DFS ordered sets and then merge them.
3934  SmallVector<ValueDFS, 8> DFSOrderedSet;
3935  convertClassToDFSOrdered(*CC, DFSOrderedSet, UseCounts, ProbablyDead);
3936 
3937  // Sort the whole thing.
3938  llvm::sort(DFSOrderedSet);
3939  for (auto &VD : DFSOrderedSet) {
3940  int MemberDFSIn = VD.DFSIn;
3941  int MemberDFSOut = VD.DFSOut;
3942  Value *Def = VD.Def.getPointer();
3943  bool FromStore = VD.Def.getInt();
3944  Use *U = VD.U;
3945  // We ignore void things because we can't get a value from them.
3946  if (Def && Def->getType()->isVoidTy())
3947  continue;
3948  auto *DefInst = dyn_cast_or_null<Instruction>(Def);
3949  if (DefInst && AllTempInstructions.count(DefInst)) {
3950  auto *PN = cast<PHINode>(DefInst);
3951 
3952  // If this is a value phi and that's the expression we used, insert
3953  // it into the program
3954  // remove from temp instruction list.
3955  AllTempInstructions.erase(PN);
3956  auto *DefBlock = getBlockForValue(Def);
3957  LLVM_DEBUG(dbgs() << "Inserting fully real phi of ops" << *Def
3958  << " into block "
3959  << getBlockName(getBlockForValue(Def)) << "\n");
3960  PN->insertBefore(&DefBlock->front());
3961  Def = PN;
3962  NumGVNPHIOfOpsEliminations++;
3963  }
3964 
3965  if (EliminationStack.empty()) {
3966  LLVM_DEBUG(dbgs() << "Elimination Stack is empty\n");
3967  } else {
3968  LLVM_DEBUG(dbgs() << "Elimination Stack Top DFS numbers are ("
3969  << EliminationStack.dfs_back().first << ","
3970  << EliminationStack.dfs_back().second << ")\n");
3971  }
3972 
3973  LLVM_DEBUG(dbgs() << "Current DFS numbers are (" << MemberDFSIn << ","
3974  << MemberDFSOut << ")\n");
3975  // First, we see if we are out of scope or empty. If so,
3976  // and there equivalences, we try to replace the top of
3977  // stack with equivalences (if it's on the stack, it must
3978  // not have been eliminated yet).
3979  // Then we synchronize to our current scope, by
3980  // popping until we are back within a DFS scope that
3981  // dominates the current member.
3982  // Then, what happens depends on a few factors
3983  // If the stack is now empty, we need to push
3984  // If we have a constant or a local equivalence we want to
3985  // start using, we also push.
3986  // Otherwise, we walk along, processing members who are
3987  // dominated by this scope, and eliminate them.
3988  bool ShouldPush = Def && EliminationStack.empty();
3989  bool OutOfScope =
3990  !EliminationStack.isInScope(MemberDFSIn, MemberDFSOut);
3991 
3992  if (OutOfScope || ShouldPush) {
3993  // Sync to our current scope.
3994  EliminationStack.popUntilDFSScope(MemberDFSIn, MemberDFSOut);
3995  bool ShouldPush = Def && EliminationStack.empty();
3996  if (ShouldPush) {
3997  EliminationStack.push_back(Def, MemberDFSIn, MemberDFSOut);
3998  }
3999  }
4000 
4001  // Skip the Def's, we only want to eliminate on their uses. But mark
4002  // dominated defs as dead.
4003  if (Def) {
4004  // For anything in this case, what and how we value number
4005  // guarantees that any side-effets that would have occurred (ie
4006  // throwing, etc) can be proven to either still occur (because it's
4007  // dominated by something that has the same side-effects), or never
4008  // occur. Otherwise, we would not have been able to prove it value
4009  // equivalent to something else. For these things, we can just mark
4010  // it all dead. Note that this is different from the "ProbablyDead"
4011  // set, which may not be dominated by anything, and thus, are only
4012  // easy to prove dead if they are also side-effect free. Note that
4013  // because stores are put in terms of the stored value, we skip
4014  // stored values here. If the stored value is really dead, it will
4015  // still be marked for deletion when we process it in its own class.
4016  if (!EliminationStack.empty() && Def != EliminationStack.back() &&
4017  isa<Instruction>(Def) && !FromStore)
4018  markInstructionForDeletion(cast<Instruction>(Def));
4019  continue;
4020  }
4021  // At this point, we know it is a Use we are trying to possibly
4022  // replace.
4023 
4024  assert(isa<Instruction>(U->get()) &&
4025  "Current def should have been an instruction");
4026  assert(isa<Instruction>(U->getUser()) &&
4027  "Current user should have been an instruction");
4028 
4029  // If the thing we are replacing into is already marked to be dead,
4030  // this use is dead. Note that this is true regardless of whether
4031  // we have anything dominating the use or not. We do this here
4032  // because we are already walking all the uses anyway.
4033  Instruction *InstUse = cast<Instruction>(U->getUser());
4034  if (InstructionsToErase.count(InstUse)) {
4035  auto &UseCount = UseCounts[U->get()];
4036  if (--UseCount == 0) {
4037  ProbablyDead.insert(cast<Instruction>(U->get()));
4038  }
4039  }
4040 
4041  // If we get to this point, and the stack is empty we must have a use
4042  // with nothing we can use to eliminate this use, so just skip it.
4043  if (EliminationStack.empty())
4044  continue;
4045 
4046  Value *DominatingLeader = EliminationStack.back();
4047 
4048  auto *II = dyn_cast<IntrinsicInst>(DominatingLeader);
4049  bool isSSACopy = II && II->getIntrinsicID() == Intrinsic::ssa_copy;
4050  if (isSSACopy)
4051  DominatingLeader = II->getOperand(0);
4052 
4053  // Don't replace our existing users with ourselves.
4054  if (U->get() == DominatingLeader)
4055  continue;
4056  LLVM_DEBUG(dbgs()
4057  << "Found replacement " << *DominatingLeader << " for "
4058  << *U->get() << " in " << *(U->getUser()) << "\n");
4059 
4060  // If we replaced something in an instruction, handle the patching of
4061  // metadata. Skip this if we are replacing predicateinfo with its
4062  // original operand, as we already know we can just drop it.
4063  auto *ReplacedInst = cast<Instruction>(U->get());
4064  auto *PI = PredInfo->getPredicateInfoFor(ReplacedInst);
4065  if (!PI || DominatingLeader != PI->OriginalOp)
4066  patchReplacementInstruction(ReplacedInst, DominatingLeader);
4067  U->set(DominatingLeader);
4068  // This is now a use of the dominating leader, which means if the
4069  // dominating leader was dead, it's now live!
4070  auto &LeaderUseCount = UseCounts[DominatingLeader];
4071  // It's about to be alive again.
4072  if (LeaderUseCount == 0 && isa<Instruction>(DominatingLeader))
4073  ProbablyDead.erase(cast<Instruction>(DominatingLeader));
4074  // For copy instructions, we use their operand as a leader,
4075  // which means we remove a user of the copy and it may become dead.
4076  if (isSSACopy) {
4077  unsigned &IIUseCount = UseCounts[II];
4078  if (--IIUseCount == 0)
4079  ProbablyDead.insert(II);
4080  }
4081  ++LeaderUseCount;
4082  AnythingReplaced = true;
4083  }
4084  }
4085  }
4086 
4087  // At this point, anything still in the ProbablyDead set is actually dead if
4088  // would be trivially dead.
4089  for (auto *I : ProbablyDead)
4091  markInstructionForDeletion(I);
4092 
4093  // Cleanup the congruence class.
4094  CongruenceClass::MemberSet MembersLeft;
4095  for (auto *Member : *CC)
4096  if (!isa<Instruction>(Member) ||
4097  !InstructionsToErase.count(cast<Instruction>(Member)))
4098  MembersLeft.insert(Member);
4099  CC->swap(MembersLeft);
4100 
4101  // If we have possible dead stores to look at, try to eliminate them.
4102  if (CC->getStoreCount() > 0) {
4103  convertClassToLoadsAndStores(*CC, PossibleDeadStores);
4104  llvm::sort(PossibleDeadStores);
4105  ValueDFSStack EliminationStack;
4106  for (auto &VD : PossibleDeadStores) {
4107  int MemberDFSIn = VD.DFSIn;
4108  int MemberDFSOut = VD.DFSOut;
4109  Instruction *Member = cast<Instruction>(VD.Def.getPointer());
4110  if (EliminationStack.empty() ||
4111  !EliminationStack.isInScope(MemberDFSIn, MemberDFSOut)) {
4112  // Sync to our current scope.
4113  EliminationStack.popUntilDFSScope(MemberDFSIn, MemberDFSOut);
4114  if (EliminationStack.empty()) {
4115  EliminationStack.push_back(Member, MemberDFSIn, MemberDFSOut);
4116  continue;
4117  }
4118  }
4119  // We already did load elimination, so nothing to do here.
4120  if (isa<LoadInst>(Member))
4121  continue;
4122  assert(!EliminationStack.empty());
4123  Instruction *Leader = cast<Instruction>(EliminationStack.back());
4124  (void)Leader;
4125  assert(DT->dominates(Leader->getParent(), Member->getParent()));
4126  // Member is dominater by Leader, and thus dead
4127  LLVM_DEBUG(dbgs() << "Marking dead store " << *Member
4128  << " that is dominated by " << *Leader << "\n");
4129  markInstructionForDeletion(Member);
4130  CC->erase(Member);
4131  ++NumGVNDeadStores;
4132  }
4133  }
4134  }
4135  return AnythingReplaced;
4136 }
4137 
4138 // This function provides global ranking of operations so that we can place them
4139 // in a canonical order. Note that rank alone is not necessarily enough for a
4140 // complete ordering, as constants all have the same rank. However, generally,
4141 // we will simplify an operation with all constants so that it doesn't matter
4142 // what order they appear in.
4143 unsigned int NewGVN::getRank(const Value *V) const {
4144  // Prefer constants to undef to anything else
4145  // Undef is a constant, have to check it first.
4146  // Prefer poison to undef as it's less defined.
4147  // Prefer smaller constants to constantexprs
4148  // Note that the order here matters because of class inheritance
4149  if (isa<ConstantExpr>(V))
4150  return 3;
4151  if (isa<PoisonValue>(V))
4152  return 1;
4153  if (isa<UndefValue>(V))
4154  return 2;
4155  if (isa<Constant>(V))
4156  return 0;
4157  if (auto *A = dyn_cast<Argument>(V))
4158  return 4 + A->getArgNo();
4159 
4160  // Need to shift the instruction DFS by number of arguments + 5 to account for
4161  // the constant and argument ranking above.
4162  unsigned Result = InstrToDFSNum(V);
4163  if (Result > 0)
4164  return 5 + NumFuncArgs + Result;
4165  // Unreachable or something else, just return a really large number.
4166  return ~0;
4167 }
4168 
4169 // This is a function that says whether two commutative operations should
4170 // have their order swapped when canonicalizing.
4171 bool NewGVN::shouldSwapOperands(const Value *A, const Value *B) const {
4172  // Because we only care about a total ordering, and don't rewrite expressions
4173  // in this order, we order by rank, which will give a strict weak ordering to
4174  // everything but constants, and then we order by pointer address.
4175  return std::make_pair(getRank(A), A) > std::make_pair(getRank(B), B);
4176 }
4177 
4178 bool NewGVN::shouldSwapOperandsForIntrinsic(const Value *A, const Value *B,
4179  const IntrinsicInst *I) const {
4180  auto LookupResult = IntrinsicInstPred.find(I);
4181  if (shouldSwapOperands(A, B)) {
4182  if (LookupResult == IntrinsicInstPred.end())
4183  IntrinsicInstPred.insert({I, B});
4184  else
4185  LookupResult->second = B;
4186  return true;
4187  }
4188 
4189  if (LookupResult != IntrinsicInstPred.end()) {
4190  auto *SeenPredicate = LookupResult->second;
4191  if (SeenPredicate) {
4192  if (SeenPredicate == B)
4193  return true;
4194  else
4195  LookupResult->second = nullptr;
4196  }
4197  }
4198  return false;
4199 }
4200 
4201 namespace {
4202 
4203 class NewGVNLegacyPass : public FunctionPass {
4204 public:
4205  // Pass identification, replacement for typeid.
4206  static char ID;
4207 
4208  NewGVNLegacyPass() : FunctionPass(ID) {
4210  }
4211 
4212  bool runOnFunction(Function &F) override;
4213 
4214 private:
4215  void getAnalysisUsage(AnalysisUsage &AU) const override {
4223  }
4224 };
4225 
4226 } // end anonymous namespace
4227 
4229  if (skipFunction(F))
4230  return false;
4231  return NewGVN(F, &getAnalysis<DominatorTreeWrapperPass>().getDomTree(),
4232  &getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F),
4233  &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(F),
4234  &getAnalysis<AAResultsWrapperPass>().getAAResults(),
4235  &getAnalysis<MemorySSAWrapperPass>().getMSSA(),
4236  F.getParent()->getDataLayout())
4237  .runGVN();
4238 }
4239 
4240 char NewGVNLegacyPass::ID = 0;
4241 
4242 INITIALIZE_PASS_BEGIN(NewGVNLegacyPass, "newgvn", "Global Value Numbering",
4243  false, false)
4250 INITIALIZE_PASS_END(NewGVNLegacyPass, "newgvn", "Global Value Numbering", false,
4251  false)
4252 
4253 // createGVNPass - The public interface to this file.
4254 FunctionPass *llvm::createNewGVNPass() { return new NewGVNLegacyPass(); }
4255 
4257  // Apparently the order in which we get these results matter for
4258  // the old GVN (see Chandler's comment in GVN.cpp). I'll keep
4259  // the same order here, just in case.
4260  auto &AC = AM.getResult<AssumptionAnalysis>(F);
4261  auto &DT = AM.getResult<DominatorTreeAnalysis>(F);
4262  auto &TLI = AM.getResult<TargetLibraryAnalysis>(F);
4263  auto &AA = AM.getResult<AAManager>(F);
4264  auto &MSSA = AM.getResult<MemorySSAAnalysis>(F).getMSSA();
4265  bool Changed =
4266  NewGVN(F, &DT, &AC, &TLI, &AA, &MSSA, F.getParent()->getDataLayout())
4267  .runGVN();
4268  if (!Changed)
4269  return PreservedAnalyses::all();
4270  PreservedAnalyses PA;
4272  return PA;
4273 }
i
i
Definition: README.txt:29
llvm::PreservedAnalyses
A set of analyses that are preserved following a run of a transformation pass.
Definition: PassManager.h:152
llvm::VNCoercion::analyzeLoadFromClobberingMemInst
int analyzeLoadFromClobberingMemInst(Type *LoadTy, Value *LoadPtr, MemIntrinsic *DepMI, const DataLayout &DL)
This function determines whether a value for the pointer LoadPtr can be extracted from the memory int...
Definition: VNCoercion.cpp:349
set
We currently generate a but we really shouldn eax ecx xorl edx divl ecx eax divl ecx movl eax ret A similar code sequence works for division We currently compile i32 v2 eax eax jo LBB1_2 atomic and others It is also currently not done for read modify write instructions It is also current not done if the OF or CF flags are needed The shift operators have the complication that when the shift count is EFLAGS is not set
Definition: README.txt:1277
AssumptionCache.h
patchAndReplaceAllUsesWith
static void patchAndReplaceAllUsesWith(Instruction *I, Value *Repl)
Definition: NewGVN.cpp:3677
llvm::AAManager
A manager for alias analyses.
Definition: AliasAnalysis.h:1260
llvm::ValueDFS::LocalNum
unsigned int LocalNum
Definition: PredicateInfo.cpp:96
llvm::CmpInst::getSwappedPredicate
Predicate getSwappedPredicate() const
For example, EQ->EQ, SLE->SGE, ULT->UGT, OEQ->OEQ, ULE->UGE, OLT->OGT, etc.
Definition: InstrTypes.h:849
llvm::simplifyGEPInst
Value * simplifyGEPInst(Type *SrcTy, Value *Ptr, ArrayRef< Value * > Indices, bool InBounds, const SimplifyQuery &Q)
Given operands for a GetElementPtrInst, fold the result or return null.
Definition: InstructionSimplify.cpp:4700
llvm
This is an optimization pass for GlobalISel generic memory operations.
Definition: AddressRanges.h:18
llvm::sys::path::const_iterator::end
friend const_iterator end(StringRef path)
Get end iterator over path.
Definition: Path.cpp:235
llvm::tgtok::Def
@ Def
Definition: TGLexer.h:50
NewGVN::ValueDFS
Definition: NewGVN.cpp:3501
M
We currently emits eax Perhaps this is what we really should generate is Is imull three or four cycles eax eax The current instruction priority is based on pattern complexity The former is more complex because it folds a load so the latter will not be emitted Perhaps we should use AddedComplexity to give LEA32r a higher priority We should always try to match LEA first since the LEA matching code does some estimate to determine whether the match is profitable if we care more about code then imull is better It s two bytes shorter than movl leal On a Pentium M
Definition: README.txt:252
llvm::CmpInst::ICMP_EQ
@ ICMP_EQ
equal
Definition: InstrTypes.h:740
llvm::make_range
iterator_range< T > make_range(T x, T y)
Convenience function for iterating over sub-ranges.
Definition: iterator_range.h:53
llvm::set_is_subset
bool set_is_subset(const S1Ty &S1, const S2Ty &S2)
set_is_subset(A, B) - Return true iff A in B
Definition: SetOperations.h:72
llvm::SmallPtrSetImpl::erase
bool erase(PtrType Ptr)
erase - If the set contains the specified pointer, remove it and return true, otherwise return false.
Definition: SmallPtrSet.h:379
getBlockName
static std::string getBlockName(const BasicBlock *B)
Definition: NewGVN.cpp:932
llvm::DataLayout
A parsed version of the target data layout string in and methods for querying it.
Definition: DataLayout.h:113
llvm::BumpPtrAllocatorImpl::Reset
void Reset()
Deallocate all but the current slab and reset the current pointer to the beginning of it,...
Definition: Allocator.h:123
PredicateInfo.h
llvm::CmpInst::Predicate
Predicate
This enumeration lists the possible predicates for CmpInst subclasses.
Definition: InstrTypes.h:719
llvm::User::operands
op_range operands()
Definition: User.h:242
PHI
Rewrite undef for PHI
Definition: AMDGPURewriteUndefForPHI.cpp:101
IntrinsicInst.h
llvm::SimplifyQuery
Definition: InstructionSimplify.h:93
llvm::AnalysisManager::getResult
PassT::Result & getResult(IRUnitT &IR, ExtraArgTs... ExtraArgs)
Get the result of an analysis pass for a given IR unit.
Definition: PassManager.h:774
llvm::pdb::PDB_DataKind::Member
@ Member
Scalar.h
SetOperations.h
T
llvm::Function
Definition: Function.h:60
llvm::optimized_def_chain
iterator_range< def_chain_iterator< T, true > > optimized_def_chain(T MA)
Definition: MemorySSA.h:1342
llvm::DenseMapBase::lookup
ValueT lookup(const_arg_type_t< KeyT > Val) const
lookup - Return the entry for the specified key, or a default constructed value if no such entry exis...
Definition: DenseMap.h:197
P
This currently compiles esp xmm0 movsd esp eax eax esp ret We should use not the dag combiner This is because dagcombine2 needs to be able to see through the X86ISD::Wrapper which DAGCombine can t really do The code for turning x load into a single vector load is target independent and should be moved to the dag combiner The code for turning x load into a vector load can only handle a direct load from a global or a direct load from the stack It should be generalized to handle any load from P
Definition: README-SSE.txt:411
Pass.h
llvm::BitVector::set
BitVector & set()
Definition: BitVector.h:344
llvm::LocalNum
LocalNum
Definition: PredicateInfo.cpp:81
llvm::BitVector::clear
void clear()
clear - Removes all bits from the bitvector.
Definition: BitVector.h:328
llvm::DenseMapInfo< const Expression * >::getTombstoneKey
static const Expression * getTombstoneKey()
Definition: NewGVN.cpp:451
llvm::SmallVector
This is a 'vector' (really, a variable-sized array), optimized for the case when the array is small.
Definition: SmallVector.h:1181
isCopyOfAPHI
static bool isCopyOfAPHI(const Value *V)
Definition: NewGVN.cpp:987
Statistic.h
llvm::detail::DenseSetImpl::find
iterator find(const_arg_type_t< ValueT > V)
Definition: DenseSet.h:179
ErrorHandling.h
Allocator.h
llvm::BitVector::set_bits
iterator_range< const_set_bits_iterator > set_bits() const
Definition: BitVector.h:133
llvm::CmpInst::getInversePredicate
Predicate getInversePredicate() const
For example, EQ -> NE, UGT -> ULE, SLT -> SGE, OEQ -> UNE, UGT -> OLE, OLT -> UGE,...
Definition: InstrTypes.h:833
ValueTracking.h
Local.h
llvm::ValueDFS::DFSIn
int DFSIn
Definition: PredicateInfo.cpp:94
llvm::DominatorTree
Concrete subclass of DominatorTreeBase that is used to compute a normal dominator tree.
Definition: Dominators.h:166
alwaysAvailable
static bool alwaysAvailable(Value *V)
Definition: NewGVN.cpp:1005
GlobalsModRef.h
llvm::cl::Hidden
@ Hidden
Definition: CommandLine.h:139
llvm::PatternMatch::m_Br
brc_match< Cond_t, bind_ty< BasicBlock >, bind_ty< BasicBlock > > m_Br(const Cond_t &C, BasicBlock *&T, BasicBlock *&F)
Definition: PatternMatch.h:1731
llvm::DenseMapBase::erase
bool erase(const KeyT &Val)
Definition: DenseMap.h:302
llvm::Type
The instances of the Type class are immutable: once they are created, they are never changed.
Definition: Type.h:45
NL
#define NL
Definition: DetailedRecordsBackend.cpp:31
DenseMap.h
MemoryBuiltins.h
llvm::BitVector::resize
void resize(unsigned N, bool t=false)
resize - Grow or shrink the bitvector.
Definition: BitVector.h:334
llvm::sys::path::end
const_iterator end(StringRef path)
Get end iterator over path.
Definition: Path.cpp:235
llvm::MemorySSA::getLiveOnEntryDef
MemoryAccess * getLiveOnEntryDef() const
Definition: MemorySSA.h:741
llvm::sys::path::begin
const_iterator begin(StringRef path, Style style=Style::native)
Get begin iterator over path.
Definition: Path.cpp:226
llvm::BasicBlockEdge
Definition: Dominators.h:98
llvm::copy
OutputIt copy(R &&Range, OutputIt Out)
Definition: STLExtras.h:1641
llvm::DominatorTreeBase::getNode
DomTreeNodeBase< NodeT > * getNode(const NodeT *BB) const
getNode - return the (Post)DominatorTree node for the specified basic block.
Definition: GenericDomTree.h:351
llvm::Optional
Definition: APInt.h:33
llvm::DenseMapBase::count
size_type count(const_arg_type_t< KeyT > Val) const
Return 1 if the specified key is in the map, 0 otherwise.
Definition: DenseMap.h:145
Offset
uint64_t Offset
Definition: ELFObjHandler.cpp:79
llvm::SmallPtrSet< const Value *, 8 >
llvm::GVNExpression::StoreExpression::~StoreExpression
~StoreExpression() override
Hashing.h
llvm::MemoryPhi
Represents phi nodes for memory accesses.
Definition: MemorySSA.h:479
llvm::max
Expected< ExpressionValue > max(const ExpressionValue &Lhs, const ExpressionValue &Rhs)
Definition: FileCheck.cpp:337
getCopyOf
static Value * getCopyOf(const Value *V)
Definition: NewGVN.cpp:975
equalsLoadStoreHelper
static bool equalsLoadStoreHelper(const T &LHS, const Expression &RHS)
Definition: NewGVN.cpp:904
RHS
Value * RHS
Definition: X86PartialReduction.cpp:76
llvm::LoadInst::getPointerOperand
Value * getPointerOperand()
Definition: Instructions.h:261
llvm::detail::DenseSetImpl< ValueT, DenseMap< ValueT, detail::DenseSetEmpty, DenseMapInfo< ValueT >, detail::DenseSetPair< ValueT > >, DenseMapInfo< ValueT > >::insert
std::pair< iterator, bool > insert(const ValueT &V)
Definition: DenseSet.h:206
llvm::SPII::Load
@ Load
Definition: SparcInstrInfo.h:32
llvm::detail::DenseSetImpl< ValueT, DenseMap< ValueT, detail::DenseSetEmpty, DenseMapInfo< ValueT >, detail::DenseSetPair< ValueT > >, DenseMapInfo< ValueT > >::count
size_type count(const_arg_type_t< ValueT > V) const
Return 1 if the specified key is in the set, 0 otherwise.
Definition: DenseSet.h:97
llvm::Type::getInt8Ty
static IntegerType * getInt8Ty(LLVMContext &C)
Definition: Type.cpp:237
ConstantFolding.h
Use.h
llvm::DebugCounter::getCounterValue
static int64_t getCounterValue(unsigned ID)
Definition: DebugCounter.h:107
llvm::GVNExpression::CallExpression::~CallExpression
~CallExpression() override
llvm::MemoryPhi::getIncomingBlock
BasicBlock * getIncomingBlock(unsigned I) const
Return incoming basic block number i.
Definition: MemorySSA.h:540
LLVM_DEBUG
#define LLVM_DEBUG(X)
Definition: Debug.h:101
DepthFirstIterator.h
llvm::map_iterator
mapped_iterator< ItTy, FuncTy > map_iterator(ItTy I, FuncTy F)
Definition: STLExtras.h:310
llvm::DebugCounter::isCounterSet
static bool isCounterSet(unsigned ID)
Definition: DebugCounter.h:102
F
#define F(x, y, z)
Definition: MD5.cpp:55
llvm::RISCVFenceField::R
@ R
Definition: RISCVBaseInfo.h:265
llvm::BasicBlock
LLVM Basic Block Representation.
Definition: BasicBlock.h:55
AliasAnalysis.h
result
It looks like we only need to define PPCfmarto for these because according to these instructions perform RTO on fma s result
Definition: README_P9.txt:256
PointerIntPair.h
llvm::dbgs
raw_ostream & dbgs()
dbgs() - This returns a reference to a raw_ostream for debugging messages.
Definition: Debug.cpp:163
GraphTraits.h
llvm::GVNExpression::UnknownExpression
Definition: GVNExpression.h:625
llvm::DomTreeNodeBase::getDFSNumIn
unsigned getDFSNumIn() const
getDFSNumIn/getDFSNumOut - These return the DFS visitation order for nodes in the dominator tree.
Definition: GenericDomTree.h:143
SparseBitVector.h
llvm::DominatorTree::dominates
bool dominates(const BasicBlock *BB, const Use &U) const
Return true if the (end of the) basic block BB dominates the use U.
Definition: Dominators.cpp:122
llvm::DomTreeNodeBase::getDFSNumOut
unsigned getDFSNumOut() const
Definition: GenericDomTree.h:144
Arg
amdgpu Simplify well known AMD library false FunctionCallee Value * Arg
Definition: AMDGPULibCalls.cpp:187
llvm::detail::DenseSetImpl< ValueT, DenseMap< ValueT, detail::DenseSetEmpty, DenseMapInfo< ValueT >, detail::DenseSetPair< ValueT > >, DenseMapInfo< ValueT > >::end
iterator end()
Definition: DenseSet.h:174
llvm::ValueDFS::U
Use * U
Definition: PredicateInfo.cpp:99
Instruction.h
CommandLine.h
llvm::GVNExpression::int_op_inserter
Definition: GVNExpression.h:480
llvm::GVNExpression::ConstantExpression
Definition: GVNExpression.h:588
LHS
Value * LHS
Definition: X86PartialReduction.cpp:75
llvm::ConstantInt
This is the shared class of boolean and integer constants.
Definition: Constants.h:79
llvm::Instruction::getNumSuccessors
unsigned getNumSuccessors() const
Return the number of successors that this instruction has.
Definition: Instruction.cpp:803
llvm::DenseMapInfo
An information struct used to provide DenseMap with the various necessary components for a given valu...
Definition: APInt.h:34
llvm::all_of
bool all_of(R &&range, UnaryPredicate P)
Provide wrappers to std::all_of which take ranges instead of having to pass begin/end explicitly.
Definition: STLExtras.h:1590
llvm::orc::tpctypes::LookupResult
std::vector< JITTargetAddress > LookupResult
Definition: TargetProcessControlTypes.h:130
llvm::MemorySSAWrapperPass
Legacy analysis pass which computes MemorySSA.
Definition: MemorySSA.h:984
llvm::DominatorTreeBase::getRootNode
DomTreeNodeBase< NodeT > * getRootNode()
getRootNode - This returns the entry node for the CFG of the function.
Definition: GenericDomTree.h:370
llvm::PassRegistry::getPassRegistry
static PassRegistry * getPassRegistry()
getPassRegistry - Access the global registry object, which is automatically initialized at applicatio...
Definition: PassRegistry.cpp:24
llvm::DenseMapInfo< const Expression * >::isEqual
static bool isEqual(const Expression *LHS, const Expression *RHS)
Definition: NewGVN.cpp:471
NewGVN::ValueDFS::operator<
bool operator<(const ValueDFS &Other) const
Definition: NewGVN.cpp:3512
llvm::MutableArrayRef
MutableArrayRef - Represent a mutable reference to an array (0 or more elements consecutively in memo...
Definition: ArrayRef.h:28
llvm::PredicateConstraint::OtherOp
Value * OtherOp
Definition: PredicateInfo.h:77
llvm::GVNExpression::PHIExpression
Definition: GVNExpression.h:505
Constants.h
llvm::PatternMatch::match
bool match(Val *V, const Pattern &P)
Definition: PatternMatch.h:49
llvm::ExactEqualsExpression::ExactEqualsExpression
ExactEqualsExpression(const Expression &E)
Definition: NewGVN.cpp:435
isZero
static bool isZero(Value *V, const DataLayout &DL, DominatorTree *DT, AssumptionCache *AC)
Definition: Lint.cpp:524
llvm::AAResults
Definition: AliasAnalysis.h:518
llvm::getStoredValue
SDValue getStoredValue(SDValue Op)
Definition: VECustomDAG.cpp:332
llvm::MemorySSA::isLiveOnEntryDef
bool isLiveOnEntryDef(const MemoryAccess *MA) const
Return true if MA represents the live on entry value.
Definition: MemorySSA.h:737
P2
This might compile to this xmm1 xorps xmm0 movss xmm0 ret Now consider if the code caused xmm1 to get spilled This might produce this xmm1 movaps xmm0 movaps xmm1 movss xmm0 ret since the reload is only used by these we could fold it into the producing something like xmm1 movaps xmm0 ret saving two instructions The basic idea is that a reload from a spill if only one byte chunk is bring in zeros the one element instead of elements This can be used to simplify a variety of shuffle where the elements are fixed zeros This code generates ugly probably due to costs being off or< 4 x float > * P2
Definition: README-SSE.txt:278
E
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
llvm::User
Definition: User.h:44
llvm::initializeNewGVNLegacyPassPass
void initializeNewGVNLegacyPassPass(PassRegistry &)
C
(vector float) vec_cmpeq(*A, *B) C
Definition: README_ALTIVEC.txt:86
llvm::ARM_PROC::A
@ A
Definition: ARMBaseInfo.h:34
InstrTypes.h
EnableStoreRefinement
static cl::opt< bool > EnableStoreRefinement("enable-store-refinement", cl::init(false), cl::Hidden)
llvm::JumpTable::Simplified
@ Simplified
Definition: TargetOptions.h:47
llvm::createNewGVNPass
FunctionPass * createNewGVNPass()
Definition: NewGVN.cpp:4254
llvm::AnalysisUsage
Represent the analysis usage information of a pass.
Definition: PassAnalysisSupport.h:47
llvm::ms_demangle::QualifierMangleMode::Result
@ Result
llvm::simplifyBinOp
Value * simplifyBinOp(unsigned Opcode, Value *LHS, Value *RHS, const SimplifyQuery &Q)
Given operands for a BinaryOperator, fold the result or return null.
Definition: InstructionSimplify.cpp:5575
NewGVN::ValueDFS::Def
PointerIntPair< Value *, 1, bool > Def
Definition: NewGVN.cpp:3509
llvm::AMDGPU::PALMD::Key
Key
PAL metadata keys.
Definition: AMDGPUMetadata.h:486
TargetLibraryInfo.h
llvm::GVNExpression::LoadExpression::equals
bool equals(const Expression &Other) const override
Definition: NewGVN.cpp:910
AssumeBundleBuilder.h
llvm::SPII::Store
@ Store
Definition: SparcInstrInfo.h:33
DenseSet.h
false
Definition: StackSlotColoring.cpp:141
B
static GCRegistry::Add< OcamlGC > B("ocaml", "ocaml 3.10-compatible GC")
First
into llvm powi allowing the code generator to produce balanced multiplication trees First
Definition: README.txt:54
llvm::Value::getValueID
unsigned getValueID() const
Return an ID for the concrete type of this object.
Definition: Value.h:532
llvm::VNCoercion
Definition: VNCoercion.h:34
llvm::GVNExpression::Expression
Definition: GVNExpression.h:60
llvm::Instruction
Definition: Instruction.h:42
llvm::DebugCounter::shouldExecute
static bool shouldExecute(unsigned CounterName)
Definition: DebugCounter.h:74
llvm::DominatorTreeWrapperPass
Legacy analysis pass which computes a DominatorTree.
Definition: Dominators.h:302
GVNExpression.h
llvm::STATISTIC
STATISTIC(NumFunctions, "Total number of functions")
isCopyOfPHI
static bool isCopyOfPHI(const Value *V, const PHINode *PN)
Definition: NewGVN.cpp:983
PointerLikeTypeTraits.h
EnablePhiOfOps
static cl::opt< bool > EnablePhiOfOps("enable-phi-of-ops", cl::init(true), cl::Hidden)
Currently, the generation "phi of ops" can result in correctness issues.
llvm::AArch64CC::LE
@ LE
Definition: AArch64BaseInfo.h:268
llvm::UndefValue::get
static UndefValue * get(Type *T)
Static factory methods - Return an 'undef' object of the specified type.
Definition: Constants.cpp:1713
BitVector.h
llvm::GVNExpression::BasicExpression::~BasicExpression
~BasicExpression() override
llvm::CmpInst::FCMP_OEQ
@ FCMP_OEQ
0 0 0 1 True if ordered and equal
Definition: InstrTypes.h:722
SmallPtrSet.h
llvm::Instruction::getSuccessor
BasicBlock * getSuccessor(unsigned Idx) const
Return the specified successor. This instruction must be a terminator.
Definition: Instruction.cpp:815
llvm::BitVector
Definition: BitVector.h:75
llvm::DominatorTreeBase::updateDFSNumbers
void updateDFSNumbers() const
updateDFSNumbers - Assign In and Out numbers to the nodes while walking dominator tree in dfs order.
Definition: GenericDomTree.h:732
none
@ none
Definition: HWAddressSanitizer.cpp:188
PatternMatch.h
llvm::ExactEqualsExpression::getComputedHash
hash_code getComputedHash() const
Definition: NewGVN.cpp:437
llvm::MCID::Call
@ Call
Definition: MCInstrDesc.h:155
place
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 ID Predecessors according to mbb< bb27, 0x8b0a7c0 > Note ADDri is not a two address instruction its result reg1037 is an operand of the PHI node in bb76 and its operand reg1039 is the result of the PHI node We should treat it as a two address code and make sure the ADDri is scheduled after any node that reads reg1039 Use info(i.e. register scavenger) to assign it a free register to allow reuse the collector could move the objects and invalidate the derived pointer This is bad enough in the first place
Definition: README.txt:50
llvm::GVNExpression::PHIExpression::~PHIExpression
~PHIExpression() override
llvm::MemoryAccess::getBlock
BasicBlock * getBlock() const
Definition: MemorySSA.h:164
llvm::Value::use_empty
bool use_empty() const
Definition: Value.h:344
llvm::CallingConv::ID
unsigned ID
LLVM IR allows to use arbitrary numbers as calling convention identifiers.
Definition: CallingConv.h:24
Type.h
X
static GCMetadataPrinterRegistry::Add< ErlangGCPrinter > X("erlang", "erlang-compatible garbage collector")
INITIALIZE_PASS_BEGIN
INITIALIZE_PASS_BEGIN(NewGVNLegacyPass, "newgvn", "Global Value Numbering", false, false) INITIALIZE_PASS_END(NewGVNLegacyPass
llvm::PredicateInfo
Encapsulates PredicateInfo, including all data associated with memory accesses.
Definition: PredicateInfo.h:178
INITIALIZE_PASS_END
#define INITIALIZE_PASS_END(passName, arg, name, cfg, analysis)
Definition: PassSupport.h:58
llvm::DenseMapInfo< const Expression * >::isEqual
static bool isEqual(const ExactEqualsExpression &LHS, const Expression *RHS)
Definition: NewGVN.cpp:465
llvm::Instruction::isCommutative
bool isCommutative() const LLVM_READONLY
Return true if the instruction is commutative:
Definition: Instruction.cpp:796
llvm::PredicateConstraint::Predicate
CmpInst::Predicate Predicate
Definition: PredicateInfo.h:76
llvm::DOTGraphTraits
DOTGraphTraits - Template class that can be specialized to customize how graphs are converted to 'dot...
Definition: DOTGraphTraits.h:166
llvm::ExactEqualsExpression::operator==
bool operator==(const Expression &Other) const
Definition: NewGVN.cpp:439
llvm::sort
void sort(IteratorTy Start, IteratorTy End)
Definition: STLExtras.h:1538
llvm::DenseSet
Implements a dense probed hash-table based set.
Definition: DenseSet.h:268
llvm::MemorySSAWalker::getClobberingMemoryAccess
MemoryAccess * getClobberingMemoryAccess(const Instruction *I)
Given a memory Mod/Ref/ModRef'ing instruction, calling this will give you the nearest dominating Memo...
Definition: MemorySSA.h:1044
llvm::wouldInstructionBeTriviallyDead
bool wouldInstructionBeTriviallyDead(Instruction *I, const TargetLibraryInfo *TLI=nullptr)
Return true if the result produced by the instruction would have no side effects if it was not used.
Definition: Local.cpp:414
llvm::PredicateBase
Definition: PredicateInfo.h:82
DEBUG_COUNTER
DEBUG_COUNTER(VNCounter, "newgvn-vn", "Controls which instructions are value numbered")
BasicBlock.h
llvm::cl::opt< bool >
llvm::ValueDFS
Definition: PredicateInfo.cpp:93
llvm::RISCVFenceField::O
@ O
Definition: RISCVBaseInfo.h:264
llvm::PPC::Predicate
Predicate
Predicate - These are "(BI << 5) | BO" for various predicates.
Definition: PPCPredicates.h:26
llvm::StoreInst
An instruction for storing to memory.
Definition: Instructions.h:298
llvm::DebugCounter::setCounterValue
static void setCounterValue(unsigned ID, int64_t Count)
Definition: DebugCounter.h:115
llvm::getInitialValueOfAllocation
Constant * getInitialValueOfAllocation(const Value *V, const TargetLibraryInfo *TLI, Type *Ty)
If this is a call to an allocation function that initializes memory to a fixed value,...
Definition: MemoryBuiltins.cpp:459
llvm::Constant
This is an important base class in LLVM.
Definition: Constant.h:41
llvm::Instruction::eraseFromParent
SymbolTableList< Instruction >::iterator eraseFromParent()
This method unlinks 'this' from the containing basic block and deletes it.
Definition: Instruction.cpp:77
CFGPrinter.h
llvm::DenseMapBase< DenseMap< KeyT, ValueT, DenseMapInfo< KeyT >, llvm::detail::DenseMapPair< KeyT, ValueT > >, KeyT, ValueT, DenseMapInfo< KeyT >, llvm::detail::DenseMapPair< KeyT, ValueT > >::clear
void clear()
Definition: DenseMap.h:110
llvm::GVNExpression
Definition: GVNExpression.h:40
llvm::TargetLibraryInfoWrapperPass
Definition: TargetLibraryInfo.h:468
uint64_t
llvm::GVNExpression::AggregateValueExpression::~AggregateValueExpression
~AggregateValueExpression() override
llvm::simplifySelectInst
Value * simplifySelectInst(Value *Cond, Value *TrueVal, Value *FalseVal, const SimplifyQuery &Q)
Given operands for a SelectInst, fold the result or return null.
Definition: InstructionSimplify.cpp:4555
D
static GCRegistry::Add< StatepointGC > D("statepoint-example", "an example strategy for statepoint")
llvm::BitVector::any
bool any() const
any - Returns true if any bit is set.
Definition: BitVector.h:163
const
aarch64 promote const
Definition: AArch64PromoteConstant.cpp:232
llvm::SmallPtrSetImpl::end
iterator end() const
Definition: SmallPtrSet.h:408
llvm::AssumptionAnalysis
A function analysis which provides an AssumptionCache.
Definition: AssumptionCache.h:173
llvm::PreservedAnalyses::preserve
void preserve()
Mark an analysis as preserved.
Definition: PassManager.h:173
INITIALIZE_PASS_DEPENDENCY
INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
llvm::BumpPtrAllocatorImpl
Allocate memory in an ever growing pool, as if by bump-pointer.
Definition: Allocator.h:63
llvm::numbers::e
constexpr double e
Definition: MathExtras.h:53
llvm::DenseMap
Definition: DenseMap.h:714
llvm::nodes
iterator_range< typename GraphTraits< GraphType >::nodes_iterator > nodes(const GraphType &G)
Definition: GraphTraits.h:110
llvm::MemorySSA
Encapsulates MemorySSA, including all data associated with memory accesses.
Definition: MemorySSA.h:700
I
#define I(x, y, z)
Definition: MD5.cpp:58
llvm::MemorySSA::getMemoryAccess
MemoryUseOrDef * getMemoryAccess(const Instruction *I) const
Given a memory Mod/Ref'ing instruction, get the MemorySSA access associated with it.
Definition: MemorySSA.h:717
VNCoercion.h
llvm::DenseMapInfo< const Expression * >::getHashValue
static unsigned getHashValue(const Expression *E)
Definition: NewGVN.cpp:457
llvm::cl::init
initializer< Ty > init(const Ty &Val)
Definition: CommandLine.h:439
newgvn
newgvn
Definition: NewGVN.cpp:4250
llvm::SmallPtrSetImpl::begin
iterator begin() const
Definition: SmallPtrSet.h:403
ArrayRef.h
llvm::detail::DenseSetImpl< ValueT, DenseMap< ValueT, detail::DenseSetEmpty, DenseMapInfo< ValueT >, detail::DenseSetPair< ValueT > >, DenseMapInfo< ValueT > >::begin
iterator begin()
Definition: DenseSet.h:173
llvm::pdb::PDB_MemoryType::Stack
@ Stack
transform
instcombine should handle this transform
Definition: README.txt:262
llvm::DenseMapBase< DenseMap< KeyT, ValueT, DenseMapInfo< KeyT >, llvm::detail::DenseMapPair< KeyT, ValueT > >, KeyT, ValueT, DenseMapInfo< KeyT >, llvm::detail::DenseMapPair< KeyT, ValueT > >::find
iterator find(const_arg_type_t< KeyT > Val)
Definition: DenseMap.h:150
llvm::sys::path::const_iterator::begin
friend const_iterator begin(StringRef path, Style style)
Get begin iterator over path.
Definition: Path.cpp:226
assert
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
std::swap
void swap(llvm::BitVector &LHS, llvm::BitVector &RHS)
Implement std::swap in terms of BitVector swap.
Definition: BitVector.h:853
SI
StandardInstrumentations SI(Debug, VerifyEach)
llvm::GVNExpression::StoreExpression
Definition: GVNExpression.h:370
llvm::GVNExpression::VariableExpression
Definition: GVNExpression.h:552
iterator_range.h
okayForPHIOfOps
static bool okayForPHIOfOps(const Instruction *I)
Definition: NewGVN.cpp:2567
llvm::SimplifyQuery::getWithInstruction
SimplifyQuery getWithInstruction(Instruction *I) const
Definition: InstructionSimplify.h:120
llvm::WinEH::EncodingType::CE
@ CE
Windows NT (Windows on ARM)
llvm::salvageKnowledge
void salvageKnowledge(Instruction *I, AssumptionCache *AC=nullptr, DominatorTree *DT=nullptr)
Calls BuildAssumeFromInst and if the resulting llvm.assume is valid insert if before I.
Definition: AssumeBundleBuilder.cpp:293
llvm::MemorySSAAnalysis
An analysis that produces MemorySSA for a function.
Definition: MemorySSA.h:934
llvm::MemorySSA::getWalker
MemorySSAWalker * getWalker()
Definition: MemorySSA.cpp:1559
llvm::SmallPtrSetImpl::count
size_type count(ConstPtrType Ptr) const
count - Return 1 if the specified pointer is in the set, 0 otherwise.
Definition: SmallPtrSet.h:383
llvm::ArrayRecycler::clear
void clear(AllocatorType &Allocator)
Release all the tracked allocations to the allocator.
Definition: ArrayRecycler.h:104
llvm::PatternMatch::m_Value
class_match< Value > m_Value()
Match an arbitrary value and ignore it.
Definition: PatternMatch.h:76
llvm::detail::DenseSetImpl< ValueT, DenseMap< ValueT, detail::DenseSetEmpty, DenseMapInfo< ValueT >, detail::DenseSetPair< ValueT > >, DenseMapInfo< ValueT > >::clear
void clear()
Definition: DenseSet.h:92
llvm::size
auto size(R &&Range, std::enable_if_t< std::is_base_of< std::random_access_iterator_tag, typename std::iterator_traits< decltype(Range.begin())>::iterator_category >::value, void > *=nullptr)
Get the size of a range.
Definition: STLExtras.h:1571
llvm::GVNExpression::StoreExpression::equals
bool equals(const Expression &Other) const override
Definition: NewGVN.cpp:914
llvm::AMDGPU::CPol::SCC
@ SCC
Definition: SIDefines.h:307
llvm::ConstantFoldInstOperands
Constant * ConstantFoldInstOperands(Instruction *I, ArrayRef< Constant * > Ops, const DataLayout &DL, const TargetLibraryInfo *TLI=nullptr)
ConstantFoldInstOperands - Attempt to constant fold an instruction with the specified operands.
Definition: ConstantFolding.cpp:1213
llvm::Expression
Class representing an expression and its matching format.
Definition: FileCheckImpl.h:237
llvm::SmallPtrSetImplBase::clear
void clear()
Definition: SmallPtrSet.h:95
llvm::AssumptionCacheTracker
An immutable pass that tracks lazily created AssumptionCache objects.
Definition: AssumptionCache.h:202
llvm::ArrayRef
ArrayRef - Represent a constant reference to an array (0 or more elements consecutively in memory),...
Definition: APInt.h:32
llvm::isInstructionTriviallyDead
bool isInstructionTriviallyDead(Instruction *I, const TargetLibraryInfo *TLI=nullptr)
Return true if the result produced by the instruction is not used, and the instruction will return.
Definition: Local.cpp:395
llvm::VNCoercion::getConstantLoadValueForLoad
Constant * getConstantLoadValueForLoad(Constant *SrcVal, unsigned Offset, Type *LoadTy, const DataLayout &DL)
Definition: VNCoercion.cpp:507
llvm::simplifyCastInst
Value * simplifyCastInst(unsigned CastOpc, Value *Op, Type *Ty, const SimplifyQuery &Q)
Given operands for a CastInst, fold the result or return null.
Definition: InstructionSimplify.cpp:4923
llvm::min
Expected< ExpressionValue > min(const ExpressionValue &Lhs, const ExpressionValue &Rhs)
Definition: FileCheck.cpp:357
Mul
BinaryOperator * Mul
Definition: X86PartialReduction.cpp:70
llvm::any_of
bool any_of(R &&range, UnaryPredicate P)
Provide wrappers to std::any_of which take ranges instead of having to pass begin/end explicitly.
Definition: STLExtras.h:1597
Cond
SmallVector< MachineOperand, 4 > Cond
Definition: BasicBlockSections.cpp:137
llvm::LoadInst::isSimple
bool isSimple() const
Definition: Instructions.h:253
llvm::AssumptionCache
A cache of @llvm.assume calls within a function.
Definition: AssumptionCache.h:42
llvm_unreachable
#define llvm_unreachable(msg)
Marks that the current location is not supposed to be reachable.
Definition: ErrorHandling.h:143
llvm::Value::getType
Type * getType() const
All values are typed, get the type of this value.
Definition: Value.h:255
llvm::MemorySSA::getBlockDefs
const DefsList * getBlockDefs(const BasicBlock *BB) const
Return the list of MemoryDef's and MemoryPhi's for a given basic block.
Definition: MemorySSA.h:765
if
if(llvm_vc STREQUAL "") set(fake_version_inc "$
Definition: CMakeLists.txt:14
llvm::ConstantInt::isZero
bool isZero() const
This is just a convenience method to make client code smaller for a common code.
Definition: Constants.h:194
llvm::AnalysisUsage::addPreserved
AnalysisUsage & addPreserved()
Add the specified Pass class to the set of analyses preserved by this pass.
Definition: PassAnalysisSupport.h:98
llvm::Value::replaceAllUsesWith
void replaceAllUsesWith(Value *V)
Change all uses of this to point to a new Value.
Definition: Value.cpp:532
A
* A
Definition: README_ALTIVEC.txt:89
DL
MachineBasicBlock MachineBasicBlock::iterator DebugLoc DL
Definition: AArch64SLSHardening.cpp:76
llvm::ExactEqualsExpression
Definition: NewGVN.cpp:432
S
add sub stmia L5 ldr r0 bl L_printf $stub Instead of a and a wouldn t it be better to do three moves *Return an aggregate type is even return S
Definition: README.txt:210
NewGVN.h
CC
auto CC
Definition: RISCVRedundantCopyElimination.cpp:79
llvm::GVNExpression::CallExpression
Definition: GVNExpression.h:301
llvm::DominatorTreeBase::properlyDominates
bool properlyDominates(const DomTreeNodeBase< NodeT > *A, const DomTreeNodeBase< NodeT > *B) const
properlyDominates - Returns true iff A dominates B and A != B.
Definition: GenericDomTree.h:392
llvm::MemoryAccess
Definition: MemorySSA.h:142
llvm::DomTreeNodeBase< BasicBlock >
llvm::LoadInst
An instruction for reading from memory.
Definition: Instructions.h:174
llvm::DenseMapBase< DenseMap< KeyT, ValueT, DenseMapInfo< KeyT >, llvm::detail::DenseMapPair< KeyT, ValueT > >, KeyT, ValueT, DenseMapInfo< KeyT >, llvm::detail::DenseMapPair< KeyT, ValueT > >::insert
std::pair< iterator, bool > insert(const std::pair< KeyT, ValueT > &KV)
Definition: DenseMap.h:207
llvm::make_filter_range
iterator_range< filter_iterator< detail::IterOfRange< RangeT >, PredicateT > > make_filter_range(RangeT &&Range, PredicateT Pred)
Convenience function that takes a range of elements and a predicate, and return a new filter_iterator...
Definition: STLExtras.h:512
llvm::DenseMapInfo< const Expression * >::getHashValue
static unsigned getHashValue(const ExactEqualsExpression &E)
Definition: NewGVN.cpp:461
llvm::ValueDFS::Def
Value * Def
Definition: PredicateInfo.cpp:98
llvm::ConstantInt::getFalse
static ConstantInt * getFalse(LLVMContext &Context)
Definition: Constants.cpp:834
Argument.h
llvm::ArrayRecycler
Recycle small arrays allocated from a BumpPtrAllocator.
Definition: ArrayRecycler.h:28
llvm::depth_first
iterator_range< df_iterator< T > > depth_first(const T &G)
Definition: DepthFirstIterator.h:230
llvm::MCID::Select
@ Select
Definition: MCInstrDesc.h:164
runOnFunction
static bool runOnFunction(Function &F, bool PostInlining)
Definition: EntryExitInstrumenter.cpp:69
llvm::GVNExpression::Expression::~Expression
virtual ~Expression()
llvm::VNCoercion::getConstantMemInstValueForLoad
Constant * getConstantMemInstValueForLoad(MemIntrinsic *SrcInst, unsigned Offset, Type *LoadTy, const DataLayout &DL)
Definition: VNCoercion.cpp:566
llvm::CmpInst::isImpliedFalseByMatchingCmp
static bool isImpliedFalseByMatchingCmp(Predicate Pred1, Predicate Pred2)
Determine if Pred1 implies Pred2 is false when two compares have matching operands.
Definition: Instructions.cpp:4373
llvm::Instruction::isAtomic
bool isAtomic() const
Return true if this instruction has an AtomicOrdering of unordered or higher.
Definition: Instruction.cpp:642
Constant.h
llvm::CmpInst::isImpliedTrueByMatchingCmp
static bool isImpliedTrueByMatchingCmp(Predicate Pred1, Predicate Pred2)
Determine if Pred1 implies Pred2 is true when two compares have matching operands.
Definition: Instructions.cpp:4348
llvm::empty
constexpr bool empty(const T &RangeOrContainer)
Test whether RangeOrContainer is empty. Similar to C++17 std::empty.
Definition: STLExtras.h:256
llvm::ConstantInt::getTrue
static ConstantInt * getTrue(LLVMContext &Context)
Definition: Constants.cpp:827
std
Definition: BitVector.h:851
llvm::Constant::getNullValue
static Constant * getNullValue(Type *Ty)
Constructor to create a '0' constant of arbitrary type.
Definition: Constants.cpp:350
llvm::DenseMapBase< DenseMap< KeyT, ValueT, DenseMapInfo< KeyT >, llvm::detail::DenseMapPair< KeyT, ValueT > >, KeyT, ValueT, DenseMapInfo< KeyT >, llvm::detail::DenseMapPair< KeyT, ValueT > >::end
iterator end()
Definition: DenseMap.h:84
llvm::AMDGPU::SendMsg::Op
Op
Definition: SIDefines.h:348
llvm::PreservedAnalyses::all
static PreservedAnalyses all()
Construct a special preserved set that preserves all passes.
Definition: PassManager.h:158
llvm::PHINode::Create
static PHINode * Create(Type *Ty, unsigned NumReservedValues, const Twine &NameStr="", Instruction *InsertBefore=nullptr)
Constructors - NumReservedValues is a hint for the number of incoming edges that this phi node will h...
Definition: Instructions.h:2741
llvm::ArrayRef::begin
iterator begin() const
Definition: ArrayRef.h:152
llvm::VNCoercion::analyzeLoadFromClobberingStore
int analyzeLoadFromClobberingStore(Type *LoadTy, Value *LoadPtr, StoreInst *DepSI, const DataLayout &DL)
This function determines whether a value for the pointer LoadPtr can be extracted from the store at D...
Definition: VNCoercion.cpp:208
Casting.h
Function.h
llvm::GVNExpression::DeadExpression
Definition: GVNExpression.h:541
DebugCounter.h
llvm::DenseMapBase::size
unsigned size() const
Definition: DenseMap.h:99
llvm::TargetStackID::Value
Value
Definition: TargetFrameLowering.h:27
llvm::Type::getPointerTo
PointerType * getPointerTo(unsigned AddrSpace=0) const
Return a pointer to the current type.
Definition: Type.cpp:774
llvm::TargetLibraryInfo
Provides information about what library functions are available for the current target.
Definition: TargetLibraryInfo.h:222
Numbering
Global Value Numbering
Definition: NewGVN.cpp:4250
llvm::MemoryUseOrDef
Class that has the common methods + fields of memory uses/defs.
Definition: MemorySSA.h:252
llvm::SmallVectorImpl::clear
void clear()
Definition: SmallVector.h:596
llvm::ExactEqualsExpression::E
const Expression & E
Definition: NewGVN.cpp:433
llvm::MCID::Add
@ Add
Definition: MCInstrDesc.h:185
llvm::GVNExpression::LoadExpression::~LoadExpression
~LoadExpression() override
llvm::ReversePostOrderTraversal
Definition: PostOrderIterator.h:291
llvm::BitVector::reset
BitVector & reset()
Definition: BitVector.h:385
llvm::IntrinsicInst
A wrapper class for inspecting calls to intrinsic functions.
Definition: IntrinsicInst.h:46
llvm::VNCoercion::getConstantStoreValueForLoad
Constant * getConstantStoreValueForLoad(Constant *SrcVal, unsigned Offset, Type *LoadTy, const DataLayout &DL)
Definition: VNCoercion.cpp:450
llvm::makeArrayRef
ArrayRef< T > makeArrayRef(const T &OneElt)
Construct an ArrayRef from a single element.
Definition: ArrayRef.h:475
AA
llvm::NewGVNPass::run
PreservedAnalyses run(Function &F, AnalysisManager< Function > &AM)
Run the pass over the function.
Definition: NewGVN.cpp:4256
llvm::DominatorTreeAnalysis
Analysis pass which computes a DominatorTree.
Definition: Dominators.h:267
llvm::DenseMapInfo< const Expression * >::getEmptyKey
static const Expression * getEmptyKey()
Definition: NewGVN.cpp:445
llvm::BasicBlock::reverse_iterator
InstListType::reverse_iterator reverse_iterator
Definition: BasicBlock.h:89
MemorySSA.h
Instructions.h
PostOrderIterator.h
llvm::GVNExpression::AggregateValueExpression
Definition: GVNExpression.h:411
llvm::User::getNumOperands
unsigned getNumOperands() const
Definition: User.h:191
llvm::PointerIntPair< Value *, 1, bool >
llvm::ValueDFS::DFSOut
int DFSOut
Definition: PredicateInfo.cpp:95
SmallVector.h
lookup
static bool lookup(const GsymReader &GR, DataExtractor &Data, uint64_t &Offset, uint64_t BaseAddr, uint64_t Addr, SourceLocations &SrcLocs, llvm::Error &Err)
A Lookup helper functions.
Definition: InlineInfo.cpp:108
User.h
ArrayRecycler.h
Dominators.h
llvm::BitVector::find_first
int find_first() const
find_first - Returns the index of the first set bit, -1 if none of the bits are set.
Definition: BitVector.h:293
llvm::AAResultsWrapperPass
A wrapper pass to provide the legacy pass manager access to a suitably prepared AAResults object.
Definition: AliasAnalysis.h:1308
llvm::Instruction::getParent
const BasicBlock * getParent() const
Definition: Instruction.h:91
InstructionSimplify.h
llvm::SmallVectorImpl::pop_back_val
T pop_back_val()
Definition: SmallVector.h:659
llvm::Value::deleteValue
void deleteValue()
Delete a pointer to a generic Value.
Definition: Value.cpp:109
llvm::VNCoercion::analyzeLoadFromClobberingLoad
int analyzeLoadFromClobberingLoad(Type *LoadTy, Value *LoadPtr, LoadInst *DepLI, const DataLayout &DL)
This function determines whether a value for the pointer LoadPtr can be extracted from the load at De...
Definition: VNCoercion.cpp:314
llvm::GlobalsAAWrapperPass
Legacy wrapper pass to provide the GlobalsAAResult object.
Definition: GlobalsModRef.h:144
llvm::PHINode::getIncomingBlock
BasicBlock * getIncomingBlock(unsigned i) const
Return incoming basic block number i.
Definition: Instructions.h:2815
llvm::ArrayRef::size
size_t size() const
size - Get the array size.
Definition: ArrayRef.h:164
llvm::iterator_range
A range adaptor for a pair of iterators.
Definition: iterator_range.h:30
llvm::PHINode
Definition: Instructions.h:2699
llvm::GVNExpression::BasicExpression
Definition: GVNExpression.h:136
llvm::PatternMatch
Definition: PatternMatch.h:47
llvm::detail::DenseSetImpl< ValueT, DenseMap< ValueT, detail::DenseSetEmpty, DenseMapInfo< ValueT >, detail::DenseSetPair< ValueT > >, DenseMapInfo< ValueT > >::erase
bool erase(const ValueT &V)
Definition: DenseSet.h:101
llvm::SmallVectorImpl
This class consists of common code factored out of the SmallVector class to reduce code duplication b...
Definition: APFloat.h:42
llvm::reverse
auto reverse(ContainerTy &&C)
Definition: STLExtras.h:365
DenseMapInfo.h
llvm::SmallPtrSetImpl< const Value * >
llvm::AnalysisManager
A container for analyses that lazily runs them and caches their results.
Definition: InstructionSimplify.h:42
llvm::FunctionPass
FunctionPass class - This class is used to implement most global optimizations.
Definition: Pass.h:308
llvm::CallInst
This class represents a function call, abstracting a target machine's calling convention.
Definition: Instructions.h:1474
llvm::ConstantInt::isOne
bool isOne() const
This is just a convenience method to make client code smaller for a common case.
Definition: Constants.h:200
llvm::isGuaranteedNotToBePoison
bool isGuaranteedNotToBePoison(const Value *V, AssumptionCache *AC=nullptr, const Instruction *CtxI=nullptr, const DominatorTree *DT=nullptr, unsigned Depth=0)
Definition: ValueTracking.cpp:5435
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
GEP
Hexagon Common GEP
Definition: HexagonCommonGEP.cpp:171
llvm::GVNExpression::op_inserter
Definition: GVNExpression.h:243
llvm::patchReplacementInstruction
void patchReplacementInstruction(Instruction *I, Value *Repl)
Patch the replacement so that it is not more restrictive than the value being replaced.
Definition: Local.cpp:2730
llvm::AnalysisUsage::addRequired
AnalysisUsage & addRequired()
Definition: PassAnalysisSupport.h:75
From
BlockVerifier::State From
Definition: BlockVerifier.cpp:55
llvm::MemorySSAWalker
This is the generic walker interface for walkers of MemorySSA.
Definition: MemorySSA.h:1015
llvm::User::getOperand
Value * getOperand(unsigned i) const
Definition: User.h:169
llvm::simplifyCmpInst
Value * simplifyCmpInst(unsigned Predicate, Value *LHS, Value *RHS, const SimplifyQuery &Q)
Given operands for a CmpInst, fold the result or return null.
Definition: InstructionSimplify.cpp:5593
raw_ostream.h
llvm::all_equal
bool all_equal(R &&Range)
Returns true if all elements in Range are equal or when the Range is empty.
Definition: STLExtras.h:1782
llvm::SmallPtrSetImpl::contains
bool contains(ConstPtrType Ptr) const
Definition: SmallPtrSet.h:389
copy
we should consider alternate ways to model stack dependencies Lots of things could be done in WebAssemblyTargetTransformInfo cpp there are numerous optimization related hooks that can be overridden in WebAssemblyTargetLowering Instead of the OptimizeReturned which should consider preserving the returned attribute through to MachineInstrs and extending the MemIntrinsicResults pass to do this optimization on calls too That would also let the WebAssemblyPeephole pass clean up dead defs for such as it does for stores Consider implementing and or getMachineCombinerPatterns Find a clean way to fix the problem which leads to the Shrink Wrapping pass being run after the WebAssembly PEI pass When setting multiple variables to the same we currently get code like const It could be done with a smaller encoding like local tee $pop5 local copy
Definition: README.txt:101
Value.h
llvm::GVNExpression::LoadExpression
Definition: GVNExpression.h:328
InitializePasses.h
llvm::Value
LLVM Value Representation.
Definition: Value.h:74
Debug.h
llvm::TargetLibraryAnalysis
Analysis pass providing the TargetLibraryInfo.
Definition: TargetLibraryInfo.h:443
llvm::Value::users
iterator_range< user_iterator > users()
Definition: Value.h:421
llvm::BumpPtrAllocatorImpl::Deallocate
void Deallocate(const void *Ptr, size_t Size, size_t)
Definition: Allocator.h:218
Other
Optional< std::vector< StOtherPiece > > Other
Definition: ELFYAML.cpp:1247
llvm::Use
A Use represents the edge between a Value definition and its users.
Definition: Use.h:43
llvm::SmallVectorImpl::emplace_back
reference emplace_back(ArgTypes &&... Args)
Definition: SmallVector.h:923
llvm::SmallPtrSetImpl::insert
std::pair< iterator, bool > insert(PtrType Ptr)
Inserts Ptr if and only if there is no element in the container equal to Ptr.
Definition: SmallPtrSet.h:365
llvm::Intrinsic::ID
unsigned ID
Definition: TargetTransformInfo.h:38
llvm::hash_code
An opaque object representing a hash code.
Definition: Hashing.h:73
llvm::PoisonValue::get
static PoisonValue * get(Type *T)
Static factory methods - Return an 'poison' object of the specified type.
Definition: Constants.cpp:1732