LLVM  10.0.0svn
MemorySSAUpdater.cpp
Go to the documentation of this file.
1 //===-- MemorySSAUpdater.cpp - Memory SSA Updater--------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------===//
8 //
9 // This file implements the MemorySSAUpdater class.
10 //
11 //===----------------------------------------------------------------===//
13 #include "llvm/ADT/STLExtras.h"
14 #include "llvm/ADT/SetVector.h"
15 #include "llvm/ADT/SmallPtrSet.h"
18 #include "llvm/IR/DataLayout.h"
19 #include "llvm/IR/Dominators.h"
20 #include "llvm/IR/GlobalVariable.h"
21 #include "llvm/IR/IRBuilder.h"
22 #include "llvm/IR/LLVMContext.h"
23 #include "llvm/IR/Metadata.h"
24 #include "llvm/IR/Module.h"
25 #include "llvm/Support/Debug.h"
27 #include <algorithm>
28 
29 #define DEBUG_TYPE "memoryssa"
30 using namespace llvm;
31 
32 // This is the marker algorithm from "Simple and Efficient Construction of
33 // Static Single Assignment Form"
34 // The simple, non-marker algorithm places phi nodes at any join
35 // Here, we place markers, and only place phi nodes if they end up necessary.
36 // They are only necessary if they break a cycle (IE we recursively visit
37 // ourselves again), or we discover, while getting the value of the operands,
38 // that there are two or more definitions needing to be merged.
39 // This still will leave non-minimal form in the case of irreducible control
40 // flow, where phi nodes may be in cycles with themselves, but unnecessary.
41 MemoryAccess *MemorySSAUpdater::getPreviousDefRecursive(
42  BasicBlock *BB,
43  DenseMap<BasicBlock *, TrackingVH<MemoryAccess>> &CachedPreviousDef) {
44  // First, do a cache lookup. Without this cache, certain CFG structures
45  // (like a series of if statements) take exponential time to visit.
46  auto Cached = CachedPreviousDef.find(BB);
47  if (Cached != CachedPreviousDef.end()) {
48  return Cached->second;
49  }
50 
51  if (BasicBlock *Pred = BB->getSinglePredecessor()) {
52  // Single predecessor case, just recurse, we can only have one definition.
53  MemoryAccess *Result = getPreviousDefFromEnd(Pred, CachedPreviousDef);
54  CachedPreviousDef.insert({BB, Result});
55  return Result;
56  }
57 
58  if (VisitedBlocks.count(BB)) {
59  // We hit our node again, meaning we had a cycle, we must insert a phi
60  // node to break it so we have an operand. The only case this will
61  // insert useless phis is if we have irreducible control flow.
62  MemoryAccess *Result = MSSA->createMemoryPhi(BB);
63  CachedPreviousDef.insert({BB, Result});
64  return Result;
65  }
66 
67  if (VisitedBlocks.insert(BB).second) {
68  // Mark us visited so we can detect a cycle
70 
71  // Recurse to get the values in our predecessors for placement of a
72  // potential phi node. This will insert phi nodes if we cycle in order to
73  // break the cycle and have an operand.
74  for (auto *Pred : predecessors(BB))
75  if (MSSA->DT->isReachableFromEntry(Pred))
76  PhiOps.push_back(getPreviousDefFromEnd(Pred, CachedPreviousDef));
77  else
78  PhiOps.push_back(MSSA->getLiveOnEntryDef());
79 
80  // Now try to simplify the ops to avoid placing a phi.
81  // This may return null if we never created a phi yet, that's okay
82  MemoryPhi *Phi = dyn_cast_or_null<MemoryPhi>(MSSA->getMemoryAccess(BB));
83 
84  // See if we can avoid the phi by simplifying it.
85  auto *Result = tryRemoveTrivialPhi(Phi, PhiOps);
86  // If we couldn't simplify, we may have to create a phi
87  if (Result == Phi) {
88  if (!Phi)
89  Phi = MSSA->createMemoryPhi(BB);
90 
91  // See if the existing phi operands match what we need.
92  // Unlike normal SSA, we only allow one phi node per block, so we can't just
93  // create a new one.
94  if (Phi->getNumOperands() != 0) {
95  // FIXME: Figure out whether this is dead code and if so remove it.
96  if (!std::equal(Phi->op_begin(), Phi->op_end(), PhiOps.begin())) {
97  // These will have been filled in by the recursive read we did above.
98  llvm::copy(PhiOps, Phi->op_begin());
99  std::copy(pred_begin(BB), pred_end(BB), Phi->block_begin());
100  }
101  } else {
102  unsigned i = 0;
103  for (auto *Pred : predecessors(BB))
104  Phi->addIncoming(&*PhiOps[i++], Pred);
105  InsertedPHIs.push_back(Phi);
106  }
107  Result = Phi;
108  }
109 
110  // Set ourselves up for the next variable by resetting visited state.
111  VisitedBlocks.erase(BB);
112  CachedPreviousDef.insert({BB, Result});
113  return Result;
114  }
115  llvm_unreachable("Should have hit one of the three cases above");
116 }
117 
118 // This starts at the memory access, and goes backwards in the block to find the
119 // previous definition. If a definition is not found the block of the access,
120 // it continues globally, creating phi nodes to ensure we have a single
121 // definition.
122 MemoryAccess *MemorySSAUpdater::getPreviousDef(MemoryAccess *MA) {
123  if (auto *LocalResult = getPreviousDefInBlock(MA))
124  return LocalResult;
126  return getPreviousDefRecursive(MA->getBlock(), CachedPreviousDef);
127 }
128 
129 // This starts at the memory access, and goes backwards in the block to the find
130 // the previous definition. If the definition is not found in the block of the
131 // access, it returns nullptr.
132 MemoryAccess *MemorySSAUpdater::getPreviousDefInBlock(MemoryAccess *MA) {
133  auto *Defs = MSSA->getWritableBlockDefs(MA->getBlock());
134 
135  // It's possible there are no defs, or we got handed the first def to start.
136  if (Defs) {
137  // If this is a def, we can just use the def iterators.
138  if (!isa<MemoryUse>(MA)) {
139  auto Iter = MA->getReverseDefsIterator();
140  ++Iter;
141  if (Iter != Defs->rend())
142  return &*Iter;
143  } else {
144  // Otherwise, have to walk the all access iterator.
145  auto End = MSSA->getWritableBlockAccesses(MA->getBlock())->rend();
146  for (auto &U : make_range(++MA->getReverseIterator(), End))
147  if (!isa<MemoryUse>(U))
148  return cast<MemoryAccess>(&U);
149  // Note that if MA comes before Defs->begin(), we won't hit a def.
150  return nullptr;
151  }
152  }
153  return nullptr;
154 }
155 
156 // This starts at the end of block
157 MemoryAccess *MemorySSAUpdater::getPreviousDefFromEnd(
158  BasicBlock *BB,
159  DenseMap<BasicBlock *, TrackingVH<MemoryAccess>> &CachedPreviousDef) {
160  auto *Defs = MSSA->getWritableBlockDefs(BB);
161 
162  if (Defs) {
163  CachedPreviousDef.insert({BB, &*Defs->rbegin()});
164  return &*Defs->rbegin();
165  }
166 
167  return getPreviousDefRecursive(BB, CachedPreviousDef);
168 }
169 // Recurse over a set of phi uses to eliminate the trivial ones
170 MemoryAccess *MemorySSAUpdater::recursePhi(MemoryAccess *Phi) {
171  if (!Phi)
172  return nullptr;
173  TrackingVH<MemoryAccess> Res(Phi);
175  std::copy(Phi->user_begin(), Phi->user_end(), std::back_inserter(Uses));
176  for (auto &U : Uses)
177  if (MemoryPhi *UsePhi = dyn_cast<MemoryPhi>(&*U))
178  tryRemoveTrivialPhi(UsePhi);
179  return Res;
180 }
181 
182 // Eliminate trivial phis
183 // Phis are trivial if they are defined either by themselves, or all the same
184 // argument.
185 // IE phi(a, a) or b = phi(a, b) or c = phi(a, a, c)
186 // We recursively try to remove them.
187 MemoryAccess *MemorySSAUpdater::tryRemoveTrivialPhi(MemoryPhi *Phi) {
188  assert(Phi && "Can only remove concrete Phi.");
189  auto OperRange = Phi->operands();
190  return tryRemoveTrivialPhi(Phi, OperRange);
191 }
192 template <class RangeType>
193 MemoryAccess *MemorySSAUpdater::tryRemoveTrivialPhi(MemoryPhi *Phi,
194  RangeType &Operands) {
195  // Bail out on non-opt Phis.
196  if (NonOptPhis.count(Phi))
197  return Phi;
198 
199  // Detect equal or self arguments
200  MemoryAccess *Same = nullptr;
201  for (auto &Op : Operands) {
202  // If the same or self, good so far
203  if (Op == Phi || Op == Same)
204  continue;
205  // not the same, return the phi since it's not eliminatable by us
206  if (Same)
207  return Phi;
208  Same = cast<MemoryAccess>(&*Op);
209  }
210  // Never found a non-self reference, the phi is undef
211  if (Same == nullptr)
212  return MSSA->getLiveOnEntryDef();
213  if (Phi) {
214  Phi->replaceAllUsesWith(Same);
215  removeMemoryAccess(Phi);
216  }
217 
218  // We should only end up recursing in case we replaced something, in which
219  // case, we may have made other Phis trivial.
220  return recursePhi(Same);
221 }
222 
223 void MemorySSAUpdater::insertUse(MemoryUse *MU, bool RenameUses) {
224  InsertedPHIs.clear();
225  MU->setDefiningAccess(getPreviousDef(MU));
226  // In cases without unreachable blocks, because uses do not create new
227  // may-defs, there are only two cases:
228  // 1. There was a def already below us, and therefore, we should not have
229  // created a phi node because it was already needed for the def.
230  //
231  // 2. There is no def below us, and therefore, there is no extra renaming work
232  // to do.
233 
234  // In cases with unreachable blocks, where the unnecessary Phis were
235  // optimized out, adding the Use may re-insert those Phis. Hence, when
236  // inserting Uses outside of the MSSA creation process, and new Phis were
237  // added, rename all uses if we are asked.
238 
239  if (!RenameUses && !InsertedPHIs.empty()) {
240  auto *Defs = MSSA->getBlockDefs(MU->getBlock());
241  (void)Defs;
242  assert((!Defs || (++Defs->begin() == Defs->end())) &&
243  "Block may have only a Phi or no defs");
244  }
245 
246  if (RenameUses && InsertedPHIs.size()) {
248  BasicBlock *StartBlock = MU->getBlock();
249 
250  if (auto *Defs = MSSA->getWritableBlockDefs(StartBlock)) {
251  MemoryAccess *FirstDef = &*Defs->begin();
252  // Convert to incoming value if it's a memorydef. A phi *is* already an
253  // incoming value.
254  if (auto *MD = dyn_cast<MemoryDef>(FirstDef))
255  FirstDef = MD->getDefiningAccess();
256 
257  MSSA->renamePass(MU->getBlock(), FirstDef, Visited);
258  }
259  // We just inserted a phi into this block, so the incoming value will
260  // become the phi anyway, so it does not matter what we pass.
261  for (auto &MP : InsertedPHIs)
262  if (MemoryPhi *Phi = cast_or_null<MemoryPhi>(MP))
263  MSSA->renamePass(Phi->getBlock(), nullptr, Visited);
264  }
265 }
266 
267 // Set every incoming edge {BB, MP->getBlock()} of MemoryPhi MP to NewDef.
268 static void setMemoryPhiValueForBlock(MemoryPhi *MP, const BasicBlock *BB,
269  MemoryAccess *NewDef) {
270  // Replace any operand with us an incoming block with the new defining
271  // access.
272  int i = MP->getBasicBlockIndex(BB);
273  assert(i != -1 && "Should have found the basic block in the phi");
274  // We can't just compare i against getNumOperands since one is signed and the
275  // other not. So use it to index into the block iterator.
276  for (auto BBIter = MP->block_begin() + i; BBIter != MP->block_end();
277  ++BBIter) {
278  if (*BBIter != BB)
279  break;
280  MP->setIncomingValue(i, NewDef);
281  ++i;
282  }
283 }
284 
285 // A brief description of the algorithm:
286 // First, we compute what should define the new def, using the SSA
287 // construction algorithm.
288 // Then, we update the defs below us (and any new phi nodes) in the graph to
289 // point to the correct new defs, to ensure we only have one variable, and no
290 // disconnected stores.
291 void MemorySSAUpdater::insertDef(MemoryDef *MD, bool RenameUses) {
292  InsertedPHIs.clear();
293 
294  // See if we had a local def, and if not, go hunting.
295  MemoryAccess *DefBefore = getPreviousDef(MD);
296  bool DefBeforeSameBlock = DefBefore->getBlock() == MD->getBlock();
297 
298  // There is a def before us, which means we can replace any store/phi uses
299  // of that thing with us, since we are in the way of whatever was there
300  // before.
301  // We now define that def's memorydefs and memoryphis
302  if (DefBeforeSameBlock) {
303  DefBefore->replaceUsesWithIf(MD, [MD](Use &U) {
304  // Leave the MemoryUses alone.
305  // Also make sure we skip ourselves to avoid self references.
306  User *Usr = U.getUser();
307  return !isa<MemoryUse>(Usr) && Usr != MD;
308  // Defs are automatically unoptimized when the user is set to MD below,
309  // because the isOptimized() call will fail to find the same ID.
310  });
311  }
312 
313  // and that def is now our defining access.
314  MD->setDefiningAccess(DefBefore);
315 
316  // Remember the index where we may insert new phis below.
317  unsigned NewPhiIndex = InsertedPHIs.size();
318 
319  SmallVector<WeakVH, 8> FixupList(InsertedPHIs.begin(), InsertedPHIs.end());
320  if (!DefBeforeSameBlock) {
321  // If there was a local def before us, we must have the same effect it
322  // did. Because every may-def is the same, any phis/etc we would create, it
323  // would also have created. If there was no local def before us, we
324  // performed a global update, and have to search all successors and make
325  // sure we update the first def in each of them (following all paths until
326  // we hit the first def along each path). This may also insert phi nodes.
327  // TODO: There are other cases we can skip this work, such as when we have a
328  // single successor, and only used a straight line of single pred blocks
329  // backwards to find the def. To make that work, we'd have to track whether
330  // getDefRecursive only ever used the single predecessor case. These types
331  // of paths also only exist in between CFG simplifications.
332 
333  // If this is the first def in the block and this insert is in an arbitrary
334  // place, compute IDF and place phis.
335  auto Iter = MD->getDefsIterator();
336  ++Iter;
337  auto IterEnd = MSSA->getBlockDefs(MD->getBlock())->end();
338  if (Iter == IterEnd) {
339  ForwardIDFCalculator IDFs(*MSSA->DT);
341  SmallPtrSet<BasicBlock *, 2> DefiningBlocks;
342  for (const auto &VH : InsertedPHIs)
343  if (const auto *RealPHI = cast_or_null<MemoryPhi>(VH))
344  DefiningBlocks.insert(RealPHI->getBlock());
345  DefiningBlocks.insert(MD->getBlock());
346  IDFs.setDefiningBlocks(DefiningBlocks);
347  IDFs.calculate(IDFBlocks);
348  SmallVector<AssertingVH<MemoryPhi>, 4> NewInsertedPHIs;
349  for (auto *BBIDF : IDFBlocks) {
350  auto *MPhi = MSSA->getMemoryAccess(BBIDF);
351  if (!MPhi) {
352  MPhi = MSSA->createMemoryPhi(BBIDF);
353  NewInsertedPHIs.push_back(MPhi);
354  }
355  // Add the phis created into the IDF blocks to NonOptPhis, so they are
356  // not optimized out as trivial by the call to getPreviousDefFromEnd
357  // below. Once they are complete, all these Phis are added to the
358  // FixupList, and removed from NonOptPhis inside fixupDefs().
359  // Existing Phis in IDF may need fixing as well, and potentially be
360  // trivial before this insertion, hence add all IDF Phis. See PR43044.
361  NonOptPhis.insert(MPhi);
362  }
363 
364  for (auto &MPhi : NewInsertedPHIs) {
365  auto *BBIDF = MPhi->getBlock();
366  for (auto *Pred : predecessors(BBIDF)) {
368  MPhi->addIncoming(getPreviousDefFromEnd(Pred, CachedPreviousDef),
369  Pred);
370  }
371  }
372 
373  // Re-take the index where we're adding the new phis, because the above
374  // call to getPreviousDefFromEnd, may have inserted into InsertedPHIs.
375  NewPhiIndex = InsertedPHIs.size();
376  for (auto &MPhi : NewInsertedPHIs) {
377  InsertedPHIs.push_back(&*MPhi);
378  FixupList.push_back(&*MPhi);
379  }
380  }
381 
382  FixupList.push_back(MD);
383  }
384 
385  // Remember the index where we stopped inserting new phis above, since the
386  // fixupDefs call in the loop below may insert more, that are already minimal.
387  unsigned NewPhiIndexEnd = InsertedPHIs.size();
388 
389  while (!FixupList.empty()) {
390  unsigned StartingPHISize = InsertedPHIs.size();
391  fixupDefs(FixupList);
392  FixupList.clear();
393  // Put any new phis on the fixup list, and process them
394  FixupList.append(InsertedPHIs.begin() + StartingPHISize, InsertedPHIs.end());
395  }
396 
397  // Optimize potentially non-minimal phis added in this method.
398  unsigned NewPhiSize = NewPhiIndexEnd - NewPhiIndex;
399  if (NewPhiSize)
400  tryRemoveTrivialPhis(ArrayRef<WeakVH>(&InsertedPHIs[NewPhiIndex], NewPhiSize));
401 
402  // Now that all fixups are done, rename all uses if we are asked.
403  if (RenameUses) {
405  BasicBlock *StartBlock = MD->getBlock();
406  // We are guaranteed there is a def in the block, because we just got it
407  // handed to us in this function.
408  MemoryAccess *FirstDef = &*MSSA->getWritableBlockDefs(StartBlock)->begin();
409  // Convert to incoming value if it's a memorydef. A phi *is* already an
410  // incoming value.
411  if (auto *MD = dyn_cast<MemoryDef>(FirstDef))
412  FirstDef = MD->getDefiningAccess();
413 
414  MSSA->renamePass(MD->getBlock(), FirstDef, Visited);
415  // We just inserted a phi into this block, so the incoming value will become
416  // the phi anyway, so it does not matter what we pass.
417  for (auto &MP : InsertedPHIs) {
418  MemoryPhi *Phi = dyn_cast_or_null<MemoryPhi>(MP);
419  if (Phi)
420  MSSA->renamePass(Phi->getBlock(), nullptr, Visited);
421  }
422  }
423 }
424 
425 void MemorySSAUpdater::fixupDefs(const SmallVectorImpl<WeakVH> &Vars) {
428  for (auto &Var : Vars) {
429  MemoryAccess *NewDef = dyn_cast_or_null<MemoryAccess>(Var);
430  if (!NewDef)
431  continue;
432  // First, see if there is a local def after the operand.
433  auto *Defs = MSSA->getWritableBlockDefs(NewDef->getBlock());
434  auto DefIter = NewDef->getDefsIterator();
435 
436  // The temporary Phi is being fixed, unmark it for not to optimize.
437  if (MemoryPhi *Phi = dyn_cast<MemoryPhi>(NewDef))
438  NonOptPhis.erase(Phi);
439 
440  // If there is a local def after us, we only have to rename that.
441  if (++DefIter != Defs->end()) {
442  cast<MemoryDef>(DefIter)->setDefiningAccess(NewDef);
443  continue;
444  }
445 
446  // Otherwise, we need to search down through the CFG.
447  // For each of our successors, handle it directly if their is a phi, or
448  // place on the fixup worklist.
449  for (const auto *S : successors(NewDef->getBlock())) {
450  if (auto *MP = MSSA->getMemoryAccess(S))
451  setMemoryPhiValueForBlock(MP, NewDef->getBlock(), NewDef);
452  else
453  Worklist.push_back(S);
454  }
455 
456  while (!Worklist.empty()) {
457  const BasicBlock *FixupBlock = Worklist.back();
458  Worklist.pop_back();
459 
460  // Get the first def in the block that isn't a phi node.
461  if (auto *Defs = MSSA->getWritableBlockDefs(FixupBlock)) {
462  auto *FirstDef = &*Defs->begin();
463  // The loop above and below should have taken care of phi nodes
464  assert(!isa<MemoryPhi>(FirstDef) &&
465  "Should have already handled phi nodes!");
466  // We are now this def's defining access, make sure we actually dominate
467  // it
468  assert(MSSA->dominates(NewDef, FirstDef) &&
469  "Should have dominated the new access");
470 
471  // This may insert new phi nodes, because we are not guaranteed the
472  // block we are processing has a single pred, and depending where the
473  // store was inserted, it may require phi nodes below it.
474  cast<MemoryDef>(FirstDef)->setDefiningAccess(getPreviousDef(FirstDef));
475  return;
476  }
477  // We didn't find a def, so we must continue.
478  for (const auto *S : successors(FixupBlock)) {
479  // If there is a phi node, handle it.
480  // Otherwise, put the block on the worklist
481  if (auto *MP = MSSA->getMemoryAccess(S))
482  setMemoryPhiValueForBlock(MP, FixupBlock, NewDef);
483  else {
484  // If we cycle, we should have ended up at a phi node that we already
485  // processed. FIXME: Double check this
486  if (!Seen.insert(S).second)
487  continue;
488  Worklist.push_back(S);
489  }
490  }
491  }
492  }
493 }
494 
496  if (MemoryPhi *MPhi = MSSA->getMemoryAccess(To)) {
497  MPhi->unorderedDeleteIncomingBlock(From);
498  tryRemoveTrivialPhi(MPhi);
499  }
500 }
501 
503  const BasicBlock *To) {
504  if (MemoryPhi *MPhi = MSSA->getMemoryAccess(To)) {
505  bool Found = false;
506  MPhi->unorderedDeleteIncomingIf([&](const MemoryAccess *, BasicBlock *B) {
507  if (From != B)
508  return false;
509  if (Found)
510  return true;
511  Found = true;
512  return false;
513  });
514  tryRemoveTrivialPhi(MPhi);
515  }
516 }
517 
519  const ValueToValueMapTy &VMap,
520  PhiToDefMap &MPhiMap,
521  bool CloneWasSimplified,
522  MemorySSA *MSSA) {
523  MemoryAccess *InsnDefining = MA;
524  if (MemoryDef *DefMUD = dyn_cast<MemoryDef>(InsnDefining)) {
525  if (!MSSA->isLiveOnEntryDef(DefMUD)) {
526  Instruction *DefMUDI = DefMUD->getMemoryInst();
527  assert(DefMUDI && "Found MemoryUseOrDef with no Instruction.");
528  if (Instruction *NewDefMUDI =
529  cast_or_null<Instruction>(VMap.lookup(DefMUDI))) {
530  InsnDefining = MSSA->getMemoryAccess(NewDefMUDI);
531  if (!CloneWasSimplified)
532  assert(InsnDefining && "Defining instruction cannot be nullptr.");
533  else if (!InsnDefining || isa<MemoryUse>(InsnDefining)) {
534  // The clone was simplified, it's no longer a MemoryDef, look up.
535  auto DefIt = DefMUD->getDefsIterator();
536  // Since simplified clones only occur in single block cloning, a
537  // previous definition must exist, otherwise NewDefMUDI would not
538  // have been found in VMap.
539  assert(DefIt != MSSA->getBlockDefs(DefMUD->getBlock())->begin() &&
540  "Previous def must exist");
541  InsnDefining = getNewDefiningAccessForClone(
542  &*(--DefIt), VMap, MPhiMap, CloneWasSimplified, MSSA);
543  }
544  }
545  }
546  } else {
547  MemoryPhi *DefPhi = cast<MemoryPhi>(InsnDefining);
548  if (MemoryAccess *NewDefPhi = MPhiMap.lookup(DefPhi))
549  InsnDefining = NewDefPhi;
550  }
551  assert(InsnDefining && "Defining instruction cannot be nullptr.");
552  return InsnDefining;
553 }
554 
555 void MemorySSAUpdater::cloneUsesAndDefs(BasicBlock *BB, BasicBlock *NewBB,
556  const ValueToValueMapTy &VMap,
557  PhiToDefMap &MPhiMap,
558  bool CloneWasSimplified) {
559  const MemorySSA::AccessList *Acc = MSSA->getBlockAccesses(BB);
560  if (!Acc)
561  return;
562  for (const MemoryAccess &MA : *Acc) {
563  if (const MemoryUseOrDef *MUD = dyn_cast<MemoryUseOrDef>(&MA)) {
564  Instruction *Insn = MUD->getMemoryInst();
565  // Entry does not exist if the clone of the block did not clone all
566  // instructions. This occurs in LoopRotate when cloning instructions
567  // from the old header to the old preheader. The cloned instruction may
568  // also be a simplified Value, not an Instruction (see LoopRotate).
569  // Also in LoopRotate, even when it's an instruction, due to it being
570  // simplified, it may be a Use rather than a Def, so we cannot use MUD as
571  // template. Calls coming from updateForClonedBlockIntoPred, ensure this.
572  if (Instruction *NewInsn =
573  dyn_cast_or_null<Instruction>(VMap.lookup(Insn))) {
574  MemoryAccess *NewUseOrDef = MSSA->createDefinedAccess(
575  NewInsn,
576  getNewDefiningAccessForClone(MUD->getDefiningAccess(), VMap,
577  MPhiMap, CloneWasSimplified, MSSA),
578  /*Template=*/CloneWasSimplified ? nullptr : MUD,
579  /*CreationMustSucceed=*/CloneWasSimplified ? false : true);
580  if (NewUseOrDef)
581  MSSA->insertIntoListsForBlock(NewUseOrDef, NewBB, MemorySSA::End);
582  }
583  }
584  }
585 }
586 
588  BasicBlock *Header, BasicBlock *Preheader, BasicBlock *BEBlock) {
589  auto *MPhi = MSSA->getMemoryAccess(Header);
590  if (!MPhi)
591  return;
592 
593  // Create phi node in the backedge block and populate it with the same
594  // incoming values as MPhi. Skip incoming values coming from Preheader.
595  auto *NewMPhi = MSSA->createMemoryPhi(BEBlock);
596  bool HasUniqueIncomingValue = true;
597  MemoryAccess *UniqueValue = nullptr;
598  for (unsigned I = 0, E = MPhi->getNumIncomingValues(); I != E; ++I) {
599  BasicBlock *IBB = MPhi->getIncomingBlock(I);
600  MemoryAccess *IV = MPhi->getIncomingValue(I);
601  if (IBB != Preheader) {
602  NewMPhi->addIncoming(IV, IBB);
603  if (HasUniqueIncomingValue) {
604  if (!UniqueValue)
605  UniqueValue = IV;
606  else if (UniqueValue != IV)
607  HasUniqueIncomingValue = false;
608  }
609  }
610  }
611 
612  // Update incoming edges into MPhi. Remove all but the incoming edge from
613  // Preheader. Add an edge from NewMPhi
614  auto *AccFromPreheader = MPhi->getIncomingValueForBlock(Preheader);
615  MPhi->setIncomingValue(0, AccFromPreheader);
616  MPhi->setIncomingBlock(0, Preheader);
617  for (unsigned I = MPhi->getNumIncomingValues() - 1; I >= 1; --I)
618  MPhi->unorderedDeleteIncoming(I);
619  MPhi->addIncoming(NewMPhi, BEBlock);
620 
621  // If NewMPhi is a trivial phi, remove it. Its use in the header MPhi will be
622  // replaced with the unique value.
623  tryRemoveTrivialPhi(MPhi);
624 }
625 
627  ArrayRef<BasicBlock *> ExitBlocks,
628  const ValueToValueMapTy &VMap,
629  bool IgnoreIncomingWithNoClones) {
630  PhiToDefMap MPhiMap;
631 
632  auto FixPhiIncomingValues = [&](MemoryPhi *Phi, MemoryPhi *NewPhi) {
633  assert(Phi && NewPhi && "Invalid Phi nodes.");
634  BasicBlock *NewPhiBB = NewPhi->getBlock();
635  SmallPtrSet<BasicBlock *, 4> NewPhiBBPreds(pred_begin(NewPhiBB),
636  pred_end(NewPhiBB));
637  for (unsigned It = 0, E = Phi->getNumIncomingValues(); It < E; ++It) {
638  MemoryAccess *IncomingAccess = Phi->getIncomingValue(It);
639  BasicBlock *IncBB = Phi->getIncomingBlock(It);
640 
641  if (BasicBlock *NewIncBB = cast_or_null<BasicBlock>(VMap.lookup(IncBB)))
642  IncBB = NewIncBB;
643  else if (IgnoreIncomingWithNoClones)
644  continue;
645 
646  // Now we have IncBB, and will need to add incoming from it to NewPhi.
647 
648  // If IncBB is not a predecessor of NewPhiBB, then do not add it.
649  // NewPhiBB was cloned without that edge.
650  if (!NewPhiBBPreds.count(IncBB))
651  continue;
652 
653  // Determine incoming value and add it as incoming from IncBB.
654  if (MemoryUseOrDef *IncMUD = dyn_cast<MemoryUseOrDef>(IncomingAccess)) {
655  if (!MSSA->isLiveOnEntryDef(IncMUD)) {
656  Instruction *IncI = IncMUD->getMemoryInst();
657  assert(IncI && "Found MemoryUseOrDef with no Instruction.");
658  if (Instruction *NewIncI =
659  cast_or_null<Instruction>(VMap.lookup(IncI))) {
660  IncMUD = MSSA->getMemoryAccess(NewIncI);
661  assert(IncMUD &&
662  "MemoryUseOrDef cannot be null, all preds processed.");
663  }
664  }
665  NewPhi->addIncoming(IncMUD, IncBB);
666  } else {
667  MemoryPhi *IncPhi = cast<MemoryPhi>(IncomingAccess);
668  if (MemoryAccess *NewDefPhi = MPhiMap.lookup(IncPhi))
669  NewPhi->addIncoming(NewDefPhi, IncBB);
670  else
671  NewPhi->addIncoming(IncPhi, IncBB);
672  }
673  }
674  };
675 
676  auto ProcessBlock = [&](BasicBlock *BB) {
677  BasicBlock *NewBlock = cast_or_null<BasicBlock>(VMap.lookup(BB));
678  if (!NewBlock)
679  return;
680 
681  assert(!MSSA->getWritableBlockAccesses(NewBlock) &&
682  "Cloned block should have no accesses");
683 
684  // Add MemoryPhi.
685  if (MemoryPhi *MPhi = MSSA->getMemoryAccess(BB)) {
686  MemoryPhi *NewPhi = MSSA->createMemoryPhi(NewBlock);
687  MPhiMap[MPhi] = NewPhi;
688  }
689  // Update Uses and Defs.
690  cloneUsesAndDefs(BB, NewBlock, VMap, MPhiMap);
691  };
692 
693  for (auto BB : llvm::concat<BasicBlock *const>(LoopBlocks, ExitBlocks))
694  ProcessBlock(BB);
695 
696  for (auto BB : llvm::concat<BasicBlock *const>(LoopBlocks, ExitBlocks))
697  if (MemoryPhi *MPhi = MSSA->getMemoryAccess(BB))
698  if (MemoryAccess *NewPhi = MPhiMap.lookup(MPhi))
699  FixPhiIncomingValues(MPhi, cast<MemoryPhi>(NewPhi));
700 }
701 
703  BasicBlock *BB, BasicBlock *P1, const ValueToValueMapTy &VM) {
704  // All defs/phis from outside BB that are used in BB, are valid uses in P1.
705  // Since those defs/phis must have dominated BB, and also dominate P1.
706  // Defs from BB being used in BB will be replaced with the cloned defs from
707  // VM. The uses of BB's Phi (if it exists) in BB will be replaced by the
708  // incoming def into the Phi from P1.
709  // Instructions cloned into the predecessor are in practice sometimes
710  // simplified, so disable the use of the template, and create an access from
711  // scratch.
712  PhiToDefMap MPhiMap;
713  if (MemoryPhi *MPhi = MSSA->getMemoryAccess(BB))
714  MPhiMap[MPhi] = MPhi->getIncomingValueForBlock(P1);
715  cloneUsesAndDefs(BB, P1, VM, MPhiMap, /*CloneWasSimplified=*/true);
716 }
717 
718 template <typename Iter>
719 void MemorySSAUpdater::privateUpdateExitBlocksForClonedLoop(
720  ArrayRef<BasicBlock *> ExitBlocks, Iter ValuesBegin, Iter ValuesEnd,
721  DominatorTree &DT) {
723  // Update/insert phis in all successors of exit blocks.
724  for (auto *Exit : ExitBlocks)
725  for (const ValueToValueMapTy *VMap : make_range(ValuesBegin, ValuesEnd))
726  if (BasicBlock *NewExit = cast_or_null<BasicBlock>(VMap->lookup(Exit))) {
727  BasicBlock *ExitSucc = NewExit->getTerminator()->getSuccessor(0);
728  Updates.push_back({DT.Insert, NewExit, ExitSucc});
729  }
730  applyInsertUpdates(Updates, DT);
731 }
732 
734  ArrayRef<BasicBlock *> ExitBlocks, const ValueToValueMapTy &VMap,
735  DominatorTree &DT) {
736  const ValueToValueMapTy *const Arr[] = {&VMap};
737  privateUpdateExitBlocksForClonedLoop(ExitBlocks, std::begin(Arr),
738  std::end(Arr), DT);
739 }
740 
742  ArrayRef<BasicBlock *> ExitBlocks,
743  ArrayRef<std::unique_ptr<ValueToValueMapTy>> VMaps, DominatorTree &DT) {
744  auto GetPtr = [&](const std::unique_ptr<ValueToValueMapTy> &I) {
745  return I.get();
746  };
747  using MappedIteratorType =
749  decltype(GetPtr)>;
750  auto MapBegin = MappedIteratorType(VMaps.begin(), GetPtr);
751  auto MapEnd = MappedIteratorType(VMaps.end(), GetPtr);
752  privateUpdateExitBlocksForClonedLoop(ExitBlocks, MapBegin, MapEnd, DT);
753 }
754 
756  DominatorTree &DT) {
757  SmallVector<CFGUpdate, 4> RevDeleteUpdates;
758  SmallVector<CFGUpdate, 4> InsertUpdates;
759  for (auto &Update : Updates) {
760  if (Update.getKind() == DT.Insert)
761  InsertUpdates.push_back({DT.Insert, Update.getFrom(), Update.getTo()});
762  else
763  RevDeleteUpdates.push_back({DT.Insert, Update.getFrom(), Update.getTo()});
764  }
765 
766  if (!RevDeleteUpdates.empty()) {
767  // Update for inserted edges: use newDT and snapshot CFG as if deletes had
768  // not occurred.
769  // FIXME: This creates a new DT, so it's more expensive to do mix
770  // delete/inserts vs just inserts. We can do an incremental update on the DT
771  // to revert deletes, than re-delete the edges. Teaching DT to do this, is
772  // part of a pending cleanup.
773  DominatorTree NewDT(DT, RevDeleteUpdates);
774  GraphDiff<BasicBlock *> GD(RevDeleteUpdates);
775  applyInsertUpdates(InsertUpdates, NewDT, &GD);
776  } else {
778  applyInsertUpdates(InsertUpdates, DT, &GD);
779  }
780 
781  // Update for deleted edges
782  for (auto &Update : RevDeleteUpdates)
783  removeEdge(Update.getFrom(), Update.getTo());
784 }
785 
787  DominatorTree &DT) {
789  applyInsertUpdates(Updates, DT, &GD);
790 }
791 
793  DominatorTree &DT,
794  const GraphDiff<BasicBlock *> *GD) {
795  // Get recursive last Def, assuming well formed MSSA and updated DT.
796  auto GetLastDef = [&](BasicBlock *BB) -> MemoryAccess * {
797  while (true) {
798  MemorySSA::DefsList *Defs = MSSA->getWritableBlockDefs(BB);
799  // Return last Def or Phi in BB, if it exists.
800  if (Defs)
801  return &*(--Defs->end());
802 
803  // Check number of predecessors, we only care if there's more than one.
804  unsigned Count = 0;
805  BasicBlock *Pred = nullptr;
806  for (auto &Pair : children<GraphDiffInvBBPair>({GD, BB})) {
807  Pred = Pair.second;
808  Count++;
809  if (Count == 2)
810  break;
811  }
812 
813  // If BB has multiple predecessors, get last definition from IDom.
814  if (Count != 1) {
815  // [SimpleLoopUnswitch] If BB is a dead block, about to be deleted, its
816  // DT is invalidated. Return LoE as its last def. This will be added to
817  // MemoryPhi node, and later deleted when the block is deleted.
818  if (!DT.getNode(BB))
819  return MSSA->getLiveOnEntryDef();
820  if (auto *IDom = DT.getNode(BB)->getIDom())
821  if (IDom->getBlock() != BB) {
822  BB = IDom->getBlock();
823  continue;
824  }
825  return MSSA->getLiveOnEntryDef();
826  } else {
827  // Single predecessor, BB cannot be dead. GetLastDef of Pred.
828  assert(Count == 1 && Pred && "Single predecessor expected.");
829  BB = Pred;
830  }
831  };
832  llvm_unreachable("Unable to get last definition.");
833  };
834 
835  // Get nearest IDom given a set of blocks.
836  // TODO: this can be optimized by starting the search at the node with the
837  // lowest level (highest in the tree).
838  auto FindNearestCommonDominator =
839  [&](const SmallSetVector<BasicBlock *, 2> &BBSet) -> BasicBlock * {
840  BasicBlock *PrevIDom = *BBSet.begin();
841  for (auto *BB : BBSet)
842  PrevIDom = DT.findNearestCommonDominator(PrevIDom, BB);
843  return PrevIDom;
844  };
845 
846  // Get all blocks that dominate PrevIDom, stop when reaching CurrIDom. Do not
847  // include CurrIDom.
848  auto GetNoLongerDomBlocks =
849  [&](BasicBlock *PrevIDom, BasicBlock *CurrIDom,
850  SmallVectorImpl<BasicBlock *> &BlocksPrevDom) {
851  if (PrevIDom == CurrIDom)
852  return;
853  BlocksPrevDom.push_back(PrevIDom);
854  BasicBlock *NextIDom = PrevIDom;
855  while (BasicBlock *UpIDom =
856  DT.getNode(NextIDom)->getIDom()->getBlock()) {
857  if (UpIDom == CurrIDom)
858  break;
859  BlocksPrevDom.push_back(UpIDom);
860  NextIDom = UpIDom;
861  }
862  };
863 
864  // Map a BB to its predecessors: added + previously existing. To get a
865  // deterministic order, store predecessors as SetVectors. The order in each
866  // will be defined by the order in Updates (fixed) and the order given by
867  // children<> (also fixed). Since we further iterate over these ordered sets,
868  // we lose the information of multiple edges possibly existing between two
869  // blocks, so we'll keep and EdgeCount map for that.
870  // An alternate implementation could keep unordered set for the predecessors,
871  // traverse either Updates or children<> each time to get the deterministic
872  // order, and drop the usage of EdgeCount. This alternate approach would still
873  // require querying the maps for each predecessor, and children<> call has
874  // additional computation inside for creating the snapshot-graph predecessors.
875  // As such, we favor using a little additional storage and less compute time.
876  // This decision can be revisited if we find the alternative more favorable.
877 
878  struct PredInfo {
881  };
883 
884  for (auto &Edge : Updates) {
885  BasicBlock *BB = Edge.getTo();
886  auto &AddedBlockSet = PredMap[BB].Added;
887  AddedBlockSet.insert(Edge.getFrom());
888  }
889 
890  // Store all existing predecessor for each BB, at least one must exist.
893  for (auto &BBPredPair : PredMap) {
894  auto *BB = BBPredPair.first;
895  const auto &AddedBlockSet = BBPredPair.second.Added;
896  auto &PrevBlockSet = BBPredPair.second.Prev;
897  for (auto &Pair : children<GraphDiffInvBBPair>({GD, BB})) {
898  BasicBlock *Pi = Pair.second;
899  if (!AddedBlockSet.count(Pi))
900  PrevBlockSet.insert(Pi);
901  EdgeCountMap[{Pi, BB}]++;
902  }
903 
904  if (PrevBlockSet.empty()) {
905  assert(pred_size(BB) == AddedBlockSet.size() && "Duplicate edges added.");
906  LLVM_DEBUG(
907  dbgs()
908  << "Adding a predecessor to a block with no predecessors. "
909  "This must be an edge added to a new, likely cloned, block. "
910  "Its memory accesses must be already correct, assuming completed "
911  "via the updateExitBlocksForClonedLoop API. "
912  "Assert a single such edge is added so no phi addition or "
913  "additional processing is required.\n");
914  assert(AddedBlockSet.size() == 1 &&
915  "Can only handle adding one predecessor to a new block.");
916  // Need to remove new blocks from PredMap. Remove below to not invalidate
917  // iterator here.
918  NewBlocks.insert(BB);
919  }
920  }
921  // Nothing to process for new/cloned blocks.
922  for (auto *BB : NewBlocks)
923  PredMap.erase(BB);
924 
925  SmallVector<BasicBlock *, 16> BlocksWithDefsToReplace;
926  SmallVector<WeakVH, 8> InsertedPhis;
927 
928  // First create MemoryPhis in all blocks that don't have one. Create in the
929  // order found in Updates, not in PredMap, to get deterministic numbering.
930  for (auto &Edge : Updates) {
931  BasicBlock *BB = Edge.getTo();
932  if (PredMap.count(BB) && !MSSA->getMemoryAccess(BB))
933  InsertedPhis.push_back(MSSA->createMemoryPhi(BB));
934  }
935 
936  // Now we'll fill in the MemoryPhis with the right incoming values.
937  for (auto &BBPredPair : PredMap) {
938  auto *BB = BBPredPair.first;
939  const auto &PrevBlockSet = BBPredPair.second.Prev;
940  const auto &AddedBlockSet = BBPredPair.second.Added;
941  assert(!PrevBlockSet.empty() &&
942  "At least one previous predecessor must exist.");
943 
944  // TODO: if this becomes a bottleneck, we can save on GetLastDef calls by
945  // keeping this map before the loop. We can reuse already populated entries
946  // if an edge is added from the same predecessor to two different blocks,
947  // and this does happen in rotate. Note that the map needs to be updated
948  // when deleting non-necessary phis below, if the phi is in the map by
949  // replacing the value with DefP1.
951  for (auto *AddedPred : AddedBlockSet) {
952  auto *DefPn = GetLastDef(AddedPred);
953  assert(DefPn != nullptr && "Unable to find last definition.");
954  LastDefAddedPred[AddedPred] = DefPn;
955  }
956 
957  MemoryPhi *NewPhi = MSSA->getMemoryAccess(BB);
958  // If Phi is not empty, add an incoming edge from each added pred. Must
959  // still compute blocks with defs to replace for this block below.
960  if (NewPhi->getNumOperands()) {
961  for (auto *Pred : AddedBlockSet) {
962  auto *LastDefForPred = LastDefAddedPred[Pred];
963  for (int I = 0, E = EdgeCountMap[{Pred, BB}]; I < E; ++I)
964  NewPhi->addIncoming(LastDefForPred, Pred);
965  }
966  } else {
967  // Pick any existing predecessor and get its definition. All other
968  // existing predecessors should have the same one, since no phi existed.
969  auto *P1 = *PrevBlockSet.begin();
970  MemoryAccess *DefP1 = GetLastDef(P1);
971 
972  // Check DefP1 against all Defs in LastDefPredPair. If all the same,
973  // nothing to add.
974  bool InsertPhi = false;
975  for (auto LastDefPredPair : LastDefAddedPred)
976  if (DefP1 != LastDefPredPair.second) {
977  InsertPhi = true;
978  break;
979  }
980  if (!InsertPhi) {
981  // Since NewPhi may be used in other newly added Phis, replace all uses
982  // of NewPhi with the definition coming from all predecessors (DefP1),
983  // before deleting it.
984  NewPhi->replaceAllUsesWith(DefP1);
985  removeMemoryAccess(NewPhi);
986  continue;
987  }
988 
989  // Update Phi with new values for new predecessors and old value for all
990  // other predecessors. Since AddedBlockSet and PrevBlockSet are ordered
991  // sets, the order of entries in NewPhi is deterministic.
992  for (auto *Pred : AddedBlockSet) {
993  auto *LastDefForPred = LastDefAddedPred[Pred];
994  for (int I = 0, E = EdgeCountMap[{Pred, BB}]; I < E; ++I)
995  NewPhi->addIncoming(LastDefForPred, Pred);
996  }
997  for (auto *Pred : PrevBlockSet)
998  for (int I = 0, E = EdgeCountMap[{Pred, BB}]; I < E; ++I)
999  NewPhi->addIncoming(DefP1, Pred);
1000  }
1001 
1002  // Get all blocks that used to dominate BB and no longer do after adding
1003  // AddedBlockSet, where PrevBlockSet are the previously known predecessors.
1004  assert(DT.getNode(BB)->getIDom() && "BB does not have valid idom");
1005  BasicBlock *PrevIDom = FindNearestCommonDominator(PrevBlockSet);
1006  assert(PrevIDom && "Previous IDom should exists");
1007  BasicBlock *NewIDom = DT.getNode(BB)->getIDom()->getBlock();
1008  assert(NewIDom && "BB should have a new valid idom");
1009  assert(DT.dominates(NewIDom, PrevIDom) &&
1010  "New idom should dominate old idom");
1011  GetNoLongerDomBlocks(PrevIDom, NewIDom, BlocksWithDefsToReplace);
1012  }
1013 
1014  tryRemoveTrivialPhis(InsertedPhis);
1015  // Create the set of blocks that now have a definition. We'll use this to
1016  // compute IDF and add Phis there next.
1017  SmallVector<BasicBlock *, 8> BlocksToProcess;
1018  for (auto &VH : InsertedPhis)
1019  if (auto *MPhi = cast_or_null<MemoryPhi>(VH))
1020  BlocksToProcess.push_back(MPhi->getBlock());
1021 
1022  // Compute IDF and add Phis in all IDF blocks that do not have one.
1024  if (!BlocksToProcess.empty()) {
1025  ForwardIDFCalculator IDFs(DT, GD);
1026  SmallPtrSet<BasicBlock *, 16> DefiningBlocks(BlocksToProcess.begin(),
1027  BlocksToProcess.end());
1028  IDFs.setDefiningBlocks(DefiningBlocks);
1029  IDFs.calculate(IDFBlocks);
1030 
1031  SmallSetVector<MemoryPhi *, 4> PhisToFill;
1032  // First create all needed Phis.
1033  for (auto *BBIDF : IDFBlocks)
1034  if (!MSSA->getMemoryAccess(BBIDF)) {
1035  auto *IDFPhi = MSSA->createMemoryPhi(BBIDF);
1036  InsertedPhis.push_back(IDFPhi);
1037  PhisToFill.insert(IDFPhi);
1038  }
1039  // Then update or insert their correct incoming values.
1040  for (auto *BBIDF : IDFBlocks) {
1041  auto *IDFPhi = MSSA->getMemoryAccess(BBIDF);
1042  assert(IDFPhi && "Phi must exist");
1043  if (!PhisToFill.count(IDFPhi)) {
1044  // Update existing Phi.
1045  // FIXME: some updates may be redundant, try to optimize and skip some.
1046  for (unsigned I = 0, E = IDFPhi->getNumIncomingValues(); I < E; ++I)
1047  IDFPhi->setIncomingValue(I, GetLastDef(IDFPhi->getIncomingBlock(I)));
1048  } else {
1049  for (auto &Pair : children<GraphDiffInvBBPair>({GD, BBIDF})) {
1050  BasicBlock *Pi = Pair.second;
1051  IDFPhi->addIncoming(GetLastDef(Pi), Pi);
1052  }
1053  }
1054  }
1055  }
1056 
1057  // Now for all defs in BlocksWithDefsToReplace, if there are uses they no
1058  // longer dominate, replace those with the closest dominating def.
1059  // This will also update optimized accesses, as they're also uses.
1060  for (auto *BlockWithDefsToReplace : BlocksWithDefsToReplace) {
1061  if (auto DefsList = MSSA->getWritableBlockDefs(BlockWithDefsToReplace)) {
1062  for (auto &DefToReplaceUses : *DefsList) {
1063  BasicBlock *DominatingBlock = DefToReplaceUses.getBlock();
1064  Value::use_iterator UI = DefToReplaceUses.use_begin(),
1065  E = DefToReplaceUses.use_end();
1066  for (; UI != E;) {
1067  Use &U = *UI;
1068  ++UI;
1070  if (MemoryPhi *UsrPhi = dyn_cast<MemoryPhi>(Usr)) {
1071  BasicBlock *DominatedBlock = UsrPhi->getIncomingBlock(U);
1072  if (!DT.dominates(DominatingBlock, DominatedBlock))
1073  U.set(GetLastDef(DominatedBlock));
1074  } else {
1075  BasicBlock *DominatedBlock = Usr->getBlock();
1076  if (!DT.dominates(DominatingBlock, DominatedBlock)) {
1077  if (auto *DomBlPhi = MSSA->getMemoryAccess(DominatedBlock))
1078  U.set(DomBlPhi);
1079  else {
1080  auto *IDom = DT.getNode(DominatedBlock)->getIDom();
1081  assert(IDom && "Block must have a valid IDom.");
1082  U.set(GetLastDef(IDom->getBlock()));
1083  }
1084  cast<MemoryUseOrDef>(Usr)->resetOptimized();
1085  }
1086  }
1087  }
1088  }
1089  }
1090  }
1091  tryRemoveTrivialPhis(InsertedPhis);
1092 }
1093 
1094 // Move What before Where in the MemorySSA IR.
1095 template <class WhereType>
1096 void MemorySSAUpdater::moveTo(MemoryUseOrDef *What, BasicBlock *BB,
1097  WhereType Where) {
1098  // Mark MemoryPhi users of What not to be optimized.
1099  for (auto *U : What->users())
1100  if (MemoryPhi *PhiUser = dyn_cast<MemoryPhi>(U))
1101  NonOptPhis.insert(PhiUser);
1102 
1103  // Replace all our users with our defining access.
1104  What->replaceAllUsesWith(What->getDefiningAccess());
1105 
1106  // Let MemorySSA take care of moving it around in the lists.
1107  MSSA->moveTo(What, BB, Where);
1108 
1109  // Now reinsert it into the IR and do whatever fixups needed.
1110  if (auto *MD = dyn_cast<MemoryDef>(What))
1111  insertDef(MD, /*RenameUses=*/true);
1112  else
1113  insertUse(cast<MemoryUse>(What), /*RenameUses=*/true);
1114 
1115  // Clear dangling pointers. We added all MemoryPhi users, but not all
1116  // of them are removed by fixupDefs().
1117  NonOptPhis.clear();
1118 }
1119 
1120 // Move What before Where in the MemorySSA IR.
1122  moveTo(What, Where->getBlock(), Where->getIterator());
1123 }
1124 
1125 // Move What after Where in the MemorySSA IR.
1127  moveTo(What, Where->getBlock(), ++Where->getIterator());
1128 }
1129 
1131  MemorySSA::InsertionPlace Where) {
1132  return moveTo(What, BB, Where);
1133 }
1134 
1135 // All accesses in To used to be in From. Move to end and update access lists.
1136 void MemorySSAUpdater::moveAllAccesses(BasicBlock *From, BasicBlock *To,
1137  Instruction *Start) {
1138 
1139  MemorySSA::AccessList *Accs = MSSA->getWritableBlockAccesses(From);
1140  if (!Accs)
1141  return;
1142 
1143  MemoryAccess *FirstInNew = nullptr;
1144  for (Instruction &I : make_range(Start->getIterator(), To->end()))
1145  if ((FirstInNew = MSSA->getMemoryAccess(&I)))
1146  break;
1147  if (!FirstInNew)
1148  return;
1149 
1150  auto *MUD = cast<MemoryUseOrDef>(FirstInNew);
1151  do {
1152  auto NextIt = ++MUD->getIterator();
1153  MemoryUseOrDef *NextMUD = (!Accs || NextIt == Accs->end())
1154  ? nullptr
1155  : cast<MemoryUseOrDef>(&*NextIt);
1156  MSSA->moveTo(MUD, To, MemorySSA::End);
1157  // Moving MUD from Accs in the moveTo above, may delete Accs, so we need to
1158  // retrieve it again.
1159  Accs = MSSA->getWritableBlockAccesses(From);
1160  MUD = NextMUD;
1161  } while (MUD);
1162 }
1163 
1165  BasicBlock *To,
1166  Instruction *Start) {
1167  assert(MSSA->getBlockAccesses(To) == nullptr &&
1168  "To block is expected to be free of MemoryAccesses.");
1169  moveAllAccesses(From, To, Start);
1170  for (BasicBlock *Succ : successors(To))
1171  if (MemoryPhi *MPhi = MSSA->getMemoryAccess(Succ))
1172  MPhi->setIncomingBlock(MPhi->getBasicBlockIndex(From), To);
1173 }
1174 
1176  Instruction *Start) {
1177  assert(From->getSinglePredecessor() == To &&
1178  "From block is expected to have a single predecessor (To).");
1179  moveAllAccesses(From, To, Start);
1180  for (BasicBlock *Succ : successors(From))
1181  if (MemoryPhi *MPhi = MSSA->getMemoryAccess(Succ))
1182  MPhi->setIncomingBlock(MPhi->getBasicBlockIndex(From), To);
1183 }
1184 
1185 /// If all arguments of a MemoryPHI are defined by the same incoming
1186 /// argument, return that argument.
1188  MemoryAccess *MA = nullptr;
1189 
1190  for (auto &Arg : MP->operands()) {
1191  if (!MA)
1192  MA = cast<MemoryAccess>(Arg);
1193  else if (MA != Arg)
1194  return nullptr;
1195  }
1196  return MA;
1197 }
1198 
1200  BasicBlock *Old, BasicBlock *New, ArrayRef<BasicBlock *> Preds,
1201  bool IdenticalEdgesWereMerged) {
1202  assert(!MSSA->getWritableBlockAccesses(New) &&
1203  "Access list should be null for a new block.");
1204  MemoryPhi *Phi = MSSA->getMemoryAccess(Old);
1205  if (!Phi)
1206  return;
1207  if (Old->hasNPredecessors(1)) {
1208  assert(pred_size(New) == Preds.size() &&
1209  "Should have moved all predecessors.");
1210  MSSA->moveTo(Phi, New, MemorySSA::Beginning);
1211  } else {
1212  assert(!Preds.empty() && "Must be moving at least one predecessor to the "
1213  "new immediate predecessor.");
1214  MemoryPhi *NewPhi = MSSA->createMemoryPhi(New);
1215  SmallPtrSet<BasicBlock *, 16> PredsSet(Preds.begin(), Preds.end());
1216  // Currently only support the case of removing a single incoming edge when
1217  // identical edges were not merged.
1218  if (!IdenticalEdgesWereMerged)
1219  assert(PredsSet.size() == Preds.size() &&
1220  "If identical edges were not merged, we cannot have duplicate "
1221  "blocks in the predecessors");
1222  Phi->unorderedDeleteIncomingIf([&](MemoryAccess *MA, BasicBlock *B) {
1223  if (PredsSet.count(B)) {
1224  NewPhi->addIncoming(MA, B);
1225  if (!IdenticalEdgesWereMerged)
1226  PredsSet.erase(B);
1227  return true;
1228  }
1229  return false;
1230  });
1231  Phi->addIncoming(NewPhi, New);
1232  tryRemoveTrivialPhi(NewPhi);
1233  }
1234 }
1235 
1237  assert(!MSSA->isLiveOnEntryDef(MA) &&
1238  "Trying to remove the live on entry def");
1239  // We can only delete phi nodes if they have no uses, or we can replace all
1240  // uses with a single definition.
1241  MemoryAccess *NewDefTarget = nullptr;
1242  if (MemoryPhi *MP = dyn_cast<MemoryPhi>(MA)) {
1243  // Note that it is sufficient to know that all edges of the phi node have
1244  // the same argument. If they do, by the definition of dominance frontiers
1245  // (which we used to place this phi), that argument must dominate this phi,
1246  // and thus, must dominate the phi's uses, and so we will not hit the assert
1247  // below.
1248  NewDefTarget = onlySingleValue(MP);
1249  assert((NewDefTarget || MP->use_empty()) &&
1250  "We can't delete this memory phi");
1251  } else {
1252  NewDefTarget = cast<MemoryUseOrDef>(MA)->getDefiningAccess();
1253  }
1254 
1255  SmallSetVector<MemoryPhi *, 4> PhisToCheck;
1256 
1257  // Re-point the uses at our defining access
1258  if (!isa<MemoryUse>(MA) && !MA->use_empty()) {
1259  // Reset optimized on users of this store, and reset the uses.
1260  // A few notes:
1261  // 1. This is a slightly modified version of RAUW to avoid walking the
1262  // uses twice here.
1263  // 2. If we wanted to be complete, we would have to reset the optimized
1264  // flags on users of phi nodes if doing the below makes a phi node have all
1265  // the same arguments. Instead, we prefer users to removeMemoryAccess those
1266  // phi nodes, because doing it here would be N^3.
1267  if (MA->hasValueHandle())
1268  ValueHandleBase::ValueIsRAUWd(MA, NewDefTarget);
1269  // Note: We assume MemorySSA is not used in metadata since it's not really
1270  // part of the IR.
1271 
1272  while (!MA->use_empty()) {
1273  Use &U = *MA->use_begin();
1274  if (auto *MUD = dyn_cast<MemoryUseOrDef>(U.getUser()))
1275  MUD->resetOptimized();
1276  if (OptimizePhis)
1277  if (MemoryPhi *MP = dyn_cast<MemoryPhi>(U.getUser()))
1278  PhisToCheck.insert(MP);
1279  U.set(NewDefTarget);
1280  }
1281  }
1282 
1283  // The call below to erase will destroy MA, so we can't change the order we
1284  // are doing things here
1285  MSSA->removeFromLookups(MA);
1286  MSSA->removeFromLists(MA);
1287 
1288  // Optionally optimize Phi uses. This will recursively remove trivial phis.
1289  if (!PhisToCheck.empty()) {
1290  SmallVector<WeakVH, 16> PhisToOptimize{PhisToCheck.begin(),
1291  PhisToCheck.end()};
1292  PhisToCheck.clear();
1293 
1294  unsigned PhisSize = PhisToOptimize.size();
1295  while (PhisSize-- > 0)
1296  if (MemoryPhi *MP =
1297  cast_or_null<MemoryPhi>(PhisToOptimize.pop_back_val()))
1298  tryRemoveTrivialPhi(MP);
1299  }
1300 }
1301 
1303  const SmallSetVector<BasicBlock *, 8> &DeadBlocks) {
1304  // First delete all uses of BB in MemoryPhis.
1305  for (BasicBlock *BB : DeadBlocks) {
1306  Instruction *TI = BB->getTerminator();
1307  assert(TI && "Basic block expected to have a terminator instruction");
1308  for (BasicBlock *Succ : successors(TI))
1309  if (!DeadBlocks.count(Succ))
1310  if (MemoryPhi *MP = MSSA->getMemoryAccess(Succ)) {
1311  MP->unorderedDeleteIncomingBlock(BB);
1312  tryRemoveTrivialPhi(MP);
1313  }
1314  // Drop all references of all accesses in BB
1315  if (MemorySSA::AccessList *Acc = MSSA->getWritableBlockAccesses(BB))
1316  for (MemoryAccess &MA : *Acc)
1317  MA.dropAllReferences();
1318  }
1319 
1320  // Next, delete all memory accesses in each block
1321  for (BasicBlock *BB : DeadBlocks) {
1322  MemorySSA::AccessList *Acc = MSSA->getWritableBlockAccesses(BB);
1323  if (!Acc)
1324  continue;
1325  for (auto AB = Acc->begin(), AE = Acc->end(); AB != AE;) {
1326  MemoryAccess *MA = &*AB;
1327  ++AB;
1328  MSSA->removeFromLookups(MA);
1329  MSSA->removeFromLists(MA);
1330  }
1331  }
1332 }
1333 
1334 void MemorySSAUpdater::tryRemoveTrivialPhis(ArrayRef<WeakVH> UpdatedPHIs) {
1335  for (auto &VH : UpdatedPHIs)
1336  if (auto *MPhi = cast_or_null<MemoryPhi>(VH))
1337  tryRemoveTrivialPhi(MPhi);
1338 }
1339 
1341  const BasicBlock *BB = I->getParent();
1342  // Remove memory accesses in BB for I and all following instructions.
1343  auto BBI = I->getIterator(), BBE = BB->end();
1344  // FIXME: If this becomes too expensive, iterate until the first instruction
1345  // with a memory access, then iterate over MemoryAccesses.
1346  while (BBI != BBE)
1347  removeMemoryAccess(&*(BBI++));
1348  // Update phis in BB's successors to remove BB.
1349  SmallVector<WeakVH, 16> UpdatedPHIs;
1350  for (const BasicBlock *Successor : successors(BB)) {
1352  if (MemoryPhi *MPhi = MSSA->getMemoryAccess(Successor)) {
1353  MPhi->unorderedDeleteIncomingBlock(BB);
1354  UpdatedPHIs.push_back(MPhi);
1355  }
1356  }
1357  // Optimize trivial phis.
1358  tryRemoveTrivialPhis(UpdatedPHIs);
1359 }
1360 
1362  const BasicBlock *To) {
1363  const BasicBlock *BB = BI->getParent();
1364  SmallVector<WeakVH, 16> UpdatedPHIs;
1365  for (const BasicBlock *Succ : successors(BB)) {
1367  if (Succ != To)
1368  if (auto *MPhi = MSSA->getMemoryAccess(Succ)) {
1369  MPhi->unorderedDeleteIncomingBlock(BB);
1370  UpdatedPHIs.push_back(MPhi);
1371  }
1372  }
1373  // Optimize trivial phis.
1374  tryRemoveTrivialPhis(UpdatedPHIs);
1375 }
1376 
1378  Instruction *I, MemoryAccess *Definition, const BasicBlock *BB,
1379  MemorySSA::InsertionPlace Point) {
1380  MemoryUseOrDef *NewAccess = MSSA->createDefinedAccess(I, Definition);
1381  MSSA->insertIntoListsForBlock(NewAccess, BB, Point);
1382  return NewAccess;
1383 }
1384 
1386  Instruction *I, MemoryAccess *Definition, MemoryUseOrDef *InsertPt) {
1387  assert(I->getParent() == InsertPt->getBlock() &&
1388  "New and old access must be in the same block");
1389  MemoryUseOrDef *NewAccess = MSSA->createDefinedAccess(I, Definition);
1390  MSSA->insertIntoListsBefore(NewAccess, InsertPt->getBlock(),
1391  InsertPt->getIterator());
1392  return NewAccess;
1393 }
1394 
1396  Instruction *I, MemoryAccess *Definition, MemoryAccess *InsertPt) {
1397  assert(I->getParent() == InsertPt->getBlock() &&
1398  "New and old access must be in the same block");
1399  MemoryUseOrDef *NewAccess = MSSA->createDefinedAccess(I, Definition);
1400  MSSA->insertIntoListsBefore(NewAccess, InsertPt->getBlock(),
1401  ++InsertPt->getIterator());
1402  return NewAccess;
1403 }
const_iterator end(StringRef path)
Get end iterator over path.
Definition: Path.cpp:233
AccessList * getWritableBlockAccesses(const BasicBlock *BB) const
Definition: MemorySSA.h:803
bool hasNPredecessors(unsigned N) const
Return true if this block has exactly N predecessors.
Definition: BasicBlock.cpp:260
static MemoryAccess * onlySingleValue(MemoryPhi *MP)
If all arguments of a MemoryPHI are defined by the same incoming argument, return that argument...
const_iterator begin(StringRef path, Style style=Style::native)
Get begin iterator over path.
Definition: Path.cpp:224
void dropAllReferences()
Drop all references to operands.
Definition: User.h:294
This class represents lattice values for constants.
Definition: AllocatorList.h:23
AllAccessType::self_iterator getIterator()
Get the iterators for the all access list and the defs only list We default to the all access list...
Definition: MemorySSA.h:178
bool dominates(const MemoryAccess *A, const MemoryAccess *B) const
Given two memory accesses in potentially different blocks, determine whether MemoryAccess A dominates...
Definition: MemorySSA.cpp:2138
MemoryAccess * getDefiningAccess() const
Get the access that produces the memory state used by this Use.
Definition: MemorySSA.h:257
void updateExitBlocksForClonedLoop(ArrayRef< BasicBlock *> ExitBlocks, const ValueToValueMapTy &VMap, DominatorTree &DT)
Update phi nodes in exit block successors following cloning.
iterator begin() const
Definition: ArrayRef.h:136
BasicBlock * getSuccessor(unsigned Idx) const
Return the specified successor. This instruction must be a terminator.
void applyInsertUpdates(ArrayRef< CFGUpdate > Updates, DominatorTree &DT)
Apply CFG insert updates, analogous with the DT edge updates.
const AccessList * getBlockAccesses(const BasicBlock *BB) const
Return the list of MemoryAccess&#39;s for a given basic block.
Definition: MemorySSA.h:758
void changeToUnreachable(const Instruction *I)
Instruction I will be changed to an unreachable.
void insertUse(MemoryUse *Use, bool RenameUses=false)
This file contains the declarations for metadata subclasses.
bool hasValueHandle() const
Return true if there is a value handle associated with this value.
Definition: Value.h:505
void moveAfter(MemoryUseOrDef *What, MemoryUseOrDef *Where)
Represents a read-write access to memory, whether it is a must-alias, or a may-alias.
Definition: MemorySSA.h:375
NodeT * findNearestCommonDominator(NodeT *A, NodeT *B) const
findNearestCommonDominator - Find nearest common dominator basic block for basic block A and B...
int getBasicBlockIndex(const BasicBlock *BB) const
Return the first index of the specified basic block in the value list for this PHI.
Definition: MemorySSA.h:575
reverse_iterator rbegin()
Definition: BasicBlock.h:273
const Instruction * 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:137
DefsOnlyType::self_iterator getDefsIterator()
Definition: MemorySSA.h:190
void calculate(SmallVectorImpl< NodeTy *> &IDFBlocks)
Calculate iterated dominance frontiers.
iterator end()
Get an iterator to the end of the SetVector.
Definition: SetVector.h:92
bool isReachableFromEntry(const Use &U) const
Provide an overload for a Use.
Definition: Dominators.cpp:299
op_iterator op_begin()
Definition: User.h:229
void setIncomingValue(unsigned I, MemoryAccess *V)
Definition: MemorySSA.h:534
iterator begin()
Instruction iterator methods.
Definition: BasicBlock.h:268
Represents read-only accesses to memory.
Definition: MemorySSA.h:319
void removeEdge(BasicBlock *From, BasicBlock *To)
Update the MemoryPhi in To following an edge deletion between From and To.
void renamePass(BasicBlock *BB, MemoryAccess *IncomingVal, SmallPtrSetImpl< BasicBlock *> &Visited)
Definition: MemorySSA.h:822
block_iterator block_begin()
Definition: MemorySSA.h:500
MemoryUseOrDef * createDefinedAccess(Instruction *, MemoryAccess *, const MemoryUseOrDef *Template=nullptr, bool CreationMustSucceed=true)
Definition: MemorySSA.cpp:1699
A Use represents the edge between a Value definition and its users.
Definition: Use.h:55
This class consists of common code factored out of the SmallVector class to reduce code duplication b...
Definition: APFloat.h:41
Encapsulates MemorySSA, including all data associated with memory accesses.
Definition: MemorySSA.h:703
void applyUpdates(ArrayRef< CFGUpdate > Updates, DominatorTree &DT)
Apply CFG updates, analogous with the DT edge updates.
const DefsList * getBlockDefs(const BasicBlock *BB) const
Return the list of MemoryDef&#39;s and MemoryPhi&#39;s for a given basic block.
Definition: MemorySSA.h:766
void insertIntoListsForBlock(MemoryAccess *, const BasicBlock *, InsertionPlace)
Definition: MemorySSA.cpp:1592
mir Rename Register Operands
A simple intrusive list implementation.
Definition: simple_ilist.h:78
User * getUser() const LLVM_READONLY
Returns the User that contains this Use.
Definition: Use.cpp:40
bool insert(const value_type &X)
Insert a new element into the SetVector.
Definition: SetVector.h:141
MemoryUseOrDef * getMemoryAccess(const Instruction *I) const
Given a memory Mod/Ref&#39;ing instruction, get the MemorySSA access associated with it.
Definition: MemorySSA.h:720
iterator begin()
Get an iterator to the beginning of the SetVector.
Definition: SetVector.h:82
ArrayRef - Represent a constant reference to an array (0 or more elements consecutively in memory)...
Definition: APInt.h:32
void replaceAllUsesWith(Value *V)
Change all uses of this to point to a new Value.
Definition: Value.cpp:429
ValueT lookup(const KeyT &Val) const
lookup - Return the entry for the specified key, or a default constructed value if no such entry exis...
Definition: ValueMap.h:165
static void ValueIsRAUWd(Value *Old, Value *New)
Definition: Value.cpp:913
static bool ProcessBlock(BasicBlock &BB, DominatorTree &DT, LoopInfo &LI, AAResults &AA)
Definition: Sink.cpp:198
MemoryUseOrDef * createMemoryAccessBefore(Instruction *I, MemoryAccess *Definition, MemoryUseOrDef *InsertPt)
Create a MemoryAccess in MemorySSA before or after an existing MemoryAccess.
Concrete subclass of DominatorTreeBase that is used to compute a normal dominator tree...
Definition: Dominators.h:144
void replaceUsesWithIf(Value *New, llvm::function_ref< bool(Use &U)> ShouldReplace)
Go through the uses list for this definition and make each use point to "V" if the callback ShouldRep...
Definition: Value.h:300
size_type count(const key_type &key) const
Count the number of elements of a given key in the SetVector.
Definition: SetVector.h:210
MemoryAccess * getIncomingValue(unsigned I) const
Return incoming value number x.
Definition: MemorySSA.h:533
use_iterator_impl< Use > use_iterator
Definition: Value.h:351
NodeT * getBlock() const
static GCRegistry::Add< OcamlGC > B("ocaml", "ocaml 3.10-compatible GC")
const BasicBlock * getSinglePredecessor() const
Return the predecessor of this block if it has a single predecessor block.
Definition: BasicBlock.cpp:233
static constexpr UpdateKind Insert
void updatePhisWhenInsertingUniqueBackedgeBlock(BasicBlock *LoopHeader, BasicBlock *LoopPreheader, BasicBlock *BackedgeBlock)
Update MemorySSA when inserting a unique backedge block for a loop.
void set(Value *Val)
Definition: Value.h:719
AllAccessType::reverse_self_iterator getReverseIterator()
Definition: MemorySSA.h:184
LLVM Basic Block Representation.
Definition: BasicBlock.h:57
static void setMemoryPhiValueForBlock(MemoryPhi *MP, const BasicBlock *BB, MemoryAccess *NewDef)
Conditional or Unconditional Branch instruction.
Value handle that tracks a Value across RAUW.
Definition: ValueHandle.h:327
size_t size() const
size - Get the array size.
Definition: ArrayRef.h:148
DomTreeNodeBase * getIDom() const
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
void removeDuplicatePhiEdgesBetween(const BasicBlock *From, const BasicBlock *To)
Update the MemoryPhi in To to have a single incoming edge from From, following a CFG change that repl...
void addIncoming(MemoryAccess *V, BasicBlock *BB)
Add an incoming value to the end of the PHI list.
Definition: MemorySSA.h:564
std::pair< iterator, bool > insert(PtrType Ptr)
Inserts Ptr if and only if there is no element in the container equal to Ptr.
Definition: SmallPtrSet.h:370
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:112
InsertionPlace
Used in various insertion functions to specify whether we are talking about the beginning or end of a...
Definition: MemorySSA.h:788
void moveAllAfterSpliceBlocks(BasicBlock *From, BasicBlock *To, Instruction *Start)
From block was spliced into From and To.
op_iterator op_end()
Definition: User.h:231
amdgpu Simplify well known AMD library false FunctionCallee Value * Arg
Interval::pred_iterator pred_end(Interval *I)
Definition: Interval.h:115
op_range operands()
Definition: User.h:237
size_type count(ConstPtrType Ptr) const
count - Return 1 if the specified pointer is in the set, 0 otherwise.
Definition: SmallPtrSet.h:381
self_iterator getIterator()
Definition: ilist_node.h:81
void updateForClonedLoop(const LoopBlocksRPO &LoopBlocks, ArrayRef< BasicBlock *> ExitBlocks, const ValueToValueMapTy &VM, bool IgnoreIncomingWithNoClones=false)
Update MemorySSA after a loop was cloned, given the blocks in RPO order, the exit blocks and a 1:1 ma...
MemoryAccess * createMemoryAccessInBB(Instruction *I, MemoryAccess *Definition, const BasicBlock *BB, MemorySSA::InsertionPlace Point)
Create a MemoryAccess in MemorySSA at a specified point in a block, with a specified clobbering defin...
DomTreeNodeBase< NodeT > * getNode(const NodeT *BB) const
getNode - return the (Post)DominatorTree node for the specified basic block.
An intrusive list with ownership and callbacks specified/controlled by ilist_traits, only with API safe for polymorphic types.
Definition: ilist.h:388
#define llvm_unreachable(msg)
Marks that the current location is not supposed to be reachable.
void moveToPlace(MemoryUseOrDef *What, BasicBlock *BB, MemorySSA::InsertionPlace Where)
void setDefiningBlocks(const SmallPtrSetImpl< NodeTy *> &Blocks)
Give the IDF calculator the set of blocks in which the value is defined.
iterator_range< T > make_range(T x, T y)
Convenience function for iterating over sub-ranges.
A SetVector that performs no allocations if smaller than a certain size.
Definition: SetVector.h:297
void setDefiningAccess(MemoryAccess *DMA, bool Optimized=false, Optional< AliasResult > AR=MayAlias)
Definition: MemorySSA.h:299
unsigned getNumOperands() const
Definition: User.h:191
SmallPtrSet - This class implements a set which is optimized for holding SmallSize or less elements...
Definition: SmallPtrSet.h:417
BlockVerifier::State From
iterator end()
Definition: BasicBlock.h:270
This is a &#39;vector&#39; (really, a variable-sized array), optimized for the case when the array is small...
Definition: SmallVector.h:837
bool dominates(const Instruction *Def, const Use &U) const
Return true if Def dominates a use in User.
Definition: Dominators.cpp:248
Module.h This file contains the declarations for the Module class.
iterator end() const
Definition: ArrayRef.h:137
void wireOldPredecessorsToNewImmediatePredecessor(BasicBlock *Old, BasicBlock *New, ArrayRef< BasicBlock *> Preds, bool IdenticalEdgesWereMerged=true)
A new empty BasicBlock (New) now branches directly to Old.
static MemoryAccess * getNewDefiningAccessForClone(MemoryAccess *MA, const ValueToValueMapTy &VMap, PhiToDefMap &MPhiMap, bool CloneWasSimplified, MemorySSA *MSSA)
BasicBlock * getBlock() const
Definition: MemorySSA.h:159
pred_range predecessors(BasicBlock *BB)
Definition: CFG.h:124
MemoryAccess * getLiveOnEntryDef() const
Definition: MemorySSA.h:742
raw_ostream & dbgs()
dbgs() - This returns a reference to a raw_ostream for debugging messages.
Definition: Debug.cpp:132
Class that has the common methods + fields of memory uses/defs.
Definition: MemorySSA.h:247
iterator_range< user_iterator > users()
Definition: Value.h:419
unsigned getNumIncomingValues() const
Return the number of incoming edges.
Definition: MemorySSA.h:530
BasicBlock * getIncomingBlock(unsigned I) const
Return incoming basic block number i.
Definition: MemorySSA.h:543
use_iterator use_begin()
Definition: Value.h:358
block_iterator block_end()
Definition: MemorySSA.h:511
LLVM_NODISCARD bool empty() const
Definition: SmallVector.h:55
unsigned pred_size(const BasicBlock *BB)
Get the number of predecessors of BB.
Definition: CFG.h:121
#define I(x, y, z)
Definition: MD5.cpp:58
bool empty() const
Determine if the SetVector is empty or not.
Definition: SetVector.h:72
MemoryUseOrDef * createMemoryAccessAfter(Instruction *I, MemoryAccess *Definition, MemoryAccess *InsertPt)
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:332
void moveAllAfterMergeBlocks(BasicBlock *From, BasicBlock *To, Instruction *Start)
From block was merged into To.
bool isLiveOnEntryDef(const MemoryAccess *MA) const
Return true if MA represents the live on entry value.
Definition: MemorySSA.h:738
reverse_iterator rend(StringRef path)
Get reverse end iterator over path.
Definition: Path.cpp:304
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:185
iterator insert(iterator I, reference Node)
Insert a node by reference; never copies.
Definition: simple_ilist.h:159
Wrapper class to LoopBlocksDFS that provides a standard begin()/end() interface for the DFS reverse p...
Definition: LoopIterator.h:172
void removeMemoryAccess(MemoryAccess *, bool OptimizePhis=false)
Remove a MemoryAccess from MemorySSA, including updating all definitions and uses.
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
void removeBlocks(const SmallSetVector< BasicBlock *, 8 > &DeadBlocks)
Remove all MemoryAcceses in a set of BasicBlocks about to be deleted.
user_iterator user_begin()
Definition: Value.h:395
void insertDef(MemoryDef *Def, bool RenameUses=false)
Insert a definition into the MemorySSA IR.
succ_range successors(Instruction *I)
Definition: CFG.h:259
DefsOnlyType::reverse_self_iterator getReverseDefsIterator()
Definition: MemorySSA.h:196
This file exposes an interface to building/using memory SSA to walk memory instructions using a use/d...
Represents phi nodes for memory accesses.
Definition: MemorySSA.h:481
DefsList * getWritableBlockDefs(const BasicBlock *BB) const
Definition: MemorySSA.h:809
#define LLVM_DEBUG(X)
Definition: Debug.h:122
OutputIt copy(R &&Range, OutputIt Out)
Definition: STLExtras.h:1217
void moveBefore(MemoryUseOrDef *What, MemoryUseOrDef *Where)
void updateForClonedBlockIntoPred(BasicBlock *BB, BasicBlock *P1, const ValueToValueMapTy &VM)
bool use_empty() const
Definition: Value.h:342
void changeCondBranchToUnconditionalTo(const BranchInst *BI, const BasicBlock *To)
Conditional branch BI is changed or replaced with an unconditional branch to To.
bool empty() const
empty - Check if the array is empty.
Definition: ArrayRef.h:143
const BasicBlock * getParent() const
Definition: Instruction.h:66
user_iterator user_end()
Definition: Value.h:403