LLVM  6.0.0svn
GVN.cpp
Go to the documentation of this file.
1 //===- GVN.cpp - Eliminate redundant values and loads ---------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This pass performs global value numbering to eliminate fully redundant
11 // instructions. It also performs simple dead load elimination.
12 //
13 // Note that this pass does the value numbering itself; it does not use the
14 // ValueNumbering analysis passes.
15 //
16 //===----------------------------------------------------------------------===//
17 
19 #include "llvm/ADT/DenseMap.h"
21 #include "llvm/ADT/Hashing.h"
22 #include "llvm/ADT/MapVector.h"
24 #include "llvm/ADT/SetVector.h"
25 #include "llvm/ADT/SmallPtrSet.h"
26 #include "llvm/ADT/Statistic.h"
29 #include "llvm/Analysis/CFG.h"
33 #include "llvm/Analysis/Loads.h"
39 #include "llvm/IR/DataLayout.h"
40 #include "llvm/IR/Dominators.h"
41 #include "llvm/IR/GlobalVariable.h"
42 #include "llvm/IR/IRBuilder.h"
43 #include "llvm/IR/IntrinsicInst.h"
44 #include "llvm/IR/LLVMContext.h"
45 #include "llvm/IR/Metadata.h"
46 #include "llvm/IR/PatternMatch.h"
48 #include "llvm/Support/Debug.h"
54 
55 #include <vector>
56 using namespace llvm;
57 using namespace llvm::gvn;
58 using namespace llvm::VNCoercion;
59 using namespace PatternMatch;
60 
61 #define DEBUG_TYPE "gvn"
62 
63 STATISTIC(NumGVNInstr, "Number of instructions deleted");
64 STATISTIC(NumGVNLoad, "Number of loads deleted");
65 STATISTIC(NumGVNPRE, "Number of instructions PRE'd");
66 STATISTIC(NumGVNBlocks, "Number of blocks merged");
67 STATISTIC(NumGVNSimpl, "Number of instructions simplified");
68 STATISTIC(NumGVNEqProp, "Number of equalities propagated");
69 STATISTIC(NumPRELoad, "Number of loads PRE'd");
70 
71 static cl::opt<bool> EnablePRE("enable-pre",
72  cl::init(true), cl::Hidden);
73 static cl::opt<bool> EnableLoadPRE("enable-load-pre", cl::init(true));
74 
75 // Maximum allowed recursion depth.
76 static cl::opt<uint32_t>
77 MaxRecurseDepth("max-recurse-depth", cl::Hidden, cl::init(1000), cl::ZeroOrMore,
78  cl::desc("Max recurse depth (default = 1000)"));
79 
84 
85  Expression(uint32_t o = ~2U) : opcode(o) {}
86 
87  bool operator==(const Expression &other) const {
88  if (opcode != other.opcode)
89  return false;
90  if (opcode == ~0U || opcode == ~1U)
91  return true;
92  if (type != other.type)
93  return false;
94  if (varargs != other.varargs)
95  return false;
96  return true;
97  }
98 
100  return hash_combine(
101  Value.opcode, Value.type,
102  hash_combine_range(Value.varargs.begin(), Value.varargs.end()));
103  }
104 };
105 
106 namespace llvm {
107 template <> struct DenseMapInfo<GVN::Expression> {
108  static inline GVN::Expression getEmptyKey() { return ~0U; }
109 
110  static inline GVN::Expression getTombstoneKey() { return ~1U; }
111 
112  static unsigned getHashValue(const GVN::Expression &e) {
113  using llvm::hash_value;
114  return static_cast<unsigned>(hash_value(e));
115  }
116  static bool isEqual(const GVN::Expression &LHS, const GVN::Expression &RHS) {
117  return LHS == RHS;
118  }
119 };
120 } // End llvm namespace.
121 
122 /// Represents a particular available value that we know how to materialize.
123 /// Materialization of an AvailableValue never fails. An AvailableValue is
124 /// implicitly associated with a rematerialization point which is the
125 /// location of the instruction from which it was formed.
127  enum ValType {
128  SimpleVal, // A simple offsetted value that is accessed.
129  LoadVal, // A value produced by a load.
130  MemIntrin, // A memory intrinsic which is loaded from.
131  UndefVal // A UndefValue representing a value from dead block (which
132  // is not yet physically removed from the CFG).
133  };
134 
135  /// V - The value that is live out of the block.
137 
138  /// Offset - The byte offset in Val that is interesting for the load query.
139  unsigned Offset;
140 
141  static AvailableValue get(Value *V, unsigned Offset = 0) {
142  AvailableValue Res;
143  Res.Val.setPointer(V);
144  Res.Val.setInt(SimpleVal);
145  Res.Offset = Offset;
146  return Res;
147  }
148 
149  static AvailableValue getMI(MemIntrinsic *MI, unsigned Offset = 0) {
150  AvailableValue Res;
151  Res.Val.setPointer(MI);
152  Res.Val.setInt(MemIntrin);
153  Res.Offset = Offset;
154  return Res;
155  }
156 
157  static AvailableValue getLoad(LoadInst *LI, unsigned Offset = 0) {
158  AvailableValue Res;
159  Res.Val.setPointer(LI);
160  Res.Val.setInt(LoadVal);
161  Res.Offset = Offset;
162  return Res;
163  }
164 
166  AvailableValue Res;
167  Res.Val.setPointer(nullptr);
168  Res.Val.setInt(UndefVal);
169  Res.Offset = 0;
170  return Res;
171  }
172 
173  bool isSimpleValue() const { return Val.getInt() == SimpleVal; }
174  bool isCoercedLoadValue() const { return Val.getInt() == LoadVal; }
175  bool isMemIntrinValue() const { return Val.getInt() == MemIntrin; }
176  bool isUndefValue() const { return Val.getInt() == UndefVal; }
177 
179  assert(isSimpleValue() && "Wrong accessor");
180  return Val.getPointer();
181  }
182 
184  assert(isCoercedLoadValue() && "Wrong accessor");
185  return cast<LoadInst>(Val.getPointer());
186  }
187 
189  assert(isMemIntrinValue() && "Wrong accessor");
190  return cast<MemIntrinsic>(Val.getPointer());
191  }
192 
193  /// Emit code at the specified insertion point to adjust the value defined
194  /// here to the specified type. This handles various coercion cases.
195  Value *MaterializeAdjustedValue(LoadInst *LI, Instruction *InsertPt,
196  GVN &gvn) const;
197 };
198 
199 /// Represents an AvailableValue which can be rematerialized at the end of
200 /// the associated BasicBlock.
202  /// BB - The basic block in question.
204 
205  /// AV - The actual available value
207 
210  Res.BB = BB;
211  Res.AV = std::move(AV);
212  return Res;
213  }
214 
216  unsigned Offset = 0) {
217  return get(BB, AvailableValue::get(V, Offset));
218  }
220  return get(BB, AvailableValue::getUndef());
221  }
222 
223  /// Emit code at the end of this block to adjust the value defined here to
224  /// the specified type. This handles various coercion cases.
226  return AV.MaterializeAdjustedValue(LI, BB->getTerminator(), gvn);
227  }
228 };
229 
230 //===----------------------------------------------------------------------===//
231 // ValueTable Internal Functions
232 //===----------------------------------------------------------------------===//
233 
234 GVN::Expression GVN::ValueTable::createExpr(Instruction *I) {
235  Expression e;
236  e.type = I->getType();
237  e.opcode = I->getOpcode();
238  for (Instruction::op_iterator OI = I->op_begin(), OE = I->op_end();
239  OI != OE; ++OI)
240  e.varargs.push_back(lookupOrAdd(*OI));
241  if (I->isCommutative()) {
242  // Ensure that commutative instructions that only differ by a permutation
243  // of their operands get the same value number by sorting the operand value
244  // numbers. Since all commutative instructions have two operands it is more
245  // efficient to sort by hand rather than using, say, std::sort.
246  assert(I->getNumOperands() == 2 && "Unsupported commutative instruction!");
247  if (e.varargs[0] > e.varargs[1])
248  std::swap(e.varargs[0], e.varargs[1]);
249  }
250 
251  if (CmpInst *C = dyn_cast<CmpInst>(I)) {
252  // Sort the operand value numbers so x<y and y>x get the same value number.
253  CmpInst::Predicate Predicate = C->getPredicate();
254  if (e.varargs[0] > e.varargs[1]) {
255  std::swap(e.varargs[0], e.varargs[1]);
256  Predicate = CmpInst::getSwappedPredicate(Predicate);
257  }
258  e.opcode = (C->getOpcode() << 8) | Predicate;
259  } else if (InsertValueInst *E = dyn_cast<InsertValueInst>(I)) {
260  for (InsertValueInst::idx_iterator II = E->idx_begin(), IE = E->idx_end();
261  II != IE; ++II)
262  e.varargs.push_back(*II);
263  }
264 
265  return e;
266 }
267 
268 GVN::Expression GVN::ValueTable::createCmpExpr(unsigned Opcode,
270  Value *LHS, Value *RHS) {
271  assert((Opcode == Instruction::ICmp || Opcode == Instruction::FCmp) &&
272  "Not a comparison!");
273  Expression e;
274  e.type = CmpInst::makeCmpResultType(LHS->getType());
275  e.varargs.push_back(lookupOrAdd(LHS));
276  e.varargs.push_back(lookupOrAdd(RHS));
277 
278  // Sort the operand value numbers so x<y and y>x get the same value number.
279  if (e.varargs[0] > e.varargs[1]) {
280  std::swap(e.varargs[0], e.varargs[1]);
281  Predicate = CmpInst::getSwappedPredicate(Predicate);
282  }
283  e.opcode = (Opcode << 8) | Predicate;
284  return e;
285 }
286 
287 GVN::Expression GVN::ValueTable::createExtractvalueExpr(ExtractValueInst *EI) {
288  assert(EI && "Not an ExtractValueInst?");
289  Expression e;
290  e.type = EI->getType();
291  e.opcode = 0;
292 
294  if (I != nullptr && EI->getNumIndices() == 1 && *EI->idx_begin() == 0 ) {
295  // EI might be an extract from one of our recognised intrinsics. If it
296  // is we'll synthesize a semantically equivalent expression instead on
297  // an extract value expression.
298  switch (I->getIntrinsicID()) {
299  case Intrinsic::sadd_with_overflow:
300  case Intrinsic::uadd_with_overflow:
301  e.opcode = Instruction::Add;
302  break;
303  case Intrinsic::ssub_with_overflow:
304  case Intrinsic::usub_with_overflow:
305  e.opcode = Instruction::Sub;
306  break;
307  case Intrinsic::smul_with_overflow:
308  case Intrinsic::umul_with_overflow:
309  e.opcode = Instruction::Mul;
310  break;
311  default:
312  break;
313  }
314 
315  if (e.opcode != 0) {
316  // Intrinsic recognized. Grab its args to finish building the expression.
317  assert(I->getNumArgOperands() == 2 &&
318  "Expect two args for recognised intrinsics.");
319  e.varargs.push_back(lookupOrAdd(I->getArgOperand(0)));
320  e.varargs.push_back(lookupOrAdd(I->getArgOperand(1)));
321  return e;
322  }
323  }
324 
325  // Not a recognised intrinsic. Fall back to producing an extract value
326  // expression.
327  e.opcode = EI->getOpcode();
328  for (Instruction::op_iterator OI = EI->op_begin(), OE = EI->op_end();
329  OI != OE; ++OI)
330  e.varargs.push_back(lookupOrAdd(*OI));
331 
332  for (ExtractValueInst::idx_iterator II = EI->idx_begin(), IE = EI->idx_end();
333  II != IE; ++II)
334  e.varargs.push_back(*II);
335 
336  return e;
337 }
338 
339 //===----------------------------------------------------------------------===//
340 // ValueTable External Functions
341 //===----------------------------------------------------------------------===//
342 
343 GVN::ValueTable::ValueTable() : nextValueNumber(1) {}
344 GVN::ValueTable::ValueTable(const ValueTable &) = default;
346 GVN::ValueTable::~ValueTable() = default;
347 
348 /// add - Insert a value into the table with a specified value number.
350  valueNumbering.insert(std::make_pair(V, num));
351 }
352 
353 uint32_t GVN::ValueTable::lookupOrAddCall(CallInst *C) {
354  if (AA->doesNotAccessMemory(C)) {
355  Expression exp = createExpr(C);
356  uint32_t &e = expressionNumbering[exp];
357  if (!e) e = nextValueNumber++;
358  valueNumbering[C] = e;
359  return e;
360  } else if (AA->onlyReadsMemory(C)) {
361  Expression exp = createExpr(C);
362  uint32_t &e = expressionNumbering[exp];
363  if (!e) {
364  e = nextValueNumber++;
365  valueNumbering[C] = e;
366  return e;
367  }
368  if (!MD) {
369  e = nextValueNumber++;
370  valueNumbering[C] = e;
371  return e;
372  }
373 
374  MemDepResult local_dep = MD->getDependency(C);
375 
376  if (!local_dep.isDef() && !local_dep.isNonLocal()) {
377  valueNumbering[C] = nextValueNumber;
378  return nextValueNumber++;
379  }
380 
381  if (local_dep.isDef()) {
382  CallInst* local_cdep = cast<CallInst>(local_dep.getInst());
383 
384  if (local_cdep->getNumArgOperands() != C->getNumArgOperands()) {
385  valueNumbering[C] = nextValueNumber;
386  return nextValueNumber++;
387  }
388 
389  for (unsigned i = 0, e = C->getNumArgOperands(); i < e; ++i) {
390  uint32_t c_vn = lookupOrAdd(C->getArgOperand(i));
391  uint32_t cd_vn = lookupOrAdd(local_cdep->getArgOperand(i));
392  if (c_vn != cd_vn) {
393  valueNumbering[C] = nextValueNumber;
394  return nextValueNumber++;
395  }
396  }
397 
398  uint32_t v = lookupOrAdd(local_cdep);
399  valueNumbering[C] = v;
400  return v;
401  }
402 
403  // Non-local case.
406  // FIXME: Move the checking logic to MemDep!
407  CallInst* cdep = nullptr;
408 
409  // Check to see if we have a single dominating call instruction that is
410  // identical to C.
411  for (unsigned i = 0, e = deps.size(); i != e; ++i) {
412  const NonLocalDepEntry *I = &deps[i];
413  if (I->getResult().isNonLocal())
414  continue;
415 
416  // We don't handle non-definitions. If we already have a call, reject
417  // instruction dependencies.
418  if (!I->getResult().isDef() || cdep != nullptr) {
419  cdep = nullptr;
420  break;
421  }
422 
423  CallInst *NonLocalDepCall = dyn_cast<CallInst>(I->getResult().getInst());
424  // FIXME: All duplicated with non-local case.
425  if (NonLocalDepCall && DT->properlyDominates(I->getBB(), C->getParent())){
426  cdep = NonLocalDepCall;
427  continue;
428  }
429 
430  cdep = nullptr;
431  break;
432  }
433 
434  if (!cdep) {
435  valueNumbering[C] = nextValueNumber;
436  return nextValueNumber++;
437  }
438 
439  if (cdep->getNumArgOperands() != C->getNumArgOperands()) {
440  valueNumbering[C] = nextValueNumber;
441  return nextValueNumber++;
442  }
443  for (unsigned i = 0, e = C->getNumArgOperands(); i < e; ++i) {
444  uint32_t c_vn = lookupOrAdd(C->getArgOperand(i));
445  uint32_t cd_vn = lookupOrAdd(cdep->getArgOperand(i));
446  if (c_vn != cd_vn) {
447  valueNumbering[C] = nextValueNumber;
448  return nextValueNumber++;
449  }
450  }
451 
452  uint32_t v = lookupOrAdd(cdep);
453  valueNumbering[C] = v;
454  return v;
455 
456  } else {
457  valueNumbering[C] = nextValueNumber;
458  return nextValueNumber++;
459  }
460 }
461 
462 /// Returns true if a value number exists for the specified value.
463 bool GVN::ValueTable::exists(Value *V) const { return valueNumbering.count(V) != 0; }
464 
465 /// lookup_or_add - Returns the value number for the specified value, assigning
466 /// it a new number if it did not have one before.
468  DenseMap<Value*, uint32_t>::iterator VI = valueNumbering.find(V);
469  if (VI != valueNumbering.end())
470  return VI->second;
471 
472  if (!isa<Instruction>(V)) {
473  valueNumbering[V] = nextValueNumber;
474  return nextValueNumber++;
475  }
476 
477  Instruction* I = cast<Instruction>(V);
478  Expression exp;
479  switch (I->getOpcode()) {
480  case Instruction::Call:
481  return lookupOrAddCall(cast<CallInst>(I));
482  case Instruction::Add:
483  case Instruction::FAdd:
484  case Instruction::Sub:
485  case Instruction::FSub:
486  case Instruction::Mul:
487  case Instruction::FMul:
488  case Instruction::UDiv:
489  case Instruction::SDiv:
490  case Instruction::FDiv:
491  case Instruction::URem:
492  case Instruction::SRem:
493  case Instruction::FRem:
494  case Instruction::Shl:
495  case Instruction::LShr:
496  case Instruction::AShr:
497  case Instruction::And:
498  case Instruction::Or:
499  case Instruction::Xor:
500  case Instruction::ICmp:
501  case Instruction::FCmp:
502  case Instruction::Trunc:
503  case Instruction::ZExt:
504  case Instruction::SExt:
505  case Instruction::FPToUI:
506  case Instruction::FPToSI:
507  case Instruction::UIToFP:
508  case Instruction::SIToFP:
509  case Instruction::FPTrunc:
510  case Instruction::FPExt:
511  case Instruction::PtrToInt:
512  case Instruction::IntToPtr:
513  case Instruction::BitCast:
514  case Instruction::Select:
515  case Instruction::ExtractElement:
516  case Instruction::InsertElement:
517  case Instruction::ShuffleVector:
518  case Instruction::InsertValue:
519  case Instruction::GetElementPtr:
520  exp = createExpr(I);
521  break;
522  case Instruction::ExtractValue:
523  exp = createExtractvalueExpr(cast<ExtractValueInst>(I));
524  break;
525  default:
526  valueNumbering[V] = nextValueNumber;
527  return nextValueNumber++;
528  }
529 
530  uint32_t& e = expressionNumbering[exp];
531  if (!e) e = nextValueNumber++;
532  valueNumbering[V] = e;
533  return e;
534 }
535 
536 /// Returns the value number of the specified value. Fails if
537 /// the value has not yet been numbered.
539  DenseMap<Value*, uint32_t>::const_iterator VI = valueNumbering.find(V);
540  assert(VI != valueNumbering.end() && "Value not numbered?");
541  return VI->second;
542 }
543 
544 /// Returns the value number of the given comparison,
545 /// assigning it a new number if it did not have one before. Useful when
546 /// we deduced the result of a comparison, but don't immediately have an
547 /// instruction realizing that comparison to hand.
549  CmpInst::Predicate Predicate,
550  Value *LHS, Value *RHS) {
551  Expression exp = createCmpExpr(Opcode, Predicate, LHS, RHS);
552  uint32_t& e = expressionNumbering[exp];
553  if (!e) e = nextValueNumber++;
554  return e;
555 }
556 
557 /// Remove all entries from the ValueTable.
559  valueNumbering.clear();
560  expressionNumbering.clear();
561  nextValueNumber = 1;
562 }
563 
564 /// Remove a value from the value numbering.
566  valueNumbering.erase(V);
567 }
568 
569 /// verifyRemoved - Verify that the value is removed from all internal data
570 /// structures.
571 void GVN::ValueTable::verifyRemoved(const Value *V) const {
573  I = valueNumbering.begin(), E = valueNumbering.end(); I != E; ++I) {
574  assert(I->first != V && "Inst still occurs in value numbering map!");
575  }
576 }
577 
578 //===----------------------------------------------------------------------===//
579 // GVN Pass
580 //===----------------------------------------------------------------------===//
581 
583  // FIXME: The order of evaluation of these 'getResult' calls is very
584  // significant! Re-ordering these variables will cause GVN when run alone to
585  // be less effective! We should fix memdep and basic-aa to not exhibit this
586  // behavior, but until then don't change the order here.
587  auto &AC = AM.getResult<AssumptionAnalysis>(F);
588  auto &DT = AM.getResult<DominatorTreeAnalysis>(F);
589  auto &TLI = AM.getResult<TargetLibraryAnalysis>(F);
590  auto &AA = AM.getResult<AAManager>(F);
591  auto &MemDep = AM.getResult<MemoryDependenceAnalysis>(F);
592  auto *LI = AM.getCachedResult<LoopAnalysis>(F);
594  bool Changed = runImpl(F, AC, DT, TLI, AA, &MemDep, LI, &ORE);
595  if (!Changed)
596  return PreservedAnalyses::all();
599  PA.preserve<GlobalsAA>();
601  return PA;
602 }
603 
604 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
605 LLVM_DUMP_METHOD void GVN::dump(DenseMap<uint32_t, Value*>& d) const {
606  errs() << "{\n";
608  E = d.end(); I != E; ++I) {
609  errs() << I->first << "\n";
610  I->second->dump();
611  }
612  errs() << "}\n";
613 }
614 #endif
615 
616 /// Return true if we can prove that the value
617 /// we're analyzing is fully available in the specified block. As we go, keep
618 /// track of which blocks we know are fully alive in FullyAvailableBlocks. This
619 /// map is actually a tri-state map with the following values:
620 /// 0) we know the block *is not* fully available.
621 /// 1) we know the block *is* fully available.
622 /// 2) we do not know whether the block is fully available or not, but we are
623 /// currently speculating that it will be.
624 /// 3) we are speculating for this block and have used that to speculate for
625 /// other blocks.
627  DenseMap<BasicBlock*, char> &FullyAvailableBlocks,
628  uint32_t RecurseDepth) {
629  if (RecurseDepth > MaxRecurseDepth)
630  return false;
631 
632  // Optimistically assume that the block is fully available and check to see
633  // if we already know about this block in one lookup.
634  std::pair<DenseMap<BasicBlock*, char>::iterator, char> IV =
635  FullyAvailableBlocks.insert(std::make_pair(BB, 2));
636 
637  // If the entry already existed for this block, return the precomputed value.
638  if (!IV.second) {
639  // If this is a speculative "available" value, mark it as being used for
640  // speculation of other blocks.
641  if (IV.first->second == 2)
642  IV.first->second = 3;
643  return IV.first->second != 0;
644  }
645 
646  // Otherwise, see if it is fully available in all predecessors.
647  pred_iterator PI = pred_begin(BB), PE = pred_end(BB);
648 
649  // If this block has no predecessors, it isn't live-in here.
650  if (PI == PE)
651  goto SpeculationFailure;
652 
653  for (; PI != PE; ++PI)
654  // If the value isn't fully available in one of our predecessors, then it
655  // isn't fully available in this block either. Undo our previous
656  // optimistic assumption and bail out.
657  if (!IsValueFullyAvailableInBlock(*PI, FullyAvailableBlocks,RecurseDepth+1))
658  goto SpeculationFailure;
659 
660  return true;
661 
662 // If we get here, we found out that this is not, after
663 // all, a fully-available block. We have a problem if we speculated on this and
664 // used the speculation to mark other blocks as available.
665 SpeculationFailure:
666  char &BBVal = FullyAvailableBlocks[BB];
667 
668  // If we didn't speculate on this, just return with it set to false.
669  if (BBVal == 2) {
670  BBVal = 0;
671  return false;
672  }
673 
674  // If we did speculate on this value, we could have blocks set to 1 that are
675  // incorrect. Walk the (transitive) successors of this block and mark them as
676  // 0 if set to one.
677  SmallVector<BasicBlock*, 32> BBWorklist;
678  BBWorklist.push_back(BB);
679 
680  do {
681  BasicBlock *Entry = BBWorklist.pop_back_val();
682  // Note that this sets blocks to 0 (unavailable) if they happen to not
683  // already be in FullyAvailableBlocks. This is safe.
684  char &EntryVal = FullyAvailableBlocks[Entry];
685  if (EntryVal == 0) continue; // Already unavailable.
686 
687  // Mark as unavailable.
688  EntryVal = 0;
689 
690  BBWorklist.append(succ_begin(Entry), succ_end(Entry));
691  } while (!BBWorklist.empty());
692 
693  return false;
694 }
695 
696 
697 
698 
699 /// Given a set of loads specified by ValuesPerBlock,
700 /// construct SSA form, allowing us to eliminate LI. This returns the value
701 /// that should be used at LI's definition site.
704  GVN &gvn) {
705  // Check for the fully redundant, dominating load case. In this case, we can
706  // just use the dominating value directly.
707  if (ValuesPerBlock.size() == 1 &&
708  gvn.getDominatorTree().properlyDominates(ValuesPerBlock[0].BB,
709  LI->getParent())) {
710  assert(!ValuesPerBlock[0].AV.isUndefValue() &&
711  "Dead BB dominate this block");
712  return ValuesPerBlock[0].MaterializeAdjustedValue(LI, gvn);
713  }
714 
715  // Otherwise, we have to construct SSA form.
716  SmallVector<PHINode*, 8> NewPHIs;
717  SSAUpdater SSAUpdate(&NewPHIs);
718  SSAUpdate.Initialize(LI->getType(), LI->getName());
719 
720  for (const AvailableValueInBlock &AV : ValuesPerBlock) {
721  BasicBlock *BB = AV.BB;
722 
723  if (SSAUpdate.HasValueForBlock(BB))
724  continue;
725 
726  SSAUpdate.AddAvailableValue(BB, AV.MaterializeAdjustedValue(LI, gvn));
727  }
728 
729  // Perform PHI construction.
730  return SSAUpdate.GetValueInMiddleOfBlock(LI->getParent());
731 }
732 
733 Value *AvailableValue::MaterializeAdjustedValue(LoadInst *LI,
734  Instruction *InsertPt,
735  GVN &gvn) const {
736  Value *Res;
737  Type *LoadTy = LI->getType();
738  const DataLayout &DL = LI->getModule()->getDataLayout();
739  if (isSimpleValue()) {
740  Res = getSimpleValue();
741  if (Res->getType() != LoadTy) {
742  Res = getStoreValueForLoad(Res, Offset, LoadTy, InsertPt, DL);
743 
744  DEBUG(dbgs() << "GVN COERCED NONLOCAL VAL:\nOffset: " << Offset << " "
745  << *getSimpleValue() << '\n'
746  << *Res << '\n' << "\n\n\n");
747  }
748  } else if (isCoercedLoadValue()) {
749  LoadInst *Load = getCoercedLoadValue();
750  if (Load->getType() == LoadTy && Offset == 0) {
751  Res = Load;
752  } else {
753  Res = getLoadValueForLoad(Load, Offset, LoadTy, InsertPt, DL);
754  // We would like to use gvn.markInstructionForDeletion here, but we can't
755  // because the load is already memoized into the leader map table that GVN
756  // tracks. It is potentially possible to remove the load from the table,
757  // but then there all of the operations based on it would need to be
758  // rehashed. Just leave the dead load around.
759  gvn.getMemDep().removeInstruction(Load);
760  DEBUG(dbgs() << "GVN COERCED NONLOCAL LOAD:\nOffset: " << Offset << " "
761  << *getCoercedLoadValue() << '\n'
762  << *Res << '\n'
763  << "\n\n\n");
764  }
765  } else if (isMemIntrinValue()) {
766  Res = getMemInstValueForLoad(getMemIntrinValue(), Offset, LoadTy,
767  InsertPt, DL);
768  DEBUG(dbgs() << "GVN COERCED NONLOCAL MEM INTRIN:\nOffset: " << Offset
769  << " " << *getMemIntrinValue() << '\n'
770  << *Res << '\n' << "\n\n\n");
771  } else {
772  assert(isUndefValue() && "Should be UndefVal");
773  DEBUG(dbgs() << "GVN COERCED NONLOCAL Undef:\n";);
774  return UndefValue::get(LoadTy);
775  }
776  assert(Res && "failed to materialize?");
777  return Res;
778 }
779 
780 static bool isLifetimeStart(const Instruction *Inst) {
781  if (const IntrinsicInst* II = dyn_cast<IntrinsicInst>(Inst))
782  return II->getIntrinsicID() == Intrinsic::lifetime_start;
783  return false;
784 }
785 
786 /// \brief Try to locate the three instruction involved in a missed
787 /// load-elimination case that is due to an intervening store.
789  DominatorTree *DT,
791  using namespace ore;
792  User *OtherAccess = nullptr;
793 
794  OptimizationRemarkMissed R(DEBUG_TYPE, "LoadClobbered", LI);
795  R << "load of type " << NV("Type", LI->getType()) << " not eliminated"
796  << setExtraArgs();
797 
798  for (auto *U : LI->getPointerOperand()->users())
799  if (U != LI && (isa<LoadInst>(U) || isa<StoreInst>(U)) &&
800  DT->dominates(cast<Instruction>(U), LI)) {
801  // FIXME: for now give up if there are multiple memory accesses that
802  // dominate the load. We need further analysis to decide which one is
803  // that we're forwarding from.
804  if (OtherAccess)
805  OtherAccess = nullptr;
806  else
807  OtherAccess = U;
808  }
809 
810  if (OtherAccess)
811  R << " in favor of " << NV("OtherAccess", OtherAccess);
812 
813  R << " because it is clobbered by " << NV("ClobberedBy", DepInfo.getInst());
814 
815  ORE->emit(R);
816 }
817 
818 bool GVN::AnalyzeLoadAvailability(LoadInst *LI, MemDepResult DepInfo,
819  Value *Address, AvailableValue &Res) {
820 
821  assert((DepInfo.isDef() || DepInfo.isClobber()) &&
822  "expected a local dependence");
823  assert(LI->isUnordered() && "rules below are incorrect for ordered access");
824 
825  const DataLayout &DL = LI->getModule()->getDataLayout();
826 
827  if (DepInfo.isClobber()) {
828  // If the dependence is to a store that writes to a superset of the bits
829  // read by the load, we can extract the bits we need for the load from the
830  // stored value.
831  if (StoreInst *DepSI = dyn_cast<StoreInst>(DepInfo.getInst())) {
832  // Can't forward from non-atomic to atomic without violating memory model.
833  if (Address && LI->isAtomic() <= DepSI->isAtomic()) {
834  int Offset =
836  if (Offset != -1) {
837  Res = AvailableValue::get(DepSI->getValueOperand(), Offset);
838  return true;
839  }
840  }
841  }
842 
843  // Check to see if we have something like this:
844  // load i32* P
845  // load i8* (P+1)
846  // if we have this, replace the later with an extraction from the former.
847  if (LoadInst *DepLI = dyn_cast<LoadInst>(DepInfo.getInst())) {
848  // If this is a clobber and L is the first instruction in its block, then
849  // we have the first instruction in the entry block.
850  // Can't forward from non-atomic to atomic without violating memory model.
851  if (DepLI != LI && Address && LI->isAtomic() <= DepLI->isAtomic()) {
852  int Offset =
853  analyzeLoadFromClobberingLoad(LI->getType(), Address, DepLI, DL);
854 
855  if (Offset != -1) {
856  Res = AvailableValue::getLoad(DepLI, Offset);
857  return true;
858  }
859  }
860  }
861 
862  // If the clobbering value is a memset/memcpy/memmove, see if we can
863  // forward a value on from it.
864  if (MemIntrinsic *DepMI = dyn_cast<MemIntrinsic>(DepInfo.getInst())) {
865  if (Address && !LI->isAtomic()) {
867  DepMI, DL);
868  if (Offset != -1) {
869  Res = AvailableValue::getMI(DepMI, Offset);
870  return true;
871  }
872  }
873  }
874  // Nothing known about this clobber, have to be conservative
875  DEBUG(
876  // fast print dep, using operator<< on instruction is too slow.
877  dbgs() << "GVN: load ";
878  LI->printAsOperand(dbgs());
879  Instruction *I = DepInfo.getInst();
880  dbgs() << " is clobbered by " << *I << '\n';
881  );
882 
883  if (ORE->allowExtraAnalysis())
884  reportMayClobberedLoad(LI, DepInfo, DT, ORE);
885 
886  return false;
887  }
888  assert(DepInfo.isDef() && "follows from above");
889 
890  Instruction *DepInst = DepInfo.getInst();
891 
892  // Loading the allocation -> undef.
893  if (isa<AllocaInst>(DepInst) || isMallocLikeFn(DepInst, TLI) ||
894  // Loading immediately after lifetime begin -> undef.
895  isLifetimeStart(DepInst)) {
896  Res = AvailableValue::get(UndefValue::get(LI->getType()));
897  return true;
898  }
899 
900  // Loading from calloc (which zero initializes memory) -> zero
901  if (isCallocLikeFn(DepInst, TLI)) {
902  Res = AvailableValue::get(Constant::getNullValue(LI->getType()));
903  return true;
904  }
905 
906  if (StoreInst *S = dyn_cast<StoreInst>(DepInst)) {
907  // Reject loads and stores that are to the same address but are of
908  // different types if we have to. If the stored value is larger or equal to
909  // the loaded value, we can reuse it.
910  if (S->getValueOperand()->getType() != LI->getType() &&
911  !canCoerceMustAliasedValueToLoad(S->getValueOperand(),
912  LI->getType(), DL))
913  return false;
914 
915  // Can't forward from non-atomic to atomic without violating memory model.
916  if (S->isAtomic() < LI->isAtomic())
917  return false;
918 
919  Res = AvailableValue::get(S->getValueOperand());
920  return true;
921  }
922 
923  if (LoadInst *LD = dyn_cast<LoadInst>(DepInst)) {
924  // If the types mismatch and we can't handle it, reject reuse of the load.
925  // If the stored value is larger or equal to the loaded value, we can reuse
926  // it.
927  if (LD->getType() != LI->getType() &&
929  return false;
930 
931  // Can't forward from non-atomic to atomic without violating memory model.
932  if (LD->isAtomic() < LI->isAtomic())
933  return false;
934 
935  Res = AvailableValue::getLoad(LD);
936  return true;
937  }
938 
939  // Unknown def - must be conservative
940  DEBUG(
941  // fast print dep, using operator<< on instruction is too slow.
942  dbgs() << "GVN: load ";
943  LI->printAsOperand(dbgs());
944  dbgs() << " has unknown def " << *DepInst << '\n';
945  );
946  return false;
947 }
948 
949 void GVN::AnalyzeLoadAvailability(LoadInst *LI, LoadDepVect &Deps,
950  AvailValInBlkVect &ValuesPerBlock,
951  UnavailBlkVect &UnavailableBlocks) {
952 
953  // Filter out useless results (non-locals, etc). Keep track of the blocks
954  // where we have a value available in repl, also keep track of whether we see
955  // dependencies that produce an unknown value for the load (such as a call
956  // that could potentially clobber the load).
957  unsigned NumDeps = Deps.size();
958  for (unsigned i = 0, e = NumDeps; i != e; ++i) {
959  BasicBlock *DepBB = Deps[i].getBB();
960  MemDepResult DepInfo = Deps[i].getResult();
961 
962  if (DeadBlocks.count(DepBB)) {
963  // Dead dependent mem-op disguise as a load evaluating the same value
964  // as the load in question.
965  ValuesPerBlock.push_back(AvailableValueInBlock::getUndef(DepBB));
966  continue;
967  }
968 
969  if (!DepInfo.isDef() && !DepInfo.isClobber()) {
970  UnavailableBlocks.push_back(DepBB);
971  continue;
972  }
973 
974  // The address being loaded in this non-local block may not be the same as
975  // the pointer operand of the load if PHI translation occurs. Make sure
976  // to consider the right address.
977  Value *Address = Deps[i].getAddress();
978 
979  AvailableValue AV;
980  if (AnalyzeLoadAvailability(LI, DepInfo, Address, AV)) {
981  // subtlety: because we know this was a non-local dependency, we know
982  // it's safe to materialize anywhere between the instruction within
983  // DepInfo and the end of it's block.
984  ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB,
985  std::move(AV)));
986  } else {
987  UnavailableBlocks.push_back(DepBB);
988  }
989  }
990 
991  assert(NumDeps == ValuesPerBlock.size() + UnavailableBlocks.size() &&
992  "post condition violation");
993 }
994 
995 bool GVN::PerformLoadPRE(LoadInst *LI, AvailValInBlkVect &ValuesPerBlock,
996  UnavailBlkVect &UnavailableBlocks) {
997  // Okay, we have *some* definitions of the value. This means that the value
998  // is available in some of our (transitive) predecessors. Lets think about
999  // doing PRE of this load. This will involve inserting a new load into the
1000  // predecessor when it's not available. We could do this in general, but
1001  // prefer to not increase code size. As such, we only do this when we know
1002  // that we only have to insert *one* load (which means we're basically moving
1003  // the load, not inserting a new one).
1004 
1005  SmallPtrSet<BasicBlock *, 4> Blockers(UnavailableBlocks.begin(),
1006  UnavailableBlocks.end());
1007 
1008  // Let's find the first basic block with more than one predecessor. Walk
1009  // backwards through predecessors if needed.
1010  BasicBlock *LoadBB = LI->getParent();
1011  BasicBlock *TmpBB = LoadBB;
1012 
1013  while (TmpBB->getSinglePredecessor()) {
1014  TmpBB = TmpBB->getSinglePredecessor();
1015  if (TmpBB == LoadBB) // Infinite (unreachable) loop.
1016  return false;
1017  if (Blockers.count(TmpBB))
1018  return false;
1019 
1020  // If any of these blocks has more than one successor (i.e. if the edge we
1021  // just traversed was critical), then there are other paths through this
1022  // block along which the load may not be anticipated. Hoisting the load
1023  // above this block would be adding the load to execution paths along
1024  // which it was not previously executed.
1025  if (TmpBB->getTerminator()->getNumSuccessors() != 1)
1026  return false;
1027  }
1028 
1029  assert(TmpBB);
1030  LoadBB = TmpBB;
1031 
1032  // Check to see how many predecessors have the loaded value fully
1033  // available.
1035  DenseMap<BasicBlock*, char> FullyAvailableBlocks;
1036  for (const AvailableValueInBlock &AV : ValuesPerBlock)
1037  FullyAvailableBlocks[AV.BB] = true;
1038  for (BasicBlock *UnavailableBB : UnavailableBlocks)
1039  FullyAvailableBlocks[UnavailableBB] = false;
1040 
1041  SmallVector<BasicBlock *, 4> CriticalEdgePred;
1042  for (BasicBlock *Pred : predecessors(LoadBB)) {
1043  // If any predecessor block is an EH pad that does not allow non-PHI
1044  // instructions before the terminator, we can't PRE the load.
1045  if (Pred->getTerminator()->isEHPad()) {
1046  DEBUG(dbgs()
1047  << "COULD NOT PRE LOAD BECAUSE OF AN EH PAD PREDECESSOR '"
1048  << Pred->getName() << "': " << *LI << '\n');
1049  return false;
1050  }
1051 
1052  if (IsValueFullyAvailableInBlock(Pred, FullyAvailableBlocks, 0)) {
1053  continue;
1054  }
1055 
1056  if (Pred->getTerminator()->getNumSuccessors() != 1) {
1057  if (isa<IndirectBrInst>(Pred->getTerminator())) {
1058  DEBUG(dbgs() << "COULD NOT PRE LOAD BECAUSE OF INDBR CRITICAL EDGE '"
1059  << Pred->getName() << "': " << *LI << '\n');
1060  return false;
1061  }
1062 
1063  if (LoadBB->isEHPad()) {
1064  DEBUG(dbgs()
1065  << "COULD NOT PRE LOAD BECAUSE OF AN EH PAD CRITICAL EDGE '"
1066  << Pred->getName() << "': " << *LI << '\n');
1067  return false;
1068  }
1069 
1070  CriticalEdgePred.push_back(Pred);
1071  } else {
1072  // Only add the predecessors that will not be split for now.
1073  PredLoads[Pred] = nullptr;
1074  }
1075  }
1076 
1077  // Decide whether PRE is profitable for this load.
1078  unsigned NumUnavailablePreds = PredLoads.size() + CriticalEdgePred.size();
1079  assert(NumUnavailablePreds != 0 &&
1080  "Fully available value should already be eliminated!");
1081 
1082  // If this load is unavailable in multiple predecessors, reject it.
1083  // FIXME: If we could restructure the CFG, we could make a common pred with
1084  // all the preds that don't have an available LI and insert a new load into
1085  // that one block.
1086  if (NumUnavailablePreds != 1)
1087  return false;
1088 
1089  // Split critical edges, and update the unavailable predecessors accordingly.
1090  for (BasicBlock *OrigPred : CriticalEdgePred) {
1091  BasicBlock *NewPred = splitCriticalEdges(OrigPred, LoadBB);
1092  assert(!PredLoads.count(OrigPred) && "Split edges shouldn't be in map!");
1093  PredLoads[NewPred] = nullptr;
1094  DEBUG(dbgs() << "Split critical edge " << OrigPred->getName() << "->"
1095  << LoadBB->getName() << '\n');
1096  }
1097 
1098  // Check if the load can safely be moved to all the unavailable predecessors.
1099  bool CanDoPRE = true;
1100  const DataLayout &DL = LI->getModule()->getDataLayout();
1102  for (auto &PredLoad : PredLoads) {
1103  BasicBlock *UnavailablePred = PredLoad.first;
1104 
1105  // Do PHI translation to get its value in the predecessor if necessary. The
1106  // returned pointer (if non-null) is guaranteed to dominate UnavailablePred.
1107 
1108  // If all preds have a single successor, then we know it is safe to insert
1109  // the load on the pred (?!?), so we can insert code to materialize the
1110  // pointer if it is not available.
1111  PHITransAddr Address(LI->getPointerOperand(), DL, AC);
1112  Value *LoadPtr = nullptr;
1113  LoadPtr = Address.PHITranslateWithInsertion(LoadBB, UnavailablePred,
1114  *DT, NewInsts);
1115 
1116  // If we couldn't find or insert a computation of this phi translated value,
1117  // we fail PRE.
1118  if (!LoadPtr) {
1119  DEBUG(dbgs() << "COULDN'T INSERT PHI TRANSLATED VALUE OF: "
1120  << *LI->getPointerOperand() << "\n");
1121  CanDoPRE = false;
1122  break;
1123  }
1124 
1125  PredLoad.second = LoadPtr;
1126  }
1127 
1128  if (!CanDoPRE) {
1129  while (!NewInsts.empty()) {
1130  Instruction *I = NewInsts.pop_back_val();
1131  if (MD) MD->removeInstruction(I);
1132  I->eraseFromParent();
1133  }
1134  // HINT: Don't revert the edge-splitting as following transformation may
1135  // also need to split these critical edges.
1136  return !CriticalEdgePred.empty();
1137  }
1138 
1139  // Okay, we can eliminate this load by inserting a reload in the predecessor
1140  // and using PHI construction to get the value in the other predecessors, do
1141  // it.
1142  DEBUG(dbgs() << "GVN REMOVING PRE LOAD: " << *LI << '\n');
1143  DEBUG(if (!NewInsts.empty())
1144  dbgs() << "INSERTED " << NewInsts.size() << " INSTS: "
1145  << *NewInsts.back() << '\n');
1146 
1147  // Assign value numbers to the new instructions.
1148  for (Instruction *I : NewInsts) {
1149  // Instructions that have been inserted in predecessor(s) to materialize
1150  // the load address do not retain their original debug locations. Doing
1151  // so could lead to confusing (but correct) source attributions.
1152  // FIXME: How do we retain source locations without causing poor debugging
1153  // behavior?
1154  I->setDebugLoc(DebugLoc());
1155 
1156  // FIXME: We really _ought_ to insert these value numbers into their
1157  // parent's availability map. However, in doing so, we risk getting into
1158  // ordering issues. If a block hasn't been processed yet, we would be
1159  // marking a value as AVAIL-IN, which isn't what we intend.
1160  VN.lookupOrAdd(I);
1161  }
1162 
1163  for (const auto &PredLoad : PredLoads) {
1164  BasicBlock *UnavailablePred = PredLoad.first;
1165  Value *LoadPtr = PredLoad.second;
1166 
1167  auto *NewLoad = new LoadInst(LoadPtr, LI->getName()+".pre",
1168  LI->isVolatile(), LI->getAlignment(),
1169  LI->getOrdering(), LI->getSyncScopeID(),
1170  UnavailablePred->getTerminator());
1171  NewLoad->setDebugLoc(LI->getDebugLoc());
1172 
1173  // Transfer the old load's AA tags to the new load.
1174  AAMDNodes Tags;
1175  LI->getAAMetadata(Tags);
1176  if (Tags)
1177  NewLoad->setAAMetadata(Tags);
1178 
1179  if (auto *MD = LI->getMetadata(LLVMContext::MD_invariant_load))
1180  NewLoad->setMetadata(LLVMContext::MD_invariant_load, MD);
1181  if (auto *InvGroupMD = LI->getMetadata(LLVMContext::MD_invariant_group))
1182  NewLoad->setMetadata(LLVMContext::MD_invariant_group, InvGroupMD);
1183  if (auto *RangeMD = LI->getMetadata(LLVMContext::MD_range))
1184  NewLoad->setMetadata(LLVMContext::MD_range, RangeMD);
1185 
1186  // We do not propagate the old load's debug location, because the new
1187  // load now lives in a different BB, and we want to avoid a jumpy line
1188  // table.
1189  // FIXME: How do we retain source locations without causing poor debugging
1190  // behavior?
1191 
1192  // Add the newly created load.
1193  ValuesPerBlock.push_back(AvailableValueInBlock::get(UnavailablePred,
1194  NewLoad));
1195  MD->invalidateCachedPointerInfo(LoadPtr);
1196  DEBUG(dbgs() << "GVN INSERTED " << *NewLoad << '\n');
1197  }
1198 
1199  // Perform PHI construction.
1200  Value *V = ConstructSSAForLoadSet(LI, ValuesPerBlock, *this);
1201  LI->replaceAllUsesWith(V);
1202  if (isa<PHINode>(V))
1203  V->takeName(LI);
1204  if (Instruction *I = dyn_cast<Instruction>(V))
1205  I->setDebugLoc(LI->getDebugLoc());
1206  if (V->getType()->isPtrOrPtrVectorTy())
1207  MD->invalidateCachedPointerInfo(V);
1209  ORE->emit(OptimizationRemark(DEBUG_TYPE, "LoadPRE", LI)
1210  << "load eliminated by PRE");
1211  ++NumPRELoad;
1212  return true;
1213 }
1214 
1217  using namespace ore;
1218  ORE->emit(OptimizationRemark(DEBUG_TYPE, "LoadElim", LI)
1219  << "load of type " << NV("Type", LI->getType()) << " eliminated"
1220  << setExtraArgs() << " in favor of "
1221  << NV("InfavorOfValue", AvailableValue));
1222 }
1223 
1224 /// Attempt to eliminate a load whose dependencies are
1225 /// non-local by performing PHI construction.
1226 bool GVN::processNonLocalLoad(LoadInst *LI) {
1227  // non-local speculations are not allowed under asan.
1228  if (LI->getParent()->getParent()->hasFnAttribute(Attribute::SanitizeAddress))
1229  return false;
1230 
1231  // Step 1: Find the non-local dependencies of the load.
1232  LoadDepVect Deps;
1233  MD->getNonLocalPointerDependency(LI, Deps);
1234 
1235  // If we had to process more than one hundred blocks to find the
1236  // dependencies, this load isn't worth worrying about. Optimizing
1237  // it will be too expensive.
1238  unsigned NumDeps = Deps.size();
1239  if (NumDeps > 100)
1240  return false;
1241 
1242  // If we had a phi translation failure, we'll have a single entry which is a
1243  // clobber in the current block. Reject this early.
1244  if (NumDeps == 1 &&
1245  !Deps[0].getResult().isDef() && !Deps[0].getResult().isClobber()) {
1246  DEBUG(
1247  dbgs() << "GVN: non-local load ";
1248  LI->printAsOperand(dbgs());
1249  dbgs() << " has unknown dependencies\n";
1250  );
1251  return false;
1252  }
1253 
1254  // If this load follows a GEP, see if we can PRE the indices before analyzing.
1255  if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(LI->getOperand(0))) {
1256  for (GetElementPtrInst::op_iterator OI = GEP->idx_begin(),
1257  OE = GEP->idx_end();
1258  OI != OE; ++OI)
1259  if (Instruction *I = dyn_cast<Instruction>(OI->get()))
1260  performScalarPRE(I);
1261  }
1262 
1263  // Step 2: Analyze the availability of the load
1264  AvailValInBlkVect ValuesPerBlock;
1265  UnavailBlkVect UnavailableBlocks;
1266  AnalyzeLoadAvailability(LI, Deps, ValuesPerBlock, UnavailableBlocks);
1267 
1268  // If we have no predecessors that produce a known value for this load, exit
1269  // early.
1270  if (ValuesPerBlock.empty())
1271  return false;
1272 
1273  // Step 3: Eliminate fully redundancy.
1274  //
1275  // If all of the instructions we depend on produce a known value for this
1276  // load, then it is fully redundant and we can use PHI insertion to compute
1277  // its value. Insert PHIs and remove the fully redundant value now.
1278  if (UnavailableBlocks.empty()) {
1279  DEBUG(dbgs() << "GVN REMOVING NONLOCAL LOAD: " << *LI << '\n');
1280 
1281  // Perform PHI construction.
1282  Value *V = ConstructSSAForLoadSet(LI, ValuesPerBlock, *this);
1283  LI->replaceAllUsesWith(V);
1284 
1285  if (isa<PHINode>(V))
1286  V->takeName(LI);
1287  if (Instruction *I = dyn_cast<Instruction>(V))
1288  // If instruction I has debug info, then we should not update it.
1289  // Also, if I has a null DebugLoc, then it is still potentially incorrect
1290  // to propagate LI's DebugLoc because LI may not post-dominate I.
1291  if (LI->getDebugLoc() && LI->getParent() == I->getParent())
1292  I->setDebugLoc(LI->getDebugLoc());
1293  if (V->getType()->isPtrOrPtrVectorTy())
1294  MD->invalidateCachedPointerInfo(V);
1296  ++NumGVNLoad;
1297  reportLoadElim(LI, V, ORE);
1298  return true;
1299  }
1300 
1301  // Step 4: Eliminate partial redundancy.
1302  if (!EnablePRE || !EnableLoadPRE)
1303  return false;
1304 
1305  return PerformLoadPRE(LI, ValuesPerBlock, UnavailableBlocks);
1306 }
1307 
1308 bool GVN::processAssumeIntrinsic(IntrinsicInst *IntrinsicI) {
1309  assert(IntrinsicI->getIntrinsicID() == Intrinsic::assume &&
1310  "This function can only be called with llvm.assume intrinsic");
1311  Value *V = IntrinsicI->getArgOperand(0);
1312 
1313  if (ConstantInt *Cond = dyn_cast<ConstantInt>(V)) {
1314  if (Cond->isZero()) {
1315  Type *Int8Ty = Type::getInt8Ty(V->getContext());
1316  // Insert a new store to null instruction before the load to indicate that
1317  // this code is not reachable. FIXME: We could insert unreachable
1318  // instruction directly because we can modify the CFG.
1319  new StoreInst(UndefValue::get(Int8Ty),
1321  IntrinsicI);
1322  }
1323  markInstructionForDeletion(IntrinsicI);
1324  return false;
1325  }
1326 
1327  Constant *True = ConstantInt::getTrue(V->getContext());
1328  bool Changed = false;
1329 
1330  for (BasicBlock *Successor : successors(IntrinsicI->getParent())) {
1331  BasicBlockEdge Edge(IntrinsicI->getParent(), Successor);
1332 
1333  // This property is only true in dominated successors, propagateEquality
1334  // will check dominance for us.
1335  Changed |= propagateEquality(V, True, Edge, false);
1336  }
1337 
1338  // We can replace assume value with true, which covers cases like this:
1339  // call void @llvm.assume(i1 %cmp)
1340  // br i1 %cmp, label %bb1, label %bb2 ; will change %cmp to true
1341  ReplaceWithConstMap[V] = True;
1342 
1343  // If one of *cmp *eq operand is const, adding it to map will cover this:
1344  // %cmp = fcmp oeq float 3.000000e+00, %0 ; const on lhs could happen
1345  // call void @llvm.assume(i1 %cmp)
1346  // ret float %0 ; will change it to ret float 3.000000e+00
1347  if (auto *CmpI = dyn_cast<CmpInst>(V)) {
1348  if (CmpI->getPredicate() == CmpInst::Predicate::ICMP_EQ ||
1349  CmpI->getPredicate() == CmpInst::Predicate::FCMP_OEQ ||
1350  (CmpI->getPredicate() == CmpInst::Predicate::FCMP_UEQ &&
1351  CmpI->getFastMathFlags().noNaNs())) {
1352  Value *CmpLHS = CmpI->getOperand(0);
1353  Value *CmpRHS = CmpI->getOperand(1);
1354  if (isa<Constant>(CmpLHS))
1355  std::swap(CmpLHS, CmpRHS);
1356  auto *RHSConst = dyn_cast<Constant>(CmpRHS);
1357 
1358  // If only one operand is constant.
1359  if (RHSConst != nullptr && !isa<Constant>(CmpLHS))
1360  ReplaceWithConstMap[CmpLHS] = RHSConst;
1361  }
1362  }
1363  return Changed;
1364 }
1365 
1367  auto *ReplInst = dyn_cast<Instruction>(Repl);
1368  if (!ReplInst)
1369  return;
1370 
1371  // Patch the replacement so that it is not more restrictive than the value
1372  // being replaced.
1373  // Note that if 'I' is a load being replaced by some operation,
1374  // for example, by an arithmetic operation, then andIRFlags()
1375  // would just erase all math flags from the original arithmetic
1376  // operation, which is clearly not wanted and not needed.
1377  if (!isa<LoadInst>(I))
1378  ReplInst->andIRFlags(I);
1379 
1380  // FIXME: If both the original and replacement value are part of the
1381  // same control-flow region (meaning that the execution of one
1382  // guarantees the execution of the other), then we can combine the
1383  // noalias scopes here and do better than the general conservative
1384  // answer used in combineMetadata().
1385 
1386  // In general, GVN unifies expressions over different control-flow
1387  // regions, and so we need a conservative combination of the noalias
1388  // scopes.
1389  static const unsigned KnownIDs[] = {
1394  combineMetadata(ReplInst, I, KnownIDs);
1395 }
1396 
1398  patchReplacementInstruction(I, Repl);
1399  I->replaceAllUsesWith(Repl);
1400 }
1401 
1402 /// Attempt to eliminate a load, first by eliminating it
1403 /// locally, and then attempting non-local elimination if that fails.
1404 bool GVN::processLoad(LoadInst *L) {
1405  if (!MD)
1406  return false;
1407 
1408  // This code hasn't been audited for ordered or volatile memory access
1409  if (!L->isUnordered())
1410  return false;
1411 
1412  if (L->use_empty()) {
1414  return true;
1415  }
1416 
1417  // ... to a pointer that has been loaded from before...
1418  MemDepResult Dep = MD->getDependency(L);
1419 
1420  // If it is defined in another block, try harder.
1421  if (Dep.isNonLocal())
1422  return processNonLocalLoad(L);
1423 
1424  // Only handle the local case below
1425  if (!Dep.isDef() && !Dep.isClobber()) {
1426  // This might be a NonFuncLocal or an Unknown
1427  DEBUG(
1428  // fast print dep, using operator<< on instruction is too slow.
1429  dbgs() << "GVN: load ";
1430  L->printAsOperand(dbgs());
1431  dbgs() << " has unknown dependence\n";
1432  );
1433  return false;
1434  }
1435 
1436  AvailableValue AV;
1437  if (AnalyzeLoadAvailability(L, Dep, L->getPointerOperand(), AV)) {
1438  Value *AvailableValue = AV.MaterializeAdjustedValue(L, L, *this);
1439 
1440  // Replace the load!
1441  patchAndReplaceAllUsesWith(L, AvailableValue);
1443  ++NumGVNLoad;
1444  reportLoadElim(L, AvailableValue, ORE);
1445  // Tell MDA to rexamine the reused pointer since we might have more
1446  // information after forwarding it.
1447  if (MD && AvailableValue->getType()->isPtrOrPtrVectorTy())
1448  MD->invalidateCachedPointerInfo(AvailableValue);
1449  return true;
1450  }
1451 
1452  return false;
1453 }
1454 
1455 // In order to find a leader for a given value number at a
1456 // specific basic block, we first obtain the list of all Values for that number,
1457 // and then scan the list to find one whose block dominates the block in
1458 // question. This is fast because dominator tree queries consist of only
1459 // a few comparisons of DFS numbers.
1460 Value *GVN::findLeader(const BasicBlock *BB, uint32_t num) {
1461  LeaderTableEntry Vals = LeaderTable[num];
1462  if (!Vals.Val) return nullptr;
1463 
1464  Value *Val = nullptr;
1465  if (DT->dominates(Vals.BB, BB)) {
1466  Val = Vals.Val;
1467  if (isa<Constant>(Val)) return Val;
1468  }
1469 
1470  LeaderTableEntry* Next = Vals.Next;
1471  while (Next) {
1472  if (DT->dominates(Next->BB, BB)) {
1473  if (isa<Constant>(Next->Val)) return Next->Val;
1474  if (!Val) Val = Next->Val;
1475  }
1476 
1477  Next = Next->Next;
1478  }
1479 
1480  return Val;
1481 }
1482 
1483 /// There is an edge from 'Src' to 'Dst'. Return
1484 /// true if every path from the entry block to 'Dst' passes via this edge. In
1485 /// particular 'Dst' must not be reachable via another edge from 'Src'.
1487  DominatorTree *DT) {
1488  // While in theory it is interesting to consider the case in which Dst has
1489  // more than one predecessor, because Dst might be part of a loop which is
1490  // only reachable from Src, in practice it is pointless since at the time
1491  // GVN runs all such loops have preheaders, which means that Dst will have
1492  // been changed to have only one predecessor, namely Src.
1493  const BasicBlock *Pred = E.getEnd()->getSinglePredecessor();
1494  assert((!Pred || Pred == E.getStart()) &&
1495  "No edge between these basic blocks!");
1496  return Pred != nullptr;
1497 }
1498 
1499 // Tries to replace instruction with const, using information from
1500 // ReplaceWithConstMap.
1501 bool GVN::replaceOperandsWithConsts(Instruction *Instr) const {
1502  bool Changed = false;
1503  for (unsigned OpNum = 0; OpNum < Instr->getNumOperands(); ++OpNum) {
1504  Value *Operand = Instr->getOperand(OpNum);
1505  auto it = ReplaceWithConstMap.find(Operand);
1506  if (it != ReplaceWithConstMap.end()) {
1507  assert(!isa<Constant>(Operand) &&
1508  "Replacing constants with constants is invalid");
1509  DEBUG(dbgs() << "GVN replacing: " << *Operand << " with " << *it->second
1510  << " in instruction " << *Instr << '\n');
1511  Instr->setOperand(OpNum, it->second);
1512  Changed = true;
1513  }
1514  }
1515  return Changed;
1516 }
1517 
1518 /// The given values are known to be equal in every block
1519 /// dominated by 'Root'. Exploit this, for example by replacing 'LHS' with
1520 /// 'RHS' everywhere in the scope. Returns whether a change was made.
1521 /// If DominatesByEdge is false, then it means that we will propagate the RHS
1522 /// value starting from the end of Root.Start.
1523 bool GVN::propagateEquality(Value *LHS, Value *RHS, const BasicBlockEdge &Root,
1524  bool DominatesByEdge) {
1526  Worklist.push_back(std::make_pair(LHS, RHS));
1527  bool Changed = false;
1528  // For speed, compute a conservative fast approximation to
1529  // DT->dominates(Root, Root.getEnd());
1530  const bool RootDominatesEnd = isOnlyReachableViaThisEdge(Root, DT);
1531 
1532  while (!Worklist.empty()) {
1533  std::pair<Value*, Value*> Item = Worklist.pop_back_val();
1534  LHS = Item.first; RHS = Item.second;
1535 
1536  if (LHS == RHS)
1537  continue;
1538  assert(LHS->getType() == RHS->getType() && "Equality but unequal types!");
1539 
1540  // Don't try to propagate equalities between constants.
1541  if (isa<Constant>(LHS) && isa<Constant>(RHS))
1542  continue;
1543 
1544  // Prefer a constant on the right-hand side, or an Argument if no constants.
1545  if (isa<Constant>(LHS) || (isa<Argument>(LHS) && !isa<Constant>(RHS)))
1546  std::swap(LHS, RHS);
1547  assert((isa<Argument>(LHS) || isa<Instruction>(LHS)) && "Unexpected value!");
1548 
1549  // If there is no obvious reason to prefer the left-hand side over the
1550  // right-hand side, ensure the longest lived term is on the right-hand side,
1551  // so the shortest lived term will be replaced by the longest lived.
1552  // This tends to expose more simplifications.
1553  uint32_t LVN = VN.lookupOrAdd(LHS);
1554  if ((isa<Argument>(LHS) && isa<Argument>(RHS)) ||
1555  (isa<Instruction>(LHS) && isa<Instruction>(RHS))) {
1556  // Move the 'oldest' value to the right-hand side, using the value number
1557  // as a proxy for age.
1558  uint32_t RVN = VN.lookupOrAdd(RHS);
1559  if (LVN < RVN) {
1560  std::swap(LHS, RHS);
1561  LVN = RVN;
1562  }
1563  }
1564 
1565  // If value numbering later sees that an instruction in the scope is equal
1566  // to 'LHS' then ensure it will be turned into 'RHS'. In order to preserve
1567  // the invariant that instructions only occur in the leader table for their
1568  // own value number (this is used by removeFromLeaderTable), do not do this
1569  // if RHS is an instruction (if an instruction in the scope is morphed into
1570  // LHS then it will be turned into RHS by the next GVN iteration anyway, so
1571  // using the leader table is about compiling faster, not optimizing better).
1572  // The leader table only tracks basic blocks, not edges. Only add to if we
1573  // have the simple case where the edge dominates the end.
1574  if (RootDominatesEnd && !isa<Instruction>(RHS))
1575  addToLeaderTable(LVN, RHS, Root.getEnd());
1576 
1577  // Replace all occurrences of 'LHS' with 'RHS' everywhere in the scope. As
1578  // LHS always has at least one use that is not dominated by Root, this will
1579  // never do anything if LHS has only one use.
1580  if (!LHS->hasOneUse()) {
1581  unsigned NumReplacements =
1582  DominatesByEdge
1583  ? replaceDominatedUsesWith(LHS, RHS, *DT, Root)
1584  : replaceDominatedUsesWith(LHS, RHS, *DT, Root.getStart());
1585 
1586  Changed |= NumReplacements > 0;
1587  NumGVNEqProp += NumReplacements;
1588  }
1589 
1590  // Now try to deduce additional equalities from this one. For example, if
1591  // the known equality was "(A != B)" == "false" then it follows that A and B
1592  // are equal in the scope. Only boolean equalities with an explicit true or
1593  // false RHS are currently supported.
1594  if (!RHS->getType()->isIntegerTy(1))
1595  // Not a boolean equality - bail out.
1596  continue;
1597  ConstantInt *CI = dyn_cast<ConstantInt>(RHS);
1598  if (!CI)
1599  // RHS neither 'true' nor 'false' - bail out.
1600  continue;
1601  // Whether RHS equals 'true'. Otherwise it equals 'false'.
1602  bool isKnownTrue = CI->isMinusOne();
1603  bool isKnownFalse = !isKnownTrue;
1604 
1605  // If "A && B" is known true then both A and B are known true. If "A || B"
1606  // is known false then both A and B are known false.
1607  Value *A, *B;
1608  if ((isKnownTrue && match(LHS, m_And(m_Value(A), m_Value(B)))) ||
1609  (isKnownFalse && match(LHS, m_Or(m_Value(A), m_Value(B))))) {
1610  Worklist.push_back(std::make_pair(A, RHS));
1611  Worklist.push_back(std::make_pair(B, RHS));
1612  continue;
1613  }
1614 
1615  // If we are propagating an equality like "(A == B)" == "true" then also
1616  // propagate the equality A == B. When propagating a comparison such as
1617  // "(A >= B)" == "true", replace all instances of "A < B" with "false".
1618  if (CmpInst *Cmp = dyn_cast<CmpInst>(LHS)) {
1619  Value *Op0 = Cmp->getOperand(0), *Op1 = Cmp->getOperand(1);
1620 
1621  // If "A == B" is known true, or "A != B" is known false, then replace
1622  // A with B everywhere in the scope.
1623  if ((isKnownTrue && Cmp->getPredicate() == CmpInst::ICMP_EQ) ||
1624  (isKnownFalse && Cmp->getPredicate() == CmpInst::ICMP_NE))
1625  Worklist.push_back(std::make_pair(Op0, Op1));
1626 
1627  // Handle the floating point versions of equality comparisons too.
1628  if ((isKnownTrue && Cmp->getPredicate() == CmpInst::FCMP_OEQ) ||
1629  (isKnownFalse && Cmp->getPredicate() == CmpInst::FCMP_UNE)) {
1630 
1631  // Floating point -0.0 and 0.0 compare equal, so we can only
1632  // propagate values if we know that we have a constant and that
1633  // its value is non-zero.
1634 
1635  // FIXME: We should do this optimization if 'no signed zeros' is
1636  // applicable via an instruction-level fast-math-flag or some other
1637  // indicator that relaxed FP semantics are being used.
1638 
1639  if (isa<ConstantFP>(Op1) && !cast<ConstantFP>(Op1)->isZero())
1640  Worklist.push_back(std::make_pair(Op0, Op1));
1641  }
1642 
1643  // If "A >= B" is known true, replace "A < B" with false everywhere.
1644  CmpInst::Predicate NotPred = Cmp->getInversePredicate();
1645  Constant *NotVal = ConstantInt::get(Cmp->getType(), isKnownFalse);
1646  // Since we don't have the instruction "A < B" immediately to hand, work
1647  // out the value number that it would have and use that to find an
1648  // appropriate instruction (if any).
1649  uint32_t NextNum = VN.getNextUnusedValueNumber();
1650  uint32_t Num = VN.lookupOrAddCmp(Cmp->getOpcode(), NotPred, Op0, Op1);
1651  // If the number we were assigned was brand new then there is no point in
1652  // looking for an instruction realizing it: there cannot be one!
1653  if (Num < NextNum) {
1654  Value *NotCmp = findLeader(Root.getEnd(), Num);
1655  if (NotCmp && isa<Instruction>(NotCmp)) {
1656  unsigned NumReplacements =
1657  DominatesByEdge
1658  ? replaceDominatedUsesWith(NotCmp, NotVal, *DT, Root)
1659  : replaceDominatedUsesWith(NotCmp, NotVal, *DT,
1660  Root.getStart());
1661  Changed |= NumReplacements > 0;
1662  NumGVNEqProp += NumReplacements;
1663  }
1664  }
1665  // Ensure that any instruction in scope that gets the "A < B" value number
1666  // is replaced with false.
1667  // The leader table only tracks basic blocks, not edges. Only add to if we
1668  // have the simple case where the edge dominates the end.
1669  if (RootDominatesEnd)
1670  addToLeaderTable(Num, NotVal, Root.getEnd());
1671 
1672  continue;
1673  }
1674  }
1675 
1676  return Changed;
1677 }
1678 
1679 /// When calculating availability, handle an instruction
1680 /// by inserting it into the appropriate sets
1681 bool GVN::processInstruction(Instruction *I) {
1682  // Ignore dbg info intrinsics.
1683  if (isa<DbgInfoIntrinsic>(I))
1684  return false;
1685 
1686  // If the instruction can be easily simplified then do so now in preference
1687  // to value numbering it. Value numbering often exposes redundancies, for
1688  // example if it determines that %y is equal to %x then the instruction
1689  // "%z = and i32 %x, %y" becomes "%z = and i32 %x, %x" which we now simplify.
1690  const DataLayout &DL = I->getModule()->getDataLayout();
1691  if (Value *V = SimplifyInstruction(I, {DL, TLI, DT, AC})) {
1692  bool Changed = false;
1693  if (!I->use_empty()) {
1694  I->replaceAllUsesWith(V);
1695  Changed = true;
1696  }
1697  if (isInstructionTriviallyDead(I, TLI)) {
1699  Changed = true;
1700  }
1701  if (Changed) {
1702  if (MD && V->getType()->isPtrOrPtrVectorTy())
1703  MD->invalidateCachedPointerInfo(V);
1704  ++NumGVNSimpl;
1705  return true;
1706  }
1707  }
1708 
1709  if (IntrinsicInst *IntrinsicI = dyn_cast<IntrinsicInst>(I))
1710  if (IntrinsicI->getIntrinsicID() == Intrinsic::assume)
1711  return processAssumeIntrinsic(IntrinsicI);
1712 
1713  if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
1714  if (processLoad(LI))
1715  return true;
1716 
1717  unsigned Num = VN.lookupOrAdd(LI);
1718  addToLeaderTable(Num, LI, LI->getParent());
1719  return false;
1720  }
1721 
1722  // For conditional branches, we can perform simple conditional propagation on
1723  // the condition value itself.
1724  if (BranchInst *BI = dyn_cast<BranchInst>(I)) {
1725  if (!BI->isConditional())
1726  return false;
1727 
1728  if (isa<Constant>(BI->getCondition()))
1729  return processFoldableCondBr(BI);
1730 
1731  Value *BranchCond = BI->getCondition();
1732  BasicBlock *TrueSucc = BI->getSuccessor(0);
1733  BasicBlock *FalseSucc = BI->getSuccessor(1);
1734  // Avoid multiple edges early.
1735  if (TrueSucc == FalseSucc)
1736  return false;
1737 
1738  BasicBlock *Parent = BI->getParent();
1739  bool Changed = false;
1740 
1741  Value *TrueVal = ConstantInt::getTrue(TrueSucc->getContext());
1742  BasicBlockEdge TrueE(Parent, TrueSucc);
1743  Changed |= propagateEquality(BranchCond, TrueVal, TrueE, true);
1744 
1745  Value *FalseVal = ConstantInt::getFalse(FalseSucc->getContext());
1746  BasicBlockEdge FalseE(Parent, FalseSucc);
1747  Changed |= propagateEquality(BranchCond, FalseVal, FalseE, true);
1748 
1749  return Changed;
1750  }
1751 
1752  // For switches, propagate the case values into the case destinations.
1753  if (SwitchInst *SI = dyn_cast<SwitchInst>(I)) {
1754  Value *SwitchCond = SI->getCondition();
1755  BasicBlock *Parent = SI->getParent();
1756  bool Changed = false;
1757 
1758  // Remember how many outgoing edges there are to every successor.
1760  for (unsigned i = 0, n = SI->getNumSuccessors(); i != n; ++i)
1761  ++SwitchEdges[SI->getSuccessor(i)];
1762 
1763  for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
1764  i != e; ++i) {
1765  BasicBlock *Dst = i->getCaseSuccessor();
1766  // If there is only a single edge, propagate the case value into it.
1767  if (SwitchEdges.lookup(Dst) == 1) {
1768  BasicBlockEdge E(Parent, Dst);
1769  Changed |= propagateEquality(SwitchCond, i->getCaseValue(), E, true);
1770  }
1771  }
1772  return Changed;
1773  }
1774 
1775  // Instructions with void type don't return a value, so there's
1776  // no point in trying to find redundancies in them.
1777  if (I->getType()->isVoidTy())
1778  return false;
1779 
1780  uint32_t NextNum = VN.getNextUnusedValueNumber();
1781  unsigned Num = VN.lookupOrAdd(I);
1782 
1783  // Allocations are always uniquely numbered, so we can save time and memory
1784  // by fast failing them.
1785  if (isa<AllocaInst>(I) || isa<TerminatorInst>(I) || isa<PHINode>(I)) {
1786  addToLeaderTable(Num, I, I->getParent());
1787  return false;
1788  }
1789 
1790  // If the number we were assigned was a brand new VN, then we don't
1791  // need to do a lookup to see if the number already exists
1792  // somewhere in the domtree: it can't!
1793  if (Num >= NextNum) {
1794  addToLeaderTable(Num, I, I->getParent());
1795  return false;
1796  }
1797 
1798  // Perform fast-path value-number based elimination of values inherited from
1799  // dominators.
1800  Value *Repl = findLeader(I->getParent(), Num);
1801  if (!Repl) {
1802  // Failure, just remember this instance for future use.
1803  addToLeaderTable(Num, I, I->getParent());
1804  return false;
1805  } else if (Repl == I) {
1806  // If I was the result of a shortcut PRE, it might already be in the table
1807  // and the best replacement for itself. Nothing to do.
1808  return false;
1809  }
1810 
1811  // Remove it!
1812  patchAndReplaceAllUsesWith(I, Repl);
1813  if (MD && Repl->getType()->isPtrOrPtrVectorTy())
1814  MD->invalidateCachedPointerInfo(Repl);
1816  return true;
1817 }
1818 
1819 /// runOnFunction - This is the main transformation entry point for a function.
1820 bool GVN::runImpl(Function &F, AssumptionCache &RunAC, DominatorTree &RunDT,
1821  const TargetLibraryInfo &RunTLI, AAResults &RunAA,
1822  MemoryDependenceResults *RunMD, LoopInfo *LI,
1823  OptimizationRemarkEmitter *RunORE) {
1824  AC = &RunAC;
1825  DT = &RunDT;
1826  VN.setDomTree(DT);
1827  TLI = &RunTLI;
1828  VN.setAliasAnalysis(&RunAA);
1829  MD = RunMD;
1830  VN.setMemDep(MD);
1831  ORE = RunORE;
1832 
1833  bool Changed = false;
1834  bool ShouldContinue = true;
1835 
1836  // Merge unconditional branches, allowing PRE to catch more
1837  // optimization opportunities.
1838  for (Function::iterator FI = F.begin(), FE = F.end(); FI != FE; ) {
1839  BasicBlock *BB = &*FI++;
1840 
1841  bool removedBlock = MergeBlockIntoPredecessor(BB, DT, LI, MD);
1842  if (removedBlock)
1843  ++NumGVNBlocks;
1844 
1845  Changed |= removedBlock;
1846  }
1847 
1848  unsigned Iteration = 0;
1849  while (ShouldContinue) {
1850  DEBUG(dbgs() << "GVN iteration: " << Iteration << "\n");
1851  ShouldContinue = iterateOnFunction(F);
1852  Changed |= ShouldContinue;
1853  ++Iteration;
1854  }
1855 
1856  if (EnablePRE) {
1857  // Fabricate val-num for dead-code in order to suppress assertion in
1858  // performPRE().
1859  assignValNumForDeadCode();
1860  bool PREChanged = true;
1861  while (PREChanged) {
1862  PREChanged = performPRE(F);
1863  Changed |= PREChanged;
1864  }
1865  }
1866 
1867  // FIXME: Should perform GVN again after PRE does something. PRE can move
1868  // computations into blocks where they become fully redundant. Note that
1869  // we can't do this until PRE's critical edge splitting updates memdep.
1870  // Actually, when this happens, we should just fully integrate PRE into GVN.
1871 
1872  cleanupGlobalSets();
1873  // Do not cleanup DeadBlocks in cleanupGlobalSets() as it's called for each
1874  // iteration.
1875  DeadBlocks.clear();
1876 
1877  return Changed;
1878 }
1879 
1880 bool GVN::processBlock(BasicBlock *BB) {
1881  // FIXME: Kill off InstrsToErase by doing erasing eagerly in a helper function
1882  // (and incrementing BI before processing an instruction).
1883  assert(InstrsToErase.empty() &&
1884  "We expect InstrsToErase to be empty across iterations");
1885  if (DeadBlocks.count(BB))
1886  return false;
1887 
1888  // Clearing map before every BB because it can be used only for single BB.
1889  ReplaceWithConstMap.clear();
1890  bool ChangedFunction = false;
1891 
1892  for (BasicBlock::iterator BI = BB->begin(), BE = BB->end();
1893  BI != BE;) {
1894  if (!ReplaceWithConstMap.empty())
1895  ChangedFunction |= replaceOperandsWithConsts(&*BI);
1896  ChangedFunction |= processInstruction(&*BI);
1897 
1898  if (InstrsToErase.empty()) {
1899  ++BI;
1900  continue;
1901  }
1902 
1903  // If we need some instructions deleted, do it now.
1904  NumGVNInstr += InstrsToErase.size();
1905 
1906  // Avoid iterator invalidation.
1907  bool AtStart = BI == BB->begin();
1908  if (!AtStart)
1909  --BI;
1910 
1911  for (SmallVectorImpl<Instruction *>::iterator I = InstrsToErase.begin(),
1912  E = InstrsToErase.end(); I != E; ++I) {
1913  DEBUG(dbgs() << "GVN removed: " << **I << '\n');
1914  if (MD) MD->removeInstruction(*I);
1915  DEBUG(verifyRemoved(*I));
1916  (*I)->eraseFromParent();
1917  }
1918  InstrsToErase.clear();
1919 
1920  if (AtStart)
1921  BI = BB->begin();
1922  else
1923  ++BI;
1924  }
1925 
1926  return ChangedFunction;
1927 }
1928 
1929 // Instantiate an expression in a predecessor that lacked it.
1930 bool GVN::performScalarPREInsertion(Instruction *Instr, BasicBlock *Pred,
1931  unsigned int ValNo) {
1932  // Because we are going top-down through the block, all value numbers
1933  // will be available in the predecessor by the time we need them. Any
1934  // that weren't originally present will have been instantiated earlier
1935  // in this loop.
1936  bool success = true;
1937  for (unsigned i = 0, e = Instr->getNumOperands(); i != e; ++i) {
1938  Value *Op = Instr->getOperand(i);
1939  if (isa<Argument>(Op) || isa<Constant>(Op) || isa<GlobalValue>(Op))
1940  continue;
1941  // This could be a newly inserted instruction, in which case, we won't
1942  // find a value number, and should give up before we hurt ourselves.
1943  // FIXME: Rewrite the infrastructure to let it easier to value number
1944  // and process newly inserted instructions.
1945  if (!VN.exists(Op)) {
1946  success = false;
1947  break;
1948  }
1949  if (Value *V = findLeader(Pred, VN.lookup(Op))) {
1950  Instr->setOperand(i, V);
1951  } else {
1952  success = false;
1953  break;
1954  }
1955  }
1956 
1957  // Fail out if we encounter an operand that is not available in
1958  // the PRE predecessor. This is typically because of loads which
1959  // are not value numbered precisely.
1960  if (!success)
1961  return false;
1962 
1963  Instr->insertBefore(Pred->getTerminator());
1964  Instr->setName(Instr->getName() + ".pre");
1965  Instr->setDebugLoc(Instr->getDebugLoc());
1966  VN.add(Instr, ValNo);
1967 
1968  // Update the availability map to include the new instruction.
1969  addToLeaderTable(ValNo, Instr, Pred);
1970  return true;
1971 }
1972 
1973 bool GVN::performScalarPRE(Instruction *CurInst) {
1974  if (isa<AllocaInst>(CurInst) || isa<TerminatorInst>(CurInst) ||
1975  isa<PHINode>(CurInst) || CurInst->getType()->isVoidTy() ||
1976  CurInst->mayReadFromMemory() || CurInst->mayHaveSideEffects() ||
1977  isa<DbgInfoIntrinsic>(CurInst))
1978  return false;
1979 
1980  // Don't do PRE on compares. The PHI would prevent CodeGenPrepare from
1981  // sinking the compare again, and it would force the code generator to
1982  // move the i1 from processor flags or predicate registers into a general
1983  // purpose register.
1984  if (isa<CmpInst>(CurInst))
1985  return false;
1986 
1987  // We don't currently value number ANY inline asm calls.
1988  if (CallInst *CallI = dyn_cast<CallInst>(CurInst))
1989  if (CallI->isInlineAsm())
1990  return false;
1991 
1992  uint32_t ValNo = VN.lookup(CurInst);
1993 
1994  // Look for the predecessors for PRE opportunities. We're
1995  // only trying to solve the basic diamond case, where
1996  // a value is computed in the successor and one predecessor,
1997  // but not the other. We also explicitly disallow cases
1998  // where the successor is its own predecessor, because they're
1999  // more complicated to get right.
2000  unsigned NumWith = 0;
2001  unsigned NumWithout = 0;
2002  BasicBlock *PREPred = nullptr;
2003  BasicBlock *CurrentBlock = CurInst->getParent();
2004 
2006  for (BasicBlock *P : predecessors(CurrentBlock)) {
2007  // We're not interested in PRE where the block is its
2008  // own predecessor, or in blocks with predecessors
2009  // that are not reachable.
2010  if (P == CurrentBlock) {
2011  NumWithout = 2;
2012  break;
2013  } else if (!DT->isReachableFromEntry(P)) {
2014  NumWithout = 2;
2015  break;
2016  }
2017 
2018  Value *predV = findLeader(P, ValNo);
2019  if (!predV) {
2020  predMap.push_back(std::make_pair(static_cast<Value *>(nullptr), P));
2021  PREPred = P;
2022  ++NumWithout;
2023  } else if (predV == CurInst) {
2024  /* CurInst dominates this predecessor. */
2025  NumWithout = 2;
2026  break;
2027  } else {
2028  predMap.push_back(std::make_pair(predV, P));
2029  ++NumWith;
2030  }
2031  }
2032 
2033  // Don't do PRE when it might increase code size, i.e. when
2034  // we would need to insert instructions in more than one pred.
2035  if (NumWithout > 1 || NumWith == 0)
2036  return false;
2037 
2038  // We may have a case where all predecessors have the instruction,
2039  // and we just need to insert a phi node. Otherwise, perform
2040  // insertion.
2041  Instruction *PREInstr = nullptr;
2042 
2043  if (NumWithout != 0) {
2044  // Don't do PRE across indirect branch.
2045  if (isa<IndirectBrInst>(PREPred->getTerminator()))
2046  return false;
2047 
2048  // We can't do PRE safely on a critical edge, so instead we schedule
2049  // the edge to be split and perform the PRE the next time we iterate
2050  // on the function.
2051  unsigned SuccNum = GetSuccessorNumber(PREPred, CurrentBlock);
2052  if (isCriticalEdge(PREPred->getTerminator(), SuccNum)) {
2053  toSplit.push_back(std::make_pair(PREPred->getTerminator(), SuccNum));
2054  return false;
2055  }
2056  // We need to insert somewhere, so let's give it a shot
2057  PREInstr = CurInst->clone();
2058  if (!performScalarPREInsertion(PREInstr, PREPred, ValNo)) {
2059  // If we failed insertion, make sure we remove the instruction.
2060  DEBUG(verifyRemoved(PREInstr));
2061  PREInstr->deleteValue();
2062  return false;
2063  }
2064  }
2065 
2066  // Either we should have filled in the PRE instruction, or we should
2067  // not have needed insertions.
2068  assert (PREInstr != nullptr || NumWithout == 0);
2069 
2070  ++NumGVNPRE;
2071 
2072  // Create a PHI to make the value available in this block.
2073  PHINode *Phi =
2074  PHINode::Create(CurInst->getType(), predMap.size(),
2075  CurInst->getName() + ".pre-phi", &CurrentBlock->front());
2076  for (unsigned i = 0, e = predMap.size(); i != e; ++i) {
2077  if (Value *V = predMap[i].first)
2078  Phi->addIncoming(V, predMap[i].second);
2079  else
2080  Phi->addIncoming(PREInstr, PREPred);
2081  }
2082 
2083  VN.add(Phi, ValNo);
2084  addToLeaderTable(ValNo, Phi, CurrentBlock);
2085  Phi->setDebugLoc(CurInst->getDebugLoc());
2086  CurInst->replaceAllUsesWith(Phi);
2087  if (MD && Phi->getType()->isPtrOrPtrVectorTy())
2088  MD->invalidateCachedPointerInfo(Phi);
2089  VN.erase(CurInst);
2090  removeFromLeaderTable(ValNo, CurInst, CurrentBlock);
2091 
2092  DEBUG(dbgs() << "GVN PRE removed: " << *CurInst << '\n');
2093  if (MD)
2094  MD->removeInstruction(CurInst);
2095  DEBUG(verifyRemoved(CurInst));
2096  CurInst->eraseFromParent();
2097  ++NumGVNInstr;
2098 
2099  return true;
2100 }
2101 
2102 /// Perform a purely local form of PRE that looks for diamond
2103 /// control flow patterns and attempts to perform simple PRE at the join point.
2104 bool GVN::performPRE(Function &F) {
2105  bool Changed = false;
2106  for (BasicBlock *CurrentBlock : depth_first(&F.getEntryBlock())) {
2107  // Nothing to PRE in the entry block.
2108  if (CurrentBlock == &F.getEntryBlock())
2109  continue;
2110 
2111  // Don't perform PRE on an EH pad.
2112  if (CurrentBlock->isEHPad())
2113  continue;
2114 
2115  for (BasicBlock::iterator BI = CurrentBlock->begin(),
2116  BE = CurrentBlock->end();
2117  BI != BE;) {
2118  Instruction *CurInst = &*BI++;
2119  Changed |= performScalarPRE(CurInst);
2120  }
2121  }
2122 
2123  if (splitCriticalEdges())
2124  Changed = true;
2125 
2126  return Changed;
2127 }
2128 
2129 /// Split the critical edge connecting the given two blocks, and return
2130 /// the block inserted to the critical edge.
2131 BasicBlock *GVN::splitCriticalEdges(BasicBlock *Pred, BasicBlock *Succ) {
2132  BasicBlock *BB =
2134  if (MD)
2135  MD->invalidateCachedPredecessors();
2136  return BB;
2137 }
2138 
2139 /// Split critical edges found during the previous
2140 /// iteration that may enable further optimization.
2141 bool GVN::splitCriticalEdges() {
2142  if (toSplit.empty())
2143  return false;
2144  do {
2145  std::pair<TerminatorInst*, unsigned> Edge = toSplit.pop_back_val();
2146  SplitCriticalEdge(Edge.first, Edge.second,
2148  } while (!toSplit.empty());
2149  if (MD) MD->invalidateCachedPredecessors();
2150  return true;
2151 }
2152 
2153 /// Executes one iteration of GVN
2154 bool GVN::iterateOnFunction(Function &F) {
2155  cleanupGlobalSets();
2156 
2157  // Top-down walk of the dominator tree
2158  bool Changed = false;
2159  // Needed for value numbering with phi construction to work.
2160  // RPOT walks the graph in its constructor and will not be invalidated during
2161  // processBlock.
2163  for (BasicBlock *BB : RPOT)
2164  Changed |= processBlock(BB);
2165 
2166  return Changed;
2167 }
2168 
2169 void GVN::cleanupGlobalSets() {
2170  VN.clear();
2171  LeaderTable.clear();
2172  TableAllocator.Reset();
2173 }
2174 
2175 /// Verify that the specified instruction does not occur in our
2176 /// internal data structures.
2177 void GVN::verifyRemoved(const Instruction *Inst) const {
2178  VN.verifyRemoved(Inst);
2179 
2180  // Walk through the value number scope to make sure the instruction isn't
2181  // ferreted away in it.
2183  I = LeaderTable.begin(), E = LeaderTable.end(); I != E; ++I) {
2184  const LeaderTableEntry *Node = &I->second;
2185  assert(Node->Val != Inst && "Inst still in value numbering scope!");
2186 
2187  while (Node->Next) {
2188  Node = Node->Next;
2189  assert(Node->Val != Inst && "Inst still in value numbering scope!");
2190  }
2191  }
2192 }
2193 
2194 /// BB is declared dead, which implied other blocks become dead as well. This
2195 /// function is to add all these blocks to "DeadBlocks". For the dead blocks'
2196 /// live successors, update their phi nodes by replacing the operands
2197 /// corresponding to dead blocks with UndefVal.
2198 void GVN::addDeadBlock(BasicBlock *BB) {
2201 
2202  NewDead.push_back(BB);
2203  while (!NewDead.empty()) {
2204  BasicBlock *D = NewDead.pop_back_val();
2205  if (DeadBlocks.count(D))
2206  continue;
2207 
2208  // All blocks dominated by D are dead.
2210  DT->getDescendants(D, Dom);
2211  DeadBlocks.insert(Dom.begin(), Dom.end());
2212 
2213  // Figure out the dominance-frontier(D).
2214  for (BasicBlock *B : Dom) {
2215  for (BasicBlock *S : successors(B)) {
2216  if (DeadBlocks.count(S))
2217  continue;
2218 
2219  bool AllPredDead = true;
2220  for (BasicBlock *P : predecessors(S))
2221  if (!DeadBlocks.count(P)) {
2222  AllPredDead = false;
2223  break;
2224  }
2225 
2226  if (!AllPredDead) {
2227  // S could be proved dead later on. That is why we don't update phi
2228  // operands at this moment.
2229  DF.insert(S);
2230  } else {
2231  // While S is not dominated by D, it is dead by now. This could take
2232  // place if S already have a dead predecessor before D is declared
2233  // dead.
2234  NewDead.push_back(S);
2235  }
2236  }
2237  }
2238  }
2239 
2240  // For the dead blocks' live successors, update their phi nodes by replacing
2241  // the operands corresponding to dead blocks with UndefVal.
2243  I != E; I++) {
2244  BasicBlock *B = *I;
2245  if (DeadBlocks.count(B))
2246  continue;
2247 
2249  for (BasicBlock *P : Preds) {
2250  if (!DeadBlocks.count(P))
2251  continue;
2252 
2253  if (isCriticalEdge(P->getTerminator(), GetSuccessorNumber(P, B))) {
2254  if (BasicBlock *S = splitCriticalEdges(P, B))
2255  DeadBlocks.insert(P = S);
2256  }
2257 
2258  for (BasicBlock::iterator II = B->begin(); isa<PHINode>(II); ++II) {
2259  PHINode &Phi = cast<PHINode>(*II);
2261  UndefValue::get(Phi.getType()));
2262  }
2263  }
2264  }
2265 }
2266 
2267 // If the given branch is recognized as a foldable branch (i.e. conditional
2268 // branch with constant condition), it will perform following analyses and
2269 // transformation.
2270 // 1) If the dead out-coming edge is a critical-edge, split it. Let
2271 // R be the target of the dead out-coming edge.
2272 // 1) Identify the set of dead blocks implied by the branch's dead outcoming
2273 // edge. The result of this step will be {X| X is dominated by R}
2274 // 2) Identify those blocks which haves at least one dead predecessor. The
2275 // result of this step will be dominance-frontier(R).
2276 // 3) Update the PHIs in DF(R) by replacing the operands corresponding to
2277 // dead blocks with "UndefVal" in an hope these PHIs will optimized away.
2278 //
2279 // Return true iff *NEW* dead code are found.
2280 bool GVN::processFoldableCondBr(BranchInst *BI) {
2281  if (!BI || BI->isUnconditional())
2282  return false;
2283 
2284  // If a branch has two identical successors, we cannot declare either dead.
2285  if (BI->getSuccessor(0) == BI->getSuccessor(1))
2286  return false;
2287 
2288  ConstantInt *Cond = dyn_cast<ConstantInt>(BI->getCondition());
2289  if (!Cond)
2290  return false;
2291 
2292  BasicBlock *DeadRoot =
2293  Cond->getZExtValue() ? BI->getSuccessor(1) : BI->getSuccessor(0);
2294  if (DeadBlocks.count(DeadRoot))
2295  return false;
2296 
2297  if (!DeadRoot->getSinglePredecessor())
2298  DeadRoot = splitCriticalEdges(BI->getParent(), DeadRoot);
2299 
2300  addDeadBlock(DeadRoot);
2301  return true;
2302 }
2303 
2304 // performPRE() will trigger assert if it comes across an instruction without
2305 // associated val-num. As it normally has far more live instructions than dead
2306 // instructions, it makes more sense just to "fabricate" a val-number for the
2307 // dead code than checking if instruction involved is dead or not.
2308 void GVN::assignValNumForDeadCode() {
2309  for (BasicBlock *BB : DeadBlocks) {
2310  for (Instruction &Inst : *BB) {
2311  unsigned ValNum = VN.lookupOrAdd(&Inst);
2312  addToLeaderTable(ValNum, &Inst, BB);
2313  }
2314  }
2315 }
2316 
2318 public:
2319  static char ID; // Pass identification, replacement for typeid
2320  explicit GVNLegacyPass(bool NoLoads = false)
2321  : FunctionPass(ID), NoLoads(NoLoads) {
2323  }
2324 
2325  bool runOnFunction(Function &F) override {
2326  if (skipFunction(F))
2327  return false;
2328 
2329  auto *LIWP = getAnalysisIfAvailable<LoopInfoWrapperPass>();
2330 
2331  return Impl.runImpl(
2332  F, getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F),
2333  getAnalysis<DominatorTreeWrapperPass>().getDomTree(),
2334  getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(),
2335  getAnalysis<AAResultsWrapperPass>().getAAResults(),
2336  NoLoads ? nullptr
2337  : &getAnalysis<MemoryDependenceWrapperPass>().getMemDep(),
2338  LIWP ? &LIWP->getLoopInfo() : nullptr,
2339  &getAnalysis<OptimizationRemarkEmitterWrapperPass>().getORE());
2340  }
2341 
2342  void getAnalysisUsage(AnalysisUsage &AU) const override {
2346  if (!NoLoads)
2349 
2354  }
2355 
2356 private:
2357  bool NoLoads;
2358  GVN Impl;
2359 };
2360 
2361 char GVNLegacyPass::ID = 0;
2362 
2363 // The public interface to this file...
2365  return new GVNLegacyPass(NoLoads);
2366 }
2367 
2368 INITIALIZE_PASS_BEGIN(GVNLegacyPass, "gvn", "Global Value Numbering", false, false)
2376 INITIALIZE_PASS_END(GVNLegacyPass, "gvn", "Global Value Numbering", false, false)
Legacy wrapper pass to provide the GlobalsAAResult object.
BinaryOp_match< LHS, RHS, Instruction::And > m_And(const LHS &L, const RHS &R)
Definition: PatternMatch.h:550
uint64_t CallInst * C
SymbolTableList< Instruction >::iterator eraseFromParent()
This method unlinks &#39;this&#39; from the containing basic block and deletes it.
Definition: Instruction.cpp:69
FunctionPass * createGVNPass(bool NoLoads=false)
Create a legacy GVN pass.
Definition: GVN.cpp:2364
static cl::opt< bool > EnableLoadPRE("enable-load-pre", cl::init(true))
A parsed version of the target data layout string in and methods for querying it. ...
Definition: DataLayout.h:109
bool isUndefValue() const
Definition: GVN.cpp:176
static ConstantInt * getFalse(LLVMContext &Context)
Definition: Constants.cpp:523
class_match< Value > m_Value()
Match an arbitrary value and ignore it.
Definition: PatternMatch.h:72
This class is the base class for the comparison instructions.
Definition: InstrTypes.h:850
AnalysisUsage & addPreserved()
Add the specified Pass class to the set of analyses preserved by this pass.
raw_ostream & errs()
This returns a reference to a raw_ostream for standard error.
Helper class for SSA formation on a set of values defined in multiple blocks.
Definition: SSAUpdater.h:39
Diagnostic information for missed-optimization remarks.
Provides a lazy, caching interface for making common memory aliasing information queries, backed by LLVM&#39;s alias analysis passes.
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:219
void addIncoming(Value *V, BasicBlock *BB)
Add an incoming value to the end of the PHI list.
This instruction extracts a struct member or array element value from an aggregate value...
static PassRegistry * getPassRegistry()
getPassRegistry - Access the global registry object, which is automatically initialized at applicatio...
static AvailableValue getMI(MemIntrinsic *MI, unsigned Offset=0)
Definition: GVN.cpp:149
size_type size() const
Definition: MapVector.h:57
unsigned Offset
Offset - The byte offset in Val that is interesting for the load query.
Definition: GVN.cpp:139
DiagnosticInfoOptimizationBase::Argument NV
static Type * makeCmpResultType(Type *opnd_type)
Create a result type for fcmp/icmp.
Definition: InstrTypes.h:1074
PassT::Result & getResult(IRUnitT &IR, ExtraArgTs... ExtraArgs)
Get the result of an analysis pass for a given IR unit.
Definition: PassManager.h:687
Compute iterated dominance frontiers using a linear time algorithm.
Definition: AllocatorList.h:24
PointerTy getPointer() const
#define LLVM_DUMP_METHOD
Mark debug helper function definitions like dump() that should not be stripped from debug builds...
Definition: Compiler.h:449
bool isAtomic() const
Return true if this instruction has an AtomicOrdering of unordered or higher.
This is the interface for a simple mod/ref and alias analysis over globals.
void Initialize(Type *Ty, StringRef Name)
Reset this object to get ready for a new set of SSA updates with type &#39;Ty&#39;.
Definition: SSAUpdater.cpp:54
LLVM_ATTRIBUTE_ALWAYS_INLINE size_type size() const
Definition: SmallVector.h:136
uint32_t lookupOrAddCmp(unsigned Opcode, CmpInst::Predicate Pred, Value *LHS, Value *RHS)
Returns the value number of the given comparison, assigning it a new number if it did not have one be...
Definition: GVN.cpp:548
iterator end()
Definition: Function.h:582
void AddAvailableValue(BasicBlock *BB, Value *V)
Indicate that a rewritten value is available in the specified block with the specified value...
Definition: SSAUpdater.cpp:67
bool operator==(const Expression &other) const
Definition: GVN.cpp:87
This class represents a function call, abstracting a target machine&#39;s calling convention.
bool isNonLocal() const
Tests if this MemDepResult represents a query that is transparent to the start of the block...
This file contains the declarations for metadata subclasses.
An immutable pass that tracks lazily created AssumptionCache objects.
uint32_t lookup(Value *V) const
Returns the value number of the specified value.
Definition: GVN.cpp:538
A cache of .assume calls within a function.
AtomicOrdering getOrdering() const
Returns the ordering constraint of this load instruction.
Definition: Instructions.h:233
bool onlyReadsMemory(ImmutableCallSite CS)
Checks if the specified call is known to only read from non-volatile memory (or not access memory at ...
1 1 1 0 True if unordered or not equal
Definition: InstrTypes.h:876
void deleteValue()
Delete a pointer to a generic Value.
Definition: Value.cpp:93
bool hasFnAttribute(Attribute::AttrKind Kind) const
Return true if the function has the attribute.
Definition: Function.h:254
unsigned second
This class implements a map that also provides access to all stored values in a deterministic order...
Definition: MapVector.h:38
BasicBlock * getSuccessor(unsigned i) const
bool properlyDominates(const DomTreeNodeBase< NodeT > *A, const DomTreeNodeBase< NodeT > *B) const
properlyDominates - Returns true iff A dominates B and A != B.
STATISTIC(NumFunctions, "Total number of functions")
A debug info location.
Definition: DebugLoc.h:34
Analysis pass which computes a DominatorTree.
Definition: Dominators.h:232
bool isCoercedLoadValue() const
Definition: GVN.cpp:174
An instruction for reading from memory.
Definition: Instructions.h:164
const BasicBlock * getEnd() const
Definition: Dominators.h:84
Hexagon Common GEP
Value * getCondition() const
idx_iterator idx_end() const
unsigned replaceDominatedUsesWith(Value *From, Value *To, DominatorTree &DT, const BasicBlockEdge &Edge)
Replace each use of &#39;From&#39; with &#39;To&#39; if that use is dominated by the given edge.
Definition: Local.cpp:1816
Use * op_iterator
Definition: User.h:209
iterator end()
Get an iterator to the end of the SetVector.
Definition: SetVector.h:93
Value * getMemInstValueForLoad(MemIntrinsic *SrcInst, unsigned Offset, Type *LoadTy, Instruction *InsertPt, const DataLayout &DL)
If analyzeLoadFromClobberingMemInst returned an offset, this function can be used to actually perform...
Definition: VNCoercion.cpp:475
LLVMContext & getContext() const
Get the context in which this basic block lives.
Definition: BasicBlock.cpp:33
op_iterator op_begin()
Definition: User.h:214
static Constant * getNullValue(Type *Ty)
Constructor to create a &#39;0&#39; constant of arbitrary type.
Definition: Constants.cpp:207
iterator begin()
Instruction iterator methods.
Definition: BasicBlock.h:252
std::pair< iterator, bool > insert(const std::pair< KeyT, ValueT > &KV)
Definition: DenseMap.h:176
void dump() const
Support for debugging, callable in GDB: V->dump()
Definition: AsmWriter.cpp:3606
bool match(Val *V, const Pattern &P)
Definition: PatternMatch.h:49
AnalysisUsage & addRequired()
#define INITIALIZE_PASS_DEPENDENCY(depName)
Definition: PassSupport.h:51
bool isVolatile() const
Return true if this is a load from a volatile memory location.
Definition: Instructions.h:217
static cl::opt< uint32_t > MaxRecurseDepth("max-recurse-depth", cl::Hidden, cl::init(1000), cl::ZeroOrMore, cl::desc("Max recurse depth (default = 1000)"))
void getAnalysisUsage(AnalysisUsage &AU) const override
getAnalysisUsage - This function should be overriden by passes that need analysis information to do t...
Definition: GVN.cpp:2342
static cl::opt< bool > EnablePRE("enable-pre", cl::init(true), cl::Hidden)
bool isDef() const
Tests if this MemDepResult represents a query that is an instruction definition dependency.
const DataLayout & getDataLayout() const
Get the data layout for the module&#39;s target platform.
Definition: Module.cpp:361
bool runOnFunction(Function &F) override
runOnFunction - Virtual method overriden by subclasses to do the per-function processing of the pass...
Definition: GVN.cpp:2325
Option class for critical edge splitting.
int getBasicBlockIndex(const BasicBlock *BB) const
Return the first index of the specified basic block in the value list for this PHI.
void clear()
Remove all entries from the ValueTable.
Definition: GVN.cpp:558
unsigned getNumArgOperands() const
Return the number of call arguments.
bool isClobber() const
Tests if this MemDepResult represents a query that is an instruction clobber dependency.
A Use represents the edge between a Value definition and its users.
Definition: Use.h:56
PointerType * getPointerTo(unsigned AddrSpace=0) const
Return a pointer to the current type.
Definition: Type.cpp:639
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:251
This class consists of common code factored out of the SmallVector class to reduce code duplication b...
Definition: APFloat.h:42
MemoryDependenceResults & getMemDep() const
Definition: GVN.h:61
bool isIntegerTy() const
True if this is an instance of IntegerType.
Definition: Type.h:197
An analysis that produces MemoryDependenceResults for a function.
void setName(const Twine &Name)
Change the name of the value.
Definition: Value.cpp:284
std::vector< NonLocalDepEntry > NonLocalDepInfo
Analysis pass that exposes the LoopInfo for a function.
Definition: LoopInfo.h:820
bool isSimpleValue() const
Definition: GVN.cpp:173
Interval::succ_iterator succ_begin(Interval *I)
succ_begin/succ_end - define methods so that Intervals may be used just like BasicBlocks can with the...
Definition: Interval.h:106
Instruction * clone() const
Create a copy of &#39;this&#39; instruction that is identical in all ways except the following: ...
static void patchReplacementInstruction(Instruction *I, Value *Repl)
Definition: GVN.cpp:1366
#define F(x, y, z)
Definition: MD5.cpp:55
Type * getType() const
All values are typed, get the type of this value.
Definition: Value.h:245
bool insert(const value_type &X)
Insert a new element into the SetVector.
Definition: SetVector.h:142
The core GVN pass object.
Definition: GVN.h:46
IntType getInt() const
bool canCoerceMustAliasedValueToLoad(Value *StoredVal, Type *LoadTy, const DataLayout &DL)
Return true if CoerceAvailableValueToLoadType would succeed if it was called.
Definition: VNCoercion.cpp:15
Expression(uint32_t o=~2U)
Definition: GVN.cpp:85
void andIRFlags(const Value *V)
Logical &#39;and&#39; of any supported wrapping, exact, and fast-math flags of V and this instruction...
#define DEBUG_TYPE
Definition: GVN.cpp:61
iterator begin()
Get an iterator to the beginning of the SetVector.
Definition: SetVector.h:83
MDNode * getMetadata(unsigned KindID) const
Get the metadata of given kind attached to this Instruction.
Definition: Instruction.h:190
DiagnosticInfoOptimizationBase::setExtraArgs setExtraArgs
static AvailableValue getLoad(LoadInst *LI, unsigned Offset=0)
Definition: GVN.cpp:157
hash_code hash_value(const APFloat &Arg)
See friend declarations above.
Definition: APFloat.cpp:4428
unsigned getOpcode() const
Returns a member of one of the enums like Instruction::Add.
Definition: Instruction.h:121
LoadInst * getCoercedLoadValue() const
Definition: GVN.cpp:183
static GVN::Expression getEmptyKey()
Definition: GVN.cpp:108
An instruction for storing to memory.
Definition: Instructions.h:306
bool isMinusOne() const
This function will return true iff every bit in this constant is set to true.
Definition: Constants.h:209
void add(Value *V, uint32_t num)
add - Insert a value into the table with a specified value number.
Definition: GVN.cpp:349
void replaceAllUsesWith(Value *V)
Change all uses of this to point to a new Value.
Definition: Value.cpp:428
void takeName(Value *V)
Transfer the name from V to this value.
Definition: Value.cpp:290
iterator begin()
Definition: Function.h:580
static unsigned getHashValue(const GVN::Expression &e)
Definition: GVN.cpp:112
Concrete subclass of DominatorTreeBase that is used to compute a normal dominator tree...
Definition: Dominators.h:134
Value * getOperand(unsigned i) const
Definition: User.h:154
Interval::succ_iterator succ_end(Interval *I)
Definition: Interval.h:109
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:202
void initializeGVNLegacyPassPass(PassRegistry &)
bool isVoidTy() const
Return true if this is &#39;void&#39;.
Definition: Type.h:141
const BasicBlock & getEntryBlock() const
Definition: Function.h:564
an instruction for type-safe pointer arithmetic to access elements of arrays and structs ...
Definition: Instructions.h:837
void getAAMetadata(AAMDNodes &N, bool Merge=false) const
Fills the AAMDNodes structure with AA metadata from this instruction.
BasicBlock * SplitCriticalEdge(TerminatorInst *TI, unsigned SuccNum, const CriticalEdgeSplittingOptions &Options=CriticalEdgeSplittingOptions())
If this edge is a critical edge, insert a new node to split the critical edge.
#define P(N)
initializer< Ty > init(const Ty &Val)
Definition: CommandLine.h:404
Value * GetValueInMiddleOfBlock(BasicBlock *BB)
Construct SSA form, materializing a value that is live in the middle of the specified block...
Definition: SSAUpdater.cpp:95
SmallVector< uint32_t, 4 > varargs
Definition: GVN.cpp:83
const NonLocalDepInfo & getNonLocalCallDependency(CallSite QueryCS)
Perform a full dependency query for the specified call, returning the set of blocks that the value is...
uint64_t getZExtValue() const
Return the constant as a 64-bit unsigned integer value after it has been zero extended as appropriate...
Definition: Constants.h:149
A set of analyses that are preserved following a run of a transformation pass.
Definition: PassManager.h:153
* if(!EatIfPresent(lltok::kw_thread_local)) return false
ParseOptionalThreadLocal := /*empty.
void setDebugLoc(DebugLoc Loc)
Set the debug location information for this instruction.
Definition: Instruction.h:277
const BasicBlock * getSinglePredecessor() const
Return the predecessor of this block if it has a single predecessor block.
Definition: BasicBlock.cpp:217
void insertBefore(Instruction *InsertPos)
Insert an unlinked instruction into a basic block immediately before the specified instruction...
Definition: Instruction.cpp:75
LLVM Basic Block Representation.
Definition: BasicBlock.h:59
PointerIntPair - This class implements a pair of a pointer and small integer.
PHITransAddr - An address value which tracks and handles phi translation.
Definition: PHITransAddr.h:36
The instances of the Type class are immutable: once they are created, they are never changed...
Definition: Type.h:46
BinaryOp_match< LHS, RHS, Instruction::Or > m_Or(const LHS &L, const RHS &R)
Definition: PatternMatch.h:556
This class holds the mapping between values and value numbers.
Definition: GVN.h:68
Conditional or Unconditional Branch instruction.
This file provides the interface for LLVM&#39;s Global Value Numbering pass which eliminates fully redund...
static GVN::Expression getTombstoneKey()
Definition: GVN.cpp:110
static Value * ConstructSSAForLoadSet(LoadInst *LI, SmallVectorImpl< AvailableValueInBlock > &ValuesPerBlock, GVN &gvn)
Given a set of loads specified by ValuesPerBlock, construct SSA form, allowing us to eliminate LI...
Definition: GVN.cpp:702
This is an important base class in LLVM.
Definition: Constant.h:42
LLVM_ATTRIBUTE_ALWAYS_INLINE iterator begin()
Definition: SmallVector.h:116
static bool isEqual(const GVN::Expression &LHS, const GVN::Expression &RHS)
Definition: GVN.cpp:116
const Instruction & front() const
Definition: BasicBlock.h:264
A manager for alias analyses.
#define A
Definition: LargeTest.cpp:12
bool mayHaveSideEffects() const
Return true if the instruction may have side effects.
Definition: Instruction.h:500
Diagnostic information for applied optimization remarks.
Interval::pred_iterator pred_begin(Interval *I)
pred_begin/pred_end - define methods so that Intervals may be used just like BasicBlocks can with the...
Definition: Interval.h:116
unsigned getNumIndices() const
bool isUnordered() const
Definition: Instructions.h:264
Represent the analysis usage information of a pass.
op_iterator op_end()
Definition: User.h:216
Analysis pass providing a never-invalidated alias analysis result.
Predicate
This enumeration lists the possible predicates for CmpInst subclasses.
Definition: InstrTypes.h:860
PointerIntPair< Value *, 2, ValType > Val
V - The value that is live out of the block.
Definition: GVN.cpp:136
MemIntrinsic * getMemIntrinValue() const
Definition: GVN.cpp:188
FunctionPass class - This class is used to implement most global optimizations.
Definition: Pass.h:298
Interval::pred_iterator pred_end(Interval *I)
Definition: Interval.h:119
Value * getPointerOperand()
Definition: Instructions.h:270
Value * getLoadValueForLoad(LoadInst *SrcVal, unsigned Offset, Type *LoadTy, Instruction *InsertPt, const DataLayout &DL)
If analyzeLoadFromClobberingLoad returned an offset, this function can be used to actually perform th...
Definition: VNCoercion.cpp:363
GVNLegacyPass(bool NoLoads=false)
Definition: GVN.cpp:2320
static void reportLoadElim(LoadInst *LI, Value *AvailableValue, OptimizationRemarkEmitter *ORE)
Definition: GVN.cpp:1215
static UndefValue * get(Type *T)
Static factory methods - Return an &#39;undef&#39; object of the specified type.
Definition: Constants.cpp:1320
static PreservedAnalyses all()
Construct a special preserved set that preserves all passes.
Definition: PassManager.h:159
A wrapper analysis pass for the legacy pass manager that exposes a MemoryDepnedenceResults instance...
void printAsOperand(raw_ostream &O, bool PrintType=true, const Module *M=nullptr) const
Print the name of this Value out to the specified raw_ostream.
Definition: AsmWriter.cpp:3538
INITIALIZE_PASS_END(RegBankSelect, DEBUG_TYPE, "Assign register bank of generic virtual registers", false, false) RegBankSelect
A memory dependence query can return one of three different answers.
DominatorTree & getDominatorTree() const
Definition: GVN.h:59
unsigned first
Intrinsic::ID getIntrinsicID() const
Return the intrinsic ID of this intrinsic.
Definition: IntrinsicInst.h:51
static void reportMayClobberedLoad(LoadInst *LI, MemDepResult DepInfo, DominatorTree *DT, OptimizationRemarkEmitter *ORE)
Try to locate the three instruction involved in a missed load-elimination case that is due to an inte...
Definition: GVN.cpp:788
void markInstructionForDeletion(Instruction *I)
This removes the specified instruction from our various maps and marks it for deletion.
Definition: GVN.h:54
A function analysis which provides an AssumptionCache.
bool isPtrOrPtrVectorTy() const
Return true if this is a pointer type or a vector of pointer types.
Definition: Type.h:224
Value * MaterializeAdjustedValue(LoadInst *LI, GVN &gvn) const
Emit code at the end of this block to adjust the value defined here to the specified type...
Definition: GVN.cpp:225
A SetVector that performs no allocations if smaller than a certain size.
Definition: SetVector.h:298
bool doesNotAccessMemory(ImmutableCallSite CS)
Checks if the specified call is known to never read or write memory.
This is the common base class for memset/memcpy/memmove.
Iterator for intrusive lists based on ilist_node.
unsigned getNumOperands() const
Definition: User.h:176
SmallPtrSet - This class implements a set which is optimized for holding SmallSize or less elements...
Definition: SmallPtrSet.h:423
#define E
Definition: LargeTest.cpp:27
This is the shared class of boolean and integer constants.
Definition: Constants.h:84
void emit(DiagnosticInfoOptimizationBase &OptDiag)
Output the remark via the diagnostic handler and to the optimization record file. ...
#define B
Definition: LargeTest.cpp:24
iterator end()
Definition: BasicBlock.h:254
bool dominates(const Instruction *Def, const Use &U) const
Return true if Def dominates a use in User.
Definition: Dominators.cpp:234
Provides information about what library functions are available for the current target.
Predicate
Predicate - These are "(BI << 5) | BO" for various predicates.
Definition: PPCPredicates.h:27
const MemDepResult & getResult() const
size_type count(const KeyT &Key) const
Definition: MapVector.h:132
#define INITIALIZE_PASS_BEGIN(passName, arg, name, cfg, analysis)
Definition: PassSupport.h:48
A collection of metadata nodes that might be associated with a memory access used by the alias-analys...
Definition: Metadata.h:642
LLVM_NODISCARD T pop_back_val()
Definition: SmallVector.h:385
static Constant * get(Type *Ty, uint64_t V, bool isSigned=false)
If Ty is a vector type, return a Constant with a splat of the given value.
Definition: Constants.cpp:560
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...
pred_range predecessors(BasicBlock *BB)
Definition: CFG.h:110
Value * MaterializeAdjustedValue(LoadInst *LI, Instruction *InsertPt, GVN &gvn) const
Emit code at the specified insertion point to adjust the value defined here to the specified type...
Definition: GVN.cpp:733
static ConstantInt * getTrue(LLVMContext &Context)
Definition: Constants.cpp:516
bool isCommutative() const
Return true if the instruction is commutative:
Definition: Instruction.h:416
void setOperand(unsigned i, Value *Val)
Definition: User.h:159
raw_ostream & dbgs()
dbgs() - This returns a reference to a raw_ostream for debugging messages.
Definition: Debug.cpp:132
void swap(llvm::BitVector &LHS, llvm::BitVector &RHS)
Implement std::swap in terms of BitVector swap.
Definition: BitVector.h:923
const Module * getModule() const
Return the module owning the function this instruction belongs to or nullptr it the function does not...
Definition: Instruction.cpp:57
hash_code hash_combine(const Ts &...args)
Combine values into a single hash_code.
Definition: Hashing.h:602
Represents an AvailableValue which can be rematerialized at the end of the associated BasicBlock...
Definition: GVN.cpp:201
typename SuperClass::iterator iterator
Definition: SmallVector.h:328
iterator_range< user_iterator > users()
Definition: Value.h:395
hash_code hash_combine_range(InputIteratorT first, InputIteratorT last)
Compute a hash_code for a sequence of values.
Definition: Hashing.h:480
An opaque object representing a hash code.
Definition: Hashing.h:72
bool isMallocLikeFn(const Value *V, const TargetLibraryInfo *TLI, bool LookThroughBitCast=false)
Tests if a value is a call or invoke to a library function that allocates uninitialized memory (such ...
iterator insert(iterator I, T &&Elt)
Definition: SmallVector.h:482
void verifyRemoved(const Value *) const
verifyRemoved - Verify that the value is removed from all internal data structures.
Definition: GVN.cpp:571
void append(in_iter in_start, in_iter in_end)
Add the specified range to the end of the SmallVector.
Definition: SmallVector.h:398
void erase(Value *v)
Remove a value from the value numbering.
Definition: GVN.cpp:565
static bool isLifetimeStart(const Instruction *Inst)
Definition: GVN.cpp:780
static bool isZero(Value *V, const DataLayout &DL, DominatorTree *DT, AssumptionCache *AC)
Definition: Lint.cpp:529
LLVM_ATTRIBUTE_ALWAYS_INLINE iterator end()
Definition: SmallVector.h:120
unsigned GetSuccessorNumber(const BasicBlock *BB, const BasicBlock *Succ)
Search for the specified successor of basic block BB and return its position in the terminator instru...
Definition: CFG.cpp:72
const DebugLoc & getDebugLoc() const
Return the debug location for this node as a DebugLoc.
Definition: Instruction.h:280
bool MergeBlockIntoPredecessor(BasicBlock *BB, DominatorTree *DT=nullptr, LoopInfo *LI=nullptr, MemoryDependenceResults *MemDep=nullptr)
Attempts to merge a block into its predecessor, if possible.
unsigned getAlignment() const
Return the alignment of the access that is being performed.
Definition: Instructions.h:226
Instruction * getInst() const
If this is a normal dependency, returns the instruction that is depended on.
Value * getStoreValueForLoad(Value *SrcVal, unsigned Offset, Type *LoadTy, Instruction *InsertPt, const DataLayout &DL)
If analyzeLoadFromClobberingStore returned an offset, this function can be used to actually perform t...
Definition: VNCoercion.cpp:343
LLVM_NODISCARD bool empty() const
Definition: SmallVector.h:61
Value * getArgOperand(unsigned i) const
getArgOperand/setArgOperand - Return/set the i-th call argument.
StringRef getName() const
Return a constant reference to the value&#39;s name.
Definition: Value.cpp:218
const Function * getParent() const
Return the enclosing method, or null if none.
Definition: BasicBlock.h:108
bool isCallocLikeFn(const Value *V, const TargetLibraryInfo *TLI, bool LookThroughBitCast=false)
Tests if a value is a call or invoke to a library function that allocates zero-filled memory (such as...
SyncScope::ID getSyncScopeID() const
Returns the synchronization scope ID of this load instruction.
Definition: Instructions.h:245
#define I(x, y, z)
Definition: MD5.cpp:58
bool mayReadFromMemory() const
Return true if this instruction may read memory.
static AvailableValue get(Value *V, unsigned Offset=0)
Definition: GVN.cpp:141
uint32_t opcode
Definition: GVN.cpp:81
PassT::Result * getCachedResult(IRUnitT &IR) const
Get the cached result of an analysis pass for a given IR unit.
Definition: PassManager.h:706
bool exists(Value *V) const
Returns true if a value number exists for the specified value.
Definition: GVN.cpp:463
LLVM_NODISCARD std::enable_if<!is_simple_type< Y >::value, typename cast_retty< X, const Y >::ret_type >::type dyn_cast(const Y &Val)
Definition: Casting.h:323
idx_iterator idx_begin() const
void preserve()
Mark an analysis as preserved.
Definition: PassManager.h:174
bool isUnconditional() const
friend hash_code hash_value(const Expression &Value)
Definition: GVN.cpp:99
uint32_t lookupOrAdd(Value *V)
lookup_or_add - Returns the value number for the specified value, assigning it a new number if it did...
Definition: GVN.cpp:467
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:166
Value * getSimpleValue() const
Definition: GVN.cpp:178
Analysis pass providing the TargetLibraryInfo.
iterator_range< df_iterator< T > > depth_first(const T &G)
Multiway switch.
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
unsigned getNumSuccessors() const
Return the number of successors that this terminator has.
const BasicBlock * getStart() const
Definition: Dominators.h:80
Represents a particular available value that we know how to materialize.
Definition: GVN.cpp:126
static bool IsValueFullyAvailableInBlock(BasicBlock *BB, DenseMap< BasicBlock *, char > &FullyAvailableBlocks, uint32_t RecurseDepth)
Return true if we can prove that the value we&#39;re analyzing is fully available in the specified block...
Definition: GVN.cpp:626
0 0 0 1 True if ordered and equal
Definition: InstrTypes.h:863
bool isInstructionTriviallyDead(Instruction *I, const TargetLibraryInfo *TLI=nullptr)
Return true if the result produced by the instruction is not used, and the instruction has no side ef...
Definition: Local.cpp:293
LLVM Value Representation.
Definition: Value.h:73
succ_range successors(BasicBlock *BB)
Definition: CFG.h:143
static AvailableValueInBlock getUndef(BasicBlock *BB)
Definition: GVN.cpp:219
void removeInstruction(Instruction *InstToRemove)
Removes an instruction from the dependence analysis, updating the dependence of instructions that pre...
OptimizationRemarkEmitter legacy analysis pass.
#define DEBUG(X)
Definition: Debug.h:118
PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM)
Run the pass over the function.
Definition: GVN.cpp:582
IRTranslator LLVM IR MI
bool hasOneUse() const
Return true if there is exactly one user of this value.
Definition: Value.h:408
Predicate getSwappedPredicate() const
For example, EQ->EQ, SLE->SGE, ULT->UGT, OEQ->OEQ, ULE->UGE, OLT->OGT, etc.
Definition: InstrTypes.h:974
This is an entry in the NonLocalDepInfo cache.
A container for analyses that lazily runs them and caches their results.
BasicBlock * BB
BB - The basic block in question.
Definition: GVN.cpp:203
static void patchAndReplaceAllUsesWith(Instruction *I, Value *Repl)
Definition: GVN.cpp:1397
Legacy analysis pass which computes a DominatorTree.
Definition: Dominators.h:261
bool isMemIntrinValue() const
Definition: GVN.cpp:175
A wrapper pass to provide the legacy pass manager access to a suitably prepared AAResults object...
const TerminatorInst * getTerminator() const LLVM_READONLY
Returns the terminator instruction if the block is well formed or null if the block is not well forme...
Definition: BasicBlock.cpp:120
void setIncomingValue(unsigned i, Value *V)
AvailableValue AV
AV - The actual available value.
Definition: GVN.cpp:206
Value * SimplifyInstruction(Instruction *I, const SimplifyQuery &Q, OptimizationRemarkEmitter *ORE=nullptr)
See if we can compute a simplified version of this instruction.
static IntegerType * getInt8Ty(LLVMContext &C)
Definition: Type.cpp:174
void combineMetadata(Instruction *K, const Instruction *J, ArrayRef< unsigned > KnownIDs)
Combine the metadata of two instructions so that K can replace J.
Definition: Local.cpp:1703
The optimization diagnostic interface.
bool use_empty() const
Definition: Value.h:322
#define D
Definition: LargeTest.cpp:26
static AvailableValue getUndef()
Definition: GVN.cpp:165
static bool isOnlyReachableViaThisEdge(const BasicBlockEdge &E, DominatorTree *DT)
There is an edge from &#39;Src&#39; to &#39;Dst&#39;.
Definition: GVN.cpp:1486
A wrapper class for inspecting calls to intrinsic functions.
Definition: IntrinsicInst.h:44
const BasicBlock * getParent() const
Definition: Instruction.h:66
This instruction inserts a struct field of array element value into an aggregate value.
bool HasValueForBlock(BasicBlock *BB) const
Return true if the SSAUpdater already has a value for the specified block.
Definition: SSAUpdater.cpp:63
bool isCriticalEdge(const TerminatorInst *TI, unsigned SuccNum, bool AllowIdenticalEdges=false)
Return true if the specified edge is a critical edge.
Definition: CFG.cpp:88
MemDepResult getDependency(Instruction *QueryInst)
Returns the instruction on which a memory operation depends.